0 387 28942 9

Neuroanatomical
Tract-Tracing 3
Molecules, Neurons, and Systems
Neuroanatomical
Tract-Tracing 3
Molecules, Neurons, and Systems
Edited by
Laszlo Zaborszky
Rutgers University
Newark, NJ, USA
Floris G. Wouterlood
Vrije University
Amsterdam, The Netherlands
José Luis Lanciego

University of Navarra
Pamplona, Spain
Laszlo Zaborszky Floris G. Wouterlood José Luis Lanciego
Center for Molecular and Department of Anatomy Neurosciences Division
Behavioral Neuroscience Vrije University Medical Center Center for Applied
Rutgers University Amsterdam, The Netherlands Medical Research
Newark, NJ, USA (CIMA)
University of Navarra
Pamplona, Spain
Cover illustration: Reconstruction of functional connectivity between neurons based on the

temporal coherence of spiking activity recorded extracellulary from the somatosensory cortex
of a rat. See Fig. 20.8 on page 666.
Library of Congress Control Number: 2005932857

Printed on acid-free paper.
ISBN-10: 0-387-28941-0 e-ISBN: 0-387-28942-9
ISBN-13: 978-0387-28941-0

C 2006 Springer Science+Business Media, Inc.
All rights reserved. This work may not be translated or copied in whole or in part without the
written permission of the publisher (Springer Science+Business Media, Inc., 233 Spring Street,
New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly
analysis. Use in connection with any form of information storage and retrieval, electronic
adaptation, computer software, or by similar or dissimilar methodology now known or hereafter
developed is forbidden.
The use in this publication of trade names, trademarks, service marks, and similar terms, even
if they are not identified as such, is not to be taken as an expression of opinion as to whether
or not they are subject to proprietary rights.
Printed in the United States of America. (TB/KYO)
9 8 7 6 5 4 3 2 1
springer.com
To Sarah Z for her 18th birthday
Preface
Between the first edition of Neuroanatomical Tract-Tracing Methods in 1981

and the current, third edition, neuroscience has witnessed a total transi-
tion into the information age. Scientists, whether they wanted it or not,
have turned digital. Today, everyone is linked up with worldwide computer
communication networks, with local and worldwide digital environments
offering vastly increased speed and accuracy of data acquisition and pro-
cessing. Communication and exchange of information between scientists
worldwide is a matter of seconds. The electronic dissemination of research
data has become routine. Publication of scientific results has changed from
the typewritten manuscript to electronic online submission. Search engines,
PubMed-like services, and electronic notification and delivery services are
making life more convenient for scientists. What is not on the Web does not exist.
Do we still need books?
We think positively about books. In the first place, it is common sense
to have at hand a printed technical protocol in the setting of a laboratory
engaged in experimental neuroscience. Although the workbench protocol
does not necessarily have to be a book, it is nonetheless helpful to have a
book at hand that not only provides the technical protocol, but also explains
why the protocol is designed as it is, what the alternatives are, and what their
consequences are. The Web is a wonderful yet particularly fluid medium
in which things change very quickly. Data that were here today are gone
tomorrow. A book has a longer time constant, which sometimes is beneficial.
The first two editions of the book (Heimer and Robards, 1981; Heimer
and Zaborszky, 1989—both published by Plenum Press) had a tremendous
impact on neuroscience. They are still among the frequently consulted
books in the laboratory. We feel that the moment has arrived to pursue
a third, thoroughly updated version of this landmark book, in order to con-
tinue the line originated by Neuroanatomical Tract-Tracing Methods. The target
audience remains the graduate students and young investigators working in
the laboratory, seeking fast, complete, up-to-date, and immediately applica-
ble information about techniques, written by acknowledged experts in the
field.
Since the last edition, several methods that were in their infancy 15 years
ago have become routine, older methods have experienced a renaissance,
and newly emerging techniques need validation. Molecular techniques, such
as genomics and proteomics, have become established methods, which allow
for the study of gene expression of recorded and traced neurons (chapters
by Ginsberg, Griffith, Stornetta, and their colleagues). The simultaneous
vii
viii PREFACE
development of new fluorescence probes, single- and multiphoton confo-

cal laser scanning microscopes, and vastly increased computer processing
power have contributed to a renaissance of fluorescence methods (chapters
by Lanciego, Molnar, Reiner, and Wouterlood). In vivo tractography as well
as structural and functional imaging techniques allow for the study of the liv-
ing human brain with a degree of detail never dreamed possible (chapter by
Amunts and Zilles). Immunocytochemistry using pre- and postembedding
electron microscopy for identifying neuroactive substances still remains an
art which has to be tailored to one’s needs (chapters by Sesack, Ottersen, and
their coworkers). Viral tracers for the analysis of neural circuits have become
established in some laboratories (chapter by Geerling and his colleagues).
Several chapters briefly discuss how databases help in acquisition, analy-
sis, modeling, and integration of complex cross-scale data sets (chapters by
Zilles, Bjaalie, and Nadasdy). Readers interested in these topics are referred
to a recently published textbook edited by Koslow and Subramaniam (2005)
that summarizes efforts in this field led by the Human Brain Project of the
National Institutes of Health.
As neuroscience research progresses, we witness drastic changes in how
methods are used. The previous editions of this series reflected the reduc-
tionist approach to study neuroscience which was characteristic for the previ-
ous century. Even in the second edition, only 2 out of 13 chapters combined
techniques that crossed the traditional borders of anatomy and physiol-
ogy. In the present edition, most of the chapters describe methods, which
allow for the integration of molecular, cellular, and system level data, re-
flecting a holistic-integrative approach to neuroscience in the twenty-first
century. Specifically, using sophisticated combinations of tracing methods,
the molecular and genetic identity of a neuron (chapters by Ginsberg and
his colleagues), as well as the synaptology of any circuitry can be accu-
rately determined (chapter by Sesack and her coworkers). Using extracel-
lular, juxtacellular, and intracellular recordings (chapters by Duque and
Zaborszky, and Sik), anatomical features can be correlated with electroen-
cephalographic (EEG), multiunit activity (MUA), local field potentials, and
intrinsic membrane characteristics. Recent advances in voltage-sensitive dye
imaging (chapter by Petersen) and two-photon calcium imaging (chapter
by Goldberg et al.) are promising techniques for studying the spatiotem-
poral dynamics of hundreds of neurons in the living brain. Sophisticated
statistical designs (Avendaño) and expanding computational approaches
have the potential to capture full three-dimensional (3D) relations of neu-
ronal and architectonic features of entire brain systems (chapters by Ascoli,
Bjaalie, and their colleagues). The last chapter (by Nadasdy et al.) predicts
that within the next 10 years the complete 3D vectorial database of the rat
brain will be available to address specific questions about hidden organiza-
tion principles of the nervous system. However, these authors also address
the gap that exists in our understanding between “structural” and “func-
tional” connectivity (e.g., Friston et al., 1993). We hope that students of this
book will bridge this gap eventually, leading to a better understanding of
PREFACE ix
how the human brain functions in health, aging, and disease. This is our
ultimate goal.
ACKNOWLEDGMENT
It has been an honor as well as a pleasure to select contributors for the

current edition of Neuroanatomical Tract-Tracing Methods and to produce with
them this book. However, this book would never have been written with-
out several generations of scientists designing and optimizing the meth-
ods discussed in the chapters. Among all those whose shared legacy is the
current technological standard, we would like to specially mention the re-
cently deceased Sanford Palay (Sandy) (1992), who saw the first synapse
using electron microscopy and Theodor Blackstad (see, e.g., Blackstad and
Bjaalie, 1988) whose contribution was instrumental to computational neu-
roanatomy as we understand it today. We had the great privilege to work
or interact with them. However, we learned the most from our mentors,
Lennart Heimer and Enrico Mugnaini, pioneers in tract-tracing and cellu-
lar neuroscience. They were not only mentors and teachers, but friends as
well. We dedicate this book to them.
We thank the authors for their contributions and patience during the
somewhat lengthy editorial process. They exerted great effort in writing
their chapters, and also provided essential feedback by cross-reviewing each
manuscript. We are also indebted to Drs. James Tepper (Rutgers University),
Harry Uylings (The Netherlands Brain Research Institute), and Rolf Kötter
(C. and O. Vogt Brain Research Institute, Düsseldorf) who as external re-
viewers read earlier versions of some of the chapters. We have the fortune to
have on our side Kathleen Lyon, Senior Biosciences Editor of Springer, who
spearheaded this edition with great enthusiasm. Last, but not least, we would
like to acknowledge Professors Ian Creese and Paula Tallal, codirectors of
the Center for Molecular and Behavioral Neuroscience, Rutgers University,
for encouragement. The National Institutes of Health (L.Z.) gave gener-
ous financial support over many years. Elizabeth Hur, helping the editorial
process at Rutgers University, was the first Graduate Student who read and
benefited from this book. We hope many will follow.
Laszlo Zaborszky
Jose L. Lanciego
REFERENCES
Blackstad, T. W., and Bjaalie, J. G., 1988, Computer programs for neuroanatomy: three-
dimensional reconstruction and analysis of populations of cortical neurons and other
bodies with a laminar distribution. Comput. Biol. Med. 18:321–340.
x PREFACE
Friston, K. J., Frith, C. D., Liddle, P. F., and Frackowiak, R. S., 1993, Functional connectivity: the
principal-component analysis of large (PET) data sets. J. Cereb. Blood Flow Metab. 13:5–14.
Heimer, L., and Robards, M., 1981, Neuroanatomical Tract-Tracing Methods, New York: Plenum
Press, p. 567.
Heimer, L., and Zaborszky, L., 1989, Neuroanatomical Tract-Tracing Methods 2. Recent Progress,
New York: Plenum Press, p. 408.
Koslow, S. H., and Subramaniam, S., 2005, Databasing the Brain: From Data to Knowledge, Hoboken:
Wiley-Liss, p. 466.
Palay, S. L., 1992, A concatenation of accidents, In: Samson, F. S., Adelman, G. (eds.), The
Neurosciences: Paths of Discovery, II, Boston: Birkhauser, pp. 191–212.
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
1. Short Retrospection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Lennart Heimer
2. Preembedding Immunoelectron Microscopy: Applications for

Studies of the Nervous System . . . . . . . . . . . . . . . . . . . . . . . 6
Susan R. Sesack, LeeAnn H. Miner, and Natalia Omelchenko
3. Postembedding Immunogold Cytochemistry of Membrane

Molecules and Amino Acid Transmitters in the Central
Nervous System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Thomas Misje Mathiisen, Erlend Arnulf Nagelhus, Bahareh Jouleh,
Reidun Torp, Didrik Sølie Frydenlund, Maria-Niki Mylonakou,
Mahmood Amiry-Moghaddam, Luciene Covolan, Jo Kristian Utvik,
Bjørg Riber, Karen Marie Gujord, Jorunn Knutsen, Øivind Skare,
Petter Laake, Svend Davanger, Finn-Mogens Haug, Eric Rinvik,
and Ole Petter Ottersen
4. Cell and Tissue Microdissection in Combination with Genomic

and Proteomic Applications . . . . . . . . . . . . . . . . . . . . . . . . 109
Stephen D. Ginsberg, Scott E. Hemby, Elliott J. Mufson,
and Lee J. Martin
5. Molecules and Membrane Activity: Single-Cell RT-PCR and

Patch-Clamp Recording from Central Neurons . . . . . . . . . . . . 142
William H. Griffith, Sun-Ho Han, Brian A. McCool,
and David Murchison
6. Merging Structure and Function: Combination of In Vivo

Extracellular and Intracellular Electrophysiological Recordings
with Neuroanatomical Techniques . . . . . . . . . . . . . . . . . . . . 175
Attila Sı́k
7. Juxtacellular Labeling of Individual Neurons In Vivo:

From Electrophysiology to Synaptology . . . . . . . . . . . . . . . . . 197
Alvaro Duque and Laszlo Zaborszky
xi
xii CONTENTS
8. Nonradioactive In Situ Hybridization in Combination with

Tract-Tracing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Ruth L. Stornetta and Patrice G. Guyenet
9. Viral Tracers for the Analysis of Neural Circuits . . . . . . . . . . . 263

Joel C. Geerling, Thomas C. Mettenleiter, and Arthur D. Loewy
10. Dextran Amines: Versatile Tools for Anterograde and Retrograde

Studies of Nervous System Connectivity . . . . . . . . . . . . . . . . . 304
Anton Reiner and Marcia G. Honig
11. Multiple Neuroanatomical Tract-Tracing: Approaches for

Multiple Tract-Tracing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
José L. Lanciego and Floris G. Wouterlood
12. Tract-Tracing in Developing Systems and in Postmortem Human

Material Using Carbocyanine Dyes . . . . . . . . . . . . . . . . . . . . 366
Zoltán Molnár, Daniel Blakey, Irina Bystron, and
Rosalind S. E. Carney
13. Combined Fluorescence Methods to Determine Synapses in

the Light Microscope: Multilabel Confocal Laser Scanning
Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
14. Advances in Understanding Cortical Function through

Combined Voltage-Sensitive Dye Imaging, Whole-Cell
Recordings, and Analysis of Cellular Morphology . . . . . . . . . . 436
Carl C. H. Petersen
15. From Dendrites to Networks: Optically Probing the Living Brain

Slice and Using Principal Component Analysis to Characterize
Neuronal Morphology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452
Jesse H. Goldberg, Farid-Hamzei-Sichani, Jason MacLean,
Gabor Tamas, Rochelle Urban, and Rafael Yuste
16. Stereology of Neural Connections: An Overview . . . . . . . . . . . 477

Carlos Avendaño
17. Three-Dimensional Computerized Reconstruction from Serial

Sections: Cell Populations, Regions, and Whole Brain . . . . . . . 530
Jan G. Bjaalie and Trygve B. Leergaard
CONTENTS xiii
18. Atlases of the Human Brain: Tools for Functional
Neuroimaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566
Katrin Amunts and Karl Zilles
19. Neuron and Network Modeling . . . . . . . . . . . . . . . . . . . . . . 604

Giorgio A. Ascoli and Ruggero Scorcioni
20. Functional Connectivity of the Brain: Reconstruction from Static

and Dynamic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 631
Zoltan Nadasdy, Gyorgy Buzsaki, and Laszlo Zaborszky
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682
1
Short Retrospection
LENNART HEIMER
When Laszlo Zaborszky and I dedicated the 1989 tract-tracing book to Alf
Brodal, Walle J. H. Nauta, and János Szentágothai, we did it in part to em-
phasize the maxim that tract-tracing is a crucial part of the overall effort to
understand the functions of the central nervous system. We also wanted to
highlight the fact—repeatedly illustrated in the history of neuroscience—
that breakthroughs in neurobiology are often the result of histotechnical
advances. The scientists saluted above began their careers in neuroanatomy
with efforts to improve techniques for tracing neuronal connections. Al-
though they were standing on the shoulders of giants who had grabbled
with histotechnical problems of tract-tracing long before they arrived on
the scene, I would like to begin this short retrospection in the middle of the
twentieth century with reference to the book New Research Techniques of Neu-
roanatomy, edited by William Windle and published by Charles C. Thomas in
1957. Here, Nauta presented his silver staining method by summarizing his
continuing efforts to improve the silver technique for the tracing of degen-
erating axons. What was often referred to as the “suppressive” Nauta–Gygax
method (Nauta and Gygax, 1954) was by far the most popular tract-tracing
method at the time. The word “suppressive” underscored its ability to sup-
press the staining of normal fibers while still revealing the main parts of
degenerating fibers, an attractive feature which greatly facilitated the trac-
ing of axons undergoing degeneration as a result of an experimental lesion.
When the next tract-tracing book: Contemporary Research Methods in Neu-
roanatomy, edited by Nauta and Ebbesson and published by Springer Verlag,
appeared more than a decade later in 1970, the silver methods for the stain-
ing of degenerating axons were still the methods of choice for most scientists
interested in tracing neuronal connections. In fact, one third of the chapters
in the Nauta–Ebbesson book dealt with the Nauta–Gygax method and its
modifications, whereas another third described various modifications of the
Golgi technique, which at that time experienced a revival as a tract-tracing
method. Many of the major pathways of the brain and spinal cord were
convincingly described for the first time by the aid of experimental silver
methods during this period. The secret of Nauta’s success in tract-tracing is
LENNART HEIMER • Departments of Neurosurgery and Neuroscience, University of

Virginia Health Science Center, P.O. Box 800212, Charlottesville, VA 22908
1
2 LENNART HEIMER
most easily explained by reference to his unfailing persistence and ability to

spend a large part of his career to improving and promoting the art of silver
staining. To those unfamiliar with the tinker-and-toil process of developing
new techniques, Nauta’s never-ending quest for histotechnical perfection
might understandably have evoked the image of an obsessive mind. Nauta
vividly described the ups and downs of developing tract-tracing methods in
his “farewell” article in the Journal of Neuroscience (Nauta, 1993).
An important reason for the success of the silver method was the fact that
its most sensitive modifications, which became available at the end of the
1960s, made it possible to trace pathways throughout the central nervous
system and to visualize the entire axonal projections including their termi-
nations at the light microscopic level. The areas of terminations could be
conveniently confirmed and investigated with the aid of the electron micro-
scope, provided the sampling of the tissue was guided by a silver method
sensitive enough to stain the terminal fields. With some justification, there-
fore, the silver methods can be called the first modern methods for the
tracing of pathways in the central nervous system. This is in large part Walle
Nauta’s legacy.
When I was introduced to the silver methods around 1960, it seemed rea-
sonable to suggest that we needed a new anatomical technique every 10 or
15 years in order to keep good progress. Little did I know what was about
to happen. If the 1960s are remembered for the introduction of the Falck–
Hillarp fluorescence method for the tracing and mapping of monoamine
neurons in the central nervous system (Falck et al., 1962)—in itself a mile-
stone on the road toward establishing the field of chemical neuroanatomy—
the 1970s were a time of unprecedented renewal in the field of tract-tracing
with anterograde and retrograde axonal transport of tracers replacing the
silver techniques as the methods of choice in tract-tracing (Cowan and
Cuenod, 1975). The novel axonal transport methods could be effectively
exploited on the backdrop of an already established basic anatomical frame-
work, thanks in large part to earlier tract-tracing studies aided by silver meth-
ods. Although the silver methods lost their dominant role as tract-tracing
methods during the last quarter of the twentieth century, they experience a
dramatic revival as a tool for neurotoxic assessment (Beltramino et al., 1993).
Then followed in the 1980s in rapid succession a host of new methods
for the anatomical and chemical mapping of pathways and neuronal mi-
crocircuits, including immunohistochemical methods, in situ hybridization
histochemistry, transneuronal tracing using live viruses, and the identifica-
tion of successive links in chains of neurons using a combination of different
techniques at both the light and electron microscopic levels. Computational
neuroanatomy is gaining momentum, and as reflected in several chapters
in this volume, sophisticated combinations of anatomical and physiologi-
cal techniques have become routine. The application of voltage-sensitive
dye imaging has opened up the possibility to study the function and archi-
tecture of neuronal networks in living animals, and the genetic expression
of recorded and traced neurons can be studied. We may now attempt to
SHORT RETROSPECTION 3
elucidate Vogt’s classic theory of Pathoklise (Vogt, 1925), or “selective vul-
nerability,” with molecular biological techniques. Many of the recent pi-
oneering discoveries in neuroscience are in large part the result of this
unprecedented development of neuroscience methods, and for the neu-
roscientists who came of age in the last quarter of the twentieth century,
no problem of neuronal connectivity seems too difficult. For those willing
to venture beyond the mere routine application of standard methods, the
possibilities are almost unlimited, as they are determined primarily by the
scientist’s own imagination and technical skills.
The successes in experimental tract-tracing during the second half of the
last century did not translate into similar gains in the study of connections
in the human brain, although the phenomenal achievements in the field
of chemical neuroanatomy have to some extent compensated for this de-
ficiency in tract-tracing per se. Methods based on postmortem diffusion of
different chemical substances have been tried (Haber, 1988; Sparks et al.,
2000), but the slow rate of diffusion has limited the usefulness of these meth-
ods in the large human brain. The use of postmortem silver techniques
or myelin stains in patients with focal lesions (Grafe and Leonard, 1981;
Mesulam, 1979; Miklossy and Van der Loos, 1991) is another approach, but
the inability to control the location and size of human brain lesions has
so far restricted the value of these methods. Nonetheless, it appears that
postmortem silver staining of human brains with focal lesions or degen-
erative disorders is underutilized, maybe in part because the art of silver
staining is gradually disappearing as the younger generation of experimen-
tal neuroanatomists have embraced the axonal transport methods. But the
situation is not without its silver lining. The expertise of silver staining has
been uniquely preserved in some laboratories, especially by Robert Switzer
(Switzer, 2000), whose activities and writings suggest that it would be worth-
while to more actively try to introduce modern silver methods on a routine
basis in neuropathology laboratories.
In vivo tractography through the method of diffusion tensor imaging (DTI;
Basser et al., 2000), based on the tendency of water molecules to diffuse in
the direction of myelinated fiber bundles, has become the new buzzword
in clinical neuroscience. Results obtained by this method in the living hu-
man brain so far are spectacular (Behrens et al., 2003; Catani et al., 2002).
In vivo tractography will in all likelihood have a great impact in clinical
neuroscience. In spite of the great expectations surrounding in vivo fiber
tractography, however, it is unrealistic to expect that new pathways, defined
by origin and termination, will be discovered by DTI alone. Only white mat-
ter bundles can be visualized, and there is currently no possibility to recon-
struct intracortical connections. The axons of many myelinated pathways,
furthermore, often intermingle with axons belonging to other pathways,
which makes it difficult to disentangle individual myelinated pathways, even
if efforts to do just that have had some success (Catani et al., 2002).
Whatever limitations DTI may have in discovering new pathways, the tech-
nique has served as a powerful incentive to study fiber tracts in the human
4 LENNART HEIMER
brain; neurosurgeons and neuroradiologists, in particular, are promoting

blunt dissection of the white substance as part of their education and daily
activities (Kier et al., 2004; Türe et al., 2000; Yaşargil, 2004). However, dis-
section of fiber tracts, even if aided by the operation microscope, has some
obvious drawbacks as soon as the fiber bundles in question are intermin-
gling with other fiber tracts, and this, needless to say, is the rule rather than
the exception in the CNS.
DTI has caught the imagination of basic and clinical neuroscientists alike.
It is already yielding dividends in clinical neuroscience, and it has reinforced
the value of brain dissection as an integral part of studying the major path-
ways in the human brain. It may also have raised the awareness among clini-
cal neuroscientists that a naturally occurring lesion may sometimes provide
a golden opportunity to trace connections in the human brain (see ear-
lier discussion of silver methods). Maybe one day an even more powerful
method, or a combination of methods, will fulfill Mesulam’s dream of a
new frontier in the field of tract-tracing in the human brain (Mesulam,
2005).
Let us return to experimental tract-tracing methods in the broadest sense
of the term. This includes studying all aspects of neuronal circuitry using
anatomical, physiological, histochemical, molecular biological, and neu-
roimaging methods. The histotechnical revolution, which gained momen-
tum in the last quarter of the twentieth century, is reflected by dramatic
progress in the basic and clinical neurosciences. This in turn, has raised the
hope that some day in the not too distant future, the most crippling brain
disorders might be understood and successfully treated. For those chosen to
carry the torch of neuroscience in the early part of the twenty-first century,
this book is a godsend considering the often difficult problems of choosing
and applying the right technique or combination of techniques. Great ambi-
tions were needed to produce a coherent volume on tract-tracing methods
in the broadest sense of the term. The editors, the authors, and the publisher
deserve our congratulations.
REFERENCES
Basser, P. J., Pajevic, S., Pierpaoli, C., Duda, J., and Aldroubi, A., 2000, In vivo fiber tractography
using DT-MRI data, Magn. Reson. Med. 44:625–632.
Behrens, T. E. J., Johansen-Berg, H., Woolrich, M. W., Smith, S. M., Wheeler-Kinshott, C. A.
M., Boulby, P. A., Barker, G. J., Sillery, S. L., Sheehan, K., Ciccarelly, O., Thompson, A. J.,
Brady, J. M., and Matthews, P. M., 2003, Non-invasive mapping of connections between
human thalamus and cortex using diffusion imaging, Nat. Neurosci. 6:750–757.
Beltramino, C. A., DeOlmos, J., Gallyas, F., Heimer, L., and Zaborszky, L., 1993, Silver impreg-
nation as a tool for neurotoxic assessment, In: Erinoff, L. (ed.), Assessing Neurotoxicity of
Drugs of Abuse. Rockville, Maryland: NIDA Mongraph series, Vol. 136, pp. 101–132.
Catani, M., Howard, R. J., Pajevic, S., and Jones, D. K., 2002, Virtual in vivo interactive dissection
of white matter fasciculi in the human brain, Neuroimage 17:77–94.
Cowan, W. M., and Cuenod, M., 1975, The Use of Axonal Transport for Studies of Neuronal Connec-
tivity, Amsterdam: Elsevier.
SHORT RETROSPECTION 5
Falck, B., Hillarp, N. -Å., Thieme, G., and Torp, A., 1962, Fluorescence of catecholamines and
related compounds condensed with formaldehyde, J. Histochem. Cytochem. 10:348–354.
Grafe, M. F., and Leonard, C. M., 1980, Successful silver impregnation of degenerating axons
after long survivals in the human brain, J. Neuropathol. Exp. Neurol. 39:555–574.
Haber, S., 1988, Tracing intrinsic fiber connections in postmortem human brain with WGA-
HRP, J. Neurosci. Methods 23:15–22.
Kier, E. L., Staib, L. H., Davis, L. M., and Bronen, R. A., 2004, MR imaging of the temporal
stem: anatomic dissection tractography of the uncinate fasciculus, inferior occipitofrontal
fasciculcus, and Meyer’s loop of the optic radiation, Am. J. Neuroradiol. 25:677–691.
Mesulam, M. M., 1979, Tracing neural connections of human brain with selective silver im-
pregnation. Observations on geniculocalcarine, spinothalamic, and entorhinal pathways,
Arch. Neurol. 36:814–818.
Mesulam, M. M., 2005, Imaging connectivity in the human cerebral cortex: the next frontier?
Ann. Neurol. 57:5–7.
Miklossy, J., and Van Der Loos, H., 1991, The long-distance effects of brain lesions: visualizing
of myelinated pathways in the human brain using polarizing and fluorescence microscopy,
J. Neuropathol. Exp. Neurol. 50:1–15.
Nauta, W. H. J., 1993, Some early travails of tracing axonal pathways in the brain, J. Neurosci.
13:1337–1345.
Nauta, W. H. J., and Gygax, P. A., 1954, Silver impregnation of degenerating axons in the central
nervous system: a modified technic, Stain Technol. 29:91–93.
Sparks, D. L., Lue, L. -F., Martin, T. A., and Rogers, J., 2000, Neural tract tracing using Di-I:
a review and a new method to make fast Di-I faster in human brain, J. Neurosci. Methods
103:3–10.
Switzer, R. C., III, 2000, Application of silver degeneration stains for neurotoxicity testing,
Toxicol. Pathol. 28:70–83.
Türe, U., Yas,argil, M. G., Friedman, A. H., Al-Mefty, O., 2000, Fiber Dissection technique: lateral
aspect of the brain, Neurosurgery 47:417–427.
Vogt, O., 1925, Der Begriff der Pathoklise, J. Psychol. Neurol. (Lpz) 31:245–255.
Yaşargil, M. G., 2004, Impact of temporal lobe surgery, J. Neurosurg. 101:725–738.
2
Preembedding
Immunoelectron Microscopy:
Applications for Studies of
the Nervous System
SUSAN R. SESACK, LEEANN H. MINER, and
NATALIA OMELCHENKO
INTRODUCTION
APPLICATIONS
General Applications and Appraisal of the Methods
Ultrastructural Immunolocalization of Neurobiological Proteins
Analyses of Synaptic Connections
PRINCIPLES OF THE METHODS
Animals
Tract-Tracing
Phenotypic Labeling
Intracardial Perfusion
Immunocytochemistry
Tissue Preparation for TEM
Tissue Sampling
SUMMARY OF ADVANTAGES AND LIMITATIONS
Advantages
Limitations: Sources of False-Positive Errors
Limitations: Sources of False-Negative Errors
PROSPECTS FOR THE FUTURE
APPENDIX
Recipes for Standard Buffers
Fixation
Immunolabeling
Tissue Preparation for Electron Microscopy
REFERENCES
SUSAN R. SESACK, LEEANN H. MINER, AND NATALIA OMELCHENKO • De-

partment of Neuroscience, University of Pittsburgh, 446 Crawford Hall, Pittsburgh, PA 15260
6
PREEMBEDDING IMMUNOELECTRON MICROSCOPY 7
Abstract: This chapter addresses the basic applications of tract-tracing and preem-
bedding immunoperoxidase and immunogold–silver labeling for transmission elec-
tron microscopy, focusing primarily on identifying the cellular and subcellular lo-
calization of proteins of relevance to neurotransmission and on defining synaptic
connectivity within neuronal circuits. Information is provided regarding the use
of preembedding immunoperoxidase and immunogold techniques to identify the
cellular and subcellular localization of neuronal receptors and transporters. The
chapter also describes in detail a triple-labeling approach designed by our labora-
tory for identifying synaptic inputs to neuronal cell populations defined both by
their projection targets and by their transmitter phenotype. Protocols presented in
the Appendix are designed to enable researchers trained in small animal surgery,
immunocytochemistry, electron microscopy, and appropriate laboratory safety pro-
cedures to perform ultrastructural investigations similar to those described here.
Keywords: electron microscopy, immunocytochemistry, immunogold, immunoper-
oxidase, preembedding, tract-tracing, ultrastructure
I. INTRODUCTION
Having been available for 70 years, it seems reasonable to ask what trans-
mission electron microscopy (TEM) can add to the investigation of the ner-
vous system in the new millennium. Today, the basic synaptic organization of
most brain regions is known, as is the morphological detail of most neurons
and support cells. Chapters in previous volumes of this series have addressed
the applications of TEM for studies of brain/neuronal structure, patterns of
degeneration, and identification of synaptic connectivity using tract-tracing,
immunocytochemistry, and electrophysiological cell filling (Heimer and
Robards, 1981; Heimer and Zaborszky, 1989). In the chapter by Wouterlood
of the current volume, a sophisticated technique for using confocal mi-
croscopy to define synaptic contacts is described, raising further questions
about how long TEM will endure as a staple approach for neuroscience in-
vestigation. Is it the case that TEM will increasingly become a legacy method
used only by aesthete scholars eager to have quality photomicrographic ev-
idence at the highest magnifications possible?
Obviously, we believe that this is not the case and that TEM, in combina-
tion with procedures to identify discrete pathways and proteins at cellular
and subcellular levels, is as powerful and up to date as ever. Of course, TEM is
not without its disadvantages; it is expensive, time-consuming, requires con-
siderable technical skill, and is capable of generating false-positive and false-
negative results, the latter even in the hands of experienced researchers. In
the current chapter, we will present a general assessment of the progress
made in TEM studies of the nervous system since the last volume of this
series and provide technical and interpretational guidelines that are crit-
ical for planning modern TEM experiments. We will briefly characterize
relatively common procedures and refer readers to other chapters where
these are described in greater detail. In addition, we will supply sometimes
8 SUSAN R. SESACK, LEEANN H. MINER, and NATALIA OMELCHENKO
hard-to-find information that has not yet been described in this series or
needs to be updated. This chapter focuses on preembedding immunocy-
tochemical methods for TEM. The reader is referred to the chapter by
Mathisen et al. (this volume) for a description of postembedding techniques
that are also used for immunoelectron microscopy.
II. APPLICATIONS
A. General Applications and Appraisal of the Methods
The higher magnification and resolution afforded by the TEM method of

analysis presents distinct advantages compared to light microscopic (LM)
techniques. TEM allows the observation of cell ultrastructure and how this
morphology changes with natural or extrinsically imposed events such as ag-
ing, drug treatment, or stress. For such questions, the application of LM and
Golgi impregnation is also quite useful, particularly for measuring dendritic
branching and for counting spine density on identified cell types (Irwin
et al., 2002; Kolb et al., 2003). However, TEM in combination with unbiased
stereological measurements can more accurately and more quickly provide
counts of overall spine number in a given brain region, as well as the num-
ber of axon terminals, dendrites, and synapses in that area. As such, TEM
application of the physical disector method has become an important tool
for analyzing how structure and synaptology change with experience. This
methodology can also be combined with immunocytochemistry (Beaulieu
et al., 1994). Such applications will not be addressed here, and the reader is
directed to the existing literature on physical disector methods for counting
neuronal elements (Howard, 1990; Mayhew, 1992; von Bartheld, 2002) and
to the chapter by Avendaño in this volume.
With relatively few exceptions, most antigens can be immunolabeled by
both LM and TEM, although the latter allows for a more precise assessment
of the subcellular position of antigens, and the relationship of labeled struc-
tures to each other can be determined more precisely with TEM techniques.
This makes the latter method essential for the most definitive determination
of the cellular and subcellular distribution of neurotransmitters, receptors,
transporters, enzymes, metabolites, and so on (see section “Ultrastructural
Immunolocalization of Neurobiological Proteins”). TEM is also suitable for
more recent studies of protein trafficking (Bloch et al., 2003; Ravary et al.,
2001), although LM approaches have the advantage of being able to follow
these intracellular processes in real or near-real time.
TEM allows the visualization of the fine characteristics of synapses occur-
ring between neuronal structures identified by tract-tracing or immunocy-
tochemistry. Hence, it continues to find utility as a tool for studying commu-
nication between functional neural systems. For this particular application,
TEM approaches have gained in analytical power since the last release of
neuroanatomical tract-tracing methods, due to the ability to recognize at
least three markers on the same sections. Although LM techniques such as
that described in the chapter by Wouterlood (this volume) may eventually
provide comparable information, TEM will continue to represent the prin-
cipal method for detailed synaptic connectivity studies of neuronal circuitry.
Below, we address in more detail the important technical issues that deal
with the studies of this type.
B. Ultrastructural Immunolocalization of Neurobiological Proteins
The functional impact of any given neurotransmitter is determined by

the activation of receptor subtypes with particular affinities and activities
and often by the removal of the transmitter from the extracellular space
by plasma membrane transporters. Although these general principles of
neurotransmission apply to most neuronal systems, differences in the ex-
tent to which receptors and transporters are inserted in the plasma mem-
brane can cause distinct cellular responses to neurochemicals that are highly
regulated. Hence, a powerful application of ultrastructural immunocyto-
chemistry is the localization of neurobiologically relevant proteins (Cornea-
Hebert et al., 1999; Ferguson et al., 2003; Garzón et al., 1999; Glass et al., 2004;
Hanson and Smith, 1999; Huang and Pickel, 2002; Pickel and Chan, 1999;
Pickel et al., 2004; Miner et al., 2000, 2003a, 2003c; Riad et al., 2000). The
information obtained from ultrastructural descriptions of their distribution
is crucial to our understanding of the operation of these systems.
1. General Issues of Pre- and Postembedding

Immunocytochemical Methods
The main techniques used for immunologically based electron-dense

staining of these proteins are preembedding immunoperoxidase, preem-
bedding immunogold–silver, and postembedding immunogold. These are
“indirect” methods of labeling that involve a “primary” unconjugated anti-
body directed against the antigen of interest. The primary antibody is an
unmodified protein that cannot be visualized by TEM. Therefore, the pres-
ence of the primary antibody is detected by using a secondary antibody that
is directed against species-specific antigenic sites on the primary antibody.
This secondary antibody is conjugated to a molecule that will allow the cre-
ation of an electron-dense product that can be readily visualized. Some of
the labeling molecules are enzymes whose reaction product can be detected
by the addition of a substrate, whereas others are electron-dense heavy met-
als that can be visualized directly. Because the end reaction products from
the various immunocytochemical techniques possess distinguishing charac-
teristics, a combination of the labeling methods can be used to identify mul-
tiple proteins and determine their cellular or subcellular relationships to
each other. For instance, preembedding immunoperoxidase labeling of one
antigen can be followed by preembedding immunogold–silver (Fig. 2.1A)
or postembedding immunogold labeling of another antigen. Also, both
B C
Figure 2.1. Electron micrographs showing axon terminals in the rat PFC singly la-
beled by immunogold–silver for NET (NET-t) or dually labeled for NET and by
immunoperoxidase for TH (NET + TH-t). Axons dually labeled for NET and TH
contain the majority of gold–silver particles along the plasma membrane (arrow-
heads in A). Axons singly labeled for NET contain immunogold–silver particles
mainly within the cytoplasm and occasionally form symmetric synapses (small ar-
rows) onto unlabeled dendrites (ud in B) or spines (us in C). In (C), the spine also
receives an asymmetric synapse (large arrow) from an unlabeled terminal (ut). Scale
bar represents 0.5 µm.
preembedding (Yi et al., 2001) and postembedding (see chapter by Mathisen

et al., this volume) immunogold procedures can make use of different-sized
gold or gold–silver particles to perform dual-labeling studies.
For the preembedding techniques, the immunocytochemical labeling is
performed on thick sections (50 µm), prior to embedding in the plastic resin
necessary for cutting the ultrathin (60 nm) sections used for TEM. These
preembedding techniques allow satisfactory visualization of most proteins
and are becoming somewhat standardized, as described in detail in the Ap-
pendix. However, proteins situated in areas that may be inaccessible to anti-
bodies are not consistently recognized with these methods. Such locations
include dense membrane barriers like the synaptic complex (e.g., many
receptors and transporters) or within vesicles (e.g., the synthetic enzyme
dopamine β-hydroxylase).
For the postembedding procedure, immunocytochemistry is performed
after the thick sections are embedded in plastic and sliced into ultrathin
sections. Due to the thinness of the sections, the experimenter can be
confident that antibodies contact most portions of any given cell. There-
fore, the major advantage of the postembedding technique is that it allows
the visualization of proteins within areas that are typically not accessible
with the preembedding procedures. For instance, receptors located within
pre- and postsynaptic densities can be visualized (Hanson and Smith, 1999;
Nusser et al., 1995). However, many antigens become denatured during
the embedding process, rendering this technique unfeasible for the local-
ization of many proteins. Another potential disadvantage is that postem-
bedding gold reagents may interact nonspecifically with the surface of ul-
trathin sections, resulting in background staining and interfering with un-
equivocal identification of specifically labeled structures (Van Haeften and
Wouterlood, 2000). For a complete review of the applications and technical
considerations regarding postembedding methods, the reader is referred
to the chapter by Mathisen et al. (in this volume).
2. Preembedding Immunoperoxidase Labeling
The preembedding immunoperoxidase technique involves localizing

horseradish peroxidase enzyme activity. The horseradish peroxidase
molecule is easily conjugated to antibodies, and because it is relatively small,
it can readily penetrate membranes of cells fixed with aldehydes. When the
peroxidase molecule is exposed to a hydrogen peroxide substrate and the
capturing agent 3,3 -diaminobenzidine (DAB), the oxidation products of
the DAB will form a brown reaction product. This insoluble product be-
comes electron dense following chelation by osmium tetroxide (Fig. 2.1A).
a. Strengths and Weaknesses
The major benefit of the immunoperoxidase technique is its superior

level of sensitivity. This is particularly true given the various techniques for
signal amplification with this method (Hsu et al., 1981; Ordronneau et al.,
1981). Compared to the preembedding immunogold approach (described
below), immunoperoxidase maximizes the labeling of antigen (Chan et al.,
1990), which presents a considerable advantage if the amount of the protein
of interest in a given structure is low. Hence, if the desired end point is
to demonstrate that a particular neuronal element contains a particular
protein, this is the method of choice.
The main disadvantage of immunoperoxidase labeling is that the catalytic
activity of the enzyme can result in the accumulation of reaction product
that is capable of diffusion away from the original site of activity and, thus,
the location of the antigen (Novikoff et al., 1972). In this case, immunoper-
oxidase is not the best choice if the goal of the study is to determine the
exact subcellular location of a protein. Another potential disadvantage of
immunoperoxidase is that the accumulation of excessive amounts of reac-
tion product can obscure the visualization of cytoplasmic organelles and
synaptic specializations, which limits the determination of cellular structure
or the presence of synaptic contacts, and challenges the identification of
synapse type. In some cases, the crystalline reaction product can become
sufficiently excessive to break through membranes and leak into surround-
ing structures. Consequently, one must take great care when considering
structures lying immediately adjacent to heavily labeled profiles to ensure
that all membranes that separate the profiles are intact.
Another potential drawback of immunoperoxidase staining is that some

cellular elements contain proteins with endogenous peroxidase-like activ-
ity. Such activity is present in red blood cells. As a result, suboptimal per-
fusions that leave behind many blood cells within brain capillaries can be
problematic, in that they have the capacity to use up reagents intended
for immunoperoxidase labeling with subsequent reduction in the density of
specific labeling. In addition, some glial cells and even some neurons appear
to have endogenous peroxidase-like activity (Conradi, 1981; Srebro, 1972;
Svensson et al., 1984). Therefore, one may wish to incubate tissue sections in
hydrogen peroxide (see Appendix, section “Sodium Borohydride and Hy-
drogen Peroxide Treatments”) prior to immunocytochemical procedures
in order to quench this endogenous peroxidase activity.
b. Applications and Analysis
The immunoperoxidase technique is well suited for qualitative analysis

of the distribution of proteins. In other words, the presence or absence of a
protein can be readily determined using this labeling procedure (Fig. 2.1A).
Moreover, if the aim of an experiment is to establish the general localization
of a protein (e.g., expression in soma, dendrites, or axons), this procedure
is optimal, given its sensitivity. Quantitative analysis of immunoperoxidase-
labeled tissue is typically limited to counting the number of labeled struc-
tures within a given area or volume. Because the peroxidase reaction product
is capable of spreading over some distance, determination of the subcellular
localization of proteins is imprecise with this technique. Therefore, if knowl-
edge of exact protein location is desired, this procedure is not the most viable
choice. Likewise, the density of antigens within labeled structures cannot be
discerned with precision, although estimates of light or heavy localization
are possible. It should be noted that this method is valuable as a first step
in localizing proteins. Because immunoperoxidase labeling is the most sen-
sitive immunocytochemical technique and the least costly in terms of time
and money, it is practical to use this technique to gain knowledge about the
overall amount of an antigen and its general distribution in order to guide
further labeling with less sensitive techniques. Some proteins, particularly
those in low abundance, often do show “hot spots” of immunoreactivity that
strongly suggest a certain subcellular pattern of localization (Aoki et al., 2001;
Doly et al., 2004; Mi et al., 2000). Hence, this may be the best first approxima-
tion that is available in the event that the less sensitive immunogold–silver
methods do not produce consistently detectable labeling.
3. Preembedding Immunogold–Silver Labeling
Immunogold labeling exploits the conjugation of colloidal gold particles

to immunoglobulins (DeMey and Moeremans, 1986). Therefore, the tissue
is exposed to the primary antibody and then to a secondary antibody that
has been conjugated to a gold particle. Colloidal gold probes may be pre-
pared in a variety of sizes (from ∼1 to 150 nm). For electron microscopy,
gold probes below 30 nm are typically used. For example, it has been shown
that gold particles in the range of 10 nm can be used without silver enhance-
ment for preembedding subcellular immunolocalization (Zaborszky et al.,
2004). However, the use of ultrasmall gold particles (∼1 nm) conjugated
to secondary antibodies has the advantage of greater tissue penetration for
preembedding studies (Chan et al., 1990; Pickel et al., 1993). Because parti-
cle of this size are not readily visualized by electron microscopy, the gold is
then enlarged by allowing silver to complex with it. It should be noted that,
because antibody penetration is not an issue for postembedding immuno-
labeling, the gold particles conjugated to the secondary antibodies used for
that method are much larger (10–20 nm), and can be easily distinguished
by electron microscopy without further enhancement procedures.
a. Strengths and Weaknesses
In contrast to the precipitates formed in immunoperoxidase staining,

silver-enhanced gold particles are more electron dense, more discrete, and
spherical (Figs. 2.1–2.3). Hence, they are quite easily identified. Their dis-
crete nature and the fact that the gold particle remains attached to the sec-
ondary antibody means that diffusion away from the source of the antigen is
not a problem as it can be for immunoperoxidase. Thus, the predominant
strength of this technique is that it permits a more precise view of anti-
gen localization (Figs. 2.1–2.3). For instance, this method will distinguish
whether a protein is inserted within the plasma membrane or within the
cytoplasm (see especially Fig. 2.1) (Ferguson et al., 2003; Garzón et al., 1999;
Miner et al., 2000, 2003c; Pickel and Chan, 1999). It is possible that the silver-
enhanced gold particles will settle at areas that are slightly removed from
the antigen. However, the distance in question is typically in the range of
20–25 nm (Paspalas and Goldman-Rakic, 2004) (e.g., along the inner or the
outer plasma membrane). If a more refined level of protein localization is
required, postembedding immunogold labeling may be necessary (see the
chapter by Mathisen et al.). Having discrete particles also allows for a certain
degree of quantification with the immunogold–silver method. Particles can
be counted, measured as number per unit area or per unit volume, and
assigned to different compartments such as cytoplasmic or plasmalemmal.
Although the preembedding immunogold–silver technique is a powerful
method for localizing proteins, it has limitations. One weakness is that the
level of labeling achieved for a given amount of antigen is generally lower
than with immunoperoxidase. It has been estimated that the immunogold–
silver approach is perhaps one order of magnitude less sensitive than im-
munoperoxidase (Chan et al., 1990). This is likely due to decreased pen-
etration into the tissue section of the secondary antibody with the gold
particle conjugated to it and the absence of methods for signal amplification.
B C
Figure 2.2. Electron micrographs of the rat PFC showing immunogold–silver labeling
for SERT in axon terminals (SERT-t) forming asymmetric synapses (large arrows)
onto unlabeled dendrites (ud) or spines (us). The majority of gold–silver particles
for SERT are located along the plasma membrane (arrowheads), although some are
also distributed within the cytoplasm. Occasionally, gold–silver particles for SERT
are found in close proximity to sites of synaptic contact (white arrow). Scale bar
represents 0.5 µm.
Another possible weakness is that preembedding immunogold may be bet-

ter suited to detecting plasmalemmal antigens when they are located at
extrasynaptic than at synaptic sites (Figs. 2.2 and 2.3) (Bernard et al., 1997).
Some studies indicate that preembedding immunogold is unable to detect
A B
Figure 2.3. Electron micrographs from the rat PFC illustrating immunogold–silver
labeling for 5HT2A receptors (5HT2A R) in dendritic shafts (5HT2A R-d) and spines
(5HT2A R-s) receiving synaptic input (arrows) from unlabeled terminals (ut). Arrow-
heads indicate immunogold–silver particles associated with the plasma membrane.
Those with asterisks are within or near the postsynaptic density. Scale bar represents
0.5 µm.
proteins that are embedded with in the synaptic complex, including the
synaptic cleft (Baude et al., 1995; Bernard et al., 1997). However, other pub-
lished studies show acceptable labeling of synaptic proteins (Chen et al.,
2004; Kulik et al., 2003; Wong et al., 2002) (see also Fig. 2.3). Conversely, the
postembedding immunogold method is considered the preferred method
for visualizing proteins within synaptic specializations (Bernard et al., 1997).
Most likely, this is due to the fact that the synaptic complex has been sec-
tioned at 60 nm, exposing antigenic sites within the complex. Another po-
tential weakness of the preembedding immunogold–silver technique is that
the silver enhancement of the gold particles is sensitive to several experimen-
tal factors, and therefore care must be taken to obtain reliable gold–silver
labeling (see section “Choice of Markers”).
b. Applications and Analysis
The ultrastructural immunolocalization of neurobiological proteins is a

powerful method for identifying the subcellular distribution of receptors
and transporters that may or may not be situated at predicted locations
(Ferguson et al., 2003; Hanson and Smith, 1999; Masson et al., 1999). Re-
gardless of whether such studies match or deviate from expectations, they
elucidate the anatomical substrates for the actions of neurotransmitters.
For example, we have utilized the preembedding immunogold–
silver method to investigate the subcellular localization of monoamine
transporters in the rat cortex (Miner et al., 2000, 2003c). In the case of
the serotonin transporter (SERT), immunogold–silver particles are typically
found on the plasma membrane in extrasynaptic locations (Fig. 2.2). Occa-
sionally, particles are localized close to the synaptic specialization (Fig. 2.2C).
Surprisingly, the norepinephrine transporter (NET) is localized predomi-
nantly in the cytoplasm (∼75%), with only a minority of gold–silver particles
found along the plasma membrane (Fig. 2.1). We have also performed a pre-
liminary investigation (unpublished data) of the subcellular localization of
the serotonin (5-hydroxytryptamine, 5HT) 2A receptor subtype (5HT2A R)
in cortical cells utilizing an antibody that was used previously to localize this
receptor by immunoperoxidase (Miner et al., 2003a). The preembedding
gold–silver method shows that this receptor is localized to both extrasynaptic
and synaptic plasmalemmal sites (Fig. 2.3).
In addition to characterizing the typical distribution of neurobiologically
relevant proteins, these techniques allow the provocative documentation of
changes following physiological or pathological treatments. For instance,
alterations in the subcellular distribution of various proteins have been ob-
served in response to genetic, pharmacological, or environmental manip-
ulations (Dumartin et al., 1998, 2000; Glass et al., 2004; Miner et al., 2003b,
2004; Nusser et al., 1999; Riad et al., 2004). The findings from such studies
provide valuable insights into the function and trafficking of these impor-
tant proteins.
For analysis of tissue in which the subcellular localization of gold–silver

particles, and hence antigen, is the dependent measure, it is customary to
count the total number of gold particles in the structure of interest, and then
to assign each gold particle to a particular compartment (e.g., cytoplasm,
smooth endoplasmic reticulum, vesicle, plasma membrane). One can then
assess as a percentage the tendency of gold particles to accumulate within a
particular compartment. It is also possible to determine whether gold par-
ticles are localized to the inside or outside edge of a membrane, although
as mentioned earlier, the exact position of the relatively large preembed-
ding immunogold–silver particles may not reflect the antigen distribution
with perfect fidelity. More refined localization requires postembedding im-
munogold methods.
4. Dual-Labeling Procedures
A major strength of the immunocytochemical procedures discussed here

is that several procedures may be combined in order to visualize two anti-
gens in the same tissue section. The type of dual-labeling method used
most frequently in our laboratory is preembedding immunoperoxidase
and immunogold–silver. The immunoperoxidase reaction product is dense,
dark, and flocculent, whereas the silver-enhanced gold particles are fully
black and particulate. Therefore, the two types of markers can be readily
distinguished, even in the same structure. In our experience, we prefer to use
a “parallel” labeling method in which the tissue is incubated simultaneously
in the primary antibodies for both antigens. Then, the immunoperoxidase
labeling is performed, followed by the immunogold–silver procedures. The
order of manipulations can also be switched so that immunogold–silver is
performed first (Katona et al., 2001). However, in this case, an additional
step of gold toning may be needed to stabilize the immunogold–silver prior
to peroxidase treatments (Arai et al., 1992). Either dual-labeling combina-
tion necessitates that the primary antibodies be raised in different species,
but it allows the experimenter to determine the presence of one antigen and
the precise subcellular localization of the other. For example, in our studies
of the localization of NET in the cortex, we found that axons containing
immunoperoxidase labeling for the catecholamine synthetic enzyme, tyro-
sine hydroxylase (TH), contain preembedding immunogold–silver labeling
for NET primarily along the plasma membrane (Fig. 2.1A) (Miner et al.,
2003c). This is in distinct contrast to axons that contain NET and not TH
(Fig. 2.1B,C), where NET is predominantly cytoplasmic.
Other TEM methods for labeling two antigens in the same tissue sections
include combinations of immunoperoxidase chromogens with different
physical characteristics (Charara et al., 1996; Norgren and Lehman, 1989;
Smith et al., 1994; Zaborszky and Heimer, 1989; Zhou and Grofova, 1995)
(see section “Choice of Markers”) and combinations of preembedding
immunogold–silver techniques (Yi et al., 2001). For the latter procedure,
silver intensification of one antigen is performed, followed by silver inten-
sification of a second antigen. The labeling for each can be distinguished
based on the size of the gold–silver particles, with the particles for the first
antigen being larger, because they are exposed to two enhancement sessions.
This procedure allows the subcellular localization of two antigens within the
same tissue.
Finally, two antigens can be visualized simultaneously by combining pre-
embedding immunoperoxidase or immunogold–silver with postembedding
immunogold procedures. An advantage of this technique is that two primary
antibodies that were generated within the same species can be used, because
the osmication and dehydration performed prior to plastic embedding de-
stroys the antigenicity of the first primary antibody. However, it should be
noted that this is feasible only if those same steps do not denature the antigen
to be labeled by the postembedding method (see above).
C. Analyses of Synaptic Connections
In order to determine whether neurons of interest are synaptically con-

nected, it is necessary to distinguish particular populations of cells based on
their specific protein content and/or on their topographic projections. Al-
though neurochemical phenotypes can typically be defined in naive animals,
the identification of neuronal connections requires surgical intervention to
introduce tract-tracing agents, at least at present (see section “Prospects for
the Future”). Here, we discuss the principles of study design that combine
tract-tracing and neurochemical definitions with TEM analysis.
It is beyond the scope of this chapter to compare the effectiveness and
methodology for all the available anterograde and retrograde tracing ap-
proaches, many of which have been described in previous volumes of this
series (Gerfen et al., 1989; Pickel and Milner, 1989; Skirboll et al., 1989;
Steward, 1981; Warr et al., 1981; Zaborszky and Heimer, 1989). Relevant
information regarding different methods of intraparenchymal injections,
including iontophoretic and pressure injections have also been covered
earlier (Alheid et al., 1981). Here, we will focus on the tracers that are used
most commonly in ultrastructural studies, and particularly those that com-
bine well with each other and with techniques to label phenotypic markers.
The main requirements for such tracers are (1) sensitivity in labeling as
many of the neurons and processes that contribute to a pathway as possi-
ble, (2) specificity in the direction of transport, (3) recognizable filling of
the parts of neurons that will be examined by TEM, (4) compatibility with
ultrastructural preservation, and (5) compatibility with other markers. If
multiple tracers need to be combined, the choices may become more lim-
ited. In order to avoid the performance of multiple survival surgeries, we
use tracers with similar optimal survival times. The clear understanding of
individual tracer competence is a critical issue for planning TEM experi-
ments, and the inability of any particular tracer or combination of tracers to
fulfill all of the above requirements is the main reason why the results may
need to be verified by alternative tracer approaches.
1. Anterograde Tract-Tracing
a. Anterograde Degeneration
The induction of anterograde degeneration following brain lesion is the

oldest tracing method available, but it still has excellent utility for some in-
vestigations. The three main lesion approaches are physical disruption (e.g.,
suction), cytotoxic injection, and electrolytic lesion (Moore, 1981), with cy-
totoxic injections having the advantage of being directed only to cell bodies
and dendrites in the region while sparing fibers of passage. Physical disrup-
tion and electrolytic lesions provide a more certain disruption of all cellular
elements at the same time, and the electrolytic approach can be calibrated
to control the degree of current spread. After a survival time that must be
determined empirically for each pathway, the axon terminals of the lesioned
neurons will show signs of degeneration that can be readily distinguished
in the electron microscope (Mugnaini and Friedrich, 1981). In the best
case, the terminals will demonstrate the characteristics of electron-dense
degeneration (Fig. 2.4A): shrunken and distorted boundaries, darkened
cytoplasm, disrupted organelles like mitochondria and synaptic vesicles, and
frequent engulfment on nonsynaptic sides by glial processes (Mugnaini and
Friedrich, 1981; Sesack and Pickel, 1992). Despite this extensive disruption
of structure, axon terminals will retain their synaptic connections with tar-
get cells until they are completely removed by glia, and this provides the
degeneration method its utility in studies of synaptic connections.
Anterograde degeneration has two major advantages and several disad-
vantages of note. First, in our experience, it is the most effective method
of anterograde tracing for TEM in the rodent, assuming that disruption of
fibers of passage is not an issue for the pathway under study (Sesack and
Pickel, 1992). Depending on the size of the lesion, many of the axons of
interest will be affected. Moreover, the method does not require any chemi-
cal or immunolabeling, and affected terminals will be observed throughout
the thickness of ultrathin sections. This is in contrast to tracers that require
immunocytochemical detection, whose recognition will be limited to the
tissue plastic interface (see section “General Sampling Issues”). Second,
anterograde degeneration can be combined with immunocytochemical
methods, including sensitive immunoperoxidase approaches (e.g., avidin–
biotin peroxidase; ABC (Hsu et al., 1981) to label target neurons or conver-
gent structures so that the two methods together provide excellent likeli-
hood of finding hypothesized synaptic connections (i.e., low false-negative
rate). In this regard, it should be noted that the two methods can be fur-
ther combined with preembedding immunogold–silver for triple-labeling
studies.
B C
Figure 2.4. Electron micrographic examples of anterograde tract-tracing in the rat

forebrain. (A) Two degenerating axon terminals (dt) in the PFC following an elec-
trolytic lesion in the mediodorsal thalamus form asymmetric synapses (large arrows)
onto spines either labeled by immunogold–silver for calbindin (CB-s) or unlabeled
(us). The latter receives an additional symmetric synapse (small arrow) from an
unlabeled terminal (ut). (B, C) Axon terminals in the amygdala (B) or NAc (C)
containing immunogold–silver labeling for PHA-L (PHA-L-t in B) or BDA (BDA-t
in C) transported anterogradely from the PFC or paraventricular thalamus, respec-
tively, synapse onto unlabeled spines. In (C) immunoperoxidase labeling for TH is
evident within axons (TH-a) in the adjacent neuropil. Scale bar in (A) represents
0.5 µm for (A) and (B); scale bar in (C) represents 0.5 µm for (C).
There are several disadvantages to the anterograde degeneration method:
1. The correct survival time after lesion is critically important. At short

times, it will be hard to recognize terminals undergoing degeneration,
whereas at long times, synaptic contacts will be disrupted and the af-
fected terminals will be eliminated. Deviations of a half day can make
a difference, so that the optimal timing for a particular pathway must
be established empirically.
2. The method is not applicable for all connections or in all species.
Monoamine pathways are particularly difficult to label with this ap-
proach, and early studies that attempted to use the method produced
erroneous results. A detailed analysis of the difficulties associated with
anterograde degeneration for detection of monoamine axons is avail-
able (Zaborszky et al., 1979). In addition, not all pathways appear to
undergo a process of anterograde degeneration that can be easily de-
tected by TEM. For example, in a prior study of efferents from the
prefrontal cortex (PFC), we were able to label the projection to the
nucleus accumbens (NAc) using anterograde degeneration but not
the fibers known to innervate the ventral tegmental area (VTA), de-
spite examination of lengthy survival times (Sesack and Pickel, 1992).
With regard to other species, the rat is particularly amenable to the an-
terograde degeneration method because of its small size and the ease
of lesioning the majority of a pathway nearly simultaneously. In pri-
mates, the anterograde degeneration process can proceed at different
rates in different neurons, making the selection of an optimal survival
time nearly impossible.
3. False-positive results can be generated if the lesion disrupts fibers not
originating from but passing through the lesioned area en route to the
target of interest.
4. Anterograde degeneration cannot be used for phenotypic character-
ization of afferents. Such features as size, normal morphology, and
immunoreactivity for particular markers will be compromised.
b. Phaseolus vulgaris Leucoagglutinin (PHA-L)
PHA-L is a plant lectin that has many superior properties for anterograde
tract-tracing, including specificity for the anterograde direction and the
ability to fill the processes of neurons that take it up in a Golgi-like fash-
ion (Gerfen and Sawchenko, 1984; Wouterlood and Groenewegen, 1985;
Wouterlood and Jorritsma-Byham, 1993). A detailed description of the ap-
plication of PHA-L tract-tracing can be found in Gerfen et al. (1989), and
combination of this method with transmitter identification in postsynaptic
targets has been described in a previous chapter in this series (Zaborszky
and Heimer, 1989). Here, we will focus primarily on the advantages and
disadvantages of this tracer for TEM studies.
The requirement of PHA-L for iontophoretic injection to stimulate uptake
is both an advantage and a disadvantage of this method. Iontophoresis is
advantageous because it contributes to the selective uptake into soma and
dendrites and, hence, the anterograde specificity. It also helps to create the
smaller injection sites that are characteristic of this tracer, which may be
desirable or undesirable, depending on the target of interest. Moreover,
the equipment for iontophoretic application is somewhat expensive, and so
this can limit the availability of the PHA-L method.
Regarding other advantages, PHA-L rarely undergoes retrograde trans-
port and only in certain systems (Shu and Peterson, 1988), making it an
otherwise exclusive anterograde tracer for most pathways. Another impor-
tant advantage is that it is not taken up by fibers of passage. To date, there
are no reports of PHA-L failing to label any particular pathway. Further-
more, PHA-L remains within the neurons that take it up for considerable
periods of time (up to 10 weeks) (Wouterlood et al., 1990), facilitating stud-
ies in which the introduction of the tracer and sacrifice of the animal must
be separated by some period of time. In the early stages after injection of
PHA-L, the extent of anterograde transport increases with time, so that for
TEM studies, it is generally advisable to wait 10–14 days following injection.
PHA-L is transported into all branches of an axon, making it possible to use
this tracer to study clearly identified local collaterals within the region of
injection. Although biotinylated dextran amine (BDA) has the same ability
(Melchitzky et al., 2001), its greater potential to undergo retrograde trans-
port and trafficking into the collateral branches of retrogradely labeled cells
(Wouterlood and Jorritsma-Byham, 1993) (see below) may compromise its
usefulness for studying local collaterals. Finally, the PHA-L method can be
combined with: BDA as a second anterograde tracer (Dolleman-Van der
Weel et al., 1994; French and Totterdell, 2002, 2003), retrograde tract-tracing
agents, or immunocytochemistry to identify target phenotype. In this regard,
PHA-L can be localized by either immunoperoxidase or immunogold–silver
methods (Fig. 2.4B), with the other marker being used to label additional
tracers or transmitters.
The PHA-L method has two principal disadvantages:
1. It is reported to have less sensitivity compared to BDA (Wouterlood and
Jorritsma-Byham, 1993), although we have not found this difference
to be particularly remarkable in our own studies of brain connectiv-
ity. Lower sensitivity is, in part, due to the relatively small injection
sites achieved with the iontophoretic technique. However, the major
reason for lower sensitivity is the requirement for immunolabeling to
visualize PHA-L. This lowers the sensitivity for TEM studies in which
ultrastructural preservation and penetration of the tissue surface must
be carefully balanced (see section “General Sampling Issues”).
2. Small injection sites limit the usefulness of the method for studies in
primates. Of course, PHA-L is the preferred method when injections
must be confined to a small target.
c. Biotinylated Dextran Amine (BDA)
BDA (10,000 MW) is another example of a highly successful method for

anterograde tract-tracing (Brandt and Apkarian, 1992; Reiner et al., 2000;
Veenman et al., 1992), as described in more detail in the chapter by Reiner
and Honig. The tracer can be injected as a 10% solution in 0.01 M phosphate
buffer, pH 7.4, either by iontophoresis or by pressure, enabling the investi-
gator to make large or small injections as desired. The ability to make large
injections is particularly important for tract-tracing in primates. BDA utilizes
a simple detection method, in which brain sections are simply incubated in
ABC reagents (Hsu et al., 1981) without the need for immunolabeling proce-
dures. This provides BDA with greater sensitivity than PHA-L (Wouterlood
and Jorritsma-Byham, 1993). However, when needed, BDA can be labeled by
immunogold–silver methods (Fig. 2.4C) using an antibody directed against
biotin (Pinto et al., 2003).
At the same time, BDA presents with some rather significant disadvan-
tages. Specifically, it can be transported retrogradely and via fibers of pas-
sage (Reiner et al., 2000; Van Haeften and Wouterlood, 2000). Moreover, the
BDA that undergoes retrograde transport can subsequently travel into the
collaterals of these neurons and give rise to false-positive results when used
for anterograde tracing (Chen and Aston-Jones, 1998; Reiner et al., 2000).
This requires that investigators check carefully for retrograde transport of
BDA into any brain area that is also afferent to the region of interest.
2. Retrograde Tract-Tracing
For LM studies, it is usually sufficient that retrograde tracers transport

into the soma and proximal dendrites of labeled cells. However, for TEM
studies of synaptic inputs to identified cell populations, there may be the
added requirement that a retrograde tract-tracing agent also traffic into the
distal dendrites of neurons, if the afferent under study synapses distally. This
poses one of the most difficult challenges in studies of synaptic connectivity.
To our knowledge, only a few tracers possess this ability, and then only in
certain cells or under special circumstances. Here, we review three tracers
that have been reported to infiltrate distal dendrites following retrograde
transport. A fourth retrograde tracer, cholera toxin subunit B, has also been
reported in the literature to show usefulness in ultrastructural analyses, and
interested readers are directed to several reviews on this subject (Bruce and
Grofova, 1992; Ericson and Blomqvist, 1988; Llewellyn-Smith et al., 1990).
In our hands, we have observed no obvious advantages with cholera toxin
over the tracers that follow.
a. BDA
A smaller molecular form of BDA (3000 MW) is available for use as a ret-
rograde tracer that can be delivered either by pressure or by iontophoresis
using a different buffer: 10% solution in 0.1 M sodium citrate–HCl (Reiner
et al., 2000). Like the use of BDA for anterograde tract-tracing, large injec-
tions of BDA can be made, making this technique useful for primate studies.
Retrogradely transported BDA can be detected with a simple ABC method,
and there is no evidence that it undergoes transneuronal transport (Brandt
and Apkarian, 1992; Rajakumar et al., 1993). However, the ability of BDA
(3000 MW) to also undergo anterograde transport limits the usefulness of
this tracer in multitracing experiments, and the tracer can also be taken up
by fibers of passage (Reiner et al., 2000; Veenman et al., 1992).
With regard to the need to visualize the distal dendrites of labeled cells,
a method has been introduced for improving axonal uptake of BDA by first
applying NMDA N-methyl-d-aspartate to lesion the soma and dendrites at
the injection site that might compete for uptake of BDA ( Jiang et al., 1993).
However, in our experience, this procedure does not work for all pathways.
For example, when we attempted to use BDA as a retrograde tracer in the
PFC to VTA pathway, the prior injection of NMDA into the VTA resulted
in BDA diffusion throughout a considerable portion of the midbrain with
minimal retrograde transport (unpublished data).
b. FluoroGold (FG)
FG has been utilized for some time as an effective retrograde tracer for LM
studies (Schmued and Fallon, 1986; Skirboll et al., 1989). More recently, FG
has also been adapted as a highly effective tracer for TEM (Chang et al., 1990;
Naumann et al., 2000; Van Bockstaele et al., 1994). FG can be delivered as a
2% solution in 0.1 M cacodylate buffer, pH 7.5, by iontophoretic or pressure
injections. FG is a highly effective and sensitive retrograde tracer that, to
our knowledge, has never failed to label any particular cell population.
Neither does this tracer diffuse out of retrogradely labeled cells into adjacent
neurons (Novikova et al., 1997; Schmued and Fallon, 1986; Van Bockstaele
et al., 1994). Moreover, FG appears to be an exclusively retrograde tracer, at
least in the systems for which we have used it to date. Following forebrain or
midbrain injections of FG, we have systematically examined sections in the
midbrain or pons by TEM and found no evidence for anterograde transport
of FG into axon terminals, despite the known existence of available pathways
for such transport. The inherent fluorescence of FG also provides a nice
benefit to this method, in that the appropriateness of the injection sites and
transport can be checked immediately after the brain is sectioned. Misplaced
injections or cases of poor transport can then be discarded without loss of
more costly reagents.
At the TEM level, FG is seen as being diffusely distributed within the
cytoplasm of retrogradely labeled cells or concentrated within lysosomes
(Fig. 2.5). In certain neurons, the passage of FG into distal dendrites is
excellent, particularly for cells with modest dendritic branching such as
monoamine neurons (Carr and Sesack, 2000b; Van Bockstaele and Pickel,
1995). Within these populations, FG appears to undergo equivalent pene-
tration into the distal dendrites of cells that have different axonal targets
but share similar dendritic morphology. For example, we have analyzed
random ultrastructural samples of dopamine (DA) neurons in the VTA that
were retrogradely labeled with FG from the NAc or from the PFC and found
no difference in the mean or distribution of the cross-sectional diameters
of the dendrites containing FG (unpublished observations). This suggests
that FG was transported to an equivalent extent into large and small caliber
dendrites in the two populations.
Unfortunately, FG fails to be transported into the distal most dendrites
of cell classes with complex dendritic trees, in particular, those with ex-
tensive dendritic spines (e.g., cortical pyramidal cells). This difficulty is
compounded by the need for immunocytochemistry to detect FG for TEM
studies (Chang et al., 1990), with the reduced antibody penetration that
accompanies the requirement for ultrastructural preservation. To some ex-
tent, this limitation is mitigated by the fact that FG induces the formation
Figure 2.5. Electron micrograph of the rat VTA showing dendrites containing im-
munoperoxidase for FG (FG-d) retrogradely transported from the NAc. The labeling
is diffusely distributed to the cytoplasm in one case, but is concentrated in a lysosome
(open arrow) in the other. The latter dendrite is dually labeled by immunogold—
silver for GABA (FG + GABA-d). Scale bar represents 0.5 µm.
of lysosomes in labeled cells, which can aid in the detection of this tracer
(Schmued et al., 1989). However, labeled lysosomes do not always appear
in a particular plane of section. Moreover, individual dendrites can exhibit
markedly different densities of FG content (Figs. 2.5 and 2.14), with the
lowest densities falling near the limit of detection. Another disadvantage of
FG is that it can be taken up by fibers of passage (Dado et al., 1990). The
iontophoretic application method reduces this drawback (Pieribone and
Aston-Jones, 1988), and in our experience, this problem is less with FG than
with other retrograde tracers such as cholera toxin (see also Llewellyn-Smith
et al., 1990). Nevertheless, it is an important limitation that must be taken
into account when interpreting the results of studies using FG. Finally, FG
has only limited usefulness for long-term studies. After 1 week, the intensity
of labeling begins to diminish progressively (Novikova et al., 1997), and FG
can be cytotoxic in some cases (Garrett et al., 1991; Naumann et al., 2000).
c. Pseudorabies Virus (PRV)
The use of live viruses that replicate within neurons (Fig. 2.6) is an es-
sential tool for retrograde tracing of multisynaptic connections in the ner-
vous system, as detailed in the chapter by Geerling et al. and as previously
Figure 2.6. Electron micrograph of the rat PFC showing a neuronal perikaryon that is
infected with PRV (PRV-p) following transport of this tracer from the NAc. Numerous
viral particles (arrowheads) are evident in the cytoplasm and nucleus (n). Scale bar
represents 0.5 µm.
described in reviews and the primary literature (Aston-Jones and Card, 2000;
Card and Enquist, 1994; Card et al., 1993; McLean et al., 1989; Strick and
Card, 1992). In collaboration with J. Patrick Card, we have adapted this
technique to label the distal dendrites of neurons infected with virus through
a first-order process (Fig. 2.7) in order to study identified synaptic inputs
to these dendrites in dual-labeling TEM studies (Figs. 2.8–2.10) (Carr et al.,
1999; Carr and Sesack, 2000a). The technique utilizes an attenuated Bartha
strain of PRV and has a number of advantages for retrograde tract-tracing.
PRV is an exclusively retrograde tracer, although viral strains that undergo
anterograde transport are also available (Aston-Jones and Card, 2000). PRV
has a high affinity for brain proteins and is rapidly sequestered by neurons at
intraparenchymal injection sites, which are typically rather small. The latter
may be disadvantageous if large injections are needed to fill a region of inter-
est. Uptake of PRV by fibers passing through the injection site is lower than
typically observed with most retrograde tracers (Aston-Jones and Card, 2000;
Chen et al., 1999), although such transport can occur and must be taken
into consideration for each study. Viral transport is the only tract-tracing
method in which the sensitivity is not limited by the amount of tracer incor-
porated into cells at the time of injection. The ability of the virus to replicate
progeny (Fig. 2.6), to traffic viral particles into the dendritic tree (Fig. 2.7),
and to cause infected neurons to synthesize viral-specific proteins that con-
centrate in the soma and dendrites (Figs. 2.8–2.10) provides a degree of
signal amplification that is Golgi-like in nature and unique among tracers
(Card et al., 1990). In essence, these processes are driven by the necessity
for the virus to repeat its life cycle in newly infected cells. Hence, replicated
virus is trafficked into the dendrites to facilitate transneuronal passage at
sites of synaptic afferent contact (Aston-Jones and Card, 2000; Card et al.,
1993). Reactive astrocytes that are drawn to infected neurons take up any
Figure 2.7. Schematic diagram illustrating the application of PRV for first-order ret-
rograde transport that labels the distal dendrites of neurons with complex dendritic
trees. PRV is taken up by axon terminals at the site of intraparenchymal injection
(the NAc in this example) and transported retrogradely (to the PFC in this case).
The virus then enters the nucleus (n), in which the virus replicates itself and man-
ufactures viral-specific proteins. Viral particles then acquire two membrane coats
from the Golgi apparatus (ga; only one coat is shown here) and then travel into the
dendritic tree seeking synaptic sites (*) at which to pass out of the cell to infect other
neurons.
infectious particles that are not transported into axon terminals, and thereby
limit nonsynaptic spread. The astrocytes themselves are unable to replicate
infectious virus (Aston-Jones and Card, 2000; Card et al., 1993).
For retrograde tracing in cells with simple dendritic branching patterns,
like VTA neurons, we have shown (Carr and Sesack, 2000a) that the ability of
PRV to label distal dendrites (Fig. 2.8) is comparable to FG (Fig. 2.5). How-
ever, for the identification of the thin distal dendrites and dendritic spines
of cortical pyramidal neurons, infection with PRV (Figs. 2.9 and 2.10) is
superior to any other tracing method in our experience (Carr et al., 1999;
McLean et al., 1989). Moreover, the PRV method is readily amenable to dual-
labeling studies of the synaptic inputs to these dendrites (Fig. 2.10) (Carr
et al., 1999). Although detection of PRV is based on immunocytochemistry,
Figure 2.8. Electron micrograph of the rat VTA showing distal dendrites that are:
singly labeled by immunoperoxidase for the retrograde tracer PRV (PRV-d) trans-
ported from the NAc, singly labeled by immunogold–silver for GABA (GABA-d), or
dually labeled for both markers (PRV + GABA-d). An unlabeled dendrite (ud) is
shown for comparison. The PRV-d receives a symmetric synapse (small arrow) from
an unlabeled terminal (ut). A second unlabeled terminal is apposed to the PRV +
GABA-d without synapsing in this section. Immunogold–silver labeling for GABA is
also evident in glial processes (*). Scale bar represents 0.5 µm.
the primary antibodies are directed against multiple epitopes of viral pro-
teins and so are quite sensitive, even for TEM studies.
Despite the decided advantages of retrograde tract-tracing with PRV for
TEM studies of synaptic connectivity, there are substantial limitations that
are important to note. It is sometimes difficult to precisely localize PRV injec-
tion sites, due to rapid sequestration and transport of the virus (Aston-Jones
and Card, 2000). Viruses cause progressive infections that eventually kill the
subjects. However, the survival times that are used for most tracing exper-
iments are typically earlier than this period (Kelly and Strick, 2000), and
the application of viral tracing for TEM requires only short times needed
for first-order infection. Viruses are living elements and so can have consid-
erable differences in their neuroinvasiveness that reflect growth methods,
storage conditions, titer, and other factors (Aston-Jones and Card, 2000;
Card et al., 1991, 1999). Another significant limitation to viral tracing is the
need for biosafety containment facilities in which to inject and house ani-
mals. PRV is a swine pathogen that is not infectious to humans, but infected
rodents need to be kept isolated from other animals (Strick and Card, 1992).
Herpes simplex virus type 1 and rabies virus are infectious to humans and
therefore require even more extensive biosafety precautions and facilities
Figure 2.9. Electron micrograph of the rat PFC showing immunogold–silver for PRV
in spines (PRV-s) and distal dendrites (PRV-d) of pyramidal neurons retrogradely
labeled from the contralateral PFC. Immunoreactivity for PRV extends from the
parent dendrite into the spine in (A) but not in (B). Both spines receive asymmetric
synapses from unlabeled terminals (ut). Scale bar represents 0.5 µm.
A B
Figure 2.10. Electron micrographs of the rat PFC showing immunogold–silver for
PRV in the spines of pyramidal neurons (PRV-s) retrogradely labeled from the con-
tralateral PFC and immunoperoxidase labeling for tyrosine hydroxylase (TH) in
axon terminals (TH-t). (A). The PRV-s receives an asymmetric synapse (large arrow)
from an unlabeled terminal (ut) and is apposed by a TH-t that does not form a
synapse in this section. (B). The TH-t forms a symmetric synapse (small arrow) onto
the PRV-s. Scale bar represents 0.5 µm.
that may be prohibitive for some laboratories (Kelly and Strick, 2000; Strick
and Card, 1992).
One of the most attractive aspects of viral tracing, namely its transneu-
ronal transport, may lead to false-positive results when this method is used
for TEM studies. Avoiding such outcomes requires careful attention to sur-
vival times (Aston-Jones and Card, 2000) and control experiments to ensure
that second-order infection has not occurred within the region of interest at
the time when TEM studies are performed (Carr et al., 1999). For example, in
our prior analysis of cortical pyramidal neurons infected by first-order trans-
port, we monitored second-order infection of GABA local circuit neurons
whose synapses onto the pyramidal cells are proximal and therefore likely
to be the earliest to pass virus. Survival times at which substantial infection
of GABA cells was detected were not used in the TEM analysis of synaptic
inputs to pyramidal neurons.
Another notable drawback to PRV tract-tracing is the fact that some neu-
rons may not label with this virus. For example, our repeated attempts to
label the known projection from the PFC to the VTA with this method were
unsuccessful. The most probable explanation is competition for uptake of
the virus at the injection site (Card et al., 1999), with afferents having the
largest terminal density being most likely to transport quantities of virus
sufficient to mount an infection. Indeed, our injections of PRV into the
VTA lead to extensive retrograde transport into brainstem sites, seemingly
at the expense of forebrain areas.
As a final consideration of the viral tracing method for use in studying
synaptic connectivity, it is worth considering why we have emphasized TEM
investigation of first-order transport rather than the better known and more
easily applied method of LM analysis of transsynaptic transport. For certain
experiments, we agree that carefully timed LM studies of viral trafficking
are sufficient to provide evidence of synaptic connections between given
cell populations. However, for other circuits, there may be multiple path-
ways that can transport virus, making it difficult without further manipula-
tions to be certain which course viruses took in moving between synaptically
connected cells (Fig. 2.11). For example, if investigating whether the PFC
synapses onto DA neurons in the VTA that project to the NAc, one might
inject PRV into the NAc and wait for second-order transport into the PFC
by way of the VTA. However, following NAc injections, the labeling of PFC
neurons could arise from connections within the basolateral amygdala or
the paraventricular thalamus rather than the VTA (Problem 1). Indeed, our
inability to label the PFC by first-order viral uptake from the VTA (see above)
suggests that these alternate pathways would be the more probable routes
for virus to reach the PFC. Another likely circuit would involve transneu-
ronal transport of virus into PFC pyramidal neurons via intrinsic synapses
onto neighboring pyramidal cells that underwent first-order infection from
the NAc (Problem 2). Finally, and perhaps most importantly, even if selective
lesions could confine the virus into passing transsynaptically via the VTA,
the method would not reveal the phenotype of VTA cells involved in this
Figure 2.11. Schematic diagram illustrating the use of PRV for multisynaptic tract-
tracing to test whether PFC neurons synapse onto DA cells (D) in the VTA that
project to the NAc. Interpretational problems arise (see section “Retrograde Tract-
Tracing”) when the virus passes transneuronally through alternate pathways or via
GABA cells (G).
transneuronal passage. For example, second-order infection of PFC neurons

would most likely occur due to uptake by mesoaccumbens GABA neurons
that receive PFC synaptic input (Problem 3) (Carr and Sesack, 2000a). Re-
cent technological developments are beginning to solve the latter problem,
for example in creating a form of PRV whose replication is dependent on
Cre-mediated recombinant events that can be isolated to neurons with spe-
cific phenotype (DeFalco et al., 2001). However, if such an approach is not
available for the complex circuit being investigated, the use of TEM to an-
alyze first-order retrograde transport of virus remains a viable option for
examining synaptic inputs to identified cell populations.
3. Identification of Neurochemical Phenotype
For tract-tracing experiments conducted for TEM analysis, it is often im-

portant to identify the neurochemical phenotype of the neurons/dendrites
targeted by the traced pathway. For example, we wished to determine the
extent to which projections from the PFC innervate DA or GABA neurons in
the VTA (Carr and Sesack, 2000a; Sesack and Pickel, 1992). For such studies,
antibodies directed against transmitters, enzymes, receptors, transporters,
calcium-binding proteins, or peptides that are unique identifiers of the tar-
get cells can be used. Combinations of immunocytochemical markers are
then needed to label both the tracer and the phenotypic marker. As de-
scribed above in section “Dual-Labeling Procedures,” the most typical com-
bination we use is preembedding immunoperoxidase and immunogold–
silver, although other combinations are also available. It is important to
consider that the peroxidase method is more sensitive than gold–silver by ap-
proximately one order of magnitude (Chan et al., 1990). Hence, we typically
use peroxidase or immunoperoxidase to label the least abundant antigen
(e.g., BDA or PHA-L) and immunogold–silver to label the more plentiful
antigen (e.g., TH or GABA).
In other cases, it may be desirable to identify the neurochemical pheno-
type of neurons containing retrograde tracer from a known injection site.
Often such studies are conducted at the LM level, taking advantage of the
many available fluorescent markers for double and triple labeling (Skirboll
et al., 1989) (see also the chapter by Wouterlood). However, some pheno-
types are difficult to detect by LM because a particular marker is not present
in sufficient quantity to clearly define the soma. In our experience, this
has been a particular problem with GABA in projection neurons. Although
the levels of GABA are high in local circuit neurons, they are markedly
less robust in projection neurons, such as those in midbrain and basal gan-
glia structures. Often it is difficult to define these cells using LM methods.
Use of GAD antibodies can help with an LM definition of this population,
although even this approach requires colchicine treatment to boost GAD
levels, and colchicine can interfere with transport of retrograde tracer (Ford
et al., 1995). Hence, TEM becomes a useful alternative for determining the
extent to which GABA cells contribute to particular projections, as GABA
levels are sufficient to detect by TEM. For this purpose, we have followed
the protocol originated by Van Bockstaele (Van Bockstaele et al., 1994; Van
Bockstaele and Pickel, 1995) to demonstrate the strength of the GABA pro-
jection from the VTA to the PFC (Carr and Sesack, 2000b).
Finally, some experiments may call for the identification of neurochemical
phenotype in axons that innervate particular brain regions as determined
by anterograde tract-tracing. In our experience, this is a particularly diffi-
cult issue to resolve using dual-labeling preembedding methods. Although
evidence of dually labeled axons is often obtained, it appears that the inci-
dence of such antigen coexpression is underestimated by the combination
of immunoperoxidase and immunogold–silver that our laboratory typically
uses. This issue is further discussed below in section “Limitations: Sources
of False-Negative Errors”.
4. Triple-Labeling Studies
TEM studies can be designed that analyze several combinations of tract-

tracing agents and neurochemical markers. The general methodological
principles for combining tracing with neurotransmitter phenotype iden-

tification for LM and TEM have been described in a previous volume of
this series (Zaborszky and Heimer, 1989). Here, we deal specifically with
experiments in which investigators seek to define the synaptic target of a
particular anterogradely labeled pathway both in terms of its neurochemi-
cal phenotype and in terms of its own projection targets. This requires the
combination of two tracers, one anterograde and one retrograde, with a
phenotypic marker.
As an example of such an application, we will refer to our findings with
regard to the afferent regulation of midbrain DA neurons. The major inputs
and outputs of the substantia nigra (SN) and VTA have been known for some
time (Oades and Halliday, 1987; Phillipson, 1979; Swanson, 1982), although
more minor projections are still the subject of investigation. It is known that
the major forebrain projections of the VTA arise from separate popula-
tions of neurons, both DA and non-DA, presumably GABA cells (Carr and
Sesack, 2000b; Swanson, 1982; Van Bockstaele and Pickel, 1995). DA cells
innervating the NAc regulate locomotion and motivated behaviors, whereas
DA neurons projecting to the PFC modulate cognitive and affective func-
tions. To understand the morphological basis of behavioral control of DA
cell activity, it is important to develop a detailed picture of the specific affer-
ents that synapse onto different populations of VTA DA neurons. Afferents
can be identified by anterograde tract-tracing or, in some cases, phenotypic
markers, if these derive from single sources (e.g., acetylcholine). Cell pop-
ulations can be defined on the basis of retrograde tract-tracing from known
target areas (e.g., PFC and NAc), and neurotransmitter identity of these pop-
ulations can be delineated based on immunocytochemistry for transmitter
markers. In the first study of its kind, our laboratory used a combination
of anterograde and retrograde tract-tracing with immunocytochemistry to
show that excitatory projections from the PFC synapse selectively onto DA
and not GABA neurons that project back to the PFC and onto GABA but
not DA cells that innervate the NAc (Carr and Sesack, 2000a). Such synaptic
specificity may explain some of the unique functional properties of mesopre-
frontal and mesoaccumbens neurons and help to advance understanding
of the circumstances and mechanisms for their behavioral activation.
More recently, we have performed a similar study of the inputs to the
VTA from the brainstem laterodorsal tegmentum (LDT), which, unlike the
PFC, synapses onto DA neurons that innervate the NAc (Omelchenko and
Sesack, 2005a). The LDT projection contains a mixed neurochemical phe-
notype, and with this in mind, we have also completed an analysis of cholin-
ergic inputs to different VTA cell populations (Omelchenko and Sesack,
2005b), as the cholinergic innervation derives predominantly from the LDT
(Hallanger and Wainer, 1988; Oakman et al., 1995).
The success of these experiments depends on the combination of im-
munoperoxidase and immunogold–silver labeling methods, and on the use
of one of these markers, typically immunoperoxidase, to label two of the
three desired components that are known to be segregated in different
Figure 2.12. Schematic diagrams showing the labeling procedures used for triple
labeling in the experiments described here. (A) Two primary antibodies raised in
different species against FG and phenotypic protein 1 (Prot 1) are co-applied, and
then a biotinylated secondary antibody against the first species is added, followed by
the avidin–biotin peroxidase complex. The latter will also bind to the biotin in the
anterograde tracer BDA. Following peroxidase histochemistry, a gold-conjugated
secondary antibody against the second species is added, and the bound gold parti-
cles are silver enhanced. (B) Three primary antibodies raised in different species
against FG, a protein labeling the afferent pathway (Prot 1; e.g., PHA-L or a unique
phenotypic marker such as VAChT) and a phenotypic protein labeling the target
(Prot 2) are co-applied. A mixture of biotinylated secondary antibodies against the
first two species is then applied, followed by the avidin–biotin peroxidase complex.
Following peroxidase histochemistry, a gold-conjugated secondary antibody against
the third species is added, and the bound gold particles are silver enhanced.
neuronal compartments. In the most typical case, BDA is used as the

anterograde tracer, and FG is the retrograde tracer of choice. Immunoper-
oxidase labeling for FG in soma and dendrites is accomplished using the
ABC method, which will by design also label the axons that contain BDA
(Fig. 2.12A). Immunogold–silver is then used to label the phenotypic marker
in the soma and dendrites. In this case, the primary antibodies against FG
and the phenotypic protein must be raised in separate species. The method
can also be adapted for the use of PHA-L as the anterograde tracer or the
use of a phenotypic marker that labels a class of afferent axons from a known
source, for example the vesicular acetylcholine transporter (VAChT) to label
cholinergic afferents to the VTA that are known to derive from the brain-
stem tegmentum. In this case, there are three primary antibodies, and each
must either be from a different species (Fig. 2.12B) or two can be from the
same species as long as they will be segregated into different compartments
(e.g., rabbit anti-FG and rabbit anti-PHA-L).
It is important for these experiments that appropriate controls are run
to verify the segregation of immunoperoxidase markers. Sections pro-
cessed only for FG should reveal peroxidase product for this tracer only
in dendrites, either diffused in the cytoplasm or concentrated in lysosomes
(Figs. 2.5 and 2.13), and not in axons. Similarly, sections processed only
for the anterograde tracer should reveal peroxidase product for this tracer
diffusely distributed within axons and not soma or dendrites (Fig. 2.13).
Figure 2.13. Schematic diagram illustrating the differential distribution of tracers

and phenotypic markers within different compartments in an experiment in which
anterograde and retrograde tract-tracing are combined. Peroxidase or immunoper-
oxidase is used to label both the retrograde tracer FG within soma and dendrites
and the anterograde tracer BDA or PHA-L (or a unique phenotypic marker) within
axon terminals. Control experiments are needed for each system under study to en-
sure that the tracers label only their respective compartments. Immunoperoxidase is
typically diffusely distributed within these compartments but is occasionally concen-
trated within lysosomes in the case of FG. Finally, preembedding immunogold–silver
is used to label antigens unique for different neuronal phenotypes in the retrogradely
labeled cell population(s).
Figure 2.14. Electron micrographs of the rat VTA showing the juxtaposition of axon
terminals labeled by peroxidase for BDA (BDA-t) anterogradely transported from
the LDT, dendrites singly labeled by immunoperoxidase for FG (FG-d) retrogradely
transported from the NAc, singly labeled by immunogold–silver for TH (TH-d), or
dually labeled for both markers (FG + TH-d). (A) The BDA-t forms an asymmetric
synapse (large arrow) onto the FG-d. In (B) the BDA-t synapses onto the TH-d, while
the FG + TH-d receives synaptic input from an unlabeled terminal (ut). Scale bar
represents 0.5 µm.
This is especially important for BDA, which is capable of some degree of

retrograde transport in certain neuronal pathways. The segregation of the
immunogold–silver is less of a concern, as it may appear in soma and den-
drites (Fig. 2.13) as well as axons, for example in the case of GABA.
Using this approach, it is possible to see synaptic contacts on several differ-
ent dendrite populations. For example, the anterograde tracer may occur
within axon terminals that synapse onto dendrites containing only the retro-
grade tracer and not the phenotypic marker (Fig. 2.14A) or onto dendrites
that are phenotypically labeled but do not contain the retrograde tracer
(Fig. 2.14B). However, in fortuitous cases, the axons containing anterograde
tracer are observed to synapse onto dendrites that contain both the retro-
grade tracer and the phenotypic marker (Fig. 2.15A). Alternatively, axons
labeled by a phenotypic marker (e.g., VAChT) will synapse onto dendrites
that have these characteristics (Fig. 2.15B).
For this method to provide a useful estimate of the extent to which
synapses occur between identified inputs and outputs of a region, it is neces-
sary to perform a certain degree of analysis in serial sections. In many cases,
labeled axons may be closely apposed to labeled dendrites without synapsing
Figure 2.15. Electron micrographs of the rat VTA showing axon terminals labeled by
peroxidase for BDA (BDA-t in A) anterogradely transported from the LDT or for
VAChT (VAChT-t in B) forming asymmetric synapses (large arrows) onto dendrites
dually labeled by immunoperoxidase for FG retrogradely transported from the PFC
or the NAc, respectively, and immunogold–silver for TH (FG + TH-d). In (A) a
second FG + TH-d does not receive synaptic input in this section. Scale bar represents
0.5 µm.
onto them in one plane of section. In this case, synapses may be revealed in
immediately adjacent sections. Serial sections also assist the determination
of whether low levels of gold–silver particles are repeated over the same
structure in adjacent sections and therefore likely to be specific. However,
in this case, it would be preferable to examine sections closer to rather than
further from the surface (see Fig. 2.16). In order to thoroughly investigate
the possible presence of synapses onto each identified cell population, it
is also necessary to sample extensive amounts of tissue. This is particularly
true if the investigator uses mesh grids as recommended in section “General
Sampling Issues,” as the metal obscures part of the tissue being examined.
As further discussed in that section below, extensive sampling is necessary
to address the possibility that failure to find a particular synapse type is not
due simply to underrepresentation in the sample.
Other investigators have developed a triple-labeling approach that utilizes
two different anterograde tract-tracing agents and examines whether both
inputs converge onto a common target neuron (French and Totterdell,
2002, 2003). In this case, the target neuron is defined not by neurochemical
phenotype but by morphological type, as assessed by its uptake of locally
administered BDA. The success of this method relies on careful comparative
analyses between LM and TEM in order to identify different qualities of
immunoperoxidase for the two anterograde tracers and identify regions
of probable synaptic input to the target neuron. These regions are then
analyzed by TEM to verify the presence of synapses.
III. PRINCIPLES OF THE METHODS
Although tract-tracing and immunocytochemical methods can be com-

bined in a variety of ways, the major steps of the TEM preembedding
procedures are usually the same. First, the animal is deeply anesthetized
until all pain reflexes are gone. The animal then undergoes transcardial
perfusion with aldehydes to fix proteins. The brain blocks are sliced with
a Vibratome; some method is used to enhance antibody penetration; and
free-floating brain sections are incubated in antibody solutions. Tissue lipids
are then fixed with osmium tetroxide, and sections are dehydrated and em-
bedded in plastic resin. After ultrathin sectioning and counterstaining, the
tissue is then examined by a TEM. Many of these procedures are common
to LM immunolabeling and so involve similar technical concerns: antibody
specificity, prevention of nonspecific labeling, penetration enhancement,
etc. However, it is important to appreciate that TEM studies are not simply
LM experiments with some extra steps. The experiments must be designed
with TEM in mind, which typically involves altering procedures in order to
balance maximal preservation of ultrastructure with optimal detection of
the desired antigens.
A. Animals
The methods presented here are designed primarily for small rodents,
although they can be adapted for larger animals. Moreover, many of the
immunocytochemical and sampling procedures are fully applicable to other
species once fixation is completed and brain sections are cut. An important
question to consider for TEM studies is how many animals are typically
needed. The answer varies depending on the details of the experiment. If
tract-tracing is part of the design, additional animals are often required to
allow for misplaced injections. If a new antibody is being tested, pilot studies
of optimal fixatives and dilutions are needed. Once all the parameters are
optimized, the number of animals required may depend on the results that
are obtained. Abundant proteins or highly robust synaptic connections will
typically be observed in multiple samples from within and across animals.
In that case, three animals each showing similar protein localization or a
similar frequency of synapse detection may be sufficient. The more the data
seem to vary within or between animals, the greater the sampling required
to ensure that the conclusion being developed accurately represents reality.
A question that relates to the issue of variability is the extent to which

animals with different densities of labeling should be represented in the
overall sample. Obviously, one would not be expected to produce a large
sample from an animal in which the immunodetection procedures were
clearly suboptimal, just to say that animals were sampled equivalently. Such
an approach is sure to produce false-negative results. Conversely, if the tis-
sue from one particular animal seems to have been blessed with the most
optimal immunolabeling and ultrastructural preservation, it is tempting to
overrepresent this animal in the sample, arguing that it is actually more rep-
resentative of reality. Of course, a balance needs to be struck between the de-
sire for absolute rigor and the search for “truth” in neuroanatomy. Animals
with poor immunolabeling should be excluded from quantitative analyses,
and conclusions should not be based on data from single animals. Moreover,
clearly stating in published work how the sample was collected and how ex-
tensively each animal was represented in the sample should allow colleagues
to make their own conclusions regarding the validity of the findings.
B. Tract-Tracing
The principles of experimental design for combining anterograde and

retrograde tract-tracing with immunochemical markers have been exten-
sively described in the sections above. Here, we wish to present some method-
ological issues associated with animal welfare. The surgery necessary to intro-
duce multiple tract-tracing agents can be rather long. However, we chose to
combine tracers that have similar survival times (e.g., BDA and FG) specif-
ically so that they can be injected during a single survival surgery. Most
institutions require extensive justification for performing multiple survival
surgeries, and for the best interests of the animals, we endeavor to avoid
such designs. Investigators must choose an anesthetic regimen that can be
maintained over several hours. We prefer to use a mixture of ketamine
(34 mg/kg), xylazine (7 mg/kg), and acepromazine (1 mg/kg) that is in-
jected i.m. For long surgeries, it is also imperative that the animals’ tem-
perature be monitored with a rectal probe and maintained at 36.5◦ C using
a thermostatically regulated heating pad. It is also important to monitor
animals closely during recovery from surgery and to administer analgesic if
they show any signs of pain or discomfort. We recommend butorphanol, 2
mg/kg, s.c. every 8 h as needed.
C. Phenotypic Labeling
In the sections above, we have also extensively described the principles

of immunocytochemistry as related to detection of neurochemical pheno-
type. One issue that was not addressed is the use of intracerebral (typically
intraventricular) injections of colchicine to enhance the content of certain
proteins or peptides so that they may be more readily detected by LM or TEM
(Dube and Pelletier, 1979; Ford et al., 1995; Graybiel and Chesselet, 1984;
Ribak et al., 1978). By disrupting axonal transport (Paulson and McClure,

1975), colchicine leads to accumulation of peptide in soma and sometimes
in dendrites, thereby improving the sensitivity of antigen detection. Most
commonly, colchicine is injected into the lateral ventricles and allowed to
perfuse the brain through the ventricular system. However, local injections
of colchicine within a brain region can reveal the most likely sources of
peptide-containing afferents that can then be verified with retrograde tract-
tracing (Arluison et al., 1994). In either case, colchicine interferes with ax-
onal transport of tracing agents (Monti-Graziadei and Berkley, 1991), and
so if the goal of an experiment is to examine phenotypic markers in ret-
rogradely labeled cells, it is necessary to inject the tracer first and perform
intracranial injection of colchicine in a second survival surgery. Because
colchicine treatment is distressful for animals, it is recommended that the
second survival time be no more than 24–48 h and that butorphanol is given
at 2 mg/kg, s.c. every 8 h to relieve suffering. It should also be borne in mind
that the full consequences of colchicine treatment are not yet understood.
Data indicate that the drug may evoke abnormal gene expression, protein
synthesis, and morphological changes within neurons (Pirnik et al., 2003;
Rho and Swanson, 1989; Yan and Ribak, 1999).
D. Intracardial Perfusion
1. Pretreatments
Certain neuronal elements contain endogenous metals (e.g., zinc in glu-

tamate nerve terminals) that can complex with silver during the silver
enhancement steps for preembedding immunogold–silver labeling. Hence,
animals are first treated with a zinc chelator to minimize this source of spu-
rious labeling (Veznedaroglu and Milner, 1992).
2. Choice of Fixative
Immunocytochemical staining methods necessitate labeling antigens in

fixed cellular material. The goal of tissue fixation for immunocytochemical
processing is the preservation of tissue in a state as close to natural as possi-
ble while maintaining the ability of the antigen to react with the antibody.
Rapid, thorough preservation of the brain is required. A detailed descrip-
tion of the fixation procedures used in our laboratory is provided in the
Appendix. However, it should be noted that the characteristics of some an-
tibodies allow optimal labeling only following certain types of fixation. For
instance, the antibody against the neurotransmitter DA is generated against
a peptide sequence that is conjugated to glutaraldehyde (Chagnaud et al.,
1987). Because of this trait, labeling with this antibody is feasible only fol-
lowing fixation using high amounts of glutaraldehyde. On the other hand,
certain fixatives may denature antigens, resulting in little to no staining.
Therefore, the optimal fixative for each antigen–antibody combination must

be determined empirically.
The standard fixative for TEM is typically some combination of 4%
formaldehyde and glutaraldehyde in concentrations from 0.05 to 1% de-
pending on compatibility with primary antibodies. However, in our expe-
rience, certain antibodies actually produce better immunolabeling in tis-
sue fixed with 2% formaldehyde and 3.75% acrolein, a related aldehyde
that gives excellent tissue preservation for TEM. This can be true even for
antibodies that label only poorly in glutaraldehyde-fixed sections, such as
those generated against the monoamine plasma membrane transporters de-
scribed above. Hence, we consider it worth the effort to attempt acrolein fixa-
tion in testing new antibodies, although many laboratories avoid this chem-
ical because the hazardous risks associated with it are considered greater
than for glutaraldehyde. The basic procedure for acrolein perfusion was
presented in Leranth and Pickel (1989). Here in the Appendix, we provide
information on how to safely mix acrolein solutions and perform intracar-
dial perfusions in rodents without undue risk to personnel. Of course, each
laboratory should become familiar with the material safety data sheet for
acrolein as for any hazardous chemical.
Regardless of the fixative employed, the speed at which fixation of brain
tissue is accomplished is essential for the best ultrastructural preservation. In
our experience, the time between the opening of the animal’s diaphragm
and the introduction of fixative should be as short as possible (20–30 s).
Extensive saline rinsing to remove blood cells only delays the time at which
fixative is introduced. Hence, we recommend the inclusion of heparin with
the initial saline rinse to quickly remove blood cells and prevent clotting
within the vasculature. It is also important to avoid introducing air bubbles
into the system, as these might also block vessel perfusion.
3. Postfixation and Sectioning
A well-fixed brain should contain no visible blood and be firm to the

touch. It should be ready for sectioning on a Vibratome after a short post-
fixation period (30–60 min). Postfixation is usually performed in the final
fixative that was perfused through the animal. Inappropriately fixed tissue
will often be difficult to section, and although the brain may be further
hardened by postfixation overnight, such prolonged exposure to fixative
can render antigens of interest inaccessible to primary antibodies. In our
opinion, a brain that does not section well should be abandoned unless it
is quite valuable. Cutting on a Vibratome is the recommended method of
sectioning brain material, as it avoids any artifacts that would be associated
with frozen sectioning. As presented previously in this series (Leranth and
Pickel, 1989), it is recommended that sections be treated with a brief in-
cubation in 1% sodium borohydride in order to stop fixation (by reducing
any aldehydes still exposed to the tissue) and reduce background labeling.
Finally, in the event that immunocytochemical labeling cannot proceed im-
mediately, or the investigator wishes to save sets of tissue for future analysis
(not unusual in the case of tract-tracing studies), it is possible to store sec-
tions in a cryoprotectant solution (see Appendix, section “Cryoprotection
and Tissue Storage”). However, pilot studies should be run to determine
whether this storage alters antigenicity for any given protein.
E. Immunocytochemistry
1. Penetration Enhancement
For many immunocytochemical experiments, materials or procedures are

introduced to the tissue sections to disrupt cellular membranes in order
to potentiate antibody penetration, and thus increase the level of labeling.
However, because the membranes must remain visibly intact for electron mi-
croscopic examination, only the mildest techniques for enhancing antibody
penetration can be employed. The typical compounds used for TEM are de-
tergents such as Triton X-100 (0.04%) or surfactants like PhotoFlo (0.1%).
Some investigators use a slightly higher concentration of these reagents but
only briefly expose tissue to them during the normal serum blocking pro-
cedure. Alternatively, the lower concentrations can be used throughout the
primary antibody incubation. In our experience, the use of detergents like
Triton tends to reduce background nonspecific labeling, which is advanta-
geous. On the other hand, we have found a few proteins whose antigenicity
is actually reduced by detergent treatment (Luedtke et al., 1999; Sesack and
Snyder, 1995), most likely due to disruption of transmembrane domains
upon solubilization. Hence, penetration enhancement can also be accom-
plished by treating tissue with a cryoprotectant and subjecting it to rapid
freezing and thawing using either liquid nitrogen or a −80◦ C freezer. The
recipe for the latter procedure is given in the Appendix.
2. Choice of Markers
TEM involves passing a beam of electrons through the tissue specimen.

Therefore, only markers that can trap electrons and thus appear “electron
dense” can be visualized with this method. In contrast to the use of varied
colors for LM techniques, there are only a few types of markers that can be
distinguished from each other at the TEM level. These include autoradiog-
raphy (i.e., silver grains developed in a photographic emulsion), gold with
or without silver enhancement, and peroxidase chromogens. Autoradio-
graphic immunolabeling has been discussed in detail in a previous chapter
in this series (Pickel and Milner, 1989). This method has high sensitivity,
and the product is easy distinguishable from immunogold and immunoper-
oxidase markers. However, it can take 3–12 months to develop autoradio-
graphic material for TEM. Moreover, it is not entirely suited for subcellular
localization studies, as radioactivity can spread to produce silver grains a

short distance away from antigen sites. In recent years, immunoautoradiog-
raphy has generally been replaced by methods that are quicker to develop
and provide more discrete localization.
In the sections above, we have provided general information regard-
ing the advantages and disadvantages of using immunoperoxidase versus
immunogold–silver for preembedding studies of protein localization or
synaptic connectivity. Here, we discuss more specific details of the proce-
dures within these two general classes.
a. Preembedding Immunoperoxidase
One of the earliest types of immunoperoxidase labeling involved the appli-

cation of a soluble complex of peroxidase enzyme bound to antiperoxidase
antibodies, termed PAP (for peroxidase antiperoxidase) (Sternberger,
1974). This procedure, which has been previously discussed in this series
(Pickel, 1981), involves primary antibody binding to tissue antigens, fol-
lowed by application of a secondary antibody raised in another species and
directed against the species in which the primary antibody was raised, and
finally addition of the PAP complex for which the antiperoxidase is raised in
the same species as the primary antibody. Following exposure of the perox-
idase enzyme to a chromogen substrate (e.g., DAB) and H2 O2 , an electron-
dense reaction product is formed. The density of the peroxidase product
can then be further enhanced by osmication of the tissue ( Johansson and
Backman, 1983). Moreover, an additional round of exposure to the sec-
ondary antibody and the PAP complex produces amplification of the per-
oxidase signal (Ordronneau et al., 1981). Subsequent efforts to enhance
even further the incorporation of peroxidase, and thus the sensitivity of the
method, lead to the introduction of the ABC technique (Hsu et al., 1981). For
this procedure, the secondary antibody is conjugated to several molecules
of biotin. The tissue is then exposed to a solution containing avidin, which
binds to biotin with high affinity, and biotinylated horseradish peroxidase.
Similar to the other types of immunoperoxidase procedures, exposure of
the enzyme to a chromogen results in a flocculent, electron-dense reaction
product. Although most laboratories now use the ABC method as their sole
immunoperoxidase technique, we continue to find applications for the PAP
method, for example when using immunogold–silver to localize compounds
like BDA that would otherwise produce false dual labeling if exposed to ABC
(Pinto and Sesack, 1998). In addition, the presence of endogenous biotin
in some glial cells (Yagi et al., 2002) may present a circumstance in which
the PAP method would be preferred to ABC.
As previously reviewed in this series, there are several chromogens that
can be used for immunoperoxidase, the most common being DAB, benzi-
dine dihydrochloride, tetramethylbenzidine, VIP peroxidase, and SG per-
oxidase (both from Vector Laboratories Burlingame, CA) (Warr et al., 1981;
Zaborszky and Heimer, 1989; Zhou and Grofova, 1995). DAB is the most
commonly used chromogen for TEM studies, as it produces the smallest
crystalline size and therefore the most diffuse, flocculent reaction product
that can fill even small neuronal processes. It is also more stable for the TEM
processing steps than tetramethylbenzidine. More recently, VIP peroxidase
has shown excellent sensitivity and stability for TEM studies, and its distinc-
tive rosette-like appearance may provide easier recognition within dendrites
where the diffuse nature of the DAB reaction product may sometimes be
difficult to discern (Zhou and Grofova, 1995). Benzidine dihydrochloride
appears to be less sensitive than DAB or VIP peroxidase and its precipitates
are not always uniform in appearance (Zhou and Grofova, 1995). Neverthe-
less, it is still a useful marker for abundant antigens in large processes such as
soma and proximal dendrites (Charara et al., 1996). Tetramethylbenzidine is
reported to have greater sensitivity than other chromogens, but its instability
in aqueous solutions and alcohol requires that it be stabilized for use in TEM
(Llewellyn-Smith et al., 1993; Marfurt et al., 1988; Rye et al., 1984). Such stabi-
lization procedures can result in some loss of sensitivity. Because these differ-
ent immunoperoxidase chromogens produce precipitates of different size,
texture, and appearance, they can be used in combination for dual-labeling
studies (Norgren and Lehman, 1989; Smith et al., 1994; Zaborszky and
Heimer, 1989; Zhou and Grofova, 1995). However, it may not always be pos-
sible to distinguish the presence of two different peroxidase products within
the same neuronal structure, especially axon terminals (Zhou and Grofova,
1995) (see also section “Limitations: Sources of False-Negative Errors”).
Another important consideration for TEM studies is the strength of the
immunoperoxidase reaction. We have experimented with changing the con-
centration of the DAB chromogen, the concentration of the H2 O2 , and the
duration of the incubation. In our experience, the main determinant of
the amount of peroxidase product generated is the incubation time. The
time in solution should be chosen empirically: inadequate exposure will not
label antigens deep in the tissue, whereas overexposure will obscure ultra-
structural detail. The peroxidase product in overstained profiles may also
pierce the plasma membrane and spread into adjacent profiles. This is usu-
ally readily detectable in the TEM as peroxidase product in the vicinity of
broken membranes. In the event that the sensitivity of immunoperoxidase
is needed for a study but a clear view of cellular detail is also critical, the
DAB precipitate can be intensified by metallic silver grains through an ar-
gyrophil III reaction (Gallyas, 1982). This procedure has been applied with
high sensitivity in TEM studies (Liposits et al., 1984) and has been further
modified to allow enhancement of DAB reaction product that is deliber-
ately produced at a low level so as not to obscure subcellular detail (Smiley
and Goldman-Rakic, 1993). A further variant of this approach is to perform
silver–gold enhancement of a nickel-DAB precipitate, and this method pro-
duces a reaction product that can be distinguished from nonenhanced DAB
for dual-labeling TEM studies (Hajszan and Zaborszky, 2002). In this regard,
it should be noted that it is not uncommon to find light silver labeling of
immunoperoxidase in tissue that has been dually labeled by the standard

ABC and preembedding immunogold–silver methods. However, the size of
these particles is generally quite small and therefore readily distinguished
from specific immunogold–silver particles.
b. Preembedding Immunogold–Silver
The protocols for immunogold–silver labeling are rather different from

the immunoperoxidase steps, and the distinction begins at the time of sac-
rifice. As noted above, it is recommended that the animal should be treated
with a zinc chelator prior to perfusion in order to prevent silver inten-
sification of endogenous zinc (Veznedaroglu and Milner, 1992). Another
approach using l-cysteine exposure to reduce tissue argyrophilia in fixed
sections has also been developed (Smiley and Goldman-Rakic, 1993).
Following incubation of sections in primary antibody, they are exposed to
secondary antibodies conjugated to “ultrasmall” gold particles in the range
of 1 µm. There are several commercial sources of these antibodies as well
as the silver enhancement solutions that are matched to them. The most
common sources of silver reagents used for preembedding immunogold–
silver are Amersham Biosciences Corps (Piscataway, NJ), Nanoprobes Inc.
(Yaphank, NY), and Aurion (Electron Microscopy Sciences, Fort Washing-
ton, PA). Regarding dual- or triple-labeling studies such as those described
here (sections “Dual-Labeling Procedures” and “Triple-Labeling Studies”),
it is important to note that Electron Microscopy Sciences offers a line of
gold-conjugated secondary antibodies raised in donkey. We have found that
having all secondary antibodies from donkey helps to avoid species cross-
reaction.
A critical determinant for the success of preembedding immunogold–
silver is the size of the gold–silver particles, and hence the total incubation
time in silver reagents. This issue has been extensively covered in the original
reviews of the method (Chan et al., 1990; Pickel et al., 1993). Briefly, at short
incubation times, labeled structures appear light gold at the LM level. As
silver intensification proceeds, the color moves increasingly toward brown
and finally black. In addition, the levels of nonspecific gold–silver deposit
increase with time. The LM appearance of specific gold–silver labeling that
is considered optimal for TEM is generally in the range of brown. The
total incubation time to achieve this is empirical and based largely on the
primary antibody and the density of the antigen. It is recommended that
each experiment involve a timed series (4–12 min) on a few test sections
and then bulk processing of the remaining tissue sections at two different
time points, one that is deemed optimal by LM (clear detection of the
structures of interest with minimal evident background) and one that is
1–2 min shorter.
The silver processing steps also require extensively clean glassware, which
usually involves acid washing. In our experience, it is more convenient,
though admittedly more costly, to use disposable cultured cell well plates
for this purpose. In addition, there must be no metal ions in the tissue
at the time when the enhancement procedure is performed. Any metal
ions present in the tissue will be silver intensified in addition to the gold
particles, resulting in nonspecific labeling. Therefore, the tissue must be
manipulated using nonmetal instruments. For this purpose, our laboratory
employs wooden applicator sticks.
Another important issue is the temperature at which the silver reaction
is conducted. The amount of time necessary for the silver enhancement is
highly dependent on the temperature of the silver solution (i.e., the warmer
the solution, the shorter the silver enhancement time required). In our lab-
oratory, we store the silver solutions in the refrigerator, but allow them to
reach room temperature prior to use. Furthermore, once the silver enhance-
ment has been performed, the solutions to which the tissue is subsequently
exposed must be at room temperature. In other words, the osmium tetrox-
ide solution must be at an ambient temperature prior to osmication of
silver-intensified immunogold-labeled tissue. Otherwise, it is possible that
tissue expansion and contraction upon temperature changes will dislodge
silver-enhanced gold particles from the tissue.
3. Parallel Versus Serial Antibody Incubations
In our experience, the most efficient way to perform dual or triple im-
munocytochemical labeling is to incubate sections in primary antibodies
raised in different species in a parallel manner, followed by serial appli-
cations of secondary antibodies for immunoperoxidase and immunogold–
silver respectively. Of course, it is also possible to apply primary antibodies in
a sequential manner, performing all of the immunoperoxidase procedures
before incubation in the second primary antibody. However, we have ob-
served at least one case in which low levels of peroxidase reaction product
were washed out during subsequent lengthy antibody incubations. Hence,
we continue to prefer the parallel versus the serial approach.
4. Antibody Dilutions
The optimal dilution of primary antibody will vary considerably and must
be empirically determined for each antigen–antibody combination and for
each brain region. For example, in the VTA where TH immunoreactivity in
soma and dendrites is abundant, anti-TH antibodies can be used at higher
dilution than in the forebrain where the levels of TH in axons are lower.
Moreover, the concentration of primary antibody often has to be several
times higher for immunogold–silver than for immunoperoxidase due to
the lower sensitivity of the former method. For initial pilot studies, a range
of antibody dilutions should be tested that include concentrations published
in the literature. Additional recommendations for dilution testing have re-

cently been published (Saper, 2003).
5. Antibody Controls
It is essential for any laboratory using immunocytochemical methods to

establish the specificity of the antibodies they are employing. The appro-
priate control experiments for immunocytochemical detection of antigens
for TEM are similar to those required for LM or for confocal microscopy
and have been discussed elsewhere (Saper, 2003). Briefly, lack of immunola-
beling must be observed following: (1) omission of the primary antiserum,
(2) exclusion of the secondary antiserum, (3) preadsorption of the primary
antibody with the antigen prior to exposure to the tissue, and, where pos-
sible, (4) incubation of tissue sections from transgenic mice in which the
antigen has been “knocked out.”
When performing double- and triple-labeling procedures, special atten-
tion should be paid to ensure absence of cross-reaction between secondary
antibodies. Obviously, the best results can be achieved if all secondary anti-
bodies used in the study are obtained from the same species. However, the
immunological similarity between certain species (e.g., sheep and goat) can
still lead to problems, and we have sometimes noted cross-reaction of certain
antibodies against immunologically different species (e.g., anti-rabbit IgG
labeling rat primary antibodies). For biotinylated IgG, we recommend the
use of antibodies with minimal species cross-reaction, for example those
obtained from Jackson ImmunoResearch Laboratories, West Grove, PA.
However, as gold-conjugated antibodies are unlikely to be developed for
minimal species cross-reaction, it is still necessary to perform the following
control experiment. Incubate tissue sections in a mixture of primary anti-
bodies that are known to have distinct compartmentalization or are known
not to colocalize (e.g., rabbit anti-TH and mouse anti-GABA in striatal sec-
tions). Then divide the tissue into two sets and perform dual immunoperox-
idase and immunogold–silver with one of the two secondary antibodies left
out of each set. Evidence of markers in inappropriate compartments or un-
expected dual labeling of structures will indicate that a particular secondary
antibody labels more than its respective species.
F. Tissue Preparation for TEM
1. Osmication
Treatment of tissue sections with osmium tetroxide provides fixation of

lipids by rendering them insoluble prior to dehydration and plastic embed-
ding. Osmium also imparts a heavy metal stain to the lipids, enhancing the
contrast of membranes. However, osmium is a strong oxidizing agent that
can alter the antigenicity of many proteins, making it a challenge to immuno-
cytochemically label such proteins following plastic embedding (Hemming
et al., 1983). Consequently, investigators have developed methods to sub-
stitute other chemicals for osmium to achieve lipid fixation without loss of
antigenicity (Phend et al., 1995).
For preembedding procedures, the greatest challenge with osmium is that
it can oxidize silver metal to silver salt with resultant loss of the silver used
to enhance immunogold labeling. In our experience, loss of silver during
the osmication step is primarily associated with the presence of chloride
ions in the buffering solution. The use of PBS (phosphate-buffered saline)
or PB buffers that have been pH adjusted with HCl introduces chloride
anions that can form a salt with silver cations, and this seems to greatly
speed the oxidation of silver metal to silver salt. Indeed, when inadvertently
using solutions that contained chloride, we have witnessed the oxidation of
silver in brain sections to nondetectable levels within 1–2 min. Conversely,
scrupulous avoidance of chloride ions (or other halides that can readily
form silver salts) can reduce this problem and slow the rate of oxidation so
that it is minimal within the typical times used for osmication. Hence, the
buffer mixed with osmium tetroxide should be PB that has been pH adjusted
with phosphoric acid and contains no chloride ions. Using this reagent, it
should not be necessary to shorten the osmication time.
2. Dehydration
Preparation of tissue for electron microscopy requires fixation of lipids

followed by extensive dehydration. During these procedures, the volume
of tissue decreases (Hillman and Deutsch, 1978), suggesting that a portion
of the extracellular space has been lost. Therefore, caution must be exer-
cised in determining some measurements, such as distance between labeled
structures, and it should always be acknowledged that such measurements
are only semi-quantitative in nature.
3. Counterstaining
Tissue to be examined by electron microscopy is typically stained with

heavy metals (first osmium and then uranyl and lead) in order to increase
the electron density of membranes and enhance contrast. Many procedures
call for application of uranyl acetate en bloc, meaning on sections during
the alcohol dehydration steps (usually with the 70%). However, such treat-
ment can sometimes make it difficult to identify sparse immunoperoxidase
labeling, such as that associated with low abundance proteins. For this rea-
son, we routinely apply heavy metal stains only after ultrathin sectioning and
only on a portion of the grids. In this case, the investigator has the option
to omit uranyl acetate and stain ultrathin sections only with lead citrate in
order to visualize low amounts of immunoperoxidase.
G. Tissue Sampling
1. General Sampling Issues
In preembedding tissue labeled by immunoperoxidase or immunogold–

silver, it is crucial to examine only the surface of sections, which is rep-
resented by the interface between the tissue and the plastic embedding
medium. Because large amounts of penetration enhancers are not feasible
for immunolabeling of tissue for TEM, antibodies do not have access to
the full extent of the section thickness. Thus, maximal antibody penetra-
tion is limited to a few microns of the tissue surface. Moreover, even if one
restricts sampling to the tissue surface, the actual protein levels will be un-
derestimated with these techniques, due to the chemical fixation and low
concentrations of penetration enhancers. Hence, immunocytochemistry for
TEM involves a compromise that balances immunodetection with morpho-
logical integrity. Immunoreactivity that is excessive on the surface may not
penetrate more than a few microns (Fig. 2.16). Conversely, morphological
preservation may be excellent within the depths of the tissue section but
compromised at the surface where antibody penetration is greatest. More-
over, the extent to which immunoreagents penetrate the tissue is not equal
between methods, with preembedding immunogold–silver generally pene-
trating less well than immunoperoxidase (Fig. 2.16) (Chan et al., 1990).
Knowing that immunoreactivity is confined to the outer surface of flat-
embedded sections, the investigator must decide what approach to take
regarding ultramicrotomy. Some investigators actually turn their thick sec-
tions on edge and cut ultrathin sections perpendicular to (i.e., at 90◦ from)
the original plane of sectioning. This approach has the advantage of al-
lowing the experimenter to measure precisely the distance from the tissue
surface where immunolabeling becomes weak or nondetectable. However,
in addition to being a considerable technical challenge, this method has
the disadvantage of causing lost perspective regarding how the area being
sectioned relates to typical LM views of the region of interest. It is also
possible to cut the tissue surface at an oblique angle so that the zone con-
taining optimal antibody penetration and tissue morphology is “stretched”
over many sections. However, in our experience, this reduces the size of the
useful zone within each ultrathin section and creates the need to collect
many more sections in order to obtain a sufficient sample size. Hence, our
approach has always been to embed the thick sections in plastic as flat as
possible (by placing them under glass slides and heavy lead bricks) and to
perform ultramicrotomy en face.
In order to accomplish the goals of analysis at the tissue surface, we further
recommend collecting ultrathin sections on grids that contain at least three
sections, each with part tissue and part embedding resin. Focusing the anal-
ysis on the middle section allows serial examination in sections both deeper
and more superficial to the central one (Fig. 2.16). Using mesh grids allows
the creation of a grid map for estimating the area analyzed and relocating
Figure 2.16. Schematic diagrams showing the sampling strategy used for preem-
bedding immunoelectron microscopic studies (see text, section “General Sampling
Issues”).
regions of interest during serial section analysis or in the event that better
photomicrographs are required. Mesh grids are also generally tougher and
stand up well to repeat handling. Of course, loss of desired regions under
the metal mesh is inevitable with these grids. If this becomes a problem for a
study, then slot grids are recommended so that all of the tissue is supported
on a nonobscuring film like formvar. However, slot grids are generally more
fragile and it can be difficult to relocate specific areas of interest if it becomes
necessary after a viewing session.
At low magnification, it is advisable to take a photomicrograph or make
a drawing of the ultrathin section and assign letter and number coordi-
nates to each grid square. At 3000× magnification, our laboratory chooses
grid squares that (a) contain at least 25%, but no more than 75% tissue
or (b) share an edge (not a corner) with a square that meets the first cri-
terion (Fig. 2.16). These criteria ensure that the TEM sample is collected
at a relatively uniform distance from the tissue surface. Recording which
grid squares are to be analyzed and the approximate amount of tissue that
each contains (25, 50, 75, or 100%) allows estimation of the area of tissue
analyzed. For example, thin bar copper 400 mesh grids (Electron Mi-
croscopy Sciences) have squares that are 55 µm on a side, or 3025 µm2
area. This is then multiplied by the number of squares analyzed and the
proportion of tissue that each square contained to derive an estimate of the
total tissue area sampled.
Selected squares are then analyzed at higher magnification (10–30,000×).
For each square, it is good to record the number of fields that contain specific
peroxidase and/or specific gold–silver labeling. We define a “field” as the
area delimited by the photographic brackets on the microscope. In this case,
either fields that are photographed or simply analyzed for content without
photo documentation can be recorded for the number of events per unit
area (area of the bracketed region). In the case of receptor or transporter
studies, this gives an estimate of the density of immunolabeling (i.e., number
of labeled profiles per unit area). Profiles can be further analyzed for the
position of gold–silver particles in relation to the plasma membrane or other
structures. For synaptic connectivity, fields are analyzed for whether labeled
processes (usually axons) contact unlabeled or labeled targets (usually soma,
dendrites, or spines) or make no obvious contacts. A coordinate relocation
system on the microscope can be used to examine serial sections to verify
whether labeling of profiles is specific (e.g., number of gold–silver particles
per profile or per unit area) and whether synaptic specializations are present
at points of contact.
2. Criteria for Immunoperoxidase
In general, specific labeling using immunoperoxidase staining is easy to

discern as a flocculent, dense precipitate within labeled structures. Occa-
sionally, nonspecific or “background” labeling may also be present. Unfor-
tunately, there are no established principles for estimating the amount of
background staining in immunoperoxidase-labeled tissue. In fact, in some
instances, nonspecific labeling may be somewhat difficult to determine, be-
cause it is typically related to the primary antibody. In other words, if the
primary antibody is omitted, the background staining disappears. It is for
this reason that simply omitting the primary antibody is not an acceptable
test for specificity; rather, it is merely a control for the specificity of the sec-
ondary antibody. Specific immunoperoxidase staining should be confined
to structures that have the potential to synthesize the antigen. For instance,
labeling in cells that do not express the mRNA for the protein has a high
likelihood of indicating nonspecific or background reactivity and must be
examined with caution. Background labeling also tends to be more diffuse
and less dense than specific peroxidase product. Nevertheless, we have often
noted that diffuse cellular immunoperoxidase labeling that is visualized by
LM is not always evident by TEM. Conversely, TEM may detect immunore-
activity that is too weak to produce a visible signal in LM. In any event, the
presence of suspected background labeling necessitates rigorous tests for
specificity of the primary antibody.
3. Criteria for Immunogold–Silver
Immunogold–silver labeling is sometimes associated with a higher proba-

bility of background labeling than is immunoperoxidase, necessitating the
establishment of consistent criteria for determining specific immunoreactiv-
ity. Such criteria will depend on the localization and density of the antigen as
well as the level of background immunogold–silver labeling. Basically, exper-
imenters should be confident that their criterion for specific immunogold–
silver labeling would not be met by randomly distributed gold–silver par-
ticles throughout the tissue, and this in turn is affected by the size of the
structure of interest. For example, the probability that three gold–silver par-
ticles would distribute randomly within a proximal dendrite is high, but the
presence of three random particles within an axon terminal is a low prob-
ability event. Similarly, the probability that two gold–silver particles within
an axon terminal would both be randomly distributed to the plasma mem-
brane is a low probability event compared to the same particles localized
to the cytoplasm. Hence, for our published studies of plasma membrane
transporters, we have set our criteria for specific immunogold–silver label-
ing within axons as at least two particles on the plasmalemma or three in
total (including those in the cytoplasm) (Miner et al., 2000, 2003c). These
criteria are somewhat more conservative than other laboratories (Garzón
et al., 1999; Pickel and Chan, 1999), and so we must acknowledge a higher
likelihood of false-negative outcomes in our studies. On the other hand, we
can more confidently assert the absence of false-positive results.
IV. SUMMARY OF ADVANTAGES AND LIMITATIONS
A. Advantages
Preembedding immunoperoxidase methods involving signal amplifica-

tion have the advantage of superior sensitivity for the localization of sparse
antigens. This is particularly true for TEM, as the greater resolving power
of electron microscopy allows detection of low levels of peroxidase reaction
product that may appear too diffuse for LM identification. The more discrete
and nondiffusible marker associated with the preembedding immunogold–
silver technique provides the advantage of indicating the precise subcellular
localization of antigens, including neurotransmitter receptors and trans-
porters. Discrete gold particles, with or without silver enhancement, are also
advantageous because they are more readily quantified than the precipitate
formed by peroxidase reaction. When combined, the immunoperoxidase
and immunogold methods provide a powerful means for localizing two or
more antigens in relation to each other. With TEM detection, these ap-
proaches allow the identification of specific synaptic relationships formed
between cellular elements that are labeled by tract-tracing and/or neuro-
chemical phenotype. If two antigens are known to be localized to separate
neuronal compartments (e.g., anterograde and retrograde tracing agents
in dendrites and axons, respectively), then the combination of preembed-
ding immunoperoxidase and immunogold methods can be used for triple-
labeling studies, specifically the determination of whether afferents into a
region of interest synapse onto populations of cells identified both by their
neurotransmitter phenotype and major axonal target.
B. Limitations: Sources of False-Positive Errors
For any TEM study involving tract-tracing and immunocytochemistry,

consideration must be given to potential sources of false-positive and false-
negative results. Common sources of false-positive errors are cross-reaction
of the primary antibody with unknown proteins, cross-reaction of secondary
antibodies with inappropriate species, nonspecific immunolabeling, or erro-
neous transport of tract-tracing agents. The evidence supporting specificity
of immunoreagents for TEM is the same as for LM studies and should be
demonstrated with appropriate controls prior to the experiment. Much has
been written on this subject and on sources of nonspecific/background la-
beling. The reader is referred elsewhere for a consideration of these issues
(Saper, 2003) (see also section “Antibody Controls”).
For experiments involving tract-tracing, the investigator should be famil-
iar with the LM literature on afferents and efferents of each region of interest
and hence the potential sources of false-positive results if tracer injections
spread beyond the target. Such cases should be eliminated if tracer spread
would involve an adjacent pathway that is not the one under study. An-
other common source of false-positive labeling in tract-tracing studies of
the type described here is uptake of retrograde tracer by fibers of passage.
The extent of this problem varies with different retrograde tracers, but the
tracer with the least amount of uptake into passing fibers is PRV. If per-
forming retrograde tract-tracing from a region in which uptake by fibers
of passage is likely to produce false-positive results, it is recommended that
PRV be used, and that potential contributors to false positives are systemat-
ically checked in control experiments. Finally, and as described above, use
of BDA for anterograde tracing has the potential to produce false-positive
results if it undergoes retrograde transport and subsequent anterograde
trafficking into collateral axons. When using BDA, care should be taken to
examine the most likely brain regions in which retrograde transport might
occur. Evidence of retrograde transport indicates that collateral transport
is a possibility; such cases should be discarded, and PHA-L should then be
used as the preferred anterograde tracer.
C. Limitations: Sources of False-Negative Errors
False-negative results are a common concern in studies using TEM im-

munocytochemistry due to the limited penetration of immunochemicals in
tissue prepared with minimal or no detergent. This drawback especially af-
fects the detection of immunogold reagents, which typically penetrate less
deeply into the tissue than immunoperoxidase compounds (Chan et al.,
1990). Although these limitations cannot be completely avoided, their im-
pact can be minimized by utilizing the less sensitive immunogold–silver
method to label antigens in high abundance, confining the analysis to the
surface of the sections where both peroxidase and gold–silver markers are
present, and analyzing serial sections for all profiles with sparse labeling. An-
terograde and retrograde tract-tracing can also contribute to false-negative
results, because all neurons in a population of interest are unlikely to be la-
beled in any given study. The use of multiple animals with slightly different
injection sites can help to overcome this limitation, as can the use of large
injections where possible. Finally, for retrograde tract-tracing with FG, an
additional source of false-negative results discussed earlier (section “Retro-
grade Tract-Tracing”) is that the tracer may be confined to lysosomes and
not spread diffusely within dendrites. Extensive sampling, examination of
dendrites in serial section, and following dendrites cut longitudinally to see
whether lysosomes are present can help to minimize this problem.
An additional methodological consideration for the methods presented
here is the ability to detect the presence of multiple markers within one
profile. Unfortunately, the limitations of preembedding immunocytochem-
istry are such that it may not be possible to detect the colocalization of
markers in every structure in which these antigens are present, particularly
when the neuronal structures are small. This limitation may result from
imperfect antibody penetration, spatial interference when relatively large
reagents compete for access to antigens within confined spaces, differential
location of antigens (e.g., cytoplasmic versus vesicular), disproportional con-
centration of antigens, and unequal sensitivity of immunoperoxidase and
immunogold–silver methods. In our experience, such impediments typically
have a greater impact on the dual labeling of axon terminals as opposed to
dendrites. Sometimes, the problem can be addressed by switching the or-
der in which immunoperoxidase and immunogold labeling are performed
(Katona et al., 2001) or using a sequential method of incubation in primary
antibodies rather than the parallel method recommended here. Adjusting
the concentrations of the primary antibody may also help, for example, re-
ducing the concentration of the antibody directed against the antigen in
greatest abundance. However, this issue remains a limitation of preembed-
ding methods and one for which postembedding methodologies (Charara
et al., 1996; Smith et al., 1996) may be superior for avoiding false-negative
outcomes.
V. PROSPECTS FOR THE FUTURE
It is hoped that microarray studies (see chapter by Ginsberg et al.) will

eventually identify unique phenotypic markers for neuronal pathways and
so render obsolete the need for tract-tracing. Having such protein markers
would avoid the need for survival surgery in animals. Moreover, identify-
ing pathways based on unique protein signature is likely to have greater
sensitivity compared to tract-tracing, in that antibodies have the potential to
detect all the neuronal structures that contain that protein, whereas tracer
injections virtually never include all the cells and/or processes that con-
tribute to a given projection. Of course, limited antibody penetration will
still present difficulties for interpretation of TEM studies of protein localiza-
tion or synapse identification. A second potential solution to the limitations
of tract-tracing would be the introduction of genetically altered animals in
which marker proteins (e.g., green fluorescent protein) are expressed in
cells of interest (Zhao et al., 2004) that are otherwise difficult to study by
standard methods, for example GABA neurons that project from the VTA
to the PFC. The ability to be certain that the marker protein is expressed in
all such cells and their processes will allow a more complete analysis of their
synaptic organization at both cell body and nerve terminal levels.
APPENDIX
Here, we will present a full protocol for the immunoperoxidase and

immunogold–silver procedures for TEM. This can be followed for single-,
dual-, or triple-labeling experiments.
A. Recipes for Standard Buffers
1. 0.2 M Sodium Phosphate Buffer
1000 ml distilled water

21.8 g sodium phosphate dibasic
6.4 g sodium phosphate monobasic
pH to 7.3 with phosphoric acid, not HCl (see section “Osmication” )
Dilute 1:1 with water to make 0.1 M sodium phosphate buffer (PB)
2. 0.01 M Phosphate-Buffered Saline (PBS)

50 ml 0.2 M PB
9 g sodium chloride
3. 0.1 M Tris-Buffered Saline (TBS)

12.1 g trizma base
9 g sodium chloride
pH to 7.6 with HCl
B. Fixation
1. Rat Preparation for Immunogold Procedure
Animals are first anesthetized with 60 mg/kg pentobarbital i.p. and then
given 1 g/kg i.p. of diethyldithiocarbamic acid (Sigma, St. Louis, MO) for
15 min prior to aldehyde perfusion. During this treatment, animals should
be carefully monitored for seizures, which can be induced by chelation of
zinc. In the event that seizure activity is detected, animals should be given
supplemental doses of anesthetic.
2. Perfusion and Fixatives
Correctly prepared fixatives for perfusion should be filtered and then

checked to ensure that they are clear and colorless.
a. 3.8% Acrolein, 2% Formaldehyde in 0.1 M PB
Rats are perfused with 10 ml heparin saline (1000 U/ml; Elkins-Sinn,

NJ), followed by 50 ml of 3.8% acrolein and 2% formaldehyde, followed by
200–400 ml of 2% formaldehyde in 0.1 M PB. Coronal blocks of brain are
postfixed in 2% formaldehyde for 30–60 min.
In a well-ventilated hood, prepare 2% formaldehyde in 0.1 M phosphate
buffer as follows. Heat 500 ml of ultrapure water in a 1-l glass beaker to
60–65◦ C. Do not exceed 65◦ C. Turn off the heat and add 20 g of EM grade
granular paraformaldehyde (Electron Microscopy Sciences; Fort Washing-
ton, PA), stirring constantly. Stir for several minutes and then add small
volumes of 1 N NaOH, stirring for several minutes after each addition until
the solution is mostly clear. A few granules of paraformaldehyde might still
be present. Filter through a Buchner funnel and an aspiration flask using #3
filter paper. Also filter 500 ml of 0.2 M PB. Transfer the solution to a beaker.
For one rat, measure 48.1 ml of the 2% solution of freshly depolymerized
paraformaldehyde (i.e., formaldehyde) into a 100 ml graduated cylinder. We
recommend the use of acrolein from Electron Microscopy Sciences because
it is supplied in 2 ml single-use glass ampoules and so does not require
storage of opened containers of acrolein. Wearing gloves and eye goggles
use a 5-cm3 syringe and an 18-G needle (the large gauge is needed because
of the high vapor pressure) to remove 1.9 ml of acrolein from the glass
Figure 2.17. Equipment setup for intracardial perfusion with acrolein by using a
peristaltic pump. The rack above a collection container allows recovery of perfusate
for safe disposal, and the entire assembly is placed within a laminar flow hood to
contain acrolein vapors.
ampoule. Add acrolein to the graduated cylinder; seal tightly with parafilm
and invert several times to mix. For two rats, use two ampoules of acrolein
and add 3.8 ml of acrolein to 96.2 ml of 2% formaldehyde.
The perfusion system that is needed for this fixative consists of a peri-
staltic pump and tubes that are attached to a three-way stopcock with two
inlets and one outlet, all assembled in a well-ventilated laminar flow hood
(Figs. 2.17 and 2.18). This system allows the delivery of two to three differ-
ent solutions without the introduction of air bubbles that might block brain
capillaries. Although not shown in the figure, we recommend using a metal
clamp stand to ensure that the graduated cylinder containing acrolein does
not inadvertently tip over.
The rat is placed on a perfusion rack that is set over a plastic container
in order to collect the blood with acrolein perfusate. Full descriptions of
Figure 2.18. Tubing arrangement for acrolein perfusion. The use of a three-way stop-
cock allows for two inlet lines and one outlet. The first inlet tube is primed for
acrolein, and the stopcock is then switched to shut off the acrolein and allow flow
through the second inlet tube. This is first rinsed with water to remove residual
acrolein and then primed with heparin saline. The outlet tube is run into the hep-
arin saline, which circulates while the animal’s abdomen and thorax are exposed.
The cannula from the outlet tube is then inserted through the base of the left ven-
tricle into the aorta and clamped in place. The stopcock is then switched to allow
acrolein to perfuse the animal. This shuts off the flow of heparin saline, making it
safe to move the inlet tube from the heparin saline into the formaldehyde. Once
the correct volume of acrolein is perfused, the stopcock is switched one final time
to complete the perfusion with formaldehyde.
the surgical procedure can be found elsewhere (Friedrich and Mugnaini,

1981; Leranth and Pickel, 1989). Briefly, once the anesthetized rat’s pain
reflexes have ceased completely, tape down the arms and tail. Turn the
peristaltic pump to run at a flow rate of approximately 50 ml/min. Cut a
wide opening in the lower abdomen; find the xiphoid process and clamp
with a regular hemostat. Holding the hemostat cut up either side of the
rib cage, cut open the diaphragm and remove any membrane surrounding
the heart. Slit open the right atrium and then the base of the left ventricle.
Push the outflow tubing with the cannula through the left ventricle and up
into the aorta; clamp the cannula into the aorta with a vascular hemostat.
The cannula can also be clamped onto the left ventricle if preferred, but in
either case the hemostat should be propped to keep it from twisting. Imme-
diately switch the stopcock to introduce the acrolein with 2% formaldehyde
and turn up the perfusion speed to 90 ml/min; pump through 50 ml of
acrolein. In the meantime, transfer the heparin inflow tubing into the plain
2% formaldehyde. After acrolein, switch the stopcock to pump through
200 ml of formaldehyde and turn down the pump speed to 80 ml/min.
When handling acrolein, always wear gloves and eye protection. Vials
should only be opened and the perfusion should only be performed in a
well-ventilated hood with the shield lowered. Acrolein remaining in the am-
poules and any fluids containing acrolein after the perfusion is completed
should be collected into a labeled glass waste container that is stored in a
flammable liquids cabinet until it can be properly disposed. Use a funnel
to empty the blood/perfusate into the glass storage bottle and rinse several
times with water until the collection container is safe to remove from the
hood for final cleaning. Any items that contact acrolein (vials, ampoules,
syringes, etc.) should be kept in the hood at least overnight until the solu-
tion has evaporated (even a small amount of acrolein put in the trash will
soon become evident to anyone in the room). If a spill occurs outside the
hood, evacuate the room immediately and call chemical safety. If possible,
turn on the hood, as this may help to clear the acrolein vapors and contain
them to the affected room. Use a safety shower to wash any skin or clothing
that contacts acrolein.
b. 0.05–1% Glutaraldehyde, 4% Formaldehyde in 0.1 M PB
Rats are perfused first with heparin saline as above, followed by 500 ml
of the para/glut fixative. Coronal blocks are postfixed in 4% formaldehyde
for 30–60 min.
In a well-ventilated hood, prepare 4% formaldehyde in 0.1 M phosphate
buffer as follows and as described in detail above. Heat 500 ml of ultra-
pure water. Add 40 g of EM grade granular paraformaldehyde, followed by
small volumes of 1 N NaOH until clear. Filter solution, followed by 500 ml
of 0.2 M PB. Transfer the solution to a beaker. For the standard 0.2% glu-
taraldehyde, add 8 ml of 25% EM grade glutaraldehyde to a liter of the 4%
freshly depolymerized paraformaldehyde.
3. Vibratome Sectioning
After perfusion, remove the brain from the skull and cut it into thick
blocks that contain the brain regions of interest. Postfix for the times
described above, and then transfer to 0.1 M PB and section on a Vibratome.
For the best sectioning, use well-fixed brains, fill the Vibratome with 0.1 M
PB that has been cooled to 4◦ C, use fresh, sharp blades, and set the slicer to
low speed and wide amplitude. Section at 40–60 µm and collect sections in
serial order into cell wells containing 0.1 M PB.
4. Sodium Borohydride and Hydrogen Peroxide Treatments
Rinse sections in 0.1 M PB and divide into multiple conditions as de-

sired. Incubate sections for 30 min in 1% sodium borohydride in PB. As
sections will float to the top, maintain hydration by mixing them back down
occasionally. Rinse sections extensively in PB until all bubbles are gone. If
endogenous peroxidase activity has the potential to confound interpreta-
tion in the study, treat sections for 15 min with 3% H2 O2 in 0.1 M TBS and
then rinse extensively in this buffer.
5. Cryoprotection and Tissue Storage
For long-term storage in a solution that is compatible with later EM anal-

ysis, place sections into the following cryoprotectant (generously provided
by Darlene Melchitzky) and freeze at -20◦ C. Other potential cryoprotectant
solutions may also serve this purpose (Rosene et al., 1986).
Storage cryoprotectant
300 ml ethylene glycol
300 ml glycerol
100 ml 0.2 M PB
300 ml ultrapure water
C. Immunolabeling
Immunolabeling procedures are performed on free-floating sections at

room temperature with constant shaking unless otherwise specified.
1. Primary Antibody Steps and Penetration Enhancement
a. Optional, Freeze–Thaw Procedure (protocol generously provided

by Dr. Yoland Smith)
Place sections in cryoprotectant for 20 min.

Freeze thaw cryoprotectant
200 ml 0.2 M PB
80 ml glycerol
200 g sucrose
qs final volume to 1000 ml
Solution can be stored in −20◦ C freezer.
Place sections into −80◦ C freezer for 20 min.
Thaw sections at room temperature for 10 min each in the following
solutions: cryoprotectant at 100, 70, 50, and 30% diluted in PBS.
Rinse sections in PBS (3 × 5 min).
Rinse sections in TBS (3 × 5 min).
b. Blocking Solution
87 ml TBS
10 ml 0.4% Triton X-100 (Sigma) (optional)
Final concentration is 0.04% Triton.
3 ml normal serum
1 g BSA
Place sections in blocking solution for 30 min.
Incubate sections in primary antibody made up in blocking solution
overnight at room temperature or over two nights at 4◦ C.
Rinse sections in 0.1 M TBS (1 min then 3 × 10 min).
If single labeling for immunogold–silver is being performed, proceed to
section “Immunogold Labeling.” Otherwise, follow the next steps.
2. Immunoperoxidase Labeling
Mix biotinylated secondary antibody in blocking solution. Typical final

concentrations are 1:100 to 1:400.
Incubate sections in secondary antibody for 30 min.
Mix ABC complex solution using the Vectastain Elite kit (Vector Labo-
ratories) by adding two drops each of A and B solutions to 10 ml of
0.1 M TBS. Allow to stand at least 30 min before use. Take care not
to contaminate the dropper bottles and do not overmix or vortex the
solution.
Incubate sections in ABC solution for 30–120 min.
Prepare the DAB (Sigma) immediately before use. To 100 ml of 0.1 M
TBS, add 22 mg of DAB and 10 µl of 30% H2 O2 . Filter the solution.
Incubate sections in DAB for 3–6 min, depending on the desired strength
of the reaction, keeping in mind that copious labeling in the light
microscope may appear overblown by TEM.
Stop the reaction by rinsing sections in 0.1 M TBS (1 min then 3 × 5 min).
For tissue to be plastic embedded after immunoperoxidase labeling, trans-
fer sections to 0.1 M PB (2 × 5 min) and proceed to section “Tissue
Preparation for Electron Microscopy.” Otherwise, follow the next steps.
3. Immunogold Labeling
Rinse sections in 0.01 M PBS (1 min then 3 × 5 min).

Place sections in washing buffer for 30 min.
Washing buffer
93.5 ml 0.01 M PBS
0.8 g BSA
0.5 ml fish gelatin (comes with the gold secondary antibody)
6 ml normal serum
Incubate sections in gold-conjugated secondary antibody diluted 1:50 in
washing buffer for 2–4 h or overnight.
Rinse sections in washing buffer (1 min then 3 × 5 min).
Rinse sections in 0.01 M PBS (1 min then 3 × 5 min).
In a ventilated hood, fix sections in 2% glutaraldehyde in PBS for 10 min.
Remove silver enhancement solutions from the refrigerator (see below).
Rinse sections in 0.01 M PBS (1 min, then 3 × 5 min) until odor of
glutaraldehyde is gone.
0.2 M sodium citrate buffer
5.88 g sodium citrate dihydrate
pH to 7.4 using citric acid
0.2 M citric acid
4.2 g citric acid monohydrate
Keep refrigerated.
4. Silver Enhancement
Fill a 12-well cultured cell plate and a 24-well plate as shown in Figure
2.19. For the 0.1 M PB in the last two rows of wells, make sure to use solution
that is not contaminated by chloride ions, as these are the final steps before
osmication. Divide up the tissue in batches in the first row of 0.01 M PBS
depending on the number of silver enhancement times to be tested. Do
not proceed until all the wells (except silver) are filled with solution and
the silver enhancement solutions are at room temperature. Do not allow
sections to sit for long periods in the citrate buffer, as this has only weak
buffering capacity that is not optimal for morphological preservation.
Once all is ready, fill the second row of the 24-well culture plate with
equal drops of IntenSE M kit A and B (Amersham, Arlington Heights, IL)
Figure 2.19. Schematic illustration of cell culture wells showing the order of reagents
used for silver intensification at different time points.
solutions. The exact volume will depend on the number and size of the
sections, but usually 10 drops each of A and B are sufficient for 4–8 sections.
Using a wooden applicator stick with a pointed end (snap in half from the
ends to get a clean break), transfer sections to 0.2 M sodium citrate buffer
in the second row of the 12-well cell plate. Rinse sections for 1 min. Using
a new wooden applicator stick (to minimize contamination by phosphate),
transfer sections to the third row of the 12-well cell plate and rinse for 1 min.
Now transfer sections to the citrate buffer in the first row of the 24-well cell
plate. Rinse for 1 min. Then transfer sections to the silver solution (second
row of the 24-well cell plate) and start a timer.
Gently swirl the plates and time the silver enhancement reaction carefully,
as differences of 30 s are significant. If needed, use a dissecting microscope
to watch the progress of silver intensification. Stop the silver enhancement
reaction by rinsing in buffer. Do not return sections to the silver once they
are removed. Transfer sections into the third through sixth rows of the 24-
well cell plate in succession (1 min each for citrate buffer and longer for PB).
Sections can remain in the final row of 0.1 M PB until dishes for osmication
are ready.
In practice, we have found it useful to perform the steps above for a pi-
lot determination of optimal silver enhancement times on a small number
of sections (see also section “Choice of Markers”). We usually run test sec-
tions at wide intervals (e.g., 4, 8, and 12 min), mount these on slides, and
examine them by LM. The most optimal silver enhancement time is that
which produces clearly labeled neuronal elements, typically with a golden-
brown color, with minimal background (e.g., few silver-enhanced gold par-
ticles in the white matter). Pale gold labeling is probably underdeveloped;
black labeling is considered overdeveloped and likely to be associated with
greater background particles (see also Chan et al., 1990). For the actual sil-
ver enhancement of the experimental sections, we recommend dividing the
sections into half and running them at two different times, 1–2 min apart.
Subsequent TEM analysis may reveal that one of these tissue sets has better
labeling characteristics with regard to specific and nonspecific labeling.
D. Tissue Preparation for Electron Microscopy
Rinse sections in 0.1 M PB (2 × 5 min) in Coors dishes. Place dishes

in a well-ventilated hood. Prepare 2% OsO4 in 0.1 M PB by mixing equal
volumes of 4% OsO4 and 0.2 M PB. Make sure sections are lying flat in the
Coors wells then slowly draw off the phosphate buffer without disturbing
the sections. Gently add OsO4 taking care not to twist or fold the sections.
Incubate in OsO4 for 1 h in the hood.
The remaining steps for tissue preparation are standard in electron mi-
croscopy (Friedrich and Mugnaini, 1981) and will not be presented in detail
here. Briefly, after rinsing several times with PB, sections are dehydrated at
10-min intervals through a standard series of increasing strength ethanol so-
lutions (30%, 50%, 70%, 95%, 100%, 100%), then twice in 100% propylene
oxide or acetone, followed by embedding resin equally mixed with propy-
lene oxide or acetone. The sections are left for several hours or overnight
and then infiltrated with pure resin for 2–4 h. Sections are then embed-
ded between sheets of commercial plastic. Once polymerized at 60◦ C for
at least 18–24 h, the resin-embedded sections are trimmed and cut on an
ultramicrotome, and ultrathin sections are collected onto mesh or coated
slot grids.
REFERENCES
Alheid, G. F., Edwards, S. B., Kitai, S. T., Park, M. R., and Switzer, R. C. I., 1981, Methods for
delivering tracers, In: Heimer, L., and RoBards, M. J. (eds.), Neuroanatomical Tract-Tracing
Methods, New York: Plenum Press, pp. 91–116.
Aoki, C., Miko, I., Oviedo, H., Mikeladze-Dvali, T., Alexandre, L., Sweeney, N., and Bredt, D. S.,
2001, Electron microscopic immunocytochemical detection of PSD-95, PSD-93, SAP-102,
and SAP-97 at postsynaptic, presynaptic, and nonsynaptic sites of adult and neonatal rat
visual cortex, Synapse 40:239–257.
Arai, R., Kojima, Y., Geffard, M., Kitahama, K., and Maeda T., 1992, Combined use of silver stain-
ing of the retrograde tracer WGAapoHRP-Au and pre-embedding immunocytochemistry
for electron microscopy: demonstration of dopaminergic terminals in synaptic contact
with striatal neurons projecting to the substantia nigra in the rat, J. Histochem. Cytochem.
40:889–892.
Arluison, M., Brochier, G., Vankova, M., Leviel, V., Villalobos, J., and Tramu G., 1994, Demon-
stration of peptidergic afferents to the bed nucleus of the stria terminalis using local
injections of colchicine. A combined immunohistochemical and retrograde tracing study,

Brain Res. Bull. 34:319–337.
Aston-Jones, G., and Card, J. P., 2000, Use of pseudorabies virus to delineate multisynaptic
circuits in brain: opportunities and limitations, J. Neurosci. Methods 103:51–61.
Baude, A., Nusser, Z., Molnár, E., McIlhinney, R. A. J., and Somogyi, P., 1995, High-resolution
immunogold localization of AMPA type glutamate receptor subunits at synaptic and non-
synaptic sites in rat hippocampus, Neuroscience 69:1031–1055.
Beaulieu, C., Campistron, G., and Crevier, C., 1994, Quantitative aspects of the GABA circuitry
in the primary visual cortex of the adult rat, J. Comp. Neurol. 339:559–572.
Bernard, V., Somogyi, P., and Bolam, J. P., 1997, Cellular, subcellular, and subsynaptic distribu-
tion of AMPA-type glutamate receptor subunits in the neostriatum of the rat, J. Neurosci.
17:819–833.
Bloch, B., Bernard, V., and Dumartin, B., 2003, “In vivo” intraneuronal trafficking of G protein
coupled receptors in the striatum: regulation by dopaminergic and cholinergic environ-
ment, Biol. Cell 95:477–488.
Brandt, H. M., and Apkarian, A. V., 1992, Biotin-dextran: a sensitive anterograde tracer for
neuroanatomic studies in rat and monkey, J. Neurosci. Methods 45:35–40.
Bruce, K., and Grofova, I., 1992, Notes on a light and electron microscopic double-labeling
method combining anterograde tracing with Phaseolus vulgaris leucoagglutinin and retro-
grade tracing with cholera toxin subunit B, J. Neurosci. Methods 45:23–33.
Card, J. P., and Enquist, L. W., 1994, Use of pseudorabies virus for definition of synaptically
linked populations of neurons, In: Adolph, K. W. (ed.), Methods in Molecular Genetics, New
York: Academic Press, pp. 363–382.
Card, J. P., Enquist, L. W., and Moore, R. Y., 1999, Neuroinvasiveness of pseudorabies virus
injected intracerebrally is dependent on viral concentration and terminal field density, J.
Comp. Neurol. 407:438–452.
Card, J. P., Rinaman, L., Lynn, R. B., Lee, B.-H., Meade, R. P., Miselis, R. R., and Enquist,
L. W., 1993, Pseudorabies virus infection of the rat central nervous system: ultrastructural
characterization of viral replication, transport, and pathogenesis, J. Neurosci. 13:2515–2539.
Card, J. P., Rinaman, L., Schwaber, J. S., Miselis, R. R., Whealy, M. E., Robbins, A. K., and
Enquist, L. W., 1990, Neurotropic properties of pseudorabies virus: uptake and transneu-
ronal passage in the rat central nervous system, J. Neurosci. 10:1974–1994.
Card, J. P., Whealy, M. E., Robbins, A. K., Moore, R. Y., and Enquist, L. W., 1991, Two α-
herpesvirus strains are transported differently in the rodent visual system, Neuron 6:957–
969.
Carr, D. B., O’Donnell, P., Card, J. P., and Sesack, S. R., 1999, Dopamine terminals in the rat
prefrontal cortex synapse on pyramidal cells that project to the nucleus accumbens, J.
Neurosci. 19:11049–11060.
Carr, D. B., and Sesack, S. R., 2000a, Projections from the rat prefrontal cortex to the ventral
tegmental area: target specificity in the synaptic associations with mesoaccumbens and
mesocortical neurons, J. Neurosci. 20:3864–3873.
Carr, D. B., and Sesack, S. R., 2000b, GABA-containing neurons in the rat ventral tegmental
area project to the prefrontal cortex, Synapse 38:114–123.
Chagnaud, J. L., Mons, N., Tuffet, S., Grandier-Vazeilles, X., and Geffard, M., 1987, Monoclonal
antibodies against glutaraldehyde-conjugated dopamine, J. Neurochem. 49:487–494.
Chan, J., Aoki, C., and Pickel, V. M., 1990, Optimization of differential immunogold–silver and
peroxidase labeling with maintenance of ultrastructure in brain sections before plastic
embedding, J. Neurosci. Methods 33:113–127.
Chang, H. T., Kuo, H., Whittaker, J. A., and Cooper, N. G. F., 1990, Light and electron micro-
scopic analysis of projection neurons retrogradely labeled with Fluoro-Gold: notes on the
application of antibodies to Fluoro-Gold, J. Neurosci. Methods 35:31–37.
Charara, A., Smith, Y., and Parent, A., 1996, Glutamatergic inputs from the pedunculopontine
nucleus to midbrain dopaminergic neurons in primates: Phaseolus vulgaris leucoagglutinin
anterograde labeling combined with postembedding glutamate and GABA immunohisto-
chemistry, J. Comp. Neurol. 364:254–266.
Chen, S., and Aston-Jones, G., 1998, Axonal collateral–collateral transport of tract tracers in
brain neurons: false anterograde labelling and useful tool, Neuroscience 82:1151–1163.
Chen, L., Boyes, J., Yung, W. H., and Bolam, J. P., 2004, Subcellular localization of GABAB
receptor subunits in rat globus pallidus, J. Comp. Neurol. 474:340–352.
Chen, S., Yang, M., Miselis, R. R., and Aston-Jones, G., 1999, Characterization of transsynaptic
tracing with central application of pseudorabies virus, Brain Res. 838:171–183.
Conradi, N. G., 1981, Endogenous peroxidatic activity in the cerebral and cerebellar cortex of
normal adult rats, Acta Neuropathol. S7:3–6.
Cornea-Hebert, V., Riad, M., Wu, C., Singh, S. K., and Descarries, L., 1999, Cellular and subcel-
lular distribution of the serotonin 5-HT2A receptor in the central nervous system of the
adult rat, J. Comp. Neurol. 409:187–209.
Dado, R. J., Burstein, R., Cliffer, K. D., and Giesler, G. J., 1990, Evidence that Fluoro-Gold can
be transported avidly through fibers of passage, Brain Res. 553:329–333.
DeFalco, J., Tomishima, M., Liu, H., Zhao, C., Cai, X., Marth, J. D., Enquist, L., and Friedman, J.
M., 2001, Virus-assisted mapping of neural inputs to a feeding center in the hypothalamus,
Science 291:2608–2613.
DeMey, J., and Moeremans, M., 1986, The preparation of colloidal gold probes and their use as
a marker in electron microscopy, In: Koehler, J. K. (ed.), Advanced Techniques in Biological
Electron Microscopy III, New York: Springer-Verlag, pp. 229–271.
Dolleman-Van der Weel, M. J., Wouterlood, F. G., and Witter, M. P., 1994, Multiple antero-
grade tracing, combining Phaseolus vulgaris leucoagglutinin with rhodamine- and biotin-
conjugated dextran amine, J. Neurosci. Methods 51:9–21.
Doly, S., Madeira, A., Fischer, J., Brisorgueil, M. J., Daval, G., Bernard, R., Verge, D., and Conrath,
M., 2004, The 5-HT2A receptor is widely distributed in the rat spinal cord and mainly
localized at the plasma membrane of postsynaptic neurons, J. Comp. Neurol. 472:496–511.
Dube, D., and Pelletier, G., 1979, Effect of colchicine on the immunohistochemical localiza-
tion of somatostatin in the rat brain: light and electron microscopic studies, J. Histochem.
Cytochem. 27:1577–1581.
Dumartin, B., Caillé, I., Gonon, F., and Bloch, B., 1998, Internalization of D1 dopamine receptor
in striatal neurons in vivo as evidence of activation by dopamine agonists, J. Neurosci.
18:1650–1661.
Dumartin, B., Jaber, M., Gonon, F., Caron, M. G., Giros, B., and Bloch, B., 2000, Dopamine tone
regulates D1 receptor trafficking and delivery in striatal neurons in dopamine transporter-
deficit mice, Proc. Natl. Acad. Sci. 97:1879–1884.
Ericson, H., and Blomqvist, A., 1988, Tracing of neuronal connections with cholera toxin
subunit B: light and electron microscopic immunohistochemistry using monoclonal anti-
bodies, J. Neurosci. Methods 24:225–235.
Ferguson, S. M., Savchenko, V., Apparsundaram, S., Zwick, M., Wright, J., Heilman, C. J., Yi,
H., Levey, A. I., and Blakely, R. D., 2003, Vesicular localization and activity-dependent
trafficking of presynaptic choline transporters, J. Neurosci. 23:9697–9709.
Ford, B., Holmes, C. J., Mainville, L., and Jones, B. E., 1995, GABAergic neurons in the
rat pontomesencephalic tegmentum: codistribution with cholinergic and other tegmen-
tal neurons projecting to the posterior lateral hypothalamus, J. Comp. Neurol. 363:177–
196.
French, S. J., and Totterdell, S., 2002, Hippocampal and prefrontal cortical inputs monosynap-
tically converge with individual projection neurons of the nucleus accumbens, J. Comp.
Neurol. 446:151–165.
French, S. J., and Totterdell, S., 2003, Individual nucleus accumbens-projection neurons receive
both basolateral amygdala and ventral subicular afferents in rats, Neuroscience 119:19–31.
Friedrich, V. L., and Mugnaini, E., 1981, Preparation of neural tissues for electron microscopy,
In: Heimer, L., and Robards, M. J. (eds.), Neuroanatomical Tact-Tacing Methods, New York:
Plenum Press, pp. 345–375.
Gallyas, F., 1982, Suppression of the argyrophil III reaction by mercapto compounds (a pre-
requisite for the intensification of certain histochemical reactions by physical developers),
Acta Histochem. 70:99–105.
Garrett, W. T., McBride, R. L., Williams, J. K. J., and Feringa, E. R., 1991, Fluoro-Gold’s toxicity
makes it inferior to True Blue for long-term studies of dorsal root ganglion neurons and
motoneurons, Neurosci. Lett. 128:137–139.
Garzón, M., Vaughan, R. A., Uhl, G. R., Kuhar, M. J., and Pickel, V. M., 1999, Cholinergic axon
terminals in the ventral tegmental area target a subpopulation of neurons expressing low
levels of the dopamine transporter, J. Comp. Neurol. 410:197–210.
Gerfen, C. R., and Sawchenko, P. E., 1984, An anterograde neuroanatomical tracing method
that shows the detailed morphology of neurons, their axons and terminals: immunohis-
tochemical localization of an axonally transported plant lectin, Phaseolus vulgaris leucoag-
glutinin (PHA-L), Brain Res. 290:219–238.
Gerfen, C. R., Sawchenko, P. E., and Carlsen, J., 1989, The PHA-L anterograde axonal tracing
method, In: Heimer, L., and Zaborszky, L. (eds.), Neuroanatomical Tract-Tracing Methods 2:
Recent Progress, New York: Plenum Press, pp. 19–47.
Glass, M. J., Kruzich, P. J., Kreek, M. J., and Pickel, V. M., 2004, Decreased plasma membrane
targeting of NMDA-NR1 receptor subunit in dendrites of medial nucleus tractus solitarius
neurons in rat self-administering morphine, Synapse 53:191–201.
Graybiel, A. M., and Chesselet, M. F., 1984, Compartmental distribution of striatal cell bodies
expressing [Met]enkephalin-like immunoreactivity, Proc. Natl. Acad. Sci. 81:7980–7984.
Hajszan, T., and Zaborszky, L., 2002, Direct catecholaminergic-cholinergic interactions in the
basal forebrain. III. Adrenergic innervation of choline acetyltransferase-containing neu-
rons in the rat, J. Comp. Neurol. 449:141–157.
Hallanger, A. E., and Wainer, B. H., 1988, Ascending projections from the pedunculopontine
tegmental nucleus and the adjacent mesopontine tegmentum in the rat, J. Comp. Neurol.
274:483–515.
Hanson, J. E., and Smith, Y., 1999, Group I metabotropic glutamate receptors at GABAergic
synapses in monkeys, J. Neurosci. 19:6488–6496.
Heimer, L., and Robarts, M., 1981, Neuroanatomical Tract-Tracing Methods, New York: Plenum
Press, p. 567.
Heimer, L., and Zaborszky, L., 1989, Neuroanatomical Tract-Tracing Methods 2. Recent Progress, New
York: Plenum Press, p. 408.
Hemming, F. J., Mesguich, P., Morel, G., and Dubois, P. M., 1983, Cryoultramicrotomy versus
plastic embedding: comparative immunocytochemistry of rat anterior pituitary cells, J.
Microsc. 131:25–34.
Hillman, H., and Deutsch, K., 1978, Area changes in slices of rat brain during preparation for
histology or electron microscopy, J. Microsc. 114:77–84.
Howard, V., 1990, Stereological techniques in biological electron microscopy, In: Hawkes, P. W.,
and Valdrè, U. (eds.), Biophysical Electron Microscopy. Basic Concepts and Modern Techniques.
Bologna, Italy: Academic Press, pp. 479–508.
Hsu, S.-M., Raine, L., and Fanger, H., 1981, Use of avidin–biotin-peroxidase complex (ABC)
in immunoperoxidase techniques: a comparison between ABC and unlabeled antibody
(PAP) procedures, J. Histochem. Cytochem. 29:577–580.
Huang, J., and Pickel, V. M., 2002, Serotonin transporters (SERTs) within the rat nucleus of
the solitary tract: subcellular distribution and relation to 5HT2A receptors, J. Neurocytol.
31:667–679.
Irwin, S. A., Idupulapati, M., Gilbert, M. E., Harris, J. B., Chakravarti, A. B., Rogers, E. J., Crisos-
tomo, R. A., Larsen, B. P., Mehta, A., Alcantara, C. J., Patel, B., Swain, R. A., Weiler, I. J.,
Oostra, B. A., and Greenough, W. T., 2002, Dendritic spine and dendritic field character-
istics of layer V pyramidal neurons in the visual cortex of fragile-X knockout mice, Am. J.
Med. Genet. 111:140–146.
Jiang, X., Johnson, R. R., and Burkhalter, A., 1993, Visualization of dendritic morphology of
cortical projection neurons by retrograde axonal labeling, J. Neurosci. Methods 50:45–60.
Johansson, O., and Backman, J., 1983, Enhancement of immunoperoxidase staining using
osmium tetroxide, J. Neurosci. Methods 7:185–193.
Katona, I., Rancz, E. A., Acsady, L., Ledent, C., Mackie, K., Hajos, N., and Freund, T. F., 2001,
Distribution of CB1 cannabinoid receptors in the amygdala and their role in the control
of GABAergic transmission, J. Neurosci. 21:9506–9518.
Kelly, R. M., and Strick, P. L., 2000, Rabies as a transneuronal tracer of circuits in the central
nervous system, J. Neurosci. Methods 103:63–71.
Kolb, B., Gorny, G., Li, Y., Samaha, A. N., and Robinson, T. E., 2003, Amphetamine or cocaine
limits the ability of later experience to promote structural plasticity in the neocortex and
nucleus accumbens, Proc. Natl. Acad. Sci. 100:10523–10528.
Kulik, Á., Vida, I., Luján, R., Haas, C. A., López-Bendito, G., Shigemoto, R., and Frotscher, M.,
2003, Subcellular localization of metabotropic GABAB receptor subunits GABAB1a/b and
GABAB2 in the rat hippocampus, J. Neurosci. 23:11026–11035.
Leranth, C., and Pickel, V. M., 1989, Electron microscopic pre-embedding double immunos-
taining methods, In: Heimer, L., and Zaborsky, L. (eds.), Neuroanatomical Tract Tracing 2,
New York: Plenum Press, pp. 129–172.
Liposits, Z., Setalo, G., and Flerko, B., 1984, Application of the silver-gold intensified 3,3 -
diaminobenzidine chromogen to the light and electron microscopic detection of the
luteinizing hormone-releasing hormone system of the rat brain, Neuroscience 13:513–
525.
Llewellyn-Smith, I. J., Minson, J. B., Wright, A. P., and Hodgson, A. J., 1990, Cholera toxin
B-gold, a retrograde tracer that can be used in light and electron microscopic immunocy-
tochemical studies, J. Comp. Neurol. 294:179–191.
Llewellyn-Smith, I. J., Pilowsky, P. M., and Minson, J. B., 1993, The tungstate-stabilized tetram-
ethylbenzidine reaction for light and electron microscopic immunocytohemistry and for
revealing biocytin filled neurons, J. Neurosci. Methods 46:27–40.
Luedtke, R. R., Griffin, S. A., Conroy, S. S., Jin, X., Pinto, A., and Sesack, S. R., 1999, Immunoblot
and immunohistochemical comparison of murine monoclonal antibodies specific for the
rat D1a and D1b dopamine receptor subtypes, J. Neuroimmunol. 101:170–187.
Marfurt, C. F., Turner, D. F., and Adams, C. E., 1988, Stabilization of tetramethylbenzidine
(TMB) reaction product at the electron microscopic level by ammonium molybdate, J.
Neurosci. Methods 25:215–223.
Masson, J., Riad, M., Chaudhry, F., Darmon, M., Aidouni, Z., Conrath, M., Giros, B., Hamon,
M., Storm-Mathisen, J., Descarries, L., and El Mestikaw, S., 1999, Unexpected localization
of the Na+-dependent-like orphan transporter, Rxt1, on synaptic vesicles in the rat central
nervous system, Eur. J. Neurosci. 11:1349–1361.
Mayhew, T. M., 1992, A review of recent advances in stereology for quantifying neural structure,
J. Neurocytol. 21:313–328.
McLean, J. H., Shipley, M. T., and Bernstein, D. I., 1989, Golgi-like, transneuronal retrograde
labelling with CNS injections of herpes simplex virus type 1, Brain Res. Bull. 22:867–881.
Melchitzky, D. S., González-Burgos, B., Barrionuevo, G., and Lewis, D. A., 2001, Synaptic targets
of the intrinsic axon collaterals of supragranular pyramidal neurons in monkey prefrontal
cortex, J. Comp. Neurol. 430:209–221.
Mi, Z. P., Jiang, P., Weng, W. L. F. P., Narayanan, V., and Lagenaur, C. F., 2000, Expression of a
synapse-associated membrane protein, P84/SHPS-1, and its ligand, IAP/CD47, in mouse
retina, J. Comp. Neurol. 416:335–344.
Miner, L. A. H., Backstrom, J. R., Sanders-Bush, E., and Sesack, S. R., 2003a. Ultrastructural
localization of serotonin2A receptors in the middle layers of the rat prelimbic prefrontal
cortex, Neuroscience 116:107–117.
Miner, L. A. H., Benmansour, S., Moore, F. W., Blakely, R. D., Morilak, D. A., Frazer, A., and
Sesack, S. R., 2004, Chronic treatment with a selective serotonin reuptake inhibitor reduces
the total and plasmalemmal serotonin transporter immunoreactivity in axon terminals
within the rat prefrontal cortex, Soc. Neurosci. Abstr. 29:54.55.
Miner, L. A. H., Moore, F. W., Jedema, H. P., Grace, A. A., and Sesack, S. R., 2003b. Ultrastructural
localization of the norepinephrine transporter in the prefrontal cortex of chronically
stressed rats, Soc. Neurosci. Abstr. 28:506.505.
Miner, L. A. H., Schroeter, S., Blakely, R. D., and Sesack, S. R., 2000, Ultrastructural localization
of the serotonin transporter in superficial and deep layers of the rat prefrontal cortex and
its spatial relationship to dopamine terminals, J. Comp. Neurol. 427:220–234.
Miner, L. A. H., Schroeter, S., Blakely, R. D., and Sesack, S. R., 2003c. Ultrastructural localization
of the norepinephrine transporter in superficial and deep layers of the rat prelimbic
prefrontal cortex and its spatial relationship to probable dopamine terminals, J. Comp.
Neurol. 466:478–494.
Monti-Graziadei, A. G., and Berkley, K. J., 1991, Effects of colchicine on retrogradely-
transported WGA-HRP, Brain Res. 565:162–166.
Moore, R. Y., 1981, Methods for selective, restricted lesion placement in the central nervous
system, In: Heimer, L., and Robards, M. J. (eds.), Neuroanatomical Tract-Tracing Methods,
Mugnaini, E., and Friedrich, V. L., Jr., 1981, Electron microscopy: identification and study
of normal and degenerating neural elements by electron microscopy, In: Heimer, L.,
and Robards, M. J. (eds.), Neuroanatomical Tract-Tracing Methods, New York: Plenum Press,
pp. 377–406.
Naumann, T., Härtig, W., and Frotscher, M., 2000, Retrograde tracing with Fluoro-Gold: differ-
ent methods of tracer detection at the ultrastructural level and neurodegenerative changes
of back-filled neurons in long-term studies, J. Neurosci. Methods 103:11–21.
Norgren, R. B. J., and Lehman, M. N., 1989, A double-label pre-embedding immunoperoxidase
technique for electron microscopy using diaminobenzidine and tetramethylbenzidine as
markers, J. Histochem. Cytochem. 37:1283–1289.
Novikoff, A., Novikoff, P., Quintana, N., and Davis, C., 1972, Diffusion artifacts in 3,3 -
diaminobenzidine cytochemistry, J. Histochem. Cytochem. 20:745–749.
Novikova, L., Novikov, L., and Kellerth, J. O., 1997, Persistent neuronal labeling by retrograde
fluorescent tracers: a comparison between Fast Blue, Fluoro-Gold and various dextran
conjugates, J. Neurosci. Methods 74:9–15.
Nusser, Z., Ahmad, Z., Tretter, V., Fuchs, K., Wisden, W., Sieghart, W., and Somogyi, P., 1999,
Alterations in the expression of GABAA receptor subunits in the cerebellar granule cells
after the disruption of the alpha6 subunit gene, Eur. J. Neurosci. 11:1685–1697.
Nusser, Z., Roberts, J. D., Baude, A., Richards, J. G., Sieghart, W., and Somogyi, P., 1995,
Immunocytochemical localization of the α1 and β2/3 subunits of the GABAA receptor in
relation to specific GABAergic synapses in the dentate gyrus, Eur. J. Neurosci. 7:630–646.
Oades, R. D., and Halliday, G. M., 1987, Ventral tegmental (A10) system: neurobiology. 1.
Anatomy and connectivity, Brain Res. Rev. 12:117–165.
Oakman, S. C., Faris, P. L., Kerr, P. E., Cozzari, C., and Hartman, B. K., 1995, Distribution of pon-
tomesencephalic cholinergic neurons projecting to substantia nigra differs significantly
from those projecting to ventral tegmental area, J. Neurosci. 15:5859–5869.
Omelchenko, N., and Sesack, S. R., 2005, Laterodorsal tegmental projections to identified cell
populations in the rat ventral tegmental area, J. Comp. Neurol. 483:217–235.
Omelchenko, N., and Sesack, S. R., 2005b, cholinergic axons in the rat ventral tegmental are
synapse preferentially onto mesoaccumbens dopamine neurons, J. Comp. Neurol.
Ordronneau, P., Lindström, P. B.-M., and Petrusz, P., 1981, Four unlabeled antibody bridge
techniques: a comparison, J. Histochem. Cytochem. 29:1397–1404.
Paspalas, C. D., and Goldman-Rakic, P. S., 2004, Microdomains for dopamine volume neuro-
transmission in primate prefrontal cortex, J. Neurosci. 24:5292–5300.
Paulson, J. C., and McClure, W. O., 1975, Inhibition of axoplasmic transport by colchicine,
podophyllotoxin, and vinblastine: an effect on microtubules, Ann. N. Y. Acad. Sci. 253:517–
527.
Phend, K. D., Rustioni, A., and Weinberg, R. J., 1995, An osmium-free method of epon embed-
ment that preserves both ultrastructure and antigenicity for post-embedding immunocy-
tochemistry, J. Histochem. Cytochem. 43:283–292.
Phillipson, O. T., 1979, Afferent projections to the ventral tegmental area of Tsai and interfas-
cicular nucleus: a horseradish peroxidase study in the rat, J. Comp. Neurol. 187:117–144.
Pickel, V. M., 1981, Immunocytochemical methods, In: Heimer, L., and RoBards, M. (eds.),
Neuroanatomical Tract-Tracing Methods, New York: Plenum Press, pp. 483–509.
Pickel, V. M., and Chan, J., 1999, Ultrastructural localization of the serotonin transporter in
limbic and motor compartments of the nucleus accumbens, J. Neurosci. 19:7356–7366.
Pickel, V. M., Chan, J., and Aoki, C., 1993, Electron microscopic immunocytochemical la-
belling of endogenous and/or transported antigens in rat brain using silver-intensified
one-nanometre colloidal gold, In: Cuello, A. (ed.), Immunohistochemistry II, Chichester:
John Wiley & Sons, pp. 265–280.
Pickel, V. M., Chan, J., Kash, T. L., Rodriguez, J. J., and Mackie, K., 2004, Compartment-specific
localization of cannabinoid (CB1) and mu-opiod receptors in rat nucleus accumbens,
Neuroscience 127:101–112.
Pickel, V. M., and Milner, T. A., 1989, Interchangeable uses of autoradiographic and peroxidase
markers for electron microscopic detection of neuronal pathways and transmitter-related
antigens in single sections, In: Heimer, L., and Zaborszky, L. (eds.), Neuroanatomical Tract-
Tracing Methods 2: Recent Progress, New York: Plenum Press, pp. 97–127.
Pieribone, V. A., and Aston-Jones, G., 1988, The iontophoretic application of Fluoro-Gold for
the study of afferents to deep brain nuclei, Brain Res. 475:259–271.
Pinto, A., Jankowski, M., and Sesack, S. R., 2003, Projections from the paraventricular nucleus
of the thalamus to the rat prefrontal cortex and nucleus accumbens shell: ultrastructural
characteristics and spatial relationships with dopamine afferent, J. Comp. Neurol. 459:142–
155.
Pinto, A., and Sesack, S. R., 1998, Paraventricular thalamic afferents to the rat prefrontal
cortex and nucleus accumbens shell: synaptic targets and relation to dopamine afferents,
Soc. Neurosci. Abstr. 24:1595.
Pirnik, Z., Mikkelsen, J. D., and Kiss, A., 2003, Fos induction in the rat deep cerebellar and
vestibular nuclei following central administration of colchicine: a qualitative and quanti-
tative time-course study, Brain Res. Bull. 61:63–72.
Rajakumar, N., Elisevich, K., and Flumerfelt, B. A., 1993, Biotinylated dextran: a versatile an-
terograde and retrograde neuronal tracer, Brain Res. 607:47–53.
Ravary, A., Muzerelle, A., Darmon, M., Murphy, D. L., Moessner, R., Lesch, K. P., and Gaspar,
P., 2001, Abnormal trafficking and subcellular localization of an N-terminally truncated
serotonin transporter protein, Eur. J. Neurosci. 13:1349–1362.
Reiner, A., Veenman, C. L., Medina, L., Jiao, Y., Del Mar, N., and Honig, M. G., 2000, Pathway
tracing using biotinylated dextran amines, J. Neurosci. Methods 103:23–37.
Rho, J. H., and Swanson, L. W., 1989, A morphometric analysis of functionally defined sub-
populations of neurons in the paraventricular nucleus of the rat with observations on the
effects of colchicine, J. Neurosci. 9:1375–1388.
Riad, M., Garcia, S., Watkins, K. C., Jodoin, N., Doucet, E., Langlois, X., El Mestikaw, S., Hamon,
M., and Descarries, L., 2000, Somatodendritic localization of the 5-HT1A and preterminal
axonal localization of 5-HT1B serotonin receptors in the adult rat brain, J. Comp. Neurol.
417:181–194.
Riad, M., Zimmer, L., Rbah, L., Watkins, K. C., Hamon, M., and Descarries, L., 2004, Acute
treatment with the antidepressant fluoxetine internalizes 5-HT1A autoreceptors and re-
duces the in vivo binding of the PET radioligand[18F]MPFF in the nucleus raphe dorsalis
of rat, J. Neurosci. 24:5420–5426.
Ribak, C. E., Vaughn, J. E., and Saito, K., 1978, Immunocytochemical localization of glutamic
acid decarboxylase in neuronal somata following colchicine inhibition of axonal transport,
Brain Res. 140:315–332.
Rosene, D. L., Roy, N. J., and Davis, B. J., 1986, A cryoprotection method that facilitates cutting
frozen sections of whole monkey brains for histological and histochemical processing
without freezing artifact, J. Histochem. Cytochem. 34:1301–1315.
Rye, D. B., Saper, C. B., and Wainer, B. H., 1984, Stabilization of the tetramethylbenzidine
(TMB) reaction product: application for retrograde and anterograde tracing, and combi-
nation with immunohistochemistry, J. Histochem. Cytochem. 32:1145–1153.
Saper, C. B., 2003, Magic peptides, magic antibodies: guidelines for appropriate controls for
immunohistochemistry, J. Comp. Neurol. 465:161–163.
Schmued, L. C., and Fallon, J. H., 1986, Fluoro-Gold: a new fluorescent retrograde axonal
tracer with numerous unique properties, Brain Res. 377:147–154.
Schmued, L. C., Kyriakidis, K., Fallon, J. H., and Ribak, C. E., 1989, Neurons containing ret-
rogradely transported Fluoro-Gold exhibit a variety of lysosomal profiles: a combined
brightfield, fluorescence, and electron microscopic study, J. Neurocytol. 18:333–343.
Sesack, S. R., and Pickel, V. M., 1992, Prefrontal cortical efferents in the rat synapse on unla-
beled neuronal targets of catecholamine terminals in the nucleus accumbens septi and
on dopamine neurons in the ventral tegmental area, J. Comp. Neurol. 320:145–160.
Sesack, S. R., and Snyder, C. L., 1995, Cellular and subcellular localization of syntaxin-like
immunoreactivity in the rat striatum and cortex, Neuroscience 67:993–1007.
Shu, S. Y., and Peterson, G. M., 1988, Anterograde and retrograde axonal transport of Phaseolus
vulgaris leucoagglutinin (PHA-L) from the globus pallidus to the striatum of the rat, J.
Skirboll, L. R., Thor, K., Helke, C., Hökfelt, T., Robertson, B., and Long, R., 1989, Use of
retrograde fluorescent tracers in combination with immunohistochemical methods, In:
Heimer, L., and Zaborszky, L. (eds.), Neuroanatomical Tract-Tracing Methods 2: Recent Progress,
Smiley, J. F., and Goldman-Rakic, P. S., 1993, Silver-enhanced diaminobenzidine-sulfide
(SEDS): a technique for high-resolution immunoelectron microscopy demonstrated with
monoamine immunoreactivity in monkey cerebral cortex and caudate, J. Histochem. Cy-
tochem. 41:1393–1404.
Smith, Y., Bennett, B. D., Bolam, J. P., Parent, A., and Sadikot, A. F., 1994, Synaptic relation-
ship between dopaminergic afferents and cortical or thalamic input in the sensorimotor
territory of the striatum in monkey, J. Comp. Neurol. 344:1–19.
Smith, Y., Charara, A., and Parent, A., 1996, Synaptic innervation of midbrain dopaminergic
neurons by glutamate-enriched terminals in the squirrel monkey, J. Comp. Neurol. 364:231–
253.
Srebro, Z., 1972, Ultrastructural localization of peroxidase activity in Gomori-positive glia, Acta
Anat. 83:388–397.
Sternberger, L. A., 1974, Immunocytochemistry. Englewood Cliffs, NJ: Prentice-Hall.
Steward, O., 1981, Horseradish peroxidase and fluorescent substances and their combination
with other techniques, In: Heimer, L., and Robards, M. J. (eds.), Neuroanatomical Tract-
Tracing Methods, New York: Plenum Press, pp. 279–310.
Strick, P. L., and Card, J. P., 1992, Transneuronal mapping of neural circuits with alpha her-
pesviruses, In: Bolam, J. (ed.), Experimental Neuroanatomy: A Practical Approach, New York:
IRL Press, pp. 81–101.
Svensson, B. A., Rastad, J., and Westman, J., 1984, Endogenous peroxidase-like activity in the
feline dorsal column nuclei and spinal cord, Exp. Brain Res. 55:325–332.
Swanson, L. W., 1982, The projections of the ventral tegmental area and adjacent regions: a
combined fluorescent retrograde tracer and immunofluorescence study in the rat, Brain
Res. Bull. 9:321–353.
Van Bockstaele, E. J., and Pickel, V. M., 1995, GABA-containing neurons in the ventral tegmental
area project to the nucleus accumbens in rat brain, Brain Res. 682:215–221.
Van Bockstaele, E. J., Wright, A. M., Cestari, D. M., and Pickel, V. M., 1994, Immunolabeling of
retrogradely transported Fluoro-Gold: sensitivity and application to ultrastructural analysis
of transmitter-specific mesolimbic circuitry, J. Neurosci. Methods 55:65–78.
Van Haeften, T., and Wouterlood, F. G., 2000, Neuroanatomical tracing at high resolution, J.
Veenman, C. L., Reiner, A., and Honig, M. G., 1992, Biotinylated dextran amine as an
anterograde tracer for single- and double-labeling studies, J. Neurosci. Methods 41:239–
254.
Veznedaroglu, E., and Milner, T. E., 1992, Elimination of artifactual labeling of hippocampal
mossy fibers seen following pre-embedding immunogold–silver technique by pretreatment
with zinc chelator, Microsc. Res. Tech. 23:100–101.
von Bartheld, C., 2002, Counting particles in tissue sections: choices of methods and importance
of calibration to minimize biases, Histol. Histopathol. 17:639–648.
Warr, W. B., de Olmos, J. S., and Heimer, L., 1981, Horseradish peroxidase: the basic procedure,
In: Heimer, L., and Robards, M. J. (eds.), Neuroanatomical Tract-Tracing Methods, New York:
Wong, H. K., Liu, X. B., Matos, M. F., Chan, S. F., Perez-Otano, I., Boysen, M., Cui, J., Nakanishi,
N., Trimmer, J. S., Jones, E. G., Lipton, S. A., and Sucher, N. J., 2002, Temporal and regional
expression of NMDA receptor subunit NR3A in the mammalian brain, J. Comp. Neurol.
450:303–317.
Wouterlood, F. G., Goede, P. H., and Groenewegen, H. J., 1990, The in situ detectability of the
neuroanatomical tracer Phaseolus vulgaris-leucoagglutinin (PHA-L), J. Chem. Neuroanat.
3:11–18.
Wouterlood, F. G., and Groenewegen, H. J., 1985, Neuroanatomical tracing by use of Phaseolus
vulgaris leucoagglutnin (PHA-L): electron microscopy of PHA-L filled neuronal somata,
dendrites, axons and axon terminals, Brain Res. 326:188–191.
Wouterlood, F. G., and Jorritsma-Byham, B., 1993, The anterograde neuroanatomical tracer
biotinylated dextran-amine: comparison with the tracer Phaseolus vulgaris leucoagglutinin
in preparations for electron microscopy, J. Neurosci. Methods 48:75–87.
Yagi, T., Terada, N., Baba, T., and Ohno, S., 2002, Localization of endogenous biotin-containing
proteins in mouse Bergmann glial cells, Histochem. J. 34:567–572.
Yan, X. X., and Ribak, C. E., 1999, Alteration of GABA transporter expression in the rat cerebral
cortex following needle puncture and colchicine injection, Brain Res. 816:317–328.
Yi, H., Leunissen, J. L. M., Shi, G.-M., Gutekunst, C.-A., and Hersch, S. M., 2001, A novel
procedure for pre-embedding double immunogold–silver labeling at the ultrastructural
level, J. Histochem. Cytochem. 49:279–284.
Zaborszky, L., and Heimer, L., 1989, Combinations of tracer techniques, especially HRP and
PHA-L, with transmitter identification for correlated light and electron microscopic stud-
ies, In: Heimer, L., and Zaborszky, L. (eds.), Neuroanatomical Tract-Tracing Methods 2: Recent
Progress, New York: Plenum Press, pp. 49–96.
Zaborszky, L., Léránth, C., and Palkovits, M., 1979, Light and electron microscopic identifica-
tion of monoaminergic terminals in the central nervous system, Brain Res. Bull. 4:99–117.
Zaborszky, L., Rosin, D. L., and Kiss, J., 2004, Alpha-adrenergic receptor (α2A) is colocalized
in basal forebrain cholinergic neurons: a light and electron microscopic double immuno-
labeling study, J. Neurocytol. 33:265–276.
Zhao, S., Maxwell, S., Jimenez-Beristain, A., Vives, J., Kuehner, E., Zhao, J., O’Brien, C., de
Felipe, C., Semina, E., and Li, M., 2004, Generation of embryonic stem cells and transgenic
mice expressing green fluorescence protein in midbrain dopaminergic neurons, Eur. J.
Neurosci. 19:1133–1140.
Zhou, M., and Grofova, I., 1995, The use of peroxidase substrate Vector VIP in electron micro-
scopic single and double antigen localization, J. Neurosci. Methods 62:149–158.
3
Postembedding Immunogold
Cytochemistry of Membrane
Molecules and Amino Acid
Transmitters in the Central
Nervous System
THOMAS MISJE MATHIISEN, ERLEND
ARNULF NAGELHUS, BAHAREH JOULEH,
REIDUN TORP, DIDRIK SØLIE
FRYDENLUND, MARIA-NIKI MYLONAKOU,
MAHMOOD AMIRY-MOGHADDAM,
LUCIENE COVOLAN, JO KRISTIAN UTVIK,
BJØRG RIBER, KAREN MARIE GUJORD,
JORUNN KNUTSEN, ØIVIND SKARE,
PETTER LAAKE, SVEND DAVANGER,
FINN-MOGENS HAUG, ERIC RINVIK, and
OLE PETTER OTTERSEN
INTRODUCTION
RESOLUTION
QUANTITATION
THOMAS MISJE MATHIISEN, ERLEND ARNULF NAGELHUS, BAHAREH

JOULEH, REIDUN TORP, DIDRIK SØLIE FRYDENLUND, MARIA-NIKI
MYLONAKOU, MAHMOOD AMIRY-MOGHADDAM, LUCIENE COVOLAN, JO
KRISTIAN UTVIK, BJØRG RIBER, KAREN MARIE GUJORD, JORUNN KNUTSEN,
ØYVIND SKARE, PETTER LAAKE, SVEND DAVANGER, FINN-MOGENS HAUG,
ERIC RINVIK, AND OLE PETTER OTTERSEN • Centre for Molecular Biology
and Neuroscience, and Nordic Centre for Research on Water Imbalance Related Disorders
(WIRED), Institute of Basic Medical Sciences, University of Oslo, PO Box 1105 Blindern,
N-0317 Oslo, Norway
72
POSTEMBEDDING IMMUNOGOLD CYTOCHEMISTRY 73
CONTROLS
APPLICATIONS
POSTEMBEDDING PROCEDURES
Fixation
Dehydration and Embedding
Immunoincubation
APPENDIX I: A POSTEMBEDDING IMMUNOGOLD PROCEDURE FOR
MEMBRANE PROTEINS
Tissue Preparation
Immunoincubation
Solutions
Protocol for Postembedding Immunogold Labeling Using
Ultrasmall Gold Particles Coupled to Fab Fragments (Secondary
Antibodies) and Silver Intensification
APPENDIX II: COUNTING IMMUNOPARTICLES BY DIGITAL IMAGE
ANALYSIS
REFERENCES
Abstract: This chapter deals with procedures for postembedding labeling of brain
sections embedded in epoxy or methacrylate resins and focuses on protocols
that are based on freeze substitution of chemically fixed tissue. When optimized
for the target epitope, such protocols offer a high labeling efficiency and allow
simultaneous visualization of several antigens by use of different-sized gold particles.
Postembedding labeling can be combined with anterograde tracing, permitting the
identification of transmitter and postsynaptic receptors of identified axons. By use
of tailor-made model systems, antibody selectivity can be monitored in a quantitative
manner and under conditions that are representative of the immunocytochemical
procedure. Such model systems also allow the generation of calibration curves for
assessment of the cellular and subcellular concentration of soluble antigens. When
used in conjunction with computer programs for automated acquisition and analysis
of gold particles, the postembedding immunogold procedure provides an accurate
representation of the cellular and subcellular distribution of proteins and small com-
pounds such as transmitter amino acids. The present chapter provides a quantitative
analysis and critical discussion of how changes in incubation parameters influence
the labeling intensity. Postembedding immunogold cytochemistry stands out as a
powerful technique for analysis of the chemical architecture of the central nervous
system and has proved useful for investigating disease processes at the molecular
level.
Keywords: aquaporins, glutamate, glutamate receptors, quantitation, resolution,
specificity testing
I. INTRODUCTION
The ultimate goal in immunocytochemistry is to be able to determine the

exact number and position of a given target molecule in a biological tissue.
74 THOMAS MISJE MATHIISEN et al.
A priori, this requires access to monospecific antibodies that bind with a 1:1
stoichiometry to the target antigen, and a reporter system that allows accu-
rate localization and quantitation of the primary antibodies. These are ideal
conditions that cannot be met in practice. However, they can be approached
by use of markers that are amenable to quantitative electron microscopic
analysis.
Colloidal gold particles (Faulk and Taylor, 1971; Roth, 1996; van den Pol,
1989) have proven to be the most versatile markers for this purpose. They are
electron dense, allowing easy identification and quantitation in the electron
microscope, and can be prepared in many different sizes, permitting simul-
taneous detection of several different antigens. Most importantly, colloidal
gold particles can be coupled directly to the primary or secondary antibody
so as to afford a close spatial relation to the target antigen. These features
set immunogold cytochemistry apart from the peroxidase–antiperoxidase
method and other enzyme-based immunocytochemical techniques. The lat-
ter techniques typically rely on the analysis of an electron-dense reaction
product that is difficult to quantify and that may diffuse away from the site
of formation.
Colloidal gold particles may, in principle, be applied in two different ways:
in preembedding or postembedding mode. In the preembedding mode, the
antibodies and immunogold reagents are applied to permeabilized tissue
that is subsequently embedded in a resin suitable for electron microscopic
analysis. In the postembedding mode, the immunoreagents are applied di-
rectly onto ultrathin sections of resin-embedded tissue or cells. The latter
approach allows immunodetection only of those antigen molecules that are
exposed at the surface of the section. This implies that the proportion of
antigen molecules that is available for antibody binding is severely restricted
when compared with the preembedding mode.
So why use the postembedding mode? The major advantage offered
by the postembedding mode is that each antigen molecule that occurs
at the surface of the section should stand the same chance of being im-
munodetected, regardless of its cellular or subcellular localization. This
contrasts with the situation in the preembedding mode, where diffusion
barriers may constrain the labeling and distort the relationship between
antigen concentration and gold particle density. Thus, for the purpose of
quantitation, the postembedding procedure is generally considered as the
superior of the two modes of immunogold cytochemistry. Preembedding
procedures have their own set of advantages that will not be considered
here (see Sesack et al., this volume). We also need to emphasize that cryo-
electron microscopy is outside the scope of the present chapter, which deals
exclusively with postembedding immunogold labeling of resin-embedded
sections. The chapter is focused on experience gained in our own labora-
tory and is not intended to provide a balanced overview of the historical
development of the technique. A description of the pioneering work is
found elsewhere (Griffiths, 1993; Maunsbach and Afzelius, 1999; Roth,
1996).
II. RESOLUTION
As stated in the Introduction, the resolution of the postembedding im-

munogold technique exceeds by far the resolution offered by enzyme-based
techniques. Thus, with the former technique, the marker is a well-defined
particle that is attached to the antigen through an antibody bridge, rather
than a reaction product that may be deposited at a distance from its site of
formation.
Indeed, it is the length of the antibody bridge and the dimension of
the colloidal particle that restrict the resolution of the postembedding im-
munogold procedure. Theoretically, the distance between the epitope and
the center of the gold particle should correspond to the radius of the parti-
cle plus the diameters of the interposed immunoglobulins (IgGs) (Fig. 3.1).
Using 15-nm gold particles and a primary and secondary IgG (each with an
efficient diameter of ∼8 nm), this theoretical distance should be ∼23 nm.
Obviously, the distance will be shorter if colloidal gold is coupled to the
primary antibody directly or by way of a secondary Fab complex rather than
a secondary IgG.
The theoretical prediction as to lateral resolution is borne out by ex-
periments based on the use of tailor-made model antigens (Fig. 3.2). The
model antigens were embedded in the same resin as that used for the tis-
sue, ensuring identical conditions. Further, the antigens were prepared to
form discrete bodies with a distinct demarcation from the surrounding
resin. Hence, the distance between the margin of the body and the centers
Figure 3.1. Simplified diagram of a two-step, postembedding immunogold proce-

dure. Triangles represent the antigen against which the primary antibody (A) was
raised. The secondary antibody (B) is coupled to a colloidal gold particle (Au). The
radius (r ) of the gold particle is 7.5 nm. This adds up to a maximum distance of
about 23 nm between the epitope and the gold particle center (given an effective
IgG diameter of 8 nm). (From Ottersen, 1989a.)
Figure 3.2. Lateral resolution of current immunogold procedure (15-nm gold par-
ticles). Values along the x-axis in B denote distance from centers of gold particles
to the margin of antigen-containing bodies (A). Background level of labeling is
reached ∼28 nm off the bodies. The data were based on the analysis of 30 bodies
containing glutaraldehyde-fixed L-aspartate as a model antigen. Scale bar: 0.3 µm.
(From Matsubara et al., 1996.)
of the gold particles should be representative of the lateral resolution of

the postembedding immunogold procedure. The gold particle density was
found to reach background level at ∼28 nm off the margin of the test body
(Fig. 3.2B) in good agreement with the theoretical prediction.
Obviously, the above-mentioned theoretically and experimentally deter-
mined values represent the maximum distance between an epitope and the
respective gold particle. In practice, many particles are likely to end up
closer to the epitope, due to a restricted rotational freedom. It must also be
remembered that the projected distance between the gold particle and the
epitope may be considerably shorter than the real distance.
This notwithstanding, the fact that the size of the antibody bridge signif-
icantly limits the resolution of the postembedding immunogold technique
has practical consequences in several experimental settings. Two examples
will be used to illustrate this. These examples will also show how the ef-
fective resolution can be improved by resorting to tailor-made statistical
procedures.
The first example is representative of a common problem in neurobi-
ology: the need to distinguish between two plasma membranes that are
closely apposed to one another. A close membrane apposition is typical of
central synapses where the pre- and postsynaptic membranes are separated
by a synaptic cleft of ∼20 nm. This distance is less than the theoretical and
experimental maximum values between the epitope and the gold particle
center. In other words, a gold particle overlying the presynaptic membrane
might reflect antibody binding to an epitope in the postsynaptic membrane,
Figure 3.3. Comparison of postembedding (A) and preembedding (B) labeling of

AQP4 in perivascular astrocyte end feet membranes (double arrows indicate width
of end foot). End, endothelial cell. (A) In the postembedding mode, gold particles
are deposited on either side of the membrane—the maximum distance from the
membrane reflecting the size of the antibody bridge (cf. Figs. 3.1 and 3.2). (B) In
preembedding mode (ultrasmall gold particles enhanced by silver), all particles end
up at the cytoplasmic aspect, due to the constraints imposed by the membrane (the
epitope is located at the intracellular tail of AQP4). Note variable size and confluence
of particles, typical of the preembedding mode. (A, B: From Nielsen et al., 1997.)
(C) Analysis of AQP4 immunogold labeling by recording the distribution of gold
particles along an axis perpendicular to the labeled plasma membrane (postembed-
ding labeling as in A). The ordinate indicates number of gold particles per bin (bin
width, 5 nm). The peak coincides with the plasma membrane (0 corresponds to
midpoint of membrane) and the particle density approaches background level at
∼50 nm from the membrane (inside negative). This section was labeled from both
sides, explaining why some particles are located further off than the theoretically and
experimentally determined maximum distance between epitope and gold particle
(see text). (C: From Nagelhus et al., 1998.)
and vice versa. So how is it possible to determine whether a given antigen is

localized pre- and/or postsynaptically?
This question can be resolved by recording the distribution of gold
particles along an axis perpendicular to the synaptic membranes (here
defined as z-axis). Depending, i.a., on the rotational freedom of the epi-
tope and the number of gold particles recorded, the average position of
an antigen along this axis can be determined with a precision of ∼1 nm,
Figure 3.3. (cont).
defined by the standard error of the z-value of the peak particle density (Fig.
3.3; also see Nagelhus et al., 1998, 2004). By this approach, it was possible to
demonstrate that the BK potassium channel is expressed in presynaptic but
not in postsynaptic membranes of hippocampal synapses (Hu et al., 2001;
Fig. 3.4). The same approach was exploited recently to distinguish between
closely apposed membranes in the olfactory bulb (Panzanelli et al., 2004)
and has also been used, in a different context, to identify the relative posi-
tions of molecules engaged in glutamate receptor complexes (Valtschanoff
and Weinberg, 2001). The common practice of labeling both sides of the
section may decrease the precision of this approach, as the intersections of
a membrane with the two surfaces of the 50- to 100-nm-thick tissue section
are rarely superimposed in the image (Fig. 3.5).
The second example of a biological problem that requires due attention
to the size of the antibody bridge relates to the analysis of synaptic vesicles
(particularly the small, clear vesicles that have a diameter of ∼50 nm). The
question was whether glutamate is enriched in the synaptic vesicles of gran-
ule cell dendritic spines of the olfactory bulb (Didier et al., 2001). These
spines are presynaptic to the mitral cell dendrites and display a high density
of gold particles signaling glutamate. But this signal could reflect metabolic
glutamate, rather than a vesicular pool of transmitter glutamate. Due to the
small dimensions of the clear synaptic vesicles, one cannot attach signifi-
cance to individual gold particles: even a particle located at the center of a
vesicle could theoretically depend on an epitope external to the vesicle in
question.
To circumvent this problem, measurements were made of the intercenter
distances between each gold particle and the nearest synaptic vesicle. It
turned out that short distances were overrepresented compared with ran-
dom distributions of gold particles, supporting the idea that glutamate is
associated with synaptic vesicles (Didier et al., 2001).
The two examples discussed above show that statistical analyses of
large numbers of gold particles can partly compensate for the inaccuracy
Figure 3.4. Analysis of gold particle distribution in closely apposed synaptic mem-
branes. (A) Electron micrograph showing the distribution of BK channels (10-nm
particles) and NMDA receptors (NMDARs; 15-nm particles) in double-labeled sec-
tions from the stratum radiatum of CA1. t, terminal; arrowheads indicate extent of
postsynaptic density. (B) To determine whether the two epitopes were associated
with the pre- and/or postsynaptic membranes, the vertical distribution of particles
was analyzed by the approach described in Fig. 3.3. The peak density of particles
coincided with the presynaptic membrane in the case of BK channels and with the
postsynaptic membrane in the case of NMDARs. The dimensions of the synaptic cleft
and postsynaptic density are indicated below the abscissa. (From Hu et al., 2001.)
introduced by the antibody bridge. Other factors that affect the effective
resolution of the postembedding technique are discussed in Appendix II.
How could resolution be further improved? There is a marginal gain by
coupling small gold particles directly to the primary antibodies (or Fab
fragments of primary antibodies), rather than to the secondary ones. A sub-
stantial increase in resolution would be achieved by visualizing (by negative
staining) the antibody bridge between the epitope and the gold particle.
But an even larger step toward the “ultimate goal” of defining the precise
Figure 3.5. Distribution of gold particles signaling δ2-glutamate receptors at synapses

between parallel fibers (Pf ) and Purkinje cell spines (s) of rat cerebellum. The
synapse at top center is obliquely cut, the two rows of particles representing recep-
tors exposed at opposite surfaces of the section (the section was labeled from both
sides). Asterisks indicate glial lamellae. The labeling is highly selective: only seven
particles in this field are not associated with any postsynaptic density. Of these, one
is found within a spine (arrow) and two within other intracellular compartments
(arrowheads). The linear density of gold particles (particles per micrometer mem-
brane) in parallel fiber–Purkinje cell synapses was 19.7, compared to 0 in all other
types of cerebellar synapse. Inset: Higher magnification of an immunolabeled paral-
lel fiber synapse (postsynaptic density delimited by arrowheads). M, mitochondrion.
Gold particles, 15 nm. Scale bars: 0.5 µm; 0.1 µm (inset). (From Landsend et al.,
1997.)
position of the target antigen is provided by the combination of freeze frac-

ture and immunogold labeling (Fujimoto, 1995; Rash et al., 2004, Hagiwara
et al., 2005). With this combination of techniques, the immunogold proce-
dure is employed to determine the molecular identity of intramembrane
particles (IMPs) visualized in metal replicas. This approach has proved
particularly useful in the case of proteins that are clustered in the plasma
membranes, such as connexins and aquaporins. For molecules with a more
scattered distribution, the analysis may be hampered by limited sensitivity.
This notwithstanding, the freeze fracture–immunogold labeling technique
has enormous potential for determining the exact position and number of
receptors, transporters, or other molecules in the plasma membranes of
neural cells. The extent to which this potential can be realized depends on
the ability to classify IMPs according to their sizes and structural features. A
further refinement of the freeze fracture approach is required to define a
sufficiently broad range of membrane molecules.
III. QUANTITATION
“Quantitative immunocytochemistry” is considered by many as a contra-

diction in terms. It is true that we are still far away from the “ultimate goal”
of being able to determine the accurate numbers of molecules by means of
immunocytochemistry. However, immunogold procedures have brought us
closer to this goal and should open avenues for further advances, particularly
if combined with appropriate calibration systems or with freeze fracture
techniques. A thorough discussion of quantitative aspects of immunocyto-
chemistry is provided by Griffiths (1993).
As set out in the Introduction, the use of gold particles facilitates quanti-
tation. Gold particles represent an “all or none signal,” setting them apart
from the less easily quantifiable reaction product of enzyme-based immuno-
cytochemistry. The particles can be readily identified and counted in the
electron microscope, and computer programs have been designed for au-
tomated acquisition and analysis of their distribution (Blackstad et al., 1990;
Haug et al., 1994, 1996; Monteiro-Leal et al., 2003; Ruud and Blackstad,
1999). In most cases, each gold particle deposit can be regarded as the re-
sult of an independent antigen–antibody reaction, permitting the use of
simple statistics. Hence, one would predict a linear relationship between
the gold particle density and the number of available antigen molecules
(Ottersen, 1989b; also see Posthuma et al., 1988).
The major obstacle to quantitation resides not in the counting and analysis
of gold particles but in the nature of the underlying event: the antibody–
antigen coupling. A conditio sine qua non for a meaningful quantitation of a
sample of antigen molecules is that each molecule in the sample faces the
same likelihood of encountering and binding to an antibody molecule. In
practice, this requirement is difficult to fulfill. Preembedding procedures
pose particular problems, as diffusion barriers imposed by membranes and
other tissue constituents will bias the access of immunoreagents to the tar-
get molecules. This bias remains even with optimum permeabilization of
the tissue in question. With the postembedding immunogold technique,
the problem of diffusion constraints is eliminated, as the sample of target
antigens is restricted to those molecules that are available at the cut surface
of the section. Diffusion to the interior of the section is effectively hindered
by the resin. Thus, an attractive feature of the postembedding approach is
that all molecules in the sample are equally likely to be visualized by the
immunocytochemical procedure.
The situation in practice is probably not as simple (Griffiths, 1993). Access
to antibodies may be skewed by the surface relief of the section, and epitopes
may be obscured by protein–protein interactions. Also, sterical hindrance
between the rather bulky immunoreagents may reduce the probability of
Gold particles (µm2)


Figure 3.6. A calibration system tailor-made for postembedding immunogold cyto-
chemistry (Ottersen, 1989b) demonstrates linear relationship between antigen con-
centration (mmol/l fixed glutamate in A; glutamine in B) and gold particle density
(recorded over antigen-containing bodies such as those shown in Fig. 3.7). The lin-
ear relationship tends to break down at very high (biologically irrelevant) antigen
concentrations (B). The lower ends of the plots are shown at larger scale in insets.
The slopes and correlation coefficients for the two regression lines were 12.4 and
0.99 (glutamate, all data points included) and 5.4 and 1.00 (glutamine, estimated
for data points shown in inset). (From Ottersen et al., 1992.)
labeling at sites of very high antigen concentrations. This notwithstanding,

by use of model antigens it has been shown that a linear relationship indeed
can be obtained between antigen concentration and gold particle density.
We have used conjugated amino acids as model antigens (Ottersen, 1987,
1989b). Incorporation of radiolabeled amino acids allowed the concentra-
tion of antigen to be determined. Antigens in known concentrations were
embedded in resin and sectioned for postembedding immunogold labeling
and electron microscopy. For a series of different model antigens, it could
be shown that the gold particle density was positively and linearly correlated
with the calculated concentration of antigen molecules (Ottersen, 1989b;
Ottersen et al., 1992; Fig. 3.6). As predicted, in some cases the linearity was
found to break down at very high antigen levels, probably as a result of
steric hindrance.
The sections that were used to establish the correlation between the anti-
gen concentration and the gold particle density could be incubated together
with tissue sections and under exactly the same conditions as these. We
thus had in our hands a calibration system that allowed us to determine
the approximate concentration of antigen in different cells and organelles.
This approach was used, inter alia, to assess the concentration of glutamate
in nerve terminals, neuronal cell bodies, and astrocytes (Ottersen, 1989b;
Ottersen et al., 1992), and in organelles such as mitochondria and synaptic
vesicles (Shupliakov et al., 1992).
Aprerequisite for an adequate use of calibration curves is knowledge of
the fraction of antigens that are retained in the tissue after fixation. In the
case of small molecules such as neurotransmitter amino acids, this is not a
trivial problem, as substantial amounts may be lost from the tissue during
the fixation procedure. With optimum fixation, using 1% glutaraldehyde or
more, the retention of free amino acids exceeds 80%. This could be shown
by equilibrating the tissue with tracer amounts of radiolabeled amino acids
prior to fixation by immersion or perfusion (Storm-Mathisen and Ottersen,
1990).
When used in conjunction with appropriate model systems such as the
one described above, postembedding immunocytochemistry permits an as-
sessment of amino acid concentrations in cell compartments down to the
level of synaptic vesicles and other organelles. In principle, this approach
is applicable to all antigens that are available in pure form and that can be
incorporated in calibration systems that are representative of the mode of
antigen expression in vivo (see Griffiths, 1993, for a detailed discussion).
Plasma membrane proteins represent a more difficult case in regard to
quantitation than small organic molecules or proteins that are distributed
in the aqueous interior of the cell. For membrane proteins, a representative
calibration system should be based on model membranes containing known
concentrations of the protein in question. Ideally, the model membranes
should have a composition similar to that of the membrane in which the pro-
tein is expressed in vivo. Obviously, these antigen-containing model mem-
branes would have to be embedded and sectioned in parallel with the tissue
and subjected to simultaneous immunoprocessing. It has proved difficult to
develop model systems that meet all of these requirements. Unfortunately,
therefore, postembedding immunogold cytochemistry of membrane pro-
teins remains semiquantitative rather than quantitative.
To circumvent the difficulties entailed in developing a calibration sys-
tem for membrane proteins, one may take advantage of quantitative im-
munoblotting and stereological data. Quantitative immunoblotting pro-
vides an estimate of the amount of a given protein per volume unit of tissue,
whereas stereological analyses can be used to determine the total distribu-
tion surface for that protein (such as the total astrocytic surface in the case
of a membrane protein that is restricted to astrocytes). In this way, one may
calculate the average number of protein molecules per unit area of plasma
membrane, as has been done for astroglial glutamate transporters (Lehre
and Danbolt, 1998). This value can be correlated to the linear density of
gold particles signaling the molecule in question. In principle, one ends up
with a “correction factor” that can be used subsequently to translate gold
particle densities into densities of target proteins.
Instead of calibrating the immunogold signal to biochemical data, one
can relate the number of gold particles to functional parameters, such
as the magnitude of synaptic currents in the case of postsynaptic recep-
tors. Nusser et al. (1998) used this approach to assess the number of
alpha-amino-3-hydroxy-5-methyl-4-isoxazole-propionic acid (AMPA) recep-
tors at hippocampal synapses. One gold particle in postembedding-labeled
methacrylate sections was found to correspond to 2.3 functional receptors,
allowing estimates of synaptic receptor content.
As pointed out above, the major challenge in quantitation is to provide

an accurate conversion factor between the number of gold particles and
the number of tissue antigens. However, the analysis of particle distribution
is not trivial, even if one refrains from assumptions regarding the under-
lying pattern of antigen distribution. Computer-based methods are indis-
pensable if one embarks on large projects that involve sampling of sev-
eral animals, blocks, sections, and images. In early studies in our labo-
ratory, we digitized photographic prints by means of a digitizing table
and counted particles with an electronic pen under control of the com-
puter programs Morforel (Blackstad et al., 1990) and Palirel (Ruud and
Blackstad, 1999). Today, our main tool for quantifying immunogold label-
ing is IMGAP (IMmuno-Gold-Analysis-Program), created in our laboratory
(Haug et al., 1994, 1996). This program was developed in collaboration
with SIS (Soft Imaging Systems Gmbh, Münster, Germany) as an extension
to their product analySIS. IMGAP permits automated acquisition of gold
particles, taking advantage of their high electron density and defined sizes.
Appendix II describes the typical workflow when using IMGAP. The web site
http://www.med.uio.no/imb/stat/immunogold/index.html is a service for
methods and programs for statistical analysis of immunogold data.
Simplified methods are applicable when the biological problem at hand is
limited to that of comparing immunogold labeling patterns in the same sets
of compartments across different cells (Mayhew et al., 2004). In this situation,
no information may be required about compartment size or membrane
length.
Rapid assessments of gold particle distributions can be obtained by use
of simple stereological methods. One example from our own laboratory is
described in Landsend et al. (1997), based on preparations such as that
shown in Fig. 3.5. Using a stereological approach, Lucocq et al. (2004)
claimed that counting 100–200 particles on each of two grids may be suffi-
cient to produce a rough estimate of the gold particle distribution over as
many as 10–16 different compartments. This should be kept in mind when
designing an immunogold analysis so as to avoid excessive and pointless
counting of gold particles.
Finally, it should be emphasized that the freeze fracture–immunogold
technique represents a quite different approach to quantitation, holding
great promise for the future (for review, see Rash et al., 2004). Accurate quan-
titation should be feasible for any protein that can be immunodetected in
freeze fracture replicas and that has structural features that set it apart from
other IMPs in the same membrane domain. The limited ability to differ-
entiate between different IMPs, based on their morphological appearance,
remains a problem in this regard.
Closely related to the issue of quantitation is the term labeling efficiency.
This term refers to the ratio between the number of gold particles and
the number of antigen molecules that is available for immunolabeling.
Labeling efficiency depends on many factors that have been addressed in
studies of cryosections (e.g., Griffiths and Hoppeler, 1986). As pointed out
by Griffiths (1993), it is probable that the labeling efficiency is always below
100% in postembedding-labeled sections. A rough estimate of the labeling
efficiency can be obtained by analysis of membrane domains that contain
known densities of the target proteins (derived, e.g., from correlative freeze
fracture analyses). In a postembedding immunogold analysis of AMPA re-
ceptors in hippocampal synapses, it was shown that the number of gold
particles increased by one particle per ∼15-nm increment in the length of
synaptic profile (Takumi et al., 1999). Based on the known size of AMPA
receptors and conservative estimates of their spacing, these data suggest
that there is a close to 1:1 stoichiometry between gold particles and AMPA
receptors at the section surface, given optimum experimental conditions.
Factors that affect labeling efficiency include quality of antibody, fixa-
tion procedure, embedding medium, and incubation parameters (Griffiths,
1993; Matsubara et al., 1996). Several of these factors are addressed below, in
the discussion of our “standard” postembedding immunogold procedure.
Optimization of the postembedding immunogold procedure for a given
antigen is very much a question of obtaining maximum labeling efficiency.
High labeling efficiency is of critical importance in postembedding immuno-
gold analyses because of the restricted sample of accessible epitopes, and be-
comes a decisive factor in analyses of membrane proteins that are expressed
at low densities. Indeed, the choice of immunocytochemical procedure (pre-
or postembedding) should always be based on available information on the
prevalence of the antigen at hand.
IV. CONTROLS
The need for appropriate controls in immunocytochemistry can hardly

be overemphasized. As a discussion of the general principles for assessing
antibody selectivity is outside the scope of the present chapter, we will restrict
ourselves to procedures that are specific for postembedding immunogold
cytochemistry.
Testing of selectivity must be done in conditions that are representative of
the conditions of the immunocytochemical procedure. This is because the
conformation of the epitope and the nature of the antibody–epitope interac-
tion may be influenced by a number of factors, such as the choice of fixative
and resin and the selection of incubation parameters. In other words, show-
ing that the antibody identifies a single band at the appropriate molecular
weight in immunoblots cannot be taken to imply that the antibody pro-
vides selective labeling in the postembedding mode. Immunocytochemistry
and antibody testing should be performed in parallel and under identical
conditions.
This criterion can be met in the case of small molecules such as amino
acids and larger antigens that are available in pure form (Davanger et al.,
1994; Ottersen, 1987). The target antigen and structurally related molecules
can be embedded in resin, sectioned, and immunoincubated together with
Gln
Asp
None
Gly
Tau
Glu
Gaba
Figure 3.7. Test system designed to monitor antibody selectivity under the condi-
tions of the immunocytochemical procedure. The target antigens and structurally
homologous molecules were embedded in resin and incorporated in a test sand-
wich with alternating brain sections used as spacers. Ultrathin cross sections of this
sandwich were incubated together with the tissue sections. The test antigens appear
as dense bodies. Bodies identified by arrows are enlarged in the right part of the
figure. In this case, the test antigens were prepared by coupling glutamate (Glu),
glutamine (Gln), and structurally related amino acids (standard abbreviations) to
brain macromolecules in the presence of glutaraldehyde. The test section shown
here and the accompanying tissue section (Fig. 3.8) were double labeled for glu-
tamate (15-nm particles) and glutamine (30-nm particles) using a modification of
the procedure of Wang and Larsson (1985) (see Ottersen et al., 1992, for details).
Quantitative analysis of this test section (Ottersen et al., 1992) confirmed that the
antisera react selectively with the target antigen in the actual conditions used for
fixation, embedding, and immunoincubation. The quantitative analysis also showed
that the present double-labeling procedure (using two antibodies from the same
species) distinguishes between the two antigens with negligible interference. (From
Ottersen et al., 1992)
Figure 3.8. Section from the molecular layer of rat cerebellum incubated together
with the test section shown in Fig. 3.7. The high selectivity obtained in the test section
should be representative of the selectivity in the tissue section. Small gold particles
signaling fixed glutamate are enriched in parallel fiber terminals (pf ), whereas large
gold particles signaling glutamine mainly decorate glial processes (g) and Purkinje
cell dendritic spines (s). Asterisk, possible climbing fiber; arrows, intercellular clefts.
Scale bar: 0.4 µm. (From Ottersen et al., 1992.)
the tissue sections. This approach provides a direct and reliable validation of
antibody selectivity. In the example shown here (Figs. 3.7 and 3.8), it could
be documented that antibodies to glutamate and glutamine distinguish
between the respective amino acids, despite their close structural similarity.
The degree of cross-reactivity could be determined quantitatively (Ottersen
et al., 1992; Fig. 3.7). It is important that the test antigens are exposed to
the same fixative as the target antigen in the tissue. Specifically, in the case
of amino acids, these must be conjugated to brain proteins by glutaralde-
hyde before embedding and testing (Ottersen, 1987; Storm-Mathisen et al.,
1983). In this way, the test antigens will mimic the complexes that are formed
in the tissue during perfusion fixation, when the fixative (glutaraldehyde)
cross-links free amino acids to brain macromolecules.
Positive controls such as that discussed above document the ability of the
antibody to differentiate between structurally similar epitopes. However,
negative controls are required to ascertain that the immunogold signal rep-
resents antibody binding to the target antigen rather than unspecific la-
beling. Such controls are particularly important when the target antigen is
believed to reside in nuclei, mitochondria, postsynaptic densities, or other
sites that promote unspecific binding due to high protein concentrations.
The most powerful negative control is provided by the availability of animals
with a selective knockout of the gene encoding the target protein. Pend-
ing knockout animals, transfection experiments (comparing cells with and
without the antigen in question) constitute a useful substitute. One must
not put too much emphasis on standard absorption experiments (involving
neutralization of the primary antibody by application of an excess of the
immunizing peptide), as these do not differentiate between specific and
unspecific binding of the antibody clone in question (for a comprehensive
discussion of specificity controls, see Holmseth et al., 2005).
V. APPLICATIONS
It is outside the scope of the present chapter to provide a comprehensive

discussion of the range of biological problems to which the postembedding
immunogold technique can be successfully applied. In our own laboratory,
we have found this technique to be particularly useful for the following
purposes:
1. Demonstration of protein colocalization by use of double labeling
with two different-sized gold particles (Fig. 3.9; also see Fig. 3.15).
This application takes advantage of the discrete sizes of gold particles
and the fact that double labeling can be successfully performed,
with minimum cross-reactivity (Fig. 3.7), even when the two primary
antisera are derived from the same species (Ottersen et al., 1992; Wang
and Larsson, 1985).
2. Demonstration of receptors and amino acid transmitter (e.g., gluta-
mate) in the same synapses by use of double labeling (Matsubara et al.,
1996; Takumi et al., 1999; Fig. 3.10). This procedure requires the use
of glutaraldehyde in the fixative to retain the amino acid in question
(see section “Postembedding Procedures”).
3. Combination of postembedding labeling and anterograde tracing to
identify transmitters and receptors in specific fiber projections ( Ji et al.,
1991; Ragnarson et al., 1998, 2003; Rinvik and Ottersen, 1993; Fig. 3.11).
4. Investigation of disease mechanisms at high resolution. Examples:
analysis of glutamate redistribution in experimental stroke to explore
the mechanisms of excitotoxic cell death (Torp et al., 1993), and
analysis of the mechanisms of β-amyloid generation in Alzheimer’s
disease and relevant animal models (Torp et al., 2000, 2003; Fig. 3.12).
5. Phenotypic analyses of transgene animals to demonstrate changes
in subcellular expression of neuronal or glial proteins (Amiry-
Moghaddam et al., 2003a,b; Amiry-Moghaddam and Ottersen, 2003;
Kohr et al., 2003; Neely et al., 2001; Rossi et al., 2002).
VI. POSTEMBEDDING PROCEDURES
The hallmark of postembedding immunocytochemistry is that the tissue is

embedded in a resin prior to the immunocytochemical procedure. As most
Figure 3.9. Semiquantitative analysis of subcellular expression patterns of two pro-

tein antigens. Double immunogold labeling of Kir4.1 (10-nm particles) and AQP4
(20-nm particles) in vitreal (A, B), perisynaptic (C), and perivascular (D) Müller
cell (M) membranes. (A, B) The M end foot (M) is selectively labeled at its vitreal
aspect (both aspects indicated by double arrow). The vitreal plasma membrane (be-
tween dashed lines in B) is obliquely cut, allowing labeling at the two sides of the
section to be distinguished. Small arrows indicate Kir4.1 labeling. Asterisks indicate
corresponding points at the vitreal surface (part of A is enlarged in B). (C) Weak
Kir4.1 (arrows) and AQP4 labeling is found in the thin Müller cell processes (M)
that surround photoreceptor terminals (Pt). (D) Perivascular end feet (M) shows
a polarized distribution of gold particles (double arrow, compare with A). End, en-
dothelial cell; P, pericyte. (E) Kir4.1 immunolabeling of an M microvillus (asterisk).
The gold particles (arrows) are restricted to its basal part which is devoid of large
particles signaling AQP4. IS, inner segment of photoreceptors. Scale bars: 0.5 µm
(A); 0.1 µm (B, E); 0.25 µm (C, D). The graph shows linear densities of gold particles
in different membrane domains of M. (Modified from Nagelhus et al., 1999.)
Figure 3.10. Double immunogold labeling of transmitter and receptor in synapses

from stratum radiatum of rat hippocampus. Large gold particles signal fixed glu-
tamate, and small particles signal NMDA receptors. Arrowheads indicate extent of
postsynaptic densities. t, nerve terminal. Scale bar: 200 nm. (From Takumi et al.,
1999.)
resins are hydrophobic, the tissue water has to be replaced by an organic

solvent before the infiltration step. Thus, the sequence of steps is as follows:
1. Fixation
2. Dehydration
3. Embedding (infiltration and polymerization of resin)
4. Sectioning
5. Immunoincubation
6. Electron microscopic analysis of sections
Each of the steps above can be performed in many different ways. In fact,
there are about as many recipes as there are laboratories in the field. The
reasons why there is such a plethora of procedures are that each laboratory
has a unique set of needs and that the protocol has to be tailored to the
antigen at hand. The latter point cannot be overemphasized: for each new
target antigen, one must be prepared to modify the procedure for optimum
results. This is why postembedding immunocytochemistry is oftentimes chal-
lenging and sometimes frustrating—and very rewarding when it works.
Any researcher who enters into the field of postembedding immunocyto-
chemistry will soon realize that many antibodies will never work, regardless
of fixation, embedding, and incubation conditions. Unfortunately, the per-
formance of a given antibody in postembedding immunogold analyses can-
not be predicted from its performance in immunoblots or immunofluores-
cence. An element of “trial and error” is inevitable.
Appendix I provides an outline of the protocol that is currently being used
in our own laboratory for the initial screening of a novel protein antigen.
Optimization of this protocol for a given protein may require substantial
modifications, as emphasized below.
We will first provide a general discussion of the major steps of the proce-
dure, starting with tissue fixation.
A. Fixation
As for immunocytochemistry in general, the choice of fixative is a trade-

off between the need to preserve tissue ultrastructure and the need to
Figure 3.11. Combination of anterograde labeling and postembedding immunogold

cytochemistry. Spinocerebellar terminals were identified by anterograde transport
of horseradish peroxidase conjugated to wheat germ agglutinin (HRP-WGA). The
HRP reaction product is indicated by arrowheads. The anterogradely labeled mossy
fiber terminal (Mf) is immunopositive for glutamate (A) but immunonegative for
aspartate (B; section adjacent to that in A). The absence of aspartate immunolabeling
is not a false-negative result as aspartate-containing model conjugates, incubated
together with the tissue section, showed strong immunogold labeling (inset in B).
m, mitochondria; asterisks, granule cell dendritic digits. Scale bars: 0.2 µm (A, B);
0.6 µm (inset). (From Ji et al., 1991.)
Figure 3.12. (A, B) Plasma membrane domains show coexpression of presenilin 1

(PS1) and amyloid (Aβ 42). Double immunogold labeling reveals colocalization of
Aβ (large particles) and PS1 (small particles) in a discrete patch of neuronal plasma
membrane in somatosensory cortex of an aged canine. Framed area in A is enlarged
in B. D, dendrite. Scale bars: 0.3 µm (A); 0.2 µm (B). (From Torp et al., 2000.)
(C) Postembedding immunogold labeling of amyloid plaque in the hippocampus
of a transgenic mouse model of Alzheimer’s disease (3XTg-AD). Arrowheads point
to extracellular deposit of amyloid. Intracellular compartments (asterisk) are devoid
of gold particles. Antibody to Aβ 42. T, terminal. Scale bar: 0.4 µm (R. Torp, 2006).
preserve the antigen in a form that can be recognized by the specific antibod-
ies. Strong fixatives, with a high concentration of glutaraldehyde, provide
the best ultrastructure, whereas weak fixatives, with little or no glutaralde-
hyde, provide optimum preservation of antigenicity and hence the strongest
immunocytochemical signal. The concentration of glutaraldehyde is more
critical than the concentration of formaldehyde, as glutaraldehyde is bi-
valent (it has two reactive aldehyde groups). As such, glutaraldehyde is a
very efficient cross-linker. Cross-linking of tissue macromolecules affords
good ultrastructure but may severely distort or mask the target epitopes. In
our hands, it is impossible to predict the amount of glutaraldehyde that a
given antigen will tolerate. For some membrane proteins, glutaraldehyde
concentrations as high as 1% have proved to be compatible with a strong

immunogold signal. Our standard protocol (i.e., the protocol used for the
first screening of a novel protein target) is based on the use of 0.1% glu-
taraldehyde in combination with 4% formaldehyde.
Visualization of neuroactive amino acids (such as glutamate, γ-amino
butyric acid (GABA), and glycine) represents a special case (Ottersen, 1987;
Somogyi et al., 1986; Somogyi and Hodgson, 1985), as these small molecules
are lost from the tissue unless they are irreversibly cross-linked to tissue
macromolecules (Dale et al., 1986; Ottersen, 1989a; Storm-Mathisen et al.,
1983). Thus, immunocytochemical analysis of amino acids and other small
molecules with free amino groups (including glutathione; Hjelle et al., 1994)
requires the inclusion of significant amounts of glutaraldehyde. We have
shown that 1% glutaraldehyde is sufficient to retain the major proportion
of free amino acids. Concentrations as low as 0.1% may yield a good signal,
permitting double labeling of a transmitter amino acid and its respective
receptor (e.g., visualization of presynaptic glutamate and postsynaptic AMPA
receptors in the same synapses; Takumi et al., 1999). However, the lower the
glutaraldehyde concentration, the greater the risk for a skewed retention of
amino acids across the different cell compartments.
B. Dehydration and Embedding
For protein antigens in the central nervous system, the combination of

freeze substitution and embedding in methacrylate resins has become very
popular (Griffiths, 1993; Humbel and Schwarz, 1989; Matsubara et al., 1996;
Nusser et al., 1998; van Lookeren Campagne et al., 1991). This is also the
standard procedure in our own laboratory. This combination of techniques
is designed to preserve the original conformation of the protein antigen.
For analyses of brain, we prefer to fix the tissue before freezing rather than
at later stages in the process.
The first step is to cryoprotect the fixed tissue specimen by immersion in
glycerol or sucrose. The specimen is then frozen, usually by plunging it into
liquid propane. Freezing must be obtained as quickly as possible, in order
to avoid loss of ultrastructure due to the formation of ice crystals (Griffiths,
1993). Liquid propane has a higher temperature than liquid helium but
is considered superior to the latter since it allows a faster dissipation of
heat from the tissue specimen. Obviously, the size of the specimen must be
restricted in order to permit rapid freezing. In practice, we usually refrain
from exceeding 1 mm in any dimension.
Once frozen, the tissue specimen is transferred from the cryofixation unit
to a cryosubstitution unit. This unit supports three sequential processes:
1. Substitution of methanol for ice
2. Substitution of resin for methanol (i.e., infiltration)
3. Polymerization of resin by ultraviolet (UV) light
The overall effect of processing the tissue in the cryosubstitution unit is

to replace ice with polymerized resin. The substitutions occur at very low
temperatures and are therefore very slow, requiring several days to complete.
In the cryosubstitution unit, contrast is conferred by exposure of the tissue to
uranyl acetate, which has affinity for biological membranes. Uranyl acetate
takes the place of osmium tetroxide, commonly used to provide contrast
when the tissue is embedded in epoxy resins. Osmium tetroxide absorbs
light and can be used only in low concentrations in combination with UV
polymerization.
The freeze substitution procedure outlined above differs significantly
from the classical procedure for dehydration and embedding. The classi-
cal procedure is based on postfixation of the tissue in osmium tetroxide,
followed by dehydration, infiltration in an epoxy resin, and polymerization.
These steps typically take place at room temperature (RT), except for the
polymerization process which is run at 60◦ C, depending on the type of epoxy
resin.
Although procedures based on the use of epoxy resins can be modified
for visualization of proteins in the postembedding mode (Phend et al., 1995;
Salio et al., 2005), the freeze substitution procedure is usually considered as
superior. Our own experience with a number of protein antigens is that the
latter procedure provides a much stronger immunosignal and higher signal-
to-noise ratio than do procedures based on epoxy resins. Several factors
have been proposed to explain why freeze substitution provides such an
advantage:
1. All steps, including polymerization, are run at low temperatures, re-

ducing the risk of protein denaturation.
2. Methacrylate resins can accommodate a certain amount of water, possi-
bly allowing for a partial preservation of the proteins’ hydration shells.
3. The polymerization of methacrylate resins does not engage the pro-
teins to the same extent as does the polymerization of epoxy resins,
reducing the risk of distorting or masking the target epitopes.
4. Due in part to their unique polymerization features, methacrylate
resins provide for a more pronounced surface relief than do epoxy
resins. In other words, proteins at the cut surface tend to stick out of
the plane of section rather than being bisected.
In sum, these mechanisms may help retain the original conformation

of the target protein and preserve its antigenicity (Griffiths, 1993). The
freeze substitution procedure is also well suited for nonproteinaceous anti-
gens such as neuroactive amino acids, allowing for double labeling analyses
(Fig. 3.10; also see Matsubara et al., 1996).
A range of methacrylate resins is available. The different resins differ
primarily by their ability to accommodate water and by their fluidity. In
principle, given the fact that proteins are normally expressed in an aqueous
environment, high hydrophilicity should translate into a better preservation
of protein conformation. For glutamate receptors and a number of plasma
membrane transporters in the central nervous system, we have observed
rather minor differences between different methacrylate resins when it
comes to the strength of the immunosignal. Our standard procedure is
based on the use of Lowicryl HM20, which is one of the more hydrophobic
methacrylates. Very hydrophilic methacrylates may be difficult to work with
due to inadvertent swelling of the sections once they are exposed to water.
C. Immunoincubation
Postembedding immunogold labeling of ultrathin sections usually com-

prises the following steps:
1. “Etching” of section to increase the availability of epitopes at the section
surface
2. Blocking of unspecific binding of antibodies
3. Application of primary antibody
4. Application of secondary IgG, Fab, or protein A coupled to colloidal
gold
The goal is to end up with a distribution of gold particles that truly reflects
the distribution of target epitopes, given the constraints imposed by the size
of the antibody bridge (Fig. 3.1). To minimize variability, the sections and
grids to be analyzed should be included in the same batch for simultaneous
incubation.
Appendix I shows the immunoincubation procedure that is in current
use as our standard laboratory protocol. The standard protocol has been
modified over the years (Chaudhry et al., 1995; Hjelle et al., 1994; Nagelhus
et al., 2005) and has been influenced strongly by protocols published from
other laboratories (e.g., Nusser et al., 1998; van Lookeren Campagne et al.,
1991). In fact, as is the case for immunocytochemical procedures in general,
postembedding procedures evolve continuously as parameters are tested out
and as resins and reagents improve.
The outcome of an immunoincubation depends on the combined effect
of a number of different steps. In fact, the enormous number of permu-
tations that can be obtained by combining variations of the different steps
precludes a bona fide scientific approach to optimization of the procedure.
This is unfortunate, as several steps in current use may have been carried
over from other experimental settings, which may or may not be represen-
tative of the procedure in question.
For the purpose of the present chapter, we have performed a system-
atic variation of key parameters to assess their relative importance for the
strength of the immunogold signal (measured as the number of gold par-
ticles per micrometer membrane following immunogold visualization of
the water channel protein AQP4). As shown in Fig. 3.13, the choice of sec-
ondary IgG/colloidal gold conjugate profoundly affects the signal strength
Figure 3.13. Quantitative assessment of how incubation parameters influence label-

ing efficiency. Preparations similar to that showed in Fig. 3.3A, postembedding la-
beled with an antibody to AQP4, were subjected to automated analysis of gold particle
density by the procedure outlined in Appendix II. The labeling intensity (particles
per micrometer perivascular membrane) depended on the combination of incu-
bation parameters (A–H). Reference conditions (A) were identical to those of the
“standard” protocol in Appendix I. These were sodium ethanolate (etching), 2%
HSA (blocking), 0.04 M NaCl (in buffer), 0.1% Triton (T), 15-nm gold particles.
B–H differed from A in the following ways. B: no sodium ethanolate; C: 10-nm gold
particles; D: no HSA; E: 0.01% T; F: 0.2% milk powder in lieu of HSA; G: 0.12 M
NaCl; H: identical to G, but analyzed by a different person (to check for consistency).
Sixty membrane segments were analyzed for each combination of parameters.
(compare A and C). Secondary IgG coupled to 10-nm particles leads to a

higher linear density of gold particles than does a secondary IgG coupled
to 15-nm particles. This effect relates primarily to the size of gold particle
(rather than differences between the two IgGs) as qualitative analyses of a
wider range of particle sizes point to a negative correlation between gold
particle size and signal strength. Steric hindrance may be one of the several
factors that underlie this effect.
A logical extension of the latter observations would be to minimize the
gold particle size to maximize the labeling efficiency. Gold particles that
are less than 5 nm in diameter are not easily discerned in postembedding-
labeled sections, calling for silver enhancement if smaller particles are used.
Application of nanogold particles in combination with silver enhancement
indeed leads to a significant increase in gold particle density, as shown in
postembedding immunogold labeling of glutamate receptors (Matsubara
et al., 1996; also see Fig. 3.17). However, such preparations pose difficul-
ties in quantitative analyses: The silver-enhanced particles are less uniform
than gold particles in regard to size and shape, and their stoichiometric
relationship to the target antigens is less well defined.
Another parameter that may affect signal strength is the concentration
of NaCl in the buffer of the primary and secondary antibody steps. It is
well known that antibody–antigen binding depends on a number of differ-
ent forces, including ion bonds between oppositely charged residues in the
antibody and antigen molecules. Thus, ions in the buffer (primarily Na+
and Cl− ions) would be expected to compete with antigen–antibody bind-
ing. This prediction was borne out in our postembedding analysis of AMPA
glutamate receptors (Matsubara et al., 1996) and led us to lower the NaCl
concentration of the buffer (typically to one third of that of physiological
saline) whenever an increased signal strength was wanted. In the case of
AQP4, the effect of reducing the salt concentration is negligible (compare
A with G in Fig. 3.13), underlining the fact that the significance of the
different parameters depends on the nature of the target antigen.
Blocking of unspecific binding is an essential step in postembedding
immunocytochemistry as in immunocytochemistry in general. Unspecific
binding may be caused by a wide range of mechanisms. Obviously, when
aldehydes are used for fixation, there is a risk that reactive aldehyde groups
remain in the tissue and that these groups attach the immunoreagents to
the section. For this reason, we always include glycine and TRIS in the first
blocking buffer. Both of these molecules have free amino groups that would
bind to and neutralize any free aldehyde groups at the section surface.
We also regularly use human serum albumin (HSA) to prevent unspecific
binding of IgG. Fortunately, the inclusion of up to 2% of HSA does not cause
any inadvertent reduction in the strength of the specific signal (compare
A with D in Fig. 3.13). Milk powder is also in common use as a blocking
agent, but our experience is that it is difficult to adjust its concentration
so as to provide a reduction of background labeling without affecting the
specific signal. In fact, in our quantitative analysis of AQP4 immunolabeling
we recorded a reduction in gold particle density when exchanging 2% HSA
with 0.2% milk powder (compare A with F in Fig. 3.13). The use of well-
defined blocking agents (such as HSA) is encouraged for the purpose of
standardization and reproducibility. Very low background labeling can be
obtained with adequate blocking procedures (cf. Figs. 3.5 and 3.14).
The etching step is probably the most problematic step to justify in the
postembedding immunogold protocol. As alluded to above, the purported
aim of this step is to improve access to the target epitopes. However, it is
not clear to what extent the commonly used etching procedure (a short
immersion in sodium ethanolate; see Appendix I) removes resin and un-
masks surface epitopes. In fact, our quantitative analysis of AQP4 suggests
that there is no gain in immunogold signal by including this step (compare
A with B in Fig. 3.13). Leaving out the etching steps also allows for excellent
labeling of glutamate receptors and intracellular enzymes (Figs. 3.15 and
3.16). It should be emphazised that the comparison in Fig. 3.13 was done
in the presence of Triton. Triton might help unmask epitopes at the section
surface (by dissolving lipids) and could easily obscure any positive effect
of the etching procedure. Pending a systematic analysis of the interaction
Figure 3.14. High signal-to-noise ratio afforded by optimization of postembedding

immunogold procedure. Gold particles signal expression of AQP4 in the subforni-
cal organ of rat. Immunogold particles identify AQP4 along the entire glial lamel-
lae except at the membrane domains engaged in gap or adhaerens type junctions
(arrows) or contacting neuronal elements (double-headed arrow in inset). The ves-
sel (V) and associated basal laminae (asterisks) are devoid of AQP4 immunolabeling.
Co, collagen; Fi, fibroblast; Gf, glial filaments; PVS, perivascular space. Inset: unla-
beled synapses (arrowheads) sandwiched between glial lamellae. The adjacent glial
processes are polarized with respect to AQP4 expression (double-headed arrow).
De, dendrite. Scale bars: 1 µm. (Micrographs by E. Nagelhus taken From Nielsen
et al., 1997).
between etching and Triton application, the possibility exists that the etch-
ing step represents an unjustified adoption to methacrylate resins of a step
that has been shown to work well with epoxy resins. In the absence of a
proven effect, the etching step should be omitted as it significantly detracts
from the quality of the ultrastructure. Again it should be recalled that the
effect of etching (or lack thereof) may not be the same for all antigens.
The above discussion amply documents that immunocytochemistry is still
an art and not a science. Quantitative analyses such as that shown in Fig. 3.13
can be used to monitor the effect of changing an individual parameter, but
the magnitude of effect, if any, may depend on the other parameters and
on the nature of the antigen. The take-home message is that the postem-
bedding immunogold procedure must be tailored to one’s needs and to
Figure 3.15. Omission of etching (conditions as in Fig. 3.13B but with 0.01% Triton)
is compatible with high labeling efficiency. The sections were first immunolabeled
with an antibody against tyrosine hydroxylase (20-nm gold particles), and after 1 h
in formaldehyde vapor at 80◦ C (procedure of Wang and Larsson, 1985) the sections
were immunolabeled with antibodies recognizing NMDA receptor subunits A and
B (10-nm gold particles). From ventral tegmental area of rat. (Micrograph by E.
Rinvik, 2006.)
Figure 3.16. Omission of etching allows high labeling efficiency and good ultrastruc-
tural preservation (compare with Fig. 3.10). Same procedure as in Fig. 3.15, but
single labeling with antibodies to AMPA receptor subunits 2 and 3. From ventral
tegmental area of rat. (Micrograph by E. Rinvik, 2006.)
one’s targets. This is often a major challenge but a challenge well worth
taking because of the wealth of information that can be gained when an
immunogold experiment succeeds.
APPENDIX I: A POSTEMBEDDING IMMUNOGOLD

PROCEDURE FOR MEMBRANE PROTEINS
This procedure permits double labeling with antibodies to glutamate,

GABA, or other neuroactive amino acids (Takumi et al., 1999).
A. Tissue Preparation
Note. Steps 5–10 are carried out in a computer-controlled cryosubstitution

unit.
1. Anesthetize the animal. Perfuse transcardially with 2% dextran (MW
70,000) in 0.1 M sodium phosphate buffer (PB; pH 7.4, 15 s at RT)
followed by a mixture of glutaraldehyde (0.1%) and formaldehyde
(4%; freshly depolymerized from paraformaldehyde) in the same
buffer (for rats, 50 ml/min for 20 min at RT).
2. Leave the brain in situ overnight (4◦ C).
3. Isolate tissue specimens from brain, cryoprotect by immersion in in-
creased concentrations of glycerol (10, 20, and 30% in PB), 0.5 h for
each concentration (RT), and then overnight in 30% (4◦ C).
4. Place the tissue on the specimen pin and plunge into propane cooled
to −170◦ C by LN2 in a cryofixation unit (Reichert KF80, Vienna,
Austria).
5. Transfer the specimens to 1.5% uranyl acetate dissolved in anhydrous
methanol (- 90◦ C) in a cryosubstitution unit (AFS; Reichert). After
30 h, raise the temperature stepwise (4◦ C increment per hour) from
−90 to −45◦ C.
6. Wash the samples three times with anhydrous methanol.
7. Infiltrate with Lowicryl HM 20 resin (Polysciences, Inc., Warrington,
PA 18976. Cat# 15924) at −45◦ C.
(a) Lowicryl/methanol: 1:1, 2 h
(b) Lowicryl/methanol: 2:1, 2 h
(c) Pure Lowicryl: 2 h
(d) Pure Lowicryl: overnight
8. Change to freshly prepared Lowicryl and move the Reichert capsules
with the specimens to the Lowicryl-filled gelatin capsules in the G-
chamber.
9. Transfer the capsules to a container filled with ethanol.
10. Polymerize with UV light. Start at −45◦ C (24 h), and then increase
temperature to 0◦ C (increment 5◦ C/h). Complete the polymerization
at 0◦ C (35 h).
11. Prepare ultrathin sections at 80- to 100-nm thickness and mount on
nickel rids or gold-coated grids for immunogold cytochemistry.
B. Immunoincubation
1. Etch sections in saturated sodium ethanolate (can be omitted; cf. text)

for 2–5 s.
2. Rinse well in distilled water for 3× short and 1 × 10 min. Let dry and
check that the sections are in place.
3. Place grids into a grid support plate (Leica cat. no. 705698) and im-
merse in 50 mM glycine in TBST (Tris-buffered saline with 0.1% Triton
X-100 or 0.01% Triton; cfr. text) for 10 min.
4. Preincubate in TBST containing 2% HSA for 10 min.
5. Incubate in primary antibody diluted in TBST containing 2% HSA
for 2 h, overnight in RT.
6. Rinse in TBST for 3× short, 1 × 10 min and 3× short, 1 × 10 min.
7. Preincubate in TBST containing 2% HSA for 10 min.
8. Incubate in secondary antibody (IgGs coupled to colloidal gold par-
ticles). Dilute as recommended from the company and with 2% HSA
and 0.05% polyethyleneglycol (PEG) for 1 h.
9. Rinse briefly 6× in distilled water, dry sections.
10. Incubate in 5% uranyl acetate in 40% ethanol for 90 s.
12. Incubate in lead citrate for 90 s.
C. Solutions
Sodium ethanolate: Add 100 g NaOH in 700 ml 100% ethanol.

TBST: 100 ml 0.05 M Tris buffer, pH 7.4 (pH adjusted with HCl); 900 ml
ultrafiltered (UF) water containing 0.9% NaCl; 1 g Triton X-100.
Lead citrate: Dissolve 1.33 g lead nitrate and 1.76 g sodium citrate in 30 ml
UF water. Stir for 30 min. Add 8 ml 1 M NaOH and fill up to 50 ml with
UF water. Aliquot in 10 ml syringes and store protected from light in
refrigerator.
D. Protocol for Postembedding Immunogold Labeling Using

Ultrasmall Gold Particles Coupled to Fab Fragments (Secondary
Antibodies) and Silver Intensification (Fig. 3.17)
1. Rinse grids in MilliQ filtered (18.2 M) water for 1 min

2. TBST containing 0.1% sodium borohydride and 50 mM glycine
3. Blocking buffer 10 min
4. Primary antiserum 2 h or overnight
A B C
Figure 3.17. Silver enhancement provides increased labeling intensity of synaptic

vesicle proteins. (A) Postembedding labeling of SV2 using 10-nm colloidal gold cou-
pled to the secondary antibody. (B) Same as in A, but with 1.4-nm gold particles
(“Nanogold”) coupled to a secondary Fab fragment and followed by silver enhance-
ment. (C) Same as in B, but with antibody to synaptophysin. Note clusters of small
silver deposits in B, and larger silver deposits due to longer reaction time in C. Silver
intensification gives silver deposits of varying sizes, prohibiting double-labeling ex-
periments. (Micrographs by S. Davanger, unpublished data.)
5. TBST 1 min
6. Blocking buffer 3× 1 min
7. Nanogold Fab 1:40 in blocking buffer containing PEG 2000
8. Phosphate-buffered saline (PBS) 3× 1 min
9. Glutaraldehyde 1% in PBS 3 min
10. Water 2 × 1 min
11. HQ silver Nanoprobes (N.B. safelight!)
13. Dry sections
14. Uranyl acetate 2% 40 min
16. Dry sections
17. Lead citrate 0.3% 3 min
19. Dry sections
TBST (1000 ml): 100 ml 0.05 M Tris, pH 7.4; 900 ml MilliQ water with
0.9% NaCl; 1 g Triton X-100.
Blocking Buffer (1500 µl): 2% (30 µl) normal goat serum; 1% (125 µl of
12% stock solution) bovine serum albumin; 0.5% (7.5 µl) Tween 20;
TBST 1337.5 µl.
APPENDIX II: COUNTING IMMUNOPARTICLES

BY DIGITAL IMAGE ANALYSIS
Computer support may substantially reduce the time to completion of

large projects, as illustrated by the IMGAP software referred to in the
text. The latter combines interactive and automated procedures for image
storage and retrieval, particle counting over organelles and membranes,
calculation of areal and linear densities, and conversion of the final data set
to a statistics program for further analyses. The following workflow does not
distinguish between native analySIS functions and in-house extensions.
1. “Define” an “IMGAP project”: Create a project folder with section

folders to hold all images, ancillary information, and results. Define
project-wide pick lists of Field types (see point 6 for a definition
of Field in this context) and Object types (“objects” are transected
spines, postsynaptic densities, etc.). Select measurements (native to
analySIS or defined in IMGAP) to be automatically carried out on
each object.
2. Choose the electron optic magnification that provides the best com-
promise between digital resolution, required field of view, and the
size of the digital image file. Since lossy compression methods should
be avoided, projects with many large images may run into several
gigabytes, not a problem with today’s storage facilities.
3. Capture TIFF images with any suitable camera and software and sub-
sequently import them into IMGAP, or capture images with analySIS
while keeping IMGAP active for online access to the project-wide pick
list of image labels and possibly other services.
4. Freehand draw transected organelles as Regions of Interest (ROIs),
and membranes as curves, attaching a type label to each such graphic
object.
5. After background smoothing and global intensity thresholding, de-
tect and classify particles in up to three size classes.
6. Inspect the result and revise if necessary, by changing parameters for
automated procedures or by interactively deleting, adding, or split-
ting particles. In particular, although particle detection is partly auto-
mated, its quality depends 100% on the operator.
Each ROI, curve, and particle is now uniquely identified and geo-
metrically described by analySIS. Together with analySIS, IMGAP en-
sures that they are automatically stored with the corresponding image
in “revisable format” and automatically displayed in the overlay when-
ever an image is reloaded. In IMGAP, an image with associated ROIs,
curves, and particles, and other associated information, is termed a
field (i.e., “field of view”).
7. Automatically aggregate and export particle measurements (data on
each particle in the project) to SPSS. Use to create transverse his-
tograms for each “type” of membrane and for “quality control” of the
detected particles (cf. point 6).
8. Automatically calculate areal density of particles per ROI, aggregate
over the project, and export to SPSS.
9. Prior to calculating linear densities along membranes, set a distance
filter to eliminate particles too far off the curve.
10. Export ROI and curve measurements for further graphical and statis-
tical analyses.
11. In the SPSS data set, each object (ROI, curve, or particle) has one line
and each measurement or other characteristic one column. Check the
data set for errors as recommended in standard statistics textbooks
and software manuals. Each measurement is traceable to the under-
lying image, and so errors may be diagnosed and corrected.
12. For images in need of corrections, repeat steps 4–6. Then go on to
steps 7 and 8.
13. Further analyses are performed in SPSS or other software (the SPSS
formats are compatible with a range of other statistics software, but
more manual calculations may also be required). Also see web site
http://www.med.uio.no/imb/stat/immunogold/index.html.
An important source of error in the above procedures should be noted.
Due to simplistic particle-detection and quality-control algorithms, small
falsely positive particles may easily go unnoticed unless the operator is
alert to this possibility in step 6 and has been duly trained to optimize the
parameters for steps 2, 3, and 5. Increased intensity resolution (14 rather
than 12 bits) may improve on this, as emphasized by Monteiro-Leal et al.
(2003), and in addition, the algorithms for particle detection and quality
control should be improved.
Limited digital resolution and operator error during interactive drawing
(step 4) may add error, presumably random. Obviously, both the automat-
ically detected particles and the interactively drawn curves will be more
accurate with increasing resolution. Compared to the “basic uncertainty
zone” which surrounds immunogold attached via primary and secondary
antibodies, the present inaccuracy may be thought insignificant. However,
when planning histograms with 1-nm bins, select the electron optic magni-
fication with a view to the resulting object pixel dimensions (the size of a
pixel projected to the specimen). To illustrate this, our MegaView III cam-
era (trademark of Soft Imaging Systems) has a nominal xy resolution just
above 1280 × 1024 pixels. Mounted on our Tecnai 12 it renders a specimen
area of 2.1 µm × 1.6 µm, with an object pixel size of 1.5 nm at a nominal
electron optical magnification of 49,000×. (As an aside, with these param-
eters, 10-nm gold particles come out with diameters around 4–6 pixels and
the particle detection algorithm still works satisfactorily.)
While a digital resolution of 1280 × 1024 pixels is small in relation to dig-
ital cameras in this year’s (2005) consumer market, the price–performance
ratio on cameras for electron microscopy remains two orders of magnitude
higher. If you need higher digital resolution, automated image montage, in
analySIS represented by the MIA module, may be an alternative, although
in our hands its success rate is often less than 100%.
Acknowledgments. Support is acknowledged from the Norwegian Re-

search Council, the Nordic Council (the centre of excellence programme in
molecular medicine), and EU (projects QLG3-CT-2001-02089 and LSHM-
CT-2005-005320).
REFERENCES
Amiry-Moghaddam, M., Otsuka, T., Hurn, P. D., Traystman, R. J., Haug, F. M., Froehner, S. C.,
Adams, M. E., Neely, J. D., Agre, P, Ottersen, O. P., and Bhardwaj, A., 2003a, An alpha-
syntrophin-dependent pool of AQP4 in astroglial end-feet confers bidirectional water flow
between blood and brain, Proc. Natl. Acad. Sci. USA 100(4):2106–2111.
Amiry-Moghaddam, M., and Ottersen, O. P., 2003, The molecular basis for water transport in
brain, Nat. Rev. Neurosci. 4(12):991–1001.
Amiry-Moghaddam, M., Williamson, A., Palomba, M., Eid, T., de Lanerolle, N. C., Nagelhus,
E. A., Adams, M. E., Froehner, S. C., Agre, P., and Ottersen, O. P., 2003b, Delayed K+
clearance associated with aquaporin-4 mislocalization: phenotypic defects in brains of
alpha-syntrophin-null mice, Proc. Natl. Acad. Sci. USA 100(23):13615–13620.
Blackstad, T. W., Karagulle, T., and Ottersen, O. P., 1990, MORFOREL, a computer program
for two-dimensional analysis of micrographs of biological specimens, with emphasis on
immunogold preparations, Comput. Biol. Med. 20(1):15–34.
Chaudhry, F. A., Lehre, K. P., van Lookeren Campagne, M., Ottersen, O. P., Danbolt, N. C.,
and Storm-Mathisen, J., 1995, Glutamate transporters in glial plasma membranes: highly
differentiated localizations revealed by quantitative ultrastructural immunocytochemistry,
Neuron 15(3):711–720.
Dale, N., Ottersen, O. P., Roberts, A., and Storm-Mathisen, J., 1986, Inhibitory neurones of
a motor pattern generator in Xenopus revealed by antibodies to glycine, Nature (Lond)
324:255–257.
Davanger, S., Hjelle, O. P., Babaie, E., Larsson, L. I., Hougaard, D., Storm-Mathisen, J.,
and Ottersen, O. P., 1994, Colocalization of gamma-aminobutyrate and gastrin in the
rat antrum: an immunocytochemical and in situ hybridization study, Gastroenterology
107(1):137–148.
Didier, A., Carleton, A., Bjaalie, J. G., Vincent, J. D., Ottersen, O. P., Storm-Mathisen, J., and
Lledo, P. M., 2001, A dendrodendritic reciprocal synapse provides a recurrent excitatory
connection in the olfactory bulb, Proc. Natl. Acad. Sci. USA 98:6441–6446.
Faulk, W. P., and Taylor, G. M., 1971, An immunocolloid method for the electron microscope,
Immunochemistry 11:1081–1083.
Fujimoto, K., 1995, Freeze-fracture replica electron microscopy combined with SDS digestion
for cytochemical labeling of integral membrane proteins. Application to the immunogold
labeling of intercellular junctional complexes, J . Cell Sci. 108:3443–3449.
Griffiths, G., 1993, Fine Structure Immunocytochemistry, Berlin: Springer-Verlag, p. 459.
Griffiths, G., and Hoppeler, H., 1986, Quantitation in immunocytochemistry: correlation of
immunogold labeling to absolute number of membrane antigens, J . Histochem. Cytochem.
34(11):1389–1398.
Hagiwara, A., Fukazawa, Y., Deguchi-Tawarada, M., Ohtsuka, T., and Shigemoto, R., 2005, Dif-
ferential distribution of release-related proteins in the hippocampal CA3 area as revealed
by freeze-fracture replica labeling, J. Comp. Neurol. 489(2): 195–216.
Haug, F. M., Desai, V. D., Nergaard, P. O., Laake, J., and Ottersen, O. P., 1994, Particle-counting
in immunogold labelled ultrathin sections by transmission electron microscopy and image
analysis, Anal. Cell. Pathol. 6:197.
Haug, F. M. S., Desai, V., Nergaard, P. O., and Ottersen, O. P., 1996, Quantifying immunogold
labelled neurotransmitters and receptors by image analysis, Soc. Neurosci. Abstr. 22(1):581.
Hjelle, O. P., Chaudhry, F. A., and Ottersen, O. P., 1994, Antisera to glutathione: characterization
and immunocytochemical application to the rat cerebellum, Eur. J . Neurosci. 6(5):793–
804.
Holmseth, S., Dehnes, Y., Bjornsen, L. P., Boulland, J. L., Furness, D. N., Bergles, D., and
Danbolt, N. C., 2005, Specificity of antibodies: unexpected cross-reactivity of antibodies
direct against the excitatory amino acid transporter 3(EAAT3), Neuroscience 136: 649–660.
Hu, H., Shao, L. R., Chavoshy, S., Gu, N., Trieb, M., Behrens, R., Laake, P., Pongs, O., Knaus,
H. G., Ottersen, O. P., and Storm, J. F., 2001, Presynaptic Ca2+ -activated K+ channels in
glutamatergic hippocampal terminals and their role in spike repolarization and regulation
of transmitter release, J . Neurosci. 21(24):9585–9597.
Humbel, B. M., and Schwartz, H., 1989, Freeze-substitution for immunochemistry, In: Verkleij,
A. J., and Leunissen, J. L. M. (eds.), Immunogold Labelling in Cell Biology, Boca Raton, FL:
CRC Press, pp. 115–134.
Ji, Z. Q., Aas, J. E., Laake, J., Walberg, F., and Ottersen, O. P., 1991, An electron microscopic,
immunogold analysis of glutamate and glutamine in terminals of rat spinocerebellar fibers,
J . Comp. Neurol. 307(2):296–310.
Kohr, G., Jensen, V., Koester, H. J., Mihaljevic, A. L., Utvik, J. K., Kvello, A., Ottersen, O.
P., Seeburg, P. H., Sprengel, R., and Hvalby, O., 2003, Intracellular domains of NMDA
receptor subtypes are determinants for long-term potentiation induction, J . Neurosci.
23(34):10791–10799.
Landsend, A. S., Amiry-Moghaddam, M., Matsubara, A., Bergersen, L., Usami, S., Wenthold,
R. J., and Ottersen, O. P., 1997, Differential localization of delta glutamate receptors in
the rat cerebellum: coexpression with AMPA receptors in parallel fiber-spine synapses and
absence from climbing fiber-spine synapses, J . Neurosci. 17(2):834–842.
Lehre, K. P., and Danbolt, N. C., 1998, The number of glutamate transporter subtype molecules
at glutamatergic synapses: chemical and stereological quantification in young adult rat
brain, J . Neurosci. 18(21):8751–8757.
Lucocq, J. M., Habermann, A., Watt, S., Backer, J. M., Mayhew, T. M., and Griffiths, G., 2004, A
rapid method for assessing the distribution of gold labeling on thin sections, J . Histochem.
Cytochem. 52(8):991–1000.
Matsubara, A., Laake, J. H., Davanger, S., Usami, S., and Ottersen, O. P., 1996, Organization
of AMPA receptor subunits at a glutamate synapse: a quantitative immunogold analysis of
hair cell synapses in the rat organ of Corti, J . Neurosci. 16(14):4457–4467.
Maunsbach, A. B., and Afzelius, B. A., 1999, Biomedical Electron Microscopy: Illustrated Methods and
Interpretations, San Diego: Academic Press, 548 p.
Mayhew, T. M., Griffiths, G., and Lucocq, J. M., 2004, Applications of an efficient method for
comparing immunogold labelling patterns in the same sets of compartments in different
groups of cells, Histochem. Cell Biol. 122(2):171–177.
Monteiro-Leal, L. H., Troster, H., Campanati, L., Spring, H. F., and Trendelenburg, M., 2003,
Gold finder: a computer method for fast automatic double gold labeling detection, count-
ing, and color overlay in electron microscopic images, J . Struct. Biol. 141(3): 228–239.
Nagelhus, E. A., Horio, Y., Inanobe, A., Fujita, A., Haug, F. M., Nielsen, S., Kurachi, Y., and
Ottersen, O. P., 1999, Immunogold evidence suggests that coupling of K+ siphoning and
water transport in rat retinal Muller cells is mediated by a coenrichment of Kir4.1 and
AQP4 in specific membrane domains, Glia 26(1):47–54.
Nagelhus, E. A., Mathiisen, T. M., Bateman, A. C., Haug, F. M., Ottersen, O. P., Grubb, J. H.,
Waheed, A., and Sly, W. S., 2005, Carbonic anhydrase XIV is enriched in specific membrane
domains of retinal pigment epithelium, Muller cells, and astrocytes, Proc. Natl. Acad. Sci.
USA 102(22):8031–8035.
Nagelhus, E. A., Mathiisen, T. M., and Ottersen, O. P., 2004, Aquaporin-4 in the central nervous
system: cellular and subcellular distribution and coexpression with KIR4.1, Neuroscience
129(4):905–913.
Nagelhus, E. A., Veruki, M. L., Torp, R., Haug, F. M., Laake, J. H., Nielsen, S., Agre, P., and
Ottersen, O. P., 1998, Aquaporin-4 water channel protein in the rat retina and optic
nerve: polarized expression in Muller cells and fibrous astrocytes, J . Neurosci. 18(7): 2506–
2519.
Neely, J. D., Amiry-Moghaddam, M., Ottersen, O. P., Froehner, S. C., Agre, P., and Adams, M.
E., 2001, Syntrophin-dependent expression and localization of aquaporin-4 water channel
protein, Proc. Natl. Acad. Sci. USA 98(24):14108–14113.
Nielsen, S., Nagelhus, E. A., Amiry-Moghaddam, M., Bourque, C., Agre, P., and Ottersen, O.
P., 1997, Specialized membrane domains for water transport in glial cells: high-resolution
immunogold cytochemistry of aquaporin-4 in rat brain, J . Neurosci. 17(1):171–180.
Nusser, Z., Lujan, R., Laube, G., Roberts, J. D., Molnar, E., and Somogyi, P., 1998, Cell type and
pathway dependence of synaptic AMPA receptor number and variability in the hippocam-
pus, Neuron 21(3):545–559.
Ottersen, O. P., 1987, Postembedding light- and electron microscopic immunocytochemistry of
amino acids: description of a new model system allowing identical conditions for specificity
testing and tissue processing, Exp. Brain Res. 69(1):167–174.
Ottersen, O. P., 1989a, Quantitative electron microscopic immunocytochemistry of amino
acids, Anat. Embryol. 180:1–15.
Ottersen, O. P., 1989b, Postembedding immunogold labelling of fixed glutamate: an electron
microscopic analysis of the relationship between gold particle density and antigen con-
centration, J . Chem. Neuroanat. 2(1):57–66.
Ottersen, O. P., Zhang, N., and Walberg, F., 1992, Metabolic compartmentation of glutamate
and glutamine: morphological evidence obtained by quantitative immunocytochemistry
in rat cerebellum, Neuroscience 46:519–534.
Panzanelli, P., Homanics, G. E., Ottersen, O. P., Fritschy, J. M., and Sassoe-Pognetto, M., 2004,
Pre- and postsynaptic GABA receptors at reciprocal dendrodendritic synapses in the ol-
factory bulb, Eur. J . Neurosci. 20(11):2945–2952.
Phend, K. D., Rustioni, A., and Weinberg, R. J., 1995, An osmium-free method of epon embed-
ment that preserves both ultrastructure and antigenicity for post-embedding immunocy-
tochemistry, J . Histochem. Cytochem. 43(3):283–292.
Posthuma, G., Slot, J. W., Veenendaal, T., and Geuze, H. J., 1988, Immunogold determina-
tion of amylase concentrations in pancreatic subcellular compartments, Eur. J . Cell Biol.
46(2):327–335.
Ragnarson, B., Ornung, G., Grant, G., Ottersen, O. P., and Ulfhake, B., 2003, Glutamate and
AMPA receptor immunoreactivity in Ia synapses with motoneurons and neurons of the
central cervical nucleus, Exp. Brain Res. 149(4):447–457.
Ragnarson, B., Ornung, G., Ottersen, O. P., Grant, G., and Ulfhake, B., 1998, Ultrastructural
detection of neuronally transported choleragenoid by postembedding immunocytochem-
istry in freeze-substituted Lowicryl HM20 embedded tissue, J . Neurosci. Methods 80(2):129–
136.
Rash, J. E., Davidson, K. G., Yasumura, T., and Furman, C. S., 2004, Freeze-fracture and im-
munogold analysis of aquaporin-4 (AQP4) square arrays, with models of AQP4 lattice
assembly, Neuroscience 129(4):915–934.
Rinvik, E., and Ottersen, O. P., 1993, Terminals of subthalamonigral fibres are enriched with
glutamate-like immunoreactivity: an electron microscopic, immunogold analysis in the
cat, J . Chem. Neuroanat. 6(1):19–30.
Rossi, P., Sola, E., Taglietti, V., Borchardt, T., Steigerwald, F., Utvik, J. K, Ottersen, O. P., Kohr,
G., and D’Angelo, E., 2002, NMDA receptor 2 (NR2) C-terminal control of NR open
probability regulates synaptic transmission and plasticity at a cerebellar synapse, J . Neurosci.
22(22):9687–9697.
Roth, J., 1996, The silver anniversary of gold: 25 years of the colloidal gold marker system for
immunocytochemistry and histochemistry, Histochem. Cell Biol. 106(1):1–8.
Ruud, H. K., and Blackstad, T. W., 1999, PALIREL, a computer program for analyzing particle-
to-membrane relations, with emphasis on electron micrographs of immunocytochemical
preparations and gold labeled molecules, Comput. Biomed. Res. 32(2):93–122.
Salio, C., Lossi, L., Ferrini, F., Merighi, A., 2005, Ultrastructural evidence for a pre- and postsy-
naptic localization of full length trkB receptors in substantia gelatinosa (lamina II) of rat
and mouse spinal cord, Eur. J. Neurosci. 22(8): 1951–1966.
Shupliakov, O., Brodin, L., Cullheim, S., Ottersen, O. P., and Storm-Mathisen, J., 1992, Im-
munogold quantification of glutamate in two types of excitatory synapse with different
firing patterns, J . Neurosci. 12(10):3789–3803.
Somogyi, P., Halasy, K., Somogyi, J., Storm-Mathisen, J., and Ottersen, O. P., 1986, Quantification
of immunogold labelling reveals enrichment of glutamate in mossy and parallel fibre
terminals in cat cerebellum, Neuroscience 19(4):1045–1050.
Somogyi, P., and Hodgson, A. J., 1985, Antisera to gamma-aminobutyric acid: III. Demonstration
of GABA in Golgi-impregnated neurons and in conventional electron microscopic sections
of cat striate cortex, J . Histochem. Cytochem. 33(3):249–257.
Storm-Mathisen, J., Leknes, A. K., Bore, A. T., Vaaland, J. L., Edminson, P., Haug, F.-M. S., and
Ottersen, O. P., 1983, First visualization of glutamate and GABA in neurones by immuno-
cytochemistry, Nature 301:517–520.
Storm-Mathisen, J., and Ottersen, O. P., 1990, Antibodies and fixatives for the immunocyto-
chemical localization of glycine, In: Ottersen, O. P., and Storm-Mathisen, J. (eds.), Glycine
Neurotransmission, Chichester: John Wiley & Sons, pp. 281–301.
Takumi, Y., Ramirez-Leon, V., Laake, P., Rinvik, E., and Ottersen, O. P., 1999, Different modes
of expression of AMPA and NMDA receptors in hippocampal synapses, Nat. Neurosci.
2(7):618–624.
Torp, R., Arvin, B., Le Peillet, E., Chapman, A. G., Ottersen, O. P., and Meldrum, B.
S., 1993, Effect of ischemia and reperfusion on the extra- and intracellular distribu-
tion of glutamate, glutamine, aspartate, and GABA in the rat hippocampus, with a
note on the effect of the sodium channel blocker BW1003C87, Exp. Brain Res. 96:365–
376.
Torp, R., Head, E., Milgram, N. W., Hahn, F., Ottersen, O. P., and Cotman, C. W., 2000, Ul-
trastructural evidence of fibrillar beta-amyloid associated with neuronal membranes in
behaviorally characterized aged dog brains, Neuroscience 96:495–506.
Torp, R., Ottersen, O. P., Cotman, C. W., and Head, E., 2003, Identification of neuronal plasma
membrane microdomains that colocalize beta-amyloid and presenilin: implications for
beta-amyloid precursor protein processing, Neuroscience 120(2):291–300.
Valtschanoff, J. G., and Weinberg, R. J., 2001, Laminar organization of the NMDA receptor
complex within the postsynaptic density, J . Neurosci. 21(4):1211–1217.
van den Pol, A. N., 1989, Neuronal imaging with colloidal gold, J . Microsc. 155(1):27–59.
van Lookeren Campagne, M., Oestreicher, A. B., van der Krift, T. P., Gispen, W. H., and Verkleij,
A. J., 1991, Freeze-substitution and Lowicryl HM20 embedding of fixed rat brain: suitability
for immunogold ultrastructural localization of neural antigens, J . Histochem. Cytochem.
39(9):1267–1279.
Wang, B. L., and Larsson, L. I., 1985, Simultaneous demonstration of multiple antigens by
indirect immunofluorescence or immunogold staining. Novel light and electron micro-
scopical double and triple staining method employing primary antibodies from the same
species, Histochemistry 83(1):47–56.
4
Cell and Tissue
Microdissection in
Combination with Genomic
and Proteomic Applications
STEPHEN D. GINSBERG, SCOTT E. HEMBY,
ELLIOTT J. MUFSON, and LEE J. MARTIN
INTRODUCTION
GENE EXPRESSION PROFILING USING FIXED TISSUES
Antemortem and Postmortem Variables
Acridine Orange Histofluorescence and Bioanalysis
REGIONAL MICRODISSECTION METHODS
SINGLE-CELL MICROASPIRATION METHODS
LASER CAPTURE MICRODISSECTION
Introduction
Positive Extraction
Negative Extraction
TRACT-TRACING COMBINED WITH DISCRETE CELL
MICRODISSECTION
RNA AMPLIFICATION
aRNA
TC RNA Amplification
MICROARRAY ANALYSIS OF MICRODISSECTED SAMPLES
LCM IN COMBINATION WITH PROTEOMIC APPLICATIONS
ADVANTAGES/LIMITATIONS
STEPHEN D. GINSBERG • Center for Dementia Research, Nathan Kline Institute,

and Departments of Psychiatry and Physiology and Neuroscience, New York University School
of Medicine, Orangeburg, NY 10962 SCOTT E. HEMBY • Department of Physiology
and Pharmacology, Wake Forest University School of Medicine, Winston-Salem, NC 27157
ELLIOTT J. MUFSON • Department of Neurological Sciences, Rush University Medical
Center, Chicago, IL 60612 LEE J. MARTIN • Division of Neuropathology, Departments
of Pathology and Neuroscience, Johns Hopkins University School of Medicine, Baltimore, MD
21205
109
110 STEPHEN D. GINSBERG et al.
APPENDIX: DETAILED METHODOLOGY

Acridine Orange Histofluorescence
Microaspiration and LCM
Tract-Tracing for Use with Microdissection
aRNA Amplification
TC RNA Amplification
SELDI-TOF
Supplies/Manufacturers
REFERENCES
Abstract: The combination of tissue microdissection protocols including discrete

cell microaspiration and laser capture microdissection with high throughput gene
expression profiling platforms such as cDNA microarrays and oligonucleotide mi-
croarrays enables the simultaneous assessment of many individual elements from a
single cell or a population of homogeneous cells. This chapter outlines in detail the
theoretical and practical background for selecting the appropriate tissues and condi-
tions amenable to expression profiling. In addition, this report illustrates the usage
of microdissection strategies and RNA amplification methodologies in concert with
array technologies using tissues harvested from the central nervous system obtained
from animal models of neurodegeneration and postmortem human brain tissues.
Keywords: brain, expression profiling, laser capture microdissection, microarray,

molecular fingerprint, RNA amplification, SELDI-TOF
I. INTRODUCTION
The brain is a complex structure with heterogeneous neuronal (e.g.,

pyramidal neurons and interneurons) and nonneuronal (e.g., glial cells,
epithelial cells, and vascular elements) cell populations. Advances in molec-
ular biology provide the tools needed to sample gene expression from spe-
cific homogeneous cell populations within defined brain regions without
potential contamination of adjacent neuronal subtypes and nonneuronal
cells, and are an important goal of twenty-first century neuroscience. How-
ever, gene expression profiling of homogeneous populations of cells is a
difficult task that demands a multidisciplinary approach including molecu-
lar biology, cell biology, neuroanatomy, and biomedical engineering. Indi-
vidual cell types are likely to have unique patterns or a mosaic of gene and
protein expression under normative conditions that is likely to be altered
in pathological states. For example, distinct cortical and subcortical regions
may serve entirely different functions and may be differentially affected in
neurodegenerative diseases (Galvin, 2004; Ginsberg et al., 1999b). Indeed,
the molecular basis of why certain neuronal cell populations are vulner-
able to neurodegeneration, often termed “selective vulnerability,” can be
elucidated by discrete cell analysis more readily than by utilizing regional
and total brain preparations (Ginsberg and Che, 2005). Thus, the pattern
CELL AND TISSUE MICRODISSECTION 111
of genomic and proteomic expression in a subpopulation of homogeneous
cells or single cells is more likely to be informative than the pattern in a whole
tissue homogenate, assuming the target population is well defined. With the
advent of modern molecular and cellular techniques, it is now possible to
isolate and study genomic DNA, RNA species, and proteins from microdis-
sected tissue sources. At present, an optimal methodology is to evaluate sin-
gle cells, identified either physiologically in living preparations (Eberwine
et al., 1992; Tkatch et al., 2000) or by immunocytochemical or histochemical
procedures in fixed cells in vitro or in vivo (Galvin and Ginsberg, 2004; Gins-
berg and Che, 2004; Ginsberg et al., 2004; Hemby et al., 2003; Kamme et al.,
2003; Mufson et al., 2002; Van Deerlin et al., 2002). Unfortunately, the quan-
tity of RNA harvested from a single cell, estimated to be ∼0.1–1.0 pg, is not
sufficient for standard RNA extraction procedures (Phillips and Eberwine,
1996; Sambrook and Russell, 2001). Both exponential polymerase chain
reaction (PCR) based analyses (Becker et al., 1996; D’Amore et al., 1997)
and linear RNA amplification including amplified antisense RNA (aRNA)
(Eberwine et al., 1992, 2001; Ginsberg et al., 1999a, 2000) and the newly
developed terminal continuation (TC) RNA amplification (Che and Gins-
berg, 2004, 2005; Ginsberg and Che, 2004; Ginsberg et al., 2004) have been
used in combination with single-cell microdissection procedures to enable
the use of microarray analysis (Eberwine et al., 2001; Ginsberg and Che,
2004). RNA amplification is a series of elaborate molecular-based methods
used to amplify genomic signals in a linear fashion from minute quantities
of starting materials for microarray analysis and other downstream genetic
applications (Fig. 4.1). In this chapter, we illustrate the utility of combining
discrete cell microdissection methodologies with RNA amplification for use
in microarray analyses as well as pairing laser capture microdissection (LCM)
with proteomic profiling for single cell and/or population cell resolution at
the protein level. Utilization of tract-tracing methods in combination with
gene expression analysis and proteomic profiling is also presented.
II. GENE EXPRESSION PROFILING USING FIXED TISSUES
A. Antemortem and Postmortem Variables
Assessment of single cell and homogeneous cell populations in optimally

prepared, perfused fixed animal tissues as well as fixed postmortem hu-
man brain tissues is desirable due to the abundance of animal and human
brain tissues that are archived within individual laboratories and brain banks
and because of the use of relevant animal models to further understand
disease mechanisms. At present, no consensus protocol exists for the fix-
ation and/or extraction of brain tissues obtained from animals or from
postmortem human tissues. Several laboratories have evaluated the effects
of different fixation protocols on RNA quality, ease of tissue microdissec-
tion, and success of microarray analysis (Bahn et al., 2001; Coombs et al.,
Figure 4.1. Schematic overview of the experimental design. Outline of the general
procedures used to perform microdissection combined with high-throughput gene
expression analysis using array platforms.
1999; Goldsworthy et al., 1999; Van Deerlin et al., 2000, 2002; Vincent et al.,
2002).
Despite potential advantages of discrete cell RNA amplification technol-
ogy, several caveats must be considered when undertaking such studies in
brain tissue. One factor is postmortem interval (PMI), or the time that
elapses between time of death and preservation of the tissue sample. PMI is
particularly relevant when obtaining postmortem human materials, as ani-
mal models can be fixed rapidly using perfusion techniques. Investigators
must be cognizant of many factors including PMI and the time from dissec-
tion to tissue preservation that may affect the quality and quantity of recov-
ered nucleic acids and proteins. Moreover, the choice of fixative for tissue
preservation is an important factor affecting RNA stability. Fixatives include
aldehydes (e.g., formalin, paraformaldehyde, and glutaraldehyde), alcohols
(e.g., ethanol and methanol), oxidizing agents, and picrates. In general,
fixatives either create cross-links or exert a precipitative effect that may alter
the native structure of macromolecules. With regards to neuroscience, alde-
hydes and alcohols are the most commonly used fixatives. Aldehydes induce
cross-linkage of lysine residues formed in proteins, and alcohols are protein
denaturants. The means by which RNA is preserved is unknown but likely
involves the inactivation of degradative enzymes. The choice of fixative must
be balanced between optimizing tissue morphology and preserving nucleic
acid integrity for evaluation. As reviewed by Van Deerlin et al. (2000, 2002),
ethanol and depolymerized 4% paraformaldehyde-based fixatives provide
optimal results for molecular-based studies. Another factor is the agonal
state of the human cases examined and the presence of overlapping neuro-
logic conditions. Agonal state refers to the nature and time period between
the onset of the terminal phase of an illness and death. The agonal state of
a patient prior to death can have profound effects on several parameters in-
cluding tissue pH, RNA stability, and protein degradation (Bahn et al., 2001;
Leonard et al., 1993; Van Deerlin et al., 2000, 2002). For example, hypoxia,
pneumonia, and protracted coma have been associated with alterations in
RNA and protein levels (Barton et al., 1993; Hynd et al., 2003; Tomita et al.,
2004). Therefore, numerous variables, including antemortem characteris-
tics, agonal state, duration of fixation, and length of storage, are relevant
parameters that should be considered prior to the initiation of molecular
studies that utilized human postmortem tissues.
B. Acridine Orange Histofluorescence and Bioanalysis
Of critical importance in discrete cell RNA assessment, as well as other

molecular procedures, is the evaluation of RNA quality and quantity. A use-
ful and relatively quick method for assessing RNA quality in tissue sections
prior to performing expression profiling studies is the use of acridine orange
(AO) histofluorescence. AO is a fluorescent dye that intercalates selectively
into nucleic acids (Mikel and Becker, 1991; von Bertalanffy and Bickis, 1956)
and has been used to detect RNA and DNA in brain tissues (Ginsberg and
Che, 2004; Topaloglu and Sarnat, 1989; Vincent et al., 2002; Zoccarato et
al., 1999). Upon excitation in the ultraviolet spectra, AO that intercalates
into RNA emits an orange-red fluorescence, whereas AO that intercalates
into DNA emits a yellowish-green fluorescence. AO can also be combined
with immunocytochemistry within tissue sections to double label cytoplas-
mic RNAs and specific antigens of interest, and is compatible with confocal
microscopy (Ginsberg et al., 1997). In brain tissue sections, the pale back-
ground of white matter tracts that lack abundant nucleic acids contrasts
AO-positive neurons. Nonneuronal cells tend to have less AO histofluores-
cence as compared to neurons and brain tumor cells (Sarnat et al., 1987),
suggesting that there is less overall RNA. It is important to note that indi-
vidual RNA species (e.g., rRNA, tRNA, and mRNA) cannot be delineated
by AO histofluorescence. Rather, this method provides a simple diagnos-

tic test that can be performed on adjacent tissue sections to ensure the
likelihood that an individual case has abundant RNA prior to performing
expensive microdissection and microarray studies. A more definitive exami-
nation of RNA quality can be obtained via bioanalysis (e.g., 2100 Bioanalyzer,
Agilent Technologies), which employs capillary gel electrophoretic method-
ologies to detect RNA quality and abundance (Che and Ginsberg, 2004,
2005; Ginsberg and Che, 2004). Bioanalysis enables visualization of results
in an electropherogram and/or digital gel formats, and provides a means
of RNA assessment at relatively high sensitivity. Investigators can also eval-
uate DNA and protein quality and abundance using bioanalysis platforms
(Freeman and Hemby, 2004).
III. REGIONAL MICRODISSECTION METHODS
Microdissection of individual cells is performed to enable downstream

gene expression profiling. Provided that procedures are performed on
fresh, frozen, or well-fixed tissue sections and ribonuclease (RNase) free con-
ditions are employed, both immunocytochemical and histochemical proce-
dures can be utilized to identify specific cell(s) of interest (Ginsberg and
Che, 2004, 2005). Several different methodologies have been used to aspi-
rate individual cells or groups of cells including single-cell microaspiration
and LCM techniques.
In addition to single-cell microdissection, regional dissections can also be
performed, which may be useful to the investigator. Regional analysis is a
powerful approach for the identification of transcripts that are enriched in
a specific region, lamina, or nuclei that differ from adjacent or connected
regions. Groups of related cells from discrete regions of brain or spinal
cord can be readily dissected from paraffin-embedded tissue sections (e.g.,
5–6 µm thick) or frozen tissue sections (e.g., 20–40 µm thick) by an ex-
perienced neuroanatomist. Unstained sections can be utilized, but optimal
cellular resolution occurs using sections prepared for immunocytochem-
ical or histochemical (e.g., Nissl stain) procedures. We have had success
in scraping away areas of the tissue section that were not desired to reveal
only the well-defined region of interest (Ginsberg and Che, 2002, 2004;
Hemby et al., 2002). Regional dissections can be performed on fresh, fixed,
or thawed tissue blocks using a stereomicroscope along with a scalpel or
micropunch. A caveat is that these approaches are highly operator depen-
dent, and can be difficult to reproduce across samples. RNA is extracted
from the resulting tissue for downstream applications, such as cDNA array
analysis, quantitative real-time PCR (qPCR), serial analysis of gene expres-
sion (SAGE), and differential display, among others (Che and Ginsberg,
2005; Ginsberg and Che, 2002; Lein et al., 2004; Zhao et al., 2001). Regional
dissections can also be used as the input source for protein in the case of
fresh and/or frozen tissues for proteomics-based applications as well as for
conventional neurochemical and immunoblotting procedures (Freeman
and Hemby, 2004; Mouledous et al., 2003; Palkovits, 1989). An advantage
of regional analyses is that limited RNA amplification is necessary to gener-
ate significant hybridization signal intensity. For example, microdissection
of the basal forebrain, hippocampal formation, midbrain, and nucleus ac-
cumbens has been performed to generate regional expression profiles in
normal brains and in pathological conditions including Alzheimer’s disease
(AD) and cocaine self-administration (Backes and Hemby, 2003; Fasulo and
Hemby, 2003; Ginsberg and Che, 2002, 2004; Tang et al., 2003). A disadvan-
tage of regional dissection procedures is the lack of single-cell resolution, as
neurons, nonneuronal cells, vascular elements, and epithelial cells will be
included in the dissection.
IV. SINGLE-CELL MICROASPIRATION METHODS
Discrimination and isolation of adjacent cell types from one another is

critical because this enables the selection of relatively pure populations
of individual cells and/or populations for subsequent analysis and avoids
potential contamination from a variety of sources including glia, vascular
epithelia, and other nonneuronal cells within the brain. One method of
isolating individual cells or populations of homogeneous cells is termed
single-cell microaspiration. Single-cell microaspiration entails visualizing an
individual cell (or cells) using an inverted microscope connected to a micro-
manipulator, microcontrolled vacuum source, and an imaging workstation
on an air table. Electrophysiology rigs can also be modified to aspirate cells
from fixed tissue sections with minor modifications. Handheld and syringe-
pump-driven vacuum sources can also be utilized; however, they are difficult
to control and may cause inadvertent damage to the tissue section. Indi-
vidual cells are carefully aspirated from the tissue section of interest, and
placed in microfuge tubes for subsequent RNA amplification (Fig. 4.2). This
methodology results in accurate dissection of the neurons of interest with
minimal disruption of the surrounding neuropil (Ginsberg, 2001; Ginsberg
et al., 2004; Hemby et al., 2002; Mufson et al., 2002). An advantage of utilizing
a single-cell microaspiration technique is the extremely high cellular (and
potentially subcellular, compartmental, and/or dendritic) level of resolu-
tion for aspiration of single elements (Crino et al., 1998; Ginsberg and Che,
2005; Hemby et al., 2003). Disadvantages include the relative difficulty of
performing the aspirating technique, experimenter error, and the lengthy
time allotment necessary to perform microaspiration, especially if multiple
cells are being acquired from different brain tissue sections. Moreover, in-
vestigators should be aware of the degree of heterogeneity of the cells of
interest and the extent to which small numbers of cells may or may not be
representative of a population of interest.
A key aspect of the success of single-cell and single-population gene
expression analysis is that different cell types can be discriminated based
Figure 4.2. Microaspiration of single neurons. (A) Representative photomicrograph

illustrating the placement of a human postmortem tissue section of the basal fore-
brain onto the microaspiration apparatus. (B) Section immunolabeled with an anti-
neurofilament antibody depicting a representative layer II entorhinal cortex stellate
cell obtained postmortem from a normal control human brain and the same section
following microdissection of the immunostained neuron (C). Scale bar: 25 µm. (D)
Human anterior nucleus basalis neuron visualized by dual immunolabeling for the
cholinergic marker p75NTR (brown cell bodies) and galanin (black punctate fibers).
The microaspirating pipette can be visualized in the left plane of the field shown
in (D). Photographs of the same tissue section are shown following microaspiration
of the cholinergic neuron at low (E) and higher magnification (F). Scale bars in
(D)–(F): 40 µm.
on their molecular fingerprint. For example, populations of neurons that

express proteins selectively such as cholinergic basal forebrain neurons
(Mufson et al., 2002, 2003) or midbrain dopaminergic neurons (Fasulo and
Hemby, 2003; Tang et al., 2003) are amenable to single-cell RNA amplifica-
tion and subsequent cDNA array analysis. Cells that lack a distinct or selective
phenotypic signature can be analyzed using a variety of Nissl and immunocy-
tochemical stains for downstream genetic applications (Ginsberg and Che,
2004, 2005; Kamme et al., 2003). Although histological stains are typically
not specific to an individual cell type or protein, much information can be
gleaned by utilizing classical histological preparations in conjunction with
contemporary protein (e.g., immunocytochemistry) and molecular biolog-
ical methodologies. For example, hematoxylin and eosin (H&E) staining
has been performed in combination with microdissection and PCR-based
strategies as well as microarray platforms using RNA amplification methods
(Becker et al., 1996; Goldsworthy et al., 1999; To et al., 1998). Moreover, we
have demonstrated that several Nissl stains including cresyl violet, H&E, and
thionin perform as well as immunocytochemistry in terms of hybridization
signal intensity detection when employing cDNA array analysis (Ginsberg
and Che, 2004). In contrast, several dyes that bind to RNAs directly, such
as AO and silver stain, do not perform well in combination with microdis-
section and subsequent cDNA array analysis (Ginsberg and Che, 2004).
V. LASER CAPTURE MICRODISSECTION
A. Introduction
The implementation of high-throughput microaspiration devices over

the last few years has enabled rapid accession of single cells and homo-
geneous cellular populations for downstream genomic and proteomic anal-
yses. Specifically, LCM is a strategy for acquiring histochemically and/or im-
munocytochemically labeled cells from in vivo and in vitro sources (Dolter
and Braman, 2001; Ehrig et al., 2001; Goldsworthy et al., 1999; Lu et al., 2004).
LCM has become a widely used and reproducible technique that was devel-
oped originally at the NIH (Bonner et al., 1997; Emmert-Buck et al., 1996).
There are two principal means of LCM: positive extraction and negative
extraction.
B. Positive Extraction
Positive extraction (a method used by the PixCell IIe from Arcturus) em-
ploys a laser source directly on the cell(s) of interest for the purpose of
microaspiration. There are four steps in positive extraction methods for
capturing cells under direct visualization and recovering biomolecules. Af-
ter locating the cells of interest in a tissue section, a small plastic cap (e.g.,
CapSure or CapSure HS LCM Cap) coated with a special thermoplastic film
is placed over the area of tissue containing the cell targets. A nondestruc-
tive, low-power, near-infrared laser pulse is then directed through the cap
at the target cell. The pulsed laser energy causes localized activation of the
thermoplastic film that extends, embraces, and adheres to the target cell.
Raising the thermoplastic cap separates targeted cells, now attached to the
film, from surrounding undisturbed tissue (Fig. 4.3). Populations of cells
attached to the cap are suitable for microscopic examination and down-
stream genetic analysis.
C. Negative Extraction
Negative extraction (or noncontact laser extraction) procedures employ

a laser source to cut around the area of interest within a tissue section, and
the microdissected material is catapulted into a microfuge tube (a method
utilized by the PALM system, PALM Microlaser Technologies). A variety of
conditions can modify the consistent success of cell capture, including tis-
sue fixation. Tissues can be fresh, frozen, or fixed in alcohol or aldehydes
Figure 4.3. LCM of granule cells. Photomicrographs of a microaspiration of cresyl vi-

olet stained human hippocampal granule cells using LCM from a 6-µm thick ethanol
fixed tissue section. (A) Section prior to LCM. (B) The cap is removed following
laser pulses over desired cells, leaving spaces where microdissected granule cells
originally resided. (C) Captured cells are visualized by placing the cap on a clean
slide for contrast. Scale bar in (A) and (B): 25 µm; (C): 30 µm.
(Goldsworthy et al., 1999; Su et al., 2004). Other parameters include optimal

section thickness (<10–14 µm) and the type of glass slide (e.g., uncoated,
charged, poly-l-lysine, and gelatin-coated) used to mount the tissue sec-
tions for subsequent microdissection. Both positive and negative extraction
methods allow captured cells and their processes to be examined microscop-
ically to confirm the identity and quality of isolated cell population(s). This
quality control step ensures validity of the results obtained from downstream
analysis. Single cells as well as dozens to hundreds of cells can be collected
by LCM instrumentation. RNA, DNA, and protein can be extracted from
microdissected cells and utilized as input sources for downstream applica-
tions such as microarray analysis, qPCR, as well as proteomics (Ehrig et al.,
2001; Fend et al., 1999; Suarez-Quian et al., 1999). We have utilized LCM
to microdissect a variety of neuronal populations including hippocampal
CA1 and CA3 neurons, dentate gyrus granule cells, and spinal motor neu-
rons (see below) from mouse brains and postmortem human brains (Che
and Ginsberg, 2004, 2005; Ginsberg and Che, 2002, 2004, 2005). LCM has
been increasingly utilized to collect cells for downstream proteomic analyses
including two-dimensional gel electrophoresis, tandem mass spectroscopy,
and antibody-based protein chips (Craven et al., 2002; Freeman and Hemby,
2004; Mouledous et al., 2003; Simone et al., 2000). In summary, the ability
to access DNA, RNA, and protein from microdissected tissue samples via
LCM-based technologies represents an exciting new avenue for studying
homogeneous populations of brain cells.
VI. TRACT-TRACING COMBINED WITH DISCRETE

CELL MICRODISSECTION
The combination of discrete cell dissection and RNA amplification allows

the investigator to make specific assertions about disease- or drug-induced
changes in gene and protein expression with unique certainty. Tract-tracing
methodologies extend this capability by providing the means to identify and
isolate specific processes and/or pathways of interest based on connectivity.
Various tracers are employed to label neurons and neuronal processes in
anterograde, retrograde, and bidirectional vectors. A discourse on the ad-
vantages and disadvantages of individual tracers is beyond the scope of this
chapter. However, selection criteria depend on the experimental paradigm
and the cell, region, or tissue type of interest (see these chapters in the cur-
rent book for additional detail) (Lanciego and Wouterlood, 2006; Molnar
et al., 2006; Reiner and Honig, 2006).
Hemby and colleagues have undertaken a series of studies to evaluate
the effects of psychotropic compounds on midbrain dopaminergic neurons
as defined by their axonal targets (Backes and Hemby, 2003; Fasulo and
Hemby, 2003). For example, in order to explore gene expression changes
in tegmental-accumbal dopamine neurons following cocaine administra-
tion, the retrograde tracer Fluorogold (FG) was iontophoretically injected
into the nucleus accumbens of rats. Following a 2-week period to allow
for sufficient transport of the tracer, rats were sacrificed and the localiza-
tion of injections was assessed by fluorescence microscopy. As depicted in
Fig. 4.4A, a number of ventral midbrain neurons in the area corresponding
B C
D E
Figure 4.4. Tract-tracing in combination with microdissection and array analysis.

(A) Representative section from rat midbrain for identification of FG-labeled cells.
FG (4%) was iontophoresed into the nucleus accumbens of rats 2 weeks prior
to sacrifice. Photomicrograph reveals significant midbrain FG-labeling within the
ventral tegmental area of Tsai. (B) A representative section within the midbrain
is shown following immunocytochemistry using an anti-FG antibody suitable for
microaspiration. Scale bar: 20 µm. (C) FG-immunoreactive individual neurons were
microdissected from the tissue section. (D) The six cells in B (1–6) were amplified
by two rounds of aRNA and labeled with 32 P-CTP. aRNA was run on a 1% dena-
turing gel [numbers above each lane correspond to neurons in (B) and (C)]. (E)
Radiolabeled aRNA from neuron #2 was used as a probe for identifying candidate
cDNAs on a custom-designed array (E1) and subcloned differential display prod-
ucts (E2). Key (E1) top row; neurofilament-L (arrow); casein kinase II b; H67559;
AA069725; T89891; AA076650; pulmonary surfactant associate protein (control);
heme oxygenase 1 (control): bottom row; CG1 protein precursor; H89874; H70730;
H89236; syntaxin (SYT; arrow); T92612; T90579; stathmin. Key (E2): blank; CSA1b;
YC3EA; YC3EB; blank; brain derived neurotrophic factor (BDNF); vector (pCR II);
vector (pBS). Accession numbers correspond to individual expressed sequence-
tagged cDNAs (ESTs).
to the ventral tegmental area were FG-positive. Since not all of the projecting
cells of the ventral midbrain are dopaminergic, the midbrain sections pro-
cessed for tyrosine hydroxylase immunofluorescence to ensure assessment
of dopamine-containing neurons using a mouse anti-tyrosine hydroxylase vi-
sualized with a Cy5 conjugated donkey anti-mouse secondary antibody. The
procedure of dual labeling provides both certainty of anatomical connectiv-
ity and antigen specificity of the cells of interest. Alternatively, if cell-specific
antigens are not available, antibodies directed against FG can be used to
identify labeled neurons for microdissection and subsequent downstream
genetic analyses (Fig. 4.4). When using a nonspecific antigen or histochem-
ical stain to identify labeled neurons, it is imperative to further characterize
the cell type post hoc using a validation technique such as qPCR. Multi-
ple tracers can also be used within the same subject to identify different cell
populations based on connectivity. Utility of employing multiple fluorescent
tracers is dependent on the absorption/emission spectra. For example, we
have used up to four tracers in rhesus monkeys to identify different popu-
lations of midbrain dopaminergic cells based on different projection paths
(i.e., dorsolateral prefrontal cortex, orbitofrontal cortex, nucleus accum-
bens, and caudate/putamen) (Freeman and Hemby, 2004).
Investigators must be cognizant of various caveats when using tract-tracing
methodologies. For example, the use of iontophoretic injections limits neu-
ronal damage and the potential interpretational confound of labeling fibers
en passant. The ability to iontophorese tracers may be limited by the chemi-
cal nature of the tracer and/or the vehicle required to solubilize the tracer.
In addition, the influence of tracer uptake on neuronal function remains
a relevant question. To date, equivocal data imply that tracers may damage
RNA and/or protein integrity of cells in which the tracer is sequestered
(Emsley et al., 2001; Franklin and Druhan, 2000). Therefore, additional
dose-response and toxicity studies are warranted to examine the extent to
which various tracers may influence RNA and protein expression in neu-
ronal populations.
VII. RNA AMPLIFICATION
A. aRNA
In order to generate a significant amount of RNA sufficient to perform

microarray analysis and related high-throughput genetic readouts, an RNA
amplification technique is often required when attempting expression pro-
filing from single neurons, groups of neurons, or microdissected regions.
PCR-based amplification methods are not optimal, as exponential amplifi-
cation can skew the original quantitative relationships between genes from
an initial population (Kacharmina et al., 1999). Linear RNA amplification
is another strategy that has been used successfully to generate enough in-
put RNA for robust hybridization signal intensity on array platforms. The
initial method of linear amplification termed aRNA amplification was devel-

oped by Eberwine and colleagues, and involves a T7 RNA polymerase-based
amplification procedure that enables quantitation of the relative abun-
dance of gene expression levels from identified single cells and populations
(Eberwine et al., 1992, 2001; Kacharmina et al., 1999; Phillips and Eberwine,
1996). The resultant amplified aRNA maintains a proportional representa-
tion of the size and complexity of the initial input mRNAs (Eberwine et al.,
1992; VanGelder et al., 1990). aRNA amplification entails the hybridiza-
tion of a 66 basepair oligonucleotide primer consisting of 24 thymidine
triphosphates (TTPs) and a T7 RNA polymerase promoter sequence [oligo
d(T)T7] to mRNAs and conversion to an mRNA–cDNA hybrid by reverse
transcriptase (Tecott et al., 1988; VanGelder et al., 1990) (Fig. 4.5). Upon con-
version of the mRNA–cDNA hybrid to double-stranded cDNA, a functional
Figure 4.5. aRNA amplification scheme. An oligo d(T)T7 primer is hybridized to

polyA+ mRNAs and a double-stranded mRNA–cDNA hybrid is formed by reverse
transcription. The double-stranded mRNA-cDNA hybrid is then converted into
double-stranded cDNA. Following the removal of tertiary structures and drop di-
alyzing the double-stranded cDNA against RNase-free water, the first round of aRNA
synthesis occurs via in vitro transcription (IVT) using T7 RNA polymerase and NTPs.
The second round of aRNA amplification begins by annealing random hexamers to
the newly formed aRNA, and synthesizing a cDNA strand. The oligo (dT)T7 primer is
then reintroduced and a double-stranded cDNA template is formed. aRNA probes
are then generated with fluorescent, biotin, or radiolabeled second-round aRNA
products.
T7 RNA polymerase promoter is formed. aRNA synthesis occurs with the ad-
dition of T7 RNA polymerase and nucleotide triphosphates (NTPs). Each
round of aRNA results in an approximate 1000-fold amplification from the
original amount of each polyadenylated [poly(A)+] mRNA in the sample
(Eberwine et al., 1992, 2001). Two rounds of aRNA are typically necessary to
generate sufficient quantities of aRNA for subsequent downstream analyses.
aRNA products are biased toward the 3 end of the transcript due to the prim-
ing at the poly(A)+ RNA tail (Kacharmina et al., 1999; Phillips and Eberwine,
1996). This 3 bias exists for all amplified aRNA products and relative levels
of gene expression can be compared (Che and Ginsberg, 2004; Madison and
Robinson, 1998; Phillips and Eberwine, 1996). Moreover, amplified aRNA
products tend not to be of full length (Ginsberg et al., 1999a; Kacharmina
et al., 1999; Phillips and Eberwine, 1996). Although aRNA is a laborious
and difficult procedure, we have generated successful results obtained from
microaspirated cells from animal model and postmortem human brain tis-
sues utilizing a wide variety of array platforms (Ginsberg et al., 1999a, 2000;
Hemby et al., 2002, 2003; McClain et al., 2005).
Several different strategies have been employed by independent labora-
tories to evaluate and improve linear RNA amplification efficiency (Iscove
et al., 2002; Klur et al., 2004; Matz et al., 1999; Wang et al., 2000). The principal
obstacle is the problematic second strand cDNA synthesis. This impediment
is not specific to the aRNA protocol. Rather, this issue is endemic to all cur-
rent RNA amplification methods. Key factors to improving RNA amplifica-
tion include increasing the efficiency of second-strand cDNA synthesis and
allowing for flexibility in the placement of bacteriophage transcriptional
promoter sequences.
B. TC RNA Amplification
We have developed a new linear RNA amplification procedure that uti-

lizes a method of terminal continuation. TC RNA essentially consists of
synthesizing first-strand cDNA complementary to the RNA template, subse-
quently generating second-strand cDNA complementary to the first-strand
cDNA, and finally IVT using the double-stranded cDNA as template (Che
and Ginsberg, 2004, 2005) (Fig. 4.6). Synthesis of the first-strand cDNA
complementary to template mRNA entails the use of two oligonucleotide
primers: a poly d(T) primer and a TC primer. The poly d(T) primer is similar
to conventional primers that exploit the poly(A)+ sequence present on most
mRNAs. The TC primer consists of an oligonucleotide sequence at the 5 ter-
minus and a short span of three cytidine triphosphates (CTPs) or guanosine
triphosphates (GTPs) at the 3 terminus. In this manner, single-strand cDNA
synthesis can be initiated by annealing a second oligonucleotide primer
complementary to the attached oligonucleotide (Che and Ginsberg, 2004).
By providing a known sequence at the 3 region of first-strand cDNA and
a primer complementary to it, hairpin loops will not form. Second-strand
Figure 4.6. Overview of the TC RNA amplification method. (A) A TC primer (con-
taining a bacteriophage promoter sequence for sense orientation) and a poly d(T)
primer are added to the mRNA population to be amplified (green rippled line).
First-strand (blue line) synthesis occurs as an mRNA–cDNA hybrid and is formed
after reverse transcription and terminal continuation of the oligonucleotide primers.
Following RNase H digestion to remove the original mRNA template strand, second-
strand (red line) synthesis is performed using Taq polymerase. The resultant double-
stranded product is utilized as template for IVT, yielding high-fidelity, linear RNA
amplification of sense orientation (green rippled lines). (B) Schematic similar to
(A), illustrating the TC RNA amplification procedure amplifying RNA in the anti-
sense orientation (yellow rippled lines).
cDNA synthesis can be performed with robust DNA polymerases, such as

Taq, and the TC reaction is highly efficient. One round of amplification is
sufficient for downstream genetic analyses (Che and Ginsberg, 2004; Gins-
berg and Che, 2004). Furthermore, TC RNA transcription can be driven us-
ing a promoter sequence attached to either the 3 or the 5 oligonucleotide
primers. Therefore, transcript orientation can be in an antisense orientation
(similar to conventional aRNA methods) when the bacteriophage promoter
sequence is placed on the poly d(T) primer or in a sense orientation when
the promoter sequence is attached to the TC primer, depending upon the
design of the experimental paradigm (Fig. 4.6). TC RNA amplification of-
fers high sensitivity, flexibility, and throughput capabilities for downstream
genetic analyses. Following TC RNA amplification, a large proportion of
genes can be assessed quantitatively as evidenced by bioanalysis and cDNA
microarray analysis in mouse and human postmortem brain tissues (Che
and Ginsberg, 2004; Ginsberg and Che, 2002, 2004, 2005; Mufson et al.,
2002). Robust linear amplification is consistently observed. Amplification
efficiency of approximately 2500- to 3000-fold is demonstrated with com-
mercially available purified mRNAs, and approximately 1000- to 1500-fold
amplification is found after one round using biological samples of RNA
extracted from a variety of brain sources (Che and Ginsberg, 2004). Results
indicate a high degree of expression level similarity for high, moderate,
and low expressed genes using the TC RNA amplification method. The
threshold of detection of genes with low hybridization signal intensity is
also greatly increased, as many genes that are at the limit of detection us-
ing conventional aRNA can be readily observed with the TC method (Che
and Ginsberg, 2004). Importantly, increased sensitivity appears greatest for
genes with relatively low abundance. Moreover, background hybridization
is significantly attenuated when using TC RNA amplification (Ginsberg and
Che, 2002, 2004; Mufson et al., 2002).
VIII. MICROARRAY ANALYSIS OF MICRODISSECTED SAMPLES
Once an RNA amplification procedure is utilized to increase the input

source of RNA species, biotinylated, fluorescent, or radiolabeled probes can
be generated for subsequent hybridization to microarray platforms. Tech-
nical advances have fostered the development of high-density microarrays
that allow for high-throughput analysis of hundreds to thousands of genes
simultaneously. Synthesis of cDNA microarrays entails adhering cDNAs or
ESTs to solid supports such as glass slides, plastic slides, or nylon membranes
(Brown and Botstein, 1999; Eisen and Brown, 1999). A parallel technology
uses photolithography to adhere oligonucleotides to array media (Lock-
hart et al., 1996). Gene expression is assayed by harvesting total RNA or
mRNA from sample tissues, labeling either by radioactive or by fluorescent
methods, and hybridizing the labeled probes to arrays (Fig. 4.7). Arrays
are washed to remove nonspecific background hybridization, and imaged
using a laser scanner for biotinylated/fluorescently labeled probes and a
phosphor imager for radioactively labeled probes. The specific signal inten-
sity (minus background) of amplified RNA bound to each probe set (e.g.,
oligonucleotides or cDNAs/ESTs) is expressed as a ratio of the total hy-
bridization signal intensity of the array, thereby minimizing variations due
to differences in the specific activity of the probe and the absolute quantity
of probe present. Gene expression data collected using single cells and/or
homogeneous populations via RNA amplification and microarray analysis
do not allow absolute quantitation of mRNA levels, but generate an expres-
sion profile of the relative changes in mRNA levels (Eberwine et al., 2001;
Galvin and Ginsberg, 2004; Ginsberg and Che, 2002; Ginsberg et al., 2004;
Hemby et al., 2003; Madison and Robinson, 1998; Mufson et al., 2002). Rela-
tive changes in individual mRNAs are analyzed by univariate statistics [e.g.,
analysis of variance (ANOVA) with post hoc Neumann–Keuls test] for indi-
vidual comparisons (Ginsberg et al., 1999a, 2000; Hemby et al., 2002; Mufson
et al., 2002). Differential expression greater than approximately twofold is
accepted conventionally as relevant for further examination (Freeman and
Hemby, 2004; Galvin and Ginsberg, 2004; Ginsberg et al., 2004; Hemby et al.,
2003; Mirnics et al., 2000). Differentially expressed genes can be clustered
Figure 4.7. Representative array platforms. (A) A custom-designed cDNA array with
30 lanes is depicted. cDNAs are stained with bromophenol blue to show equal loading
(top panel). The same array (A; bottom panel) is shown following hybridization with
radiolabeled aRNA from a single CA1 neuron. Note the differential expression and
abundance of cDNAs. (B) A portion of a high-density cDNA microarray, illustrating
aRNA probes generated from neurofibrillary tangle (NFT)-bearing neurons (first
panel; red), normal CA1 neurons (second panel; green), and an overlay of both
(third panel). Yellow shows similar intensities for NFTs and normal neurons, green
indicates a down regulation in NFTs relative to normal CA1 neurons, and red denotes
an up regulation.
into functional protein categories for multivariate coordinate gene expres-

sion analyses (Freeman and Hemby, 2004; Ginsberg and Che, 2002; Kotlyar
et al., 2002). Computational analysis is critical for optimal use of microar-
rays due to the enormous volume of data that is generated from a single
probe. Additionally, access to relational databases is desirable, especially
when evaluating hundreds of ESTs that may or may not be linked to genes
(and subsequent proteins) of known function.
IX. LCM IN COMBINATION WITH PROTEOMIC APPLICATIONS
Genomic and proteomic expression studies of tissues can be confounded

easily because the cells of interest, for example, neurons, exist within a
heterogeneous environment that contains many types of cells. LCM allows
the isolation of neurons on a single-cell basis for cell-specific analysis (see
section “Laser Capture Microdissection”). Once captured, these relatively
pure cell populations can be analyzed using a variety of methods, including
downstream proteomic applications. Specifically, intact proteins and mRNA
can be recovered from LCM captured cells and analyzed quantitatively. For
protein studies, the surface-enhanced laser desorption/ionization time-of-
flight mass spectrometry (SELDI-TOF MS) approach is an excellent way to
evaluate neuron proteomics using the ProteinChip Biology System (PBSII,
Ciphergen Biosystems) (Issaq et al., 2002). The PBSII uses SELDI-TOF
MS to retain proteins on a solid-phase chromatographic surface that are
subsequently ionized and detected by TOF MS. Protein profiles can be
generated from as few as 25–50 cells (Paweletz et al., 2001). The SELDI-TOF
MS technology consists of three major components: the ProteinChip
array, the chip reader apparatus, and the software. The ProteinChip array
is a 10-mm-wide × 80-mm-long platform having 8 (or 16) 2-mm spots
comprising a specific chromatographic surface. Each spot contains either
a chemically (e.g., anionic, cationic, hydrophobic, and hydrophilic) or
biochemically (e.g., antibody and receptor) treated surface for retaining
entire classes of proteins or single target proteins, respectively. Chemically
treated surfaces retain whole classes of proteins, while surfaces treated
with biochemical agent (e.g., antibody or other type of affinity reagent)
serve as bait and will interact with a specific target protein. Biochemically
treated arrays are custom-made by the user. Sample (1–10 µl of protein
extract from captured neurons) is applied to the surface, with protein
specificity being achieved via the surface treatment and the application of
solvents/buffers and washes. After an energy-absorbing molecule solution
is added, the array is inserted into the ProteinChip reader to measure the
molecular weight and relative amounts of bound proteins. The reader is
a laser desorption ionization mass spectroscopy instrument equipped with
a pulsed ultraviolet nitrogen laser. Laser activation of the sample causes its
desorption/ionization and liberation of gaseous ions from the ProteinChip
arrays. The ions enter the TOF MS module that measures the mass-to-charge
ratio of each protein. Protein detection is displayed as a series of peaks.
The readout generated by the TOF MS analysis is a trace showing the
relative abundance versus the molecular weights of the detected proteins.
The software converts the peak trace into a simulated one-dimensional
gel electrophoresis display to identify differences in protein abundances
between samples. This technology can be used to determine in normal
and degenerating neurons at specific structural stages patterns of protein
expression and the levels of specific proteins and specific posttranslationally
modified proteins (e.g., cleaved, phosphorylated, and acetylated proteins).
A practical example illustrating the union of LCM and proteomics using
postmortem human spinal cord is provided in Fig. 4.8. LCM can be used
to isolate individual spinal motor neurons, yielding a pure cell preparation
for downstream proteomic applications (Fig. 4.8A,B). Motor neurons are
Figure 4.8. LCM and SELDI-TOF MS analysis of human ALS motor neurons. (A)
Visualization of spinal cord motor neurons (arrows) in a Ponceau S stained cryostat
section of human lumbar cord. Scale bar: 75 µm. (B) Human spinal cord section
after harvesting motor neurons via LCM. The open empty circles in the section
(arrows) show where the motor neurons were formerly located. Scale bar: 100 µm.
(C) Confirmation of cell capture by direct visualization of caps with isolated motor
ideal for LCM because they are relatively large neurons with a low packing
density. For example, we have used Ponceau S stained tissue sections to
isolate target motor neurons from the surrounding neuropil for SELDI-TOF
analysis. Captured cells can be viewed microscopically for confirmation (Fig.
4.8C). Moreover, Western blotting can be used to characterize the purity of
human LCM samples. Astrocyte contamination, as assessed by glial fibrillary
acidic protein (GFAP), is negligible. A high level of the neuronal nuclear
protein NeuN and a very low level of GFAP in motor neuron cell lysates
confirm the neuronal purity of the LCM samples (Fig. 4.8D). Even with long
exposure times GFAP levels are barely detectable in motor neuron samples.
Conversely, when cells with an astrocyte morphology are captured, the GFAP
level is high and NeuN was not detectable. These pure motor neuron and
astrocyte populations can be used for precise downstream molecular analysis
of cell-specific events.
An example of the high resolution afforded by these types of applica-
tions is that the cell death protein, cleaved caspase-3, can be measured di-
rectly in human motor neurons obtained postmortem from normal control
brains and subjects with amyotrophic lateral sclerosis (ALS). Approximately
14,000–15,000 motor neurons were isolated from fresh cryostat sections
(stained with Ponceau S) from control lumbar spinal cords (three different
cases for a total of ∼45,000 motor neurons) and approximately 8000–10,000
motor neurons from ALS spinal cords (three different cases for a total of
∼30,000 motor neurons) that were in the somatodendritic attritional stage
of degeneration (Martin, 1999). Cleaved caspase-3 antibody (Cell Signaling
Technology) was covalently bound to the surface of preactivated
ProteinChip arrays (PS2 arrays, affinity capture surfaces). Covalently bound
←
Figure 4.8. (Cont.) neurons. The harvested motor neurons are surrounded by ther-
moplastic film. Scale bar: 225 µm. (D) Assessment of the purity of cell isolation by
Western blotting of lysates of LCM acquired cells for NeuN and GFAP. Astrocyte con-
tamination, as assessed by GFAP, is negligible. The high level of NeuN and very low
level of GFAP in motor neuron cell lysates confirm the neuronal purity of the LCM
samples. Even with long exposure times GFAP levels were only barely detectable in
motor neuron samples from ALS cases. (E) Protein profiling in human ALS and
control motor neurons by SELDI-TOF MS. PS2 ProteinChip arrays were used to
isolate and quantify cleaved caspase-3. After sample preparation, the ProteinChip
arrays were analyzed by laser desorption ionization TOF MS. For comparison pur-
poses, the software of the SELDI Ciphergen system displays the data as a spectra
view. Recombinant cleaved caspase-3 served as a positive control for identifying the
molecular weights of the cleaved subunits. (F) Quantification of cleaved caspase-3
in human motor neurons. To identify differences in protein abundances between
control and ALS cases, the software converts the peak trace into a simulated one-
dimensional gel electrophoresis display to measure protein abundance. The values
are mean ± standard deviation (SD). The measurements are normalized to parallel
analyses of NeuN levels in the lysates. ALS motor neurons have significantly elevated
( p < 0.001) levels of cleaved caspase-3 compared to age-matched controls (ANOVA
with post hoc Neumann–Keuls test).
immunoglobulin (IgG) served as an antibody negative control. Crude cell

lysates of captured motor neurons were applied to the ProteinChip with
bound antibody, incubated, washed, and analyzed in a Ciphergen Pro-
teinChip reader. Purified recombinant active caspase-3 (Medical and Biolog-
ical Laboratories) was used as a positive control (Fig. 4.8E, upper retentate
map) and it displayed prominent peaks at ∼11.3, 13.8, and 14 kDa. Pep-
tides corresponding to cleaved caspase-3 were found in ALS motor neurons
(Fig. 4.8E, middle retentate map, peaks at ∼11 and 14 kDa). No caspase-3
signal was observed in ALS or control cases with the nonspecific IgG bound
to the chip. In age and postmortem delayed-matched control motor neu-
rons, peaks of similar molecular weight were either not above background
or were low (Fig. 4.8E, control motor neurons, lower retentate map). Quan-
tification of cleaved caspase-3 (13.8 kDa protein) levels in control and ALS
spinal motor neurons revealed highly significant ( p < 0.001) increases in
ALS motor neurons (Fig. 4.8F). The immunoassay results had remarkably
low variability, likely due in part to the homogeneous population of motor
neurons that was accrued via LCM.
Cell-based assays are critical for evaluating changes in protein cells under-
going degeneration. The use of tissue homogenate-based assays is subopti-
mal for this purpose because tissue homogenates cannot afford sufficient
resolution. Thus, the interpretation of homogenate-based assays of tissues
with a heterogeneous cellular composition is suspect. The strategy for mea-
suring proteins in cells acquired via LCM represents a major step forward
in the analysis of cell-specific degenerative events (Freeman and Hemby,
2004; Mouledous et al., 2003). LCM and proteomic approaches are feasi-
ble and practical to apply, providing the availability of the equipment and
service maintenance. The cellular resolution attained by LCM-based tech-
nologies for downstream proteomic applications is optimal for these types
of in vivo investigations. LCM dramatically decreases the noise in the assays
by minimizing contaminating cells. The integrity of the proteins and pep-
tide fragments to be analyzed is maintained (Freeman and Hemby, 2004;
Mouledous et al., 2003). A pitfall of LCM-based technologies is that they are
labor-intensive, and the number of captured cells required for quantitative
signal detection is significant. However, the data gleaned by these types of
studies represent definitive cell-specific events.
In summary, the analysis of human material as well as of appropriate ani-
mal models of neurodegeneration will provide direct results on the molecu-
lar events occurring within diseased cells. Employing LCM in combination
with SELDI-TOF is ideal for dealing with asynchrony of neurodegenera-
tion by providing structural–molecular correlations on neurons sampled at
similar (and different) stages of degeneration. Human tissue experiments
must be controlled at several levels with disease-specific controls that are
matched for age, agonal state, and postmortem delay (Bahn et al., 2001;
Hynd et al., 2003; Van Deerlin et al., 2000, 2002). Moreover, interregional
controls within the same case are necessary to rule out the possibility that ob-
served changes are due to agonal state and tissue autolysis. Parallel studies
of neurodegeneration in optimally prepared animal models can provide
valuable side-by-side comparisons of relevant molecules. Thus, molecular
profiles can be then brought into the context of the structural phenotype
of the observed degeneration. Moreover, if captured cells (using regional,
microaspiration, and/or LCM-based technologies) obtained from an ani-
mal model display expression profile(s) that differ vastly from the human
condition it is designed to model, applicability comes into question. Ulti-
mately, a particular animal model may be deemed to be inappropriate for
further study within the context of single cell or homogeneous population
cell analysis based on disparities in genomic and proteomic profiles from a
human condition that they were designed to mimic.
X. ADVANTAGES/LIMITATIONS
A variety of tissues and cells can be used to extract mRNA for gene profil-
ing experiments. When employing mRNA as a starting material, one cannot
overemphasize the importance of the preservation of RNA integrity. RNA
species are particularly sensitive to degradation by RNase. RNases are found
in virtually every cell type, and they retain their activity over a broad pH range
(Blumberg, 1987; Farrell, 1998). Thus, RNase-free precautions are essential
for all microdissection-based studies. All biological samples require prompt
handling, either through rapid RNA extraction, flash freezing, or through
fixation to minimize degradation.
Reproducibility of single-cell expression profiling is a critical param-
eter that is improving. Advances at the level of tissue dissection, RNA
amplification, microarray platforms, and developing powerful statistical
methods will ultimately lead to greater utility and flexibility of these tech-
nologies. Recent advances include the utilization of pooled populations of
individual cell types to reduce variability in expression levels yet maintain an
expression profile for a single cell type. The likelihood of generating highly
reproducible data is increased greatly by replicate array analysis of aliquots
of the same amplified RNA sample. Validation of array results is important,
and several independent alternative techniques are quite useful to repro-
duce changes seen on an array platform such as qPCR, SAGE, and/or in
situ hybridization, among others.
When deciding whether or not to employ microaspiration and/or high-
throughput array technologies, the most important aspect to consider is
the question the researcher is interested in answering, and determining
the method(s) that would be best suited to perform the experiment. Once
a researcher has decided that an array experiment is appropriate, much
consideration needs to go into sample size and preparation, tissue and/or
cell quality, and importantly, input amount of RNA that will likely be gen-
erated. If the input source is a small sample of population of cells captured
by LCM, then an RNA amplification method is requisite. A researcher then
needs to calculate laboratory and technical effort, cost, and goals in order
to determine the commitment level that will be needed to carry out array
experiments. Sample preparation, RNA amplification, array hybridization,
and array analysis usually require a long-term commitment, as many inves-

tigators have found out much to their dismay. A qPCR experiment would
be more useful, for instance, if a researcher is trying to assess the regula-
tion of a single gene product (or splice variants/isoforms of an individual
gene family). An array experiment may yield the desired result, along with
a plethora of potential data on dozens, hundreds, thousands of genes that
may not be germane to central hypothesis. Quantitation of array platforms
is typically relative, whereas qPCR can be more direct, using cycle threshold
calculations as well as copy number (but this is difficult and not typically
feasible for the casual qPCR user). qPCR can be reliable and cost effective,
provided that the primer design is performed optimally. Alternatively, the
solution hybridization afforded by RNase protection assays cannot be under-
estimated. RNase protection assays are especially useful when input sources
of RNA are abundant, such as with in vitro paradigms. However, RNase
protection assays in tissue sections, particularly fixed tissues, are not highly
recommended. In summary, cDNA and oligonucleotide arrays are spectac-
ular tools for high-throughput analyses within a myriad of paradigms and
tissue sources. qPCR is a useful medium to low-throughput method that di-
rectly assays genes of specific interest. Our laboratory strategy is to combine
the use of both assays, by defining expression profiling patterns on microar-
ray platforms and validating individual gene level changes by independent
qPCR analyses (Ginsberg and Che, 2004, 2005; Ginsberg et al., 2004).
The combination of discrete cell microdissection procedures with mi-
croarray technologies allows for high-resolution, high-throughput expres-
sion profiling of dozens to hundreds to thousands of genes and proteins
simultaneously from a single neuron or from a group of similar neurons.
The next level of understanding of cellular and molecular mechanisms un-
derlying normative function and pathological conditions lies in the ability
to combine these aforementioned technologies with appropriate models
to recapitulate the structure and connectivity of these systems in vivo and
in vitro. Complex biological processes are not likely to be governed solely
by the action of a single isolated gene. Rather, coordinate interactions of
a multiplicity of genes may regulate normative function. When these gene
programs or mosaics undergo increased or decreased expression during the
lifespan, they may contribute to the mechanisms underlying disease patho-
genesis. Single-cell and population-cell profiling techniques coupled with
microarray platforms have the potential to quantify simultaneous expres-
sion levels of numerous genes and proteins in a given cell, thereby allowing
for previously unobserved gene interactions, and ultimately protein interac-
tions, to become more evident. Independent verification of individual gene
level changes discovered by microarray analysis by alternate techniques is
a critical component to a research program. Thus, a combination of mul-
tidisciplinary approaches is ideal for verification of gene expression level
alterations, with the explicit knowledge that the sum of the evaluations may
be more informative and reflect the actual biology of the system than an
individual method.
APPENDIX: DETAILED METHODOLOGY
A. Acridine Orange Histofluorescence
This protocol was developed for AO histochemistry using animal model

and human postmortem tissues embedded in paraffin (Ginsberg et al., 1997,
1998; Mikel and Becker, 1991). Briefly, tissue sections are deparaffinized in
xylene, graded through a descending ethanol series, and placed in distilled
water for 5 min. The sections are placed in a 0.2 M dibasic sodium phos-
phate/0.1 M citric acid (SC; pH 4.0) solution for 5 min prior to staining
with AO (10 µg/ml; Sigma) in SC for 15 min. The sections are rinsed three
times in the SC buffer, immersed in 50% ethanol in phosphate-buffered
saline (PBS; 0.12 M; pH 7.4) for 2 min, cleared in xylene, and mounted
with an antifading medium (Vectashield, Vector Laboratories). To reduce
the intense autofluorescence of lipofuscin granules that are abundant in
senescent human brain, selected tissue sections can be pretreated with ei-
ther 0.05% potassium permanganate in PBS for 20 min followed by 0.2%
potassium metabisulfite/0.2% oxalic acid in PBS for 30 s (Guntern et al.,
1992) or 0.3% Sudan Black B (Sigma; w/vol in 70% ethanol) for 10 min
(Yao et al., 2003) prior to AO histochemistry.
B. Microaspiration and LCM
Our laboratory utilizes a Nikon inverted microscope with MetaMorph

5.0 software (Universal Imaging Corporation), an Eppendorf micromanip-
ulator, and an Eppendorf Transjector for single-cell microaspiration. Im-
munostained or histochemically stained tissue sections are not coverslipped
or counterstained and are immersed in RNase-free 0.1 M Tris (pH 7.4)
(Ginsberg et al., 1999a, 2000; Hemby et al., 2002, 2003). Individual cells are
carefully aspirated from the tissue section and placed in microfuge tubes for
subsequent TC RNA amplification. LCM is performed on immunostained
or histochemically stained tissue sections that are dehydrated in an ascend-
ing series of ethanol and placed in fresh xylenes for a minimum of 15 min.
LCM is performed using a PixCell IIe instrument (Arcturus). Caps contain-
ing desired captured cells are inverted into a microfuge tube containing
Trizol reagent (Invitrogen) prior to initiating TC RNA amplification.
C. Tract-Tracing for Use with Microdissection
Fluorogold (4%; Fluorochrome Inc.) is dissolved in isotonic saline and

backfilled into an autoclaved glass micropipette with the tip tapered
to ∼12–15 µm. FG is iontophoresed into brain regions via 5 µA of current
with a 5-s on/off cycle for 10 min. Micropipettes should remain in place
5 min following infusion to allow proper diffusion into the neural tissue
and prevent diffusion up the injection tract. FG is visualized using fluores-

cence microscopy with fluorescence excitation filter set at 340–380 nm and a
barrier filter at 430 nm (Schmued and Heimer, 1990). Microaspiration and
LCM are performed on uncoverslipped tissue sections as described above.
D. aRNA Amplification
For aRNA amplification, an oligo d(T)T7 primer (20 ng/µl) is hybridized

directly to poly(A)+ mRNAs (Tecott et al., 1988). A double-stranded mRNA–
cDNA hybrid is formed by reverse-transcribing the primed mRNAs with
dNTPs (1 mM) and 10 U reverse transcriptase (AMVRT, Sekigaku) for 3 h
at 42◦ C (VanGelder et al., 1990). The double-stranded mRNA–cDNA hybrid
is converted into double-stranded cDNA by heat denaturing for 5 min at
85◦ C followed by the addition of dNTPs (1 mM) and 10 U T4 DNA poly-
merase (Invitrogen) and 10 U Klenow (Invitrogen), forming a functional
T7 RNA polymerase promoter. Following the removal of tertiary structures
and drop dialyzing the double-stranded cDNA against RNase-free water, the
first round of aRNA synthesis occurs using 2000 U T7 RNA polymerase (Epi-
centre) and NTPs at 37◦ C for 4 h (Eberwine et al., 2001). The second round
of aRNA amplification begins by annealing random hexamers to the newly
formed aRNA, and synthesizing a cDNA strand. The oligo d(T)T7 primer
is then reintroduced, which binds to the poly(A)+ sequence on the newly
synthesized cDNA strand, and a double-stranded cDNA template is formed.
aRNA is then tagged with fluorescent, biotinylated, or radiolabeled reagents
to enable hybridization to the desired cDNA microarray, oligonucleotide
platform, or membrane-based array.
E. TC RNA Amplification
TC RNA amplification consists of immersing microdissected cells or re-

gional dissections in 250 µl of proteinase K solution (50 µg/ml; Ambion)
for 12 h at 37◦ C prior to extraction in Trizol reagent (Invitrogen). RNAs are
reverse-transcribed in the presence of the poly d(T) primer (10 ng/µl) and
TC primer (10 ng/µl) in 1X first strand buffer (Invitrogen), 1 mM dNTPs,
5 mM dithiothreitol (DTT), 20 U of RNase inhibitor, and 5 U reverse tran-
scriptase (Superscript III; Invitrogen) (Che and Ginsberg, 2004). The syn-
thesized single-stranded cDNAs are converted into double-stranded cDNAs
by adding into the reverse transcription reaction the following: 10 mM Tris
(pH 8.3), 50 mM KCl, 1.5 mM MgCl2 , and 0.5 U RNase H (Invitrogen)
in a total volume of 99 µl. Samples are placed in a thermal cycler and
second-strand synthesis proceeds as follows: RNase H digestion step 37◦ C,
10 min; denaturation step 95◦ C, 3 min, annealing step 50◦ C, 3 min; elonga-
tion step 75◦ C, 30 min. 5 U (1 µl) Taq polymerase (PE Biosystems) is added
to the reaction at the initiation of the denaturation step (i.e., hot start)
(Che and Ginsberg, 2004). The reaction is terminated with 5 M ammonium
acetate. The samples are extracted in phenol:chloroform:isoamyl alcohol
(25:24:1) and ethanol precipitated with 5 µg of linear acrylamide (Am-
bion) as a carrier. The solution is centrifuged at 14,000 rpm and the pellet is
washed once with 95% ethanol and air-dried. The cDNAs are resuspended
in 20 µl of RNase-free H2 O and drop dialyzed on 0.025 µm filter mem-
branes (Millipore) against 50 ml of 18.2 M RNase-free H2 O for 2 h. The
sample is collected off the dialysis membrane, and hybridization probes are
synthesized by IVT using a fluorescent labeling kit (e.g., Cy3 and/or Cy5
labeling; Enzo Life Sciences) as per manufacturer’s instructions. Alterna-
tively, hybridization probes can be generated for membrane-based arrays
using 33 P incorporation in 40 mM Tris (pH 7.5), 7 mM MgCl2 , 10 mM NaCl,
2 mM spermidine, 5 mM of DTT, 0.5 mM of ATP, GTP, and CTP, 10 µM of
cold UTP, 20 U of RNase inhibitor, 1000 U T7 RNA polymerase (Epicentre),
and 40 µCi of 33 P-UTP (GE Healthcare) for 4 h at 37◦ C (Che and Ginsberg,
2004).
F. SELDI-TOF
Patients were diagnosed with ALS by neurological examination using

the El Escorial criteria (Brooks, 1994). Postmortem central nervous system
tissues from these individuals were obtained from the Division of Neu-
ropathology, Human Brain Resource Center, Johns Hopkins University
School of Medicine. Neuropathological evaluation confirmed the clinical
diagnosis of ALS (Martin, 1999). The cases studied were sporadic ALS. Post-
mortem samples of spinal cord from age-matched control individuals with-
out neurological disease (n = 3) and patients with ALS (n = 3) were selected
randomly for analysis.
At autopsy, spinal cord blocks (L5 segment) were dissected and snap-
frozen in liquid nitrogen and stored at −80◦ C. Lumbar spinal cord blocks
were sectioned at 12 µm in a cryostat, thaw-mounted onto Superfrost
charged glass microscope slides, and stored at −80◦ C. For LCM, selected
slides were stained briefly with Ponceau S prepared in a protease inhibitor
cocktail and air-dried. A PixCell II LCM system was used for acquiring mo-
tor neurons using a laser spot size of 30 or 60 µm. Motor neuron isolates
were lysed with cold 20 mM Tris-HCl (pH = 7.4) containing 10% (wt/vol) su-
crose, 20 U/ml aprotinin (trasylol), 20 µg/ml leupeptin, 20 µg/ml antipain,
20 µg/ml pepstatin A, 20 µg/ml chymostatin, 0.1 mM phenylmethylsulfonyl
fluoride, 10 mM benzamidine, 1 mM EDTA, and 5 mM EGTA. All of the
protease inhibitors were purchased from Sigma. Protein concentration was
determined by bicinchoninic acid protein assay kit (Pierce).
PS2 ProteinChip arrays (Ciphergen) were used for SELDI-TOF. PS2 arrays
are recommended for use in covalent immobilization of biomolecules for
the subsequent capture of target proteins from complex biological samples.
PS2 arrays have spots that are preactivated with epoxide chemistry that co-
valently bind to free primary amine groups on the surface of biomolecules
(e.g., antibodies) for immunoassays. The stably immobilized biomolecules
capture proteins from biological samples through specific, noncovalent in-

teractions. The PS2 surface is especially recommended when the aim is
to include sensitive detection, low nonspecific binding, and target protein
concentrations at less than 1% of total protein.
G. Supplies/Manufacturers
18.2 M RNase-free H2 O (Nanopure Diamond, Barnstead, Dubuque,

IA)
AMVRT (Sekigaku, Falmouth, MA)
ATP (Invitrogen, Carlsbad, CA)
Acridine Orange (Sigma, St. Louis, MO)
Ammonium acetate (Sigma)
Antipain (Sigma)
Aprotinin (Sigma)
Bioanalyzer (2100, Agilent Technologies, Palo Alto, CA)
Benzamidine (Sigma)
Chymostatin (Sigma)
Citric acid (Sigma)
Cleaved caspase-3 antibody (Cell Signaling Technology, Beverly, MA)
Caspase-3 active recombinant (Molecular and Biological Laboratories
Wobum, MA)
Cresyl violet (Sigma)
dNTPs (Invitrogen)
DTT (Sigma)
EDTA (Sigma)
EGTA (Sigma)
Filter membranes (Millipore, Billerica, MA)
1X First strand buffer (Invitrogen)
Fluorescent labeling kit (Enzo Life Sciences, Farmingdale, NY)
Fluorogold (Fluorochrome Inc., Englewood, CO)
Klenow (Invitrogen)
Leupeptin (Sigma)
Linear acrylamide (Ambion)
MetaMorph 5.0 software (Universal Imaging Corp., Downingtown, PA)
Micromanipulator (Brinkmann-Eppendorf, Westbury, NY)
NTPs (Invitrogen)
Oxalic acid (Sigma)
PALM (PALM Microlaser Technologies, Bernried Germany)
P33 -UTP (GE Healthcare, Piscataway, NJ)
PBSII (Ciphergen Biosystems, Fremont, CA)
Pepsatin (Sigma)
Phenol:chloroform:isoamyl alcohol (Invitrogen)
Phenylmethsulfonyl fluoride (Sigma)
PixCell IIe LCM (Arcturus, Mountain View, CA)
Ponceau S (Sigma)
Potassium metabisulfate (Sigma)
Potassium permanganate (Sigma)
Protein assay kit (Pierce, Rockford, IL)
Proteinase K (Ambion, Austin, TX)
Purified mRNAs (Invitrogen)
Reverse transcriptase (Superscript III, Invitrogen)
RNase H (Invitrogen)
RNase inhibitor (Invitrogen)
Spermidine (Sigma)
Sudan Black B (Sigma)
Sucrose (Sigma)
Taq polymerase (PE Biosystems, Foster City, CA)
T4 DNA polymerase (Invitrogen)
T7 RNA polymerase (Epicentre, Madison, WI)
Tris (Sigma)
Trizol reagent (Invitrogen)
UTP (Invitrogen)
Vectasheild (Vector Laboratories, Burlingame, CA)
Acknowledgments. We thank Shaoli Che, M.D., Ph.D., and Scott E.

Counts, Ph.D., for their continued efforts on these projects. Support for
these projects comes from the NINDS (NS34100, LJM; NS43939, SDG;
NS48447, SD6), NIA (AG10668, EJM & SDG; AG14449, EJM & SDG;
AG21661, EJM AG05146, LJM; AG16282, LJM), NCI (CA94520; SDG),
NIDA (DA013772, SEH; DA013234, SEH), NIMH (MH074313, SEH) and
Alzheimer’s Association (SDG). We also express our appreciation to the
families of the patients studied here, who made this research possible.
REFERENCES
Backes, E., and Hemby, S. E., 2003, Discrete cell gene profiling of ventral tegmental dopamine
neurons after acute and chronic cocaine self-administration, J. Pharmacol. Exp. Ther. 307:
450–459.
Bahn, S., Augood, S. J., Ryan, M., Standaert, D. G., Starkey, M., and Emson, P. C., 2001, Gene
expression profiling in the post-mortem human brain—no cause for dismay, J. Chem.
Neuroanat. 22:79–94.
Barton, A. J. L., Pearson, R. C. A., Najlerahim, A., and Harrison, P. J., 1993, Pre- and postmortem
influences on brain RNA, J. Neurochem. 61:1–11.
Becker, I., Becker, K.-F., Röhrl, M. H., Minkus, G., Schütze, K., and Höfler, H., 1996, Single-cell
mutation analysis of tumors from stained histologic slides, Lab. Invest. 75:801–807.
Blumberg, D. D., 1987, Creating a ribonuclease-free environment, Methods Enzymol. 152:20–24.
Bonner, R. F., Emmert-Buck, M., Cole, K., Pohida, T., Chuaqui, R., Goldstein, S., and Liotta,
L. A., 1997, Laser capture microdissection: molecular analysis of tissue, Science 278:1481–
1483.
Brooks, B. R., 1994, El Escorial World Federation of Neurology criteria for the diagnosis of
amyotrophic lateral sclerosis, J. Neurol. Sci. 124:96–107.
Brown, P. O., and Botstein, D., 1999, Exploring the new world of the genome with DNA mi-
croarrays, Nat. Genet. 21(Suppl):33–37.
Che, S., and Ginsberg, S. D., 2004, Amplification of transcripts using terminal continuation,
Lab. Invest. 84:131–137.
Che, S., and Ginsberg, S. D., 2006, RNA amplification methodologies, In: Progress in RNA
Research, Nova Science Publishing, in press.
Coombs, N. J., Gough, A. C., and Primrose, J. N., 1999, Optimisation of DNA and RNA extrac-
tion from archival formalin-fixed tissue, Nucleic Acids Res. 27:e12.
Craven, R. A., Totty, N., Harnden, P., Selby, P. J., and Banks, R. E., 2002, Laser capture mi-
crodissection and two-dimensional polyacrylamide gel electrophoresis: evaluation of tissue
preparation and sample limitations, Am. J. Pathol. 160:815–822.
Crino, P. B., Khodakhah, K., Becker, K., Ginsberg, S. D., Hemby, S., and Eberwine, J. H., 1998,
Presence and phosphorylation of transcription factors in dendrites, Proc. Natl. Acad. Sci.
U.S.A. 95:2313–2318.
D’Amore, F., Stribley, J. A., Ohno, T., Wu, G., Wickert, R. S., Delabie, J., Hinrichs, S. H., and
Chan, W. C., 1997, Molecular studies on single cells harvested by micromanipulation from
archival tissue sections previously stained by immunohistochemistry or nonisotopic in situ
hybridization, Lab. Invest. 76:219–224.
Dolter, K. E., and Braman, J. C., 2001, Small-sample total RNA purification: laser capture
microdissection and cultured cell applications, Biotechniques 30:1358–1361.
Eberwine, J., Kacharmina, J. E., Andrews, C., Miyashiro, K., McIntosh, T., Becker, K., Barrett, T.,
Hinkle, D., Dent, G., and Marciano, P., 2001, mRNA expression analysis of tissue sections
and single cells, J. Neurosci. 21:8310–8314.
Eberwine, J., Yeh, H., Miyashiro, K., Cao, Y., Nair, S., Finnell, R., Zettel, M., and Coleman,
P., 1992, Analysis of gene expression in single live neurons, Proc. Natl. Acad. Sci. U.S.A.
89:3010–3014.
Ehrig, T., Abdulkadir, S. A., Dintzis, S. M., Milbrandt, J., and Watson, M. A., 2001, Quantitative
amplification of genomic DNA from histological tissue sections after staining with nuclear
dyes and laser capture microdissection, J. Mol. Diagn. 3:22–25.
Eisen, M. B., and Brown, P. O., 1999, DNA arrays for analysis of gene expression, Methods
Enzymol. 303:179–205.
Emmert-Buck, M. R., Bonner, R. F., Smith, P. D., Chuaqui, R. F., Zhuang, Z., Goldstein, S.
R., Weiss, R. A., and Liotta, L. A., 1996, Laser capture microdissection, Science 274:998–
1001.
Emsley, J. G., Lu, X., and Hagg, T., 2001, Retrograde tracing techniques influence reported
death rates of adult rat nigrostriatal neurons, Exp. Neurol. 168:425–433.
Farrell, R. E., Jr, 1998, RNA Methodologies, 2nd edition, San Diego: Academic Press.
Fasulo, W. H., and Hemby, S. E., 2003, Time-dependent changes in gene expression profiles of
midbrain dopamine neurons following haloperidol administration, J. Neurochem. 87:205–
219.
Fend, F., Emmert-Buck, M. R., Chuaqui, R., Cole, K., Lee, J., Liotta, L. A., and Raffeld, M.,
1999, Immuno-LCM: laser capture microdissection of immunostained frozen sections for
mRNA analysis, Am. J. Pathol. 154:61–66.
Franklin, T. R., and Druhan, J. P., 2000, The retrograde tracer fluoro-gold interferes with the
expression of fos-related antigens, J. Neurosci. Methods 98:1–8.
Freeman, W. M., and Hemby, S. E., 2004, Proteomics for protein expression profiling in neu-
roscience, Neurochem. Res. 29:1065–1081.
Galvin, J. E., 2004, Neurodegenerative diseases: pathology and the advantage of single-cell
profiling, Neurochem. Res. 29:1041–1051.
Galvin, J. E., and Ginsberg, S. D., 2004, Expression profiling and pharmacotherapeutic devel-
opment in the central nervous system, Alzheimer Dis. Assoc. Disord. 18:264–269.
Ginsberg, S. D., 2001, Gene expression profiling using single cell microdissection combined
with cDNA microarrays, In: Geschwind, D. H. (ed.), DNA Microarrays: The New Frontier in
Gene Discovery and Gene Expression Analysis, Washington: Society for Neuroscience Press,
pp. 61–70.
Ginsberg, S. D., and Che, S., 2002, RNA amplification in brain tissues, Neurochem. Res. 27:981–
992.
Ginsberg, S. D., and Che, S., 2004, Combined histochemical staining, RNA amplification,
regional, and single cell analysis within the hippocampus, Lab. Invest. 84:952–962.
Ginsberg, S. D., and Che, S., 2005, Expression profile analysis within the human hippocampus:
comparison of CA1 and CA3 pyramidal neurons, J. Comp. Neurol. 487:107–118.
Ginsberg, S. D., Crino, P. B., Hemby, S. E., Weingarten, J. A., Lee, V. M.-Y., Eberwine, J. H., and
Trojanowski, J. Q., 1999a, Predominance of neuronal mRNAs in individual Alzheimer’s
disease senile plaques, Ann. Neurol. 45:174–181.
Ginsberg, S. D., Crino, P. B., Lee, V. M.-Y., Eberwine, J. H., and Trojanowski, J. Q., 1997,
Sequestration of RNA in Alzheimer’s disease neurofibrillary tangles and senile plaques,
Ann. Neurol. 41:200–209.
Ginsberg, S. D., Elarova, I., Ruben, M., Tan, F., Counts, S. E., Eberwine, J. H., Trojanowski,
J. Q., Hemby, S. E., Mufson, E. J., and Che, S., 2004, Single cell gene expression anal-
ysis: implications for neurodegenerative and neuropsychiatric disorders, Neurochem. Res.
29:1054–1065.
Ginsberg, S. D., Galvin, J. E., Chiu, T.-S., Lee, V. M.-Y., Masliah, E., and Trojanowski, J. Q.,
1998, RNA sequestration to pathological lesions of neurodegenerative disorders, Acta Neu-
ropathol. 96:487–494.
Ginsberg, S. D., Hemby, S. E., Lee, V. M.-Y., Eberwine, J. H., and Trojanowski, J. Q., 2000,
Expression profile of transcripts in Alzheimer’s disease tangle-bearing CA1 neurons, Ann.
Neurol. 48:77–87.
Ginsberg, S. D., Schmidt, M. L., Crino, P. B., Eberwine, J. H., Lee, V. M.-Y., and Trojanowski,
J. Q., 1999b, Molecular pathology of Alzheimer’s disease and related disorders, In: Peters,
A., and Morrison, J. H. (eds.), Cerebral Cortex, Vol. 14: Neurodegenerative and Age-Related
Changes in Structure and Function of Cerebral Cortex, New York: Kluwer Academic/Plenum,
pp. 603–653.
Goldsworthy, S. M., Stockton, P. S., Trempus, C. S., Foley, J. F., and Maronpot, R. R., 1999,
Effects of fixation on RNA extraction and amplification from laser capture microdissected
tissue, Mol. Carcinog. 25:86–91.
Guntern, R., Bouras, C., Hof, P. R., and Vallet, P. G., 1992, An improved thioflavine S method
for staining neurofibrillary tangles and senile plaques in Alzheimer’s disease, Experientia
48:8–10.
Hemby, S. E., Ginsberg, S. D., Brunk, B., Arnold, S. E., Trojanowski, J. Q., and Eberwine, J. H.,
2002, Gene expression profile for schizophrenia: discrete neuron transcription patterns
in the entorhinal cortex, Arch. Gen. Psychiatry 59:631–640.
Hemby, S. E., Trojanowski, J. Q., and Ginsberg, S. D., 2003, Neuron-specific age-related de-
creases in dopamine receptor subtype mRNAs, J. Comp. Neurol. 456:176–183.
Hynd, M. R., Lewohl, J. M., Scott, H. L., and Dodd, P. R., 2003, Biochemical and molecular
studies using human autopsy brain tissue, J. Neurochem. 85:543–562.
Iscove, N. N., Barbara, M., Gu, M., Gibson, M., Modi, C., and Winegarden, N., 2002, Represen-
tation is faithfully preserved in global cDNA amplified exponentially from sub-picogram
quantities of mRNA, Nat. Biotechnol. 20:940–943.
Issaq, H. J., Veenstra, T. D., Conrads, T. P., and Felschow, D., 2002, The SELDI-TOF MS ap-
proach to proteomics: protein profiling and biomarker identification, Biochem. Biophys.
Res. Commun. 292:587–592.
Kacharmina, J. E., Crino, P. B., and Eberwine, J., 1999, Preparation of cDNA from single cells
and subcellular regions, Methods Enzymol. 303:3–18.
Kamme, F., Salunga, R., Yu, J., Tran, D. T., Zhu, J., Luo, L., Bittner, A., Guo, H. Q., Miller,
N., Wan, J., and Erlander, M., 2003, Single-cell microarray analysis in hippocampus CA1:
demonstration and validation of cellular heterogeneity, J. Neurosci. 23:3607–3615.
Klur, S., Toy, K., Williams, M. P., and Certa, U., 2004, Evaluation of procedures for amplification
of small-size samples for hybridization on microarrays, Genomics 83:508–517.
Kotlyar, M., Fuhrman, S., Ableson, A., and Somogyi, R., 2002, Spearman correlation identifies
statistically significant gene expression clusters in spinal cord development and injury,
Neurochem. Res. 27:1133–1140.
Lanciego, J. L., and Wouterlood, F. G., 2006, Multiple neuroanatomical tract-tracing, In:
Zaborszky, L., Wouterlood, F., and Lanciego, J. L. (eds.), Neuroanatomical Tract Tracing
3: Molecules-Neurons-Systems, Springer New York, 336–365.
Lein, E. S., Zhao, X., and Gage, F. H., 2004, Defining a molecular atlas of the hippocampus using
DNA microarrays and high-throughput in situ hybridization, J. Neurosci. 24:3879–3889.
Leonard, S., Logel, J., Luthman, D., Casanova, M., Kirch, D., and Freedman, R., 1993, Biological
stability of mRNA isolated from human postmortem brain collections, Biol. Psychiatry.
33:456–466.
Lockhart, D. J., Dong, H., Byrne, M. C., Follettie, M. T., Gallo, M. V., Chee, M. S., Mittmann,
M., Wang, C., Kobayashi, M., Horton, H., and Brown, E. L., 1996, Expression monitoring
by hybridization to high-density oligonucleotide arrays, Nat. Biotechnol. 14:1675–1680.
Lu, L., Neff, F., Dun, Z., Hemmer, B., Oertel, W. H., Schlegel, J., and Hartmann, A., 2004, Gene
expression profiles derived from single cells in human postmortem brain, Brain Res. Brain.
Res. Protoc. 13:18–25.
Madison, R. D., and Robinson, G. A., 1998, lRNA internal standards quantify sensitivity and
amplification efficiency of mammalian gene expression profiling, Biotechniques 25:504–
514.
Martin, L. J., 1999, Neuronal death in amyotrophic lateral sclerosis is apoptosis: possible con-
tribution of a programmed cell death mechanism, J. Neuropathol. Exp. Neurol. 58:459–471.
Matz, M., Shagin, D., Bogdanova, E., Britanova, O., Lukyanov, S., Diatchenko, L., and Chenchik,
A., 1999, Amplification of cDNA ends based on template-switching effect and step-out PCR,
Nucleic Acids Res. 27:1558–1560.
McClain, K. L., Cai, Y.-H., Hicks, J., Peterson, L. E., Yan, X.-T., Che, S., and Ginsberg, S. D.,
2005, Expression profiling using human tissues in combination with RNA amplification
and microarray analysis: assessment of Langerhans cell histiocytosis, Amino Acids, 28:279–
290.
Mikel, U. V., and Becker, R. L., Jr., 1991, A comparative study of quantitative stains for DNA in
image cytometry, Anal. Quant. Cytol. Histol. 13:253-260.
Mirnics, K., Middleton, F. A., Marquez, A., Lewis, D. A., and Levitt, P., 2000, Molecular charac-
terization of schizophrenia viewed by microarray analysis of gene expression in prefrontal
cortex, Neuron 28:53–67.
Molnar, Z., Blakey, D., Bystron, I., and Carney, R. S. E., 2006, Tract-tracing in developing
systems and in post mortem human material using carbocyanine dyes, In: Zaborszky, L.,
Wouterlood, F., and Lanciego, J. L. (eds.), Neuroanatomical Tract Tracing 3: Molecules-Neurons-
Systems, New York: Springer, 366–393.
Mouledous, L., Hunt, S., Harcourt, R., Harry, J. L., Williams, K. L., and Gutstein, H. B., 2003,
Proteomic analysis of immunostained, laser-capture microdissected brain samples, Elec-
trophoresis 24:296–302.
Mufson, E. J., Counts, S. E., and Ginsberg, S. D., 2002, Single cell gene expression profiles of
nucleus basalis cholinergic neurons in Alzheimer’s disease, Neurochem. Res. 27:1035–1048.
Mufson, E. J., Ginsberg, S. D., Ikonomovic, M. D., and DeKosky, S. T., 2003, Human cholinergic
basal forebrain: chemoanatomy and neurologic dysfunction, J. Chem. Neuroanat. 26:233–
242.
Palkovits, M., 1989, Microdissection in combination with biochemical microassays as a tool in
tract tracing, In: Heimer, L., and Zaborszky, L. (eds.), Neuroanatomical Tract Tracing Methods
2: Recent Progress, New York: Plenum Press, pp. 299–310.
Paweletz, C. P., Liotta, L. A., and Petricoin, E. F., III, 2001, New technologies for biomarker anal-
ysis of prostate cancer progression: laser capture microdissection and tissue proteomics,
Urology 57:160–163.
Phillips, J., and Eberwine, J. H., 1996, Antisense RNA amplification: a linear amplification
method for analyzing the mRNA population from single living cells, Methods Enzymol.
Suppl. 10:283–288.
Reiner, A. J., and Honig, M. G., 2006, Dextran amines: versatile tools for anterograde and
retrograde studies of nervous system connectivity, In: Zaborszky, L., Wouterlood, F., and
Lanciego, J. L. (eds.), Neuroanatomical Tract Tracing 3: Molecules-Neurons-Systems, New York:
Springer, 304–335.
Sambrook, J., and Russell, D.W., 2001, Molecular Cloning: A Laboratory Manual, 3rd edition, Cold
Spring Harbor: Cold Spring Harbor Laboratory Press.
Sarnat, H. B., Curry, B., Rewcastle, N. B., and Trevenen, C. L., 1987, Gliosis and glioma distin-
guished by acridine orange, Can. J. Neurol. Sci. 14:31–35.
Schmued, L. C., and Heimer, L., 1990, Iontophoretic injection of fluoro-gold and other fluo-
rescent tracers, J. Histochem. Cytochem. 38:721–723.
Simone, N. L., Remaley, A. T., Charboneau, L., Petricoin, E. F., 3rd, Glickman, J. W., Emmert-
Buck, M. R., Fleisher, T. A., and Liotta, L. A., 2000, Sensitive immunoassay of tissue cell
proteins procured by laser capture microdissection, Am. J. Pathol. 156:445–452.
Su, J. M., Perlaky, L., Li, X. N., Leung, H. C., Antalffy, B., Armstrong, D., and Lau, C. C., 2004,
Comparison of ethanol versus formalin fixation on preservation of histology and RNA in
laser capture microdissected brain tissues, Brain Pathol. 14:175–182.
Suarez-Quian, C. A., Goldstein, S. R., Pohida, T., Smith, P. D., Peterson, J. I., Wellner, E., Ghany,
M., and Bonner, R. F., 1999, Laser capture microdissection of single cells from complex
tissues, Biotechniques 26:328–335.
Tang, W. X., Fasulo, W. H., Mash, D. C., and Hemby, S. E., 2003, Molecular profiling of midbrain
dopamine regions in cocaine overdose victims, J. Neurochem. 85:911–924.
Tecott, L. H., Barchas, J. D., and Eberwine, J. H., 1988, In situ transcription: specific synthesis
of complementary DNA in fixed tissue sections, Science 240:1661–1664.
Tkatch, T., Baranauskas, G., and Surmeier, D. J., 2000, Kv4.2 mRNA abundance and A-type
K(+) current amplitude are linearly related in basal ganglia and basal forebrain neurons,
J. Neurosci. 20:579–588.
To, M. D., Done, S. J., Redston, M., and Andrulis, I. L., 1998, Analysis of mRNA from microdis-
sected frozen tissue sections without RNA isolation, Am. J. Pathol. 153:47–51.
Tomita, H., Vawter, M. P., Walsh, D. M., Evans, S. J., Choudary, P. V., Li, J., Overman, K. M., Atz,
M. E., Myers, R. M., Jones, E. G., Watson, S. J., Akil, H., and Bunney W. E. Jr., 2004, Effect of
agonal and postmortem factors on gene expression profile: quality control in microarray
analyses of postmortem human brain, Biol. Psychiatry. 55:346–352.
Topaloglu, H., and Sarnat, H. B., 1989, Acridine orange-RNA fluorescence maturing neurons
in the perinatal rat brain, Anat. Rec. 224:88–93.
Van Deerlin, V. M. D., Ginsberg, S. D., Lee, V. M.-Y., and Trojanowski, J. Q., 2000, Fixed post
mortem brain tissue for mRNA expression analysis in neurodegenerative diseases, In:
Geschwind, D. H. (ed.), DNA Microarrays: The New Frontier in Gene Discovery and Gene Expres-
sion Analysis, Washington, DC: Society for Neuroscience, pp. 118–128.
Van Deerlin, V. M. D., Ginsberg, S. D., Lee, V. M.-Y., and Trojanowski, J. Q., 2002, The use of
fixed human post mortem brain tissue to study mRNA expression in neurodegenerative
diseases: applications of microdissection and mRNA amplification, In: Geschwind, D. H.,
and Gregg, J. P. (eds.), Microarrays for the Neurosciences: An Essential Guide, Boston: MIT
Press, pp. 201–235.
VanGelder, R., von Zastrow, M., Yool, A., Dement, W., Barchas, J., and Eberwine, J., 1990,
Amplified RNA (aRNA) synthesized from limited quantities of heterogeneous cDNA, Proc.
Natl. Acad. Sci. U.S.A. 87:1663–1667.
Vincent, V. A., DeVoss, J. J., Ryan, H. S., and Murphy, G. M., Jr., 2002, Analysis of neuronal gene
expression with laser capture microdissection, J. Neurosci. Res. 69:578–586.
von Bertalanffy, L., and Bickis, I., 1956, Identification of cytoplasmic basophilia (ribonucleic
acid) by fluorescence microscopy, J. Histochem. Cytochem. 4:481–493.
Wang, E., Miller, L. D., Ohnmacht, G. A., Liu, E. T., and Marincola, F. M., 2000, High-fidelity
mRNA amplification for gene profiling, Nat. Biotechnol. 18:457–459.
Yao, P. J., O’Herron, T. M., and Coleman, P. D., 2003, Immunohistochemical characterization
of clathrin assembly protein AP180 and synaptophysin in human brain, Neurobiol. Aging
24:173–178.
Zhao, X., Lein, E. S., He, A., Smith, S. C., Aston, C., and Gage, F. H., 2001, Transcriptional pro-
filing reveals strict boundaries between hippocampal subregions, J. Comp. Neurol. 441:187–
196.
Zoccarato, F., Cavallini, L., and Alexandre, A., 1999, The pH-sensitive dye acridine orange as a
tool to monitor exocytosis/endocytosis in synaptosomes, J. Neurochem. 72:625–633.
5
Molecules and Membrane
Activity: Single-Cell RT-PCR
and Patch-Clamp Recording
from Central Neurons
WILLIAM H. GRIFFITH, SUN-HO HAN,
BRIAN A. McCOOL, and DAVID MURCHISON
INTRODUCTION
METHODOLOGICAL CONSIDERATIONS
Types of RNA Amplification for Single-Cell PCR
Harvesting Single Cells for Molecular Analysis
Primer and Probe Design
Gel-Based Identification Versus Real-Time Fluorescent
PCR Analysis
Quantitation and Validation Experiments
APPLICATIONS
Single-Cell Identification by mRNA Expression of
Phenotypic Markers
Single-Cell mRNA Expression and Patch-Clamp Recording
Single-Cell mRNA Expression, Patch-Clamp Recording, and
Fluorescent Calcium Measurements in Young and Aged Cells
ADVANTAGES AND LIMITATIONS
Advantages
Limitations
APPENDIX: DETAILED METHODS
Neuronal Preparations
Reverse Transcription
WILLIAM H. GRIFFITH AND DAVID MURCHISON • Department of Medi-

cal Pharmacology and Toxicology, College of Medicine, Texas A&M University System
Health Science Center, College Station, TX 77843–1114 SUN-HO HAN • Department
of Pharmacology, University of California, Irvine, CA 92697–4625 BRIAN A.
McCOOL • Department of Physiology and Pharmacology, Wake Forest University School
of Medicine, Winston-Salem, NC 27157–1083
142
MOLECULES AND MEMBRANE ACTIVITY 143
Solutions for PCR
Listing of Primers and Probes
Special Methods for Validation Experiment (Cloning, DNA
Preparation, Restriction Enzymes, and Sequencing)
Additional Methods for RT-PCR
REFERENCES
Abstract: This chapter summarizes methods for characterizing mRNA expression

and electrophysiological properties of central neurons using patch-clamp record-
ing and single-cell reverse-transcription/polymerase chain reaction (scRT-PCR). A
simple scRT-PCR protocol can be used to identify neurons by the expression of
phenotypic marker mRNAs. The combination of these methods allows for the corre-
lation of functional properties with molecular expression. Somewhat more complex
methods are available for quantitation of mRNA expression. Both traditional gel-
based PCR identification and real-time fluorescent PCR identification methods can
be employed. Advantages and requirements of various methods are discussed. Dif-
ferent types of tissue preparations are presented with emphasis on methods used
in our laboratories for acutely dissociated or cultured basal forebrain and amygdala
neurons. The basal forebrain contains a heterogeneous population of cholinergic
and GABAergic neurons, while the amygdala displays neurons with a complex re-
ceptor subunit composition. Investigation of neurons with this type of molecular
diversity benefits from techniques such as scRT-PCR for cell identification. We also
illustrate how these PCR methods can be combined with more complex experimen-
tal protocols, such as calcium buffering measurements using fluorescent dyes in
dissociated neurons from aged animals. The capacity to combine scRT-PCR with a
variety of experimental protocols allows the identification of unique cell types and
relationships between physiology and gene expression.
Keywords: aging, amygdala, basal forebrain, calcium-binding proteins, ChAT, fura-2
fluorescence, GAD, ion channel subunits, receptor subunits
I. INTRODUCTION
Single-cell reverse-transcription/polymerase chain reaction (scRT-PCR)

and antisense RNA amplification methodologies have generated a great
deal of excitement in the field of neuroscience because molecular expres-
sion profiles can now be determined in conjunction with a variety of experi-
mental applications. Some of these approaches include combining electro-
physiological recording and mRNA analysis (Dixon et al., 2000; Eberwine
et al., 1992; Hinkle et al., 2004; Lambolez et al., 1992; Monyer and Lambolez,
1995; Sucher et al., 2000; Surmeier et al., 1996), extensive gene profiling of
single cells using macro- and microarrays (Eberwine et al., 2001; Hinkle et al.,
2004), and analyzing changes in single-cell gene expression during disease
states, such as Alzheimer’s disease (Chow et al., 1998; Mufson et al., 2002).
In addition, heterogenous cell types can be identified from seemingly ho-
mogeneous cell populations (Eberwine et al., 2001; Elowitz et al., 2002; Han
144 WILLIAM H. GRIFFITH et al.
et al., 2002; Monyer and Markram, 2004; Pape et al., 2001). Combinatorial ap-
proaches with molecular, genetic, and functional experimental paradigms
are beginning to describe an almost unimaginable neuronal diversity and
to reveal the challenges facing those who would like to understand the func-
tional significance of that diversity.
Improved access to molecular methods has allowed electrophysiologists,
among others, to greatly increase the resolving power of their experiments
by correlational analysis of functional and molecular data from individual
cells. In particular, scRT-PCR techniques have been used as a relatively easy
and inexpensive way to examine the selected molecular expression patterns
of neurons. For example, many early experiments correlated the functional
properties of ion channels with specific mRNA expression patterns in the
brain because, in most cases, the subunit compositions of native ligand-
and voltage-gated ion channels were unknown. Some of these studies in-
cluded glutamate receptors (Audinat et al., 1996; Bochet et al., 1994; Jonas
et al., 1994; Lambolez et al., 1992), GABA receptors (Ruano et al., 1997;
Santi et al., 1994), dopamine receptors (Surmeier et al., 1996), and voltage-
dependent potassium (Martina et al., 1998; Song et al., 1998) and calcium
channels (Bargas et al., 1994; Plant et al., 1998). The adaptability of the scRT-
PCR technique makes it amenable to a variety of experimental approaches,
while providing a level of molecular information intermediate between his-
tochemistry (see Stornetta and Guyenet, this volume) and expression array
technologies (see Ginsberg et al., this volume).
Because of the very small amount of genetic material in a single cell, some
type of amplification is necessary in order to detect gene expression. In this
review, we will discuss several methods, but we will focus on a simplified scRT-
PCR protocol that we use in our laboratory to collect molecular data from
acutely dissociated or cultured rat basal forebrain or amygdala neurons that
have been electrophysiologically characterized by patch-clamp recording.
Theoretically, the enzyme reverse transcriptase makes one or a few cDNA
copies of each expressed mRNA in a cell. These cDNAs are less prone to
degradation than are the original transcripts and can be amplified by PCR.
Each cDNA of interest is targeted by a sequence-specific primer and is copied
by a thermostable DNA polymerase. Temperature cycling produces a con-
trolled exponential amplification of the cDNA, as each subsequent copy
increases the number of templates. After a certain amount of amplifica-
tion, the PCR reaction products become detectable. We will describe both
traditional gel-based detection and the more recently developed real-time
fluorescent detection, as well as quantitative considerations.
As with any investigative technique, there are limitations and assumptions
associated with the interpretation of data from scRT-PCR experiments that
must be taken into account. We will discuss these as they relate to our efforts
to identify neuronal cell types and detect channel or receptor subunit com-
position in functionally characterized cells. Cell identification is particularly
important in parts of the brain with diverse neuronal populations. For exam-
ple, the basal forebrain contains both cholinergic and GABAergic neurons
(Panula et al., 1984; Rye et al., 1984; Sarter and Bruno, 2002) that innervate
the hippocampus, olfactory cortex, and cerebral cortex (Fibiger, 1982; Mesu-
lam et al., 1983; Zaborszky et al., 1986). These cells have been implicated in
cognitive processes, such as attention and some forms of memory (Bartus et
al., 1982; Olton et al., 1991; Sarter and Bruno, 2000), and changes occur in
these cells with age and Alzheimer’s disease (Chow et al., 1998; Coyle et al.,
1983; Decker, 1987; Fischer et al., 1989; Mufson et al., 2002). Basal forebrain
neurons have proven to be a valuable model for investigating changes in
ion channel function and calcium homeostasis during aging (Griffith et al.,
2000). In contrast, neurons in the lateral/basolateral amygdala are pheno-
typically more homogeneous and consist of principal glutamatergic neurons
and GABAergic interneurons (McDonald, 1985; McDonald et al., 1989). This
brain region is intimately associated with the regulation of emotional behav-
iors like anxiety or fear (Fanselow and LeDoux, 1999; Killcross et al., 1997)
and plays a central role in drug-seeking behaviors (See et al., 2003). Single-
cell RT-PCR studies have revealed a surprisingly complex pattern of gene ex-
pression within individual glutamatergic or GABAergic lateral/basolateral
amygdala neurons (Floyd et al., 2003; McCool and Farroni, 2001). Most of the
examples we use to illustrate this review were acquired from these systems.
II. METHODOLOGICAL CONSIDERATIONS
A. Types of RNA Amplification for Single-Cell PCR
Due to its low abundance, mRNA must be amplified in order to be de-

tected in single neurons. Several strategies to detect mRNA expression in
single cells have been devised, but all of them depend on reverse transcrip-
tase to convert the original mRNA transcripts into cDNAs. The major differ-
ences in the techniques concern the amplification procedures, the types of
primers, and the detection methods. The antisense RNA differs from others
in that the amplification is accomplished by T7 RNA polymerase rather than
by PCR. This method does not require any PCR for the detection of ampli-
fication products, but it can be used in conjunction. The advantages of this
approach are that the amplification of the mRNA is a linear, and therefore
readily quantifiable, process and that all of the transcript species present can
be amplified simultaneously (Eberwine et al., 1992, 2001; Hinkle et al., 2004).
Consult the chapter by Ginsberg et al. in this volume for a more thorough
description of linear RNA amplification techniques. Other protocols are
based on RT-PCR and feature exponential amplification of cDNA. These in-
clude 3 -end amplification followed by gene-specific PCR (Dixon et al., 1998,
2000), or two-round PCR with nonspecific primers designed for conserved
regions of closely related genes for first round, followed by second-round
PCR with either specific primers or restriction enzyme detection (Audinat
et al., 1996; Lambolez et al., 1992; Plant et al., 1998). Another variation in-
volves multiplex PCR where amplification is conducted with several specific
primer pairs simultaneously (Edwards and Gibbs, 1994; Lindqvist et al., 2002;
Phillips and Lipski, 2000). Finally, Surmeier and colleagues (Surmeier et al.,
1996; Tkatch et al., 1998; Yan and Surmeier, 1996) have used a method in
which the cDNA yield from the RT is divided for separate amplification of
each target sequence.
There are numerous possible modifications to the above methods, many
of which have been reviewed for application to single cells (Audinat et al.,
1996; Dixon et al., 2000; Eberwine et al., 2001; Hinkle et al., 2004; Monyer
and Lambolez, 1995; Phillips and Lipski, 2000; Sucher et al., 2000; Surmeier
et al., 1996). The important point is that various methods are amenable
to a particular experimental priority or design. For example, if hundreds
of genes are being investigated, then global RNA amplification using the
non-PCR-based method of Eberwine et al. (1992) is desirable. However, this
method is laborious and technically challenging and may not be practical
for all gene expression studies. The different PCR protocols have the gen-
eral advantage of amplification specificity and sensitivity. Problems can be
encountered though, when attempting to optimize conditions for successful
multiplex PCR with several primer pairs or when many cycles of amplifica-
tion are employed. On the other hand, a straightforward RT-PCR protocol
can permit the reliable detection of a small number of moderately abundant
transcripts without requiring extra amplification steps. We use a relatively
simple “one-round” single-cell RT-PCR protocol (modified from Surmeier
et al., 1996) to detect transcripts for phenotypic markers, receptor/channel
subunits, and calcium-binding proteins (CaBP) in neurons of the basal fore-
brain and amygdala that have been functionally characterized. This allows
the correlation of functional properties with molecular expression and the
identification of different cell types.
Figure 5.1 gives an overview of our protocol for scRT-PCR of acutely dis-
sociated or cultured neurons using random hexamer primers for the RT
and specific forward and reverse primers for PCR of six target cDNAs. It
should be noted that the specific conditions for optimizing the RT-PCR re-
actions were determined from work with known quantities of purified RNA
or cloned DNA before being applied to single cells (discussed below, along
with primer design). One of the target cDNAs is a positive control, which is
usually chosen as an abundant transcript present in all neurons. We use the
enzyme glyceraldehyde 3-phosphate dehydrogenase (GAPDH) as a positive
control. For phenotypic markers, we use choline acetyltransferase (ChAT)
and glutamic acid decarboxylase (GAD). This typically enables us to investi-
gate the co-expression of three (four at most) other transcripts of interest.
We have successfully detected expression of CaBP, voltage-gated calcium
and sodium channel subunits, and NMDA and glycine receptor subunits.
More detailed information for each step illustrated in Fig. 5.1 is provided
in the following sections.
B. Harvesting Single Cells for Molecular Analysis
The first and most important aspect of an experiment designed to corre-

late physiological function with molecular expression in individual neurons
Figure 5.1. Diagram of the process for cell harvesting, reverse transcription (RT), and
polymerase chain reaction (PCR). Top panel shows aspiration of a neuron into the
recording pipette. RT was performed with random hexamers and Superscript r II
reverse transcriptase. The cDNA yield was divided into six portions and subjected to
40 PCR cycles using a PTC-100TM Programmable Thermal Controller (MJ Research,
Inc.) and agarose gel identification, or 40 PCR cycles via real-time PCR with an ABI
Prism 7700 (Applied Biosystems) and fluorescent identification. Taq polymerase and
specific primer pairs for the transcripts of interest were used for amplification (DTT,
dithiothreitol; dNTP, deoxynucleotide triphosphate).
is to have a viable preparation. Different techniques are required to main-

tain acutely dissociated neurons, cultured neurons, or acute or cultured
brain slices. It is possible to record and harvest individual neurons from
all of these preparation types, and generally, the process of harvesting cells
is similar for any preparation. It is practical to determine first what meth-
ods are required to make successful, high-quality patch-clamp recordings
from the neurons of interest and then to apply the additional procedures
necessary for RT-PCR. We will be focusing on dissociated and cultured neu-
ronal preparations in the sections below, but a fine coverage of special
methods for patch clamping in brain slices is provided by Petersen in this
volume.
The most obvious difference between initial preparation for ordinary
patch-clamp recording and recording to be followed by RT-PCR is the care
taken to assure the patch-pipette, intracellular solution and RT reaction
mixture are all RNase free. We use dedicated stocks of chemicals and ma-
terials that are kept separate from ordinary supplies. PCR-dedicated equip-
ment, such as pipetters, are also kept separately. Latex gloves are worn when
handling any of the RT-PCR-dedicated components and manipulation of
these materials (such as solution preparation) is done inside a sterile hood
equipped with UV lights. Gloves and all external surfaces are washed with
70% ethanol or commercially available RNase-inactivating solutions (e.g.,
RNaseZAP r
, Ambion) prior to introduction into the hood. If gloves are
powdered, some effort is required to wash residual powder from the exte-
rior of the glove, especially if fluorescent PCR detection is to be used (the
powder is usually highly fluorescent corn starch). Pipette and RT solutions
are prepared with nuclease-free water and sterilized (autoclaved) RNase-
free PCR tubes. Pipette tips containing an aerosol filter can be obtained
sterile and nuclease- and nucleic acid-free from the manufacturer.
We use patch-pipettes pulled from glass (Garner Glass Co., 7052 or KG-
33) that has been sterilized by baking (400◦ F, 2 h). This glass can be kept
in a sterilized petri dish inside the UV hood. We prefer a filament type
of glass (KG-33) due to the ease and reliability of filling the tip with the
pipette solution without air bubble blockage. The pipette puller and pol-
isher are washed with 70% ethanol-soaked Kimwipes r
(Kimberly–Clark)
without touching the actual heating elements. Pulled and polished pipettes
are kept in dedicated storage jars that are periodically washed with ethanol.
It is desirable to have the aperture of the patch-pipette tip as large as is con-
sistent with obtaining stable recordings. Pipette tip size varies with cell type
but is generally between 0.5 and 2.0 µm. Aliquots of intracellular pipette
solution can be kept frozen in sterile nuclease-free tubes in a dedicated con-
tainer until use. We have successfully used pipette solutions containing CsCl,
cesium acetate, cesium methylsulfonate, or potassium gluconate as the prin-
cipal salts. Other common ingredients, such as EGTA, TEA, NaCl, HEPES,
etc., do not appear to affect the RT-PCR, and we have not had to alter the
composition of our pipette solutions from what we use without RT-PCR.
RNase inhibitor can be added to the pipette solution for added assurance
against degradation of RNA. This precaution might be worthwhile in the
cases of prolonged recording or low-abundance target transcripts.
A sterile plastic syringe (1 ml) can be fabricated as a sterile pipette filler
by removing the plunger and melting the distal end over a small Bunsen
burner and allowing the end to drop off under gravitational pull in a vertical
orientation. When done correctly, this results in a narrowly tapered hollow
tip that can be cut to a desired length with a sterile scalpel blade. The
“trick” is to rotate the syringe for even heating directly in the flame (flame
< 2 in. high and < 1 in. wide) until just before the plastic sags. These fillers
are then disposable and inexpensive. In practice, it is not possible to have
a sterile extracellular bath or recording chamber, because, of course, the
preparation itself is not sterile. We do try, however, to keep the recording
chamber, inflow lines, and glassware clean by washing them in 70% ethanol
after each experiment. Efforts are made to keep dust to a minimum in the
recording room. Extracellular solutions contain sugars and so should be
made fresh daily, although we have successfully used solutions for several
days if kept refrigerated overnight. Parts of the recording setup that can
be safely wiped with ethanol (nonelectrical parts) are cleaned before each
recording session. The investigator wears gloves throughout the process and
keeps a squirt bottle of 70% ethanol handy for frequent hand rinsing.
The internal electrode wire should be rinsed or dipped in 10% bleach
(sodium hypochlorite) and allowed to dry prior to each individual record-
ing. Drying can be hastened by following the bleach with ethanol or by
using compressed gas. A small positive pressure should be applied to the
patch-pipette before lowering the tip below the bath surface and until the
target cell is contacted. Avoid negative pressure except for seal formation,
patch rupture, and cell aspiration.
Once the electrophysiological data have been gathered, the cell is har-
vested for RT by one of several means, depending on the experimental
arrangement and preparation. We are most experienced with acutely disso-
ciated or cultured basal forebrain neurons. Acutely dissociated basal fore-
brain neurons consist of the cell soma (∼7–30-µm-long axis in rat) and the
proximal processes (up to 60 µm long) that are superficially stuck to the
bottom of the glass-recording chamber (Fig. 5.2A1). These cells can be lifted
entirely from the bottom (Fig. 5.2A2), and thus away from other cells or cel-
lular debris, while maintaining the patch seal. If the aperture of the pipette
is large enough (∼2.5 µm), strong negative pressure can be applied, and
the entire cell can be carefully aspirated into the recording pipette (Fig.
5.2A3). As soon as the cell is inside, the pipette is quickly withdrawn from
the bath. Often though, the pipette aperture is just small enough that the
cell membrane and nucleolus combine to form an immovable plug. In this
case, the pipette with attached cell is moved to an area of the bath that is
visibly clear of cells and debris, and under fine manipulator control the tip
is gently pushed against the bottom of the chamber to nudge the plug into
the pipette. If this does not work, a slightly more vigorous push will partially
break the tip and allow the cell to enter the pipette. Skillful breaking of
Figure 5.2. Single-cell RT-PCR from a GAD+ basal forebrain cell. (A1–A3) Aspiration
of a basal forebrain neuron into a sterile patch-pipette is shown in sequence. Cell
soma ∼20µm on the long axis. Note that the cell was lifted off the bottom of the
recording chamber before aspiration to ensure that only a single cell was collected.
(B) Photomicrograph of the RT-PCR products from the same cell separated on an
ethidium bromide-stained 2% agarose gel. The left lane shows a band of GAD67
mRNA, whereas no ChAT mRNA was detected. Glyceraldehyde-3-phosphate dehy-
drogenase (GAPDH) mRNA was used as a positive cell marker. Molecular weight
markers are shown to the right (GAD, glutamate dehydrogenase; ChAT, choline
acetyltransferase). (From Han et al., 2002, with permission.)
the tip and coordinated application of negative pressure are necessary to

prevent aspiration of bath solution or debris. If another cell, or even process
is even partially aspirated, the pipette is discarded. We have aspirated both
bath solution and noncellular debris to process as negative controls.
For cultured neurons, the concern for loose debris or cells is minimal
because of the strong adhesion of the cells to the substrate. However, this
makes it more difficult to aspirate an entire cell. This feature has been ex-
ploited by Eberwine and colleagues (Crino and Eberwine, 1996; Miyashiro
et al., 1994), who were able to aspirate portions of dendrites and show that
there is some differential distribution of mRNAs between soma and pro-
cesses. Therefore, to obtain an accurate profile of the transcripts present in
a neuron, as much of the cell should be sampled as possible. In neuronal
cultures that are moderately dense, it is not practical to try to aspirate a
cell soma and neurites because of the numerous intercalated processes of
neighboring cells. We have used a “vacuuming” method to aspirate the cell
soma in an intact or slightly broken tip as shown in Fig. 5.3. Slightly mov-
ing the tip peels the soma off the substrate without the processes. While
dissociated neurons are probably the easiest to harvest in relative complete-
ness, Landfield and colleagues (Chen et al., 2000) have shown that almost
complete hippocampal CA1 neurons can be harvested from a “zipper” slice
Figure 5.3. Combined patch-clamp recording of spontaneous synaptic currents and

scRT-PCR from a ChAT+/GAD+ basal forebrain cell in culture. (A1–A4) Aspiration
of the cell into the recording pipette is shown in sequence. Note that only the single
cell is harvested without nearby debris. (B) Real-time RT-PCR amplification plot of
transcript expression in the same cell. Expression of GAPDH, ChAT, GAD67, and the
calcium-binding protein calretinin (CR) were detected. Calbindin and parvalbumin
were not detected in this neuron. (C) Spontaneous miniature synaptic currents
recorded from the same cell in the presence of TTX (0.5 µM). Scale bar: 20 µm.
preparation. In this case, the recording pipette is used to extract the cell
from the slice before it is transferred to the mouth of a larger collection
pipette that is used to aspirate the neuron. In other preparations though,
it might not be possible to extract any of the cells reliably, and researchers
must be content to aspirate a portion of the somatic cytoplasm (Pape et al.,
2001; Sucher and Deitcher, 1995).
Once the cell has been aspirated, we remove the pipette from the elec-
trode holder and transfer it to an expelling apparatus, consisting of a holding
platform for an RT tube and a syringe with a flexible plastic tube attached
that has been washed with ethanol. The open end of the tube is placed over
the butt-end of the pipette and the pipette tip is carefully introduced into
the RT tube without hitting the sides. The tip is gently crushed into the
bottom of the tube and slight positive pressure is applied to the syringe to
expel the contents while withdrawing the tip from the initial RT solution so
as not to create bubbles. The RT tube is immediately sealed, spun briefly
in a low-speed centrifuge, and placed on dry ice. For our protocol, the RT
tube contains 8 µl nuclease-free water, 1 µl dithiothreitol (DTT, 100 mM),
0.33 µl random hexamer primers (3 µg/µl), and 0.5 µl RNase inhibitor. A
separate tube is prepared for each cell to be collected. These tubes can be
prepared the day before and frozen, and then thawed and kept on ice until
needed. Other arrangements are possible, such as including the initial RT
ingredients in the pipette (Cao et al., 1996; Eberwine et al., 1992) or having
the complete RT mixture in the tube (Audinat et al., 1996). We freeze the
harvested cells in the initial RT solution (−80◦ C), and conduct the RT and
PCR within the next 3 days. Our protocol for RT is given in section “Reverse
Transcription.”
C. Primer and Probe Design
Successful design of PCR primers involves the identification of sequences

in the gene of interest that fulfill several criteria. Among the most impor-
tant considerations, the oligonucleotides must possess a favorable melting
temperature of the primer for dissociating from its complimentary sequence
to insure compatibility with the type of PCR to be performed. Generally, the
melting temperature should be between 55 and 70◦ C. This temperature is
dependent on the content of G/C nucleotides, which should not exceed
55%. All primers that are to be used in the same PCR should have very simi-
lar melting temperatures so that the temperature parameters of the PCR can
be optimized. Primers should be designed to avoid appreciable sequence-
specific secondary structures like hairpin- or stem-loops. Additionally, ex-
tensive complimentary sequences between different primers can sequester
these reagents away from more productive interactions with the template.
Ultimately, both inter- and intraprimer interactions can interfere with the
efficient PCR product formation that is essential for amplification of cDNAs
derived from individual neurons. Fortunately, most software packages uti-
lized for primer design automatically select primer sets using criteria that
insure that each primer possesses optimal characteristics.
For single-cell RT-PCR, there are several additional concerns that should
be addressed when designing PCR primers. First, one must consider that
different means of analyzing PCR product formation will determine op-
timal primer selection. Single-cell PCR products that are analyzed using
agarose gel electrophoresis must be large enough (> 100 bp) to distin-
guish from “primer dimers” that are common to single-cell approaches
but should be small enough (typically < 500 bp) to insure that the tar-
get cDNA is well represented in the population of reverse transcription
products. This is perhaps less of a concern when using RNA amplifica-
tion approaches, but it still deserves some attention for low-abundance
messages. Regardless, standard gel-based analysis often requires only for-
ward and reverse primers, and these can be identified with publicly avail-
able software, like the San Diego Supercomputer Center’s “Biology Work-
bench” (http://workbench.sdsc.edu/, Subramaniam, 1998). In contrast to
gel-based methods of product detection, “real-time” analysis employs an
additional fluorescent gene-specific oligonucleotide probe that, especially
when used with the 5 -exonuclease assay (Whitcombe et al., 1998), provides
exquisite sensitivity for low-abundance products. However, this approach has
very specific requirements of the PCR primers, probes, and products. For ex-
ample, the PCR product is typically ∼100 bp since the annealing/extension
phases are performed simultaneously at a temperature (60–65◦ C) that is sub-
optimal for most thermostable polymerases. In addition, the primer/probe
melting temperatures and their sequence content have more specific re-
quirements that make selection of appropriate target sequences in the gene
of interest more demanding. For primer/probe design, we have had the
most success employing commercial software specifically designed for a par-
ticular real-time PCR platform, such as Primer Express r
(Applied Biosys-
tems Inc.) for Taqman -based probe detection.
r
Primer specificity is a topic worth special consideration. It must be de-

termined that primers chosen for a given gene are highly specific for the
product of interest. For example, it may be necessary that PCR products
arising from cDNA be differentiated from those representing genomic DNA.
For single neurons, genomic sources are unlikely to influence product for-
mation ( Johansen et al., 1995). However, in cases where message abun-
dance is low and multiple rounds of PCR are necessary, primers can be
designed to span multiple intron/exon boundaries. This does not necessar-
ily require a direct knowledge of genomic structure in the specific species
being used, given that, for the vast majority of genes (> 99%) the position
of a given intron relative to coding region exons is highly conserved be-
tween humans, rats, and mice (Roy et al., 2003). Another specific concern,
especially for neurotransmitter receptors, is that many mRNAs are highly
similar to other gene products that arise either by transcriptional process-
ing (e.g., splicing of alternative exons) or by gene duplication, giving rise to
multiple, highly homologous subunits. In the former case, constraining at
least one primer-binding site within an alternative exon can provide an ef-
ficient means of differentiating splice variants. Similarly, sequence compar-
isons between homologous subunits can identify highly conserved regions
(> 85% identity/20 bp) that can be omitted from consideration when iden-
tifying promising target sequences. In many cases, nonhomologous regions
between related subunits are often concentrated in 5 - and 3 -noncoding
regions of the mRNA which are often more polymorphic than are coding re-
gions. This may require precise knowledge of the target sequence in the par-
ticular species of interest. The sequence of a primer or probe can be checked
for specificity after design by comparing it to the sequences in a gene
database.
D. Gel-Based Identification Versus Real-Time

Fluorescent PCR Analysis
The simplified aspect of our RT-PCR protocol is that there is only a single
round of PCR cycles for amplification. This limits the amount of amplifi-
cation product available for detection, but reduces the possibility of a false
or irrelevant positive amplification which increases with additional cycles of
PCR. By using serial dilutions of known quantities of purified RNA, we were
able to detect specific mRNAs from samples of as little as 2 pg/µl total RNA
after 40 cycles of PCR by gel electrophoresis. Real-time fluorescence PCR
was even more sensitive, as detailed below. It has been estimated that single
neurons contain ∼50 pg total RNA (Sucher et al., 2000), so there should be
no problem of starting with too little mRNA to detect. Detection was con-
firmed in gels by visual inspection and the separate identical scoring by two
of three investigators. For real-time PCR detection, the amplification plot
of the increase in fluorescence had to cross a threshold in the log-linear
range and above the noise before the 39th cycle. These rather stringent
requirements along with suitable controls (see section “Quantitation and
Validation Experiments”) insure that there is little likelihood of false posi-
tive amplification by this protocol. There is more chance of a false negative
amplification with this protocol perhaps, than with others that involve more
amplification, but some measures can be taken to increase the confidence
in a negative result, as discussed below.
1. Agarose Gel PCR
Following RT, targeted cDNAs were amplified with specific primer pairs
referenced in section “Listing of Primers and Probes.” Forty cycles of PCR
reactions generated products that were verified by gel electrophoresis. One
drop of sterile mineral oil (Sigma) was added to each tube to cover the reac-
tion mix and prevent evaporation and changes in salt concentrations during
PCR. Once the tubes were placed in the PCR instrument (PTC-100TM Pro-
grammable Thermal Controller, MJ Research, Inc.), the temperature was
elevated to 94◦ C for 2 min. During this period, Taq DNA polymerase 2.5 U
(Promega) was added to enable a “hot start,” which reduces the possibility of
nonspecific amplification and primer dimers. The tubes then went through
40 cycles of three steps: 94◦ C (1 min), 50◦ C (1 min), and 72◦ C (1.5 min).
After completion of all the cycles, the temperature was maintained at 4◦ C
until gel electrophoresis.
The PCR product was extracted from the mineral oil by mixing with 100 µl
of chloroform/isoamyl alcohol (24:1) and vortexing. The upper aqueous
phase contained the PCR product and this was placed in a new tube mixed
with the gel loading dye (Promega). The PCR products were loaded and
ran for 2 h on a 1.5–2% agarose gel (Sigma) at 70 V (Horizontal Gel Elec-
trophoresis System; Gibco BRL). The voltage and duration of the run are
dependent on the size of the PCR products. Products were visualized by
ethidium bromide staining and images were made using a digital camera
system (MultiImageTM Light Cabinet, Alpha Innotech Corporation). Primer
specificity was tested by RT-PCR of total RNA extracted from whole rat brain
or basal forebrain tissue sections. Gels of the PCR products were seen to
contain single bands at the expected molecular weights.
2. Semiquantitative Real-Time PCR

(TaqmanR
Probes and SYBR R
Green)
A relatively new method is available to quantitate PCR products using

real-time changes in fluorescent intensity during each PCR cycle (Gibson
et al., 1996; Heid et al., 1996). We currently use this method for our single-
cell studies because of the increased throughput, enhanced sensitivity, and
potential for quantitation. Because of the advantages, an emphasis will be
given to this technique compared to gel-based PCR. Figure 5.4 shows an
Figure 5.4. Semiquantitative RT-PCR. (A) Amplification plots of real-time PCR from
young and aged total mRNA from rat basal forebrain; GAPDH, ChAT, GAD, CB, and
CR expression are shown. Threshold cycle (C T ) for each transcript is shown as the
intersection of the solid line with the fluorescent intensity plot. (B) Standard curve
method for quantitation. A slope of –3.4 indicates a near 100% efficiency of the
PCR reactions. ChAT values are normalized to GAPDH in both young (n = 5) and
aged (n = 5) rats (∗ p < 0.05). (C) Comparative CT method for quantitation. The
validation experiment is shown to the left and plots the difference (C T ) between
the transcript and the normalizer (GAPDH). The slopes are < 0.1 and demonstrate
that the efficiencies of the target and reference are approximately constant. The
plot to the right shows the difference between the calibrator and the C T (C T )
for the same data shown in (A) (∗ p < 0.05).
amplification plot of the increase in fluorescence versus cycle number. In

this example, PCR is performed with fluorescent-labeled probes. All the
primers and probes were designed using Primer Express r
software and
their PCR products are approximately 50–150 bp, which are much smaller
than products generated for regular gel-based PCR. Forward and reverse
primers serve the same function as in gel-based detection, but an additional
25–30 bp Taqman r
probe is utilized (see section “Listing of Primers and
Probes”). The probe is designed to be complimentary to a sequence located
between the forward and the reverse primers. Taqman r
probes are con-
structed with a reporter dye (FAM, 6-carboxylfluorescein) at the 5 -end and
the fluorescence quencher dye (TAMRA, 6-carboxytetramethyl-rhodamine)
located at the 3 -end. The close contact between the two dyes is responsible
for the fluorescence quenching. However, as Taq extends DNA synthesis
from a PCR primer, it runs into the probe and displaces the probe’s 5 -end.
The resulting “Y” structure makes the 5 -end of the probe a substrate for the
polymerase’s innate 5 to 3 exonuclease activity which cleaves the probe,
separating the reporter dye from the quencher dye and increasing the flu-
orescence. As amplification increases and more 5 -reporter dye is liberated,
there is a proportional increase in fluorescence. For each PCR cycle, the
fluorescent intensity is plotted and is directly proportional to the accumula-
tion of PCR product. The PCR cycles using Taqman r
probes included initial
◦
r
steps of 50 C/2 min for optimal AmpErase UNG enzyme activity, followed
by 95◦ C/10 min for activation of AmpliTaq r
Gold DNA polymerase, and
◦ ◦
then 40 cycles of 95 C/15 s (melt), 60 C/1 min (anneal/extend).
The initial amount of template present can be determined by the number
of cycles required for fluorescence to reach detection threshold (threshold
cycle, C T ). In the case of high initial concentrations of template, relatively
few cycles would be necessary to reach threshold (C T is lower), but in the
case of low initial concentrations, more cycles are required (C T is higher).
Because each PCR cycle (after the first two or three) results in doubling
of the amount of cDNA, we can calculate the relative concentrations of
samples by the number of cycles necessary to reach threshold fluorescence.
Included in the reaction mix for real-time PCR (Taqman r
Universal Master
mix), there is a passive reference dye, ROX, which serves as an internal
control to normalize each fluorescence signal. Therefore, the normalized
reporter fluorescence is given by
Emission intensity of reporter dye (FAM)
Rn = .
Emission intensity of passive reference (ROX)
This normalization corrects for extraneous fluorescence fluctuation. In
addition, fluorescence intensities recorded in early cycles are considered
background and are averaged and subtracted from fluorescence levels in
later cycles. This is shown in the following equation:
Rn = (Rn+) − (Rn−).
Rn+ is the normalized fluorescent intensity in the first nonbackground cycle

or beyond, and Rn− is the intensity of sample without template (negative
control) or of background cycles. Finally, an amplification plot of Rn is
graphed against each cycle, as described above.
In a similar manner, SYBER r
Green also can be used as the fluorescent
indicator. One advantage of SYBER r
Green is the lower cost compared
r
to Taqman fluorescent probes. Amplification plots of PCR products using
SYBER r
Green can be generated by measuring the increase in fluorescence,
resulting from the binding of SYBER r
Green to double-stranded DNA. The
passive reference molecule is also ROX, because the excitation–emission
profile for SYBER r
Green is similar to that of the FAM dye. When using
r
SYBER Green, an additional melting curve is performed to ensure that a
single PCR product is produced. When SYBR r
Green was used, the same
primers for gel-based PCR were also utilized. The reaction mix contained the
same ingredients for Taqman r
-based detection, except 2× SYBR r
Green
PCR Master Mix (Promega) was included. The reaction parameters using
the SYBR r
Green detection protocol included 40 cycles of 95◦ C/1 min,
50◦ C/1 min, and 72◦ C/1.5 min.
Major advantages of real-time PCR include the high throughput and po-
tential for quantitative analysis. We utilize the 96-well format of the ABI
Prism 7700 sequence detection system to enable analysis of up to 14 cells
(six sequence probes) along with appropriate controls all on one plate. This
ability to analyze a relatively large number of cells is a definite advantage
when combining both electrophysiology and gene analysis.
Several methods are available to quantitate results from real-time PCR
experiments, including the standard curve method and the comparative
C T method. In the standard curve method, the initial concentration (copy
number) for each sample is calculated. The comparative C T method is sim-
ilar to the standard curve method, except it uses a single point comparison
and an arithmetic formula to achieve relative quantification. Figure 5.4 and
the section below describe a comparison between these two methods in
an experiment to quantitate tissue mRNA levels of the cholinergic marker,
ChAT, from both young and aged rats. This example also illustrates a “val-
idation” experiment required for using the comparative C T method. The
validation experiment confirms a linear amplification over an extended
RNA concentration for each transcript, thus verifying the efficiency of the
PCR. For the standard curve method, we utilized purified cloned DNA frag-
ments that were cut by different restriction enzymes from cloned plasmid
DNA. Serial dilutions of each cloned DNA fragment containing from 102 to
109 copy numbers were used to generate standard curves. The methods for
restriction enzyme analysis are described in section “Special Methods for
Validation Experiment”.
E. Quantitation and Validation Experiments
Our primary use for real-time RT-PCR is to identify phenotypic marker

mRNAs in single cells. Although we do not routinely quantitate our
single-cell results, the following example summarizes a comparison of basal

forebrain tissue mRNA levels between samples from young and aged rats
(Fig. 5.4). We utilized both the standard curve and the comparative C T
methods for quantitation. Tissues from five young and five aged Fischer 344
rats were used for total RNA purification. The RNA was quantified by spec-
trometry after removing genomic DNA. Real-time RT-PCR was conducted on
10 ng samples and amplification plots for young and aged were constructed,
as shown in Fig. 5.4A. In both young and aged samples, the GAPDH curve oc-
curs at the earliest cycle number and GAD, calbindin (CB), calretinin (CR),
and ChAT curves are detected later. The plateau fluorescence was reached
around cycle 30, showing approximately the same level of fluorescence
(Rn) regardless of sample concentrations. We set the threshold in the log-
linear part of the curve where no limiting factors influenced amplification.
From DNA concentrations in each dilution and the molecular weight of
PCR products we could calculate DNA copy number for each concentration.
For example, cloned ChAT DNA was cut by the restriction enzyme PvuII for
a total 753 bp product that contained 323 bp ChAT PCR product sequences
and 430 bp product of sections of plasmid DNA. The copy number of ChAT
DNA molecules was calculated using the molecular weight of the 753 bp
product and Avogadro’s number (6.023 × 1023 ). For ChAT, logarithmic copy
number of initial concentration of DNA versus C T was plotted as a standard
curve (Fig. 5.4B, left panel). The curve for ChAT was fit by linear regression
and had a slope of 3.36. According to the manufacturer’s specifications,
100% efficiency of PCR reactions occurs with a slope of 3.4. These data
suggest that our PCR efficiencies were close to 100%. Experiments were
repeated in triplicate, and we used the same threshold fluorescence level
that was utilized in standard curve preparation for calculating copy number
from samples of young and aged rats. The copy numbers of ChAT were
normalized by the copy number of GAPDH, and this ratio was compared
in young and aged samples (Fig. 5.4B, right panel). An independent two-
tailed t-test revealed that ChAT mRNA expression was significantly reduced
in aged tissue ( p < 0.05). Although this absolute standard curve method
is a reasonable method for semiquantification analysis, there are several
disadvantages, including cost and the number of samples for the standard
curve that must be run for each target sequence. We have found this method
to be less than ideal for single-cell analysis.
A second method of quantitation is available to overcome some of the
disadvantages of the method discussed above: the comparative C T method.
This method utilizes sample normalization and calibration for relative com-
parisons. In the example above, the C T value for ChAT is normalized by the
C T value of GAPDH. Samples are then normalized again by a “calibrator”
to enable comparisons across reaction plates. We utilized 5 ng of purified
whole brain RNA as the calibrator and the calibrator was subjected to the
same RT-PCR as the samples on each plate. Because PCR results are an ex-
ponential function, normalization was achieved by subtracting C T values.
Normalized C T (C T of sample subtracted from C T of GAPDH) is called the
C T , and it was further normalized by subtraction from the calibrator C T
value, which produced a C T value. This C T was changed to a linear
value by the conversion 2∧C T and then compared between each sample.
In our example of quantification of tissue RNA, all of the young and aged
samples were run on the same plate, and so C T values of one of the sam-
ples were utilized as calibrator and normalized to generate C T for the
other samples. Results similar to above were obtained when the data were
analyzed using the comparative C T method. The comparative C T method
demonstrated a significant decrease in ChAT mRNA expression during ag-
ing (Fig. 5.4C, right panel, p < 0.05).
In order to perform valid comparisons using the comparative C T method,
a validation experiment must be performed. Because the normalizer
(GAPDH) expression level may differ in each sample, the PCR amplification
efficiency of GAPDH and other targets have to be consistent for different
sample concentrations. This validation experiment is shown in Fig. 5.4C
(right panel). A known quantity of total RNA purified from young basal fore-
brain was used for RT and different concentrations of cDNA corresponding
to 0.06 ng to 2 µg starting RNA were subjected to PCR. All four lines were
almost parallel, and more importantly, the slopes of each line ranged from
0.02 to 0.07. According to the manufacturer’s specifications, if the slopes
are less than 0.1, the validation test is successful and GAPDH can be used
as normalizer. When semiquantitative analyses are desired, the comparative
C T method has the advantage that a single point (C T value) can be used for
comparisons, thus negating the need for a complete standard curve for each
sample. This is particularly important for single-cell RT-PCR. Numerous re-
views are available to discuss the general principles of quantitative real-time
RT-PCR (Freeman et al., 1999; Medhurst et al., 2000; Stahlberg et al., 2004)
and recently, a protocol for improved quantitative real-time RT-PCR from
single cells has been developed (Liss, 2002).
Various controls should be used to confirm the specificity of amplifica-
tion for both positive and negative results. Ideally, two positive and one
negative control should be run with each PCR, and other controls can be
run occasionally. The first positive control is the internal control to con-
firm the successful introduction of the cell into the RT tube and the sub-
sequent completion of the RT-PCR reactions. As mentioned above, we use
GAPDH transcript as the target for this control. This serves as the normalizer
for quantitative methods also. The second positive control should contain
mRNAs for all the targeted transcripts in quantities somewhat above the
limits of detection. We have used whole tissue total RNA at concentrations
of 1–5 ng/µl for this control that also serves as the calibrator (see above).
Detection of the targeted cDNA in this control confirms the success and
assesses the quality of the RT-PCR. Another control that can be used to con-
firm the specificity of the primers and success of the protocol is to test cells
with well-known expression profiles. For example, cerebellar Purkinje neu-
rons are GABAergic and we have detected GAD expression in them but not
ChAT, as expected (unpublished observation). Cloning and sequencing of
PCR products is the ultimate control for primer specificity, and is discussed
in more detail below (see section “Additional Methods for RT-PCR”).
An important negative control is conducted by processing a cell normally,
but omitting the reverse transcriptase. If amplification is detected, there
could be contamination of the PCR by genomic or other cDNA. Some in-
vestigators control for amplification of genomic DNA by the use of a DNAase
that can be inactivated by proteinase K and heat denaturation prior to
RT (Sucher and Deitcher, 1995). However, this sort of amplification is not
thought to be detectable in single cells undergoing only a single round of
40 PCR cycles. Careful primer design can minimize the potential for am-
plification of genomic DNA (see section “Primer and Probe Design”). For
real-time PCR using the Taqman r
system, any carryover cDNA contamina-
tion will contain “U” residues and is degraded prior to PCR by the enzyme
Amperase r
UNG (uracil-N-glycosylase). Other negative controls include
running the RT-PCR without aspirating a cell, after aspirating some of the
bath solution, and after aspirating noncellular debris. Ideally, the investiga-
tor conducting the PCR should be blind to the electrophysiological results
to control for bias.
III. APPLICATIONS
A. Single-Cell Identification by mRNA Expression of

Phenotypic Markers
For many experimental approaches, it is not necessary to quantitate the

PCR products. Often, merely detecting expression of target transcripts is
sufficient to correlate function with molecular identification. Neurons of-
ten can be identified by the expression of accepted phenotypic markers for
metabolism or transport of neurotransmitters. The adaptibility of RT-PCR
approaches to the identification of individual neurons makes it a useful tech-
nique for preparations such as dissociated neurons that cannot be subjected
to other procedures commonly used to label phenotypic markers. Even in
cases where a technique such as immunohistochemistry can be employed,
RT-PCR might be employed because of the relative ease of processing, the
potentially high-throughput capacity, and the ability to detect several molec-
ular species for each cell.
Approximately 90% of the neurons of rat basal forebrain are considered
to be cholinergic or GABAergic, and so we use primers designed to detect en-
zymes involved in the synthesis of acetylcholine (ChAT) and GABA (GAD).
Other well-known phenotypic markers include tyrosine hydroxylase and
vesicular glutamate transporters. A positive control marker should be tar-
geted in each cell so that its expression can validate a detection failure of the
phenotypic markers. Common positive controls include GAPDH, β-actin,
18S rRNAs, and neuron-specific enolase. It is a good practice to run a global
positive control along with each PCR to further validate detection failures.
Thus, if a particular basal forebrain neuron expressed GAPDH, but ChAT or
GAD was not detected, and the global positive control (purified whole basal
forebrain tissue mRNA) showed expression of ChAT and GAD, then the con-
fidence in the negative result is increased. A similar argument can be made
for successful detection of a target transcript from a cell validating a negative
detection of that transcript in other cells processed in the same PCR.
B. Single-Cell mRNA Expression and Patch-Clamp Recording
Combining physiological data with expression data can provide the

strongest confirmation of a neuron’s identity. This is particularly valuable in
cases of ambiguous expressions of phenotypic markers. For example, GAD
mRNA has been detected in some neurons that are clearly glutamatergic
(Cao et al., 1996) or that also express ChAT mRNA (Han et al., 2002; Tkatch
et al., 1998). Combined patch-clamp and RT-PCR has allowed us to show that
ChAT+/GAD+ neurons of the basal forebrain have physiological proper-
ties consistent with cholinergic, rather than with GABAergic cells (Han et al.,
2005). As shown in Fig. 5.5A, this acutely dissociated basal forebrain neuron
Figure 5.5. Combined patch-clamp recording of low-voltage-activated (LVA) calcium

current and scRT-PCR from a basal forebrain neuron. (A) LVA currents are gen-
erated by an inactivation protocol, consisting of different prepulses from –110 to
–50 mV, followed by a test pulse of –45 mV. The current in response to the test pulse
is shown. (B) Real-time PCR amplification plot showing the increased fluorescence
with each PCR cycle for specific transcripts for GAPDH, ChAT, GAD, and the LVA
calcium channel Cav3.2 (α1H ) subunit. Transcripts for other LVA channel subunits
(Cav3.1 and Cav3.3 ) were not detected.
Figure 5.6. Combined patch-clamp recording of NMDA-gated currents and single-

cell RT-PCR from a lateral/basolateral amygdala neuron. (A) NMDA (100 µM) and
glycine (3 µM) were applied either with or without the NR2B-selective anatagonist
ifenprodil (10 µM). The cellular contents of this neuron were then harvested and
subjected to RT-PCR with primers for the ubiquitous gene glyceraldehyde phosphate
dehydrogenase (GDH, bottom), for the 65 kD of the GABA synthetic enzyme gluta-
mate decarboxylase (GAD), and for the four NR2 subunits. Products were visualized
using standard agarose gel electrophoresis (size of products from the marker shown
at right). (B) Summary of all GAD+ neurons shows that NR2 expression was quite
heterogeneous in this neuronal population. All cells expressed more than one sub-
unit. In addition, NR2C and NR2D subunit mRNAs were detected in several neurons,
in contrast to the GAD− principal neurons that express primarily NR2A and NR2B
(Floyd et al., 2003).
displayed a low-voltage-activated (LVA) Ca2+ current typical of cholinergic

cells in response to an inactivation voltage-step protocol. Figure 5.5B shows
the real-time fluorescence amplification plot from that cell. GAPDH, ChAT,
GAD, and LVA Ca2+ channel α subunit CaV 3.2 (α1H ) were detected.
Another valuable application of combined patch-clamp and RT-PCR is to
identify and characterize the receptor subunits in native neurons. Figure
5.6A shows NMDA-activated currents and their modulation by the NR2B
receptor subunit specific inhibitor, ifenprodil, in an acutely dissociated rat
amygdala neuron. The ethidium bromide stained gel shown below detects
GAPDH, GAD, and NMDA receptor subunits 2A, 2B, and 2D, consistent with
the physiological result. The diversity of basolateral GAD+ amygdala neurons
is represented in Fig. 5.6B. Note that each cell expresses multiple subunits.
C. Single-Cell mRNA Expression, Patch-Clamp Recording, and

Fluorescent Calcium Measurements in Young and Aged Cells
Because the interface between scRT-PCR and patch clamping is the mi-
croelectrode, any application that can be combined with patch clamping
Figure 5.7. Ca2+ currents, fura-2 fluorescence ratio records, buffering curve, and
PCR amplification plot for a young neuron expressing transcripts for ChAT, GAD,
calretinin (CR), and GAPDH. (A) Ca2+ currents (top) and corresponding fura-2
fluorescent ratio records (bottom). (B) Buffering curve constructed from the data
at left. The slope of the linear portion of the curve is essentially the reciprocal of the
buffering value, 125. (C) Real-time PCR amplification plot showing the increased
fluorescence with each reaction cycle for specific transcripts for the cell marker
GAPDH, ChAT, GAD, and CR. Currents are generated by voltage steps Vh = −60
to 0 mV for different durations to create increasing Ca2+ influx. The ratio of fura-
2 fluorescence at 340 and 380 nm excitation (340/380) increases with increasing
intracellular Ca2+ concentrations.
can probably be combined with RT-PCR as well. We have taken advantage of

this versatility by including scRT-PCR with patch clamping and fura-2-based
Ca2+ -sensitive microfluorimetry to examine the Ca2+ buffering of rat basal
forebrain neurons. We used standard methods for isolating neuronal high-
voltage-activated Ca2+ currents from acutely dissociated neurons (Murchi-
son and Griffith, 1996) with the addition of 50 µM fura-2 to the pipette
solution and the application of a sterile RT-PCR protocol. The recordings
shown in Fig. 5.7A were made with a Cs-acetate-based pipette solution, but
the methods for recording and analyzing the fura-2 fluorescence signal are
the same as previously published (Murchison and Griffith, 1998). Briefly,
the Ca2+ currents are integrated to determine the amount of charge cross-
ing the membrane and this is converted to an expected concentration of
Ca2+ entry using a formula that estimates cell volume from the measured
membrane capacitance. These values are graphed against the peaks of the
calibrated fura fluorescence ratio signals. Fura-2 differentially changes the
intensity of its fluorescent emission from excitation wavelengths of 340
and 380 nm in the presence of different Ca2+ concentrations such that an
increasing ratio indicates increasing Ca2+ concentration. The resulting plot
Figure 5.8. Ca2+ currents and corresponding fura-2 fluorescence ratio records
in noncholinergic neurons identified by real-time RT-PCR from a 26-month rat
(A) and a 4-month rat (B). Recordings are as described in Fig. 5.7. PCR ampli-
fication plots show the fluorescence increases associated with each reaction cycle
for specific transcripts for the cell marker GAPDH, GAD, parvalbumin (PV), and
calretinin (CR). The cell in “A” is identified as GAD+ and PV+, while the cell in
“B” is identified as GAD+ and CR+. ChAT and CB were not detected in these
neurons.
is called the buffering curve (Fig. 5.7B) and the reciprocal of the slope ap-
proximates the value β, which represents the number of buffered Ca2+ ions
for every free ion. A larger buffering value indicates stronger cellular Ca2+
buffering. The real-time fluorescence amplification plot from this neuron
is shown in Fig. 5.7C. GAPDH, ChAT, GAD, and the CaBP CR were detected
in this cell, but the CaBPs calbindin (CB) and parvalbumin (PV) were not.
Note that there is no interference of the fura-2 fluorescence with the fluo-
rescent detection of the ABI Prism 7700 and Taqman r
fluorescent probes
because the excitation spectra do not overlap (fura-2 does not fluoresce at
the excitation wavelengths of the probes).
The above combination of methods can be applied just as well to basal
forebrain neurons from aged rats, as shown in Fig. 5.8. Young and aged
GAD+ neurons are shown for comparison. Note that both GAD+ neu-
rons display greater buffering than does the ChAT+/GAD+ neuron seen in
Fig. 5.7, despite the detection of a CaBP in each of the cells. All of the target
sequences detected in young basal forebrain neurons have been detected
in aged neurons also (Han et al., 2002, 2005).
IV. ADVANTAGES AND LIMITATIONS
Because scRT-PCR methods can be used with a number of experimental

preparations, it is worthwhile to mention several advantages and disadvan-
tages associated with various preparations. As shown in Figs. 5.2 and 5.4–
5.8, acutely dissociated neurons are often used in a variety of experiments
designed to study voltage- and ligand-gated currents. Visualization of an iso-
lated neuron has obvious advantages for enhancing electrode placement
and recording efficiency. Likewise, acutely dissociated cells are isolated and
therefore easily harvested without contamination from other cells. A sec-
ond important advantage of scRT-PCR analysis of acutely dissociated neu-
rons is that these cells are very difficult to identify with more traditional
methods, such as immunocytochemistry, because cells are not firmly at-
tached to the glass surface and are not amenable to recovery after fixation.
One disadvantage of acutely dissociated neurons is that they are removed
from their surrounding milieu and may lack relevant external influences.
This disadvantage is not unique to acutely dissociated neurons and may
be raised for any in vitro preparation; thus this limitation does not ap-
ply particularly to RT-PCR experiments. Neurons in primary tissue culture
(Fig. 5.3) have many of these same advantages for visualization and ease of
electrical recording mentioned above. Additional advantages include func-
tional synaptic circuits and cellular trophic interactions. One disadvantage
may be an increased potential for aspiration of neighboring tissue into the
harvesting pipette because individual neurons are not always isolated from
one another. Another disadvantage of primary cultures is that neurons in
culture do not necessarily represent adult expression patterns or cellular
function. Cultured neurons have been identified both electrophysiologi-
cally and immunocytochemically for many years and so it will be interesting
to determine how scRT-PCR cell identification compares with previous data.
Finally, some of the most physiological preparations available to study cellu-
lar physiology are thin brain slices. In slices, cells can be visually identified
and isolated for scRT-PCR (see references in earlier sections). Despite the
advantage of intact tissue, a disadvantage of this preparation is the fact
that the harvesting pipette must travel through surrounding tissue, and
there is an increased likelihood of contamination. Nevertheless, the utility
of scRT-PCR makes it worthwhile for application to a variety of experimental
preparations.
The following advantages and limitations apply to RT-PCR protocols in
general and to our scRT-PCR protocol as compared to other protocols.
A. Advantages
Utility: The protocol is amenable to a number of applications and is an ex-

cellent means of identifying neuronal phenotypes following functional
characterization.
Sensitivity: If the efficiencies of the RT and PCR are close to optimal, a

single copy of an mRNA target sequence should be amplified to reach
detection threshold by 40 PCR cycles using real-time fluorescence.
Simplicity: Particularly if real-time detection is employed, the efficiency of
data turnaround is higher and not as labor intensive as some protocols.
Reliability: Data are very reproducible with only a single round of PCR.
Quantifiability: By the use of real-time PCR, the relative abundance of a
transcript can be estimated with some confidence.
Stability: Once amplified, cDNA is more stable than is labile RNA.
Precision: Expression profile data can be obtained from a discrete pop-
ulation of cells, avoiding the possible misinterpretation of tissue level
profiling.
B. Limitations
Reliability: There is some potential for false negatives. If the efficiency of

the RT or PCR is not close to optimal, then some transcripts that were
present could go undetected.
Quantifiability: Different transcripts cannot be reliably compared quanti-
tatively. Rare transcripts are potentially subject to more noise due to
chance factors inherent in the first PCR cycles.
Selectivity: With our method, only a few targeted transcripts can be de-
tected from any one cell. Only cDNAs targeted with specific primers
are amplified.
Abundance: The small amount of RT product limits the number of tar-
gets that can be probed and the number of experimental replications
possible.
Indirectness: mRNA expression is shown but protein expression is un-
known.
Economy: Qualitative (gel-based) methods are the most economical but
can be the least informative, while real-time approaches are more
expensive.
V. APPENDIX: DETAILED METHODS
A. Neuronal Preparations
Adult male Fischer 344 or Sprague-Dawley rats or pregnant Sprague-

Dawley females were purchased from Harlan (Indianapolis, IN). Animals
took food and water ad libitum and were maintained on a 12 h light/dark
cycle. Handling and care of the animals were in accordance with policies of
Texas A&M University and the National Institute of Health.
Acutely dissociated basal forebrain neurons (medial septum and nu-
cleus of the diagonal band) and lateral/basolateral amygdala neurons
were obtained as described previously (McCool et al., 2003; Murchison
and Griffith, 1996). Briefly, isoflurane (Anaquest, Liberty Corner, NJ) anes-
thetized adult (2–27 mo) rats were decapitated. Coronal brain slices (400–
450 µm) were microdissected to isolate the MS/nDB or amygdala and en-
zymatically treated with either trypsin (∼0.7 mg/ml for basal forebrain;
Sigma Type XI) or pronase (∼0.5 mg/ml for amygdala, CalBiochem). After
trituration of an individual hemislice, cells were dispersed onto the glass
floor of a recording chamber and allowed to settle and stick to the glass
for 7–8 min. The recording chamber is mounted on an inverted micro-
scope and perfused at a rate of ∼2 ml/min, resulting in a bath turnover of
< 30 s. Experiments were performed at 20–21◦ C. Septal neurons were cul-
tured from 20-day embryos using established methods (Hsiao et al., 2004).
Septal slices were obtained and treated briefly with trypsin. After mechan-
ical dissociation, cells were centrifuged (1000 × g), resuspended (∼4.5 ×
106 cells/ml), and plated (∼50 µl) on 10.5 × 22 mm glass coverslips (acid
washed and coated with 50 µg/ml poly-d-lysine, Sigma) within 35-mm plastic
petri dishes. Plated cells were maintained in a “culture media” containing
equal volumes of D-MEM/F-12 and Neurobasal media (Gibco) with 2% B27
supplement (Gibco) and 1 µg/ml bovine serum albumin (Sigma). Up to
eight plastic dishes were grouped in a large 150 × 20 mm covered glass petri
dish and placed in a humidified 5% CO2 incubator.
B. Reverse Transcription
The individual tubes containing the initial reaction mixture [8 µl

nuclease-free water (Promega), 1 µl DTT (100 mM), 0.33 µl random hex-
amer primers (3 µg/µl), and 0.5 µl Rnase inhibitor (10 U/µl)] and each
aspirated cell or negative control is removed from the −80◦ C freezer and al-
lowed to warm on ice. They are then heated to 70◦ C for 10 min using a water
bath or heating block. This step denatures various unwanted proteins. The
RT tubes are then cooled on ice before adding the RT ingredients. Each tube
receives 4 µl of 5× first strand buffer (250 mM Tris-HCl, 375 mM KCl, 15 mM
MgCl2 ), 1 µl DTT (0.1 M), 0.5 µl RNase inhibitor (10 U/µl), 0.4 µl mixed
deoxynucleotide triphosphates (dNTPs 25 mM), and 1 µl Superscript r
II
reverse transcriptase (200 U/µl). This RT mixture is allowed to reach room
temperature before a 50 min exposure to 42◦ C to synthesize the cDNA. The
reaction is terminated by inactivating the reverse transcriptase with 70◦ C
for 15 min. After cooling on ice, 0.5 µl RNase H is added to each tube and
heated to 37◦ C for 20 min to remove the RNA from the RNA/DNA hybrid
to yield single-stranded cDNA. The resulting cDNA can be used for PCR
immediately or frozen for use later.
C. Solutions for PCR
Ingredients are from Invitrogen

r
unless indicated. For gel-based PCR,
the reaction mixtures contained 2.0 mM MgCl2 , 0.5 mM of each of the
dNTPs, 0.8–1.0 µM specific primers, 2.5 U Taq DNA polymerase (Promega),

5 µl 10× buffer (Promega), and 3–4 µl cDNA template from the single-
cell RT reaction. For real-time PCR, each well contained cDNA template
(3–5 µl), 2× Taqman r
universal PCR master mix (14–16 µl, AmpliTaqr
r
Gold DNA Polymerase, AmpErase UNG, dNTPs with dUTP, passive
reference dye ROX, optimized buffer component, Applied Biosystems),
0.3 µM specific forward and reverse primers, and 0.2 µM specific Taqman
r
probes.
D. Listing of Primers and Probes
1. Primers for Gel-Based PCR
The primers for ChAT, GAD, GAPDH, calbindin, calretinin, and parvalbu-
min that were used in gel-based PCR have been published previously (Han
et al., 2002). Likewise, primers for NMDA receptor subunits (Floyd et al.,
2003) and glycine receptor subunits (McCool and Farroni, 2001) have also
been published.
2. Primers and Probes for Real-Time RT-PCR
Forward and reverse primers Taqman

r probe
GAPDH 5 -CGCCCCTTCCGCTGAT-3 5 -CATGTTTGTGATGGG

5 -TGACAATCTTGAGGGAGTTG TGTGAACCACGAG-3
TCA-3
ChAT 5 -CCGGTTTGTCCTCTCCACCAG-3 5 -AGGTGCCCACAACC
5 -GGGACCACGGGTCCATAACA-3 ATGGAGATG-3
GAD 5 -ACGCCTTCGCCTGCAA-3 5 -TCCTCGAACGCGGG
5 -GGACGCAGGTTGGTAGTAT AGCGG-3
TAGGA-3
Calbindin 5 -AATTGTAGAGTTGGCCCAT 5 -CCCACCGAAGAGAAT
GTCTT-3 TTCCTGCTGC-3
5 -TCAGTTGCTGGCATCGAAAG-3
Calretinin 5 -GCAGAGCTGGCGCAGATC-3 5 -TGCCAACCGAAGAGA
5 -CCCACGTGCTGCCTGAA-3 ATTTCCTTTTGTG-3
Parvalbumin 5 -TCCTCAGATGCCAGAGACTT 5 -AGGAAACAAAGACG
GTC-3 CTGATGGCTGCT-3
5 -CCGTCCCCGTCCTTGTC-3
Cav 3.2 5 -TGCCTCCGACTGGTTTGTAAC-3 5 -TGCCGCTCCGAACGT
5 -AGTCGTCGAAGGCCTCCAA-3 TGCAG-3
E. Special Methods for Validation Experiment (Cloning, DNA
Preparation, Restriction Enzymes, and Sequencing)
1. Cloning
For ligation, each PCR product was mixed with centrifuged pGEM-T vec-
tor (50 ng, Promega), 2× rapid ligation buffer, T4 DNA ligase (3 Weiss
unit/µl, Promega), and nuclease-free water (Promega). This mixture was
incubated at room temperature for 1 h or 15◦ C for a few hours. Transfor-
mation was performed using frozen JM109 high-efficiency component cells
(Promega). Cells were thawed on ice just prior to use and mixed gently.
Cells (50 µl) were transferred to a tube, containing 2 µl of the ligation re-
action. Following gentle mixing and incubation for 20 min on ice, the tube
was heat-shocked for 45–50 s in a 42◦ C water bath. Incubation on ice was
performed for 2 min and then 950 µl SOS medium was added. After incuba-
tion for 1.5 h at 37◦ C with shaking (150 rpm), 100 µl of the transformation
culture was spread onto agar SOC plates containing ampicillin (50 µg/ml),
IPTG (isopropylthio-β-d-galactoside, Promega), and X-gal (chromogenic
substrate 5-bromo-4-chloro-3-indolyl-β-d-galactoside, Promega) for double
screening. Plates were incubated upside down overnight at 37◦ C. A white
colony indicated insertion of foreign DNA (PCR product) into a polycloning
site and these colonies were picked, plated, and incubated at 37◦ C for
overnight again for single colony isolation. A white single colony was in-
oculated into LB broth with ampicillin (50 µg/ml) and cultured overnight
for DNA preparation.
2. DNA Preparation
Large-scale DNA purification kits (Gibco BRL) were used for purify-
ing DNA from cultured cells, containing cloned DNA of GAPDH, ChAT,
GAD67, calbindin, calretinin, and parvalbumin. Overnight-cultured cells
were harvested (4000–5000 rpm, 10 min) and were suspended in a cell
suspension buffer (50 mM Tris-HCl, pH 8.0, 10 mM EDTA). A cell lysis
solution containing NaOH (200 mM) and SDS (1%) was added followed
by gentle tube inverting five times. Incubation occurred for 5 min with
neutralization buffer (potassium acetate 3.1 M, pH 5.5) and then centrifu-
gation was applied for 10 min (15, 000 × g at room temperature). The
supernatant was poured into equilibrated columns (equilibration buffer;
600 mM NaCl, 100 mM sodium acetate, pH 5.0, 0.15% Triton X-100) and
flowed by gravity. The columns were washed twice with wash buffer (800 mM
NaCl, 100 mM sodium acetate, pH 5.0) and elution buffer (1.25 M NaCl,
100 mM Tris–HCl, pH 8.5), and DNA was collected. 2-Propanol was used
for precipitation and 70% ethanol rinsed the pellet. After 10 min of air-
drying, the DNA pellet was dissolved with 200 µl TE buffer (10 mM Tris-HCl,
pH 8.0, 0.1 mM EDTA). DNA insertion was verified by gel electrophoresis

(1% agarose gel).
3. Restriction Enzyme Analysis and Sequencing
Purified DNA of cloned cultures from PCR products of GAPDH, ChAT,

GAD67, calbindin, calretinin, and parvalbumin were digested by different
restriction enzymes for verification of PCR product before sequencing and
isolation of target fragments for real-time PCR standard curves.
The restriction enzyme PstI for ChAT, NcoI for GAD, XmnI for calbindin
and parvalbumin, and SstI for calretinin were used (PstI, NcoI, SstI from
Gibco BRL; XmnI from Promega). Each of the purified DNAs was mixed
with the restriction enzyme (10 U/µl) and 10× buffers to reach final con-
centrations of 50 mM Tris-HCl (pH 8.0), 10 mM MgCl2 , and 50 mM NaCl
for PstI and SstI, and 100 mM NaCl, 6 mM Tris-HCl (pH 7.5), 6 mM MgCl2 ,
50 mM NaCl, and 1 mM DTT for NcoI and XmnI. Overnight incubation
at 37◦ C was performed and the size of the cut fragments was visualized by
electrophoresis (1.2% agarose gel). The correct bands were confirmed by
size, and the sequence of the inserted PCR product was performed using
pairs of forward and reverse primers (DNA Technologies Lab, Department
of Veterinary Pathobiology, Texas A&M University).
Different restriction enzymes were used to obtain purified cloned DNA
of PCR products for GAPDH, ChAT, GAD67, CB, CR, and PV standard
curves. EcoRI for GAPDH and PvuII for ChAT, GAD, CB, CR, and PV were
used to cut plasmid DNA. DNA fragments containing each PCR product
sequence were visualized by gel electrophoresis on low-melting agarose gel.
Bands of PCR products were sliced from the gel with a sterilized surgical
blade (Bard-Parker), and Wizard r
DNA Preps (Promega) were used for
DNA purification. Agarose slices were incubated at 70◦ C until the agarose
was completely melted. Resin (1 ml) was added and mixed thoroughly for
20 s. It was transferred to a minicolumn/syringe barrel assembled with a
vacuum manifold and vacuum was applied. The column was washed by 2 ml
of 80% 2-propanol and a vacuum was continued for 30 s to dry it. The
column was transferred to a new tube and 50 µl of water or TE buffer was
added. After 1 min incubation at room temperature, DNA was collected
following centrifugation for 20 s at 10, 000 × g. Eluted DNA was quantified
by spectrometry and kept at −20◦ C until use.
F. Additional Methods for RT-PCR
1. Cloning and Sequencing of PCR Products
Even though PCR products displayed the correct size, nonspecific ampli-
fication or contamination with similar size products may occur. Therefore,
we cloned each PCR product (ChAT, GAD, CB, CR, and PV) by ligation and
transformation. pGEM plasmids containing each of the PCR products (3 µg
total RNA purified from young rat basal forebrain) were separated by gel
electrophoresis. They were distinguished by size, being heavier than pGEM
alone but lighter than a pGEM+500 bp marker. All the plasmids came from
single colonies and were purified by DNA purification kits obtained from
Gibco BRL. Restriction enzyme analysis was performed to confirm the cor-
rect inserts of each product before sequencing. PstI for ChAT, NcoI for GAD,
XmnI for CB and PV, and SstI for CR were selected because they have only
two cutting sites outside or inside of the inserts, thus producing only two
fragments of known size. Depending on the direction of the inserts, there
were two possible sets of fragment sizes generated after restriction enzyme
treatment for each primer set. For further proof, plasmids with different
inserts for ChAT, GAD, CB, CR, and PV were sequenced using forward and
reverse primers and were shown to be a 100% match with their respective
genomic sequence. The same verification experiment was repeated using
total RNA from aged animals.
2. Optimization for Gel-Based PCR
Additional experiments were conducted to optimize the concentration

of Mg2+ for different primers with PCR. Because salt concentrations play
a critical role in amplification efficiency during PCR and different primers
may have different optimal concentrations, we examined concentrations of
0.5, 1, 2, 3, and 4 mM Mg2+ . The strongest gel-based signals for each of the
targets were detected with 2 mM Mg2+ concentration.
Acknowledgments. This study was supported in part by NIH grant

AG007805.
REFERENCES
Audinat, E., Lambolez, B., and Rossier, J., 1996, Functional and molecular analysis of glutamate-
gated channels by patch-clamp and RT-PCR at the single-cell level, Neurochem. Int. 28:119–
136.
Bargas, J., Howe, A., Eberwine, J., Cao, Y., and Surmeier, D. J., 1994, Cellular and molecu-
lar characterization of Ca2+ currents in acutely isolated, adult rat neostriatal neurons, J.
Neurosci. 14:6667–6686.
Bartus, R. T., Dean, R. L., III, Beer, B., and Lippa, A. S., 1982, The cholinergic hypothesis of
geriatric memory dysfunction, Science 217:408–414.
Bochet, P., Audinat, E., Lambolez, B., Crepel, F., Rossier, J., Iino, M., Tsuzuki, K., and Ozawa,
S., 1994, Subunit composition at the single-cell level explains functional properties of a
glutamate-gated channel, Neuron 12:383–388.
Cao, Y., Wilcox, K. S., Martin, C. E., Rachinsky, T. L., Eberwine, J., and Dichter, M. A., 1996,
Presence of mRNA for glutamic acid decarboxylase in both excitatory and inhibitory
neurons, Proc. Natl. Acad. Sci. U.S.A. 93:9844–9849.
Chen, K. C., Blalock, E. M., Thibault, O., Kaminker, P., and Landfield, P. W., 2000, Ex-
pression of alpha 1D subunit mRNA is correlated with L-type Ca2+ channel activity
in single neurons of hippocampal “zipper” slices, Proc. Natl. Acad. Sci. U.S.A. 97:4357–
4362.
Chow, N., Cox, C., Callahan, L. M., Weimer, J. M., Guo, L., and Coleman, P. D., 1998, Expression
profiles of multiple genes in single neurons of Alzheimer’s disease, Proc. Natl. Acad. Sci.
U.S.A. 95:9620–9625.
Coyle, J. T., Price, D. L., and DeLong, M. R., 1983, Alzheimer’s disease: a disorder of cortical
cholinergic innervation, Science 219:1184–1190.
Crino, P., and Eberwine, J., 1996, Molecular characterization of the dendritic growth cone:
regulated mRNA transport and local protein synthesis, Neuron 17:1173–1187.
Decker, M. W., 1987, The effects of aging on hippocampal and cortical projections of the
forebrain cholinergic system, Brain Res. Rev. 12:423–438.
Dixon, A. K., Richardson, P. J., Lee, K., Carter, N. P., and Freeman, T. C., 1998, Expression
profiling of single cells using 3 prime end amplification (TPEA) PCR, Nucleic Acids Res.
26:4426–4431.
Dixon, A. K., Richardson, P. J., Pinnock, R. D., and Lee, K., 2000, Gene-expression analysis at
the single-cell level, Trends Pharmacol. Sci. 21:65–70.
Eberwine, J., Kacharmina, J. E., Andrews, C., Miyashiro, K., McIntosh, T., Becker, K., Barrett, T.,
Hinkle, D., Dent, G., and Marciano, P., 2001, mRNA expression analysis of tissue sections
and single cells, J. Neurosci. 21:8310–8314.
Eberwine, J., Yeh, H., Miyashiro, K., Cao, Y., Nair, S., Finnell, R., Zettel, M., and Coleman,
P., 1992, Analysis of gene expression in single live neurons, Proc. Natl. Acad. Sci. U.S.A.
89:3010–3014.
Edwards, M. C., and Gibbs, R. A., 1994, Multiplex PCR: advantages, development, and applica-
tions, PCR Methods Appl. 3:S65–S75.
Elowitz, M. B., Levine, A. J., Siggia, E. D., and Swain, P. S., 2002, Stochastic gene expression in
a single cell, Science 297:1183–1186.
Fanselow, M. S., and LeDoux, J. E., 1999, Why we think plasticity underlying Pavlovian fear
conditioning occurs in the basolateral amygdala, Neuron 23:229–232.
Fibiger, H. C., 1982, The organization and some projections of cholinergic neurons of the
mammalian forebrain, Brain Res. 257:327–388.
Fischer, W., Gage, F. H., and Bjorklund, A., 1989, Degenerative changes in forebrain cholinergic
nuclei correlate with cognitive impairments in aged rats. Eur.J . Neurosci. 1:34–45.
Floyd, D. W., Jung, K. Y., and McCool, B. A., 2003, Chronic ethanol ingestion facilitates N-
methyl-d-aspartate receptor function and expression in rat lateral/basolateral amygdala
neurons, J. Pharmacol. Exp. Ther. 307:1020–1029.
Freeman, W. M., Walker, S. J., and Vrana, K. E., 1999, Quantitative RT-PCR: pitfalls and potential,
Biotechniques 26:112–125.
Gibson, U. E., Heid, C. A., and Williams, P. M., 1996, A novel method for real time quantitative
RT-PCR, Genome Res. 6:995–1001.
Griffith, W. H., Jasek, M. C., Bain, S. H., and Murchison, D., 2000, Modification of ion chan-
nels and calcium homeostasis of basal forebrain neurons during aging, Behav. Brain Res.
115:219–233.
Han, S-H., McCool, B. A., Murchison, D., Nahm, S. S., Parrish, A. R., and Griffith, W. H., 2002,
Single-cell RT-PCR detects shifts in mRNA expression profiles of basal forebrain neurons
during aging, Mol. Brain Res. 98:67–80.
Han, S-H., Murchison, D., and Griffith, W. H., 2005, Low voltage-activated calcium and fast
tetrodotoxin-resistant sodium currents define subtypes of cholinergic and noncholinergic
neurons in rat basal forebrain, Mol. Brain Res. 134:226–238.
Heid, C. A., Stevens, J., Livak, K. J., and Williams, P. M., 1996, Real time quantitative PCR,
Genome Res. 6:986–994.
Hinkle, D., Glanzer, J., Sarabi, A., Pajunen, T., Zielinski, J., Belt, B., Miyashiro, K., McIntosh,
T., and Eberwine, J., 2004, Single neurons as experimental systems in molecular biology,
Prog. Neurobiol. 72:129–142.
Hsiao, S. H., DuBois, D. W., Miranda, R. C., and Frye, G. D., 2004, Critically timed ethanol
exposure reduces GABAA R function on septal neurons developing in vivo but not in vitro,
Brain Res. 1008:69–80.
Johansen, F. F., Lambolez, B., Audinat, E., Bochet, P., and Rossier, J., 1995, Single cell RT-
PCR proceeds without the risk of genomic DNA amplification, Neurochem. Int. 26:239–
243.
Jonas, P., Racca, C., Sakmann, B., Seeburg, P. H., and Monyer, H., 1994, Differences in Ca2+
permeability of AMPA-type glutamate receptor channels in neocortical neurons caused by
differential GluR-B subunit expression, Neuron 12:1281–1289.
Killcross, S., Robbins, T. W., and Everitt, B. J., 1997, Different types of fear-conditioned be-
haviour mediated by separate nuclei within amygdala, Nature 388:377–380.
Lambolez, B., Audinat, E., Bochet, P., Crepel, F., and Rossier, J., 1992, AMPA receptor subunits
expressed by single Purkinje cells, Neuron 9:247–258.
Lindqvist, N., Vidal-Sanz, M., and Hallbook, F., 2002, Single cell RT-PCR analysis of tyrosine ki-
nase receptor expression in adult rat retinal ganglion cells isolated by retinal sandwiching,
Brain Res. Protoc. 10:75–83.
Liss, B., 2002, Improved quantitative real-time RT-PCR for expression profiling of individual
cells, Nucleic Acids Res. 30:e89.
Martina, M., Schultz, J. H., Ehmke, H., Monyer, H., and Jonas, P., 1998, Functional and molec-
ular differences between voltage-gated K+ channels of fast-spiking interneurons and pyra-
midal neurons of rat hippocampus, J. Neurosci. 18:8111–8125.
McCool, B. A., and Farroni, J. S., 2001, Subunit composition of strychnine-sensitive glycine
receptors expressed by adult rat basolateral amygdala neurons, Eur.J . Neurosci. 14:1082–
1090.
McCool, B. A., Frye, G. D., Pulido, M. D., and Botting, S. K., 2003, Effects of chronic ethanol
consumption on rat GABA(A) and strychnine-sensitive glycine receptors expressed by
lateral/basolateral amygdala neurons, Brain Res. 963:165–177.
McDonald, A. J., 1985, Immunohistochemical identification of gamma-aminobutyric acid-
containing neurons in the rat basolateral amygdala, Neurosci. Lett. 53:203–207.
McDonald, A. J., Beitz, A. J., Larson, A. A., Kuriyama, R., Sellitto, C., and Madl, J. E., 1989, Co-
localization of glutamate and tubulin in putative excitatory neurons of the hippocampus
and amygdala: an immunohistochemical study using monoclonal antibodies, Neuroscience
30:405–421.
Medhurst, A. D., Harrison, D. C., Read, S. J., Campbell, C. A., Robbins, M. J., and Pangalos, M.
N., 2000, The use of TaqMan RT-PCR assays for semiquantitative analysis of gene expression
in CNS tissues and disease models, J. Neurosci. Methods 98:9–20.
Mesulam, M. M., Mufson, E. J., Wainer, B. H., and Levey, A. I., 1983, Central cholinergic pathways
in the rat: an overview based on an alternative nomenclature (Ch1–Ch6), Neuroscience
10:1185–1201.
Miyashiro, K., Dichter, M., and Eberwine, J., 1994, On the nature and differential distribution of
mRNAs in hippocampal neurites: implications for neuronal functioning, Proc. Natl. Acad.
Sci. U.S.A. 91:10800–10804.
Monyer, H., and Lambolez, B., 1995, Molecular biology and physiology at the single-cell level,
Curr. Opin. Neurobiol. 5:382–387.
Monyer, H., and Markram, H., 2004, Interneuron diversity series: molecular and genetic tools
to study GABAergic interneuron diversity and function, Trends Neurosci. 27:90–97.
Mufson, E. J., Counts, S. E., and Ginsberg, S. D., 2002, Gene expression profiles of cholinergic
nucleus basalis neurons in Alzheimer’s disease, Neurochem. Res. 27:1035–1048.
Murchison, D., and Griffith, W. H., 1996, High-voltage activated calcium currents in basal
forebrain neurons during aging, J. Neurophysiol. 76:158–174.
Murchison, D., and Griffith, W. H., 1998, Increased calcium buffering in basal forebrain neu-
rons during aging, J. Neurophysiol. 80:350–364.
Olton, D. S., Wenk, G. L., and Markowska, A. M., 1991, Basal forebrain, memory and attention,
In: Richardson, R. T. (ed.), Activation to Acquisition: Functional Aspects of the Basal Forebrain
Cholinergic System, Boston: Birkhauser, pp. 247–262.
Panula, P., Revuelta, A. V., Cheney, D. L., Wu, J. Y., and Costa, E., 1984, An immunohistochemical
study on the location of GABAergic neurons in rat septum, J. Comp. Neurol. 222:69–80.
Pape, J. R., Skynner, M. J., Sim, J. A., and Herbison, A. E., 2001, Profiling gamma-aminobutyric
acid (GABA(A)) receptor subunit mRNA expression in postnatal gonadotropin-releasing
hormone (GnRH) neurons of the male mouse with single cell RT-PCR, Neuroendocrinology
74:300–308.
Phillips, J. K., and Lipski, J., 2000, Single-cell RT-PCR as a tool to study gene expression in
central and peripheral autonomic neurones, Auton. Neurosci. 86:1–12.
Plant, T. D., Schirra, C., Katz, E., Uchitel, O. D., and Konnerth, A., 1998, Single-cell RT-PCR and
functional characterization of Ca2+ channels in motoneurons of the rat facial nucleus, J.
Neurosci. 18:9573–9584.
Roy, S. W., Fedorov, A., and Gilbert, W., 2003, Large-scale comparison of intron positions in
mammalian genes shows intron loss but no gain, Proc. Natl. Acad. Sci. U.S.A. 100:7158–7162.
Ruano, D., Perrais, D., Rossier, J., and Ropert, N., 1997, Expression of GABA(A) receptor
subunit mRNAs by layer V pyramidal cells of the rat primary visual cortex. Eur.J . Neurosci.
9:857–862.
Rye, D. B., Wainer, B. H., Mesulam, M. M., Mufson, E. J., and Saper, C. B., 1984, Cortical
projections arising from the basal forebrain: a study of cholinergic and noncholinergic
components employing combined retrograde tracing and immunohistochemical localiza-
tion of choline acetyltransferase, Neuroscience 13:627–643.
Santi, M. R., Vicini, S., Eldadah, B., and Neale, J. H., 1994, Analysis by polymerase chain reaction
of alpha 1 and alpha 6 GABAA receptor subunit mRNAs in individual cerebellar neurons
after whole-cell recordings, J. Neurochem. 63:2357–2360.
Sarter, M., and Bruno, J. P., 2000, Cortical cholinergic inputs mediating arousal, attentional
processing and dreaming: differential afferent regulation of the basal forebrain by telen-
cephalic and brainstem afferents, Neuroscience 95:933–952.
Sarter, M., and Bruno, J. P., 2002, The neglected constituent of the basal forebrain corticopetal
projection system: GABAergic projections, Eur.J . Neurosci. 15:1867–1873.
See, R. E., Fuchs, R. A., Ledford, C. C., and McLaughlin, J., 2003, Drug addiction, relapse, and
the amygdala, Ann. NY. Acad. Sci. 985:294–307.
Song, W. J., Tkatch, T., Baranauskas, G., Ichinohe, N., Kitai, S. T., and Surmeier, D. J., 1998,
Somatodendritic depolarization-activated potassium currents in rat neostriatal cholinergic
interneurons are predominantly of the A type and attributable to coexpression of Kv4.2
and Kv4.1 subunits, J. Neurosci. 18:3124–3137.
Stahlberg, A., Hakansson, J., Xian, X., Semb, H., and Kubista, M., 2004, Properties of the reverse
transcription reaction in mRNA quantification, Clin. Chem. 50:509–515.
Subramaniam, S., 1998, The biology workbench—a seamless database and analysis environ-
ment for the biologist, Proteins 32:1–2.
Sucher, N. J., and Deitcher, D. L., 1995, PCR and patch-clamp analysis of single neurons, Neuron
14:1095–1100.
Sucher, N. J., Deitcher, D. L., Baro, D. J., Warrick, R. M., and Guenther, E., 2000, Genes and
channels: patch/voltage-clamp analysis and single-cell RT-PCR, Cell Tissue Res. 302:295–
307.
Surmeier, D. J., Song, W. J., and Yan, Z., 1996, Coordinated expression of dopamine receptors
in neostriatal medium spiny neurons, J. Neurosci. 16:6579–6591.
Tkatch, T., Baranauskas, G., and Surmeier, D. J., 1998, Basal forebrain neurons adjacent to
the globus pallidus co-express GABAergic and cholinergic marker mRNAs, Neuroreport
9:1935–1939.
Whitcombe, D., Brownie, J., Gillard, H. L., McKechnie, D., Theaker, J., Newton, C. R., and
Little, S., 1998, A homogeneous fluorescence assay for PCR amplicons: its application to
real-time, single-tube genotyping, Clin. Chem. 44:918–923.
Yan, Z., and Surmeier, D. J., 1996, Muscarinic (m2/m4) receptors reduce N- and P-type Ca2+
currents in rat neostriatal cholinergic interneurons through a fast, membrane-delimited,
G-protein pathway, J. Neurosci. 16:2592–2604.
Zaborszky, L., Carlsen, J., Brashear, H. R., and Heimer, L., 1986, Cholinergic and GABAergic
afferents to the olfactory bulb in the rat with special emphasis on the projection neurons
in the nucleus of the horizontal limb of the diagonal band, J. Comp. Neurol. 243:488–509.
6
Merging Structure and
Function: Combination of
In Vivo Extracellular and
Intracellular
Electrophysiological
Recordings with
Neuroanatomical Techniques
ATTILA SÍK
INTRODUCTION
IN VIVO EXTRACELLULAR RECORDING
IN VIVO INTRACELLULAR RECORDING AND SINGLE-CELL
LABELING
COMBINED TECHNIQUES
Advantages
Limitations
Anesthetics
Surgery and Implantation of Stimulating and Extracellular
Recording Electrodes
Intracellular Recording
Survival Time
Fixation
Visualization of the Intracellularly Filled Cells
ATTILA SÍK • Centre de Recherche Université Laval Robert-Giffard, 2601 De la

Canardiere, Québec, Québec G1J 2G3, Canada
175
176 ATTILA SÍK
Immunolabeling
Reconstruction
Equipment and Supplies
CHEMICALS
SOLUTIONS
REFERENCES
Abstract: In order to understand the involvement of activity of single neurons in the

context of the activity generated in small or larger neuronal networks, electrophysio-
logical methods and morphological techniques need to be combined. In this chapter
I describe a combination of methods designed to enable researchers to record the
electrophysiological activity of single neurons in the intact brain, to analyze the
interactions of these identified neurons with the surrounding neuronal network,
and to investigate afterward the neuroanatomical characteristics of the recorded
neurons.
Keywords: extracellular and intracellular recording, reconstruction, immunohisto-
chemistry
I. INTRODUCTION
The deciphering of the function of neurons and neuronal networks in the

intact brain is one of the major challenges in neuroscience. The electrophys-
iological properties of individual neurons and neuronal networks have been
studied in great detail for many decades. However, to understand the role
of various types of neurons in the genesis and/or the modulation of neural
network activity as it is reflected in complex EEG patterns, it is necessary to
record single-neuron activity simultaneously with the behavior of these sin-
gle cells within the context of activity of the neuronal network in which they
are embedded. In addition, the combination of intracellular and extracel-
lular electrophysiological recordings with the subsequent morphological
and neurochemical identification of the electrophysiologically character-
ized neurons provides additional information that is essential to understand
the structural-functional relationship of neuronal networks.
Combined electrophysiological/network/morphological identification is
necessary for the construction of realistic neuronal network simulation mod-
els. Such models need to take into account information about the com-
plete physiological and anatomical properties: genomic and neurochemical
content, receptor expression and other molecular information, knowledge
about the patterns of dendrite and axon arborizations, information about
the input (number, type, and origin of synapses terminating on the soma
and on the dendritic trees), the output (number of axon terminals and the
dendritic domain of postsynaptic targets), and finally information about
the types and numbers of synaptically connected partners of different cell
types. The degree of convergence and divergence of input/output among
COMBINATION OF IN VIVO ELECTROPHYSIOLOGY WITH NEUROANATOMY 177
different neuronal populations is a crucial element in neuronal network
simulations. Therefore, research has explored these questions, yet mostly
in vitro. In spite of the obvious advantages of using slice preparations (e.g.,
controlled administration of pharmacological reagents, multiple intracellu-
lar recording, etc.), the in vitro approach has the substantial disadvantage
that the neuronal network is artificially truncated. Measuring neuronal and
network activity in the in vitro hippocampal slice preparation illustrates this
point. In the intact hippocampus, the sizes, densities, and trajectories of the
axonal arborizations of inhibitory and excitatory neurons are very different
from each other (Li et al., 1993; Sik et al., 1993; Freund and Buzsaki, 1996).
Since in a slice far more excitatory axons are truncated than inhibitory
ones, the consequence of slicing is that the excitation–inhibition balance
becomes unnaturally modified in the in vitro slice compared with the intact
situation. It has been estimated that in the in vitro hippocampal slice prepa-
ration as much as 80–90% of excitatory terminals arising from an excitatory
pyramidal cell may be truncated (Li et al., 1993; Sik et al., 1993), whereas
inhibitory axon arborization remains relatively unaffected (e.g., “O-LM”,
basket cells, etc.; Sik et al., 1995). Several types of inhibitory neurons can be
seriously truncated though [e.g., “backprojection” inhibitory neurons (Sik
et al., 1994, 1995)]. Because of this artificial condition, neuronal network
properties have to be investigated in vitro with great caution. To overcome
this problem and to reliably investigate network properties of neurons, the
most recent generation of studies has been designed to obtain informa-
tion about the function of morphologically identified neurons in the intact
brain. Several studies have been published in which intracellular recording
of neurons was first obtained in anesthetized animals followed by labeling
and subsequent neurochemical analysis of the same neurons (Sik et al., 1994,
1995, 1997).
Compared to the recording in in vitro slice preparations, the major chal-
lenge in the in vivo electrophysiological experiment is that the researcher
has to perform the recordings without visual control. Furthermore, the brain
in living animals is by no means as static as in slices, which makes it very dif-
ficult to conduct sustained intracellular recordings and to successfully fill
the recorded cell and all its processes with a marker substance once the
recordings have been completed. Therefore, it may not be surprising that
the number of analyzed cells in research articles reporting in vivo intracel-
lular recording is substantially lower than in articles using in vitro approach
(Li et al., 1992; Sik et al., 1993, 1994, 1995, 1997).
In order to characterize the neuroanatomical and neurochemical proper-
ties of the electrophysiologically characterized cells, a combination of elec-
trophysiology and various neuroanatomical techniques is required. The sim-
plest technique used in the past to localize the position of the extracellular
recording was to simply dip the electrode tip prior to recording in a dye
such as methyl blue or fast green, which left in the brain tissue a mark of the
position of the electrode (Grossman and Hampton, 1968; Simons, 1978;
Takato and Goldring, 1979; Thomas and Wilson, 1965). The resolution
178 ATTILA SÍK
of this method was obviously inadequate to determine the location of

the electrode with great precision. When single-cell recording methods
emerged, a reliable technique that allowed the researcher to successfully
visualize the electrophysiologically recorded cell was highly sought after. At
first, the new marker horseradish peroxidase (HRP), being at that time pop-
ular in neuroanatomical tract-tracing (Heimer and Robards, 1981), found
a second “killer application” as an intracellular marker in neurophysiology.
HRP clearly delineates the cell’s geometry and it supplies as a bonus an
electron-dense label that highlights the labeled neuron and its processes in
ultrathin sections studied in the electron microscope (Chang et al., 1981;
Jankowska et al., 1976; Kita and Kitai, 1986; Kitai et al., 1976; Snow et al.,
1976; Tamamaki et al., 1987; Tepper et al., 1987). However, intracellular in-
jection of HRP enables only a partial analysis of the axonal arborization of
neurons because not all axons become filled in their entire trajectory. Flu-
orescent dyes like Procion yellow (Hassin, 1979; Kaneko, 1970; Kelly and
Van Essen, 1974) and the highly fluorescent Lucifer yellow were also used
with success (Stewart, 1978, 1981; Takato and Goldring, 1979). Although
Lucifer yellow produces better details as a marker compared to HRP, it has
a major disadvantage as well: as all fluorescent dyes it loses its light emission
capability after a relatively short period of UV illumination (the so-called
“bleaching” effect). Therefore, the full reconstruction of the axonal ar-
borization of intracellularly labeled neurons has been difficult, impossible,
or dependent on the ability of the researcher to apply the complicated
technique of diaminobenzidine photoconversion (Sandell and Masland,
1988). Thus, complete reconstruction of a neuron requires the intracel-
lular injection of a molecule with a low molecular weight, providing fast
and complete diffusion into small neuronal appendages, combined with
a visualization method that produces a stable, optical, and electron-dense
end product. Sensitive intracellular recording followed by anatomical iden-
tification of the cell today is based on a widely used biotin-containing
low-molecular-weight product called biocytin (Horikawa and Armstrong,
1988). Biocytin appeared to be superior to HRP in completely staining
the dendritic trees and axonal arborizations in their finest details, both
in vitro (Horikawa and Armstrong, 1988; Kawaguchi et al., 1989) and in
vivo (Kawaguchi et al., 1990; Sik et al., 1995, 1997). Recording of the elec-
trophysiological activity of single neurons followed by injection of bio-
cytin and completed with the analysis of the neuron’s interactions with
the neuronal network is used to study the complex question of neuronal
function.
II. IN VIVO EXTRACELLULAR RECORDING
Various extracellular recording methods have been developed over the

past decades with the purpose to monitor the electrical activity of neural
networks. All these methods require that a low-resistance conductor should
be placed in the region of interest. One of the simplest methods is to use a
metal wire (Hubel, 1957). Because neurons of the same electrophysiological
class generate similar action potentials, the only way to identify a given neu-
ron from extracellularly recorded spikes is to move the electrode tip closer
to its cell body (minimum distance is 20 µm in cortex) than to any other
neurons. Neuron separation by this method is guaranteed by the differential
proximity of the recording tip and one neuron, relative to the other neurons
(see for details the chapter by Duque and Zaborszky in this volume). The
substantially larger amplitude spikes, relative to the background “noise,”
guarantee neuron isolation. To record from another neuron, another elec-
trode is needed. Because electrical recording from neurons is invasive, mon-
itoring from larger numbers of neurons with the one-electrode–one-neuron
approach inevitably increases tissue damage. Thus, improved methods are
needed for the simultaneous recording of closely spaced neuronal popula-
tions with minimal damage to the hard wiring of the brain network.
The recent advent of localized multisite extracellular recording tech-
niques has dramatically increased the yield of isolated neurons (Gray et al.,
1995; McNaughton et al., 1983; Wilson and McNaughton, 1993). With only
one recording site, signals from many neurons with similar size and ori-
entation and which are at the same distance from the tip will provide the
same magnitude signal, making single-cell isolation difficult. The use of two
or more recording sites allows the triangulation of distances because the
amplitude of the recorded spike is a function of the distance between the
neuron and the electrode (Henze et al., 2000; see for details the chapter by
Nadasdy et al. in this volume). Wire tetrodes have numerous advantages over
sharp-tip single electrodes, including larger yield of units, low-impedance
recording tips, and mechanical stability. Because the recording tip needs
not to be placed in the immediate vicinity of the neuron, long-term record-
ings in behaving animals are possible. Microelectromechanical system-based
recording devices can reduce the technical limitations inherent in wire
electrodes because with the same amount of tissue displacement the num-
ber of monitoring sites can be substantially increased (Bartho et al., 2004;
Buzsaki, 2004; Csicsvari et al., 2003; Wise and Najafi, 1991). Whereas silicon
probes have the advantages of tetrode recording principles, they are sub-
stantially smaller in size. Furthermore, multiple sites can be arranged over
a longer distance, thus allowing the simultaneous recording of neuronal ac-
tivity in various cortical layers (Buzsaki and Kandel, 1998). Currently avail-
able multishank probes can record from as many as hundred well-separated
neurons.
Cortical pyramidal cells generate extracellular currents that flow mostly
parallel with their somatodendritic axis. These extracellular features allow
the separation of signals related to individual neurons. In practice, only a
small fraction of all possible neurons can be reliably separated with the cur-
rently available probes and spike-sorting algorithms. Data processing and
viewing algorithms are freely available. Neurophysiological and behavioral
data can be explored by NeuroScope (http://neuroscope.sourceforge.net)
180 ATTILA SÍK
(Buzsaki et al., 2004). Spike-sorting is performed in two steps, first automat-

ically using KlustaKwik (http://klustakwik.sourceforge.net) (Harris et al.,
2000) and then manually using Klusters (http://klusters. sourceforge.net)
(Hazan et al., 2004). A further advantage of silicon probe monitoring of elec-
trical activity is that the closely spaced recording sites have a known one or
two dimensions. The multiple site approach allows the simultaneous moni-
toring of the extracellular flow of ions with high spatial resolution. From the
measured voltages the current flow can be calculated, and the extracellular
resistivity can be used to calculate the current density. Such current-source
density measurements provide valuable information for identifying synaptic
pathways and neuronal compartments responsible for generating the locally
measured current. For example, if a spatial distribution of current-source
density of a spontaneous field pattern matches that of the evoked currents
by thalamic but not by callosal inputs, the firm conclusion can be drawn that
the spontaneous pattern is generated by thalamic afferents (Bragin et al.,
1995; Buzsáki et al., 2003; Nadasdy et al., 1998).
A critical step in the reconstruction of a functional circuit is the identifi-
cation of the anatomical nature of the recorded and spike-sorted units. This
identification is possible only with a combined program comparing extracel-
lular and intracellular spike shape and spike dynamics of morphologically
identified neurons. The method requires several steps and dedicated exper-
iments. The examples below are taken from such experiments in vivo in the
hippocampus, but the method is compatible with neocortical areas or other
structures as well, i.e., there seems to be no limit for the identification of
extracellular spikes in behaving animals.
III. IN VIVO INTRACELLULAR RECORDING

AND SINGLE-CELL LABELING
Extracellular recording of local fields and/or large numbers of neurons

provides information about the cooperative activity of neuronal assemblies,
a type of parameter that restricts the patterns of firing of individual cells.
If one is interested in the intrinsic properties of individual cells, the inputs
that an individual cell receives, the fluctuations of membrane potential, and
so forth, then intracellular recording needs to be performed.
For in vivo intracellular recording, glass pipettes with very small tip diam-
eters (<0.5 µm) called sharp electrodes are used most frequently. Recently,
patch-pipettes have been used in vivo in order to record the electrical ac-
tivity of single neurons (Ferster and Jagadeesh, 1992; Margrie et al., 2002).
More details on in vivo patch recording are presented in this volume in the
chapter contributed by Petersen.
Sharp electrodes are manufactured using glass pipettes of different di-
ameters with the aid of an electrode puller. Since in vivo recording often
requires reaching deeper brain structures, the geometry of the electrode
is different from the one used in vitro: the shanks of the pipettes must be
sufficiently long. However, long shank electrodes may be too flexible. To
reach deeper brain regions with a high degree of precision, it is necessary
that the electrode does not drift from the planned path. This is difficult to
achieve with flexible electrodes. Therefore, usually for in vivo intracellular
recording, glass electrodes with rather thick glass walls are being used. An
alternative is to use an even more rigid electrode made of quartz. The dis-
advantage using quartz is that this material requires a special and expensive
electrode puller (e.g., Shutter P2000), where the glass is melted by laser
light instead of a tungsten-heating element.
Once the electrode is pulled it is filled up with conductive electrolyte. If
no special recording condition is required, usually 0.5 M potassium acetate
solution is used for this purpose. Since the tip of the electrode is very small,
the filling of the pipette is a process that has to be done with the necessary
care and precaution. To support homogeneous filling, glass electrodes with
inner filament are used, which, by the capillary effect, drastically increases
the probability of a successful filling process of the pipette. Basic rule is
that the recording pipette needs to be filled with electrolyte without any
discontinuity (for example, air bubbles should be absent). This is not an easy
task: simply placing the fluid inside the pipette from the back will certainly
produce a certain amount of air bubbles, rendering the pipettes useless. To
avoid the situation of ending up with a batch of sharp yet useless pipettes
because of poor filling, the following steps are recommended: (1) place
small droplets of conductive electrolyte at the open end of the pipette. Hold
this still for a couple of minutes until the liquid fills the extreme tip of the
pipette completely, (2) use a small diameter filling tube that is being inserted
into the recording pipette down to its neck and then slowly and carefully fill
the neck and shank up with the electrolyte, and (3) tap gently on the side
of the pipette if bubbles are still present. This manipulation can eliminate
discontinuities altogether; alternatively, one can use a very fine tungsten or
other rigid wire to reach inside the pipette and try to remove any air bubble.
If the goal of the experiment is to anatomically identify the recorded
neurons, some dye needs to be injected into the cells. For this purpose,
a tracer needs to be incorporated in the electrolyte. This tracer can be a
fluorescent dye like Lucifer yellow, or it can be a nonfluorescent substance
like the widely used small biotinylated molecules (biocytin, neurobiotin), or
even a mixture of several dyes. Dye injection can be achieved through the
application of high pressure (Sik et al., 1993; Tamamaki and Nojyo, 1993)
or by the application of an electrical current (i.e., if the dye is polarized).
Most widely used intracellular labeling material is biocytin. This substance
is a biotin–lysine complex of low molecular weight containing about 65%
biotin, which retains a high affinity for avidin. Because of the high affin-
ity of biotin to avidin, the conventional avidin–biotin complex method is
conveniently used to reveal the recorded cell. The biotin-containing tracer
can be injected either with positive current (i.e., biocytin) (Horikawa and
Armstrong, 1988) or with both positive and negative current pulses (i.e.,
neurobiotin) (Kita and Armstrong, 1991).
182 ATTILA SÍK
IV. COMBINED TECHNIQUES
The combination of the described in vivo electrophysiological and

anatomical methods allows both the functional and the structural character-
ization of neurons. First, the intrinsic electrophysiological properties of the
neurons are investigated using intracellular recording methods (Fig. 6.1) in
a condition where all the synaptic connections are intact and the extracellu-
lar environment is undisturbed. Second, the activity of each neuron is cor-
related with the neural network activity that is recorded by the extracellular
electrode. Even though in many cases a rough classification of the recorded
neuron can be achieved by analysis of the electrophysiological traces (exci-
tatory, inhibitory neuron, etc.), the exact characterization requires labeling
of the neuron (Fig. 6.2). Via the injection of a tracer the recorded cell can
be visualized and identified neurochemically (Fig. 6.3). Because of its low
molecular weight, biocytin diffuses easily into small structures. This charac-
teristic makes the detection of complete axonal and dendritic arborization
possible (Fig. 6.4). Using an additional immunoreaction and processing
for electron microscopy, even the targets or afferents of the investigated
neurons can be studied at the ultrastructural level (Sik et al., 1995).
The power of the combined methods is demonstrated below in a hip-
pocampal inhibitory (basket) cell, but naturally the same method can be
adapted to any neuronal type. With the appreciation of the pivotal function
of inhibitory cells in the orchestration of neural activity many questions
need to be answered; for example, how inhibitory and excitatory inputs
change the membrane potential of the neuron, how different types of in-
hibitory neurons fire, how different subtypes of inhibitory cells participate
in network oscillations, how inhibitory cells are connected to excitatory ver-
sus inhibitory neurons, what is the size of an area that a single inhibitory
neuron can innervate, how many excitatory and inhibitory neurons are in-
nervated by a single inhibitory cell, etc. After analyzing the spontaneous
activity of the basket cell during theta and non-theta network oscillations
(not shown), we analyzed the response of the neuron to positive and nega-
tive current injections. The recorded inhibitory cell was firing rhythmically
when theta activity was present in the extracellular field potential (Fig. 6.5B).
The neuron showed firing frequency accommodation when positive current
was injected intracellularly and found no sign of sag (Ih ) current (Fig. 6.5A).
Figure 6.1. (A) Simultaneous recording of extracellular theta activity in the pyra-
midal cell layer of area CA1 of the rat hippocampus (EC theta) and intracellular
activity of a basket neuron (IC theta). (B) Reconstruction of the dendritic and ax-
onal arborization of the basket cell. (Reprinted with permission from Ylinen et al.,
1995.)
Figure 6.2. Photograph of the recording micropipette track (arrowheads) and the
biocytin-filled neuron (arrows) and its camera lucida reconstruction. An intracellular
recording was made from the apical shaft of a pyramidal cell at the border of stratum
radiatum (rad) and lacunosum-moleculare (l-m) in area CA1 of the rat hippocampus.
The electrode was moved beyond the dendrite during the experiment. The pipette
track is filled with reaction product caused by peroxidase activity in red blood cells.
Abbreviation: pyr, pyramidal layer. Scale bar: 50 µm. (Reprinted with permission from
Kamondi et al., 1998.)
Streptavidin Parvalbumin
Figure 6.3. Immunohistochemical identification of the filled cell. The electrophysio-

logical property of the neuron was recorded: the cell fired close to the top of theta os-
cillation and also during ripple activity (not shown). After recording and subsequent
filling of the neuron, the visualization was achieved using fluorochrome-conjugated
(Alexa 546) streptavidin (left panel). The picture was taken using a confocal laser
microscope. The parvalbumin expression of the cell (arrow) was demonstrated by
immunofluorescence using a different fluorochrome (Alexa 488) (right panel). The
cell was later reconstructed and classified as basket cell (Dumont and Sik, 2006).
Abbreviations: o, stratum oriens; p, stratum pyramidale; r, stratum radiatum. Scale
bar: 25 µm.
184 ATTILA SÍK
mm
CA1
4 alv
S T
or
1 0 -1 mm pyr
hf
ml
gr
hil
CA3 DG
gr
−60
Figure 6.4. Reconstruction of an intracellularly recorded and subsequently filled

feedback neuron in area CA1 of the rat hippocampus. After “developing” the filled
neuron using the DAB–Ni method, the complete dendritic and axonal arborization
of the neuron could be reconstructed. The cell body of the neuron is located in the
alveus; axon collaterals are present in areas CA3, CA1, and in the hilus of dentate
gyrus (DG). The upper left inset indicates the summated length of axon collateral
along the septotemporal axis. The lower left inset shows the spontaneous activity of
the neuron at resting membrane potential. Abbreviations: S, septal direction; T, tem-
poral direction; alv, alveus; or, stratum oriens; pyr, stratum pyramidale; rad, stratum
radiatum; hf, hippocampal fissure; ml, molecular layer; gr, granule cell layer; hil,
hilus. Scale bar: 100 µm. (Reprinted from Sik et al., 1994.)
→
Figure 6.5. Complete electrophysiological, neurochemical, and neuroanatomical
characterization of a hippocampal basket cell. (A) Response of the neuron to de-
polarizing and hyperpolarizing current injection. (B) Simultaneous recording of
intracellular activity of the basket neuron and extracellular activity during theta
oscillation. (C) Biocytin-labeled basket cell. (D) Fluorescent parvalbumin immuno-
labeling of the same neurons (arrow). (E) Parvalbumin-containing target of the
filled basket cell. In the inset, the white arrow indicates a putative synaptic contact
between the biocytin-filled terminal and a parvalbumin-immunoreactive neuron.
(F) Correlated electron microscopic analysis of the same bouton shows a symmetric
synapse (arrow in the inset) on the cell body. (G) Partial reconstruction of the labeled
basket cell indicating other parvalbumin-immunoreactive targets (large circles). In-
set shows the position of the neuron in the CA1 region of the hippocampus. (H)
Distribution of pyramidal cell and parvalbumin-immunoreactive inhibitory neuron
targets of the intracellularly filled basket cell in the septotemporal direction. Overall,
99 boutons in contact with 64 parvalbumin-positive cells were counted. Graph in the
A B
C D
E F
H I
Figure 6.5. (Cont.) middle shows the probability of pyramidal and parvalbumin-
immunoreactive inhibitory cells innervated by the filled neuron. (I) The 2D dis-
tribution of the interneuron–interneuron contacts is shown in H. Abbreviations: pyr,
pyramidal; PV, parvalbumin; S, position of the soma. (Adapted, with permission,
from Wang and Buzsaki, 1996, and Sik et al., 1995.)
186 ATTILA SÍK
The filled cell contained parvalbumin (PV) (Fig. 6.5C,D) and innervated
other PV-immunoreactive neurons (Fig. 6.5E). The putative synaptic con-
tacts identified under the light microscope were further analyzed using the
electron microscope (Fig. 6.5F). Indeed, the filled basket cells formed sym-
metric (inhibitory) synaptic contacts on other PV-containing neurons be-
sides terminating on pyramidal cells. The number of synaptic contacts and
synaptic targets were determined (Fig. 6.5G). Overall, 99 boutons in contact
with 64 PV-positive cells were counted. The total number of pyramidal cell
targets (∼1500) was estimated by counting the number of boutons of the
filled basket cell in each 60-µm-thick Vibratome r
section, and assuming
that a basket cell formed 9–10 boutons on a single pyramidal cell (Halasy
et al., 1996). The probability of contacts formed on PV-containing versus
pyramidal cells was calculated by dividing the number of contacted PV or
pyramidal cells by the total number of PV or pyramidal cells in the area
innervated by the axon collaterals. The probability of postsynaptic contacts,
however, decreased with the distance between the cell pairs (Fig. 6.5H,I).
Thus, with the sequential application of the aforementioned methods the
complete electrophysiological and neuroanatomical characterization of sin-
gle neuron was achieved.
A computer simulation based on the obtained quantitative data demon-
strates that inhibitory synaptic transmission could provide a suitable mech-
anism for synchronized oscillations in a sparsely connected network of
inhibitory cells. This network can, through subthreshold oscillations in
excitatory cell populations, synchronize discharges of spatially distributed
excitatory neurons (Wang and Buzsaki, 1996).
V. SUMMARY OF ADVANTAGES AND LIMITATIONS
A. Advantages
Combination of in vivo intracellular recordings with morphological meth-

ods can provide crucial and detailed information about the structure and
function of neurons. The quantitative data of connections of a single intact
neuron are indispensable for further meaningful computer simulation of
neural networks.
B. Limitations
In spite of the great advantages of the combination of in vivo intracellular

or extracellular recording with neurochemical and morphological charac-
terization of the neurons, several limitations still remain.
r Since the recording and subsequent filling of the recorded neurons are
performed blindly, this is a very time-consuming process. Tedious work
may result in only a handful of analyzed cells. Thus, if the experiment
requires dozens of cells to be analyzed, it is advisable to use alterna-
tive methods such as the in vitro preparation or in vivo juxtacellular
labeling.
r Anesthetics can alter the firing pattern of the neurons; therefore, care-
ful judgment of the chosen drug is a prerequisite for reliable electro-
physiological analysis. To overcome this problem, “juxtacellular” label-
ing was adapted to unanesthetized, drug-free animals by taking advan-
tage of the head-restrained recording technique (Lee et al., 2004, 2005).
r When the sharp electrode penetrates the cell, neurons often discharge
artificially (the so-called “injury discharge”). If the membrane does not
seal perfectly around the intercellular electrode, it can induce a higher
activity rate than that in normal circumstances, resulting in erroneous
identification of the neural activity.
r Precise application of pharmacological reagents into a small area is
difficult or impossible.
Unless one asks specific questions that can be addressed only by intracel-
lular recording (like membrane oscillations, EPSP, or IPSP measurement in
an identified cell, etc.), this powerful but time-consuming approach can be
substituted by juxtacellular labeling. Despite these limitations, we feel con-
fident that the recording of electrophysiological activity of single neurons
and analysis of their interactions with the neuronal network, in conjunction
with the acquisition of accurate structural data on the synaptic architec-
ture, provide sufficient data for realistic experimental modeling of neuronal
function.
A. Anesthetics
Different anesthetics can be used depending on the planned length of

the recording session, whether the animal has to survive, etc. In acute ter-
minal experiments (when the animal is sacrificed following the recording),
urethane is the preferred anesthetic because the anesthetic effect lasts many
hours. If the animal needs to survive longer than 12 h after the experiment,
anesthesia by a cocktail of ketamine and xylazine is recommended.
r Urethane: Prepare urethane stock solution (5 g urethane in 10 ml of
0.9% NaCl) and inject i.p. (1.3–1.4 g/kg for rat);
r Ketamine/xylazine: Inject i.p. 75 mg/kg ketamine and 10 mg/kg xylazine
(rat). If ketamine/xylazine is used in some cases, it has to be readminis-
tered during the experiment. Use ketamine only (20 mg/kg) whenever
the animal shows signs of awakening. It can be injected i.p. or intra-
venously into the tail vein. This later provides more precision of the
dosage but is harder to execute.
188 ATTILA SÍK
r Combination of urethane and ketamine/xylazine: Use 1.25 g/kg urethane

and supplement the drug with 20 mg/kg ketamine and 2 mg/kg xy-
lazine as needed.
B. Surgery and Implantation of Stimulating and Extracellular

Recording Electrodes
r Shave off the hair from the area overlying the part of the brain studied,
i.e., over the skull or the vertebrae.
r Place the animal into a sturdy stereotaxic apparatus. An antivibration
table is highly recommended to make the intracellular recordings more
stable and longer lasting.
r Use an animal thermoregulation device to keep the body temperature
constant. A low body temperature dramatically decreases the activity of
neurons.
r If the recordings will take prolonged periods, protect the animal’s eyes
with commercially available eye drops or with paraffin oil, etc. against
dehydration of the cornea.
r Cut the skin with a scalpel and drill a small hole over the region where
the recording will be performed. The hole should be as small as possible
but should also provide sufficient space for manipulation. The ideal size
is about 1 × 1 mm. If an extracellular recording electrode is placed into
the same region through the same bone window, the size of the hole
must be larger (1.5 × 1.0 mm).
r If a stimulating electrode is to be used, prepare a small bone window
over the desired stimulation region.
r Implant the stimulating electrode and fix it with acrylic cement.
r Maneuver the extracellular recording electrode into position. The po-
sition of the electrode can be checked by driving the appropriate input
pathway with the stimulating electrode. The extracellular recording
electrode should be fixed by acrylic cement; thus, no extra electrode
holder (which would occupy space) is necessary.
r Open the dura mater with the very sharp tip of a pointy surgical blade
(for example, size 11) or by the tip of a small needle (like 27 g × 1/2 ).
Be careful not to cause any bleeding. If a large blood vessel is in the
way, drill another hole or enlarge the original hole. Do not let the brain
surface dry out, e.g., by putting a drop of 0.9% NaCl into the opening.
C. Intracellular Recording
After lowering the intracellular electrode into the area of interest, cells
need to be impaled with the sharp electrode. There are differences in the
membrane structure and intracellular ion content among cells that will
result in differences of penetrability and survival of the cells following the
penetration. As a general procedure, follow the next steps:
r Pull the electrode from 2.0-mm-diameter thick wall glass capillary.
r Fill the electrode with electrolyte (0.5 M potassium acetate) containing
1–3% biocytin.
r Keep the electrodes in a humid chamber (Petri dish with some moist-
ened paper or cotton) to prevent clogging due to drying at the tip of
the electrode.
r Insert the intracellular electrode into the region of interest, and cover
the bone window with a mixture of paraffin and paraffin oil (1:1, kept
warm before the application at ∼60◦ C) to prevent drying of the area
and to reduce pulsation of the brain.
r Use quick, small steps (2 µm/step) to advance the intracellular elec-
trode.
r When the recorded potential starts decreasing, it indicates that the
electrode is pushed to a membrane. Use one of the following methods
to penetrate into the cell: (a) “buzz” [small, short lasting (1–20 ms)
electric current injection setting the electrode tip in motion causing
small vibration of the tip. Many intracellular amplifiers have this button
on their recording unit] and (b) mechanically, by very gently tapping
the electrode holder or the motor. This will move the pipette a few
micrometers, which may be sufficient to penetrate the cell membrane.
r After penetrating into the cell, apply a negative current to counteract
the depolarization caused by ion leakage through the membrane open-
ing through which the pipette tip has entered the cell. Hyperpolarize
the cell till it stops firing and keep it at this state for a couple of minutes
to allow the cell membrane to seal around the distal pipette shaft.
r Perform the electrophysiological recording. At the end of the session,
use positive (or negative if it has beneficial effect) current pulses to
inject the dye into the cell (300–500 ms at 1 Hz using 0.5–2.5 nA cur-
rent). The necessary labeling time differs from cell to cell. A labeling
time of 2–30 min is usually sufficient to obtain complete labeling of
neurons. Cells with large axonal arborization, and/or long projection
may require longer time with higher current.
D. Survival Time
The intracellularly injected dye spreads through the neuron and its pro-
cesses by active transport mechanisms and by diffusion. If the reconstruction
of complete axonal arborization is the goal, this requires a longer trans-
port/diffusion period until the dye has filled all the thin axon collaterals
down to their terminal arborizations. Neurons with axons projecting over
longer distances require lengthier survival time. Typically, the survival time
after the injection of the marker varies between 0 and 12 h. Longer survival
is not recommended if biocytin or neurobiotin is used, because enzymes
that might be activated due to the trauma may destroy these tracers. If a
longer survival time is required, injection with biotinylated dextran amine
is recommended instead of biocytin (BDA, MW 3000).
190 ATTILA SÍK
E. Fixation
Choosing the appropriate fixative is important if the intention is to de-

termine the neurochemical features of the recorded neuron or other cells
(like target or input cells). Most immunohistochemical staining procedures
require fixation with a buffered solution of formaldehyde (depolymerized
paraformaldehyde).
r Reanesthetize the animal if necessary.
r Open the thorax and insert a large diameter needle via the left ventricle
into the ascending aorta.
r Place a clamp on the descending aorta, between the liver and the lungs.
r Open the right atrium to allow blood and perfusates to flow out.
r Flush the blood using 0.9% NaCl or 0.1 M PBS until the outflowing
liquid from the heart is clear, typically 1–3 min.
r Switch the solution to the appropriate fixative (typically 4% formalde-
hyde in 0.1 M phosphate buffer, pH 7.4).
r Fix the brain for 30 min.
r Remove the brain from the skull. If the brain is too soft, postfixation
may be carried out (1 h to overnight).
F. Visualization of the Intracellularly Filled Cells
Depending on the goal, the labeled cells can be visualized with either a
fluorescent dye or a permanent marker [like 3,3 -diaminobenzidine (DAB)
peroxidase reaction product]. We recommend a fluorescence-based visual-
ization if the aim of the experiment is to further characterize the recorded
cells immunohistochemically (see section “Immunolabeling”). If the recon-
struction of the cell is the goal, the sections have to be kept in sequential
order during the whole process.
r Section the brain using a vibrating microtome (section thickness 30–
100 µm).
r Wash the fixative with PB (5×15 min).
r Treat the section with detergent if examination is planned other than
electron microscopy (0.5% Triton X-100 in PB for 30 min).
r Wash with Tris-buffered saline (TBS) (3 × 15 min).
r React the sections with either HRP containing avidin–biotin complex
(ABC; dilution 1:500 in TBS) or with fluorochrome-conjugated strep-
tavidin (dilution 1:100–1:1000 in TBS depending on the type of the
fluorescent dye) (1 h at room temperature for ABC; 5 h overnight at
4 h for fluorescence).
r Wash the reagent (3 × 15 min TBS).
r If fluorescent dye has been used, mount and coverslip with antifading
medium and analyze the finding in a fluorescence microscope using
the proper filtering. If ABC has been used, visualize the cell using the
following steps:
1. Prepare a DAB–Ni solution: measure 100 ml 0.01 M PB.

2. Add 30 mg 3,3 diaminobenzidine (DAB) (1 toxic and carcinogenic,
decontaminate spoils and leftovers with chlorine bleach; 5 light sen-
sitive, cover with aluminum foil).
3. Add 40 mg NH4 Cl.
4. Add 5 ml 0.05 M NiNH4 SO4 (Nickel ammonium sulfate; add drop-
wise under agitation).
5. Filter this DAB–Ni solution.
6. Remove the last wash, add 1 ml of DAB–Ni solution.
7. Allow the sections to be saturated with DAB–Ni (20 min room tem-
perature, dark).
8. Add 10 µl of H2 O2 solution (made up fresh by pipetting 10 µl of 30%
H2 O2 solution into 10 ml double distilled water) to start the reaction.
A black reaction product will form in all structures that contain ABC
(the reactive compound is the peroxidase). This reaction can take 5–
40 min to fully develop. Inspect the section regularly in a microscope
to monitor the progress of the reaction.
r When sufficient reaction product has formed, rinse the sections three
times in TBS.
r Mount on glass slides and dry. Do not coverslip if the procedure is
followed up with immunolabeling (see below).
G. Immunolabeling
In order to determine the neurochemical content of the recorded cell,

an immunohistochemical reaction needs to be performed. Since the DAB
or DAB–Ni reaction product masks the immunosignal of the cell, an im-
munofluorescence technique should be used in this case (Kawaguchi, 1993).
Once the cell has been visualized by a streptavidin-conjugated fluorochrome
(see above), a regular immunoreaction using fluorescence secondary anti-
body can be applied. The exact steps can be found in other books like
Immunohistochemistry (Cuello, 1993). Briefly:
r Wash out the fixative (see above).
r Wash the sections by rinsing 3 × 10 min with 0.05 M TBS.
r Block aspecific immunosignal by incubating 45 min in blocking solu-
tion (TBS containing 5% normal goat serum (NGS) or other blocking
serum; and 0.5% Triton X-100).
r Treat the sections with fluorochrome-conjugated streptavidin (dilution
1:100–1:1000 in TBS depending on the type of the fluorescent dye)
(5h overnight at 4◦ C) in the dark.
r Wash out the reagent (3 × 15 min TBS).
192 ATTILA SÍK
r Use primary antibody diluted into TBS containing 0.5% NGS (or other
serum), 0.01% sodium azide added. Incubate the sections in the incu-
bation medium in the dark (overnight 2 days at 4◦ C) to allow antibodies
to penetrate into the sections. Longer incubation results in better pene-
tration of the antibody. During incubation, place the vials or well plates
that contain the sections on a rocking plateau to ensure gentle agitation.
r Wash out the primary antibody (3 × 15 min TBS).
r Incubate the sections (in the dark) in the solution containing the
fluorochrome-conjugated secondary antibody (dilution 1:100–1:1000
in TBS for 6 h to overnight). The excitation–emission spectrum of this
fluorochrome should be of course different from that used to visualize
the labeled cell.
r Wash 3 × 15 min with TBS.
r Mount on slides, allow sections to air-dry.
r Add antifading reagent like Mowiol (has to be at room temperature)
and coverslip.
r Seal with nail polish.
r Analyze the neurochemical content of the cell using a fluorescence
microscope equipped with the proper excitation–emission filters.
r Store at 4◦ C temporarily, or at −20◦ C for the long term.
r Then the permanent visualization of the filled cells using the ABC-DAB
protocol is performed as described above.
H. Reconstruction
The entire dendritic and axonal arborization of the recorded neurons can
be reconstructed after successful intracellular labeling. If sections are air-
dried after mounting on slides, shrinkage in Z direction is substantial (about
80–90%). In the case when the real three-dimensional (3D) structure of the
neuron is important, embedding of the section in plastic resin is necessary.
The following protocol is an example using Durcupan, but other resins can
also be used, such as Araldite, Spurr, and so forth.
r Dehydrate the sections in an ascending alcohol series (50%, 70%, 90%,
2 × 100% 10 min each).
r Change to intermediate solution (100% propylene oxide if Durcupan
is used) (2 × 10 min). Note: Propylene oxide is volatile, toxic, and com-
bustible. Use a fume hood and wear gloves.
r Place the section in pure Durcupan (overnight).
r Mount the sections, coverslip, and cure in an oven at 58◦ C for 24 h.
The cells can be reconstructed using a drawing tube or a microscope-
computer equipped with Neurolucida software. The advantage of using
the latter equipment is that the 3D information is preserved in each in-
dividual section. When a drawing tube is used, the reconstruction will be
manufactured using the 2D projection of each section. Thus, with this pro-
cedure the 3D information in individual sections is lost (see further details
in the chapters by Ascoli and Scorcioni and Duque and Zaborszky in this
volume).
I. Equipment and Supplies (Some Recommended Equipment Is in

Parenthesis, But Other Items Can Be Used)
1. Equipment for Surgery and Recording
Drill (NSK Emax)

Vibration isolation system (Newport VH Isostation)
Stereotaxic apparatus (Kopf Model 920)
Thermoregulator with heating pad (CWE TC 1000)
Operating microscope (Olympus SZ series)
Light source (WPI)
Inchworm motor system (Burleigh)
2. Equipment for Data Acquisition
Digital oscilloscope (Tektronix TDS 2014)

Noise reduction device (Hum bug) or traditional Faraday cage
Amplifier (Axon Multiclamp 700A computer-controlled microelectrode
amplifier with Softpanel or Axoclamp-2B)
Analog-digital converter (Axon Digidata 1322A data acquisition system)
Data acquisition software (Axon pClamp 9.0 electrophysiology software)
Isolated pulse stimulator (A-M systems Model 2100)
Differential amplifier (A-M systems Model 3000) for recording EEG
3. Other Equipment
Micropipette puller (Sutter Instruments)
VI. CHEMICALS
Biocytin or Neurobiotin (Vector Laboratories)

Potassium acetate (Sigma)
Paraformaldehyde, glutaraldehyde, Durcupan (Electron Microscopy
Sciences)
V. SOLUTIONS
r Phosphate buffer (PB) 0.2 M pH 7.4
Stock solution A: 0.2 M NaH2 PO4
Stock solution B: 0.2 M Na2 HPO4
Add solution A to solution B in a 1:4 ratio until the pH reaches 7.4
to give 0.2 M PB.
194 ATTILA SÍK
r Tris-buffered saline (TBS) 0.05 M pH 7.4

Trizma base 0.05 M
Trizma acid 0.05 M
0.9% NaCl
Acknowledgments. This work was supported by a Canadian Institutes of

Health Research (CIHR) grant and FRSQ. The author thanks Drs. Martin
Deschênes and György Buzsáki for helpful discussion. Most of the work was
performed in Dr. Buzsáki’s laboratory at Rutgers University, NJ, USA.
REFERENCES
Bartho, P., Hirase, H., Monconduit, L., Zugaro, M., Harris, K. D., and Buzsaki, G., 2004, Char-
acterization of neocortical principal cells and interneurons by network interactions and
extracellular features, J. Neurophysiol. 92:600–608.
Bragin, A., Jando, G., Nadasdy, Z., Hetke, J., Wise, K., and Buzsaki, G., 1995, Gamma (40–
100 Hz) oscillation in the hippocampus of the behaving rat, J. Neurosci. 15:47–60.
Buzsaki, G., 2004, Large-scale recording of neuronal ensembles, Nat. Neurosci. 7:446–451.
Buzsaki, G., Hazan, L., Zugaro, M. B., and Csicsvari, J., 2004, Neuroscope: a viewer for neuro-
physiological data, Soc. Neurosci. (abstract):768.2.
Buzsaki, G., and Kandel, A., 1998, Somadendritic backpropagation of action potentials in
cortical pyramidal cells of the awake rat, J. Neurophysiol. 79:1587–1591.
Buzsáki, G., Traub, R., and Pedley, T., 2003, The cellular synaptic generation of EEG.
In: Ebersole, J.S. and Pedley, T.A. (eds.), Current Practice of Clinical Encephelography,
Philadelphia: Lippincott-Williams and Wilkins, (3rd ed.), pp. 1–11.
Chang, H. T., Wilson, C. J., and Kitai, S. T., 1981, Single neostriatal efferent axons in the globus
pallidus: a light and electron microscopic study, Science 213:915–918.
Csicsvari, J., Henze, D. A., Jamieson, B., Harris, K. D., Sirota, A., Bartho, P., Wise, K. D., and
Buzsaki, G., 2003, Massively parallel recording of unit and local field potentials with silicon-
based electrodes, J. Neurophysiol. 90:1314–1323.
Cuello, A. C., 1993, Immunohistochemistry II, IBRO handbook series: Methods in the neuro-
sciences. John Wiley & Sons, England.
Ferster, D., and Jagadeesh, B., 1992, EPSP–IPSP interactions in cat visual cortex studied with
in vivo whole-cell patch recording, J. Neurosci. 12:1262–1274.
Freund, T. F., and Buzsaki, G., 1996, Interneurons of the hippocampus, Hippocampus 6:347–470.
Gray, C. M., Maldonado, P. E., Wilson, M., and McNaughton, B. L., 1995, Tetrodes markedly
improve the reliability and yield of multiple single-unit recordings in cat striate cortex, J.
Grossman, R. G., and Hampton, T., 1968, Depolarization of cortical glial cells during electro-
cortical activity, Brain Res. 11:316–324.
Halasy, K., Buhl, E. H., Lorinczi, Z., Tamas, G., and Somogyi, P., 1996, Synaptic target selectivity
and input of GABAergic basket and bistratified interneurons in the CA1 area of the rat
hippocampus, Hippocampus 6:306–329.
Harris, K. D., Henze, D. A., Csicsvari, J., Hirase, H., and Buzsaki, G., 2000, Accuracy of tetrode
separation as determined by simultaneous intracellular and extracellular measurements,
J. Neurophysiol. 84:401–414.
Hassin, G., 1979, Pikeperch horizontal cells identified by intracellular staining, J. Comp. Neurol.
186:529–540.
Hazan, L., Zugaro, M. B., Csicsvari, J., Creese, I., and Buzsáki, G., 2004, Klusters: a graphical
application for spike sorting of extracellular units, Soc. Neurosci. (abstract):768.3.
Heimer, L., and Robards, M. J., 1981, Neuroanatomical Tract Tracing Methods, New York: Plenum
Press.
Henze, D. A., Borhegyi, Z., Csicsvari, J., Mamiya, A., Harris, K. D., and Buzsaki, G., 2000,
Intracellular features predicted by extracellular recordings in the hippocampus in vivo, J.
Neurophysiol. 84:390–400.
Horikawa, K., and Armstrong, W. E., 1988, A versatile means of intracellular labeling: injection
of biocytin and its detection with avidin conjugates, J. Neurosci. Methods 25:1–11.
Hubel, D. H., 1957, Tungsten microelectrodes for recording single units, Science 125:
549–550.
Jankowska, E., Rastad, J., and Westman, J., 1976, Intracellular application of horseradish per-
oxidase and its light and electron microscopical appearance in spinocervical tract cells,
Brain Res. 105:557–562.
Kamondi, A., Acsady, L., and Buzsaki, G., 1998, Dendritic spikes are enhanced by cooperative
network activity in the intact hippocampus, J. Neurosci. 18:3919–3928.
Kaneko, A., 1970, Physiological and morphological identification of horizontal, bipolar and
amacrine cells in goldfish retina, J. Physiol. 207:623–633.
Kawaguchi, Y., 1993, Physiological, morphological, and histochemical characterization of three
classes of interneurons in rat neostriatum, J. Neurosci. 13:4908–4923.
Kawaguchi, Y., Wilson, C. J., and Emson, P. C., 1989, Intracellular recording of identified
neostriatal patch and matrix spiny cells in a slice preparation preserving cortical inputs, J.
Kawaguchi, Y., Wilson, C. J., and Emson, P. C., 1990, Projection subtypes of rat neostriatal matrix
cells revealed by intracellular injection of biocytin, J. Neurosci. 10:3421–3438.
Kelly, J. P., and Van Essen, D. C., 1974, Cell structure and function in the visual cortex of the
cat, J. Physiol. 238:515–547.
Kita, H., and Armstrong, W., 1991, A biotin-containing compound N-(2-
aminoethyl)biotinamide for intracellular labeling and neuronal tracing studies:
comparison with biocytin, J. Neurosci. Methods 37:141–150.
Kita, H., and Kitai, S. T., 1986, Electrophysiology of rat thalamo-cortical relay neurons: an in
vivo intracellular recording and labeling study, Brain Res. 371:80–89.
Kitai, S. T., Kocsis, J. D., Preston, R. J., and Sugimori, M., 1976, Monosynaptic inputs to caudate
neurons identified by intracellular injection of horseradish peroxidase, Brain Res. 109:601–
606.
Lee, M. G., Hassani, O. K., Alonso, A., and Jones, B. E., 2005, Cholinergic basal forebrain
neurons burst with theta during waking and paradoxical sleep, J. Neurosci. 25:4365–4369.
Lee, M. G., Manns, I. D., Alonso, A., and Jones, B. E., 2004, Sleep-wake related discharge
properties of basal forebrain neurons recorded with micropipettes in head-fixed rats, J.
Li, X. -G., Somogyi, P., Tepper, J. M., and Buzsaki, G., 1992, Axonal and dendritic arborization
of an intracellularly labeled chandelier cell in the CA1 region of rat hippocampus, Exp.
Brain Res. 90:519–525.
Li, X. G., Somogyi, P., Ylinen, A., and Buzsaki, G., 1993, The hippocampal CA3 network: an in
vivo intracellular labeling study, J. Comp. Neurol. 339:181–208.
Margrie, T. W., Brecht, M., and Sakmann, B., 2002, In vivo, low-resistance, whole-cell recordings
from neurons in the anaesthetized and awake mammalian brain, Pflugers Arch. 444:491–
498.
McNaughton, B. L., O’Keefe, J., and Barnes, C. A., 1983, The stereotrode: a new technique for
simultaneous isolation of several single units in the central nervous system from multiple
unit records, J. Neurosci. Methods 8:391–397.
Nadasdy, Z., Csicsvari, J., Penttonen, M., and Buzsaki, G., 1998, Extracellular recording and
analysis of electrical activity: from single cells to ensembles. In: Eichenbaum, H., and Davis,
J. L. (eds.), Neuronal Ensembles: Strategies for Recording and Decoding, New York: Wiley-Liss,
pp. 17–55.
Sandell, J. H., and Masland, R. H., 1988, Photoconversion of some fluorescent markers to a
diaminobenzidine product, J. Histochem. Cytochem. 36:555–559.
196 ATTILA SÍK
Sik, A., Penttonen, M., and Buzsaki, G., 1997, Interneurons in the hippocampal dentate gyrus:
an in vivo intracellular study, Eur. J. Neurosci. 9:573–588.
Sik, A., Penttonen, M., Ylinen, A., and Buzsaki, G., 1995, Hippocampal CA1 interneurons: an
in vivo intracellular labeling study, J. Neurosci. 15:6651–6665.
Sik, A., Tamamaki, N., and Freund, T. F., 1993, Complete axon arborization of a single CA3
pyramidal cell in the rat hippocampus, and its relationship with postsynaptic parvalbumin-
containing interneurons, Eur. J. Neurosci. 5:1719–1728.
Sik, A., Ylinen, A., Penttonen, M., and Buzsaki, G., 1994, Inhibitory CA1-CA3-hilar region
feedback in the hippocampus, Science 265:1722–1724.
Simons, D. J., 1978, Response properties of vibrissa units in rat SI somatosensory neocortex, J.
Snow, P. J., Rose, P. K., and Brown, A. G., 1976, Tracing axons and axon collaterals of spinal
neurons using intracellular injection of horseradish peroxidase, Science 191:312–313.
Stewart, W. W., 1978, Functional connections between cells as revealed by dye-coupling with a
highly fluorescent naphthalimide tracer, Cell 14:741–759.
Stewart, W. W., 1981, Lucifer dyes—highly fluorescent dyes for biological tracing, Nature
292:17–21.
Takato, M., and Goldring, S., 1979, Intracellular marking with lucifer yellow CH and horseradish
peroxidase of cells electrophysiologically characterized as glia in the cerebral cortex of the
cat, J. Comp. Neurol. 186:173–188.
Tamamaki, N., Abe, K., and Nojyo, Y., 1987, Columnar organization in the subiculum formed
by axon branches originating from single CA1 pyramidal neurons in the rat hippocampus,
Brain Res. 412:156–160.
Tamamaki, N., and Nojyo, Y., 1993, Projection of the entorhinal layer-II neurons in the rat as
revealed by intracellular pressure-injection of neurobiotin, Hippocampus 3:471–480.
Tepper, J. M., Sawyer, S. F., and Groves, P. M., 1987, Electrophysiologically identified nigral
dopaminergic neurons intracellularly labeled with HRP: light-microscopic analysis, J. Neu-
rosci. 7:2794–2806.
Thomas, R. C., and Wilson, V. J., 1965, Precise localization of Renshaw cells with a new marking
technique, Nature 206:211–213.
Wang, X. J., and Buzsaki, G., 1996, Gamma oscillation by synaptic inhibition in a hippocampal
interneuronal network model, J. Neurosci. 16:6402–6413.
Wilson, M. A., and McNaughton, B. L., 1993, Dynamics of the hippocampal ensemble code for
space, Science 261:1055–1058.
Wise, K. D., and Najafi, K., 1991, Microfabrication techniques for integrated sensors and mi-
crosystems, Science 254:1335–1342.
Ylinen, A., Soltesz, I., Bragin, A., Penttonen, M., Sik, A., and Buzsaki, G., 1995, Intracellular
correlates of hippocampal theta rhythm in identified pyramidal cells, granule cells, and
basket cells, Hippocampus 5:78–90.
7
Juxtacellular Labeling of
Individual Neurons In Vivo:
From Electrophysiology
to Synaptology
ALVARO DUQUE and LASZLO ZABORSZKY
INTRODUCTION
GENERAL METHODOLOGY
Anesthesia
Choice of Juxtacellular Labeling Markers and Recording Solutions
Electrodes and Recording Apparatus
Juxtacellular Labeling
Control Experiments: The Neuron Recorded is the One Labeled
Histology and 3D Light and Electron Microscopy Reconstructions
Tracer Techniques in Combination with Electrophysiological
Recording and Juxtacellular Labeling
APPLICATIONS
Electrophysiological and Morphological Identification of Single
Neurons
Retrograde Labeling of Electrophysiologically Identified Neurons
Chemical Identification and Morphometry of Juxtacellularly
Labeled Neurons
Synaptology of Electrophysiologically and Chemically Identified
Neurons
Advantages
Limitations
APPENDIX
Animal Preparation Prior to Electrophysiological Recordings
ALVARO DUQUE • Department of Neurobiology, Yale University School of Medicine,

New Haven, CT 06510 LASZLO ZABORSZKY • Center for Molecular and Behavioral
Neuroscience, Rutgers, The State University of New Jersey, Newark, NJ 07102
197
198 ALVARO DUQUE and LASZLO ZABORSZKY
Animal Preparation for Electrophysiology

Electrophysiology and Labeling
Perfusion
Cutting and Pretreatment of Sections
Visualization of Biocytin-Filled Neuron and Digital Photography
Neurochemical Identification of Biocytin-Filled Neuron
Conversion of the Fluorescent Signal to DAB
Staining for a Second Antigen
Embedding for Electron Microscopy
3D Light Microscopy Reconstructions
Electron Microscopy and 3D Reconstruction from Ultrathin
Sections
REFERENCES
Abstract: This chapter summarizes details pertaining to the extracellular recording

and juxtacellular labeling method and its application to the characterization of basal
forebrain neurons. Juxtacellular labeling is compared to other single-cell labeling
techniques. Compatibility of the method with fluorescent retrograde/anterograde
neural tracing, immunohistochemical, and electron microscopy techniques is also
illustrated.
Keywords: basal forebrain, 3D reconstruction, electron microscopy, morphology,
neurochemical identification
I. INTRODUCTION
Brain regions consist of elementary local circuits and cell assemblies spe-
cialized to carry out discrete computations that eventually give rise to behav-
ior. These circuits are built from regionally specific combinations of limited
types of individual neurons (see Nadasdy et al., this volume). However, de-
pending on the computational needs of the brain regions and the particu-
lar electrophysiological, neurochemical, and hodological characteristics of
different neuronal populations, these circuits become increasingly complex
across the neuraxis. Except for a few heuristic attempts by Cajal (1911) in the
early part of the twentieth century, it was only in the 1970s when the imagina-
tive cerebellar and cortical circuitry models of Szentágothai were published
(Szentágothai, 1970, 1978). These general models were based on correlating
(mostly indirectly) the 3D spatial architecture, derived from Golgi studies,
with the synaptic pattern of putative circuits obtained from electron micro-
scopical analysis (see also Arbib et al., 1998, and partial list of Szentágothai’s
publications in Zaborszky et al., 1992). To correctly interpret the complex en-
tanglement of axons and dendrites under the electron microscope, however,
it was necessary to combine the full cell visualization of the Golgi method
(Golgi, 1883) with the power of electron microscopy to resolve individual
synapses. After the initial attempt of Blackstad (1965), the development of
gold toning of Golgi impregnated neurons (Fairen et al., 1977) paved the way
to identify the synaptic connections of individual neurons (Somogyi, 1977).
JUXTACELLULAR LABELING 199
A further development in the analysis of neural circuits was the com-
bination of intracellular electrophysiological recordings with horseradish
peroxidase (HRP) labeling of the recorded cell ( Jankowska et al., 1976; Ki-
tai et al., 1976a, b; Light and Durkovic, 1976; Snow et al., 1976; Cullheim and
Kellerth, 1978). HRP, introduced as a neuroanatomical tracer in the early
1970s (Kristensson and Olsson; 1971; LaVail and LaVail, 1972), is an enzyme
that catalyzes the reaction by which diaminobenzidine (DAB) precipitates
and forms an electron-dense product (Graham and Karnovsky, 1966). The
chapters of Somogyi and Freund in the previous edition of this series are ex-
cellent reviews on the study of the synaptic relationships of interconnected
and chemically identified neurons, using a combination of electrophysiol-
ogy, HRP filling, Golgi method (Golgi, 1883), and immunocytochemistry
(Freund and Somogyi, 1989; Somogyi and Freund, 1989).
Intracellular recordings can also be combined with fluorescent dye la-
beling and histochemical or immunocytochemical methods to identify the
transmitter of the recorded neurons (Aghajanian and Vandermaelen, 1982;
Grace and Bunney, 1983a, b). However, morphological reconstructions of
these neurons and study of their synaptology require conversion of the flu-
orescent signal to a permanent DAB end product (Buhl, 1993), a fact that
makes this avenue much too cumbersome to be routinely applied.
In vivo technical advances, including the use of voltage-sensitive or
calcium-sensitive dyes, for the study of individual synapses in the living
animal as described in the chapter of Goldberg et al. (this volume) have
the disadvantage of being applicable only to superficial layers of the cortex.
Hence, they are not useful for investigating deep subcortical structures such
as the basal forebrain or the basal ganglia. Intracellular or patch recordings
are difficult to apply in vivo when the neuron to be recorded and labeled is
in a deep subcortical structure or in an area densely packed with neuropil
and fibers. The chances of breaking the fine pipette tips used for intracel-
lular recordings increase dramatically as a function of the distance traveled
by the micropipette electrode. Moreover, in vivo intracellular approaches
are limited by the stability of the preparation. In particular, respiration and
heartbeat produce movement, which is not reduced or only minimally re-
duced by the use of vibration-free tables. Movement in the preparation
compromises the integrity of the micropipette tip, disrupts the recording,
and usually quickly kills the cell under investigation. Therefore, most intra-
cellular or patch recordings involving single-cell labeling in the basal fore-
brain or basal ganglia have taken place in vitro (Alonso et al., 1996; Nambu
and Llinas, 1997; Koos and Tepper, 1999, 2002). In vitro preparations are
very stable and have, therefore, revealed much about intrinsic electrophys-
iological properties of these neurons, but the nature of the preparation
usually does not allow full morphological reconstructions or chemical iden-
tification. Cutting afferents and efferents, however, and maintaining the
neuron alive under typical in vitro conditions raises questions as to the
validity and relevance of some of the results obtained. Moreover, behav-
ioral correlates such as electroencephalograph (EEG) cannot be obtained in
vitro.
In vivo extracellular recordings are usually easier to perform than in-

tracellular recordings. However, for many years a major obstacle in their
usefulness was the inability to label the recorded cell. A first approach to
resolve this problem was to eject a dye, such as HRP, at the end of the
recording session and label a small group of cells found in close proximity
to the pipette tip. This effort, however, fell short of identifying “the recorded
cell” although it did identify a small area where the recording took place.
By decreasing the tip size (to 1–4 µm) of glass microelectrodes filled with
0.5–1.0% HRP in 2 M NaCl and ejecting HRP with 400–1000 nA negative
current for 5–25 s after regular extracellular recordings, 30 years ago Lynch
and colleagues (Lynch et. al., 1974a, b) managed to label only 1–4 cells per
attempt. However, in cases in which only one neuron was labeled, no evi-
dence was provided that the stained neuron was actually the one recorded
from. Twenty years later, working in the thalamic reticular nucleus, Pinault
(1994) described a method for recording and labeling single neurons ex-
tracellularly and for the first time provided evidence that the labeled cell
was the one recorded from. The usefulness of this methodology, called jux-
tacellular labeling, was finally established by Pinault in a 1996 study, which
presented further evidence from different rat brain areas, including the
neocortex, thalamus, basal ganglia, basal forebrain, and cerebellum, that
the juxtacellularly labeled neuron was the one previously recorded from.
In this chapter we describe juxtacellular labeling of single neurons with
details based on our own experience, using the method mostly in the basal
forebrain of the rat. We will present our own control experiments providing
further supporting evidence for the validity of the technique. We will elabo-
rate on the usefulness of the technique by illustrating the compatibility of the
labeling procedure with various tracer and immunocytochemical methods
both at the light and at the electron microscopical levels to study the synap-
tology of chemically and electrophysiologically identified neurons, whose
activity is also correlated with EEG recordings. The advantages, limitations,
and drawbacks of the technique are also discussed.
II. GENERAL METHODOLOGY
Lesions or injections of anterograde and retrograde tracers may be ap-

plied prior to recording and labeling of single neurons (see section “Tracer
Techniques in Combination with Electrophysiological Recording and Juxta-
cellular Labeling”). These extra steps can substantially increase the amount
of information collected from a single neuron. These and all other proce-
dures involving animals, and details pertaining to animal treatment should
be in strict accordance with National Institutes of Health guidelines, which
are readily available in publications such as the “Guide for the Care and Use
of Laboratory Animals.” Before starting any experiments, protocols need to
be reviewed and approved by the respective Institutional Animal Care and
Use Committee (IACUC). Figure 7.1 illustrates the general methodology.
A B
EEG
UNIT
CPu
LGP
ic
SI
HDB
AA
Figure 7.1. Schematic diagram of general methodology. (A) shows a digital pho-
tograph of the head of a rat fixed in a stereotaxic apparatus; four open holes
on the cranium are visible. (B) shows a diagram of the rat skull with the superfi-
cial venous system superimposed on it. Special care is always taken to avoid rup-
ture of the veins to minimize bleeding, and mediolateral measurements are taken
from the midline of the superior sagittal sinus instead of from the midline of the
bone fissure. (C) illustrate different procedures, some of which like the injection
of tracers or other chemicals and the infliction of cuts or other lesions can be per-
formed hours or days before recording and juxtacellular labeling of single neurons.
(D) Single-cell extracellular recordings can be obtained concomitant with EEG
recordings from one or several cortical and subcortical areas. Abbreviations: CPu,
caudate putamen; LGP, lateral globus pallidus; SI, substantia innominata; HDB, hor-
izontal limb of the diagonal band of Broca; AA, anterior amygdaloid area; ic, internal
capsule.
A. Anesthesia
Several choices of anesthetics, including ketamine–xylazine, urethane,

pentobarbital, are available. However, transport of the tracer during juxta-
cellular filling may depend on the activity of the cell being labeled. There-
fore, for recording and labeling experiments barbiturate anesthetics may
not be a good choice, since they are known to generally suppress neuronal
activity (Richards, 1972). On the other hand, ketamine–xylazine, which is
preferred for survival surgeries, may result in significant fluctuations in anes-
thetic plane in long recording experiments, in which variations in absorp-
tion may also result in relatively unpredictable redosing schedules. Ure-
thane (ethyl carbamate) is our preferred choice for recording experiments
because it appears to have little effect on neural activity, and its primary in-
hibitory actions seem to be restricted to some small neurons in the reticular
formation (Rogers et al., 1980). These examples make it clear that choices of
anesthetic and other details depend on the particulars of the investigation.
Discussion of the effects of different anesthetics on the firing properties of
neurons is beyond the scope of this chapter. For additional details regarding
anesthesia and surgical procedures, see theAppendix.
B. Choice of Juxtacellular Labeling Markers and Recording Solutions
The most logical choice of solution for juxtacellular labeling is one

whose ionic composition is compatible with general extracellular record-
ings with the addition of biocytin [Nε-(+)-biotinyl-l-lysine, FW 372.5 g/mol;
Sigma-Aldrich Co., St. Louis, MO] or NeurobiotinTM (N-(2-aminoethyl)
biotinamide hydrochloride, FW 322.85 g/mol; Vector Laboratories Inc.,
Burlingame, CA). The usual recording solution concentration ranges from
0.5 to about 1 M NaCl and contains 0.5–5.0% biocytin or Neurobiotin.
Difficulty to dissolve biocytin, especially when used in higher concentra-
tions, is sometimes reported. This may be easily resolved by warming up the
solution, strong agitation, sonication, or any combination of these. Dissolv-
ing biocytin in water and then mixing it with the NaCl solution may help.
In addition, we have performed recordings and juxtacellular labeling us-
ing typical K-based solutions at concentrations usually used in intracellular
recordings. In these cases, we did not notice any particular advantage or
disadvantage for the labeling procedure. In general, all these solutions can
be stored for several weeks at 4◦ C or for several months at less than 0◦ C.
The use of compounds other than HRP, biocytin, or Neurobiotin for jux-
tacellular labeling is theoretically possible, but we are not aware of any par-
ticular studies trying other substances. We have tried using biotin dextran
amine [BDA(s), Molecular Probes, Eugene, OR] compounds, but the results
obtained so far are inconclusive. Therefore, the tracers of choice for juxta-
cellular labeling are biocytin and Neurobiotin. In our hands, both provide
very similar results, a finding that is in agreement with studies comparing
biocytin and Neurobiotin for intracellular labeling (Kita and Armstrong,
1991). According to Vector Labs, however, compared to biocytin and other
neuronal labels, Neurobiotin is more soluble, iontophoreses better, and re-
mains longer in cells. Neurobiotin also seems to be easier to dissolve than
biocytin when used at higher concentrations.
Whether or not the solution needs to be filtered depends on how well
the tracer is dissolved. If the solution appears completely crystal clear to
the naked eye, it may not need to be filtered. Although filtering does not
seem to have any adverse consequences. Because of the very small amounts
of solution usually prepared (in the order of a few hundred microliters), it
is advantageous to use very small filters, so that one loses the least amount
of solution when filtering. Although any filter will do, the Cameo, 3 mm–
0.22 µm acetate syringe filters, or the like are very effective.
C. Electrodes and Recording Apparatus
1. Single Unit Extracellular and Labeling Electrode/Glass
Microelectrodes are pulled from glass capillaries. Usually, 1.0–2.0 mm cap-

illaries containing a microfilament fused to the inner wall are convenient
choices because of their strength and commercial availability. The microfil-
ament is very useful because it facilitates filling of the electrode and makes
the filling more uniform, which minimizes trapped air bubbles. If there are
air bubbles trapped in the solution (they are easily seen under the micro-
scope), it is best to remove them. This can be done by gently tapping on
the electrode while holding it vertically. In the worse case, bubbles can be
removed by introducing a very thin wire into the solution in the electrode
so as to make the bubble(s) attach to it; once attached to the wire they can
be pulled out. This is done under the microscope with the electrode placed
horizontally, usually glued to a glass slide with a small piece of putty.
A tungsten wire with a small enough diameter can be prepared for this
purpose by thinning it in a caustic solution. First, attach the positive pole,
for instance that of a 9-V battery, to the wire and the negative pole to a
carbon rod. Fill a small beaker with caustic solution, introduce the carbon
rod into it, and then slowly introduce the tungsten wire into the solution.
The current passed will “eat up” the wire, thinning it. After rinsing it in
water, this wire can be used for removing bubbles.
One easy way to fill an electrode with solution is by first placing a drop
of solution on its back and waiting for a couple of minutes for the tip to
be filled by capillary action, and then the rest of the electrode can be filled
(from the back) by using, for instance, a micropipette filling needle. Some
convenient choices for micropipette filling needles are the MicroFilTM ones
sold by WPI (World Precision Instruments, Sarasota, FL).
Of several pullers available, the Narishige PE-2 (Narishige, Tokyo, Japan)
vertical puller has been a common choice, perhaps because it is an older
model, which has been around for many years and is available in many labo-
ratories. However, any puller able to handle the right size of glass capillaries
will do. The tip of the electrode is broken under a microscope to a diameter
in the range of 0.2–2.0 µm. The smaller the tip, the higher the impedance of
the electrode. High impedance results in the detection of fewer neurons, but
has the advantage that they are more likely detected only at a closer range.
Therefore, higher impedance electrodes are more selective as the possibility
of detecting more than one cell is diminished. Higher impedance electrodes
should be used for detecting smaller cells whose electrical potentials are also
smaller and need to be detected at closer ranges. In our experience, pipettes
filled with saline and 2% biocytin, with tips of about 1–1.2 µm, have very
convenient impedances of approximately 20–40 M measured in the brain.
2. Extracellular Electrode/MUA/EEG/LFP/Metal
Electroencephalographic (EEG) or multiunit activity (MUA) and local

field potential (LFP) data can be collected at the time a single cell is
recorded. The advantage is that one can then correlate the extracellular
activity of a single cell with the activity of a network. However, one should
keep in mind that the characteristics of the EEG or LFP activity are highly
dependent on the type and depth of anesthesia. Furthermore, the signals
are also different in different brain regions and are significantly related to
the cortical layer where the electrode tip is placed. Hence, the interpretation
of the strength of the relationship or correlation of the single-cell activity
and the network activity should take these and other factors (such as the
distance between the two electrodes) into account. EEG and LFP can be
collected with a host of metal electrodes (i.e., stainless steel, tungsten, etc.),
most of which are easy to make and are also readily commercially available
in bipolar and monopolar choices. Good ready-to-use metal electrodes can
be purchased from Frederick Haer & Co. (FHC, Bowdoinham, ME). If one
prefers to make them, appropriate wire can be purchased from California
Fine Wire (CFW, Grover Beach, CA). To make electrodes, the wire is cut
to the desired length (usually 5–8 cm), taking care to keep the electrode
straight. One or two millimeters of insulation is removed from one tip and a
little bit more from the other tip, where an appropriate connector needs to
be clipped or soldered. A full discussion of EEG and electrode construction
is beyond the scope of this chapter and the reader should consult one of
the many specialized papers on the subject.
3. Electrophysiological Recordings and Recording Apparatus
Both EEG and single unit signals can be simultaneously collected, ampli-
fied, filtered, and recorded using standard equipment. In our preparations,
we have used a Neurodata IR-183 amplifier (Neurodata Instruments, New
York, NY) to record and juxtacellularly label single cells. For convenience,
both single unit and EEG tracers are usually displayed simultaneously in an
oscilloscope or computer monitor and recorded directly to a computer hard
drive via an interface. Recordings on magnetic tape for secondary storage
and/or offline analysis are also common. Software and hardware combina-
tions, such as Spike 2.0 from Cambridge Electronic Design Limited (CED,
Science Park, Cambridge, England), are excellent choices for the acquisi-
tion of data via multiple channels. As computer power increases and cost
decreases, direct storage of data into hard drives and quick online analysis
with versatility of options become common practice.
D. Juxtacellular Labeling
After the collection of sufficient extracellular electrophysiological data to

permit characterization of the recorded cell, one can proceed with juxtacel-
lular labeling.
1. Getting Close to the Cell
Before labeling is attempted, the electrode needs to be in very close ap-

position to the cell being recorded; this is why the technique is called “jux-
tacellular” labeling. As the investigator advances the electrode closer to the
cell, the amplitude of the unit’s signal increases. The electrode should be
advanced slowly in small steps of 1 or 2 µm. A sudden and substantial in-
crease in baseline noise, sometimes accompanied by a transient jump in
DC level, indicates a good position where to start experimenting with the
entrainment of the cell.
2. Application of Pulses and Entrainment of the Cell
At the point when a juxtacellular position has been reached, anodal

pulses in the range of 1–10 nA are usually used to eject the dye from the
micropipette. These current pulses cause a vigorous response from the cell
called “entrainment.” An entrained cell fires action potentials at a higher
frequency than normally observed both in response to and during the depo-
larizing phase of the current pulses being applied. Sometimes the neuron
will also fire one or several spikes during the negative phase of the pulse. If
the firing rate increases substantially during both phases of the pulse, the
neuron is usually at risk of being killed. At that point the electrode should be
moved back a few micrometers and/or the intensity of the pulses should be
decreased. The experimenter can modulate entrainment to be on and off
so as to avoid cellular damage. Uncontrolled firing at very high frequencies
and during both phases of the pulse usually precedes irreparable cellular
damage especially in slower firing neurons. The labeling procedure then
requires that the experimenter pay attention to any electrophysiological
change the neuron may undergo before and during labeling. Pulse inten-
sity should be low at the start and slowly increased until the cell responds. If
the cell does not respond, the electrode is moved closer and the procedure
is repeated. Once the cell responds to the pulse, the position of the elec-
trode and/or the intensity of the pulses may need to be adjusted to avoid
cellular damage. If repeated attempts to engage a cellular response fail with
anodal pulses, cathodal pulses can be attempted. Figure 7.2 illustrates the
a*
b*
10.0
C nA
c*
d*
1.0
E mV
500 ms
F 2.5 ms
2.5 ms
a b c d
overlap
Figure 7.2. Extracellular recording and juxtacellular labeling protocol. Trace (A)
shows the spontaneous firing of a neuron in the rat basal forebrain as recorded
extracellularly. Trace (B) shows the entrainment of the neuron in response to the
current pulses shown in trace (C). Notice that firing increases during the positive
phase of the pulse but that the cell still fires spontaneously during some of the
negative phases of the pulse (see b∗ ). Also, notice that the base noise increased as
compared to that in trace (A). Trace (D) shows that the spontaneous firing rate
after cessation of the pulse is still higher than that before the entrainment, but
with time (Trace E) it returns to preentrainment levels. Trace (F) shows selected
extracellular action potentials before entrainment (a∗ ), during entrainment (b∗ ),
and after entrainment (c∗ , d∗ ). Their overlap illustrates how action potential width
increases in a reversible manner during increased discharged frequency.
spontaneous firing of a neuron in the basal forebrain and its response to
the current pulses.
The ability and the degree to which a neuron can increase firing rate prob-
ably depend on the neuron’s molecular makeup and the state of the network
where it resides. This is why some cells can entrain and fire at very high fre-
quencies, while others cannot fire at such high frequencies in response to
current pulses, although their firing rate still increases substantially, i.e., two-
or threefold from their spontaneous rates. If the spontaneous firing rate is
very low, say on the order of 1 Hz, then a sixfold increase should indeed
be considered a very substantial change, despite the fact that a cell whose
spontaneous firing rate increases just from 20 to 60 Hz appears to be more
vigorously entrained. These vast differences that we have encountered in
rat basal forebrain may be due to the diversity of cell types in this brain area.
In particular, the compositions of their cellular membranes, ionic channels
available, etc. in part dictate their very different intrinsic properties. Hence,
it seems logical that they would respond in different ways to the same basic
stimulus. Other important considerations arise, for instance, from geomet-
rical constraints, the position of the electrode tip with respect to the soma,
the shape and size of the soma, ephaptic relations, etc. In general, in our
experience, faster firing neurons are much easier to label than slower firing
ones, because they entrain easily and require less time for good quality la-
beling. Neurons whose firing rate cannot be modulated at all by the current
pulse do not get labeled.
Generally, we applied pulses using an IR-183 Neurodata amplifier (Neu-
rodata Instruments, New York, NY). Current pulses are 200 ms long (50%
duty cycle). Entrainment for only a couple of minutes may be enough to
label the soma. More complete staining of dendrites and axons requires in
the order of at least 20 min.
3. Polarity of the Pulses
Both Neurobiotin and biocytin are zwitterions and should therefore be

ejected from the micropipette with negative or positive current pulses.
Pinault reported never to have seen labeling of neurons when apply-
ing iontophoretic negative (cathodal) pulses in the range of −50 nA to
−1 µA (Pinault, 1996), an observation corroborated by others. However, in
our own recordings and labeling of basal forebrain neurons, we have occa-
sionally encountered cells that responded better, i.e., entrained better, in
response to negative current pulses than to positive (anodal) current pulses.
In these rare instances, the application of negative current pulses when using
biocytin-filled electrodes did not result, as far as we could tell, in differences
in the quality of labeling or required labeling time. Although these cells
were not neurochemically identified, their morphology appeared similar to
many other typical neurons of the basal forebrain. By chance, we did not
attempt labeling of neurons by passing negative current pulses while using
Neurobiotin-filled electrodes. However, it is important to note that Kita and

Armstrong (1991), although describing intracellular recordings and not
juxtacellular labeling, reported that Neurobiotin is selectively ejected with
positive current pulses. They also suggested that this property would be ben-
eficial to electrophysiologists using hyperpolarizing currents to stabilize the
membrane potential of neurons prior to recording.
4. Action Potential Shape and Cellular Response After Cessation of Pulse
To avoid false expectations of labeling, it is essential to make sure that

the cell does indeed respond to the current pulses. If there is no response,
the neuron will not get labeled. It is also important to corroborate that the
neuron is still firing after cessation of the current pulse protocol and that
it continues to fire as the electrode is slowly removed from its vicinity. The
width of the action potential can be used to assess the health of a neuron and
also serves to indicate that labeling has happened. In response to the current
pulses, extracellular potentials can and usually become wider as shown in
Fig. 7.2. Apparently this broadening indicates some micropuncture of the
cellular membrane, a likely mechanism for the labeling (Pinault, 1996).
After some time, extracellular action potential width does return to the level
before injection, a possible indication of healing of the membrane. Also, it
is likely that the firing rate of the neuron will be higher after cessation of
the current pulse. However, over time it should return to baseline levels
(Fig. 7.2).
E. Control Experiments: The Neuron Recorded is the One Labeled
In order to establish that the labeled neuron was indeed the neuron that
was recorded, we designed several control experiments, some of them il-
lustrated in Fig. 7.3. First, we applied current pulses 1–10 nA in intensity,
for 15–30 min, without having first detected firing from a single neuron.
This protocol resulted in no labeling of cells but if applied longer than
approximately 20 min, it usually left a small accumulation of residual tracer
in the tissue. Second, we applied current pulses of the same intensity for
the same period of time as in the previous case, in the presence of a single
unit, but without obtaining any response (entrainment) from the cell be-
ing detected. This protocol also resulted in no labeling of neurons. Third,
we applied the same protocol while detecting a single unit and got the sin-
gle unit to respond to the pulses. In this case a single-labeled neuron was
later revealed and the morphological integrity of the neuron was preserved
(Fig. 7.3A, B). Fourth, we labeled a neuron but killed it at the end of the
labeling procedure by passing a high-voltage pulse. In these cases, when
animals were perfused within approximately 1 h after the procedure, we
indeed found a labeled neuron but usually its morphological integrity was
A B
C D
E F
Figure 7.3. Control experiments: the neuron labeled is the one recorded. Delivery of
current pulses without detection of a neuron or even when detecting a neuron that
does not entrain results in no labeling of cells. (A) and (B) illustrate cases in which a
single neuron was detected, recorded, and entrained (cells in two different animals).
In each case, after juxtacellular labeling the electrode was retracted slowly while the
cell was still firing. After proper histochemical procedures, a single-labeled cell was
found. (C) and (D) illustrate cases in which after juxtacellular labeling of a single
cell the neuron was “killed” by passing a high-voltage pulse. After histochemical
procedures, a badly damaged single-labeled neuron was found in each case. (E)
illustrates a case in which a single neuron was juxtacellularly labeled. The electrode
was retracted dorsally until the cell was no longer responding to the pulses, yet these
were passed for several minutes to create enough damage to mark the electrode
track. Afterward, the electrode was totally removed from the brain and lowered
again to the same depth at 75 µm lateral to the labeled cell, and pulses were passed
again to mark a second electrode track (without detecting and entraining any other
cells). After histochemical processing, one single-labeled cell was found at the end
of a marked electrode track and 75 µm lateral to it, a second electrode track was
found. (F) illustrates a case in which two neurons were detected and both responded
(entrained) in response to current pulses. In that case, two labeled neurons were
later found.
seriously compromised (Fig. 7.3C, D). If the animal was not perfused soon
after the end of the procedure, the remains of the labeled cell sometimes
were not found. Fifth, when we entrained a neuron for approximately 10
min. and then moved the electrode some micrometers away and proceeded
to apply current pulses for another 20 or 30 min, without detecting and
entraining a new cell or affecting the previously recorded neuron, we later
found a single-labeled neuron and, most of the time, a single electrode
track in the vicinity. Sixth, if by the same token, we first labeled a neuron
for a few minutes, then retracted the electrode dorsally until the cell was no
longer detected, marked the electrode track, and then removed the elec-
trode completely out of the brain and moved it precisely, for instance, 75
µm laterally and returned the electrode to the same depth and proceeded
to mark a second track, we later found a single track at the end of which we
had a labeled neuron, together with a second track located laterally to the
first 75 µm from it (Fig. 7.3E). Finally, if a single cell was entrained properly
for labeling and then we still sporadically entrained a second neuron in the
vicinity that was detected in the background, later we found a single strongly
labeled neuron and a second weakly labeled neuron. However, if this pro-
cedure was carried out for a long time (i.e., 45 min), both neurons might
be stained equally well, and then one could not determine with certainty
which one was related to the main signal (Fig. 7.3F).
F. Histology and 3D Light and Electron Microscopy Reconstructions
1. Histology and Immunohistochemistry
Histological and immunohistochemical procedures for visualization and

study of single juxtacellularly label neurons are identical to those followed
for intracellularly labeled neurons. In short, animals are usually perfused
with saline followed by a fixative such as acrolein, paraformaldehyde, glu-
taraldehyde, or a combination of these. Picric acid, glucose, and other chem-
icals are sometimes used to improve the outcome of a particular procedure.
Protocols, including incubation times and amounts of chemicals used, are
the choice of the investigator (see example in the Appendix), and there is
a host of different recipes available in the literature.
2. 3D Light Microscopy Reconstructions
Modeling and experimental studies suggest that neuronal morphology

(dendritic and axonal branching pattern) and class-specific connectivity
play an important role in shaping network function (see chapters by As-
coli et al. and Nadasdy et al. in this volume). Two-dimensional (2D) camera
lucida tracings of neurons have been performed following Cajal but all
the focus-axis information is lost and the only view possible is that of the
plane of sectioning. Another disadvantage of this technique is that the end
product is a drawing; one cannot statistically summarize the drawing without
somehow measuring it. Fortunately, recent technological advances allow the
use of 3D computerized reconstructions to obtain precise numerical details
about morphological characteristics and constraints for different neuronal
populations. These quantitative methods are needed in order to be able to
compare and contrast different types of neurons in a logical, mathematical
form. Through mathematics, the biological constraints on the architectural
nature of the brain and of its elements can be investigated. For instance,
dendritic size, together with their extensions into specific spatial domains,
can be an indirect measure of input density. Understanding at least some
of the structural diversity of dendritic arbors, their orientation, and size can
provide essential clues about their possible connections and hence, clues
about the intricacies of dendritic function.
Axonal reconstructions of identified neurons, although more difficult
than dendritic reconstructions, provide some clues as to whether the neuron
is a projection neuron, a local interneuron, or both. En passant or terminal
varicosities can be recorded during the light microscopical reconstructions.
These data, together with available information on the number and type of
neurons in the environment of the axonal arborizing space, can be used for
deriving the connectional probability of the electrophysiologically identified
neurons (Zaborszky et al., 2002). It is likely that digital experiments will be
increasingly used with the advance of 3D cellular databases equipped with
proper warping tools for data integration (http://www.ratbrain.org; see also
chapter by Nadasdy et al. inthis volume). The data from these “in silico”
experiments can be used to generate hypotheses that in turn can be tested
in actual experiments. For the state-of-the-art stereological assessment of
neural connections, see the chapter by Avendano in this volume.
Typically, dendritic and axonal trees can be reconstructed from serial sec-
tions using computerized microscope systems. The most widely adopted
commercial system is the Neurolucida r
system (MicroBrightField Inc.;
www.microbrightfield.com), which utilizes a computer-controlled stepping
motor stage and a high-resolution miniature CRT monitor. The motorized
stage allows data acquisition to extend distances significantly larger than a
single microscope field of view. The miniature CRT is coupled with the cam-
era lucida and superimposes a computer graphic display on the actual image
of the microscopic specimen viewed through the oculars of the microscope.
This permits the acquisition of microscopic data that require high resolution
and clarity to visualize. Neurolucida r
is operated on a PC equipped with
the Microsoft Windows operating system. Digital reconstruction of single-
labeled neurons as well as analysis of the morphological data using the Neu-
rolucida software suite (Glaser and Glaser, 1990) as well as other software
tools for extracting morphomertic measurements from the reconstructed
neurons are described in detail in the chapter of Ascoli and Scorcioni
(in this volume). Neurons in the Neurolucida r
system are represented by
the x, y , and z coordinates of manually traced points; thus morphological
reconstruction is a very time-consuming process. There are other somewhat

faster, but perhaps less precise, ways to combine traditional paper–pencil
tracings with computer-assisted reconstructions of scanned images (see
Ascoli and Scorcioni, this volume). Section “Applications” gives further de-
tails on reconstruction and morphological analysis of electrophysiologically
and chemically identified neurons.
3. Electron Microscopy-3D Reconstruction from Ultrathin Sections
Electron microscopy of ultrathin sections produces high-resolution 2D

images of cellular profiles and allows identification of synapses. However,
most of the three-dimensional structural information is lost, although it
can be recovered from reconstructing serial thin sections. The advantages
of 3D reconstruction from ultrathin sections are several, including the
ability to monitor plastic changes in the postsynaptic profiles (see chap-
ter of Goldberg et al., this volume), to better understand the complex
organization of the neuropil, and to count synapses. Although the num-
ber of synapses can be deduced from 2D images using various correc-
tion factors or applying “unbiased” sampling strategies with optical dis-
sectors as described in the chapter of Avendano in this volume, estimates
of synapses can also be done from proper light microscopic reconstruc-
tions and application of an adequate electron microscopical sampling strat-
egy. Although alignment of sections can be done manually (see Goldberg
et al., this volume), misalignment between sections due to mechanical and
optical distortion often necessitates the use of computer programs. At
present, the most popular program for alignment of EM images is the Recon-
struct (V1.03.2) tool developed by Harris and Fiala (www.synapses.mcg.edu;
http://synapses.bu.edu/tools/download.htm). This program computes the
alignment from three points identified by the user in each image. A program
under development that computes the fitting position using pixel density
values results in more accurate alignment (Simon et al., 2005).
G. Tracer Techniques in Combination with Electrophysiological

Recording and Juxtacellular Labeling
1. Choice of Retrograde and Anterograde Tracers
Retrograde and anterograde tracers may be applied to the brain as an

additional step prior to the juxtacellular labeling of single cells. Here, only
some comments are presented with specific reference to juxtacellular label-
ing and the reader is referred to appropriate chapters in this volume for
additional details about the tracers (Lanciego and Wouterlood, Reiner and
Honig; Sesack et al.; and Wouterlood).
All tracer injections require animal survival to allow tracer transport from
the injection site to the target area. It may be necessary to determine best
concentrations and survival times empirically. There are many excellent flu-
orescent tracers and selecting one depends on what fluorescent tag will be
used for visualization of the single cell. If, for instance, rhodamine (red) is
used for the visualization of the single cell, then a yellow or blue (or both)
fluorescent retrograde/anterograde tracer is a convenient choice. Fluoro-
gold (Fluorochrome Inc., Denver, CO) and Fast Blue (Sigma, St. Louis,
MO) are primarily retrograde tracers. The fluorescent dextrans are com-
monly used as anterograde or retrograde tracers (Reiner and Honig, in this
volume), and these include rhodamine isothiocyanate, rhodamine B dex-
tran, lysinated tetramethylrhodamine dextran (Fluoro-Ruby), and Fluoro
Emerald. Their advantages seem to be less diffusion at the injection site and
more permanent labeling than with the corresponding free dyes (Schmued
et al., 1990; Schmued and Heimer, 1990; Schmued, 1994). BDAs can also
be used as retrograde and anterograde tracers and offer superior character-
istics and versatility (Reiner et al., 2000). Since BDAs contain biotin, their
use in combination with biocytin or Neurobiotin labeling may be tricky and
is therefore not recommended. Other more direct choices of anterograde
and retrograde compounds for fluorescent detection include True Blue,
Granular Blue, 4 ,6-diamidino-2-phenylindole, Diamidino Yellow, Nuclear
Yellow, and Rhodamine Latex Beads, among others.
2. Sources of Artifact
There are several potential sources of artifact when tracers are injected
into any part of the brain. Two particular problems are the size of the tracer
uptake zone and the possibility of uptake of the tracer by fibers of passage at
the injection site. To minimize the fibers of passage problem, the simplest
rule is to minimize brain injury. To control the size of the tracer uptake zone,
tracer injections need to be steady and slow (over 10–15 min), which in our
experience creates less tissue damage and also minimizes the spread of the
injection by avoiding stressing the tissue around the injection site. To avoid
leakage of tracers along the pipette track, the exterior of the needle needs to
be cleaned and dried before inserting into the brain and must be left in place
for at least 5 min after the end of the injection. This technique appears to
minimize damage to the tissue and also reduces the possible uptake of tracer
by fibers of passage. However, the uptake of most tracers by fibers of passage
cannot be entirely abolished. To minimize movement of the needle and
improve reproducibility of injections, the syringe in use may be mounted on
a Hamilton Chaney adapter. Also, many tracers may have a small component
of transport in the opposite direction to the one intended, such as in the
case of vestiges of retrograde transport for an otherwise mainly anterograde
tracer injection or vice versa. In addition, because of the difficulty in limiting
the injection exclusively to the area of interest, one strategy that can be used
is to apply injections of different sizes (in different animals) so that larger
injections can increase the chance of including the majority of the target
area while smaller injections cover only portions of it.
III. APPLICATIONS
In this section we will illustrate the application of juxtacellular labeling of

single neurons to the study of neural circuits of the basal forebrain of the
rat. As outlined in Fig. 7.4 and illustrated in subsequent figures, we increase
the level of complexity of the investigation of single basal forebrain neurons
by increasing the number of procedures that are performed in addition to
extracellular recording and juxtacellular labeling. These added procedures
provide information about the transmitter, dendritic and axonal arboriza-
tion pattern, and connectivity of single electrophysiologically identified neu-
rons that could not otherwise be obtained, and which in the past has been
only inferred from separate experiments. This multiplicity of methods is
One level of Two levels of Multiple levels of

complexity complexity complexity
Extracellular
recording
EEG
C1
recordings
Extracellular Juxtacellular
recording + labeling AND/OR
A B Morphological reconstructions:
Gray area: fig. 7.5 3-D - light microscopy C2
AND/OR
Retrograde fluorescent labeling
Extracellular Juxtacellular
recording + labeling + AND/OR
Neurochemical identification
AND/OR
Bulk immunolabeling
AND/OR
Synaptology:
3-D - EM reconstructions
Extracellular Juxtacellular ANY OR

recording + labeling + ALL = RESULTS
Figure 7.4. Schematic illustration of the general experimental strategy: extracellular

recording and juxtacellular labeling of single neurons in combination with several
methods that allow further cellular characterization. The added levels of complexity
increase the amount of information collected about a single neuron. Letters A, B,
C1, and C2 are illustrated in detail in Fig. 7.5.
necessary because simplified criteria, which in some brain areas have been
successfully used to differentiate among different cell types, have failed in
the basal forebrain due to the heterogeneity of electrophysiological, mor-
phological, and neurochemical characteristics found in this region. This
combination of techniques has allowed us to study the EEG correlation of
the discharge properties of neurochemically and morphologically identi-
fied neurons in the basal forebrain (Duque et al., 2000). Our studies, for
instance, indicated that both cholinergic and parvalbumin-containing neu-
rons increase firing during cortical low-voltage fast-electrical activity (LVFA),
and therefore belong to the previously established category of “F” (fast) type
basal forebrain neurons. On the other hand, neuropeptide Y positive (NPY)
neurons that showed increased firing during cortical slow waves belong to
the so-called S (slow) cell type. Whether or not this is the case for all basal
forebrain cholinergic, parvalbumin, and NPY neurons is not known. Addi-
tional morphometric and synaptology studies in progress will allow us to
build the “basic circuitries” of the basal forebrain that are critical to under-
stand its functions.
A. Electrophysiological and Morphological Identification

of Single Neurons
Figure 7.5 illustrates a typical experiment consisting of a single-cell extra-

cellular recording with juxtacellular labeling and subsequent morpholog-
ical reconstruction of the recorded neuron. This neuron was labeled for
approximately 5 min by passing +8 nA current pulses.
From single-cell electrophysiological recordings, we classified this neu-
ron as a bursty cell with a firing rate of 2.95 Hz. Its action potential shape
is triphasic with a rise time of approximately 0.36 ms and total width of
2.88 ms. Labeling of the cell permitted us to locate the soma within the
horizontal limb of the diagonal band of Broca. The soma measures 15 ×
14 µm and its shape is round with six primary dendrites. The cell is a typical
basal forebrain neuron according to a previous classification using the jux-
tacellular technique (Pang et al., 1998). Panel C1 in Fig. 7.5 illustrates how
one additional electrophysiological procedure, EEG, concomitant with the
single-cell recording further allows the classification of this neuron into the
“S” category of basal forebrain neurons. This categorization indicates that
the firing rate increases during high-amplitude slow-cortical EEG activity
and it is different from neurons that have a faster firing rate during fast
cortical EEG activity, which were termed “F” cells (Detari and Vanderwolf,
1987; Dringenberg and Vanderwolf 1998; Detari, 2000). The simultaneous
recording of EEG and single-cell electrophysiological data also revealed that
there are neurons in the basal forebrain whose activities remain unchanged
despite EEG cortical changes (Nunez, 1996).
Figure 7.5C2 illustrates how one additional procedure, 3D light mi-
croscopy reconstruction, further advances our knowledge of this single
neuron. For instance, in addition to the soma shape and location, we now
Electrophysiological characteristics
Approx. firing pattern: bursty
Number of spikes analyzed: 2800
A unit
Mean firing rate: 2.95 Hz
Mean ISI: 163.60 + 217.85 (SD) ms CV: 1.35
Spike shape: tri-phasic, rise time 0.36 ms and
total with: 2.88 ms.
Single cell electrophysiological correlate to
10 s + EEG : The cell is "S" type: fires mostly during
slow wave activity. EEG can be acquiered from
+ C1 different cortical areas at the same time. This
EEG type of data provides behavioral relevance for
an isolated type of electrophysiological activity.
EEG/frequency composition analysis

B Cell Location:
Lomb
HDB 25 25 1.75 Hz
20
20
15
Amplitude
15 10 Significance level
10 5
0
5
0 2 4 6 8
Places the neuron 0

in reference to
0 10 20 30 40 50
other brain Frequency
structures
Allows the correlation between different
behavioral states and single cell
+ electrophysiology.
C2 Neuron during 3D reconstruction

3D Light microscopy reconstructions
provide morphometrical data that cannot
be obtained otherwise. These
morphometric parameters allow the
establishment of structuro-functional
relationships and the comparison between
different types of neurons.
Reconstructions are necessary for EM
analysis and at a minimum help us
understand the organization of neuronal
circuits.
Figure 7.5. Example of extracellular recording and juxtacellular labeling in combina-

tion with EEG recording and morphological reconstruction. Arrows point to boxes
containing examples of information that can be collected about the neuron from
the different combined methods.
know the extent and direction of its dendritic tree. We know that the neuron
has axon collaterals that communicate with nearby neurons and also a main
projection axon that runs in the ventromedial direction. This extra infor-
mation allows the formulation of hypotheses and the planning of further
experiments. Do all “S” cells have axon collaterals? If so, what types of local
neurons are contacted by “S” cells? Do they all have a projection axon?
B. Retrograde Labeling of Electrophysiologically Identified Neurons
Figure 7.6 illustrates the characterization of a basal forebrain neuron

using an additional procedure to the ones illustrated in Fig. 7.5. In this case,
Figure 7.6. Juxtacellularly labeled ChAT negative, parvalbumin-negative basal fore-

brain corticopetal neuron. This neuron was not tested for NPY. (A) The cell is
retrogradely labeled with Fluorogold from the prefrontal cortex. (B) The single
biocytin-filled neuron visualized with avidin-conjugated rhodamine. (C) Coronal
section illustrating the position of this cell. (D) Partial neurolucida reconstruction
showing that this neuron has only one projection axon (in red) and no axon col-
laterals in the neighborhood. The gray spots represent cholinergic neurons in close
proximity to some dendritic appendages. Scale bar in (A) (applies to B also): 50 µm.
approximately a week prior to the single-cell recording and juxtacellular la-

beling, the animal received an injection of the retrograde tracer Fluorogold
into the prefrontal cortex. The biocytin-labeled neuron was first visualized
with avidin-conjugated rhodamine (Fig. 7.6B). This neuron was retrogradely
labeled with Fluorogold (Fig. 7.6A), indicating that it projects to the pre-
frontal cortex. Additional immunostaining protocols revealed that this par-
ticular juxtacellularly stained and retrogradely labeled neuron was negative
for both choline acetyltransferase (ChAT) and parvalbumin, markers for
cholinergic and GABAergic neurons, respectively. After conversion of the
rhodamine fluorescent signal (juxtacellular staining) to nickel-enhanced
DAB, the tissue was processed for ChAT using DAB as end product. As
the 3D light microscopy reconstruction shows (Fig 7.6D), there are three
cholinergic cell bodies in the vicinity of this electrophysiologically identi-
fied and morphologically reconstructed cell. This identified neuron was F
type and a “just” projection neuron, since only a single axon was found that
lacked local collaterals. These extra procedures augmented the information
collected about this single neuron. Now we know its single-cell electrophysi-
ological characteristics and the correlation of these to the EEG (to the state
of the animal). In addition, we know some of the transmitters or chemical
markers for which it was negative, where it projects to, its morphology, and
the important fact that it did not have axon collaterals.
C. Chemical Identification and Morphometry

of Juxtacellularly Labeled Neurons
Figure 7.7 illustrates the addition of neurochemical identification to the

characterization of a single basal forebrain neuron. In this case, the single
neuron filled with biocytin was first visualized with avidin-conjugated rho-
damine (Fig. 7.7C) and subsequently found to be positive for the calcium-
binding protein parvalbumin, visualized with fluorescein isothiocyanate
(FITC) (Fig. 7.7B). After conversion of the rhodamine fluorescent signal to
nickel-enhanced DAB, the section containing the cell body of this electro-
physiologically and chemically identified cell was immunolabeled for ChAT.
As Fig. 7.7A shows, a cholinergic cell body (blue profile) was found in the
vicinity of the reconstructed parvalbumin-positive neuron that seemed to
be approached by axon collaterals of the electrophysiologically identified
neuron.
Figure 7.8 displays some morphometric data regarding the dendritic tree
of this parvalbumin-positive neuron, which had previously been electro-
physiologically identified. Panel A in Fig. 7.8 is the dendrogram showing
the dendritic tree of this neuron drawn in the same colors as used in the
tracing of Fig. 7.7A. A dendrogram is a stylized drawing of a branched struc-
ture, e.g., axon or dendrite. The purpose of a dendrogram is to visualize the
complexity of the 3D arborization pattern of the tree in a manner that can
be easily compared among different neurons. This dendrogram shows that
this particular parvalbumin neuron has eight dendritic trees, each dividing
into several daughter branches with a total length of 8785 µm. The graph
in Panel D, which displays the mean diameter of the dendritic branches,
indicates that primary dendrites are thicker and that as dendritic order
A
D
EEG 1 mV
unit
TP 5s
EEG 1 mV
unit
10 s
Figure 7.7. (A) Partial reconstruction of a juxtacellularly labeled basal forebrain

parvalbumin-positive neuron. The axon is shown in red. The soma and one of the
dendrites are shown in black and the other dendrites in different colors. Each den-
drite matches its color in the dendrogram shown in Fig. 7.8A. The pentagon en-
closes pieces of the neuron that are found in one of the sections from the serial
reconstruction and it corresponds to the block face shown in Fig. 7.9A. (B) The neu-
ron stained for PV (FITC). (C) Biocytin-filled neuron visualized with rhodamine.
(D) Concomitant unit and cortical EEG spontaneous and tail pinch (TP) induced
activity.
A B
2 4 6 7
3 5
1
400 µm
C
6.0 Mean tortuosity per
dendritic order
5.0
Tortuosity
4.0
3.0
2.0
1.0
700 µm
1 2 3 4 5 6 7
100 300 500 Dendritic order
Mean diameter per dendritic order Mean surface area per dendritic order
D E
900
1.8
Mean surface area (µm2)
800
Dendritic diameter (µm)
1.6 700
1.4 600
500
1.2
400
1.0
300
0.8 200
0.6 100
0
0.4
1 2 3 4 5 6 7 1 2 3 4 5 6 7
Dendritic order Dendritic order
Figure 7.8. Corresponding dendrogram (A) and dendritic polar histogram (B) for
the PV neuron shown in Fig. 7.7. The numbers at the abscissa in (A) represent
length of the dendritic trees in micrometers. The various dendritic trees are colored
the same way as in Fig. 7.7 to facilitate comparison. The outer circle of the polar
histogram corresponds to a 400-µm diameter around the origin. C, D, and E illustrate
mean tortuosity, mean diameter, and mean surface area all as functions of dendritic
order, exemplifying different types of morphometric analyses that can be performed
on data acquired with Neurolucida reconstructions. All error bars correspond to the
SEM. For further explanation see text.
progresses dendritic branches become thinner, although the end portions

of the dendrites have a tendency to be a bit thicker. This trend is similar to
the situation observed in cholinergic neurons and is opposite to that in NPY
neurons whose endings are very thin (Duque et al., in preparation). The
mean surface area, shown in Fig. 7.8E, is maximum in the case of secondary
dendrites and about the same between primary and tertiary dendrites. Since
tertiary dendrites, which are thinner than secondary and primary branches,
have as much surface area as primary dendrites, it is reasonable to deduce
that tertiary dendrites are longer than primary and secondary dendrites,
which is indeed the case (not shown). Higher order dendrites have overall
much less surface area. This type of analysis helps make predictions as to
what to expect in terms of numbers of inputs, i.e., primary and tertiary den-
drites could get roughly the same number of inputs, but second-order den-
drites might get more inputs since they have on average more surface area.
The polar histogram shown in Panel B gives a characterization of the direc-
tional distribution of dendritic growth projected onto the plane of section-
ing, similar to the analysis described by McMullen and Glaser (1988). The
algorithm for polar histograms breaks up the dendritic processes into line
segments and determines the directions in which these line segments point
and their corresponding lengths. The direction of the vector is calculated
by projecting the line segment onto the plane of sectioning. The histogram
represents total length by the distance from the origin and the angle that the
vector makes with the x-axis plotted in the radial direction. Each sector in
the polar histogram is the sum of all the dendritic growth in that particular
range of angle. The dendrogram is another tool to compare quantitatively
different neurons. The dendritic tortuosity displayed in Panel C gives a ratio
between the actual length of a dendritic segment and the distance between
its endpoints. Both the tortuosity and the dendritic orientation, as mea-
sured with the polar histogram, can determine spatial correlation between
overlapping axonal and dendritic arbors, thus affecting the probability of
connections (see also Zaborszky et al., 2002; Stepanyants et al., 2004). These
analytical tools are provided in the Neurolucida software package. Addi-
tional analyses that would also permit elaborate comparison of dendritic
morphometric parameters between various neuronal types (Scorcioni et al.,
2004; Li et al., 2005) can be performed using L-Measure (Scorcioni and As-
coli, this volume), a JAVA program freely available both for download and
for Web-based usage (http://www.krasnow.gmu.edu/L-Neuron).
D. Synaptology of Electrophysiologically and

Chemically Identified Neurons
Embedding tissue sections containing electrophysiologically and chemi-

cally identified neurons into plastic allows the study of the different neurons’
synaptic relationships. For example, we can study the input to specific den-
dritic compartments of an electrophysiologically and chemically identified
neuron. Figure 7.9 shows the intermediary documentation, at the LM level,
for a small dendritic segment of the parvalbumin neuron depicted in Fig.
7.7. Figures 7.10 and 7.11 show various boutons that impinge on the den-
dritic shaft of this parvalbumin neuron. Figure 7.12 shows a 3D rendering
of the same piece of dendrite reconstructed from 66 ultrathin sections. As
Figure 7.9. Correlated light–electron microscopy of the parvalbumin neuron shown

in Fig. 7.7A. (A) Processes of this neuron contained in section 17 that was serially
thin sectioned are shown against the full arborization of this neuron displayed as
background. (Compare this panel with the enclosed processes in the pentagon of
Fig. 7.7A). Axons are in red and dendritic processes are in black. (B) Blockface
image of thick section 17. Three arrows point to a dendritic shaft that is also labeled
in (A). Arrow with asterisk points to another dendritic shaft to facilitate comparison.
(C, D) Low-magnification (190×) montage of a thin section from this material with
Figure 7.10. (A) Low-magnification electron micrograph (1400×) to show the boxed
area from Fig. 7.9E. (B) High-power electron micrograph to illustrate the dendritic
segment that has been 3D reconstructed from serial thin sections. Numbers denote
different boutons.
the 3D model shows, this dendritic segment, which is about 3 µm in length,

is surrounded by 11 boutons and most of them had synaptic appositions.
Knowing that the total length of the dendrites of this parvalbumin neu-
ron is 8785 µm, one can estimate that this neuron may receive as many as
←
Figure 7.9. (Cont.) (C) and without (D) processes as seen in the Neurolucida file
(A). The boxed area in (D) is enlarged in (E). (E) Low-power electron micrograph
(440×) showing the boxed area from (D). The boxed area in (E) is shown at higher
magnification in Fig. 7.10A. The two capillaries with asterisks are fiducial markers to
compare (D) and (E). Arrow in the boxed area points to the small dendritic piece
that is displayed in higher magnification in Figs. 7.10–7.11 and in the 3D model of
Fig. 7.12.
Figure 7.11. (A) Ultrathin section 23 in which boutons 2, 4, and 5 can be recognized
(compare Fig. 7.10B). (B) and (C) show ultrathin sections 24 and 26 from the
series of 66 sections with additional boutons 6 and 7. White arrows point to synaptic
attachments.
32,000 synapses. However, since primary, secondary, etc. dendritic branches

possess different diameters and may be contacted by different axons, for
a more accurate estimation one has to select samples from each dendritic
segment and the terminal branches of the dendritic trees. The dendrogram
depicted in Fig. 7.8A helps to design the sampling strategy. As discussed un-
der section “Chemical Identification and Morphometry of Juxtacellularly
Labeled Neurons.” dendritic length, diameter, and surface area values can
affect the spatial arrangement of incoming boutons. In addition, analyses
on electron micrographs (e.g., average bouton size) and data from tracing
studies need to be taken into consideration if we want to arrive at a realistic
estimation of the type and synaptic number than can impinge on single or
statistically derived average neuron (Zaborszky et al., 1975). Similar morpho-
metric analyses on the axonal tree in conjunction with available data on the
number of neurons that are in the space of axonal arborizations can help to
estimate the number and postsynaptic targets of the local axon collaterals of
electrophysiologically identified neurons (see Zaborszky and Duque, 2000).
Figure 7.12. 3D partial reconstruction of the dendritic segment of the parvalbumin

neuron illustrated in Fig. 7.7A. This 3D model consists of 66 ultrathin sections,
rotated along the dendrite (green color). Numbered boutons are labeled with dif-
ferent color. Boutons labeled with numbers 2, 4, and 5 can also be seen in Figs.
7.9B and 7.10B. Boutons 2, 5, 6, and 7 can be identified on Fig. 7.11B, C. The red
surface marked with CH corresponds to cholinergic dendritic profiles that are in
close vicinity to the parvalbumin dendrite.
IV. SUMMARY OF ADVANTAGES AND LIMITATIONS
The fundamental advantage of any single-cell recording technique com-

bined with the labeling of the recorded cell is that it allows the correla-
tion of physiology and morphology at the cellular level. The advantage of
juxtacellular labeling is its compatibility with a host of other techniques that,
when combined, advance our ability to characterize single neurons in many
different ways. By virtue of being extracellular, the recording can usually be

maintained for prolong periods of time and it is less prone to disruptions
due to movement. Hence, recording and labeling of neurons can be done
both in superficial, i.e., cortical, as well as in subcortical structures that are
several millimeters deep. Juxtacellular labeling allows for the staining of neu-
rons in deep subcortical structures with a success rate of usually more than
80%. Neurons labeled with biocytin or Neurobiotin can be stained for light
microscopy and/or electron microscopy using chromogens such as DAB.
Juxtacellular labeling of single neurons is compatible with immunohisto-
chemical techniques, making it possible to collect data about the chemical
identity of the recorded and labeled cell. As seen in the previous exam-
ples, these advantages do not sacrifice the possibility of fine morphological
evaluation at the light and/or the electron microscopy levels.
The staining obtained with juxtacellular labeling can reach Golgi-like
quality in the soma and dendrites. It seems that myelinated axons are always
filled. However, depending on labeling time and quality of entrainment,
some axons are not fully labeled.
A. Advantages
1. The stability of the method allows long recordings, which in turn permit
more complete electrophysiological characterizations.
2. The method allows for recording and successful labeling of neurons in
various brain structures, particularly in very deep subcortical regions,
where intracellular recordings are almost impossible to obtain.
3. At least in theory, the natural cellular membrane and intracellular
environment is less disturbed since the cell is not being impaled with
an electrode.
4. The method is species independent and compatible with immunohis-
tochemistry and electron microscopy.
B. Limitations
1. Because the recording is extracellular, postsynaptic potentials cannot

be recorded.
2. Quality of the “fill” depends on the length of time the cell is entrained
and might not be as complete as the filling with intracellular techniques
if the cell is not labeled long enough.
3. Axonal labeling may also be partial, but this again seems to depend on
labeling time.
4. Occasionally, a glial cell in the vicinity of the labeled cell may also get
labeled.
5. The combination of juxtacellular labeling with many other techniques,
although powerful in providing exquisitely detailed characterization of
single cells, can be limited in practice because it is labor-intensive and
time-consuming.
APPENDIX
As illustrated in Fig. 7.4 and throughout examples in section “Applica-

tions”, the advantage of juxtacellular labeling is its compatibility with many
other techniques, which allows increasingly detailed characterization of a
single cell. Figures 7.6–7.12 give an outline of possible steps for a sample
experiment in which a neuron can be retrogradely labeled and then tested
for several neurochemicals to be then converted to Ni-DAB, double labeled
for ChAT bulk immunostaining, reconstructed, and finally analyzed at the
EM level.
A. Animal Preparation Prior to Electrophysiological Recordings
Day 0
1. Anesthesia: Because these are survival surgeries, use a mixture of ke-

tamine (85 mg/kg) and xylazine (15 mg/kg) i.p. Do not use urethane,
as it affects gastrointestinal motility (Yuasa and Watanabe, 1994) and is
toxic to the liver and lungs (Renuka and Dani, 1983). Do not use pen-
tobarbital; it may affect retrograde and anterograde tracer transport
(Rogers et al., 1980). Gas anesthesia can also be used.
2. Stereotaxis: For stability and proper location of brain structures, anes-
thetized animals need to be placed correctly in an appropriate
stereotaxic apparatus. Wound margins and points of contact between
animal and stereotaxic apparatus are usually infiltrated with lidocaine
solution (2%) and xylocaine ointment (5%), respectively. Aseptic con-
ditions are necessary.
3. Tracer injections: Fluorogold is primarily a retrograde tracer. Prepare
it at 2.5–4% in double distilled H2 O. Pressure inject 0.05–0.3 µl/
10 min/per site using a 1 µl Hamilton syringe. Survival time: 2–14
days. Fast Blue (FB, Sigma): same as for FG.
B. Animal Preparation for Electrophysiology
Day 1
1. Anesthesia: Use urethane 1.3 g/kg i.p. Usually injected once. Caution:
Urethane is highly toxic.
2. Surgery: Place animal in a stereotaxic apparatus as described before.
Retract scalp and overlying fascia from the skull and puncture the
atlanto-occipital membrane to allow drainage of some cerebral spinal
fluid. Keep body temperature at 37–38◦ C preferably with a hot water
circulating pad. Drill small burr holes in both hemispheres, over, for
instance, the frontal cortex for EEG recordings [anteroposterior (AP)
+1.6–1.8 mm, mediolateral (ML) ± 0.5–2.0, relative to bregma] and
over the basal forebrain (AP −0.3 to −1.0), L ±2.4–3.2 mm, relative to
bregma) for single unit recordings.
C. Electrophysiology and Labeling
1. Electrode fabrication: Make recording microelectrodes from 2.0 mm

outer diameter borosilicate glass capillaries (World Precision Instru-
ments, Sarasota, FL) on a Narishige PE-2 vertical pipette puller. Break
the tip of the electrode under visual guidance to approximately 1.0 µm
in diameter. Fill the electrode with 0.5 M NaCl containing 4% biocytin.
Measure in vitro impedance and use 10–30 M electrodes.
2. Set up EEG electrodes and obtain a signal: Find and record a neuron
long enough to allow statistical significance test to be performed for
its single-unit electrophysiological characterization. Record EEG and
single-cell electrophysiology at the same time, to allow correlation
analysis.
3. Labeling: Get as close as possible to the cell. Entrain the cell by passing
current pulses 1–10 nA in intensity. Monitor entrainment and move
the electrode closer to or further away from the cell or increase or
decrease pulse intensity as necessary to maintain entrainment without
causing cellular damage. Entrain cell for at least 20 min to obtain good
labeling.
D. Perfusion
Within a few hours after the termination of the labeling protocol, perfuse
the animal transcardially.
Pass 100 ml normal saline followed by 200 ml of ice-cold fixative [4%
paraformaldehyde, 15% saturated picric acid, and 0.05% glutaraldehyde in
0.15 M phosphate buffer (PB), pH 7.4], followed by 200 ml of the same
fixative without glutaraldehyde.
Remove brain and postfix at 4◦ C overnight in the fixative without glu-
taraldehyde.
Optional: Cryoprotect in sucrose if the brain is to be cut frozen in a cryostat
or on a freezing microtome.
E. Cutting and Pretreatment of Sections
Day 2
1. Cutting: Dissect the block of tissue containing the cell. The size of the
block depends on what you expect and want to process. If long projec-
tion axons are expected, then cut the block so that it contains the area
with the terminals of the axon. Notch one side of the brain so that you
can keep track of left and right hemispheres. Cut coronal or sagittal
sections 50–60 µm thick with a Vibratome r
or freezing microtome if
only light microscopic processing is planned.
2. Rinsing: Select the sections to be processed and rinse them several
times in cold PB until the yellowish color of the picric acid disappears.
All rinses from here on should be done in cold 0.1 M PBS. Every rinse
should take about 5 min, while gently agitating the sections in a shaker.
3. Borohydride and peroxidase pretreatments: Incubate sections in 1% sodium
borohydride in PBS (to remove excess aldehydes) for 20 min. Rinse
3–5 times or until the bubbles are gone and the sections sink. Then
incubate for 10 min in 1% hydrogen peroxide in cold PBS (to block
peroxidases) and rinse again three times.
F. Visualization of Biocytin-Filled Neuron and Digital Photography
Days 3 and 4
1. Select 48 sections where you expect to find the soma of the juxtacellu-
larly labeled neuron (so that the soma will be found within 2.4 mm of
tissue, which is considered a very large margin of error). Store the rest
of the sections in PBS at 4◦ C.
2. Incubate the sections overnight (4◦ C, in a shaker) in avidin-conjugated
rhodamine (R) (1:500; Jackson ImmunoResearch Labs, West Grove,
PA). This can be done in six scintillating vials, each containing 1 ml of
solution and the corresponding sixth section of the series for a total
of eight sections (make sure the sections are all submerged in the
solution). This way, vial 1 will have sections 1, 7, 13, and so on; vial 2
will have sections 2, 8, 14, and so on.
3. Rinse the sections once or twice (to remove excess fluorescent parti-
cles).
4. Searching for the cell: Arrange the 48 sections individually in rostro-
caudal order. (This can be done in two 24-well dishes.) Then mount
the first section onto a glass slide and search for the cell under a
epifluorescence microscope. Do not cover the section with a glass
coverslip and do not use anything to enhance fluorescence. Just keep
the section wet with cold PBS. These additional steps may substantially
deteriorate the tissue and diminish the ability to do lengthy process-
ing. If you do not find the cell, jump 150 µm and mount the fourth
section and repeat the process. Instead of scanning sections in order,
this speeds up the process of finding the cell; usually you will find den-
dritic processes and then you can just follow them very quickly (maybe
through several sections) to the soma.
5. Document your finding by taking photographs, maybe at two or three
different magnifications, usually 5×, 20×, and 40×. Low-magnification
pictures can quickly determine the region where the cell is located.
More than 40× may be difficult because of the focusing over wet
mounted tissue. Fluorescence may be very intense so that the soma
may appear larger and blurry. Try to minimize this by illuminating at
less than 100% and by using filters. At this time one can also determine
if the cell is double labeled for instance with Fluorogold.
Note : We suggest using red fluorescent markers to label single cells because
the normal human eye is more capable of detecting red than any other
color.
G. Neurochemical Identification of Biocytin-Filled Neuron
Day 5
1. Incubate the section with the soma and a second control section se-
lected from the ones in storage in a monoclonal rat anti-ChAT anti-
body (Rat anticholine acetyltranferase; 1:10; 2 days at 4◦ C; Boehringer
Mannheim, Germany). Triton X can be added to help penetration of
the antibodies and improve chances of positive immunotest, i.e., 0.02%
Triton X in 0.1 M PB. This, however, may render the section useless for
EM. Store the rest of the sections in 0.1 M PBS at 4◦ C or proceed to
develop them (see conversion of fluorescent signal to DAB).
2. Rinse the section with the soma and the control section.
3. Incubate in a secondary antibody conjugated to, for instance, fluores-
cein isothiocyanate (FITC-conjugated goat anti-rat; 1:100–200; 4 h at
room temperature; Jackson ImmunoResearch Labs, West Grove, PA).
If the results of the incubation are poor, i.e., the fluorescent signal is
not very good, the incubation can be done overnight at 4◦ C in a shaker.
The problem is that in serial testing for different immunochemicals,
this can add substantial time to the processing.
Day 6
Locate again the single-labeled cell with the “red” excitation/emission

filter set of the fluorescence microscope and photograph it; change filters
and determine if the same cell also emits green fluorescent light. If it does,
then this is evidence that the juxtacellularly labeled cell expresses ChAT
and it should be photographed again. If the identified cell is not ChAT
positive, one can repeat the steps using a different primary antibody and a
different control section. This procedure can be repeated two or three times.
The secondary antibody can always be conjugated to FITC or some other
“green” fluorochrome. This does not confound the results because one
can observe how additional cells appear under “green” fluorescence in the
field of interest, including or not the identified neuron. The latter neuron
remains the only cell visible through the “red” fluorescence filter set.
H. Conversion of the Fluorescent Signal to DAB
1. Incubate the section containing the soma and all adjacent sections
of interest (that were processed in avidin-conjugated rhodamine)
in biotinylated peroxidase [1:200, “B” component of standard ABC
(Avidin–Biotin Peroxidase complex) kit] (Vector Laboratories Inc.,
Burlingame, CA) for 2–4 h at room temperature (RT). The “A”
or avidin component of the ABC kit is omitted because the single-
stained neuron already contains the avidin (from the avidin-conjugated
rhodamine).
2. Develop the neuron using 3,3 -diaminobenzidine tetrahydrochloride
as a chromogen intensified with nickel (Ni) by incubating the sections
for 10 min in a solution containing 0.05% DAB and 0.038% nickel am-
monium sulfate and then adding hydrogen peroxide to a final concen-
tration of 0.01%, while agitating for another 10–20 min. To determine
the best timing, periodically wet mount the section with the soma or
one with dendrites and see how dark they are using a regular trans-
mitted light microscope. Try to balance the result between a very dark
signal and a very low background. The overall darkness of the section
containing the soma may be different because of the extra processing.
3. Rinse thoroughly to get rid of any Ni-DAB deposits outside the labeled
neuron.
Note : The development of DAB can also be done by using 10% B-d-glucose
and glucose oxidase instead of hydrogen peroxide. In short, after incubat-
ing the sections in the “B” component of the ABC kit as described above,
incubate sections (15–25 min at RT) in a solution of 0.1 M PB containing
50 mg DAB, 40 mg ammonium chloride, 40 mg nickel ammonium sulfate,
0.4 mg glucose oxidase, and 200 mg B-d-glucose per 100 ml of solution.
If one desires to filter the solution, this must be done before adding the
glucose oxidase.
I. Staining for a Second Antigen
Days 7 and 8
1. Select the section: If the single cell was ChAT positive and one wants to
double label the material, in order to find out if axon collaterals of the
cholinergic neuron contact, for example, NPY neurons, then several
sections around the soma of the cholinergic cell should be selected.
2. Incubate the sections in a primary antibody (Rabbit anti-NPY; 1:500;
2 days at 4◦ C; Peninsula Laboratories, Inc., Belmont, CA).
3. Rinse sections two or three times, then incubate in secondary antibody
(Biotinylated Goat anti-Rabbit 1:200, 4 h RT; Jackson ImmunoResearch
Labs, West Grove, PA).
4. Incubate in ABC as indicated in the ABC kit. Develop as described

above, but using DAB only, without nickel enhancement. Develop long
enough to make the signal light brown. Do not overdevelop because
then it will be difficult to distinguish black from brown.
Note: If there are any concerns about again using avidin and biotin be-
cause of the possible detection of false signals beforehand, sections can be
blocked using the avidin–biotin blocking agents sold by Vector (follow in-
structions in the package) or the signal can be developed by the peroxidase
antiperoxidase (PAP) technique (see Pickel and Milner, 1989).
J. Embedding for Electron Microscopy
Days 9 and 10
1. Osmification: Sections to be investigated at the EM level should be os-

micated in a solution containing 1% osmium tetroxide (Electron Mi-
croscopy Sciences, Fort Washington, PA) in phosphate-buffered saline
(PBS), for approximately 30–40 min.
2. Dehydrate tissue in an ascending series of ethanols (30, 50, 70, 90,
100%). Do contrasting by treating the tissue with 1% uranyl acetate
(Electron Microscopy Sciences, Fort Washington, PA) in 70% ethanol,
for 30 min.
3. Dehydrate tissue with 1% propylene oxide (Electron Microscopy Sci-
ences, Fort Washington, PA).
4. Infiltrate sections in durcupan (Fluka Chemie AG, Buchs, Switzerland)
overnight and then flat embed them between liquid release agent-
coated (Electron Microscopy Sciences, Fort Washington, PA) micro-
scope glass slides and coverslips.
K. 3D Light Microscopy Reconstructions
Days 11–30
Neuron reconstructions at the light microscopy level can be carried out

using several methods. Because 3D reconstruction offers more information
than do 2D reconstructions obtained with typical camera lucida systems, it
is advantageous to use a computerized 3D neuron tracing system such as
the one offered by MicroBrightField Inc. (Williston, VT). Starting from the
soma, the entire neuron is reconstructed from serial sections. The number of
sections used in each case varies according to the cell. In the case of double-
labeled material it is convenient to also plot other neurons, which are in close
proximity to the axon of the juxtacellularly label cell. In the cases presented
here the Neurolucida hardware system was interfaced with a Zeiss Axioscope
microscope. Outlines of the sections, contours of structures, and fiducial
markers were drawn with a 5× Plan-NEOFLUAR objective lens. Somata,
dendritic, and axonal branches were traced with a 100×, oil immersion
ACHROPLAN objective lens.
L. Electron Microscopy and 3D Reconstruction

from Ultrathin Sections
Days 31–40
After photography and full serial reconstruction of the labeled neuron,

coverslips can be removed and pieces of interest are dissected out of the
tissue with a razor blade. Small pieces of tissue containing areas of interest
(such as boutons) are mounted onto blank durcupan blocks and trimmed
appropriately. Ribbons of ultrathin sections are cut on a Reichert Ultra-
cut E ultramicrotome and picked up onto Formvar-coated single-slot grids
(Electron Microscopy Sciences, Fort Washington, PA). In our examples, the
ultrathin sections were analyzed on a Tecnai 12 transmission electron mi-
croscope and pictures were captured using either a Gatan Ultrascan digital
camera US 4000 SP (11,000×) or conventional EM film. The films were
developed and then digitized using a Microtek Scanmaker 4 scanner and
further manipulated using Photoshop.
a. Ultrathin cutting
1. Set the block into the holder and using a blade trim it roughly.
2. Under the light microscope, determine the depth in which the area of
interest is. With a glass knife, cut away the layers superficial to the area
of interest.
3. Using a glass knife, trim the block.
4. Begin to cut ultrathin sections using a diamond knife with simultaneous
light microscopic control.
5. Mount sections onto Formvar-coated single-slot grids (copper for gen-
eral purposes or nickel if intending to do postembedding procedures).
6. Dry grids with filter paper and then place them into gridboxes.
Days 41–60
For quantifying synaptic contacts on electrophysiologically identified neu-

rons, a certain portion of the dendritic tree or of the chemically and elec-
trophysiologically identified neuron has to be reconstructed and the total
number of synapses can be extrapolated knowing the total dendritic length
and the number of synaptic boutons in the reconstructed volume. The small
dendritic segment of the parvalbumin neuron shown in Fig. 7.12 was recon-
structed from 66 ultrathin sections. The collection of digital images took
about 20 EM h and about 10-h computer time was needed to reconstruct it
in 3D.
Acknowledgments. The research summarized in this review was sup-

ported by NIH grant NSO23945 to LZ and IR25 GM60826 to LZ and AD.
The Tecnai 12 electron microscope was purchased from grant S10 RR13959
(LZ). Special thanks are due to Lennart Heimer for his comments on a
previous version of this manuscript.
REFERENCES
Aghajanian, G. K., and Vandermaelen, C. P., 1982, Intracellular identification of central no-
radrenergic and serotonergic neurons by a new double labeling procedure, J. Neurosci.
2:1786–1792.
Alonso, A., Khateb, A., Fort, P., Jones, B. E., and Muhlethaler, M., 1996, Differential oscillatory
properties of cholinergic and noncholinergic nucleus basalis neurons in guinea pig brain
slice, Eur. J. Neurosci. 8:169–182.
Arbib, M. A., Erdi, P., and Szentágothai, J., 1998, Neural Organization: Structure, Function, and
Dynamics, Cambridge: The MIT Press, p. 407.
Blackstad, T. W., 1965, Mapping of experimental axon degeneration by electron microscopy
of Golgi preparations, Z. Zellforsch. Mikrosk. Anat. 67:819–834.
Buhl, E. H., 1993, Intracellular injection in fixed slices in combination with neuroanatomical
tracing techniques and electron microscopy to determine multisynaptic pathways in the
brain, Microsc. Res. Tech. 24:15–30.
Cajal, S. R., 1911, Histologie du Système nerveux de l’Homme et des Vertébrés, Paris: A Maloine.
Cullheim, S., and Kellerth, J. O., 1978, A morphological study of the axons and recurrent axon
collaterals of cat sciatic alpha-motoneurons after intracellular staining with horseradish
peroxidase, J. Comp. Neurol. 178:537–557.
Detari, L., 2000, Tonic and phasic influence of basal forebrain unit activity on the cortical EEG,
Behav. Brain Res. 115:159–170.
Detari, L., and Vanderwolf, C. H., 1987, Activity of identified cortically projecting and other
basal forebrain neurones during large slow waves and cortical activation in anaesthetized
rats, Brain Res. 437:1–8.
Dringenberg, H. C., and Vanderwolf, C. H., 1998, Involvement of direct and indirect pathways
in electrocorticographic activation, Neurosci. Biobehav. Rev. 22:243–257.
Duque, A., Balatoni, B., Detari, L., and Zaborszky, L., 2000, EEG correlation of the discharge
properties of identified neurons in the basal forebrain, J. Neurophysiol. 84:1627–1635.
Fairen, A., Peters, A., and Saldanha, J., 1977, A new procedure for examining Golgi impregnated
neurons by light and electron microscopy, J. Neurocytol. 6:311–337.
Freund, T. M., and Somogyi, P., 1989, Synaptic relationships of Golgi-impregnated neurons
as identified by electrophysiological or immunocytochemical techniques, In: Heimer, L.,
and Zaborszky, L. (eds.), Neuroanatomical Tract-Tracing Methods 2, New York: Plenum Press,
pp. 201–238.
Glaser, J. R., and Glaser, E. M., 1990, Neuron imaging with Neurolucida—a PC-based system
for image combining microscopy, Comput. Med. Imaging Graph. 14:307–317.
Golgi, C., 1883, Recherches sur l’histologie des centers nerveux, Arch. Ital. Biol. 3:285–317.
Grace, A. A., and Bunney, B. S., 1983a, Intracellular and extracellular electrophysiology of
nigral dopaminergic neurons—1. Identification and characterization, Neuroscience 10:
301–315.
Grace, A. A., and Bunney, B. S., 1983b, Intracellular and extracellular electrophysiology of
nigral dopaminergic neurons—2. Action potential generating mechanisms and morpho-
logical correlates, Neuroscience 10:317–331.
Graham, R. C., Jr., and Karnovsky, M. J., 1966, The early stages of absorption of injected
horseradish peroxidase in the proximal tubules of mouse kidney: ultrastructural cyto-
chemistry by a new technique, J. Histochem. Cytochem. 14:291–302.
Jankowska, E., Rastad, J., and Westman, J., 1976, Intracellular application of horseradish per-
oxidase and its light and electron microscopical appearance in spinocervical tract cells,
Brain Res. 105:557–562.
Kita, H., and Armstrong, W., 1991, A biotin-containing compound N-(2-
aminoethyl)biotinamide for intracellular labeling and neuronal tracing studies:
comparison with biocytin, J. Neurosci. Methods 37:141–150.
Kitai, S. T., Kocsis, J. D., Preston, R. J., and Sugimori, M., 1976a, Monosynaptic inputs to
caudate neurons identified by intracellular injection of horseradish peroxidase, Brain Res.
109:601–606.
Kitai, S. T., Kocsis, J. D., and Wood, J., 1976b, Origin and characteristics of the cortico-caudate
afferents: an anatomical and electrophysiological study, Brain Res. 118:137–141.
Koos, T., and Tepper, J. M., 1999, Inhibitory control of neostriatal projection neurons by
GABAergic interneurons, Nat. Neurosci. 2:467–472.
Koos, T., and Tepper, J. M., 2002, Dual cholinergic control of fast-spiking interneurons in the
neostriatum, J. Neurosci. 22:529–535.
Kristensson, K., and Olsson, Y., 1971, Retrograde axonal transport of protein, Brain Res. 29:363–
365.
LaVail, J. H., and LaVail, M. M., 1972, Retrograde axonal transport in the central nervous
system, Science 176:1416–1417.
Li, Y., Brewer, D., Burke, R. E., and Ascoli, G. A., 2005, Developmental changes in spinal
motoneuron dendrites in neonatal mice, J. Comp. Neurol. 483(3):304–317.
Light, A. R., and Durkovic, R. G., 1976, Horseradish peroxidase: an improvement in intracel-
lular staining of single, electrophysiologically characterized neurons, Exp. Neurol. 53:847–
853.
Lynch, G., Deadwyler, S., and Gall, C., 1974a, Labeling of central nervous system neurons with
extracellular recording microelectrodes, Brain Res. 66:337–341.
Lynch, G., Gall, C., Mensah, P., and Cotman, C. W., 1974b, Horseradish peroxidase histochem-
istry: a new method for tracing efferent projections in the central nervous system, Brain
Res. 65:373–380.
McMullen, N. T., and Glaser, E. M., 1988, Auditory cortical responses to neonatal deafening:
pyramidal neuron spine loss without changes in growth or orientation, Exp. Brain Res.
72:195–200.
Nambu, A., and Llinas, R., 1997, Morphology of globus pallidus neurons: its correlation with
electrophysiology in guinea pig brain slices, J. Comp. Neurol. 377:85–94.
Nunez, A., 1996, Unit activity of rat basal forebrain neurons: relationship to cortical activity,
Neuroscience 72:757–766.
Pang, K., Tepper, J. M., and Zaborszky, L., 1998, Morphological and electrophysiological char-
acteristics of noncholinergic basal forebrain neurons, J. Comp. Neurol. 394:186–204.
Pickel, V., and Milner, T. A., 1989, Interchangeable uses of autoradiographic and peroxidase
markers for electronmicroscopic detection of neuronal pathways and transmitter-related
antigens in single sections, In: Heimer, L., and Zaborszky, L. (eds.), Neuroanatomical Tract-
Tracing Methods 2, New York: Plenum Press, pp. 97–128.
Pinault, D., 1994, Golgi-like labeling of a single neuron recorded extracellularly, Neurosci. Lett.
170:255–260.
Pinault, D., 1996, A novel single-cell staining procedure performed in vivo under electrophys-
iological control: morpho-functional features of juxtacellularly labeled thalamic cells and
other central neurons with biocytin or Neurobiotin, J. Neurosci. Methods 65: 113–136.
Renuka, and Dani, H. M., 1983, Effects of toxic doses of urethane on rat liver and lung micro-
somes, Toxicol. Lett. 15:61–64.
Richards, C. D., 1972, On the mechanism of barbiturate anaesthesia, J. Physiol. 227:749–767.
Rogers, R. C., Butcher, L. L., and Novin, D., 1980, Effects of urethane and pentobarbital
anesthesia on the demonstration of retrograde and anterograde transport of horseradish
peroxidase, Brain Res. 187:197–200.
Schmued, L. C., 1994, Anterograde and retrograde neuroanatomical tract tracing with fluo-
rescent compounds, Neurosci. Protoc. 50:1–15.
Schmued, L., Kyriakidis, K., and Heimer, L., 1990, In vivo anterograde and retrograde axonal
transport of the fluorescent rhodamine-dextran-amine, Fluoro-Ruby, within the CNS, Brain
Res. 526:127–134.
Scorcioni, R., Lazarewicz, M. T., and Ascoli, G. A., 2004, Quantitative morphometry of hip-
pocampal pyramidal cells: differences between anatomical classes and reconstructing lab-
oratories, J. Comp. Neurol. 473(2):177–193.
Simon, L., Noszek, A., Garab, S., and Zaborszky, L., 2005, Optimal alignment of EM serial sec-
tions. Program No. 458.113.2005. Abstract Viewer/Itinerary Planner. Washington, DC: Society
for Neuroscience. Online.
Snow, P. J., Rose, P. K., and Brown, A. G., 1976, Tracing axons and axon collaterals of spinal
neurons using intracellular injection of horseradish peroxidase, Science 191:312–313.
Somogyi, P., 1977, A specific “axo-axonal” interneuron in the visual cortex of the rat, Brain Res.
136:345–350.
Somogyi, P., and Freund, T. M., 1989, Immunocytochemistry and synaptic relationships of
physiologically characterized HRP-filled neurons, In: Heimer, L., and Zaborszky, L. (eds.),
Neuroanatomical Tract-Tracing Methods 2, New York: Plenum Press, pp. 239–264.
Stepanyants, A., Tamas, G., and Chklovskii, D. B., 2004, Class-specific features of neuronal
wiring, Neuron 43(2):251–259.
Szentágothai, J., 1970, Glomerular synapses, complex synaptic arrangements, and their oper-
ational significance, In: Schmitt, F. O. (ed.), The Neurosciences: Second Study Program, New
York, NY: The Rockefeller University Press, pp. 427–443.
Szentágothai, J., 1978, The neuron network of the cerebral cortex: a functional interpretation,
Proc. R. Soc. Lond. B 201:219–248.
Yuasa, H., and Watanabe, J., 1994, Influence of urethane anesthesia and abdominal surgery on
gastrointestinal motility in rats, Biol. Pharm. Bull. 17:1309–1312.
Zaborszky, L., Csordas, A., Buhl, D., Duque, A., Somogyi, J., and Nadasdy, Z., 2002, Computa-
tional anatomical analysis of the basal forebrain corticopetal system, In: Ascoli, A. (ed.),
Computational Neuroanatomy: Principles and Methods, NJ: Humana Press, pp. 171–197.
Zaborszky, L., and Duque, A., 2000, Local synaptic connections of basal forebrain neurons,
Behav. Brain Res. 115:143–158.
Zaborszky, L., Leranth, C., Makara, G. B., and Palkovits, M., 1975, Quantitative studies on the
supraoptic nucleus in the rat: II. Afferent fiber connections, Exp. Brain Res. 22:525–540.
Zaborszky, L., Palkovits, M., and Flerko, B., 1992, Janos Szentágothai: a life-time adventure with
the brain. An appreciation on his eightieth birthday, J. Comp. Neurol. 326:1–6.
8
Nonradioactive In Situ
Hybridization in Combination
with Tract-Tracing
RUTH L. STORNETTA and
PATRICE G. GUYENET
INTRODUCTION
METHODOLOGICAL CONSIDERATIONS
Riboprobe Design
Production of cDNA Clone
Transformation of cDNA Plasmid into Competent Cells (E. coli)
Production of Linear cDNA Template
In Vitro Transcription of cRNA from Linear DNA Template
Test for Incorporation of Nonradioactive Label Using a Dot Blot
on a Nytran Strip
Tissue Preparation
In Situ Hybridization
Immunocytochemistry for Revealing Digoxigenin and Other
Proteins or Tract-Tracers of Interest
Modifications for Double ISH
Controls
APPLICATIONS
Tract-Tracer Combined with ISH: VGlut1[specific]
mRNA-Containing Neurons in Medulla Project to Cerebellum
Tract-Tracer Combined with ISH and Immunocytochemistry:
Catecholaminergic Neurons in Rat Medulla Oblongata
Containing VGlut2 mRNA Project to Spinal Cord
Tract-Tracer Combined with Double ISH: Neurons Containing
Both Preproenkephalin and VGlut2 mRNAs Project to Phrenic
Motor Nucleus
RUTH L. STORNETTA AND PATRICE G. GUYENET • Department of Pharmacology,

University of Virginia, Charlottesville, VA 22908
237
238 RUTH L. STORNETTA and PATRICE G. GUYENET
Tract-Tracer Combined with ISH and c-Fos: Baroactivated Neurons

in Rostral Ventrolateral Medulla Contain PPE mRNA and
Project to the Spinal Cord
ISH Combined with Juxtacellular Labeling: Barosensitive, Spinally
Projecting Neurons of the Rostral Ventrolateral Medulla
Contain PPE mRNA
Viral Tracing Combined with Double ISH: Some Presympathetic
Neurons of the Rostral Ventrolateral Medulla Contain Markers
of GABA and Glycine
Advantages Over Radioactive Methods or Antibodies
Disadvantages
APPENDIX: DETAILED PROTOCOLS
In Vitro Transcription of cRNA from Linear DNA Template
Test for Incorporation of Nonradioactive Label Using a Dot Blot
on a Nytran Strip
Prehybridization
Hybridization
Rinsing Through Decreasing Salt Concentrations, RNAse A,
and High Stringency Wash
Immunocytochemistry for Revealing Digoxigenin and Other
Modifications for Double ISH
REFERENCES
Abstract: The use of in situ hybridization (ISH) for the detection of mRNAs in cell
bodies has greatly expanded our ability to detect cellular phenotypes in the central
nervous system. Riboprobes have been used in the past to identify neuropeptide
precursors, distribution of receptors, ion channels, and enzymes. More recently,
the discovery of unambiguous markers for the major ionotropic transmitters has
made possible the definitive identification of neurons involved in fast transmis-
sion. The advantages and disadvantages of different types of probes, including
DNA probes, oligonucleotides, and RNA probes for the detection of mRNAs are
described. Although in situ hybridization was pioneered with the use of radioactive
probes, nonradioactive alternatives are now readily available. The relative merits of
nonradioactive probes, specifically for combination with tract-tracing, are discussed.
This chapter focuses on in situ hybridization methods based on nonradioactive ribo-
probes and their use in combination with tract-tracing and immunocytochemistry.
Keywords: c-Fos, clone, digoxigenin, double in situ hybridization, Fluorogold, jux-

tacellular, plasmid, reverse transcription polymerase chain reaction, riboprobe
I. INTRODUCTION
The use of in situ hybridization (ISH) for the detection of mRNAs in cell
bodies has greatly expanded our ability to detect cellular phenotypes in the
central nervous system (CNS). The first descriptions of synthetic-labeled
NONRADIOACTIVE IN SITU HYBRIDIZATION 239
RNA hybridizing to DNA in tissue (Gall and Pardue, 1969; John et al., 1969)
enabled the detection of specific DNA sequences not just on a gel or blot but
in preserved intact specimens of the native tissue. Some of the first studies
to use the ISH technique in brain tissues were performed in the 1980s and
focused on the identification of various neuropeptides (Bloch et al., 1986a,
b; Hoefler et al., 1986; Lanaud et al., 1989; Pochet et al., 1981; Shivers et al.,
1986; Siegel and Young, 1985; Terenghi et al., 1987). Researchers have also
been able to localize the mRNA coding for ion channels (Baldwin et al.,
1991; Brysch et al., 1991; Hwang et al., 1992; Lenz et al., 1994; McKinnon,
1989; Perney et al., 1992; Rudy et al., 1992; Wang et al., 1994), receptors
(Goldman et al., 1986; Malherbe et al., 1990; Rogers et al., 1991; Surmeier
et al., 1992; Wada et al., 1988), and enzymes characteristic of certain neuro-
transmitters (Chesselet et al., 1987; Julien et al., 1987; Mezey, 1989; Seroogy
et al., 1989; Wuenschell et al., 1986). More recently, the discovery of unam-
biguous markers for the major ionotropic transmitters γ-aminobutyric acid
(GABA) (Esclapez et al., 1994), glycine (Poyatos et al., 1997), and glutamate
(Fremeau et al., 2004) has made possible the definitive identification of
neurons involved in fast transmission. RNA detection in neuronal cell bod-
ies is critical for the glutamate vesicular transporters 1 and 2 (VGlut1 and
VGlut2) as well as for the glycine transporter (GlyT), since these proteins
are present only in terminals and are not found in cell soma. GAD-65 and
GAD-67 are also not as readily detected in cell bodies in the brainstem as
the mRNA coding for these substances. Thus, the use of ISH has proved
to be a critical technique for the identification of the cells involved in fast
neurotransmission in the brain (Guyenet et al., 2004).
The cytoplasmic localization of mRNA is an ideal technique when used
in combination with retrograde tract-tracing, since the location and pheno-
typic identification of the projecting neurons is the goal of many tract-tracing
studies. With the recent avalanche of sequence information now available,
DNA templates may be generated for any known sequence and used for
ISH. This technique is much faster and a positive outcome more likely than
the generation of specific antibodies for a protein of interest. ISH also has
the advantage of a somatic localization, unlike many proteins that are not
present in large amounts in the soma, but transported to terminals or as-
sembled into native form in the terminals. Of course, if one is interested in
anterograde projections and determining the phenotype of terminal fields,
ISH will not be useful.
The early studies describing ISH often relied on DNA probes. These had
several difficulties including more labor-intensive cloning techniques. The
probe itself was double-stranded and therefore could reanneal after de-
naturing and reduce the available hybridization sites (Lewis and Baldino,
1990). The engineering of a convenient vector incorporating the bacterio-
phage promoter next to a multiple cloning site featuring common restric-
tion endonucleases allowed for the creation of a DNA template for the in
vitro synthesis of RNA probes (Melton et al., 1984). The RNA probe has
several advantages. It is easier to procure the clone for the DNA template.
The RNA probe is single-stranded and thus will not hybridize to itself. The
RNA–RNA hybrid formed in the tissue is stronger than a DNA–RNA hybrid

and can withstand more stringent rinsing, resulting in lower background.
The RNA–RNA hybrid will also resist the action of RNAse, another treat-
ment that will substantially lower background. The use of oligonucleotide
probes (Lewis et al., 1985) is a further development in bringing the tech-
nique of ISH into laboratories with less familiarity with molecular biological
techniques. Synthetic oligonucleotides are widely available, relatively inex-
pensive, and easy to label with either radioactive or nonradioactive methods.
Radioactive-labeled oligonucleotides have been used successfully for many
of the same mRNAs as the longer riboprobe counterparts including recep-
tors and channels (Brysch et al., 1991; Pelletier et al., 1988; Wisden et al.,
1988). The major drawbacks with oligonucleotide probes are their relative
lack of sensitivity for messages expressed at lower levels as well as the possi-
bility for nonspecificity if the detected sequence has many identical regions
to another sequence or splice variant.
Although ISH was pioneered with the use of radioactive probes, nonra-
dioactive alternatives are now readily available. The caveat for nonradioac-
tive probes is that they are generally less sensitive due to the lower incor-
poration of the digoxigenin-labeled nucleotide. However, this is not always
the case, and in direct comparisons of some riboprobes, nonradioactive
probes were reported to be equally sensitive to their radioactive counter-
parts (Clavel et al., 1991; Kreft et al., 1996; Mitchell et al., 1993; Park et al.,
1991). Low expression levels are not a problem for the neuropeptide precur-
sors, the cytoplasmic enzymes, or the vesicular transporters; however, it is a
problem for messages that are expressed at a lower level, e.g., most of the re-
ceptors and ion channels. Some of this difficulty may be overcome by making
the template longer (i.e., more base pairs), thus offering a greater amount
of potential incorporation sites for the labeled nucleotide. On the positive
side, the resolution of the nonradioactive method is higher since the signal
is expressed directly within the cytoplasm and not as silver grains in an emul-
sion media overlying the cell. However, the problem of signal-to-noise ratio
for low signal level must be realized. Also, if one is interested in quantitative
analysis of message levels, the radioactive method is essential. However, for
neuroanatomical studies, nonradioactive riboprobes and oligonucleotides
can easily be combined with more traditional immunocytochemical meth-
ods for the detection of proteins as well as tract-tracing to discover connec-
tions of cells with specific phenotypes within the CNS ( Johnson et al., 2002;
Stornetta et al., 1999, 2002, 2003; Stornetta and Guyenet, 1999). The use
of well-designed cDNA clones also provides a reliable source of material,
free from the inconsistencies, availability, and specificity issues of polyclonal
antibodies.
This chapter focuses on ISH methods based on nonradioactive riboprobes
and their use in combination with tract-tracing and immunocytochemistry.
For methods for oligonucleotides and/or radiolabeled probes, there
are many other excellent references (see Chesselet, 1990; Darby, 2000;
Valentino et al., 1987; Wilkinson, 1998; Wisden and Morris, 1994; Young,
1990).
II. METHODOLOGICAL CONSIDERATIONS
A. Riboprobe Design
While it may be easy to obtain clones from colleagues, it is also relatively

easy to design and produce a cDNA clone for a particular sequence of inter-
est. This is often faster than waiting for the clone from a busy researcher, who
may have problems shipping the clone overseas and has the distinct advan-
tage of not having to sign material transfer agreements or involvement with
other obligations. The NCBI Web site (http://www.ncbi.nlm.nih.gov/) with
access to many different genetic databases is publicly available. Once the
sequence of interest is found, the portion least likely to overlap with other
closely related sequences can be determined by using the BLAST search
(also found on the NCBI Web site) and noting where similar sequences
align. Choose the least similar portion of the sequence for the cDNA tem-
plate. We have had success with templates up to 3.3 kb in length.
B. Production of cDNA Clone
Design primers for reverse transcription polymerase chain reaction (RT-

PCR) of RNA. Primer design is often available online from companies that
offer primer synthesis. We have used the PrimerQuest tool from Integrated
DNA Technologies (http://www.idtdna.com/SciTools/SciTools.aspx) by
Steve Rozen and Helen Skaletsky with code available at http://www.genome.
wi.mit.edu/genome software/other/primer3.html. To create the template,
use polyA+ selected RNA (in our case, the RNA is from whole brain or from
medulla oblongata). Kits to extract oligo-dT isolated RNA are commercially
available from many companies. We currently use a kit from Invitrogen
(Carlsbad, CA; FastTrack 2.0 for isolation of mRNA from 0.4–1 g of fresh
tissue). Prepared RNA is also commercially available. The next step is the
RT-PCR to extract the particular sequence of interest from the sample RNA.
We have used the Titan One Tube RT-PCR kit from Roche Applied Science
(Indianapolis, IN) according to their directions with very good success. A
single band of DNA of the correct length is necessary for use as a good
template (Fig. 8.1, Step 1d). Performing a melting curve experiment on the
RT-PCR by varying the annealing temperature of the PCR reaction will help
in achieving this goal. Once the PCR product is obtained, subclone it into
a vector usable for RNA in vitro transcription. We have had very good luck
with the pCRII-TOPO vector from Invitrogen following the manufacturer’s
directions (Fig. 8.1, “RT-PCR,” Steps 1a–d).
C. Transformation of cDNA Plasmid into Competent Cells (E. coli)
Whether the cDNA plasmid is obtained from outside the laboratory or

within the laboratory as detailed in the previous steps, it must be transformed
Figure 8.1. Overview of fabrication of DNA template and digoxigenin-labeled

riboprobe. Step 1a: Annealing the DNA oligo that is 5 to 3 to the mRNA from a
sample tissue extract of RNAs and creating a cDNA “copy” of the mRNA sequence of
interest catalyzed by the reverse transcriptase enzyme. The RNA–cDNA hybrid melts
(becomes single-stranded) by raising the temperature in the thermal cycler (PCR
machine). Step 1b: Annealing the DNA oligo that is 3 to 5 to the cDNA (instead of
the mRNA) and the copying of the cDNA to double-stranded DNA catalyzed by Taq
polymerase. Once the copy is made, the cDNA–cDNA hybrid melts by raising the
temperature of the PCR machine. Step 1c: The second annealing step is repeated 25–
35 times to generate exponential quantities of cDNA. Step 1d: Agarose gel of DNA
from PCR reaction— a single band of the correct length is produced. Step 2a: Taq
polymerase adds a 3 A overhang to the cDNA it copies. This property of the cDNA
is exploited by the TOPO vector. Step 2b: The PCR product is subcloned into the
into competent cells and prepared in a reasonable quantity for further
manipulation. This can be achieved with several commercially available kits.
We currently use the One Shot Top 10 F’ chemically competent E. coli from
Invitrogen and the Wizard Plus Midipreps DNA purification system from
Promega (Madison, WI) according to the manufacturer’s instructions. Af-
ter larger scale preparation of the DNA, the sequence should be verified
before continuing, particularly in reference to the orientation of the clone
in the vector. The concentration of the DNA can be determined with a
spectrophotometer (Fig. 8.1, Steps 3a–c).
D. Production of Linear cDNA Template
Vectors useful for in vitro RNA transcription have multiple restriction

sites on either end of the cloning site of the sequence of interest. Choose
a restriction enzyme that cuts only once at the end of the sequence of in-
terest (usually in the multiple cloning site) at the opposite end from the
desired promoter site. The enzyme should leave a 5 overhang or blunt end
(Fig. 8.1, Step 2a). (Note: Enzymes that leave a 3 overhang will result in
the production of double-stranded RNA molecules, drastically reducing the
yield of the in vitro transcription reaction.) The pCRII-TOPO vector has
promoters on either side of the cloning site, thus allowing the production
of either sense template or antisense template. We normally set up a large
number of restriction enzyme reactions with about 40 µg of DNA total. Our
experience is that the restriction enzymes work better in smaller volumes
(about 20–30 µl per reaction). It is absolutely critical that the restriction
←
Figure 8.1. (Cont.) TOPO vector (antibiotic resistance areas of the TOPO vector are
indicated as “Kan” for kanamycin and “Amp” for ampicillin). The resulting circular
piece of DNA (plasmid) is then mixed with chemically competent E. coli, incubated
on ice for 5–30 min, the reaction heated to 42◦ C for 30 s, and then returned to the
ice. Step 2c: The resulting transformation (the plasmid that is now incorporated into
the E. coli) is spread on plates (previously prepared with a mix of antibiotic, agar, and
appropriate growth medium). Step 3a: After sitting in a 37◦ C oven overnight, the
plates will have small white dots (colonies of bacterial clones). Individual colonies are
picked with a sterile toothpick or wire loop and placed into an aliquot of liquid growth
media containing appropriate antibiotic. Step 3b: Screen a sample of the colonies
to determine whether the colony contains the correct plasmid before growing the
colony in large quantities. Step 3c: The “plasmid prep” is the procedure for growing
large quantities of the correct bacterial colonies, and then releasing the plasmids
from the bacteria and purifying the plasmid DNA. After this process, the plasmid
sequence must be verified. Step 4a: The correct purified plasmid DNA is linearized
with an appropriate endonuclease (restriction enzyme). This linearized DNA serves
as the template and is concentrated with ethanol and checked on an agarose gel
(L = ladder, U = uncut DNA plasmid, C = linear DNA template). Step 4b: The
linear DNA is then transcribed into digoxigenin-labeled cRNA (riboprobe) in vitro
using the appropriate RNA polymerase in a solution containing digoxigenin-labeled
UTP.
enzyme cuts the DNA to completion. Any traces of circular plasmid DNA
remaining will carry over to the in vitro transcription reaction, resulting
in long stretches of noncoding plasmid sequence transcription and incor-
poration of much of the labeled nucleotide into this “garbage” sequence.
Check the completion of the reaction by gel electrophoresis of 1 µl from
each enzyme reaction as well as 1 µl of uncut (“supercoiled”) plasmid DNA
(Fig. 8.1, Step 4a). The uncut DNA will run at different lengths than the cut
DNA. There should be only one clear band per restriction enzyme reaction,
with no bands appearing like the uncut DNA. Combine all successful restric-
tion reactions into one tube and perform a phenol–chloroform extraction
to eliminate the enzyme and any other impurities from the now-linearized
template. Concentrate the DNA template by ethanol precipitation. Measure
the concentration by spectrophotometer (Fig. 8.1, Step 4a).
E. In Vitro Transcription of cRNA from Linear DNA Template
Assemble the components of the reaction mixture (with the exception

of the enzyme) at room temperature and in the stated order to prevent
the precipitation of DNA template by spermidine in the reaction buffer.
Be aware of keeping everything as clean and RNAse-free as possible— use
gloves and sterile tips and do the reaction assembly on a clean surface. Mix
ingredients thoroughly in a sterile plastic Eppendorf tube by pipetting up
and down after each addition (see Appendix section “In Vitro Transcription
of cRNA from Linear DNA Template” for detailed recipes).
F. Test for Incorporation of Nonradioactive Label Using

a Dot Blot on a Nytran Strip
This step is necessary to determine whether the riboprobe has been la-
beled with digoxigenin and also to determine the relative amount of digox-
igenin that has been incorporated (see Appendix section “Test for Incor-
poration of Nonradioactive Label Using a Dot Blot on a Nytran Strip” for
further details).
G. Tissue Preparation
Anesthetize rats with pentobarbital and perfuse transcardially with 100 ml

phosphate-buffered saline (PBS; pH 7.4) followed by 500 ml of 4%
paraformaldehyde in 0.1 M phosphate buffer (pH 7.4) at room temper-
ature. Extract brains and postfix in the same fixative for up to 3 days at 4◦ C
(postfixation does not appear to affect ISH; however, take care for postfix-
ation time for any additional antibodies used). Section brains at 30 µm on
a vibrating microtome at room temperature in 0.1 M phosphate buffer.
(Brains may also be sunk in 25% RNAse-free sucrose and cut frozen.)
Collect sections in RNAse-free cryoprotectant [50 mM sterile phosphate
buffer, 30% ethylene glycol (Sigma-Aldrich, St. Louis, MO), 20% glycerol
(RNAse-free; Sigma-Aldrich)]in sterile 24-well tissue culture plates. Sections
may be kept in this solution at −20◦ C for up to 1 year.
H. In Situ Hybridization
1. Prehybridization
This step allows the tissues to adapt to the conditions of the hybridization
buffer as well as serving as an important blocking step to prevent or at least
decrease nonspecific hybrid formation. It is extremely important to make
the prehybridization mixture with sterile, RNAse-free solutions, and with
sterile dishes, pipette tips, etc. This is the time to be paranoid about RNAse!
(see Appendix section “Prehybridization” for details).
2. Hybridization
In this step, the riboprobe is added directly to the “prehybridization”

solution, and the conditions are optimized for hybrid formation. See Fig. 8.2
for a summary and Appendix section “Hybridization” for further details.
Figure 8.2. Overview of visualization of digoxigenin-labeled riboprobe.

3. Rinsing
The rinsing of the tissue after the hybrids are formed is necessary to
decrease the background (nonspecific hybrids) as well as to destroy any ri-
boprobe that is not hybridized. The basic idea is to take the tissue through
solutions of decreasing salt concentration as the RNA–RNA hybrid is sen-
sitive to salt concentration. Weaker bonds (i.e., nonspecific hybrids) will
not survive the lower salt concentrations. One of the rinsing steps involves
treatment with RNAse A, an enzyme that will destroy single-stranded RNA
(e.g., nonhybridized riboprobe). The final step is a high-temperature rinse
in the lowest concentration of salt solution, a condition in which only the
strongest, most specific hybrids will survive. For a summary, see Fig. 8.2,
and for details, see Appendix section “Rinsing Through Decreasing Salt
Concentrations, RNAse A, and High Stringency Wash.”
I. Immunocytochemistry for Revealing Digoxigenin and Other

After the hybrids are formed and the tissue is rinsed, the hybrids are
stable and proteins of interest may be revealed by using standard immuno-
cytochemical protocols. The digoxigenin label as well as the Fluorogold
(FG) tracer is revealed in this manner. See Fig. 8.2 for a summary and see
Appendix section “Immunocytochemistry for Revealing Digoxigenin and
Other Proteins or Tract-Tracers of Interest” for details.
J. Modifications for Double ISH
A second nonradioactive cRNA with a different sequence of interest can

be transcribed, substituting FITC-12-UTP (Roche) for the digoxigenin-11-
UTP (see Appendix protocol “In Vitro Transcription of cRNA from Linear
DNA Template” for further details). To test the FITC-labeled riboprobe on
the dot blot, substitute sheep anti-FITC peroxidase-tagged antibody (1:2000;
Roche) for the sheep alkaline phosphatase (AP) tagged anti-digoxigenin
antibody. For further details, see Appendix protocol G.1; for further details
about visualizing the FITC probe for ISH, see Appendix protocol G.2.
K. Controls
1. Hybridization with the Sense Strand of cRNA
After in vitro transcription of the sense strand of cRNA, compare the

incorporation of digoxigenin of both sense and antisense cRNAs by relative
4V 4V
sense antisense
Figure 8.3. Sense/antisense control with cRNA for vesicular glutamate transporter-2
in medial vestibular nucleus of rat medulla oblongata.
appearance on the dot blot. Use a concentration about 2–3×, the equivalent
amount of antisense cRNA, and do “side by side” ISH on tissue from the
same brains with sense and antisense riboprobes. Allow the colorization
procedure to progress for the same amount of time. There should be very
strong signal in appropriate places for the antisense riboprobe-hybridized
tissues and no signal in the sense-hybridized tissues (see Fig. 8.3). This is the
standard control for specific hybridization signal.
2. Hybridization Signal in Expected Areas but No Signal Where No

mRNA Should Be Found
We feel that the neuroanatomical consistency of the signal in areas that

are known to express the mRNA versus no signal in areas where no mRNA is
present is the best control (see Fig. 8.4). This is difficult in cases where the
mRNA is ubiquitously or very broadly expressed. In these cases, one must
rely on the sense/antisense control data. The internal consistency of the
mRNA being expressed in certain types of neurons is also helpful (e.g., in
the case of the VGlut2, we never saw the mRNA expressed in GABAergic
neurons). One of the standards we have used for many mRNAs of interest is
the lack of expression of mRNA in motor neurons (the larger motor neurons
are notorious for expressing artifactual background in immunohistological
procedures).
III. APPLICATIONS
A. Tract-Tracer Combined with ISH: VGlut1[specific]

mRNA-Containing Neurons in Medulla Project to Cerebellum
After noting the striking distribution of VGlut1 mRNA in medulla oblon-

gata in precerebellar nuclei and the fact that VGlut1 protein is expressed in
Figure 8.4. Discrete distribution of VGlut1 mRNA in rat brain and spinal cord. (A)
Clarke’s column of dorsal horn of spinal cord. (B) Lateral reticular nucleus (IO,
inferior olive). (C) Lateral reticular nucleus n
[ ote the lack of label in the IO and
the pyramidal tract (py)]. (D) External cuneate nucleus (ECu) (sp5, spinal trigem-
inal tract; cu, cuneate fasciculus). (E) Spinal trigeminal tract and nucleus X. (F)
Ectotrigeminal nucleus (E5) and paratrigeminal nucleus. (G) Linear nucleus (Lin).
(H) Medial brainstem at Bregma level −12.80 mm (mlf, medial longitudinal fasci-
culus). (I) Prepositus nucleus (Pr) (Sol, nucleus of the solitary tract). ( J) Ventral
cochlear nucleus (VC) (8n, eighth nerve). (K) Pontine nucleus (Pn) (lfp, longitudi-
nal fasciculus pons). (L) Mesencephalic trigeminal nucleus (Me5) (periaqueductal
gray, PAG). Scale bar: 100 µm for A, D, E, G, J, and L. Scale bar:200 µm for B, C, F,
H, I, and K.
mossy fibers in cerebellum (Bellocchio et al., 1998; Hisano et al., 2002),

we tested the hypothesis that most cerebellar-projecting neurons in rat
brainstem contain VGlut1 mRNA. Four 100-nl pressure injections of the
retrograde marker FG (2% in sterile saline; Fluorochrome Inc., Englewood,
CO; Schmued and Fallon, 1986) were placed 1–2 mm deep in various
Figure 8.5. Examples of colocalization of VGlut1 mRNA and Fluorogold (FG) (from
cerebellum). (A) VGlut1 mRNA in linear nucleus (brightfield). (B) FG immunore-
activity (Cy-3 revealed by fluorescence; same field as in A). Note that most of the
cells are double labeled. (C) VGlut1 mRNA in pontine nucleus (brightfield). (D) FG
immunoreactivity (Cy-3 revealed by fluorescence; same field as in C). Arrows point
to a few of the many double-labeled cells. (E) and (G) VGlut2 mRNA in inferior olive
(brightfield). (F) and (H) FG immunoreactivity (Cy-3 revealed by fluorescence; same
field as in E). Arrows point to some of the double-labeled cells. Scale bar: 50 µm for
A, B, E, and F. Scale bar: 20 µm for C, D, G, and H.
locations of the left cerebellar cortex. The brainstem was processed for
ISH for VGlut1 mRNA in combination with FG immunocytochemistry. [Im-
munocytochemical detection of FG was accomplished by incubating the
tissue with a rabbit anti-FG antibody (1:10,000; Chemicon, Temeluca, CA),
followed by a biotinylated anti-rabbit IgG (1:200; Vector, Burlingame, CA)
and visualized with streptavidin Cy3 (1:1000; Molecular Probes, Eugene,
OR)]. We found that the vast majority of FG-labeled neurons in pons and
medulla (with the exception of the inferior olive) contained VGlut1 mRNA
(Fig. 8.5).
B. Tract-Tracer Combined with ISH and Immunocytochemistry:

Catecholaminergic Neurons in Rat Medulla Oblongata
Containing VGlut2 mRNA Project to Spinal Cord
Fluorogold (2–3% in sterile saline) was pressure injected into the vicin-
ity of the intermediolateral cell column (1 mm below the entry point of
the dorsal roots) at the first and third thoracic segments bilaterally (50-nl
injections, four injections per rat). The brainstem was processed for ISH
with VGlut2 riboprobe in combination with antibodies for tyrosine hydrox-
ylase (mouse monoclonal; Sigma) and FG (described above). We found
many neurons immunoreactive for tyrosine hydroxylase that also contained
VGlut2 mRNA and FG. Some of these neurons are illustrated in Fig. 8.6.
This was the first study to demonstrate that brainstem presympathetic cate-
cholaminergic neurons are glutamatergic.
Figure 8.6. Rostral ventrolateral medulla (RVLM) catecholaminergic and noncate-

cholaminergic neurons with projection to the thoracic spinal cord contain VGlut2
mRNA. (A–C) RVLM neurons in the same field photographed under brightfield in
A show VGlut2 mRNA AP reaction product and under fluorescent light in B and C
show FG immunoreactivity (ir) revealed with Cy-3 (B) and tyrosine hydroxylase (TH)
ir revealed with Alexa 488 (C). Asterisks indicate two TH-ir bulbospinal cells that
contain VGlut2 mRNA. The arrow points to a strongly VGlut2-positive bulbospinal
cell that is devoid of TH-ir. Scale bar: 50 µm. (From Stornetta et al. 2002.)
C. Tract-Tracer Combined with Double ISH: Neurons Containing

Both Preproenkephalin and VGlut2 mRNAs Project to
Phrenic Motor Nucleus
FluoroGold was injected iontophoretically into the left ventral horn of

spinal segment C4 after recording respiratory activity at the injection site.
The rat medulla oblongata was processed for ISH with a VGlut2 digoxigenin-
tagged riboprobe and a preproenkephalin (PPE) FITC-tagged riboprobe in
combination with antibodies for FG (described above). Many bulbospinal
neurons of the rostral ventral respiratory group in the caudal medulla ob-
longata contained both VGlut2 and PPE mRNAs. Examples of these neurons
as well as the injection site are shown in Fig. 8.7. This study demonstrated
that glutamatergic neurons controlling respiratory output could also use
enkephalin as a neurotransmitter.
D. Tract-Tracer Combined with ISH and c-Fos: Baroactivated

Neurons in Rostral Ventrolateral Medulla Contain PPE mRNA and
Project to the Spinal Cord
FluoroGold was pressure injected into the first and third segments of tho-
racic spinal cord. Rats were cannulated for chronic blood pressure record-
ing and intravenous injections. Injections of hydralazine into conscious,
freely moving rats lowered blood pressure, and after 2 h animals were
anesthetized and perfused. The brainstem was processed for ISH for PPE
mRNA (detected by a digoxigenin-labeled riboprobe) in combination with
antibodies for FG (visualized with Alexa 488) and c-Fos (goat polyclonal,
Santa Cruz, visualized with streptavidin Cy3). Many bulbospinal neurons
were c-Fos positive (baroactivated) and contained PPE mRNA. Examples of
these neurons are shown in Fig. 8.8. This study demonstrates that c-Fos, a
Figure 8.7. Colocalization of VGlut2 mRNA and PPE mRNA in rostral ventral respira-
tory group (rVRG) bulbospinal neurons. Cluster of rVRG neurons (arrows) contain-
ing (A) VGlut2 mRNA (BCIP/NBT reaction product; brightfield), (B) immunore-
activity to Fluorogold (Alexa 488, epifluorescence), and (C) PPE mRNA (Cy3; epi-
fluorescence). Scale bar: 20 µm. (D) Fluorogold injection site into fourth cervical
spinal cord segment. Composite photomicrograph of center of iontophoretic de-
posit. Scale bar: 500 µm. (From Stornetta et al., 2003.)
useful marker of cell activation, can be colocalized with mRNAs of inter-

est as well as with tract-tracers. In this particular example, this was the first
demonstration of baroactivated presympathetic enkephalinergic neurons in
brainstem.
Figure 8.8. Preproenkephalin (PPE), c-Fos immunoreactive (ir) neurons in RVLM

project to spinal cord. Cluster of RVLM neurons (arrows) containing (A) Fos-ir (Cy3;
epifluorescence), (B) Fluorogold (FG)-ir (Alexa 488, epifluorescence), and (C) PPE
mRNA (digoxigenin-labeled probe, brightfield). Scale bar: 50 µm. (From Stornetta
et al., 2001.)
A B
biotinamide PPE
Figure 8.9. Blood pressure–sensitive neuron in RVLM antidromically activated from

thoracic spinal cord contains PPE mRNA. (A) Neuron in RVLM recorded in vivo
and juxtacellularly labeled with biotinamide (Alexa 488, epifluorescence). (B) Same
neuron showing PPE mRNA (digoxigenin-labeled probe, brightfield). Scale bar:
50 µm. (From Stornetta et al., 2001.)
E. ISH Combined with Juxtacellular Labeling: Barosensitive,

Spinally Projecting Neurons of the Rostral Ventrolateral
Medulla Contain PPE mRNA
Barosensitive vasomotor presympathetic cells were recorded in rostral

ventrolateral medulla in vivo in halothane-anesthetized rats. Cells were la-
beled using the juxtacellular method (described in the chapter by Duque
and Zaborszky, this volume) After transcardial perfusion, the brainstem was
sectioned and processed for ISH and PPE mRNA (detected by a digoxigenin-
labeled riboprobe) in combination with streptavidin Alexa 488 (Molecular
Probes) for the detection of the biotinamide-labeled cell. An example of one
of these PPE mRNA positive, spinally projecting (detected by antidromic ac-
tivation from the spinal cord) barosensitive neurons is shown in Fig. 8.9. The
ability to find the mRNA of physiologically identified neurons has allowed a
major breakthrough in our ability to perform functional neuroanatomy. In
this particular example, this was the first demonstration of enkephalinergic
presympathetic neurons with firing patterns inversely correlated with blood
pressure.
F. Viral Tracing Combined with Double ISH: Some Presympathetic

Neurons of the Rostral Ventrolateral Medulla Contain Markers
of GABA and Glycine
Rats were anesthetized and the adrenal gland was exposed and injected
with pseudorabies virus (PRV; provided by L. Enquist, Princeton University,
NJ). This technique is described in further detail in the chapter by Geerling
et al. (this volume). After 3 days, animals were anesthetized and perfused
transcardially. The rat medulla oblongata was processed for ISH with a GAD-
67 digoxigenin-tagged riboprobe and a GlyT2 FITC-tagged riboprobe in
A B C
PRV GAD-67 GlyT2
Figure 8.10. Presympathetic neuron in gigantocellular reticular nucleus ventral con-

taining (A) pseudorabies virus (PRV)-ir (Alexa 488, epifluorescence), (B) GAD-
67 mRNA (digoxigenin-labeled probe, brightfield), and (C) glycine transporter-2
(GlyT2) mRNA (FITC-labeled probe, Cy3, epifluorescence). Scale bar: 50 µm. (From
Stornetta et al., 2004.)
combination with antibodies for PRV (rabbit polyclonal, Enquist). An ex-

ample of a neuron infected with PRV (resulting from injection in the adrenal
gland) in an area of the gigantocellular nucleus in medulla that also contains
GAD-67 and GlyT2 mRNAs is seen in Fig. 8.10. This demonstrates that some
presympathetic neurons have the capacity for inhibitory control of sympa-
thetic outflow. A major caveat with detection of mRNAs in cells infected with
viruses is that viruses corrupt the RNA processing machinery and the native
mRNAs are replaced by viral RNAs; thus, allowing only detection of native
mRNA at very early time points in the cell’s infection with virus (this issue
is discussed in more detail in the chapter of Geerling et al., this volume).
IV. ADVANTAGES AND LIMITATIONS
A. Advantages Over Radioactive Methods or Antibodies
1. Detection of nonradioactive riboprobes can be combined with the de-

tection of common retrograde tract-tracers as well as many commer-
cially available antibodies. Nonradioactive probes are particularly well
suited for combination with multiple fluorescent tags.
2. Nonradioactive probes can be combined with other techniques such
as pseudorabies viral tracing (Stornetta et al., 2004) and juxtacellular
labeling (Schreihofer et al., 1999; Stornetta et al., 1999, 2002, 2003).
3. Nonradioactive ISH reveals the transcripts in cell bodies making pos-
sible detection of substances normally transported to or solely present
in terminals. This is essential for the identification of a particular cell
phenotype in combination with retrograde tracers.
4. Nonradioactive ISH does not require special licensing/handling for
radioactive materials.
5. Nonradioactive ISH is much faster than radioactive ISH and relative
signal-to-noise ratio is easier to control.
6. Nonradioactive riboprobes can be produced in large amounts and kept
frozen for at least 1 year before use.
7. Riboprobes can be generated for any known sequence fairly quickly

and easily (within 1–2 weeks) compared to several months for antibody
production.
8. Riboprobe specificity can be tested with sense controls as well as tissue
specificity and can also be subjected to Northern blot analysis.
B. Disadvantages
1. Fluorescent retrograde tracers such as FG fade when exposed to ISH

conditions and must be amplified with antibodies for best visualization
after ISH.
2. Incorporation of nonradioactive label is less than for radioactive ribo-
probes and thus detection of low-level messages is more difficult and
sometimes impossible with nonradioactive ISH.
3. Attention must be paid to sterile conditions as RNAse can wreak havoc
with the procedure.
4. Extra equipment and reagents for molecular biology procedures, in-
cluding centrifuges, gel electrophoretic apparatus, water baths, and
enzymes, must be acquired/borrowed .
5. The protocol requires a few extra days as well as many extra steps over
standard immunocytochemical methods.
6. The best available antibody to digoxigenin is AP-tagged and the re-
action product of nitro blue tetrazolium (NBT)/5-bromo-4-chloro-
3-indolyl phosphate, toluidine salt (BCIP) cannot be dehydrated or
covered with “permanent” mounting media. The lack of dehydration
results in the tissue looking a bit less clear than with peroxidase reac-
tion products that can be dehydrated. This can be overcome by using a
riboprobe transcribed with FITC-12-UTP (Roche) and visualized with
a sheep anti-FITC antibody tagged with peroxidase and reacted with a
traditional peroxidase substrate.
APPENDIX: DETAILED PROTOCOLS
A. In Vitro Transcription of cRNA from Linear DNA Template
A.1. In vitro transcription reaction mixture

Sterile H2 O (q.s. to 90 µl total volume)
1.5 µg DNA template (volume will depend on concentration)
9 µl 100 mM DTT (provided with enzyme)
18 µl ribonucleotides−UTP (ATP, CTP, GTP mix equal parts of 10 mM
stock; Promega)
10.8 µl 2 mM UTP (we have tried varying the concentration of the unla-
beled UTP— this concentration gives us the best results. Note that some
unlabeled UTP is necessary for the reaction to work)
2.1 µl digoxigenin-11-UTP (Roche)
18 µl 5× transcription buffer
3 µl RNA polymerase (T3, T7, or SP6; Promega)
A.2. Incubate reaction in a 500 µl Eppendorf tube in a water bath for
2 h. For T3 or T7, incubate at 37◦ C. For SP6, incubate at 40◦ C.
A.3. Add 1.5 µl RQ1 DNase (Promega) and incubate for another 20 min
at 37◦ C. (This will destroy the DNA template.)
A.4. Purify the probe with ProbeQuant G-50 micro columns (Amersham
Biosciences, Piscataway, NJ) according to the manufacturer’s instructions
or by phenol–chloroform extraction and ethanol precipitation (we use the
micro columns).
B. Test for Incorporation of Nonradioactive Label Using

a Dot Blot on a Nytran Strip
B.1. Use a thin Nytran strip (Nytran Supercharge, SPC, pore size of
0.45 µm; Schleicher and Schuell, Keene, NH) razor cut (∼0.5 cm ×6 cm)
to fit in a 5-ml plastic test tube.
B.2. Wet the strip by immersion in 100% ethanol for 3 min.
B.3. Rinse in 2X saline-sodium citrate (SSC) (20×SSC stock: 3.0 M NaCl
and 0.3 M sodium citrate in water, pH 7.0) for 3 min and air-dry on a piece
of filter paper for about 10 min.
B.4. Spot 1-µl aliquots of dilutions of control digoxigenin-labeled RNA
(Roche) along with 1-µl aliquots of dilutions of the probe to be tested on
the strip (we use 1/100, 1/200, and 1/400).
B.5. Cross-link the RNA to the Nytran membrane with UV light [we
use the Stratagene (La Jolla, CA) UV Stratalinker 1800 set on auto
cross-link].
B.6. Immerse the strip in 2–3 ml 2×SSC in the test tube for 3-min
shaking.
B.7. Immerse in 2–3 ml phosphate buffer-bovine serum albumin-triton
(PBT) [0.1% bovine serum albumin (BSA)/0.2% Triton X-100 in PBS]and
incubate for 15 min on shaker.
B.8. Incubate in anti-digoxigenin tagged with AP Fab fragments from
sheep (1:1000; Roche) in PBT for 30 min on shaker.
B.9. Rinse 3 × 5 min in PBS.
B.10. Immerse in NMT (0.1 M NaCl/50 mM MgCl2 in 0.1 M Tris, pH 9.5)
for 5-min shaking.
B.11. Immerse in solution of NBT (4.5 µl/ml of NMT) and BCIP
(3.5 µl/ml of NMT; NBT and BCIP from Roche). Spots should appear
momentarily.
B.12. Stop the reaction by rinsing in TE 8.5 (Tris-EDTA). Use the
color intensity relative to the control RNA to determine approxi-
mate concentration of probe. We find that, generally, the in vitro
translated riboprobe should be about equal to the control RNA for best
signal.
C. Prehybridization
C.1. Rinse sections in sterile PBS in sterile petri dishes. We use glass
rods fashioned by flaming glass pipettes to seal the ends and make a
hook for transferring sections between solutions. Rinse the glass hooks with
RNaseZap (Ambion Inc., Austin, TX) prior to use.
C.2. Transfer sections to prehybridization mixture (300 µl per well in a
sterile 24-well tissue culture dish). Four to eight sections will fit per well,
depending on the size of the tissue. Incubate sections in this solution for
30–60 min shaking at room temperature and then at 37◦ C for 1 h.
C.3. Prehybridization mixture (in sterile H2 O)
0.60 M NaCl
0.10 M Tris-Cl (7.5)
0.01 M EDTA
0.05% sodium pyrophosphate
5% (w/v) dextran sulfate (must vortex and heat to 37◦ C to dissolve)
0.50 mg/ml yeast total RNA (Sigma R-7125)
0.05 mg/ml yeast tRNA (Roche 109495)
1×Denhardt’s BSA [50X Denhardt’s solution: 5 g Ficoll-70 (Sigma),
5 g polyvinylpyrrolidone, 5 g BSA (Fraction V; Sigma) q.s. with H2 O to
500 ml; may be kept in frozen aliquots]
50% deionized formamide (Sigma)
0.05 mg/ml poly A (Sigma P-9403)
10 µm of the four nucleoside triphosphates
10 mM dithiothreitol (DTT; Promega)
Make up prehybridization solution in larger batches and freeze in
appropriate-size aliquots. Add 0.5 mg/ml herring sperm (Sigma D-6898-
sodium salt of ribonucleic acids from herring testes) that has been boiled
for 10 min to denature and quenched on ice just prior to use.
D. Hybridization
D.1. Either add riboprobe directly to wells or transfer the sections into a
new solution of prehybridization mixture to which the riboprobe has been
added. We found concentrations of 1–3 µl of riboprobe per well (resulting
in 50–100 pg/µl) to be most effective for a good signal-to-noise ratio.
D.2. Incubate the sections with riboprobe at room temperature on the
shaker for 15 min.
D.3. Incubate at 55–60◦ C overnight (shaking not necessary). The tissue
culture dish cover comes in handy for this— we have never had to worry
about any extra precautions to stop evaporation.
E. Rinsing Through Decreasing Salt Concentrations, RNAse A,
and High Stringency Wash
E.1. Transfer the sections to a mesh-well bottom dish (Nason Fabrications,

Fort Bragg, CA) filled with 4 × SSC/10 mM sodium thiosulfate (NaTS).
E.2. Rinse 2 × 20 min @ 37 ◦ C in this solution.
E.3. Using gloves and a work pad, change solution to 20 µg/ml RNAse A
(Sigma) in RNAse buffer [0.5 M NaCl, 10 mM Tris (pH 8.0) 1 mM EDTA].
E.4. Incubate @37 ◦ C for 30 min.
E.5. Change solution to RNAse buffer and incubate @ 37 ◦ C for 20 min.
E.6. Transfer to solution of 2 × SSC/10 mM NaTS @ 37 ◦ C for 20 min.
◦
E.7. Transfer to a solution of 0.5 × SSC @ 37 C for 20 min.
E.8. Do a final “high-stringency” rinse in 0.1 × SSC at 50–55◦ C for 30–
60 min. If background is a problem, this high-temperature rinse may be
increased in temperature up to 60◦ C.
F. Immunocytochemistry for Revealing Digoxigenin and Other

All rinses and incubations done at room temperature on shaker unless

noted otherwise.
F.1. Rinse sections 3 × 5 min in TBS [0.1 M Tris (pH 7.4)/0.15 M NaCl].
F.2. Incubate 30 min in 10% normal horse serum/0.1% Triton X-100 in
TBS.
F.3. Incubate in sheep anti-digoxigenin tagged with AP (Roche) at 1:1000
in 10% normal horse serum/0.1% Triton X-100 in TBS first for 1 h at room
temperature on the shaker and then overnight at 4◦ C shaking. Centrifuge
the digoxigenin antibody prior to use, and use only the supernatant portion.
Note: In our hands, the only antibody against digoxigenin that works well
is the sheep AP-tagged antibody from Roche.
F.4. Add other antibodies of interest to this same mixture. Fluorescent
tract-tracers such as FG (Fluorochrome Inc., Denver, CO) should be ampli-
fied by using an antibody against FG (e.g., rabbit anti-FG; Chemicon), since
the ISH procedure causes fading of these markers.
F.5. Rinse sections 2 × 10 min in TBS.
F.6. Add direct-tagged fluorescent secondary antibody appropriate to FG
or other antibodies at this time and incubate for 45–60 min.
F.7. Rinse 2 × 10 min in TBS.
F.8. Rinse 10 min in NMT (0.1 M NaCl/50 mM MgCl2 in 0.1 M Tris,
pH 9.5).
F.9. Transfer sections to a solution of NBT (4.5 µl/ml of NMT) and BCIP
(3.5 µl/ml of NMT; 300 µl to 1 ml per well in a sterile 24-well tissue culture
dish). Filter the NBT/BCIP solution before use (we use syringe microfilters
of 0.22-µm pore size). It is important that the NBT/BCIP solution be in
clean plastic or glassware. Any dust or dirt can cause the reaction to seed,
and precipitate will form. Allow this reaction to proceed at room tempera-
ture protected from light on the shaker. Check the progress of the reaction
after 30 min and then every 15 min until the dark reaction product is seen
in cell bodies in appropriate areas and not in areas where the mRNA should
not be present. This is critical to the success of the experiment. We use a
dissecting microscope to inspect the tissue, while the reaction is progress-
ing to ensure a good signal-to-noise ratio. The reaction should be stopped
immediately if any background begins to appear.
F.10. Once there is dark reaction product and very low background (this
may take between 45 min and 4 h, depending on the probe and the amount
of mRNA expression), stop the reaction by transferring the sections back
into the mesh-well dish in a solution of TE, pH 8.5.
F.11. Rinse 3 × 10 min in TE, pH 8.5.
F.12. Rinse 3 × 5 min in TBS and 1 × 5 min in 0.1 M phosphate buffer.
F.13. Mount from 0.1 M phosphate buffer. Air-dry and cover with Vec-
tashield (Vector, Burlingame, CA) or another aqueous-based mounting me-
dia. Vectashield is good for protecting any other fluorophores in the reaction
(e.g., fluorescent-tagged secondaries) from fading. Seal edges of coverslip
with nail polish. Do not dehydrate the sections in alcohols or xylenes. Note:
The NBT/BCIP reaction product is soluble in alcohols and organic solvents
and is also light sensitive. The color will change slightly as the slides are
exposed to light.
G. Modifications for Double ISH
Unfortunately, a commercially available FITC-tagged control RNA is not

available to aid in determining exact concentrations of FITC-labeled cRNA
from the dot blot test.
G.1. Dot Blot Nytran Strip Test for FITC Probe. Follow section “Test for In-
corporation of Nonradioactive Label Using a Dot Blot on a Nytran Strip”
for Steps B.1–B.7. Use sheep anti-FITC peroxidase-tagged antibody (1:2000;
Roche) in place of the anti-digoxigenin antibody in Step B.8. Substitute a
PBS rinse for the NMT rinse in Step B.10. To visualize the peroxidase tag,
substitute Vector VIP (Vector Laboratories, one drop of each kit component
per 3.3 ml of PBS) for the NBT/BCIP solution in Step B.11. Rinse in PBS
rather than in TE as shown in Step B.12.
G.2. Double ISH. Perform the ISH exactly as described in sections “Pre-
hybridization” and “Hybridization.” Add the FITC-riboprobe with the
digoxigenin–riboprobe in Step D.1. Add sheep anti-FITC peroxidase-tagged
antibody (1:2000) along with any other primary antibodies in Step F.4. After
Step F.5, insert the following steps:
a. Incubate in biotin–tyramide at 1:75 in supplied diluent (Perkin-Elmer,
Boston, MA) for 10 min in sterile 24-well tissue culture dish.
b. Transfer to mesh-well dish and rinse 3 × 5 min in TBS.
c. Incubate in streptavidin Cy3 at 1:200 ( Jackson, West Grove, PA) plus
any other secondaries as indicated in Step F.6.
d. Proceed with rest of protocol from Steps F.6 to F.13.
Acknowledgments. This work was supported by grants HL 074011 and

HL 28785 from the National Institutes of Health, Heart Lung and Blood
Institute to P.G.G.
REFERENCES
Baldwin, T. J., Tsaur, M. L., Lopez, G. A., Jan, Y. N., and Jan, L. Y., 1991, Characterization of a
mammalian cDNA for an inactivating voltage-sensitive K+ channel, Neuron 7:471–483.
Bellocchio, E. E., Hu, H., Pohorille, A., Chan, J., Pickel, V. M., and Edwards, R. H., 1998,
The localization of the brain-specific inorganic phosphate transporter suggests a specific
presynaptic role in glutamatergic transmission, J. Neurosci. 18:8648–8659.
Bloch, B., Popovici, T., Chouham, S., and Kowalski, C., 1986a, Detection of the mRNA coding
for enkephalin precursor in the rat brain and adrenal by using an “in situ” hybridization
procedure, Neurosci. Lett. 64:29–34.
Bloch, B., Popovici, T., Le Guellec, D., Normand, E., Chouham, S., Guitteny, A. F., and Bohlen,
P., 1986b, In situ hybridization histochemistry for the analysis of gene expression in the
endocrine and central nervous system tissues: a 3-year experience, J. Neurosci. Res. 16:183–
200.
Brysch, W., Creutzfeldt, O. D., Luno, K., Schlingensiepen, R., and Schlingensiepen, K. H., 1991,
Regional and temporal expression of sodium channel messenger RNAs in the rat brain
during development, Exp. Brain Res. 86:562–567.
Chesselet, M. F., 1990, In Situ Hybridization Histochemistry, Boston: CRC Press.
Chesselet, M. F., Weiss, L., Wuenschell, C., Tobin, A. J., and Affolter, H. U., 1987, Compar-
ative distribution of mRNAs for glutamic acid decarboxylase, tyrosine hydroxylase, and
tachykinins in the basal ganglia: an in situ hybridization study in the rodent brain, J. Comp.
Neurol. 262:125–140.
Clavel, C., Binninger, I., Boutterin, M. C., Polette, M., and Birembaut, P., 1991, Comparison of
four non-radioactive and 35S-based methods for the detection of human papillomavirus
DNA by in situ hybridization, J. Virol. Methods 33:253–266.
Darby, I. A., 2000, In Situ Hybridization Protocols, 2nd ed., Totowa, NJ: Humana Press.
Esclapez, M., Tillakaratne, N. J., Kaufman, D. L., Tobin, A. J., and Houser, C. R., 1994, Com-
parative localization of two forms of glutamic acid decarboxylase and their mRNAs in
rat brain supports the concept of functional differences between the forms, J. Neurosci.
14:1834–1855.
Fremeau, R. T., Voglmaier, S., Seal, R. P., and Edwards, R. H., 2004, VGLUTs define subsets of
excitatory neurons and suggest novel roles for glutamate, Trends Neurosci. 27:98–103.
Gall, J. G., and Pardue, M. L., 1969, Formation and detection of RNA-DNA hybrid molecules
in cytological preparations, Proc. Natl. Acad. Sci. U. S. A. 63:378–383.
Goldman, D., Simmons, D., Swanson, L. W., Patrick, J., and Heinemann, S., 1986, Mapping
of brain areas expressing RNA homologous to two different acetylcholine receptor alpha-
subunit cDNAs, Proc. Natl. Acad. Sci. U .S.A. 83:4076–4080.
Guyenet, P. G., Stornetta, R. L., Weston, M. C., McQuiston, T., and Simmons, J. R., 2004,
Detection of amino acid and peptide transmitters in physiologically identified brainstem
cardiorespiratory neurons, Auton. Neurosci. 114:1–10.
Hisano, S., Sawada, K., Kawano, M., Kanemoto, M., Xiong, G. X., Mogi, K., Sakata-Haga, H.,
Takeda, J., Fukui, Y., and Nogami, H., 2002, Expression of inorganic phosphate/vesicular
glutamate transporters (BNPI/VGLUT1 and DNPI/VGLUT2) in the cerebellum and pre-

cerebellar nuclei of the rat, Mol. Brain. Res. 107:23–31.
Hoefler, H., Childers, H., Montminy, M. R., Lechan, R. M., Goodman, R. H., and Wolfe, H.
J., 1986, In situ hybridization methods for the detection of somatostatin mRNA in tissue
sections using antisense RNA probes, Histochem. J. 18:597–604.
Hwang, P. M., Glatt, C. E., Bredt, D. S., Yellen, G., and Snyder, S. H., 1992, A novel K+ channel
with unique localizations in mammalian brain: molecular cloning and characterization,
Neuron 8:473–481.
John, H. A., Birnstiel, M. L., and Jones, K. W., 1969, RNA-DNA hybrids at the cytological level,
Nature 223:582–587.
Johnson, A. D., Peoples, J., Stornetta, R. L., and Van Bockstaele, E. J., 2002, Opioid circuits
originating from the nucleus paragigantocellularis and their potential role in opiate with-
drawal, Brain Res. 955:72–84.
Julien, J. F., Legay, F., Dumas, S., Tappaz, M., and Mallet, J., 1987, Molecular cloning, expression
and in situ hybridization of rat brain glutamic acid decarboxylase messenger RNA, Neurosci.
Lett. 73:173–180.
Kreft, S., Zajc-Kreft, K., Zivin, M., Sket, D., and Grubic, Z., 1996, Application of the nonra-
dioactive in situ hybridization for the localization of acetylcholinesterase mRNA in the
central nervous system of the rat; comparison to the radioactive technique, Pflugers Arch.
431:R309–R310.
Lanaud, P., Popovici, T., Normand, E., Lemoine, C., Bloch, B., and Roques, B. P., 1989, Dis-
tribution of CCK mRNA in particular regions (hippocampus, periaqueductal grey and
thalamus) of the rat by in situ hybridization, Neurosci. Lett. 104:38–42.
Lenz, S., Perney, T. M., Qin, Y., Robbins, E., and Chesselet, M. F., 1994, GABA-ergic interneurons
of the striatum express the Shaw-like potassium channel Kv3.1, Synapse 18:55–66.
Lewis, M. E., and Baldino, F., 1990, Probes for in situ hybridization histochemistry, In: Chesselet,
M. F. (ed.), In Situ Hybridization Histochemistry, Boca Raton, FL: CRC Press, pp. 1–21.
Lewis, M. E., Sherman, T. G., and Watson, S. J., 1985, In situ hybridization histochemistry with
synthetic oligonucleotides: strategies and methods, Peptides 6(Suppl. 2):75–87.
Malherbe, P., Sigel, E., Baur, R., Persohn, E., Richards, J. G., and Mohler, H., 1990, Functional
expression and sites of gene transcription of a novel alpha subunit of the GABAA receptor
in rat brain, FEBS Lett. 260:261–265.
McKinnon, D., 1989, Isolation of a cDNA clone coding for a putative second potassium channel
indicates the existence of a gene family, J. Biol. Chem. 264:8230–8236.
Melton, D. A., Krieg, P. A., Rebagliati, M. R., Maniatis, T., Zinn, K., and Green, M. R., 1984,
Efficient in vitro synthesis of biologically active RNA and RNA hybridization probes
from plasmids containing a bacteriophage SP6 promoter, Nucleic Acids Res. 12:7035–
7056.
Mezey, E., 1989, Phenylethanolamine N-methyltransferase-containing neurons in the limbic
system of the young rat, Proc. Natl. Acad. Sci. U. S. A. 86:347–351.
Mitchell, V., Gambiez, A., and Beauvillain, J. C., 1993, Fine-structural localization of
proenkephalin mRNAs in the hypothalamic magnocellular dorsal nucleus of the guinea
pig: a comparison of radioisotopic and enzymatic in situ hybridization methods at the
light- and electron-microscopic levels, Cell Tissue Res. 274:219–228.
Park, J. S., Kurman, R. J., Kessis, T. D., and Shah, K. V., 1991, Comparison of peroxidase-labeled
DNA probes with radioactive RNA probes for detection of human papillomaviruses by in
situ hybridization in paraffin sections, Mod. Pathol. 4:81–85.
Pelletier, G., Liao, N., Follea, N., and Govindan, M. V., 1988, Distribution of estrogen receptors
in the rat pituitary as studied by in situ hybridization, Mol. Cell. Endocrinol. 56:29–33.
Perney, T. M., Marshall, J., Martin, K. A., Hockfield, S., and Kaczmarek, L. K., 1992, Expression
of the mRNAs for the Kv3.1 potassium channel gene in the adult and developing rat brain,
J. Neurophysiol. 68:756–766.
Pochet, R., Brocas, H., Vassart, G., Toubeau, G., Seo, H., Refetoff, S., Dumont, J. E., and Pasteels,
J. L., 1981, Radioautographic localization of prolactin messenger RNA on histological
sections by in situ hybridization, Brain Res. 211:433–438.
Poyatos, I., Ponce, J., Aragon, C., Gimenez, C., and Zafra, F., 1997, The glycine transporter
GLYT2 is a reliable marker for glycine-immunoreactive neurons, Mol. Brain Res. 49:
63–70.
Rogers, S. W., Hughes, T. E., Hollmann, M., Gasic, G. P., Deneris, E. S., and Heinemann, S.,
1991, The characterization and localization of the glutamate receptor subunit GluR1 in
the rat brain, J. Neurosci. 11:2713–2724.
Rudy, B., Kentros, C., Weiser, M., Fruhling, D., Serodio, P., Vega-Saenz, D. M., Ellisman, M. H.,
Pollock, J. A., and Baker, H., 1992, Region-specific expression of a K+ channel gene in
brain, Proc. Natl. Acad. Sci. U. S. A. 89:4603–4607.
Schmued, L. C., and Fallon, J. H., 1986, Fluoro-gold: a new fluorescent retrograde axonal tracer
with numerous unique properties, Brain Res. 377:147–154.
Schreihofer, A.M., Stornetta, R.L., and Guyenet, P.G., 1999, Evidence for glycinergic respiratory
neurons: Bötzinger neurons express mRNA for glycenergic transporter 2, J. Comp. Neurol.
407:583–597.
Seroogy, K., Schalling, M., Brene, S., Dagerlind, A., Chai, S. Y., Hokfelt, T., Persson, H., Brown-
stein, M., Huan, R., Dixon, J., Filer, D., Schlessinger, D., and Goldstein, M., 1989, Chole-
cystokinin and tyrosine hydroxylase messenger RNAs in neurons of rat mesencephalon:
peptide/monoamine coexistence studies using in situ hybridization combined with im-
munocytochemistry, Exp. Brain Res. 74:149–162.
Shivers, B. D., Schachter, B. S., and Pfaff, D. W., 1986, In Situ hybridization for the study of
gene expression in the brain, Meth. Enzymol. 124: 497–510.
Siegel, R. E., and Young III, W. S., 1985, Detection of preprocholecystokinin and pre-
proenkephalin A mRNAs in rat brain by hybridization histochemistry using complemen-
tary RNA probes, Neuropeptides 6:573–580.
Stornetta, R. L., Akey, P. J., and Guyenet, P. G., 1999, Location and electrophysiological char-
acterization of rostral medullary adrenergic neurons that contain neuropeptide Y mRNA
in rat medulla, J. Comp. Neurol. 415:482–500.
Stornetta, R. L., and Guyenet, P. G., 1999, Distribution of glutamic acid decarboxylase mRNA
containing neurons in rat medulla projecting to thoracic spinal cord in relation to
monoaminergic brainstem neurons, J. Comp. Neurol. 407:367–380.
Stornetta, R. L., McQuiston, T. J., and Guyenet, P. G., 2004, GABAergic and glycinergic presym-
pathetic neurons of rat medulla oblongata identified by retrograde transport of pseudora-
bies virus and in situ hybridization, J. Comp. Neurol. 479:257–270.
Stornetta, R. L., Sevigny, C. P., and Guyenet, P. G., 2003, Inspiratory augmenting bulbospinal
neurons express both glutamatergic and enkephalinergic phenotypes, J. Comp. Neurol.
455:113–124.
Stornetta, R. L., Sevigny, C. P., Schreihofer, A. M., Rosin, D. L., and Guyenet, P. G., 2002, Vesic-
ular glutamate transporter DNPI/GLUT2 is expressed by both C1 adrenergic and non-
aminergic presympathetic vasomotor neurons of the rat medulla, J. Comp. Neurol. 444:207–
220.
Stornetta, R.L., Schreihofer, A.M., Pelaez, N.M., Sevigny, C.P., and Guyenet, P.G., 2001, Pre-
proenkephalin mRNA is expressed by C1 and non-C1 barosensitive bulbospinal neurons
in the rostral ventrolateral medulla of the rat, J. Comp. Neurol. 435:111–126.
Surmeier, D. J., Eberwine, J., Wilson, C. J., Cao, Y., Stefani, A., and Kitai, S. T., 1992, Dopamine
receptor subtypes colocalize in rat striatonigral neurons, Proc. Natl. Acad. Sci. U. S. A.
89:10178–10182.
Terenghi, G., Polak, J. M., Hamid, Q., O’Brien, E., Denny, P., Legon, S., Dixon, J., Minth, C. D.,
Palay, S. L., Yasargil, G., and Chan-Palay, V., 1987, Localization of neuropeptide Y mRNA in
neurons of human cerebral cortex by means of in situ hybridization with a complementary
RNA probe, Proc. Natl. Acad. Sci. U. S. A. 84:7315–7318.
Valentino, K. L., Eberwine, J. H., and Barchas, J. D., 1987, In Situ Hybridization: Applications to
Neurobiology, New York: Oxford University Press.
Wada, K., Ballivet, M., Boulter, J., Connolly, J., Wada, E., Deneris, E. S., Swanson, L. W., Heine-
mann, S., and Patrick, J., 1988, Functional expression of a new pharmacological subtype
of brain nicotinic acetylcholine receptor, Science 240:330–334.
Wang, H., Kunkel, D. D., Schwartzkroin, P. A., and Tempel, B. L., 1994, Localization of Kv1.1
and Kv1.2, two K channel proteins, to synaptic terminals, somata, and dendrites in the
mouse brain, J. Neurosci. 14:4588–4599.
Wilkinson, D. G., 1998, In Situ Hybridization: A Practical Approach, 2nd ed., Oxford: Oxford
University Press.
Wisden, W., and Morris, B. J., 1994, In Situ Hybridization Protocols for the Brain, London: Academic
Press, Harcourt Brace &Co.
Wisden, W., Morris, B. J., Darlison, M. G., Hunt, S. P., and Barnard, E. A., 1988, Distinct GABAA
receptor alpha subunit mRNAs show differential patterns of expression in bovine brain,
Neuron 1:937–947.
Wuenschell, C. W., Fisher, R. S., Kaufman, D. L., and Tobin, A. J., 1986, In situ hybridization to
localize mRNA encoding the neurotransmitter synthetic enzyme glutamate decarboxylase
in mouse cerebellum, Proc. Natl. Acad. Sci. U. S. A. 83:6193–6197.
Young III, W. S., 1990, In situ hybridization histochemistry, In: Björklund, A., Hökfelt, T.,
Wouterlood, F. G., and Van den Pol, A. N. (eds.), Analysis of Neuronal Microcircuits and
Synaptic Interactions, Amsterdam: Elsevier.
9
Viral Tracers for the Analysis
of Neural Circuits
JOEL C. GEERLING,
THOMAS C. METTENLEITER, and
ARTHUR D. LOEWY
INTRODUCTION
HISTORICAL BACKGROUND OF VIRAL TRACING
BARTHA PRV AS A NEUROANATOMICAL TRACER
Phenotypic Characterization of PRV-Labeled Neurons
Use of PRV in Conjunction with Conventional Neural Tracers
Double Retrograde Transneuronal Tracing with Two Isoforms
of PRV
PRV Fills Neuronal Dendritic Trees
Electrophysiological Recordings from Transneuronally
Labeled Neurons
Genetically Engineered PRV for Highly Specific Tracing
SPECIFIC RETROGRADE TRANSPORT OF BARTHA PRV
TRANSNEURONAL TRANSFER OF BARTHA PRV AT
SYNAPTIC TERMINALS
NEUROANATOMICAL TRACING WITH PRV— PRACTICAL
CONSIDERATIONS AND PITFALLS
Viral Tracing in the CNS
Viral Tracing in the Peripheral Nervous System
Accurate Interpretation of Viral Labeling Patterns
False-Negative Data After Viral Transneuronal Tracing
OTHER VIRUSES USED FOR TRANSNEURONAL TRACING
STUDIES
HSV1 as a Transneuronal Tracer
Rabies as a Retrograde Transneuronal Tracer
Perspectives—
HSV1 and Rabies
JOEL C. GEERLING AND ARTHUR D. LOEWY • Department of Anatomy and

Neurobiology, P.O. Box 8108, Washington University School of Medicine, 660 S. Euclid Avenue,
St. Louis, MO 63110 THOMAS C. METTENLEITER • Institute of Molecular
Biology, Friedrich-Loeffler-Institut, D-17493
263
264 JOEL C. GEERLING, THOMAS C. METTENLEITER, and ARTHUR D. LOEWY
CONCLUSION
APPENDIX
Safety and Practical Issues
Sources of Bartha PRV and Recombinant Strains
Viral Growth, Aliquots, and Storage
Dilution of PRV
Injection of PRV
Preparation of Brain Sections for Immunohistochemical Staining
Immunohistochemical Staining
Disposal of PRV-Infected Material
REFERENCES
Abstract: Viral transneuronal tracing can be used to analyze neural circuits in the
central nervous system (CNS). In particular, the pseudorabies virus (PRV) strain
Bartha, an attenuated form of a pig alphaherpesvirus, is an excellent retrograde
transneuronal tracer for labeling neural networks. This virus is transported from the
axon terminal to the cell body of an infected neuron and enters the nucleus. There, it
replicates, producing progeny virions that are distributed throughout the cytoplasm.
These new viruses are then transferred into the axon terminals of second-order neu-
rons that innervate the infected neuron, and the process is repeated. This technique
has been used to analyze CNS networks involving chains of two or more function-
ally connected neurons. Due to the high sensitivity of viral transneuronal labeling,
false-positive data can be generated, leading to potential pitfalls of interpretation—
examples are discussed in this chapter. Protocols for growing PRV and viral tracing
methodology are included.
Keywords: Bartha pseudorabies virus, herpes simplex virus, HSV1, PRV, rabies,
transneuronal, transsynaptic
I. INTRODUCTION
The identification of neural networks is fundamental to understanding

brain functions. In fact, the input and output connections of every part of the
mammalian brain have been studied by neuroanatomical tracing methods
that depend on the axonal transport of either proteins or fluorescent chemi-
cal markers. This approach, while providing important data, is limited to the
analysis of single neurons, not circuits of functionally connected neurons.
A clear operational guide regarding the mammalian brain is dependent
on the knowledge of these circuits and the genetic expression patterns of
individual neurons that form them.
Even though conventional tracing techniques label only single neurons,
they can be used in combination to analyze multisynaptic circuits. A two-
neuron circuit, for example, can be visualized by the combination of two
tracer injections into the brain of the same animal. First, a group of neu-
rons is retrogradely labeled following injection of a protein tracer, such
as cholera toxin β-subunit (CTb), into its axon terminal field. Then, a
VIRAL TRACERS FOR THE ANALYSIS OF NEURAL CIRCUITS 265
second injection of an anterograde axonal tracer, such as the plant lectin
Phaseolus vulgaris leucoagglutinin (PHA-L), is made at a central site that
is known or predicted to project to the retrogradely labeled neurons. Af-
ter several days, the brains from these animals are processed by a dou-
ble immunohistochemical procedure that allows for the light microscopic
identification of the so-called close contacts— PHA-L-labeled axon termi-
nals abutting on CTb retrogradely labeled neurons. Definitive evidence is
dependent on immunoelectron microscopy (see chapter by Sesack et al.
in this volume). Since this technique is tedious and time-consuming, few
reports have used this methodology, prompting a search for other ap-
proaches for the anatomical identification of functionally connected sets of
neurons.
One of the first transneuronal tracers was identified in the early 1980s
with the discovery that when the protein conjugate wheat germ agglutinin–
horseradish peroxidase (WGA-HRP) was injected into the eye, it produced
transneuronal labeling in second- and third-order visual relay neurons
(Gerfen et al., 1982; Itaya and van Hoesen, 1982). However, when this pro-
tein was injected into the brain, the volume of WGA-HRP injections had to
be more limited— under these conditions, only vanishingly small amounts of
WGA-HRP were transferred transsynaptically (Fig. 9.1). Other agents, such
as the nontoxic fragment C of tetanus toxin, were also found to have lim-
ited utility as transneuronal tracers (Manning et al., 1990). The approach
was abandoned for more than a decade, but transgenic mice have been
developed with neurons expressing plant lectin genes, such as WGA, and
produce transneuronal labeling in defined neural circuits (Braz et al., 2002;
Horowitz et al., 1999; Zou et al., 2001).
Other attempts to develop anterograde transneuronal tracers included
the use of radioactive amino acids (Wiesel et al., 1974). These experiments
required injections of large amounts of isotope into the target site, such as
the eye. Then, after several days or weeks, the circuit could be identified
in histological sections prepared for autoradiography. Since these studies
require highly concentrated injections of expensive isotopes, this technique
did not gain wide appeal. Even with these setbacks, the search continued
for transneuronal tracers. In 1983, Martin and Dolivo made a key discovery
when they demonstrated that viruses could be used to map central pathways
(Martin and Dolivo, 1983).
Neurotropic herpesviruses (Fig. 9.2) have now proven extremely useful
for detailed analysis of brain circuits. Briefly, viral retrograde transneuronal
tracing occurs in the following manner. Live viruses are injected into a pe-
ripheral or central nervous system (CNS) target in a laboratory animal.
After several hours, the viruses enter axon terminals innervating this struc-
ture. From here, the viruses are transported retrogradely to the parent cell
bodies, undergo replication, and produce progeny virions, which become
dispersed throughout the cytoplasm of each infected cell. These progeny
virions are then transmitted to the incoming axon terminals that inner-
vate the infected neurons, and the infectious process is repeated. Multiple
Protein transneuronal tracer Viral transneuronal tracer
(6+ hr)
Figure 9.1. Comparison of protein and viral retrograde transneuronal tracers. Pro-
tein tracers, such as WGA-HRP (at left), can be used in transneuronal labeling stud-
ies, but only a small amount of tracer is transferred to second-order neurons. On
the right, viral tracers, such as PRV, undergo replication within the second-order
neurons and become self-amplifying transneuronal markers (Kuypers and Ugolini,
1990).
Envelope
Tegument
Capsid
Core
Figure 9.2. Morphology of a herpes virion. Herpesviruses, such as HSV1 and PRV, are
composed of four structures. The core, a linear double-stranded DNA genome, is
enclosed in an icosahedral capsid shell. The tegument, a layer of more than 15 differ-
ent proteins, surrounds the capsid. The virus particle is enclosed in a lipid envelope
in which viral glycoproteins are inserted (Mettenleiter, 2003).
rounds of replication and spread produce robust transneuronal labeling
within neural circuits (Fig. 9.1).
II. HISTORICAL BACKGROUND OF VIRAL TRACING
One hundred years ago, the route of viral entry into the CNS was a matter
of debate. The conventional view held that viruses spread locally, breaching
epithelial boundaries, spaces, and fluids to enter the brain after establishing
a foothold in the periphery. Another view proposed that viruses traveled the
axonal processes of neurons as conduits into the CNS.
For herpesviruses, evidence subsequently accumulated in support of the
concept of axonal transport. Goodpasture and Teague provided the ear-
liest support for anterograde and retrograde axonal transport of viruses
(Goodpasture, 1925; Goodpasture and Teague, 1923). In 1938, Albert Sabin,
later famous for the development of polio vaccine, made the important
observation that viruses enter the brain via preferential neural pathways
(Sabin, 1938). For example, vesicular stomatitis and eastern equine en-
cephalitis viruses ravaged the olfactory pathway to produce a lethal CNS
infection, while pseudorabies virus (PRV), a herpes family virus, traveled
in the sympathetic and trigeminal pathways without apparent olfactory
infection.
The observations of Sabin were followed in the 1970s by clear demonstra-
tions of specific axonal transport of herpesviruses to neuronal cell bodies
and transneuronal spread in the CNS (Bak et al., 1977; Cook and Stevens,
1973; Kristensson et al., 1971, 1982). However, not until Martin and Dolivo
(1983) published their study using PRV was it recognized that herpesviruses
could be used as transneuronal tracers for defining neural circuits. In par-
ticular, they drew attention to the greatest advantage of using a virus as a
transneuronal tracer— it replicates in each infected neuron. Thus, viruses
can be viewed as self-amplifying markers, robustly labeling each hierarchical
level of a neural circuit, in contrast to the diminishing transsynaptic diffusion
of chemical tracers (Fig. 9.1). PRV tracing was used in two additional stud-
ies in the 1980s but, unfortunately, the investigators did not indicate the
specific viral strain, source, and dose used in their experiments (Rouiller
et al., 1986, 1989). Attempted transsynaptic tracing with a wild-type form of
PRV (Becker strain) was found to result in uncontrolled, nonspecific, and
rapidly lethal infections, preventing its use as a specific transneuronal tracer
(Strack et al., 1989b).
Ugolini and colleagues injected herpes simplex virus type 1 (HSV1) into
peripheral nerves and showed that it produced a transneuronal infection in
rat brain (Kuypers and Ugolini, 1990; Ugolini et al., 1987, 1989). However,
the virus also spreads locally and nonspecifically to adjacent glial cells and
neurons (Ugolini et al., 1987). Subsequently, false-positive transneuronal
labeling occurred in neurons connected to sites of nonspecific infection.
This latter finding raised concern over whether viral infections could be
contained within specific neural circuits, undermining enthusiasm for this

method.
Because of these nonspecific infections, viral tracing was not widely ex-
ploited until the discovery that a less virulent derivative of PRV, Bartha PRV,
produces highly specific retrograde transneuronal infections (Jansen et al.,
1993; Strack et al., 1989b; Strack and Loewy, 1990). In contrast to the fulmi-
nant infections produced by wild-type PRV, Bartha PRV infections remain
restricted mainly to synaptically linked chains of neurons and move only in
the retrograde direction. Moreover, rats survive more than twice as long—
up to 7 days— following an injection of Bartha PRV into a peripheral target
(Westerhaus and Loewy, 2001), whereas wild-type PRV kills rats within 3
days (Strack et al., 1989b). This attribute allows transneuronal propagation
to higher order neurons of a neural circuit (Enquist, 2002). Ensuing studies
revealed specific genetic alterations responsible for the retrograde speci-
ficity and reduced infectivity of Bartha PRV (see below).
Bartha PRV is an effective and well-characterized retrograde transneu-
ronal tracer that produces infections in most laboratory species (mouse, rat,
gerbil, hamster, ferret, sheep, and chicken). Unlike other transneuronal vi-
ral tracers, such as HSV1 and rabies, PRV does not infect humans (Gustafson,
1975). Thus, Bartha PRV was quickly recognized as a safe and accessible
tool for retrograde transneuronal tracing experiments. However, PRV does
not cause infections in primates. Other neurotropic viruses— HSV1 and
rabies—have been tested and used for circuit analysis in monkeys. For fur-
ther details, see section “Other Viruses Used for Transneuronal Tracing
Studies.”
Bartha PRV has been genetically modified in various ways to create new
viral tools for neural circuit analysis. Advances in herpesvirus biology, in-
cluding the availability of the complete PRV genome sequence (Klupp et
al., 2004), provide the opportunity to construct selective viral tracers, which
should allow neuroscientists to unravel specific multisynaptic pathways in
unprecedented detail.
III. BARTHA PRV AS A NEUROANATOMICAL TRACER
While Bartha PRV has primarily been used to identify chains of central
neurons innervating peripheral targets, it has also been utilized for trac-
ing circuits within the brain. In addition, viral tracing has been combined
with various other well-established neuroanatomical techniques. Innova-
tive methodologies continue to appear, making PRV an even more use-
ful neurobiological tool. Genetically engineered PRV strains allow double
transneuronal tracing experiments and the detection of transneuronally
labeled neurons in living tissue, by fluorescent protein expression, for elec-
trophysiological recording. Recombinant PRV has even been used in trans-
genic mice in an attempt to selectively label the inputs to specific neuronal
phenotypes.
A. Phenotypic Characterization of PRV-Labeled Neurons
Bartha PRV was used to provide the first direct neuroanatomical identi-
fication of brainstem and hypothalamic neurons that regulate sympathetic
outflow systems (Strack et al., 1989a, b). Beginning with this work, it was
demonstrated that peptide antigens could still be detected in infected neu-
rons. This was a fortuitous discovery, since herpesviruses terminate protein
synthesis in other cell types, via expression of the viral host shutoff protein
gene, UL41 (Smiley, 2004). However, this viral host shutoff process does not
occur in neurons (Nichol et al., 1994).
Therefore, identification of the phenotype of infected neurons, by
double-immunohistochemical labeling, can add important information
about transneuronally labeled neurons (Fig. 9.3). In addition, investigators
have successfully used in situ hybridization to demonstrate various mRNA
transcripts in PRV-labeled cells (Boldogkoi et al., 2002; Broussard et al., 1996;
Giles et al., 2001; Song and Bartness, 2001; Stornetta et al., 2004).
One caveat should be noted regarding the detection of marker molecules
within PRV-infected neurons— expression of various peptides and enzymes
within neurons infected with Bartha PRV tends to be reduced, often dramat-
ically, relative to uninfected neighboring neurons of the same phenotype.
Consequently, central injection of colchicine 24 h prior to killing an infected
rat is sometimes used to boost the labeling of axonally transported peptides.
B. Use of PRV in Conjunction with Conventional Neural Tracers
The PRV transneuronal tracing technique can also be combined with

conventional neural tracing methods. In one example, anterograde PHA-L
tracing was combined with the PRV method to map the specific central re-
gions targeted by the periaqueductal gray matter (PAG) to modulate sympa-
thetic functions. In two series of rats, PHA-L injections were made into either
the lateral PAG column, implicated in the fight-or-flight reactions, or the
ventral PAG column, which mediates the opposite behavioral responses. Two
days later, Bartha PRV was injected into the stellate sympathetic ganglion.
After an additional 4 days, select regions of the hypothalamus and brain-
stem contained PHA-L terminals contacting PRV-labeled neurons. Specific
sites, particularly the raphe magnus nucleus, through which PAG could
modulate sympathetic activity, were thus identified (Farkas et al., 1998). In
this situation, PRV tracing alone was capable of demonstrating only that
PAG neurons have multineuronal connections to the stellate ganglion— the
specific presympathetic groups through which retrograde transneuronal
PAG labeling had occurred were revealed by simultaneous anterograde
tracing.
In a second example, a key hypothalamic relay nucleus was identified in
a circuit implicated in circadian arousal functions— the pathway from the
suprachiasmatic nucleus (SCN) to the locus coeruleus (Aston-Jones et al.,
Figure 9.3. Double viral transneuronal tracing experiments can be used to identify
specific neuronal phenotypes implicated as potential candidates regulating behav-
ioral functions, such as the fight-or-flight response. In this example, orexin neurons
(blue, panel D) in the lateral hypothalamus are transneuronally double labeled with
Bartha PRV strains expressing two unique reporters: GFP-PRV from the stellate gan-
glion (green, B) and β-gal PRV from the adrenal gland (red, C). Triple-labeled cells
appear white in panel E (Geerling et al., 2003).
2001). The SCN was screened for neurons that project both transneuronally
to the locus coeruleus and directly to the specific hypothalamic nuclei,
such as the dorsomedial hypothalamic (DMH) nucleus. Bartha PRV was
injected into locus coeruleus in animals with CTb injections into various
hypothalamic nuclei. The highest number of double-labeled SCN neurons
occurred in animals with CTb injections into the DMH, as opposed to other
hypothalamic regions such as the paraventricular and lateral hypothalamic
nuclei and the preoptic region. Since the DMH provides a direct input to
locus coeruleus and DMH lesions blocked the circadian changes in locus
coeruleus neural activity, a strong case was made for the existence of an
SCN → DMH → locus coeruleus circuit (Aston-Jones et al., 2001).
C. Double Retrograde Transneuronal Tracing with

Two Isoforms of PRV
Neurons with branched axonal projections can be identified by double

retrograde tracing techniques. For example, when two different retrograde
tracers (e.g., Fluorogold and CTb) are injected into two central regions, it
is possible to identify single neurons innervating both sites by the colocal-
ization of the two tracers within the same cell.
Jansen et al. (1995a) demonstrated that this approach could be extended
to transneuronal retrograde tracing by using two unique viruses. Two Bartha-
derived PRV strains were separately injected into different sympathetic tar-
get tissues— adrenal medulla and stellate ganglion. The two recombinant
viruses were uniquely identifiable. One contained a lacZ gene insertion
within the nonessential gG gene locus. This virus could be identified by
immunohistochemical staining for β-galactosidase (the protein product of
the lacZ gene). In the other virus, the wild-type gC gene, which is mutated in
Bartha PRV, was restored. Hence, this viral strain could be uniquely identi-
fied by its expression of the wild-type gC membrane protein, which is absent
in the lacZ recombinant.
Double virus tracing can be a powerful method for detailing the exact cen-
tral sites that regulate complex behavioral activities. The double-viral tracing
technique was used to identify groups of putative central command neurons
of the sympathetic nervous system, positioned to coordinate the activation
of multiple different sympathetic target organs ( Jansen et al., 1995a). This
study demonstrated the feasibility of double viral tracing. Later, other investi-
gators applied this technique to demonstrate central neurons positioned to
synchronize both somatomotor and sympathetic activations (Kerman et al.,
2003; Krout et al., 2003), as well as brainstem neurons that could coacti-
vate inspiratory and expiratory respiratory muscles that could discharge in
parallel, which occurs during vomiting (Billig et al., 2000). Double virus
transneuronal tracing was used to demonstrate that vasopressin-containing
SCN neurons (Ueyama et al., 1999) and individual orexin neurons in the
lateral hypothalamus (Geerling et al., 2003) can coordinately regulate mul-
tiple sympathetic outflow systems. Double transneuronal labeling in orexin
neurons is shown in Fig. 9.3. A similar strategy, using two uniquely identifi-
able strains of HSV1, has been used to compare the origins of sympathetic
outflows within the spinal cord (Levatte et al., 1998).
Since the introduction of the double virus labeling method, additional

technical refinements have greatly increased its usefulness. The original
study by Jansen and colleagues was extremely inefficient, requiring 256 rats
in order to obtain eight animals with well-matched double infections. This
3% yield was partly due to the conservative viral dose used in this study, result-
ing in only a 20% success rate for single-virus studies (Strack et al., 1989b),
but also to unmatched infectivity between the two viral tracing strains (Sams
et al., 1995). These difficulties have been overcome with the addition of
several new genetically engineered viral strains (see below).
One surprising aspect of this technique is that simultaneous infections
can frequently be established by two separate viral strains, arriving from
separate sites of origin, within the same neurons. This occurrence was not
completely predictable, based on in vitro work demonstrating a cellular
phenomenon called superinfection resistance. The herpesvirus membrane
protein gD can inhibit the subsequent infection of the cell by additional
virus (Campadelli-Fiume et al., 1988). Whether or not a similar effect is
relevant for infections with Bartha PRV derivatives in vivo, a growing body
of double viral tracing data unequivocally demonstrates that, in viral tracing
paradigms, robust double infections can occur in many neurons.
Concern over superinfection resistance in double viral tracing was first
raised when it was reported that infection by a Bartha PRV-derived strain with
enhanced virulence, due to restoration of the virulence-enhancing wild-type
gI gene (Whealy et al., 1993), was shown to greatly inhibit a second transneu-
ronal infection in the same pathway by a β-galactosidase-expressing virus
injected 24 h later (Kim et al., 1999). However, when the order was reversed,
with the β-galactosidase strain injected first, it was unable to reciprocate this
strong inhibition. Whether the first effect was due to a direct superinfection
resistance mechanism in PRV-infected neurons, as opposed to indirect con-
sequences of the greatly increased virulence of the first strain, was not re-
solved. For example, strong glial reaction at the injection site or rapid injury
to the infected first-order neurons could have simply resulted in decreased
entry and retrograde transport of virus injected a day later.
Banfield et al. (2003) found that when two isogenic Bartha recombinants
were injected at the same time and site, they produced a double transneu-
ronal infection in a large number of higher order neurons. More than 75%
of the infected third-order paraventricular hypothalamic neurons expressed
reporter genes from both viruses (green and red fluorescent proteins) after
injection of a mixture of the two strains into one eye (Cano et al., 2003).
Parallel experiments were performed on cultured dorsal root ganglion
cells. These data showed that infection with one strain of PRV greatly re-
duced the susceptibility to infection by another PRV strain within 2 h and
completely prevented superinfection after 4 h. Whether this in vitro time
limit represents a similar constraint in vivo remains unknown. The robust
double infections observed when the viral exposure times were matched
exactly (by coinjecting them) combined with the indication that a small
time window may exist for superinfection suggests that optimization of the
double-virus technique requires matching viral rates of progression as
closely as possible.
For this reason, an important aspect of double-viral tracing studies is the
choice of viruses with similar transneuronal infection kinetics (Ter Horst,
2000). Optimal yields of double-infected neurons may be obtained when
viral rates of transit are most closely matched, such that the two viruses
arrive at an afferent site with minimum delay between strains. The first
two double-viral tracing studies used two viruses with significantly different
virulence characteristics ( Jansen et al., 1995a; Kim et al., 1999). In each
case, the infectivity of the lacZ/β-galactosidase strain (a minimally altered
version of Bartha PRV) was significantly reduced relative to the second strain
(a version of Bartha to which a virulence-endowing membrane glycoprotein
gene had been restored) (Kim et al., 1999; Mettenleiter et al., 1987, 1988;
Sams et al., 1995).
Following these original studies, double transneuronal tracing has ben-
efited from the genetic engineering of isogenic viruses in which insertions
of reporter genes have been targeted to the same genetic locus. For exam-
ple, inserting the green fluorescent protein (GFP) gene within the same gG
locus used for lacZ (Jons and Mettenleiter, 1997) has allowed more compa-
rable double viral infections at equivalent doses and times (Geerling et al.,
2003; Krout et al., 2003; Ueyama et al., 1999). However, discordant rates of
expression between the two reporter genes, within double-infected cells,
can result in unequal detectability of the two strains, whether or not infec-
tivity is equal. In some studies, two different promoters have been used to
drive the expression of reporter genes— the intrinsic gG promoter and the
human cytomegalovirus (CMV) immediate-early promoter. Different pro-
moters could necessitate the use of two different viral doses to match the
timing of reporter expression, even when the rates of viral spread may be
comparable (Cano et al., 2003). Differences in relative expression levels of
reporter proteins can be largely overcome by the use of the CMV promoter
in both strains, driving high levels of gene expression as early as possible
(Banfield et al., 2003). Nonetheless, any pair of viruses used in double tracing
experiments should be compared to establish similar rates of transneuronal
progression.
D. PRV Fills Neuronal Dendritic Trees
Card et al. (1993) noted that pseudorabies virions fill the entire dendritic
tree of an infected neuron, out to the distal branches. This property can
be useful for ultrastructural analysis (Carr et al., 1999; Carr and Sesack,
2000). More interestingly, dendritic filling by PRV has been exploited to
solve a long-standing problem in neuroanatomical tracing— identifying the
synaptic afferents to the distal dendrites of a group of neurons.
When the dendrites of a particular group of neurons extend into adja-
cent cytoarchitectonic regions— outside the boundaries of its parent cell
group as defined in Nissl-stained sections— it can be difficult to determine,

simply by retrograde tracer injections, which are true neural inputs to the
particular cell group (Bourgeais et al., 2003; Luppi et al., 1995). In some
instances, this problem can be overcome by injecting PRV within the center
of a group— after replication in the soma, viral progeny spread through-
out the neuron and can spread transsynaptically into even the most distal
synaptic afferent terminals (Aston-Jones and Card, 2000). This approach has
been used to demonstrate spinal lamina I afferents to the distal dendrites
of amygdala-projecting neurons in the external lateral parabrachial nucleus
( Jasmin et al., 1997). More recently, Aston-Jones et al. (2004) injected PRV
within the core of the locus coeruleus and found that the virus replicated
and spread throughout these noradrenergic neurons, producing transneu-
ronal infections in input neurons that contact the most distal dendritic
branches.
E. Electrophysiological Recordings from Transneuronally

Labeled Neurons
Various methods have been utilized to identify specific neurons for elec-
trophysiological study in brain slices. In particular, fluorescent dyes such
as Fluorogold have been used to identify retrogradely labeled neurons in
living tissue (Kangrga and Loewy, 1995).
In 2000, Smith et al. first demonstrated that electrophysiological record-
ings could be targeted to identified neurons with multisynaptic connections
to a specific target. These investigators created the Bartha-derivative PRV
152, designed to produce high, early expression of enhanced green fluo-
rescent protein (EGFP, driven by a CMV promoter). They demonstrated
that EGFP-expressing, PRV-infected neurons were easily identifiable in tis-
sue slices. Most important, despite viral infection, the electrophysiological
properties of these cells were comparable to uninfected neurons. This study
was followed by a similar demonstration of electrophysiological recording
from PRV-infected neurons, using a different GFP-expressing strain (Irnaten
et al., 2001).
These findings built confidence that recordings can be obtained from the
neurons identified as multisynaptic afferents to a specific target. However,
infected cells may show electrophysiological abnormalities (Fukuda et al.,
1983). Still, high-quality recordings have been made from visually identified
GFP-PRV neurons and this method can be quite useful for studying neural
circuits.
F. Genetically Engineered PRV for Highly Specific Tracing
Molecular biological tools have created opportunities for constructing

viruses with improved properties as neural tracers (Boldogkoi et al., 2004).
New viral tracers may allow more selective labeling within neuronal cir-
cuits. For example, the use of cell-specific conditional expression technol-
ogy should allow the targeting of infection or viral reporter gene production
to neurons of a particular phenotype. In this way, local CNS injections of
specific viruses may produce transneuronal labeling restricted to the inputs
of a functionally specific type of neuron.
The feasibility of this approach was demonstrated by DeFalco et al. (2001),
who injected a genetically engineered Cre recombinase-dependent strain
of PRV into a transgenic mouse that expresses Cre recombinase in only
one neuronal phenotype. They began with Bartha PRV and removed its
thymidine kinase (tk) gene, which is necessary for viral replication in vivo.
The tk sequence was then reinserted, along with an EGFP reporter gene,
in the nonessential gG locus (the same site used for the lacZ and GFP re-
porters described above), driven by a CMV promoter. A STOP sequence,
flanked by loxP sites, was inserted upstream from the tk and EGFP se-
quences. This STOP sequence was positioned to prevent expression of these
genes and, therefore, both viral replication and cellular expression of EGFP
reporter.
However, the Cre recombinase enzyme can join the loxP sites, removing
the intervening STOP sequence. Therefore, this replication-deficient virus,
termed Ba2001, was injected into the brains of transgenic mice expressing
Cre under the control of specific genetic promoters (neuropeptide Y or the
leptin receptor). In these mice, Ba2001 could enter many types of neurons,
but could replicate only in Cre-expressing neurons. In these specific neu-
rons, infection with replication-competent virus was reported by concurrent
EGFP expression. With these tools, the specific neural networks regulating
NPY- or leptin receptor–expressing neuronal subpopulations within the hy-
pothalamic arcuate nucleus could be selectively labeled.
This important methodological advancement raises the possibility that
designer herpesviruses could become important tools for mapping neural
circuits with unprecedented specificity. Unfortunately, in the 4-year period
since publication, these findings have not been detailed or extended. Since
certain CNS sites reported to be infected in the DeFalco report (DeFalco
et al., 2001), such as somatosensory cortex, seem incompatible with known
inputs to the hypothalamus, the potential for spontaneous viral genetic
mutation causing spurious labeling in vivo must be addressed.
The promise of custom-made viral tracers remains alluring (Boldogkoi
et al., 2004), but the enthusiasm surrounding this technology should not
cause investigators to overlook the necessity of detailed neuroanatomical
characterization of all new viral tracers. Any new viral strain should be care-
fully compared with a well-characterized strain, such as Bartha PRV, with
respect to transsynaptic specificity, kinetics of infectious spread, tropism,
and critical viral doses (Banfield et al., 2003). Even minor alterations in
what may appear to be insignificant regions of the PRV genome, such as
the nonessential gG gene locus, can result in significant differences in in-
fectivity (Cano et al., 2003; Demmin et al., 2001; Sams et al., 1995). Finally,
it is important that experimental results obtained with any viral strain be

critically compared with existing neuroanatomical data.
IV. SPECIFIC RETROGRADE TRANSPORT OF BARTHA PRV
Since the first transneuronal tracing experiments with Bartha PRV, it was
clear that this viral strain spreads preferentially, if not exclusively, in a ret-
rograde direction (from the axon terminal to the cell body— the opposite
direction of neural transmission). An early argument for retrograde speci-
ficity came from studies that showed that, when PRV was injected into skele-
tal muscle, it produced retrograde labeling in ventral horn motor neurons,
but not in the dorsal root ganglia or central somatosensory sites (Rotto-
Percelay et al., 1992).
Patterns indicative of retrograde-only transport have also been observed
after Bartha PRV injections within the CNS. After injection into the
mediodorsal nucleus of the thalamus, O’Donnell et al. (1997) noted that
infection within the cortex first occurred within deep layer neurons, consis-
tent with retrograde transport from the thalamus. Card et al. (1998) showed
that, unlike that of wild-type Becker PRV, injection of Bartha PRV in the
prefrontal cortex did not produce an anterograde transneuronal infection
in the striatum, a major efferent target. Chen et al. (1999) reported that,
even when entering fibers of passage through an injection site, Bartha PRV
did not produce anterograde transneuronal labeling.
Wild-type PRV clearly does spread anterogradely. This was known from
Sabin’s early observations of infection in the central trigeminal sensory
pathways after olfactory instillation (Sabin, 1938). Later, it was observed
that wild-type PRV injected into the eye produces a fulminant infection of
all retinorecipient sites within the brain (Card et al., 1991). Since there are
relatively slight genetic differences between wild-type and Bartha PRVs, yet
major differences in their transport properties, a search was initiated for the
specific genes responsible for anterograde infectious spread.
At least three PRV genes appear necessary for anterograde spread: gE, gI,
and Us9. All three of these genes are absent from Bartha PRV, due to a large
deletion in the unique short (Us) region of the PRV genome (Lomniczi
et al., 1984). The deletion of any one of these genes from wild-type PRV
eliminates anterograde viral transmission (Brideau et al., 2000; Card et al.,
1992; Whealy et al., 1993).
Both gE and gI are membrane glycoproteins and form a functional het-
erodimer (Mettenleiter et al., 1988; Whealy et al., 1993). These two genes
had been previously characterized as encoding important PRV virulence-
enhancing factors (Mettenleiter et al., 1987, 1988). Deletion of either gene
from wild-type PRV was shown to eliminate anterograde spread from the
retina to the retinorecipient visual sites in the brain (Card et al., 1992; Whealy
et al., 1993). Loss of anterograde spread in gE- and gI-null mutants has also
been confirmed in the olfactory pathway (Babic et al., 1996; Kritas et al.,
1994). Injection of a mixture of gE-null and gI-null PRVs within the eye,
however, resulted in restoration of a wild-type anterograde infection pattern
(Enquist et al., 1994). This implies that both mutants infect retinal ganglion
cells, but require the addition of their respective missing gene products—
upon coinfection of the same cell— for productive anterograde spread. It is
still unclear exactly how gE and gI allow anterograde transmission (Enquist
et al., 2002; Tomishima et al., 2001).
In contrast, the mechanism by which the Us9 gene product influences
anterograde transport is better characterized. As with gE and gI, absence of
the Us9 gene inhibited the anterograde spread of wild-type PRV (Brideau
et al., 2000). Tomishima and Enquist (2001) further demonstrated in vitro
that, without Us9, necessary membrane glycoproteins do not enter the axon
of an infected neuron. While other viral proteins proceed normally into
the axon, this lack of membrane protein trafficking prevents anterograde
transmission of complete, infectious virions.
While the exact molecular mechanisms required for PRV anterograde in-
fectious spread remain unknown, the studies cited above have highlighted
some of the key factors. The identification of specific genetic mutations
preventing anterograde transneuronal infections by Bartha PRV gradually
cast doubt upon the only cited evidence that this strain could produce an
anterograde infection— the delayed infection of SCN after injection into
the eye (Brideau et al., 2000; Card, 2000; Card et al., 1991, 1992; Enquist et
al., 1994; Husak et al., 2000; Moore et al., 1995; Smith et al., 2000; Whealy
et al., 1993). Careful neuroanatomical analysis, however, revealed that this
purportedly anterograde infection was actually produced by retrograde
spread via multisynaptic autonomic outflows to the eye (see discussion under
“Practical Considerations and Pitfalls”; Pickard et al., 2002; Smeraski et al.,
2004).
In summary, a great deal of collective neuroanatomical experience with
Bartha PRV indicates that this virus moves selectively in a retrograde direc-
tion. Three key PRV genes have been individually shown to be necessary
for anterograde viral spread— Bartha is deficient in each one. Together,
these findings build a strong case that Bartha PRV is a retrograde neuronal
tracer.
V. TRANSNEURONAL TRANSFER OF BARTHA PRV AT

SYNAPTIC TERMINALS
The pattern of Bartha PRV transneuronal labeling is largely consistent

with specific transfer at synapses without leakage to nearby neurons or local
axons. For example, after PRV injection into any visceral tissue or autonomic
ganglion, transneuronal labeling was consistently found in the parvocellu-
lar subdivision of the paraventricular hypothalamic nucleus— an area known
from earlier work to be a key site regulating autonomic functions. Impor-
tantly, nearby neurons lying in the intermingled magnocellular subdivision
of this nucleus, which projects solely to the posterior pituitary, were not la-
beled (Strack et al., 1989b). Such restricted labeling indicated that random
cell-to-cell spread did not occur.
This consistent pattern led to the proposal that transneuronal spread of
Bartha PRV occurred specifically through neuronal synapses (Strack and
Loewy, 1990). However, beginning with the first use of Bartha PRV as a
transneuronal tracer, it was observed that Bartha PRV can infect glia within
infected neuronal sites (Rinaman et al., 1993; Strack et al., 1989b). Although
only limited infections of astroglia occurred, this observation raised signifi-
cant concern over the potential for false-positive labeling via not only local
spread to unrelated neurons but also subsequent transneuronal propaga-
tion. Indeed, Ugolini et al. (1987) had reported that tracing with HSV1
resulted in formidable local spread to neurons within unrelated circuits.
HSV1 injected into the mouse hypoglossal nerve spread from nerve roots
in the ventral medulla to both glia and inferior olivary neurons and, via
transneuronal transfer within only a few days, to neurons in the cerebel-
lum. Although no such nonspecific infection had been reported for Bartha
PRV, this potential roadblock was carefully examined in early experiments
validating the virus as a transneuronal tracer.
In 1990, Strack and Loewy demonstrated that, after Bartha PRV was in-
jected into the eye or the skin of the ear, retrograde labeling in the sym-
pathetic superior cervical ganglion (SCG) was completely restricted to the
subset of neurons afferent to the particular site of injection. Even after
4 days, infection did not spread locally within the SCG (Strack and Loewy,
1990). A similar result was reported for the CNS by Jansen et al. (1993). Af-
ter injection of Bartha PRV into either the stellate ganglion or the adrenal
medulla, coincident with CTb injection into the other sympathetic target,
the percentage of spinal cord neurons labeled with both virus and CTb was
not different from the double-labeled proportion found after injections of
two conventional retrograde tracers (CTb and Fluorogold). This indicated
that PRV infection within this first-order afferent site remains confined to
specific sympathetic preganglionic neurons. When both experiments are
considered together, a convincing case can be made against the likelihood
that Bartha PRV produces lateral infections involving neighboring neurons
(Scenario 3 in Fig. 9.4).
Consistent with these findings, diffusion of PRV through the neuropil
may be hindered by the large size of infectious virions (200 nm) and by
its binding to cell surface heparin sulfate moieties (Aston-Jones and Card,
2000). The spread of PRV to axons and local glia, but not to adjacent
neuronal cell bodies, is also consistent with an earlier report indicating
a greater herpesvirus-binding affinity for synaptic terminals and glial cells,
relative to neuronal perikarya (Vahlne et al., 1978).
However, besides local spread to adjacent neuronal cell bodies, one po-
tential avenue of nonspecific PRV spread remains— the leakage of virions
into adjacent nonsynaptic axons (Scenario 2 in Fig. 9.4). Whereas the exper-
iments cited above (Jansen et al., 1993; Strack and Loewy, 1990) dispelled
Specific Nonspecific
2nd Order
1
2
1st Order
Target A
Figure 9.4. Viruses can produce both specific and nonspecific retrograde infections
in the CNS. (1) The primary mode of transneuronal spread, for Bartha PRV and
other viruses, is via direct transfer to the synaptic afferents of an infected neuron.
(2) Spread of a viral tracer to adjacent axons and terminals that do not synapse
upon the infected neuron may occur (see “Transneuronal Transfer of Bartha PRV at
Synaptic Terminals”). (3) Lateral leakage of virus to neighboring neurons does not
appear to occur with Bartha PRV (Jansen et al., 1993), but may present a problem
with other viral tracers, such as HSV1 (Ugolini et al., 1987).
concern over the potential for local spread to neuronal cell bodies, only
indirect tests dealt with the possibility that some of the PRV released from
an infected neuron may infect adjacent axons or axon terminals.
First, when PRV was injected into the eye or into the skin of the ear, it pro-
duced second-order transneuronal labeling in the appropriate distribution
of sympathetic preganglionic neurons in the spinal cord, as determined
by prior electrophysiological data (Strack and Loewy, 1990). The eye- and
ear-specific SCG neurons, through which transneuronal transport had
occurred, are highly intermixed. Hence, this result indicated that transfer
of PRV from infected first-order SCG neurons took place in a preferentially

transsynaptic manner, not simply by transmission of virions to all nearby
axon terminals. However, because the distributions of preganglionic neu-
rons infected after injections into eye and ear overlapped between spinal
levels T2 and T4, the possibility remained that a small proportion of the
labeled neurons in this zone was the result of nonsynaptic viral transfer in
the SCG.
This is the only tracing experiment to directly address the issue of Bartha
PRV spread to the nearby axons. Clearly, Bartha PRV is preferentially trans-
ferred to synaptic afferents, but only circumstantial evidence exists concern-
ing whether or not a small proportion of virions is nonspecifically transferred
to adjacent axons. This possibility is of potential significance, given the ex-
ponential amplification expected to occur after false-positive labeling.
An attractive theory was proposed that reconciled the observation of as-
troglial infections with a lack of spread to adjacent neurons, and that offered
a mechanism by which transsynaptic specificity may be preserved (Rinaman
et al., 1993). Virions were observed, by electron microscopy, to be prefer-
entially released from an infected neuron at sites of synaptic contact (Card
et al., 1993). These virions did not appear to breach the synapses them-
selves; they spread parasynaptically and equally infiltrated the afferent axon
terminal and the astroglial processes that form a barrier around the synaptic
region (Card et al., 1993). Thus, it was proposed that astrocytic processes
may absorb any PRV not incorporated into afferent terminals, preventing
nonspecific spread to adjacent structures (Card, 1998). Furthermore, Card
et al. (1993) noticed that, in contrast to neurons, the PRV produced within
astrocytes did not acquire a viral envelope, which is a necessary component
for infectious virions. These investigators proposed that, rather than serving
as a source of PRV production and nonspecific local spread, astrocytes limit
viral spread to nearby axons without producing normal infectious virions
(Rinaman et al., 1993).
This appealing theory may explain the transsynaptic pattern of PRV
spread and the lack of PRV spread to adjacent neurons ( Jansen et al., 1993;
Strack and Loewy, 1990), despite infection of adjacent glia (Rinaman et al.,
1993; Strack et al., 1989b). However, the spread of some amount of virus
to adjacent nonsynaptic axons remains an important possibility that cannot
be addressed by circumstantial evidence or by inferential approaches. The
issue of whether transneuronal PRV spread occurs exclusively via synaptic
afferents remains unsettled. This possibility is an important consideration
because the brain regions where nonspecific transfer presents the greatest
obstacle to interpretation are those in which viral transneuronal tracing is
most useful— sites such as the brainstem and hypothalamus, which contain
spatially intermixed, yet functionally diverse, populations of neurons.
In summary, Bartha PRV is a retrograde transneuronal tracer that is pref-
erentially taken up by synaptic terminals. Concomitant astroglial infections
do not appear to lead to local nonspecific neuronal labeling and may even
restrict PRV transfer to increase the probability that virus uptake occurs
at synaptic sites. However, whether or not PRV is transferred exclusively to
synaptic afferents remains unresolved.
VI. NEUROANATOMICAL TRACING WITH PRV—PRACTICAL

CONSIDERATIONS AND PITFALLS
Viral tracing is a highly sensitive technique. When applied judiciously,

the viral transneuronal labeling method can produce information regard-
ing central neural circuits that is unattainable by other methods. Its high
sensitivity is, however, inseparable from a significant potential for nonspe-
cific labeling of unrelated neural circuits. The need for conservative in-
terpretation of the patterns of central labeling is important because even
attenuated viral tracers are capable of infecting many different cell types in
the brain (neurons, astroglia, and ependymal cells) by various routes. An
analysis of a series of sections throughout the brain should be performed
for each PRV case in any given study to rule out the possibility that non-
specific viral labeling had occurred. For example, when PRV is injected
into peripheral targets, such as an autonomic ganglion, inspection of the
supraoptic and magnocellular paraventricular subnucleus can be used to
determine whether a viremia had occurred, since labeling in these two sites
would be the result of uptake from the vascular system. Similar screening
scenarios are important for the evaluation of other types of experiments as
well.
A. Viral Tracing in the CNS
The interpretation of data obtained from viral transneuronal infec-

tions within the CNS can be extremely difficult, compared with viral in-
jections into peripheral structures. Viral entry into the CNS from the
periphery can be isolated to a single neural channel, but the situation
is not as straightforward as in the brain. Central neurons receive input
from multiple CNS regions. Frequently, these regions are interconnected,
greatly increasing the complexity of the potential routes of viral spread
(Fig. 9.5).
In the late 1990s, it was demonstrated that Bartha PRV could be used to
define central circuits (Jasmin et al., 1997; Kaufman et al., 1996; O’Donnell
et al., 1997), although earlier studies using HSV1 in monkeys had established
the feasibility of this approach (Lynch et al., 1994; Middleton and Strick,
1994). Subsequent evaluation of PRV tracing within the brain addressed
significant concerns about this methodology, such as injection site analysis,
nonspecific spread through the cerebrospinal fluid, and viral uptake by
fibers of passage (Chen et al., 1999).
4°
3° A
D 2° B
E
1°
C
Viral injection site
Figure 9.5. The brain’s complex circuitry frequently offers multiple alternative hy-
potheses for the route by which a virus may have labeled a particular group of
neurons. This complexity can complicate the interpretation of tracing data. In this
diagram, each neuron symbolizes a neuroanatomical region with known axonal
projections to a lower order site. Common types of neural connections that can
complicate tracing data are shown as dashed lines. A simple time-course analysis of
viral progression may differentiate between the potential routes of labeling in (C),
in which two transneuronal steps separate the alternatives. However, this approach
may not provide a clear answer for (B) and cannot resolve (A). The resolution of
labeling in these situations may require additional neuroanatomical experiments.
The reciprocally connected pair of neuronal groups depicted by (D) may lead to
uncertainty as to whether a specific subset of neurons within the second-order group
was labeled (1) directly from neurons in the first-order group or (2) from their target
neurons in the infected third-order group, which had received virus from a different
subset of second-order neurons in the same region. In (E), a similar hypothetical
situation is depicted.
1. PRV Injection Site
Chen et al. (1999) directly addressed a number of potential pitfalls as-

sociated with Bartha PRV tracing within the brain. One of the critical
issues is defining an injection site. This was problematic since after PRV
was injected into brain parenchyma it rapidly entered local axons in which
virions were transported away from the injection site. Several days later,
when most PRV tracing experiments were terminated, immunohistochemi-
cal staining for PRV did not reveal the injection site. To avoid this problem,
these investigators verified that a cocktail of PRV in a 0.05% CTb solution
was useful for approximating the injection site (Chen et al., 1999).
2. Bartha PRV in the Ventricular System
Another complicating issue associated with central PRV tracing studies

is the possibility that the virus could enter the cerebrospinal fluid and
cause nonspecific infections throughout the brain. Chen et al. (1999) in-
jected Bartha PRV into the lateral ventricle of rats and found that, after 1
or 2 days, infections were confined to specific sites and not randomly dis-
tributed throughout the brain. Highly reproducible PRV labeling occurred,
within a day after ventricular injection, in a specific subset of dorsal raphe
neurons immediately beneath the cerebral aqueduct. These neurons are re-
sponsible for the serotonergic axonal plexus in the ependymal lining of the
ventricular system (Chan-Palay, 1976), as demonstrated by rapid labeling of
the same group of neurons, as well as their ependymal axonal plexus, by CTb
(mixed with PRV for injection site localization) in the same animals. The
presence of PRV and/or CTb within this specific ependymal-projecting sub-
set of dorsal raphe neurons was proposed as practical marker for screening
viral tracing cases after injection near a ventricle (Aston-Jones and Card,
2000; Chen et al., 1999). After 2 or more days postinjection, PRV-labeled
neurons were also found scattered in other regions, such as lateral septum
and hippocampus. This labeling was hypothesized to have resulted from
the uptake of PRV from infected ependymal cells that had lysed. Regard-
less of the cause for this labeling, these experiments showed that Bartha
PRV injections into the brain ventricular system did not cause widespread
infections.
3. Bartha PRV in the Vasculature
Vascular leakage of Bartha PRV is another problem, since central injec-

tions invariably cause disruption of some blood vessels. To date, six studies
have shown that when Bartha PRV is injected into the venous system of rat
in the doses used in central tracing studies it does not produce central in-
fections (Westerhaus and Loewy, 1999). Inoculation of similar amounts of
Bartha PRV directly into the arterial supply of the brain, however, has not
been tested.
4. Uptake by Fibers of Passage
Another important issue is whether Bartha PRV is taken up by fibers

of passage within the injection site. Peripheral nerves take up this virus
when high doses are used (Dobbins and Feldman, 1994). Only one study
has examined this issue for CNS injection sites. Bartha PRV was injected
into the ventrolateral medulla of rats, in the region where crossed axons
from the inferior olivary nucleus travel toward the inferior cerebellar pe-
duncle. Viral injection here resulted in robust retrograde neuronal labeling
in the contralateral inferior olivary nucleus (Chen et al., 1999). Whether
this uptake was due to entry into injured axons or whether uptake by in-
tact fibers of passage is not certain, but these findings highlight a potential
confound when PRV is used as a central tracer. Stereotaxic injections will
cause a certain degree of damage to fibers passing near the target site—
this complication should be considered when PRV is used as a central
tracer.
5. Controlled Viral Tracing in the CNS
Some means of guarding against misinterpretation of nonspecific patterns

of transneuronal infection include corroboration with previously charac-
terized neuroanatomical connections, negative control lesion experiments,
and positive control experiments excluding alternative pathways. Clearly,
viral labeling should be viewed as specific only when underlying single
neuronal connections can be verified by conventional neural tracing. If
infected neurons are found in sites incompatible with a specific retrograde
transsynaptic spread from the viral injection site, based on well-established
neuroanatomical data, labeling should be considered nonspecific. If a hy-
pothetical pathway cannot be convincingly constructed from the viral in-
jection site to a particular infected group of neurons by piecing together
well-characterized traditional tracing data, the specificity of labeling should
be questioned.
When multiple different transneuronal routes could explain viral labeling
data, combining single-neuron tracing with PRV injection can help distin-
guish the most likely pathway (Aston-Jones et al., 2001; Farkas et al., 1998).
Also, lesion of a relay point in a proposed circuit should significantly de-
crease viral infection within a transneuronally labeled group of neurons. For
example, in a study of the multineuronal circuit from the SCN to the medial
prefrontal cortex, the paraventricular thalamic nucleus was hypothesized as
the key relay point. Lesioning this cell group effectively blocked transneu-
ronal labeling of the SCN after PRV injection within medial prefrontal cor-
tex (Sylvester et al., 2002). As shown in this study, lesion effects should be
quantified and replicated in a sufficient number of cases to demonstrate a
statistically significant reduction of viral labeling in lesioned animals versus
nonlesioned animals after an identical postinjection survival time. Injection
site analysis is necessary for every experiment to screen for the possibility
that differences in labeling could be produced between lesioned and non-
lesioned groups simply because of slight differences in the placements of
viral injections. In addition, viral labeling in a positive control site should be
quantified, in both lesioned and nonlesioned groups, to verify that the in-
fection proceeded normally in alternative pathways that are not dependent
upon the lesioned site as a relay.
A positive control lesion experiment, demonstrating that lesion of a relay
within a different potential route does not reduce transneuronal labeling,
can be helpful in verifying a hypothesized multineuronal pathway (Pickard
et al., 2002; Smeraski et al., 2004). Lesions of alternative pathways can be criti-
cal for falsifying alternative hypotheses concerning the transneuronal route
between injection site and labeled neurons (see “Accurate Interpretation
of Viral Labeling Patterns”). However, the existence of multiple parallel
pathways can still complicate data interpretation in some circumstances
(Aston-Jones et al., 2001; Farkas et al., 1998).
The approaches described above are not always practical for all tracing
objectives. Sometimes, control experiments will not guarantee a clear answer
regarding the specific circuit being studied by viral transneuronal labeling.
However, cases should always be screened for indications of nonspecific CNS
labeling.
B. Viral Tracing in the Peripheral Nervous System
Viral tracing projects designed to study the central circuits controlling

motor outflow systems are considerably easier to analyze than projects de-
signed to study CNS circuits because the route of viral entry can be ex-
perimentally limited to a single outflow channel. For example, the central
parasympathetic circuits regulating pancreas, airways, or heart were stud-
ied in isolation from the sympathetic system by performing PRV tracing
experiments on rats with T1 spinal transactions. These lesions completely
eliminated the possibility of viral entry into the brain via the sympathetic
nervous system (Haxhiu et al., 1993; Loewy and Haxhiu, 1993; Ter Horst
et al., 1996). Since Bartha PRV does not enter the CNS via afferent sys-
tems, the data generated in these particular studies clearly produced in-
formation regarding the organization of central parasympathetic motor
systems.
PRV injections into sympathetic structures sometimes resulted in unex-
pected labeling of vagal motor neurons, which has been greatly reduced
after bilateral subdiaphragmatic vagotomy (Geerling et al., 2003). Even af-
ter bilateral vagotomy, occasional cases were generated with a relatively
low number of PRV-infected neurons in the dorsal vagal nucleus (e.g., 10–
20 bilaterally throughout a 1-in-5 series of the full extent of the nucleus).
This residual labeling may have been due to the failure to transect all of the
vagal fibers.
Practical considerations have required that various neuroanatomical cri-
teria be set for what constitutes a specific viral tracing infection after pe-
ripheral PRV injection (Sams et al., 1995; Strack et al., 1989b). For example,
after retrograde tracing from the stellate ganglion, Jansen et al. (1995b) dis-
covered labeled neurons in the red nucleus. Since this nucleus was mainly
considered a somatic premotor nucleus, the labeling was interpreted as
nonspecific and used to screen individual cases for nonspecific viral spread.
Although it is possible that labeling in the red nucleus resulted from specific
transneuronal labeling of a previously unknown sympathetic outflow path-
way, the admittedly subjective criterion served as a useful index for potential
nonspecific labeling in this particular study.
Finally, the neurosecretory magnocellular neurons in the paraventricular
and supraoptic nuclei should be carefully examined after any type of PRV
injection into peripheral structures. Neurons in these two nuclei project

exclusively to the capillary beds in the posterior pituitary. If any cell body
labeling is found at these two sites, it indicates that false-positive labeling
likely occurred due to uptake of PRV from the vascular system.
C. Accurate Interpretation of Viral Labeling Patterns
The data obtained from a PRV transneuronal tracing experiment can be

very complex, even after only two or three rounds of retrograde transport
and viral replication (Fig. 9.5). Exponential replication of the virus occurs,
within a geometrically increasing population of infected afferent neurons,
at every retrograde transmission. Large numbers of neurons in many sites
can be labeled over several days.
Substantial increases in the number of infected neurons can lead to incor-
rect assumptions about the routes by which viral tracer has infected a par-
ticular group of neurons. The neuroanatomical material obtained from a
PRV tracing experiment provides a complicated snapshot of all the neurons
infected with PRV up to the time of death. The labeling pattern itself does
not indicate the order in which particular neurons are infected. Determi-
nation of the multineuronal pathways indicated by such data is particularly
problematic when an infected neuronal group is connected to numerous
other infected sites. Frequently, more than one reasonable hypothesis can
be generated to explain such complicated tracing data.
1. Determining the Route(s) of Transneuronal

Labeling—Experimental Approaches
Multiple experimental approaches can test each hypothesis. First, a tem-

poral estimate of viral progression can sometimes aid the generation of
preliminary hypotheses about the hierarchical connections of an afferent
pathway (Larsen et al., 1998; Pickard et al., 2002; Smeraski et al., 2004). In this
approach, a provisional timeline is created from the comparison of infected
sites in animals killed at progressively longer postinjection time points.
Three problems, however, often prevent a simple temporal approach from
discriminating between various alternative synaptic pathways. First, the rate
of viral progression may be variable among experimental animals. Second,
practical experience has revealed that retrograde infection does not occur
in idealized waves with one stage of replication, and retrograde transfer oc-
curring at one afferent level before infected neurons appear at a higher
level. Rather, infections tend to progress in a continuous manner, with the
number of neurons infected at one hierarchical level growing even after
infected neurons have appeared in groups afferent to that site. Also, af-
ferent neurons located farthest from an infected neuron require longer
axonal transport times, resulting in a delay in labeling. This gradual spread
becomes even more of a blur as the infection spreads through higher order
afferents. When added to the variability in rate of viral progression between
different animals, this problem can obstruct the presumed logical inter-
pretation of a simple temporal approach. Third, synaptic connections in
the brain frequently do not exist in an idealized hierarchical arrangement.
Multiple retrograde avenues to an afferent site and complicated recipro-
cal connections between higher order afferents typify brain architecture
(Fig. 9.5). When two or more potential routes of viral spread are possible,
temporal analysis may be useful in distinguishing between possibilities, but
corroboration by other approaches is often necessary.
Besides a time-course analysis of viral spread, combination with conven-
tional neural tracing and lesion experiments (described above) can aid in
differentiating between alternative pathways. When more than one hypo-
thetical transneuronal route can explain the spread of virus to a labeled
group of neurons, selective lesion studies of the different routes can help
define the circuit (see above).
2. An Example: PRV Transneuronal Labeling in Retinorecipient Sites
One example plainly demonstrates the importance of this critical analyti-

cal approach to viral transneuronal tracing. When comparing the wild-type
and attenuated Bartha strain of PRV, Card et al. (1991) observed two qualita-
tively different patterns of infection after injecting two different viruses into
the vitreous body of the eye. Wild-type PRV produced a rapid infection in
all the sites targeted by the retinal output, such as the lateral geniculate nu-
cleus, superior colliculus, and SCN, consistent with the idea that this virus is
transported in the anterograde direction. Bartha PRV, however, produced a
greatly delayed infection in the SCN and did not produce an infection in the
two main retinorecipient sites— superior colliculus and lateral geniculate.
Importantly, Bartha PRV infections within the SCN occurred only after long
postinjection survival times (3–4 days). Given the high viral dose used— 10 6
plaque forming units (pfu), two orders of magnitude higher than the dose at
which Bartha PRV was originally used for retrograde transneuronal tracing
after 4 days of survival (Strack et al., 1989a, b; Strack and Loewy, 1990)—this
time frame was consistent with the time required for retrograde transneu-
ronal spread to the hypothalamus through a chain of multiple neurons.
Despite the extended time required for viral spread to the SCN, these ob-
servations were interpreted as evidence for anterograde transport of Bartha
PRV. To reconcile this interpretation with the complete lack of anterograde
transmission to the main retinal target sites, it was further assumed that
only a subpopulation of retinal ganglion cells, which project to nonvisual
sites such as the SCN, is vulnerable to a productive infection by Bartha PRV
(Card, 2000; Card et al., 1991). Despite a lack of evidence for anterograde
transneuronal spread in other studies with Bartha PRV, this interpretation
remained unchallenged.
The assumption of a very slow form of anterograde transneuronal spread,

resulting in a selective infection of the retinorecipient neurons of the SCN,
served as the basis for a number of subsequent viral tracing studies of this cir-
cuitry (Card, 2000; Hannibal et al., 2001; Moore et al., 1995; Smith et al., 2000)
and for investigations into the genetic properties conferring this unusual
property upon Bartha PRV (Brideau et al., 2000; Card et al., 1992; Enquist
et al., 1994; Husak et al., 2000; Tomishima and Enquist, 2001; Whealy et al.,
1993). The implication that Bartha PRV might be capable of anterograde
transport casts doubt on several other investigations that provided evidence
that this virus moves exclusively in a retrograde manner (Rotto-Percelay
et al., 1992; Strack et al., 1989b).
It was not until over a decade later that an alternative explanation was
tested. In the intervening years, Enquist et al. (1994), using a series of dele-
tion mutants, evaluated the importance of individual genes to the antero-
grade spread of PRV. These important studies built a strong case for the
necessity of three particular genes, deleted in Bartha PRV, for anterograde
spread of PRV (see “Specific Retrograde Transport of Bartha PRV”). Fur-
ther analysis of the retinal infections produced by viruses lacking two of
these genes indicated that mutants deficient in anterograde transport can
still infect all types of retinal ganglion cells (Enquist et al., 1994; Husak
et al., 2000). This finding did not fit well with the proposal that Bartha PRV
produces anterograde labeling in only a subset of retinorecipient nuclei by
selectively infecting a small subpopulation of retinal ganglion cells (Card
et al., 1991).
Following these reports, Pickard et al. (2000) tested the possibility that
the spread of Bartha PRV from the eye to the SCN and other sites might
not be the result of slow anterograde spread from a specific subset of retinal
ganglion cells, but, instead, due to retrograde transneuronal spread via the
autonomic nerves innervating the eye. In the original tests of Bartha PRV
transneuronal specificity, Strack and Loewy (1990) demonstrated that injec-
tion into the nearby anterior chamber of the eye produced robust retrograde
transneuronal labeling of the sympathetic outflow to the eye. In addition, a
prominent multisynaptic outflow from the SCN to the diverse sympathetic
and parasympathetic targets had been demonstrated (Ueyama et al., 1999).
Accordingly, in both the hamster (Pickard et al., 2002) and the rat (Smeraski
et al., 2004), it was shown that (1) enucleation of the eye 24 h after Bartha
PRV injection (preventing anterograde spread of virus due to degeneration
of the optic axons from the destroyed retinal ganglion cells) did not pre-
vent later infection within the SCN and other retinorecipient sites, (2) PRV
infection in autonomic preganglionic sites— the parasympathetic Edinger–
Westphal nucleus and sympathetic ganglionic and preganglionic neurons—
preceded the SCN infection, and (3) lesions of these autonomic sites prior
to PRV injection virtually eliminated infection in the SCN. In the hamster,
neurons in the retinorecipient portion of the SCN were not even the first
to be infected. Rather, their target neurons in the subparaventricular zone,
dorsal to the SCN, were infected before labeling occurred in the SCN (see
also Card et al., 1991). In addition, the first appearance of PRV within the
rat SCN did not overlap the retinohypothalamic projection (identified by
concurrent anterograde axonal labeling with CTb; Smeraski et al., 2004).
These findings clearly disproved claims that Bartha PRV could be used as
an anterograde transneuronal tracer.
3. Thorough Neuroanatomical Hypothesis-Testing
A provocative pattern of transneuronal labeling can tempt assumptions

about the nature of the underlying pathway from injection site to infected
neurons. As the preceding example demonstrates, however, such assump-
tions should not prevent the rigorous testing of alternative hypotheses. Care-
fully analyzing viral tracing data before asserting confidence in a particular
explanation can be both complicated and time-consuming. However, com-
bining basic viral tracing with thorough and prudent neuroanatomical anal-
ysis can significantly advance our knowledge of complicated circuits within
the CNS (Aston-Jones et al., 2001; Krout et al., 2003; Pickard et al., 2002;
Smeraski et al., 2004; Sylvester et al., 2002).
D. False-Negative Data After Viral Transneuronal Tracing
As with any neural tracing technique, the degree of uptake and subse-
quent labeling of afferents to an injection site is dependent, in part, upon
the amount of tracer used. Hence, with small tracer injections, a lack of
labeling can be observed in sites known to provide lighter innervation to an
injection site. For PRV tracing, this was first noted by O’Donnell et al. (1997),
when injections of Bartha PRV into the mediodorsal thalamic nucleus did
not produce the retrograde labeling that was expected, based on prior
retrograde tracing studies, within the basolateral amygdala, a light source
of innervation. Despite substantial retrograde transneuronal infections via
the dense pallidal afferents, the absence of basolateral amygdala labeling
suggested that virions either selectively avoided particular afferent system
or stochastically entered only a proportion of afferent terminals in a given
site, based on the relative amount of virus and the density of axon terminals.
This latter possibility was tested by Card et al. (1999), who injected a range
of different Bartha PRV concentrations (104 –105 pfu) into the striatum. At
2 days postinjection, a clear dose dependency was observed for extent of
viral transneuronal labeling in various sites afferent to the striatum.
Viral concentration and postinjection survival time are two critical vari-
ables that affect optimal transneuronal labeling. Since only a few papers
have dealt with this subject, it is not possible to make generalizations at
this time regarding the optimal conditions to label any given CNS circuit.
Rather, these important experimental parameters need to be empirically
determined, but a reasonable starting point for most experiments would
involve injections of ∼3000 virions of Bartha PRV and a survival range of

2–4 days.
VII. OTHER VIRUSES USED FOR TRANSNEURONAL

TRACING STUDIES
Bartha PRV remains the only virus subjected to direct tests of its speci-
ficity as a retrograde transneuronal tracer (Card et al., 1993; Chen et al., 1999;
Pickard et al., 2002; Rinaman et al., 1993; Rotto-Percelay et al., 1992; Smeraski
et al., 2004; Strack et al., 1989b; Strack and Loewy, 1990). However, various
other viruses are also used for transneuronal tracing studies. Experiments
with HSV1 and rabies have been used to produce transneuronal labeling
with varying indications of specificity. Additional direct verifications of their
directional and transsynaptic specificity could be highly useful, particularly
since, unlike PRV strains, these can be used for tracing experiments in pri-
mates. In addition, restrictions on Bartha PRV usage in countries where PRV
has been eradicated from most pig and cattle populations may leave these
viruses as the only practical options for certain laboratories.
A. HSV1 as a Transneuronal Tracer
HSV1 has been used for transneuronal tracing in various species. Different
HSV1 strains have been used for transneuronal studies in primates by Strick
and colleagues (Clower et al., 2001; Hoover and Strick, 1993, 1999; Lynch
et al., 1994; Middleton and Strick, 1994, 1996, 2001, 2002).
The transneuronal pattern of labeling produced by this virus is highly
dependent on the specific strain used for tracing (Norgren and Lehman,
1998). The SC16 strain of HSV1, used in early studies by Ugolini, produced
both retrograde and anterograde transneuronal labeling in the brainstem
and cerebellum (Ugolini et al., 1987). Another HSV1 strain, FMC, was used
for retrograde transneuronal labeling of central neurons afferent to various
autonomic targets in a series of studies by Blessing and colleagues (Blessing
et al., 1991; Ding et al., 1993; Li et al., 1992a, b, 1993; Wesselingh et al., 1989).
The patterns of infection produced by the injection of different strains
of HSV1 into monkey cortex indicated that the McIntyre-B strain prefer-
entially caused a retrograde transneuronal pattern of labeling while the
H129 strain produced an anterograde labeling pattern (Zemanick et al.,
1991). Subsequent analysis, however, revealed that neither virus is trans-
ported exclusively in one direction, despite a significant difference in di-
rectional preference. McIntyre HSV1 can produce transneuronal labeling
in the anterograde direction (Norgren et al., 1992). Also, H129 clearly pro-
duces a retrograde infection within first-order afferent neurons (Rinaman
and Schwartz, 2004). This strain, unlike Bartha PRV, was not observed to
spread transneuronally from retrogradely infected first-order afferent vagal
motor neurons in rats with lesioned vagal afferent fibers, following injection
into the stomach wall (Rinaman and Schwartz, 2004). However, the pattern
of transneuronal infection produced by H129 in this study, after a presumed
anterograde transneuronal infection within the nucleus of the solitary tract
(NTS), may be more consistent with retrograde transneuronal labeling of
neurons afferent to this site, rather than simply anterograde spread in NTS
neurons to their efferent targets [e.g., the strong infection depicted in Fig. 2
of Rinaman and Schwartz, 2004, within a dorsal part of the bed nuclei, an
NTS afferent site— versus the dense NTS innervation in a more ventral re-
gion of the bed nuclei (Ricardo and Koh, 1978)]. Further testing of the
directional specificity of transneuronal labeling produced by HSV1 strain
H129 should reveal whether or not this virus will be useful as an anterograde
transneuronal tracer.
One drawback of viral tracing with various strains of HSV1 is the
lack of neuroanatomical experiments directly addressing the specificity of
transneuronal labeling. In Ugolini’s original HSV1 tracing study, a signif-
icant degree of nonspecific local spread of virus was reported (Ugolini
et al., 1987). This nonspecific spread resulted in false-positive anterograde
transneuronal labeling. Further tracing work with this strain was then con-
ducted without direct tests of the specificity of transneuronal labeling
(Ugolini et al., 1989). The potential for nonspecific labeling by various strains
of HSV1— via both local and transneuronal routes— limits the utility of this
virus for many neural tracing objectives (Fig. 9.4).
B. Rabies as a Retrograde Transneuronal Tracer
Although the name “pseudorabies” may seem to imply a functional rela-

tionship between PRV and rabies, these two viruses are very different. Like
HSV1, PRV is a member of the Alphaherpesvirinae family of neurotropic
herpesviruses. It contains a double-stranded DNA genome, which is tran-
scribed and replicated in the cell nucleus, and can cause lytic cell death
shortly after infection or establish latency in vivo in neuronal tissue. The
reason for the name “pseudorabies” was the CNS infection it produced in
farm species at a time when few viruses (rabies being one of them) were
known to invade the brain (Aujesky, 1902).
Rabies, in contrast, is a rhabdovirus—a single-stranded, negative-sense
RNA virus that replicates in the cytoplasm. Unlike herpesviruses, rabies
infections of the CNS, while lethal, do not appear to cause widespread
cell death. Hence, this virus has been used for retrograde transneuronal
labeling in various paradigms, in both rodent and primate. The earliest
tracing study with rabies demonstrated anterograde transneuronal infec-
tion within the brain after injection of a challenge virus strain (CVS) of
rabies into mouse olfactory epithelium (Astic et al., 1993). When CVS rabies
was injected into the hypoglossal nerve, retrograde transneuronal labeling
was produced in rats without obvious nonspecific spread (Ugolini, 1995).
In particular, no infected glial cells were observed and infection did not
appear to spread locally, even several days after the onset of infection within
primary infected neurons in the hypoglossal nucleus. This result stood in
striking contrast to the nonspecific viral labeling originally observed with
HSV1 (Ugolini et al., 1987). Studies using rabies to produce transneuronal
labeling have demonstrated a potential for its use in rodent and primate
neural tracing experiments (Astic et al., 1993; Graf et al., 2002; Grantyn et
al., 2002; Kelly and Strick, 2003; Moschovakis et al., 2004; Tang et al., 1999;
Ugolini, 1995). Rabies central transneuronal tracing methodology has been
thoroughly reviewed by Kelly and Strick (2000).
C. Perspectives—HSV1 and Rabies
There are two major drawbacks associated with the use of HSV1 and rabies
viruses as transneuronal tracers. First and foremost, these viruses infect hu-
mans, representing a potential hazard to laboratory personnel and requiring
additional precautions, especially for rabies, which requires repeated vacci-
nations and strict precautions (Kelly and Strick, 2000). For transneuronal
studies in nonprimate species, the use of Bartha PRV does not present this
problem, since it does not infect humans (Gustafson, 1975). Second, infor-
mation regarding the neuroanatomical specificity of labeling produced by
these viruses is incomplete. Rabies and HSV1 have not yet been subjected
to many of the experimental tests used to characterize Bartha PRV as a neu-
roanatomical tracer (Card et al., 1993; Chen et al., 1999; Pickard et al., 2002;
Rinaman et al., 1993; Rotto-Percelay et al., 1992; Smeraski et al., 2004; Strack
et al., 1989b; Strack and Loewy, 1990).
Varying degrees of transneuronal and directional specificity have been in-
ferred from the patterns of infection observed in various tracing paradigms
(Ugolini, 1995; Ugolini et al., 1987; Zemanick et al., 1991). For HSV1, a prob-
lematic degree of nonspecific local spread, resulting in subsequent nonspe-
cific transneuronal labeling, has been described (Ugolini et al., 1987). In
contrast, some strains of rabies may spread only in the retrograde direction
in some paradigms (Kelly and Strick, 2000; Ugolini, 1995), although cer-
tain strains can clearly produce anterograde transneuronal labeling (Astic
et al., 1993). The transneuronal specificity of infection with rabies appears
promising, especially in comparison with HSV1 (Ugolini, 1995), but has yet
to be directly tested.
A high degree of transneuronal specificity may not be a prerequisite for
transneuronal tracing in primate circuits involving massively parallel cir-
cuitry, such as primate cortical, basal ganglia, cerebellar, and thalamic path-
ways. So long as the bulk of viral transneuronal transfer occurs, stochasti-
cally, in a transsynaptic manner, it is possible that nonspecific viral spread
is largely constrained to immediately adjacent portions of parallel pathways
within the same circuits. In any case, rabies and HSV1 are currently the only
viable options for transneuronal labeling experiments in primates.
VIII. CONCLUSION
Neurotropic viruses are extremely useful neural tracers for a variety of neu-
roanatomical objectives. Strains with well-characterized properties, such as
Bartha PRV, can be used to gain valuable new data about mammalian neural
circuits. When tracing studies are executed and interpreted within known
technical limitations, they can provide information currently unattainable
by other methods.
Combined advances in virology and molecular biology may allow the de-
sign of viruses that will provide selective information about particular neural
networks, revealing CNS circuitry in unprecedented detail.
APPENDIX
A. Safety and Practical Issues
Bartha PRV has been successfully used to eradicate PRV from most pig and
cattle populations in many countries, and consequently, various rules and
restrictions have been developed concerning its use. In the United States, a
BSL-2 laboratory is required for use of this agent.
B. Sources of Bartha PRV and Recombinant Strains
Bartha PRV and related recombinant strains can be obtained directly from
individual researchers who work with this virus. Investigators can contact
Drs. Arthur Loewy (USA) and Thomas Mettenleiter (Germany).
In addition, the Center for Neuroanatomy with Neurotropic Viruses was
established at the University of Pittsburgh in the summer of 2004. Under
the direction of Drs. J. Patrick Card (card@ bns.pitt.edu) and Peter Strick
(strickp@
pitt.edu), the Center will serve as a resource for investigators in-
terested in obtaining various viral strains for tracing experiments.
C. Viral Growth, Aliquots, and Storage
The broad host range of PRV in vivo is reflected in a broad spectrum of

cells that can be productively infected in vitro. One of the advantages of work-
ing with PRV is its ability to replicate to rather high titers in easily cultivable
permanent cell lines. Primarily, kidney cell lines from rodents (rabbit RK-13
cells), ruminants (Madin-Darby bovine kidney, MDBK), or porcines (PSEK,
PK-15) are used. However, other cell lines, e.g., monkey kidney cells (Vero),
are also permissive for PRV infection. The highest titers of progeny virus
are usually obtained on porcine cells, whereas RK-13 cells are preferentially
used for transfection and for establishment of transgenic cells expressing
viral genes for transcomplementation of respective viral deletion mutants.

MDBK cells are ideal for plaque titration since they produce the clearest
and most easily visible plaques. However, other cell lines can also be used.
All the three mentioned cell lines can be cultivated in Eagle’s minimum
essential medium (MEM) supplemented with either 5% (for MDBK, PSEK,
and PK-15) or 10% fetal calf serum (for RK-13). They routinely grow well on
disposable plastic tissue culture flasks and can be propagated by trypsiniza-
tion [0.8-g NaCl, 0.4-g NaCl, 1-g dextrose, 0.58-g NaHCO 3 , 0.5-g trypsin, 0.2-g
EDTA, and 1l H2 O (pH 7.1–7.2), sterile filtered]from the culture flasks, di-
lution at a ratio between 1:3 and 1:10 with fresh medium supplemented with
fetal calf serum, and reseeding.
To grow high-titered virus stocks, cells are infected with PRV at a multi-
plicity of infection between 0.01 and 0.1 pfu per cell. After 1 h of adsorption,
the inoculum is removed and the cells are overlaid with fresh medium. After
2–3 days, a complete cytopathic effect develops with cells first exhibiting a
rounded appearance, which, as viral infection progresses, leads to lysis of
the cells. After lysis of the cell monolayer, the whole culture flask is frozen
and thawed, supernatant and cell debris are collected in a plastic tube (e.g.,
Falcon 50-ml tube) and cellular debris is sedimented by low-speed centrifu-
gation (e.g., Heraeus Minifuge, 10 min, 6000 rpm). All virus isolation steps
should be performed on ice or in a refrigerated centrifuge (+4◦ C). The
supernatant is removed and immediately aliquoted, routinely in between
0.5- and 1-ml aliquots, and frozen at −70◦ C or in liquid nitrogen. Storage at
−20◦ C leads to rapid loss of infectivity.
For determination of the infectious titer, one aliquot is thawed on ice
and serial 10-fold dilutions in medium are prepared. These are then plated
onto monolayer cells preferably in 6- or 24-well tissue culture dishes. As men-
tioned above, MDBK cells are most suited for this purpose, although other
cell types can also be used. After 1-h incubation at 37◦ C, the supernatant
is removed and substituted by medium supplemented with 5% methylcel-
lulose: 10-g methylcellulose, 3.76-g autoclavable MEM powder suspended
in 390-ml H2 O are autoclaved (use magnetic stir bar that remains in a bot-
tle). After cooling to room temperature, 200 mM l-glutamine and 880-mg
NaHCO3 (dissolved in 6-ml H2 O and sterile filtered) are then added. The
stock solution is stored at 4◦ C and is diluted 1:4 in MEM/5% fetal calf serum
for use.
Cells are then incubated in a 5% CO2 atmosphere for 2–3 days. When
plaques are clearly visible, the medium is removed, the monolayer is washed
3× with PBS, and cells are then fixed with 5% formaldehyde for 20 min,
washed with PBS, and stained with 1% crystal violet in 50% ethanol for
5 min. Staining solution is removed and monolayers are washed extensively.
Plaques are white on a blue background.
Routinely, titers up to, and sometimes in excess of, 107 pfu/ml can be
obtained. If higher virus titers are required, the virus suspension can be
concentrated by ultracentrifugation (e.g., for 1 h at 22,000 rpm in a Beckman
TST-28 rotor). It is imperative that all steps are performed either on ice or at
+4◦ C. The virus pellet is gently resuspended in the desired volume of either
TBSal 2[ 00 mM NaCl, 2.6 mM KCl, 10 mM Tris-HCl (pH 7.5), 20 mM MgCl 2 ,
1.8 mM CaCl2 ]or MEM. Then, individual aliquots of 5 µl (or other desired
amount) are made in plastic microcentrifuge tubes and stored at −70◦ C.
D. Dilution of PRV
A tube containing a single aliquot of PRV is allowed to thaw on crushed

ice. For CNS injections, 2 µl of 0.1% cholera toxin β-subunit (CTb, product
103B, List Biologicals Inc., Campbell, CA) is added. The CTb is used as
#
a marker for the injection site and can also provide limited single-neuron
tracing data, in addition to viral transneuronal labeling, in the same animal.
These proportions (adding 2 µl of 0.1% CTb solution to a 5-µl viral suspen-
sion) result in a final CTb concentration of about 0.03%— slightly less than
the 0.05% solution recommended by Chen et al. (1999). In situations where
injection site determination is not required, a similar dilution can be made
with sterile Dulbecco’s Modified Eagle Medium.
It is important to note that addition of a 2-µl CTb solution dilutes the
concentration of the viral suspension, a critical variable in interpreting viral
tracing experiments (see discussions above). For example, adding 2 µl to a
5-µl suspension of 108 pfu/ml Bartha PRV will reduce the viral concentration
to 7 × 107 pfu/ml. This change is relevant to the calculation of the amount
of virus injected into experimental animals.
E. Injection of PRV
Injections are made with a glass micropipette that has been filled with
the aid of an operating microscope. The pipette is secured to a microma-
nipulator attached to a stereotaxic frame, and then advanced into a specific
brain target in a surgically prepared animal. The micropipette is attached by
polyethylene plastic tubing to an air pressure regulator so that the virus can
be ejected by applying pressure from a handheld 50-ml syringe. Commer-
cial equipment, such as the Picospritizer (General Valve Corp., Fairfield,
NJ), can be used as well. Alternatively, a glass micropipette can be glued
onto a 1-µl Hamilton microsyringe (Fisher Scientific, Pittsburgh, PA) and
used in a similar capacity. The advantage of using a glass micropipette is
that a carefully measured volume of virus can be delivered under micro-
scopic control. Generally, 40-nl injections of the solution described in sec-
tion “Dilution of PRV” have produced good results. A 40-nl injection of PRV
without added CTb (108 pfu/ml) contains about 4000 pfu. If diluted by
the addition of CTb as described above, this same volume contains about
3000 pfu. Rats receiving PRV injections should be surveyed daily for signs
and symptoms of viral infection, such as nasal inflammation, itching, and
sneezing.
F. Preparation of Brain Sections for Immunohistochemical Staining
Survival periods usually range from 3 to 4 days, for CNS injections, up to

as long as 7–8 days for peripheral injections, depending upon variables such
as the distance of the injected target from the brain and the desired extent
of retrograde transneuronal labeling. At the end of the survival period,
brain sections are processed for immunohistochemical staining in the same
manner as other neuroanatomical tracing methods.
Briefly, anesthetized animals are killed by perfusion through the heart
with saline, followed by 4% paraformaldehyde made in 0.1 M sodium phos-
phate buffer (pH 7.4). The brain is removed, stored in 4% paraformalde-
hyde fixative for 2 days or more, and sectioned at 50 µm using a freezing mi-
crotome or cryostat. Histological sections are collected and stored in plastic
tissue culture trays containing 0.1 M sodium phosphate buffer (pH 7.4) con-
taining 0.1% sodium azide, which acts as an antibacterial agent. The quality
of the histological staining tends to be reduced after longer storage periods.
G. Immunohistochemical Staining
Immunohistochemical staining follows standard protocols. For visualiz-

ing injection sites, CTb can be localized with a goat antibody supplied by
List Biologicals (Campbell, CA), used at 1:40,000, and the avidin–biotin
complex method (ABC, Vectastain kit, Vector Laboratories, Burlingame,
CA). For visualizing PRV, monoclonal antibodies can be purchased from
Chemicon (Temecula, CA). Detailed protocols for combined PRV and neu-
ropeptide immunohistochemistry appear in several reports (Geerling et al.,
2003; Oldfield et al., 2002; Sylvester et al., 2002).
H. Disposal of PRV-Infected Material
Preparatory steps are taken before the animals are perfused—

plastic bags
containing absorbent material (newspaper) are used to collect all fluids. Af-
ter CNS removal, animal carcasses and these fluids are disposed in biohaz-
ard waste containers. Cages, water bottles, and bedding are sterilized and
cleaned in a central veterinary facility. After completion of experiments,
the tissue trays containing brain sections from PRV-infected animals are
submerged in bleach until the tissue dissolves.
REFERENCES
Astic, L., Saucier, D., Coulon, P., Lafay, F., and Flamand, A., 1993, The CVS strain of rabies virus
as transneuronal tracer in the olfactory system of mice, Brain Res. 619(1–2):146–156.
circuits in brain: opportunities and limitations, J. Neurosci. Methods 103(1):51–61.
Aston-Jones, G., Chen, S., Zhu, Y., and Oshinsky, M. L., 2001, A neural circuit for circadian
regulation of arousal, Nat. Neurosci. 4(7):732–738.
Aston-Jones, G., Zhu, Y., and Card, J. P., 2004, Numerous GABAergic afferents to locus ceruleus
in the pericerulear dendritic zone: possible interneuronal pool, J. Neurosci. 24(9):2313–
2321.
Aujesky, A., 1902, Not readily distinguishable from rabies, with unknown origin (in Hungarian),
Veteranarius 25:387–396.
Babic, N., Klupp, B., Brack, A., Mettenleiter, T. C., Ugolini, G., and Flamand, A., 1996, Deletion
of glycoprotein gE reduces the propagation of pseudorabies virus in the nervous system
of mice after intranasal inoculation, Virology 219(1):279–284.
Bak, I. J., Markham, C. H., Cook, M. L., and Stevens, J. G., 1977, Intraaxonal transport of Herpes
simplex virus in the rat central nervous system, Brain Res. 136(3):415–429.
Banfield, B. W., Kaufman, J. D., Randall, J. A., and Pickard, G. E., 2003, Development of
pseudorabies virus strains expressing red fluorescent proteins: new tools for multisynaptic
labeling applications, J. Virol. 77(18):10106–10112.
Billig, I., Foris, J. M., Enquist, L. W., Card, J. P., and Yates, B. J., 2000, Definition of neuronal
circuitry controlling the activity of phrenic and abdominal motoneurons in the ferret
using recombinant strains of pseudorabies virus, J. Neurosci. 20(19):7446–7454.
Blessing, W. W., Li, Y. W., and Wesselingh, S. L., 1991, Transneuronal transport of herpes
simplex virus from the cervical vagus to brain neurons with axonal inputs to central vagal
sensory nuclei in the rat, Neuroscience 42(1):261–274.
Boldogkoi, Z., Reichart, A., Toth, I. E., Sik, A., Erdelyi, F., Medveczky, I., Llorens-Cortes, C.,
Palkovits, M., and Lenkei, Z., 2002, Construction of recombinant pseudorabies viruses
optimized for labeling and neurochemical characterization of neural circuitry, Brain Res.
Mol. Brain Res. 109(1–2):105–118.
Boldogkoi, Z., Sik, A., Denes, A., Reichart, A., Toldi, J., Gerendai, I., Kovacs, K. J., and Palkovits,
M., 2004, Novel tracing paradigms— genetically engineered herpesviruses as tools for map-
ping functional circuits within the CNS: present status and future prospects, Prog. Neurobiol.
72(6):417–445.
Bourgeais, L., Gauriau, C., Monconduit, L., Villanueva, L., and Bernard, J. F., 2003, Dendritic
domains of nociceptive-responsive parabrachial neurons match terminal fields of lamina
I neurons in the rat, J. Comp. Neurol. 464(2):238–256.
Braz, J. M., Rico, B., and Basbaum, A. I., 2002, Transneuronal tracing of diverse CNS circuits
by Cre-mediated induction of wheat germ agglutinin in transgenic mice, Proc. Natl. Acad.
Sci. U. S. A. 99(23):15148–15153.
Brideau, A. D., Card, J. P., and Enquist, L. W., 2000, Role of pseudorabies virus Us9, a type II
membrane protein, in infection of tissue culture cells and the rat nervous system, J. Virol.
74(2):834–845.
Broussard, D. L., Li, X., and Altschuler, S. M., 1996, Localization of GABAA alpha 1 mRNA
subunit in the brainstem nuclei controlling esophageal peristalsis, Brain Res. Mol. Brain
Res. 40(1):143–147.
Campadelli-Fiume, G., Arsenakis, M., Farabegoli, F., and Roizman, B., 1988, Entry of herpes
simplex virus 1 in BJ cells that constitutively express viral glycoprotein D is by endocytosis
and results in degradation of the virus, J. Virol. 62(1):159–167.
Cano, G., Passerin, A. M., Schiltz, J. C., Card, J. P., Morrison, S. F., and Sved, A. F., 2003, Anatomi-
cal substrates for the central control of sympathetic outflow to interscapular adipose tissue
during cold exposure, J. Comp. Neurol. 460(3):303–326.
Card, J. P., 1998, Practical considerations for the use of pseudorabies virus in transneuronal
studies of neural circuitry, Neurosci. Biobehav. Rev. 22(6):685–694.
Card, J. P., 2000, Pseudorabies virus and the functional architecture of the circadian timing
system, J. Biol. Rhythms 15(6):453–461.
Card, J. P., Enquist, L. W., and Moore, R. Y., 1999, Neuroinvasiveness of pseudorabies virus
injected intracerebrally is dependent on viral concentration and terminal field density, J.
Comp. Neurol. 407(3):438–452.
Card, J. P., Levitt, P., and Enquist, L. W., 1998, Different patterns of neuronal infection after
intracerebral injection of two strains of pseudorabies virus, J. Virol. 72(5):4434–4441.
Card, J. P., Rinaman, L., Lynn, R. B., Lee, B. H., Meade, R. P., Miselis, R. R., and Enquist, L.
W., 1993, Pseudorabies virus infection of the rat central nervous system: ultrastructural
characterization of viral replication, transport, and pathogenesis, J. Neurosci. 13(6):2515–
2539.
Card, J. P., Whealy, M. E., Robbins, A. K., and Enquist, L. W., 1992, Pseudorabies virus envelope
glycoprotein gI influences both neurotropism and virulence during infection of the rat
visual system, J. Virol. 66(5):3032–3041.
Card, J. P., Whealy, M. E., Robbins, A. K., Moore, R. Y., and Enquist, L. W., 1991, Two
alpha-herpesvirus strains are transported differentially in the rodent visual system, Neu-
ron 6(6):957–969.
Carr, D. B., O’Donnell, P., Card, J. P., and Sesack, S. R., 1999, Dopamine terminals in the rat
prefrontal cortex synapse on pyramidal cells that project to the nucleus accumbens, J.
Neurosci. 19(24):11049–11060.
Carr, D. B. and Sesack, S. R., 2000, Dopamine terminals synapse on callosal projection neurons
in the rat prefrontal cortex, J. Comp. Neurol. 425(2):275–283.
Chan-Palay, V., 1976, Serotonin axons in the supra- and subependymal plexuses and in the
leptomeninges; their roles in local alterations of cerebrospinal fluid and vasomotor activity,
Brain Res. 102(1):103–130.
Chen, S., Yang, M., Miselis, R. R., and Aston-Jones, G., 1999, Characterization of transsynaptic
tracing with central application of pseudorabies virus, Brain Res. 838(1–2):171–183.
Clower, D. M., West, R. A., Lynch, J. C., and Strick, P. L., 2001, The inferior parietal lobule is the
target of output from the superior colliculus, hippocampus, and cerebellum, J. Neurosci.
21(16):6283–6291.
Cook, M. L. and Stevens, J. G., 1973, Pathogenesis of herpetic neuritis and ganglionitis in mice:
evidence for intra-axonal transport of infection, Infect. Immun. 7(2):272–288.
DeFalco, J., Tomishima, M., Liu, H., Zhao, C., Cai, X., Marth, J. D., Enquist, L., and Friedman, J.
M., 2001, Virus-assisted mapping of neural inputs to a feeding center in the hypothalamus,
Science 291(5513):2608–2613.
Demmin, G. L., Clase, A. C., Randall, J. A., Enquist, L. W., and Banfield, B. W., 2001, Insertions
in the gG gene of pseudorabies virus reduce expression of the upstream Us3 protein and
inhibit cell-to-cell spread of virus infection, J. Virol. 75(22):10856–10869.
Ding, Z. Q., Li, Y. W., Wesselingh, S. L., and Blessing, W. W., 1993, Transneuronal labelling of
neurons in rabbit brain after injection of herpes simplex virus type 1 into the renal nerve,
J. Auton. Nerv. Syst. 42(1):23–31.
Dobbins, E. G., and Feldman, J. L., 1994, Brainstem network controlling descending drive to
phrenic motoneurons in rat, J. Comp. Neurol. 347(1):64–86.
Enquist, L. W., 2002, Exploiting circuit-specific spread of pseudorabies virus in the central ner-
vous system: insights to pathogenesis and circuit tracers, J. Infect. Dis. 186(Suppl. 2):S209–
S214.
Enquist, L. W., Dubin, J., Whealy, M. E., and Card, J. P., 1994, Complementation analysis
of pseudorabies virus gE and gI mutants in retinal ganglion cell neurotropism, J. Virol.
68(8):5275–5279.
Enquist, L. W., Tomishima, M. J., Gross, S., and Smith, G. A., 2002, Directional spread of an
alpha-herpesvirus in the nervous system, Vet. Microbiol. 86(1–2):5–16.
Farkas, E., Jansen, A. S., and Loewy, A. D., 1998, Periaqueductal gray matter input to cardiac-
related sympathetic premotor neurons, Brain Res. 792(2):179–192.
Fukuda, J., Kurata, T., and Yamaguchi, K., 1983, Specific reduction in Na currents after infection
with herpes simplex virus in cultured mammalian nerve cells, Brain Res. 268(2): 367–371.
Geerling, J. C., Mettenleiter, T. C., and Loewy, A. D., 2003, Orexin neurons project to diverse
sympathetic outflow systems, Neuroscience 122(2):541–550.
Gerfen, C. R., O’Leary, D. D., and Cowan, W. M., 1982, A note on the transneuronal transport of
wheat germ agglutinin-conjugated horseradish peroxidase in the avian and rodent visual
systems, Exp. Brain Res. 48(3):443–448.
Giles, M. E., Sly, D. J., McKinley, M. J., and Oldfield, B. J., 2001, A method for the identification
of pseudorabies virus protein and angiotensin AT(1A) receptor mRNA expression in the
same CNS neurons, Brain Res. Brain Res. Protoc. 8(3):153–158.
Goodpasture, E. W., 1925, The axis-cylinders of peripheral nerves as portals of entry to the
central nervous system for the virus of herpes simplex in experimentally infected rabbits,
Am. J. Pathol. 1:11–33.
Goodpasture, E. W., and Teague, O., 1923, Transmission of the virus of herpes fibrils along
nerves in experimentally infected rabbits, J. Med. Res. 44:139–184.
Graf, W., Gerrits, N., Yatim-Dhiba, N., and Ugolini, G., 2002, Mapping the oculomotor
system: the power of transneuronal labelling with rabies virus, Eur. J. Neurosci. 15(9):
1557–1562.
Grantyn, A., Brandi, A. M., Dubayle, D., Graf, W., Ugolini, G., Hadjidimitrakis, K., and
Moschovakis, A., 2002, Density gradients of trans-synaptically labeled collicular neurons
after injections of rabies virus in the lateral rectus muscle of the rhesus monkey, J. Comp.
Neurol. 451(4):346–361.
Gustafson, D. P., 1975, Pseudorabies, In: Dunne, H. W., and Leman, A. D. (eds.), Diseases of
Swine, Ames, IA: Iowa State University Press, pp. 209–223.
Hannibal, J., Vrang, N., Card, J. P., and Fahrenkrug, J., 2001, Light-dependent induction of
cFos during subjective day and night in PACAP-containing ganglion cells of the retinohy-
pothalamic tract, J. Biol. Rhythms 16(5):457–470.
Haxhiu, M. A., Jansen, A. S., Cherniack, N. S., and Loewy, A. D., 1993, CNS innervation of
airway-related parasympathetic preganglionic neurons: a transneuronal labeling study us-
ing pseudorabies virus, Brain Res. 618(1):115–134.
Hoover, J. E., and Strick, P. L., 1993, Multiple output channels in the basal ganglia, Science
259(5096):819–821.
Hoover, J. E., and Strick, P. L., 1999, The organization of cerebellar and basal ganglia outputs to
primary motor cortex as revealed by retrograde transneuronal transport of herpes simplex
virus type 1, J. Neurosci. 19(4):1446–1463.
Horowitz, L. F., Montmayeur, J. P., Echelard, Y., and Buck, L. B., 1999, A genetic approach to
trace neural circuits, Proc. Natl. Acad. Sci. U. S. A. 96(6):3194–3199.
Husak, P. J., Kuo, T., and Enquist, L. W., 2000, Pseudorabies virus membrane proteins gI and
gE facilitate anterograde spread of infection in projection-specific neurons in the rat, J.
Virol. 74(23):10975–10983.
Irnaten, M., Neff, R. A., Wang, J., Loewy, A. D., Mettenleiter, T. C., and Mendelowitz, D., 2001,
Activity of cardiorespiratory networks revealed by transsynaptic virus expressing GFP, J.
Neurophysiol. 85(1):435–438.
Itaya, S. K., and van Hoesen, G. W., 1982, WGA-HRP as a transneuronal marker in the visual
pathways of monkey and rat, Brain Res. 236(1):199–204.
Jansen, A. S., Farwell, D. G., and Loewy, A. D., 1993, Specificity of pseudorabies virus as a
retrograde marker of sympathetic preganglionic neurons: implications for transneuronal
labeling studies, Brain Res. 617(1):103–112.
Jansen, A. S., Nguyen, X. V., Karpitskiy, V., Mettenleiter, T. C., and Loewy, A. D., 1995a, Central
command neurons of the sympathetic nervous system: basis of the fight-or-flight response,
Science 270(5236):644–646.
Jansen, A. S., Wessendorf, M. W., and Loewy, A. D., 1995b, Transneuronal labeling of CNS
neuropeptide and monoamine neurons after pseudorabies virus injections into the stellate
ganglion, Brain Res. 683(1):1–24.
Jasmin, L., Burkey, A. R., Card, J. P., and Basbaum, A. I., 1997, Transneuronal labeling of a
nociceptive pathway, the spino-(trigemino-)parabrachio-amygdaloid, in the rat, J. Neurosci.
17(10):3751–3765.
Jons, A., and Mettenleiter, T. C., 1997, Green fluorescent protein expressed by recombinant
pseudorabies virus as an in vivo marker for viral replication, J. Virol. Methods 66(2):283–292.
Kangrga, I. M., and Loewy, A. D., 1995, Whole-cell recordings from visualized C1 adrenergic bul-
bospinal neurons: ionic mechanisms underlying vasomotor tone, Brain Res. 670(2):215–
232.
Kaufman, G. D., Mustari, M. J., Miselis, R. R., and Perachio, A. A., 1996, Transneuronal pathways
to the vestibulocerebellum, J. Comp. Neurol. 370(4):501–523.
Kelly, R. M., and Strick, P. L., 2000, Rabies as a transneuronal tracer of circuits in the central
nervous system, J. Neurosci. Methods 103(1):63–71.
Kelly, R. M., and Strick, P. L., 2003, Cerebellar loops with motor cortex and prefrontal cortex
of a nonhuman primate, J. Neurosci. 23(23):8432–8444.
Kerman, I. A., Enquist, L. W., Watson, S. J., and Yates, B. J., 2003, Brainstem substrates of
sympatho-motor circuitry identified using trans-synaptic tracing with pseudorabies virus
recombinants, J. Neurosci. 23(11):4657–4666.
Kim, J. S., Enquist, L. W., and Card, J. P., 1999, Circuit-specific coinfection of neurons in the rat
central nervous system with two pseudorabies virus recombinants, J. Virol. 73(11):9521–
9531.
Klupp, B. G., Hengartner, C. J., Mettenleiter, T. C., and Enquist, L. W., 2004, Complete, anno-
tated sequence of the pseudorabies virus genome, J. Virol. 78(1):424–440.
Kristensson, K., Lycke, E., and Sjostrand, J., 1971, Spread of herpes simplex virus in peripheral
nerves, Acta Neuropathol. (Berl.) 17(1):44–53.
Kristensson, K., Nennesmo, L., Persson, L., and Lycke, E., 1982, Neuron to neuron transmission
of herpes simplex virus. Transport of virus from skin to brainstem nuclei, J. Neurol. Sci.
54(1):149–156.
Kritas, S. K., Pensaert, M. B., and Mettenleiter, T. C., 1994, Role of envelope glycoproteins gI,
gp63 and gIII in the invasion and spread of Aujeszky’s disease virus in the olfactory nervous
pathway of the pig, J. Gen. Virol. 75(Pt 9):2319–2327.
Krout, K. E., Mettenleiter, T. C., and Loewy, A. D., 2003, Single CNS neurons link both
central motor and cardiosympathetic systems: a double-virus tracing study, Neuroscience
118(3):853–866.
Kuypers, H. G., and Ugolini, G., 1990, Viruses as transneuronal tracers, Trends Neurosci.
13(2):71–75.
Larsen, P. J., Enquist, L. W., and Card, J. P., 1998, Characterization of the multisynaptic neuronal
control of the rat pineal gland using viral transneuronal tracing, Eur. J. Neurosci. 10(1):128–
145.
Levatte, M. A., Mabon, P. J., Weaver, L. C., and Dekaban, G. A., 1998, Simultaneous identifica-
tion of two populations of sympathetic preganglionic neurons using recombinant herpes
simplex virus type 1 expressing different reporter genes, Neuroscience 82(4):1253–1267.
Li, Y. W., Ding, Z. Q., Wesselingh, S. L., and Blessing, W. W., 1992a, Renal and adrenal sym-
pathetic preganglionic neurons in rabbit spinal cord: tracing with herpes simplex virus,
Brain Res. 573(1):147–152.
Li, Y. W., Ding, Z. Q., Wesselingh, S. L., and Blessing, W. W., 1993, Renal sympathetic pre-
ganglionic neurons demonstrated by herpes simplex virus transneuronal labelling in
the rabbit: close apposition of neuropeptide Y-immunoreactive terminals, Neuroscience
53(4):1143–1152.
Li, Y. W., Wesselingh, S. L., and Blessing, W. W., 1992b, Projections from rabbit caudal medulla to
C1 and A5 sympathetic premotor neurons, demonstrated with phaseolus leucoagglutinin
and herpes simplex virus, J. Comp. Neurol. 317(4):379–395.
Loewy, A. D., and Haxhiu, M. A., 1993, CNS cell groups projecting to pancreatic parasympa-
thetic preganglionic neurons, Brain Res. 620(2):323–330.
Lomniczi, B., Blankenship, M. L., and Ben-Porat, T., 1984, Deletions in the genomes of pseu-
dorabies virus vaccine strains and existence of four isomers of the genomes, J. Virol.
49(3):970–979.
Luppi, P. H., Aston-Jones, G., Akaoka, H., Chouvet, G., and Jouvet, M., 1995, Afferent projec-
tions to the rat locus coeruleus demonstrated by retrograde and anterograde tracing with
cholera-toxin B subunit and Phaseolus vulgaris leucoagglutinin, Neuroscience 65(1):119–
160.
Lynch, J. C., Hoover, J. E., and Strick, P. L., 1994, Input to the primate frontal eye field from the
substantia nigra, superior colliculus, and dentate nucleus demonstrated by transneuronal
transport, Exp. Brain Res. 100(1):181–186.
Manning, K. A., Erichsen, J. T., and Evinger, C., 1990, Retrograde transneuronal transport
properties of fragment C of tetanus toxin, Neuroscience 34(1):251–263.
Martin, X., and Dolivo, M., 1983, Neuronal and transneuronal tracing in the trigeminal system
of the rat using the herpes virus suis, Brain Res. 273(2):253–276.
Mettenleiter, T. C., 2003, Pathogenesis of neurotropic herpesviruses: role of viral glycoproteins
in neuroinvasion and transneuronal spread, Virus Res. 92(2):197–206.
Mettenleiter, T. C., Schreurs, C., Zuckermann, F., Ben-Porat, T., and Kaplan, A. S., 1988, Role
of glycoprotein gIII of pseudorabies virus in virulence, J. Virol. 62(8):2712–2717.
Mettenleiter, T. C., Zsak, L., Kaplan, A. S., Ben-Porat, T., and Lomniczi, B., 1987, Role of a
structural glycoprotein of pseudorabies in virus virulence, J. Virol. 61(12):4030–4032.
Middleton, F. A., and Strick, P. L., 1994, Anatomical evidence for cerebellar and basal ganglia
involvement in higher cognitive function, Science 266(5184):458–461.
Middleton, F. A., and Strick, P. L., 1996, The temporal lobe is a target of output from the basal
ganglia, Proc. Natl. Acad. Sci. U.S.A. 93(16):8683–8687.
Middleton, F. A., and Strick, P. L., 2001, Cerebellar projections to the prefrontal cortex of the
primate, J. Neurosci. 21(2):700–712.
Middleton, F. A., and Strick, P. L., 2002, Basal-ganglia ’projections’ to the prefrontal cortex of
the primate, Cereb. Cortex 12(9):926–935.
Moore, R. Y., Speh, J. C., and Card, J. P., 1995, The retinohypothalamic tract originates from a
distinct subset of retinal ganglion cells, J. Comp. Neurol. 352(3):351–366.
Moschovakis, A. K., Gregoriou, G. G., Ugolini, G., Doldan, M., Graf, W., Guldin, W.,
Hadjidimitrakis, K., and Savaki, H. E., 2004, Oculomotor areas of the primate frontal
lobes: a transneuronal transfer of rabies virus and [14C]-2-deoxyglucose functional imag-
ing study, J. Neurosci. 24(25):5726–5740.
Nichol, P. F., Chang, J. Y., Johnson, E. M., Jr., and Olivo, P. D., 1994, Infection of sympathetic
and sensory neurones with herpes simplex virus does not elicit a shut-off of cellular pro-
tein synthesis: implications for viral latency and herpes vectors, Neurobiol. Dis. 1(1–2):
83–94.
Norgren, R. B., Jr., and Lehman, M. N., 1998, Herpes simplex virus as a transneuronal tracer,
Neurosci. Biobehav. Rev. 22(6):695–708.
Norgren, R. B., Jr., McLean, J. H., Bubel, H. C., Wander, A., Bernstein, D. I., and Lehman, M.
N., 1992, Anterograde transport of HSV-1 and HSV-2 in the visual system, Brain Res. Bull.
28(3):393–399.
O’Donnell, P., Lavin, A., Enquist, L. W., Grace, A. A., and Card, J. P., 1997, Interconnected
parallel circuits between rat nucleus accumbens and thalamus revealed by retrograde
transsynaptic transport of pseudorabies virus, J. Neurosci. 17(6):2143–2167.
Oldfield, B. J., Giles, M. E., Watson, A., Anderson, C., Colvill, L. M., and McKinley, M. J., 2002,
The neurochemical characterisation of hypothalamic pathways projecting polysynaptically
to brown adipose tissue in the rat, Neuroscience 110(3):515–526.
Pickard, G. E., Smeraski, C. A., Tomlinson, C. C., Banfield, B. W., Kaufman, J., Wilcox,
C. L., Enquist, L. W., and Sollars, P. J., 2002, Intravitreal injection of the attenuated
pseudorabies virus PRV Bartha results in infection of the hamster suprachiasmatic nu-
cleus only by retrograde transsynaptic transport via autonomic circuits, J. Neurosci. 22(7):
2701–2710.
Ricardo, J. A., and Koh, E. T., 1978, Anatomical evidence of direct projections from the nucleus
of the solitary tract to the hypothalamus, amygdala, and other forebrain structures in the
rat, Brain Res. 153(1):1–26.
Rinaman, L., Card, J. P., and Enquist, L. W., 1993, Spatiotemporal responses of astrocytes, ram-
ified microglia, and brain macrophages to central neuronal infection with pseudorabies
virus, J. Neurosci. 13(2):685–702.
Rinaman, L., and Schwartz, G., 2004, Anterograde transneuronal viral tracing of central vis-
cerosensory pathways in rats, J. Neurosci. 24(11):2782–2786.
Rotto-Percelay, D. M., Wheeler, J. G., Osorio, F. A., Platt, K. B., and Loewy, A. D., 1992, Transneu-
ronal labeling of spinal interneurons and sympathetic preganglionic neurons after pseu-
dorabies virus injections in the rat medial gastrocnemius muscle, Brain Res. 574(1–2):291–
306.
Rouiller, E. M., Capt, M., Dolivo, M., and De Ribaupierre, F., 1986, Tensor tympani reflex
pathways studied with retrograde horseradish peroxidase and transneuronal viral tracing
techniques, Neurosci. Lett. 72(3):247–252.
Rouiller, E. M., Capt, M., Dolivo, M., and De Ribaupierre, F., 1989, Neuronal organization of
the stapedius reflex pathways in the rat: a retrograde HRP and viral transneuronal tracing
study, Brain Res. 476(1):21–28.
Sabin, A. B., 1938, Progression of different nasally instilled viruses along different nervous
pathways in the same host, Proc. Soc. Exp. Biol. 38:270–275.
Sams, J. M., Jansen, A. S., Mettenleiter, T. C., and Loewy, A. D., 1995, Pseudorabies virus mutants
as transneuronal markers, Brain Res. 687(1–2):182–190.
Smeraski, C. A., Sollars, P. J., Ogilvie, M. D., Enquist, L. W., and Pickard, G. E., 2004, Suprachias-
matic nucleus input to autonomic circuits identified by retrograde transsynaptic transport
of pseudorabies virus from the eye, J. Comp. Neurol. 471(3):298–313.
Smiley, J. R., 2004, Herpes simplex virus virion host shutoff protein: immune evasion mediated
by a viral Rnase? J. Virol. 78(3):1063–1068.
Smith, B. N., Banfield, B. W., Smeraski, C. A., Wilcox, C. L., Dudek, F. E., Enquist, L. W., and
Pickard, G. E., 2000, Pseudorabies virus expressing enhanced green fluorescent protein:
a tool for in vitro electrophysiological analysis of transsynaptically labeled neurons in
identified central nervous system circuits, Proc. Natl. Acad. Sci. U. S. A. 97(16):9264–9269.
Song, C. K., and Bartness, T. J., 2001, CNS sympathetic outflow neurons to white fat that express
MEL receptors may mediate seasonal adiposity, Am. J. Physiol. Regul. Integr. Comp. Physiol.
281(2):R666–R672.
Stornetta, R. L., McQuiston, T. J., and Guyenet, P. G., 2004, GABAergic and glycinergic presym-
pathetic neurons of rat medulla oblongata identified by retrograde transport of pseudora-
bies virus and in situ hybridization, J. Comp. Neurol. 479(3):257–270.
Strack, A. M., and Loewy, A. D., 1990, Pseudorabies virus: a highly specific transneuronal cell
body marker in the sympathetic nervous system, J. Neurosci. 10(7):2139–2147.
Strack, A. M., Sawyer, W. B., Hughes, J. H., Platt, K. B., and Loewy, A. D., 1989a, A general
pattern of CNS innervation of the sympathetic outflow demonstrated by transneuronal
pseudorabies viral infections, Brain Res. 491(1):156–162.
Strack, A. M., Sawyer, W. B., Platt, K. B., and Loewy, A. D., 1989b, CNS cell groups regulating
the sympathetic outflow to adrenal gland as revealed by transneuronal cell body labeling
with pseudorabies virus, Brain Res. 491(2):274–296.
Sylvester, C. M., Krout, K. E., and Loewy, A. D., 2002, Suprachiasmatic nucleus projection to
the medial prefrontal cortex: a viral transneuronal tracing study, Neuroscience 114(4):1071–
1080.
Tang, Y., Rampin, O., Giuliano, F., and Ugolini, G., 1999, Spinal and brain circuits to motoneu-
rons of the bulbospongiosus muscle: retrograde transneuronal tracing with rabies virus,
J. Comp. Neurol. 414(2):167–192.
Ter Horst, G. J., 2000, Transneuronal retrograde dual viral labelling of central autonomic
circuitry: possibilities and pitfalls, Auton. Neurosci. 83(3):134–139.
Ter Horst, G. J., Hautvast, R. W., De Jongste, M. J., and Korf, J., 1996, Neuroanatomy of cardiac
activity-regulating circuitry: a transneuronal retrograde viral labelling study in the rat, Eur.
J. Neurosci. 8(10):2029–2041.
Tomishima, M. J., and Enquist, L. W., 2001, A conserved alpha-herpesvirus protein necessary
for axonal localization of viral membrane proteins, J. Cell Biol. 154(4):741–752.
Tomishima, M. J., Smith, G. A., and Enquist, L. W., 2001, Sorting and transport of alpha
herpesviruses in axons, Traffic 2(7):429–436.
Ueyama, T., Krout, K. E., Nguyen, X. V., Karpitskiy, V., Kollert, A., Mettenleiter, T. C., and
Loewy, A. D., 1999, Suprachiasmatic nucleus: a central autonomic clock, Nat. Neurosci.
2(12):1051–1053.
Ugolini, G., 1995, Specificity of rabies virus as a transneuronal tracer of motor networks: trans-
fer from hypoglossal motoneurons to connected second-order and higher order central
nervous system cell groups, J. Comp. Neurol. 356(3):457–480.
Ugolini, G., Kuypers, H. G., and Simmons, A., 1987, Retrograde transneuronal transfer of
herpes simplex virus type 1 (HSV 1) from motoneurones, Brain Res. 422(2):242–256.
Ugolini, G., Kuypers, H. G., and Strick, P. L., 1989, Transneuronal transfer of herpes virus from
peripheral nerves to cortex and brainstem, Science 243(4887):89–91.
Vahlne, A., Nystrom, B., Sandberg, M., Hamberger, A., and Lycke, E., 1978, Attachment of
herpes simplex virus to neurons and glial cells, J. Gen. Virol. 40(2):359–371.
Wesselingh, S. L., Li, Y. W., and Blessing, W. W., 1989, PNMT-containing neurons in the rostral
medulla oblongata (C1, C3 groups) are transneuronally labelled after injection of herpes
simplex virus type 1 into the adrenal gland, Neurosci. Lett. 106(1–2):99–104.
Westerhaus, M. J., and Loewy, A. D., 1999, Sympathetic-related neurons in the preoptic region
of the rat identified by viral transneuronal labeling, J. Comp. Neurol. 414(3):361–378.
Westerhaus, M. J., and Loewy, A. D., 2001, Central representation of the sympathetic nervous
system in the cerebral cortex, Brain Res. 903(1–2):117–127.
Whealy, M. E., Card, J. P., Robbins, A. K., Dubin, J. R., Rziha, H. J., and Enquist, L. W., 1993,
Specific pseudorabies virus infection of the rat visual system requires both gI and gp63
glycoproteins, J. Virol. 67(7):3786–3797.
Wiesel, T. N., Hubel, D. H., and Lam, D. M., 1974, Autoradiographic demonstration of ocular-
dominance columns in the monkey striate cortex by means of transneuronal transport,
Brain Res. 79(2):273–279.
Zemanick, M. C., Strick, P. L., and Dix, R. D., 1991, Direction of transneuronal transport of
herpes simplex virus 1 in the primate motor system is strain-dependent, Proc. Natl. Acad.
Sci. U.S.A. 88(18):8048–8051.
Zou, Z., Horowitz, L. F., Montmayeur, J. P., Snapper, S., and Buck, L. B., 2001, Genetic tracing
reveals a stereotyped sensory map in the olfactory cortex, Nature 414(6860):173–179.
10
Dextran Amines: Versatile
Tools for Anterograde and
Retrograde Studies of
Nervous System Connectivity
ANTON REINER and MARCIA G. HONIG
BACKGROUND
METHODS USING DEXTRAN AMINES
What Are the Dextran Amines?
Anterograde Labeling with Dextran Amines
Retrograde Labeling with Dextran Amines
Retrograde Collateral Labeling with Dextran Amines
Combining BDA10kDa Anterograde Labeling with Other
Neuroanatomical Tracers
Combining BDA3kDa Retrograde Labeling with Other
Combining Dextran Amine Labeling with Immunohistochemistry
Merits of Dextran Amines Relative to Other Tracers
Pitfalls and Solutions
APPENDIX
Step-by-Step BDA Protocol for LM Single-Labeling Studies
Step-by-Step BDA Protocol for LM Double-Labeling Studies
Step-by-Step BDA Protocol for EM Studies
Step-by-Step BDA Protocol for EM Double-Labeling Studies
Expected Results
REFERENCES
Abstract: Dextran amines are versatile and sensitive tools for anterograde and ret-
rograde investigation of neural connectivity. Because of their tolerance of diverse
ANTON REINER AND MARCIA G. HONIG • Department of Anatomy and Neuro-

biology, University of Tennessee Health Science Center, 855 Monroe Avenue, Memphis, TN
38163
304
DEXTRAN AMINES 305
fixatives, they are ideal for various light and electron microscopic studies. They can
be iontophoretically or pressure-injected, and then depending on the type of dextran
amine used and the type of detection method, visualized by transmitted light mi-
croscopy, fluorescence microscopy, or electron microscopy. High-molecular-weight
biotinylated dextran amines (BDAs; 10 kDa) yield sensitive and exquisitely detailed
labeling of axons and terminals, while low-molecular-weight BDAs (3 kDa) yield
sensitive and detailed Golgi-like retrograde labeling of neurons. Labeling with the
BDAs can be visualized with an avidin-biotinylated horseradish peroxidase (ABC)
procedure followed by a standard or a metal-enhanced diaminobenzidine (DAB)
reaction, or with any of several fluorescent probes that bind to biotin. Fluorescent
dextran amines can be directly visualized by fluorescence microscopy or rendered
suitable for transmitted light or electron microscopic viewing by immunohistochem-
ical detection of the given fluorophore. The variety of dextran amines and the
methods for their visualization make them well-suited for multiple-label studies.
The dextran amines can also be combined with other anterograde or retrograde
tracers, or intracellular labeling, and the disparate markers separately visualized
either by multicolor DAB or by DAB–VIP r labeling, or by multiple fluorescence
viewing. In the same manner, pathway tracing with dextran amines and immuno-
histochemical labeling can be combined. The dextran amines are thus flexible and
valuable pathway-tracing tools that have gained widespread popularity since being
introduced.
Keywords: anterograde tracer, connectivity, dextran amines, double-label, neu-
roanatomical mapping, retrograde tracer
I. BACKGROUND
The selective silver impregnation of degenerating axons as a tool for

anterograde tracing of neural pathways revolutionized neuroanatomy by
making it possible to accurately and objectively delineate neural circuitry
(Fink and Heimer, 1967; Nauta and Gygax, 1954). The anterograde trac-
ing of neural pathways was further accelerated with the development of
the autoradiographic detection of anterogradely transported radioactive
amino acids (autoradiography, ARG) (Cowan et al., 1972; Edwards, 1972;
Hendrickson and Edwards, 1978), which proved to be a more reliable
method than had been the silver stains of degenerating fibers. The lim-
ited anatomical resolution of axons and terminals provided by ARG-tracing
methods was, however, a serious disadvantage. This limitation was over-
come in the Phaseolus vulgaris leucoagglutinin (PHA-L) method, which pro-
vided superbly sensitive and selective anterograde labeling of axons and
their terminals with limited uptake by fibers of passage at the injection
site, with the added advantage that PHA-L–labeled axons and terminals
could be visualized at both LM and EM levels (Gerfen et al., 1989; Ger-
fen and Sawchenko, 1984; Sesack et al., 2006, this volume; Zaborszky and
Heimer, 1989). Two drawbacks of the PHA-L method, namely that ion-
tophoretic injection methods are required and that PHA-L proved unre-
liable in the hands of many investigators, led to the quest for yet better
306 ANTON REINER and MARCIA G. HONIG
pathway-tracing agents (Groenewegen and Wouterlood, 1990; Schmued

et al., 1990; Veenman et al., 1992, 1995).
Shortly after the introduction of PHA-L, a number of investigators demon-
strated the utility of fluorophore-bound dextran–lysine conjugates (called
dextran amines) as pathway-tracing agents (Fritzsch and Wilm, 1990; Glover
et al., 1986; Nance and Burns, 1990; Schmued et al., 1990). Fluorescent dex-
tran amines rely on the ease with which dextran is transported by axons,
and the fact that the lysine moiety makes it possible to fix the transported
fluorophore-conjugated dextran in situ with standard aldehyde fixatives,
to allow visualization of the cellular localization of the transported dex-
tran amine by fluorescence microscopy. These tracers were found to com-
bine high sensitivity and detailed resolution with simplicity and reliability
in their delivery and uptake. While the fluorescent dextran amines were
revealed to be in several respects superior to ARG and PHA-L, they were
not well suited for studies involving detailed mapping of neuronal projec-
tion systems, primarily because fluorophores fade with viewing. This limita-
tion was overcome by the development of the biotinylated dextran amines
(BDAs), which possess the favorable transport properties of fluorescent dex-
tran amines and can be visualized at the transmitted light level with a stan-
dard avidin-biotinylated horseradish peroxidase (HRP) (ABC) procedure
(Brandt and Apkarian, 1992; Rajakumar et al., 1993; Reiner et al., 2000;
Veenman et al., 1992, 1995). Moreover, with the availability of antibodies
against fluorophores such as rhodamine and fluorescein, it became possi-
ble to use peroxidase immunolabeling to also detect fluorophore-bound
dextran amines (Reiner et al., 2000, 2003). Further studies have revealed
that depending on its molecular weight and the pH of the delivery vehi-
cle, dextran amines could be used either as anterograde or as retrograde
pathway-tracing agents (Fritzsch, 1993; Kaneko et al., 1996; Medina et al.,
1997) in a wide variety of species (Albert et al., 1999; Davila et al., 2002;
Fritzsch, 1993; Guirado et al., 1999; Kenigfest et al., 2000; Lanuza et al., 1998;
Lopes-Correa et al., 1998; Scalia et al., 1997; Striedter, 1994; Veenman et al.,
1992; Wang et al., 2004). In addition, owing to its tolerance of diverse fixa-
tives, BDA can be visualized at the LM or EM level, and it can be combined
with various other pathway-tracing or immunohistochemical methods. The
purpose of this chapter is to provide an overview of the uses of the dextran
amines, and to provide simple protocols for their use.
II. METHODS USING DEXTRAN AMINES
A. What Are the Dextran Amines?
Dextrans are biologically inert hydrophilic polysaccharides that possess

high water solubility, low toxicity, low immunogenicity, and resistance to
cleavage by most endogenous glycosidases (Haugland, 1996). Dextrans of
molecular weights ranging from 3 to 2000 kDa, conjugated to biotin or a wide
DEXTRAN AMINES 307
variety of fluorophores, are commercially available from Molecular Probes
(Eugene, OR), Sigma Chemical Company (St. Louis, MO), or Vector Labo-
ratories (Burlingame, CA). For studies of nervous system connectivity, dex-
trans with covalently bound lysine residues are used, since the lysines allow
the dextran to be conjugated to surrounding biomolecules by paraformalde-
hyde or glutaraldehyde fixation (Reiner et al., 2000). Dextrans transport an-
terogradely and retrogradely in neurons and their axons, but unlike DiI and
DiO are not transported in fixed nervous tissue (Fritzsch and Wilm, 1990).
Dextrans of up to 70 kDa have been used in studies of nervous system con-
nectivity, with smaller molecular weight dextrans transporting more quickly
than larger. Moreover, 3-kDa dextran seems to transport more effectively in
a retrograde direction than does 10-kDa dextran. The mechanism of BDA
uptake appears to be pinocytotic for intact neurons ( Jiang et al., 1993). The
dextran amines are, however, also taken up by damaged neurons and ax-
ons (Glover et al., 1986; Lei et al., 2004; Todorova and Rodziewicz, 1995;
Veenman et al., 1992). Dextran–lysine (called dextran amine) conjugated
to biotin is referred to as BDA, while the fluorescent conjugates of dextran
amines have been identified by either an abbreviated form of their name or
a descriptive name. For example, 10-kDa tetramethylrhodamine-conjugated
dextran amine, which we will, for clarity and consistency, here abbreviate as
RDA10kDa, has also been called fluororuby (Schmued et al., 1990). Similarly,
10-kDa fluorescein dextran amine (FDA10kDa) has been called fluoroemer-
ald (Novikova et al., 1997). Mixes of BDA10kDa and RDA10kDa, sometimes
called miniruby (Liu et al., 1993; Novikova et al., 1997), or BDA3kDa and
RDA3kDa, sometimes called microruby (Liu et al., 1993; Novikova et al.,
1997), are also commercially available as are BDA10kDa–FDA10kDa mixes
and BDA3kDa–FDA3kDa mixes (sometimes called mini- and microemer-
ald) (Novikova et al., 1997). Such mixtures of biotinylated and fluorophore-
conjugated dextran amines allow fluorescence viewing to rapidly screen
injection site and transport efficacy before proceeding to the steps needed
for producing a permanent label suitable for transmitted light microscopic
viewing of the BDA localization. Finally, in addition to different molecular
weight FDA and RDA, Molecular Probes also offers dextran amines conju-
gated to a wide array of additional fluorophores, including the intensely
fluorescent, fade-resistant Alexa fluorophores, which lend themselves to
sensitive detection and differentiation of labeled structures, especially us-
ing confocal laser scanning microscopy (CLSM).
B. Anterograde Labeling with Dextran Amines
The 10,000 Da form of dextran amine is preferentially transported an-

terogradely, with yet higher weight forms transporting less effectively than
10 kDa (Fritzsch, 1993; Kaneko et al., 1996; Medina et al., 1997; Schmued
et al., 1990; Veenman et al., 1992). A dextran amine of 10 kDa can be used
in the same type of LM single-label anterograde pathway-tracing studies as
A B
SC
C D
SC
Figure 10.1. Iontophoretic BDA10kDa injection into primary visual cortex (area 17)
of rat and resulting labeling in a series of transverse sections. Image A shows a
low-power view of the BDA10kDa injection site in area 17 (large arrow), showing
an anterograde projection to area 18 (small arrow). The superior colliculus (SC)
is located in the lower left of the image. Image B shows a high-power view of the
labeling in area 18 shown by the arrow in A. Note that labeled fibers in area 18
are located predominantly in layers II/III and V, as would be expected on the basis
of the known projection of area 17 to 18 in rats (Peters, 1985). Some retrogradely
labeled neurons are also evident in area 18, mainly in layers II/III and V, as would
be expected on the basis of the known projection of area 18 to 17 (Peters, 1985).
Although it is possible that the retrogradely labeled neurons contribute BDA10kDa-
labeled collaterals to the fiber labeling in area 18, such labeled collaterals must be far
fewer than the anterogradely labeled fibers since the distribution of labeled fibers
and terminals in 18 conforms to that expected for anterograde labeling from 17 and
not to that expected for labeled collaterals of layers II/III and V neurons (Peters,
1985). (C) Golgi-like retrograde labeling of a neuron in layer III of area 18 following
the injection shown in image A. Image D shows labeled fibers and terminals in the
superficial layers of the ipsilateral SC following the injection shown in image A. The
magnification in D is the same as in C. (From Veenman et al., 1992.)
PHA-L, biocytin, neurobiotin, CTb (cholera toxin B subunit), and WGA-

HRP (HRP-conjugated wheat germ agglutinin), and it possesses several fea-
tures that make it advantageous. First, a 10-kDa dextran amine reliably yields
anterograde labeling after either iontophoretic or pressure injection into
the nervous system (Figs. 10.1A,B,D and 10.2B). Second, it is easy to make
small and well-defined injections of the dextran amine, thereby confining
the injection to the region whose projections are being investigated and/or
study the topographic order to the projection from the given region (Figs.
10.1A and 10.2A). Third, and as noted previously, the morphological detail
DEXTRAN AMINES 309
A B
Hp
CC
2mm 100 m
C D
S
S
0.5 m
Figure 10.2. LM images of coronal sections showing iontophoretic BDA10kDa in-

jection site in rat somatosensory cortex (A) and resulting anterograde labeling in
ipsilateral striatum (B), and EM images of anterogradely labeled terminals in con-
tralateral striatum after BDA10kDa injection into rat motor cortex (C, D). Image A
shows a low-power view of an iontophoretic injection of BDA into rat somatosensory
cortex (large arrow), showing anterogradely labeled fibers that cross in the corpus
callosum (CC) above the hippocampus (Hp) and project to homotypic contralat-
eral somatosensory cortex (small arrow). A blood vessel is present in the middle of
the terminal field in the contralateral cortex. Image B shows a high-power view of
labeled axons with varicosities in ipsilateral rat striatum following the iontophoretic
injection of BDA into somatosensory cortex shown in A. Images C and D show EM
views of anterogradely labeled terminals in striatum after injection of BDA10kDa into
the contralateral motor cortex. Both terminals make asymmetric axospinous contact
with the spines (S) of presumptive striatal projection neurons. The BDA+ terminals
in C and D are densely labeled but clearly contain numerous unlabeled small clear
vesicles. The tissue was prepared using freshly made 3.5% paraformaldehyde–0.6%
glutaraldehyde–15% saturated picric acid as fixative. Both C and D are at the same
magnification. F [ rom Veenman et al., 1992 (A, B) and Reiner et al., 2003 (C, D).]
of the labeling is often exquisite (Figs. 10.1B,D and 10.2B). Fourth, the de-
tailed labeling of terminals provided by dextran amine of 10 kDa, and the
fact that BDA readily tolerates glutaraldehyde fixation with no significant
attenuation of labeling makes BDA10kDa well-suited especially to character-
izing the morphology and cellular targets of given brain regions at the LM
and EM levels (Fig. 10.2C,D; Guirado et al., 1999; Reiner et al., 2003, 2004;
Van Haeften and Wouterlood, 2000; Veenman and Reiner, 1996; Wouterlood
and Jorritsma-Byham, 1993; Wright et al., 1999, 2001). With diaminobenzi-
dine (DAB) visualization, terminals anterogradely labeled with BDA10kDa
tend to have dense labeling of the axoplasm on the surface of vesicular
membranes, so that synaptic vesicles are clearly recognizable. Moreover,
BDA10kDa seems to have advantages over such other sensitive anterograde
tracers as PHA-L and CTB in the simplicity of the visualization procedure,
since both PHA-L and CTB require a multistep immunohistochemical pro-
cedure, while BDA requires only a one-step ABC procedure. The simpler
and briefer visualization procedures for BDA are also beneficial for the
preservation of ultrastructural detail (Van Haeften and Wouterlood, 2000;
Wouterlood and Jorritsma-Byham, 1993). Note that BDA10kDa, however, is
not exclusively an anterograde tracer and yields some retrograde labeling
(Fig. 10.1B,C; Reiner et al., 2000; Veenman et al., 1992; Vercelli et al., 2000).
C. Retrograde Labeling with Dextran Amines
Low-molecular-weight dextran amine (i.e., 3000 MW), when injected

within an acidic vehicle (e.g., 10% BDA in 0.1 M sodium citrate–HCl,
pH 3.0), is mainly transported retrogradely (Fritzsch, 1993; Kaneko et al.,
1996; Medina et al., 1997; Veenman et al., 1992). There is evidence that neu-
ronal activity in terminals enhances their pinocytotic uptake of BDA3kDa,
yielding more intense retrograde labeling (Jiang et al., 1993). Since 3-kDa
dextran amines are transported twice as rapidly as 10-kDa dextran amines
(Fritzsch, 1993), shorter survival times are needed for BDA3kDa than that
for BDA10kDa. Dextran amine of 3 kDa can be used in the same types of LM
single-label retrograde pathway-tracing studies as HRP, Fluorogold, biocytin,
neurobiotin, CTB, and WGA-HRP for routine mapping of the distribution
of the neurons projecting to a given neural region and their topographic
order. The labeling intensity and fixation tolerance that make BDA10kDa ad-
vantageous for anterograde labeling also make BDA3kDa advantageous for
both LM and EM retrograde labeling studies (Figs. 10.2A–C and 10.3A,B).
Moreover, because of the Golgi-like retrograde labeling it yields and its glu-
taraldehyde tolerance, BDA3kDa is especially useful for studies of the LM
architecture of labeled neurons and for EM studies of their axodendritic
and axospinous inputs (Fig. 10.4; Lei et al., 2004). Note that 3-kDa dex-
tran amine is, however, not exclusively a retrograde tracer and yields some
anterograde labeling even when injected within an acidic vehicle ( Jiao
et al., 2000; Medina and Reiner, 1997).
D. Retrograde Collateral Labeling with Dextran Amines
The sensitivity of BDA10kDa creates one of its major disadvantages as

an anterograde tracer—the retrograde labeling of neurons with BDA10kDa
DEXTRAN AMINES 311
Figure 10.3. The schematic shown in A illustrates the circuit devoted to vocal learning
in songbirds, as viewed in the sagittal plane. The serially connected components of
this circuit are (1) the “cortical” area called HVC, which receives direct input from
avian auditory cortex; (2) area X of the medial striatum (X); (3) the dorsolateral
medial nucleus of the thalamus (DLM); (4) a “cortical” region termed the lateral
magnocellular anterior nidopallium (LMAN); and (5) the “cortical” area called the
robust nucleus of the arcopallium (RA), which projects to the vocal motoneurons of
the brainstem. The images in B–D show fields of view within area X containing neu-
rons retrogradely labeled from DLM with BDA3kDa. The image in B was captured
using differential interference contrast microscopy and shows two such neurons in
area X of male zebra finch visualized with an ABC/DAB procedure. These neurons
have the morphological characteristics of pallidal neurons, which include being rel-
atively large (12–14 µm) and possessing aspiny dendrites. The images shown in C
and D are from the same individual field within area X, captured using CLSM, from
a zebra finch that received a BDA3kDa injection into DLM. Image C shows DTAF-
labeled BDA3kDa+ neurons, while D shows the TRITC-labeled LANT6+ neurons.
All neurons that were labeled for BDA3kDa also contained LANT6, as indicated by
the arrows. The results in C and D show that area X neurons that project to DLM also
possess the pallidal trait of containing the neurotensin-related hexapeptide LANT6
(Lys8 -Asp9 -neurotensin8−13 ). (From Reiner et al., 2004.)
can, in some cases, be so complete as to label the axon collaterals of neurons

retrogradely labeled by it (Chen and Aston-Jones, 1998). If those neurons
have axon collaterals in the site to which the region injected with BDA10kDa
projects (e.g., if the BDA-labeled neurons reside within the field of antero-
gradely labeled BDA+ axons and terminals; Fig. 10.1A,B), then there may
be no simple way to reliably distinguish between anterograde labeling and
labeling of the collaterals of retrogradely labeled neurons. It would thus be
uncertain whether a given labeled axon in the terminal field arises from
the BDA injection site or is a collateral of a retrogradely labeled neuron.
1mm
20 m
+S
--t
0.5 m
Figure 10.4. Example of a BDA3kDa injection into substantia nigra (A), and the LM
level (B) and EM level (C) retrograde labeling of striatal neurons obtained. Image B
shows perikarya and dendrites of striatonigral neurons labeled with BDA3kDa, while
image C shows that this labeling included that of spines. A BDA3kDa-labeled spine
(+s) is evident in C, and the labeled spine receives an asymmetric synaptic contact
from an unlabeled terminal (−t) with the round vesicles characteristic of excitatory
input from cortex. (From Lei et al., 2004.)
One way to deal with this problem is to ascertain the projection targets of
any population of neurons retrogradely labeled with BDA10kDa. In some
cases, this information may be available from the literature, while in others
it may be necessary to obtain it by anterograde labeling from the neuronal
DEXTRAN AMINES 313
population in question. It may be the case that the retrogradely labeled neu-
rons are known not to have collaterals in the region of the BDA anterograde
labeling, in which case the confound potentially created by collateral label-
ing is obviated. In addition, the fact that BDA10kDa does not always yield
extensive retrograde labeling also somewhat mitigates this problem (Brandt
and Apkarian, 1992; Reiner et al., 1993, 2000; Veenman et al., 1992).
It is possible, however, to turn this disadvantage of BDA into a useful tool
if one is interested in the collateral projections of the retrogradely labeled
neurons (Chen and Aston-Jones, 1998). For example, we have shown that
BDA3kDa injected into the pyramidal tract (PT) at pontine levels yields
extensive Golgi-like labeling of pyramidal neurons in layer V of sensory
and motor cortex in rats (Fig. 10.5A,B; Lei et al., 2004; Reiner et al., 2003).
Accompanying this retrograde labeling is labeling of the collateral projec-
tions of these neurons to the striatum (Fig. 10.5C,D). Thus, this approach
can be used to selectively label the corticostriatal projection of the PT-type
cortical neurons. There are, in fact, at least two types of corticostriatal pro-
jection neurons in rats (Cowan and Wilson, 1994; Kincaid and Wilson, 1996;
Wilson, 1987). One of these two types is the PT type, while the other type is
termed the intratelencephalically projecting type (IT type), since it projects
only within telencephalon. Because the laminar and regional distributions
of the PT- and IT-type neurons in cortex are overlapping, it would be difficult
to selectively label the input to ipsilateral striatum from only one of them
by means of an injection of BDA10kDa directly into the cortex. Injection
of BDA3kDa into the PT, however, provides a means for selectively label-
ing the PT-type neuron terminals in ipsilateral striatum and ascertaining
their morphology and synaptic targets at the LM and EM levels (Lei et al.,
2004; Reiner et al., 2003). It should be possible to use this same collateral
labeling approach to selectively study the output of a particular neuronal
population to one or more specific targets in circumstances in which that
neuronal population (1) is intermingled with other neuronal populations
within a field and whose projection is thus difficult to selectively label by an-
terograde labeling from the cell bodies of origin by direct injection of the
region containing the cell bodies and (2) has a projection to at least one re-
gion that none of the surrounding other neuronal population in the region
does. This unique projection then can be the target of a BDA3kDa injection
to yield selective retrograde collateral labeling of the projection of that neu-
ronal population. The approach for this collateral labeling with BDA3kDa
is otherwise not different from that for conventional BDA anterograde or
retrograde labeling.
E. Combining BDA10kDa Anterograde Labeling with Other

BDA10kDa can also be used in LM double-label studies in which one

population of axons and terminals is labeled with BDA10kDa and (1) a
A B
50 m
C D
25 m 0.5 m
Figure 10.5. BDA3kDa injected into the axons of the pyramidal tract of rat at pontine
levels, as shown schematically in A, yields extensive retrograde labeling of layer V
cortical pyramidal neurons in ipsilateral sensory–motor cortex (B). The pyramidal
neurons are labeled so thoroughly that their intrastriatal collaterals (small arrows
in C), which arise from the pyramidal neuron axons as they pass through the stria-
tum (large arrow in C), are extensively labeled. This allows selective visualization
of the corticostriatal terminals of the pyramidal tract-type cortical neurons at the
LM (C) and EM levels (D). The pyramidal tract neurons give rise to large terminals
that make asymmetric synaptic contacts with the spines (S) of striatal neurons (D),
which are identifiable by their size and the presence of spine apparatus (asterisk).
The terminal shown is relatively large and associated with a perforated postsynaptic
density (arrow). (From Reiner et al., 2003.)
second population of axons is labeled with PHA-L, CTB, RDA10kDa, or

FDA10kDa or (2) a population of neurons is retrogradely labeled with HRP,
FluoroGold, CTB, RDA3kDa, FDA3kDa, or a viral tracer, such as pseudora-
bies, or labeled by intracellular filling (Alisky and Tolbert, 1994; Aston-Jones
and Card, 2000; Dolleman-van der Weel et al., 1994; Lanciego and Wouter-
lood, 1994; Luo et al., 2001). The former type of study is useful, for ex-
ample, to determine the topographic order in a projection system, with
the two anterograde tracers injected side by side in the target of interest
(Ojima and Takayanagi, 2001), while the latter studies serve to determine if
DEXTRAN AMINES 315
input from a given source ends on neurons projecting to a particular target
structure.
Such LM double-label studies require using two distinct methods for vi-
sualizing the two different populations of labeled structures. Compatible
distinct pairs of labeling methods commonly used at the transmitted light
level include two-color DAB (Antal et al., 1990; Hancock, 1986; Hsu and
Soban, 1982; Medina et al., 1997; Veenman et al., 1992), and DAB in com-
bination with Vector Laboratories very intense purple (VIP r
) (Gonzalo
et al., 2001; Lanciego et al., 1997; Lanciego and Gimenez-Amaya, 1999).
Since BDA can be detected using a fluorophore-conjugated antibiotin or
a fluorophore-conjugated avidin, BDA10kDa and a second tracer can also
be detected with pairs of distinct fluorophores (e.g., rhodamine and fluo-
rescein), and viewed using conventional epi-illumination fluorescence mi-
croscopy or CLSM ( Jiao et al., 2000).
Alternatively, two fluorescent dextran amines such as RDA10kDa and
FDA10kDa can be used for double anterograde labeling, or anterograde
labeling with RDA10kDa or FDA10kDa can be combined with retrograde
labeling with FDA3kDa or RDA3kDa, respectively. Such material can be di-
rectly viewed by fluorescence microscopy or CLSM, by separate permanent
labels suitable for transmitted light viewing produced using antisera directed
against rhodamine and fluorescein and two-color DAB, or by DAB and VIP
(Kaneko et al., 1996). A variant approach of injecting two tracers in one
animal involves coinjecting an anterograde tracer such as BDA10kDa with
a retrograde tracer such as CTB at the same site (Coolen et al., 1999). This
makes it possible to ascertain the inputs to (by retrograde CTB labeling)
and the outputs of (by anterograde BDA10kDa labeling) a single site, with
alternating sections labeled for each marker or different markers used to
visualize the labeling in the same sections.
BDA10kDa, RDA10kDa, or FDA10kDa can also be used in EM double-
label studies to determine the morphology of two defined types of axons
with respect to each other or their targets (in combination with each other
or PHA-L), or to determine the identity of the target structure of a defined
type of axon (using a retrograde tracer to label the target structure and
dextran amine to label the axons and terminals) (Alisky and Tolbert, 1994;
Lanciego et al., 1998a,b; Luo et al., 2001; Van Haeften and Wouterlood,
2000). Such EM double-label studies usually (but not always) require us-
ing pairs of markers that are distinct at the EM level, for example, DAB
and silver-intensified immunogold, DAB and benzidine dihydrochloride
(BDHC), DAB and silver-intensified DAB, or DAB and VIP r
(Anderson
et al., 1991, 1994; Groenewegen and Wouterlood, 1990; Van Haeften and
Wouterlood, 2000; Wouterlood et al., 1993; Zhou and Grofova, 1995). Note
that the same EM marker (e.g., DAB) can be used for both structures to
be visualized, if BDA-labeled input to a population of neurons retrogradely
labeled with a second tracer is under study, and the retrogradely labeled
neurons do not give rise to local collaterals with the same morphology as
the BDA-labeled terminals.
F. Combining BDA3kDa Retrograde Labeling with Other

BDA3kDa can be used in LM double-label studies in which a set of

perikarya is labeled with BDA3kDa and (1) a population of axons is labeled
using PHA-L, CTB, RDA10kDa, or FDA10kDa or (2) a second population
of neurons is retrogradely labeled, in this case with HRP, Fluorogold, CTB,
RDA3kDa, FDA3kDa, or a viral tracer such as pseudorabies or by intracel-
lular filling. Such LM double-label studies again require using two distinct
methods for visualizing the two different populations of labeled structures,
such as two-color DAB (Antal et al., 1990; Hancock, 1986; Hsu and Soban,
1982; Medina et al., 1997; Veenman et al., 1992), DAB in combination with
VIP (Gonzalo et al., 2001), or two fluorophores ( Jiao et al., 2000). As true for
fluorophore-conjugated 10-kDa dextran amines, fluorophore-conjugated 3-
kDa dextran amines can be detected using antisera directed against the
fluorophores (typically rhodamine or fluorescein), and BDA3kDa itself
can be detected with an avidin-conjugated fluorophore or a fluorophore-
conjugated antibiotin ( Jiao et al., 2000; Kaneko et al., 1996). Again, a variant
approach of injecting two tracers in one animal involves coinjecting a retro-
grade tracer such as BDA3kDa and an anterograde tracer such as PHA-L at
the same site at the same time (Coolen et al., 1999). This makes it possible to
ascertain the inputs to (by retrograde BDA3kDa labeling) and the outputs
of (by anterograde PHA-L labeling) a single site, with alternating sections
labeled for each marker or different markers used to visualize the labeling
in the same sections.
BDA3kDa can also be used in EM double-label studies of inputs to
BDA3kDa-labeled neurons, and markers that are distinct at the EM level are
then typically needed to distinguish the two classes of tracer-labeled struc-
tures (Alisky and Tolbert, 1994; Anderson et al., 1991, 1994; Groenewegen
and Wouterlood 1990; Lanciego et al., 1998a,b; Van Den Pol and Decavel,
1990; Van Haeften and Wouterlood, 2000; Wouterlood et al., 1993; Zaborszky
and Heimer, 1989). Here too, using only one EM marker (e.g., DAB) for
both categories of labeled structure is possible if the BDA3kDa retrogradely
labeled neurons do not give rise to local collaterals with the same morphol-
ogy as the anterogradely labeled input under study.
G. Combining Dextran Amine Labeling with Immunohistochemistry
BDA can also be used in LM double-label studies in which one popula-

tion of axons and terminals is labeled with BDA10kDa or one population of
neuronal perikarya is labeled with BDA3kDa, and a second population of
axons/terminals, dendrites, and/or perikarya is labeled immunohistochem-
ically for a neurotransmitter, neuropeptide, receptor, or any other molecule
unique to its members (Figs. 10.3C,D and 10.6A–C; Jiao et al., 2000; Lei et al.,
2004; Luo et al., 2001; Reiner et al., 2000, 2003, 2004; Veenman et al., 1992).
DEXTRAN AMINES 317
Figure 10.6. Injection of BDA10kDa into the subthalamic nucleus in pigeon yields
retrograde labeling in globus pallidus. The labeled pallidosubthalamic projection
neuron in image A was visualized using ABC/DAB and differential interference
contrast microscopy. Note that it has a large perikaryon and long smooth dendrites.
Images B and C are of a single field within globus pallidus, captured using CLSM.
The BDA10kDa-labeled pallidal neuron was visualized with TRITC-conjugated an-
tibiotin (B). In the same sections, ENK+ terminals were immunolabeled using a
DTAF-conjugated secondary antiserum (C). The dendrite of the BDA-labeled pal-
lidosubthalamic neuron in B is surrounded and contacted by ENK+ terminals (ar-
rows) of striatal origin in C, as is characteristic of pallidal neurons that project to the
subthalamic nucleus. The asterisk in C marks the site of the BDA-labeled neuron
shown in B. (From Jiao et al., 2000.)
Such studies are performed using two distinct markers for visualizing the
two different populations of labeled structures, as in the case of combin-
ing BDA pathway tracing with another neuroanatomical tracer. Here too,
it is possible to use fluorescent dextran amines rather than biotinylated if
fluorescence viewing of labels is desired. Note that it may be necessary to
colchicine-treat animals near the end of the dextran amine survival time
to boost perikaryal immunoreactivity for the antigen of interest if it tends to
be rapidly shipped out of the cell body via the axon (Anderson and Reiner,
1990).
BDA can also be used in EM double-label studies in which one popula-
tion of axons and terminals is labeled with BDA10kDa or one population
of neuronal perikarya is labeled with BDA3kDa, and a second population
of axons/terminals, dendrites, and/or perikarya is labeled immunohisto-
chemically for the distinctive neurotransmitter, neuropeptide, receptor, or
other unique molecules it contains (Lei et al., 2004). Such EM double-label
studies are carried out in a manner similar to that for double-label studies in-
volving BDA and a second neuroanatomical tracer. Here too, using only one
EM marker (e.g., DAB) for both the BDA-labeled and the immunolabeled
structures is possible if the BDA10kDa anterogradely labeled terminals do
not have the same morphology as any of the immunolabeled terminals in
the field under study, or if the BDA3kDa-labeled perikarya under study do
not give rise to any terminals that resemble immunolabeled terminals in the
field under study (Lei et al., 2004).
III. ADVANTAGES AND LIMITATIONS
A. Merits of Dextran Amines Relative to Other Tracers
The efficacy of BDA as an anterograde or a retrograde tracer at the LM

level has been confirmed for all major vertebrate groups, including monkeys
(Brandt and Apkarian, 1992; Lanciego et al., 1998a), cats (Richmond et al.,
1994), rats (Brandt and Apkarian, 1992; Veenman et al., 1992; Wouterlood
and Jorritsma-Byham, 1993), birds (Striedter, 1994; Tellegen et al., 2001;
Veenman et al., 1992; Wang et al., 2004), reptiles (Davila et al., 2002; Guirado
et al., 1999; Kenigfest et al., 2000; Lanuza et al., 1998), amphibians (Fritzsch,
1993; Scalia et al., 1997), bony fish (Albert et al., 1999; Lopes-Correa et al.,
1998; Torres et al., 2002; Xue et al., 2001), and jawless fish (Pombal et al.,
1999). The efficacy of the dextran amines has also been shown in several
different neural regions, in the intact CNS, in the PNS, in partially dissected
preparations, in brain slices, and with both iontophoretic and pressure injec-
tion methods (Fritzsch, 1993; Novikov, 2001; Reiner et al., 2000; Veenman et
al., 1992; Wang et al., 2004). Iontophoretic BDA injections yield small well-
defined injection sites and extensive and clear labeling of axonal projec-
tion systems over distances of at least several centimeters from the injection
site (Figs. 10.1 and 10.2). A relatively wide range of pipette tip diameters
DEXTRAN AMINES 319
(10–50 µm) and current levels (2–5 µA) for iontophoretic injections and a
wide range of survival times appear to be effective for transport of BDA. Pres-
sure injection of BDA10kDa also yields extensive and detailed anterograde
labeling, while pressure injection of acidic BDA3kDa yields extensive retro-
grade labeling. Finally, BDA labeling can readily be combined with other
anterograde or retrograde labeling methods or with immunohistochemical
labeling methods, and it can be used at the EM level, as noted above. BDA
combines the advantages of amine-conjugated dextrans (sensitivity) and of
biotinylated compounds (permanent label). For these reasons, the use of
BDA10kDa for anterograde pathway tracing and BDA3kDa for retrograde
pathway tracing has come to be widespread. The availability of fluorescent
dextran amines or of mixes of fluorescent and BDAs adds further simplicity
and versatility to dextran amines as pathway-tracing tools.
BDA10kDa appears similar to PHA-L and RDA10kDa in its sensitivity,
although some have reported a superior sensitivity of BDA10kDa (Novikov,
2001). BDA10kDa has an advantage over PHA-L in that BDA10kDa requires
only an ABC procedure followed by a DAB reaction to be visualized, whereas
a multistep immunohistochemical procedure is required to visualize PHA-L
(Van Haeften and Wouterlood, 2000; Veenman et al., 1992; Wouterlood and
Jorritsma-Byham, 1993). Further, some authors have reported that PHA-L
labeling can be capricious (Groenewegen and Wouterlood, 1990; Schmued
et al., 1990), while we and others have found that BDA10kDa consistently
yields excellent anterograde labeling (Lanciego et al., 2000; Reiner et al.,
2000; Veenman et al., 1992). Biocytin and neurobiotin also can be used for
anterograde labeling and possess many of the advantages of BDA10kDa,
but both biocytin and neurobiotin are quickly catabolized and therefore
effective tracers over only short survival times (Izzo, 1991; King et al., 1989;
Kita and Armstrong, 1991; Lapper and Bolam, 1991; Novikov, 2001). Finally,
BDA10kDa is also a more sensitive anterograde tracer than is WGA-HRP
(Ferguson et al., 2001).
BDA3kDa is more sensitive than HRP for retrograde labeling and is a
more reliable and effective retrograde tracer than biocytin or neurobiotin
(Kaneko et al., 1996; Lapper and Bolam, 1991; Medina et al., 1997). BDA
appears to be more sensitive as both retrograde and anterograde tracers
than CTB (Chen and Aston-Jones, 1998), although CTB is highly sensi-
tive and useful (Reiner et al., 1996; Shimizu et al., 1994). For EM studies,
BDA3kDa is advantageous over CTB due to the simplicity and brevity of
the labeling procedure, which benefits EM ultrastructure, and due to the
greater sensitivity and cellular detail provided by 3-kDa dextran amines (Van
Haeften and Wouterlood, 2000; Wouterlood and Jorritsma-Byham, 1993).
While Fluorogold may be as or more sensitive than BDA3kDa in terms of
numbers of neurons retrogradely labeled from an injection of the same
size, BDA3kDa yields more extensive dendritic labeling, and Fluorogold is
toxic to both injected and retrogradely labeled neurons (Naumann et al.,
2000; Schmued et al., 1993; Schmued and Fallon, 1986). BDA3kDa appears
to be more robust than HRP or WGA-HRP as a retrograde neural tracer,
except for retrograde labeling from peripheral muscle (Faulkner et al.,

1997; Richmond et al., 1995; Todorova and Rodziewicz, 1995). In the case of
retrograde labeling from muscle, dextran amines label fewer neurons than
do other retrograde tracers, but do yield excellent dendritic labeling of the
few motoneurons that are labeled.
The primary advantage of BDA over fluorescent dextran amines is that
when visualized with DAB, BDA provides a permanent, electron-dense label
rather than a transient, fluorescent one. Chang (1991) has, however, shown
that Lucifer Yellow-conjugated dextran amine can be labeled immunohisto-
chemically with a permanent, electron-dense label using antisera against Lu-
cifer Yellow. Similarly, RDA can be visualized as a permanent label immuno-
histochemically with antisera against tetramethylrhodamine, while FDA can
be visualized by antifluorescein immunolabeling (Kaneko et al., 1996). Fi-
nally, while BDA10kDa appears to be a favorable anterograde tracer and
BDA3kDa a favorable retrograde tracer, investigators must bear in mind
that they need to optimize the procedures outlined here for their system.
B. Pitfalls and Solutions
1. Absence of Labeling for BDA After Peripheral Target Injection
While the 3-kDa dextran amines are excellent as retrograde tracers when
injected into neural targets, they appear relatively poor for retrograde
labeling from peripheral muscle (Faulkner et al., 1997; Richmond et al., 1993;
Todorova and Rodziewicz, 1995). After muscle injection, dextran amines la-
bel fewer neurons than do many other retrograde tracers but do produce
excellent dendritic labeling of the few motoneurons labeled. This attribute
of dextran amines may reflect their tendency not to diffuse far from the
injection and/or to a dependence on uptake by damaged nerve terminals.
2. Absence of Labeling for BDA After Nervous System Injection
There are three main reasons why BDA labeling might fail: a faulty
ABC/DAB procedure for visualizing BDA, a failure to actually inject BDA,
and too short a survival time. If there is no BDA labeling anywhere in the
brain after ABC/DAB processing, including the injection target, then a de-
fect in processing would be suspected. Carrying out tests of the ABC/DAB
reagents by using immunohistochemistry to visualize some plentiful anti-
gen (e.g., substance P) in sections in which the BDA labeling failed will
reveal whether the fault lies with the ABC/DAB procedure. If the immuno-
labeling test yields no DAB-labeled structures, the investigator would need
to systematically determine which ABC/DAB step or reagent is problem-
atic. One potential source of problems in the ABC/DAB procedure is the
use of sodium azide, which inhibits the peroxidase reaction, in the diluting
DEXTRAN AMINES 321
solution. On the other hand, if the immunohistochemistry yields plentiful
labeling and the track created by the microsyringe or micropipette used for
the BDA injection is evident, then defective ejection of BDA or defective
BDA would be suspected. To distinguish between these two possibilities, the
investigator needs to consider several possibilities. Was the lot of BDA bad?
Was it inappropriately stored? Was the pH of the BDA solution correct? Was
the pipette or syringe clogged? Was the current polarity correct during the
iontophoretic injection? Finally, if the injection site is labeled but transport
is present only near the injection site, a larger injection or longer survival
time is the likely solution.
3. Retrograde Labeling for BDA
The tendency of BDA10kDa, and similar-weight fluorescent dextran

amines, to retrogradely label neurons, at least with some injections and for
some brain regions, gives rise to the main limitation of BDA10kDa evident
to us. The retrograde labeling with BDA10kDa seems greater after axonal
damage and may be greater in amphibians and reptiles than in mammals. It
is now clear that axon collaterals of such retrogradely labeled neurons can
also be labeled. The presence of these labeled axon collaterals could then
make it difficult to know whether labeled axons and terminals in a partic-
ular structure result from anterograde labeling of neurons at the injection
site or from the retrogradely labeled neurons. A similar problem also ex-
ists for other sensitive anterograde tracers that result in some retrograde
labeling, such as WGA-HRP, PHA-L, CTB, biocytin, and neurobiotin (Chen
and Aston-Jones, 1998; Glover et al., 1986; Groenewegen and Wouterlood,
1990; Izzo, 1991; King et al., 1989; Kita and Armstrong, 1991; Nance and
Burns, 1990; Schmued et al., 1990; Shu and Peterson, 1988; Woodson et al.,
1991). In practice, this problem should not prevent a clear interpretation
of the results of BDA injections, since such collateral labeling with dextran
amine of 10 kDa should be minor in comparison with the anterograde la-
beling and since the retrogradely labeled neurons in most instances will not
project to the same targets as the neurons at the injection site. Nonetheless,
some caution must be exercised in interpreting the results of anterograde
labeling using any of the above tracers.
4. Fibers of Passage
Another factor that can complicate the interpretation of the results ob-
tained with various anterograde or retrograde tracers is the potential up-
take of tracer by fibers of passage. This possibility has not been investigated
extensively in the case of BDA, although it is clear that, like other trac-
ers, BDA is taken up by axons that have been damaged by the injection
(Brandt and Apkarian, 1992; Glover et al., 1986; Lei et al., 2004; Todorova
and Rodziewicz, 1995; Veenman et al., 1992). BDA also appears to be taken
up pinocytotically by intact axon terminals at the injection site ( Jiang et al.,
1993). Iontophoretic injection of tracer has been reported to minimize up-
take by fibers of passage, in studies using the fluorescent retrograde tracer
Fluorogold, since axonal damage is largely avoided by the thin glass mi-
cropipettes used for iontophoresis (Schmued and Heimer, 1990). It may
thus be beneficial to use iontophoresis for dextran amine injection if the
possibility of uptake by fibers of passage needs to be minimized.
APPENDIX
A. Step-by-Step BDA Protocol for LM Single-Labeling Studies
1. Injection
1. Intact vertebrates are deeply anesthetized and secured in a stereotaxic

device or appropriate headholder, while semi-intact preparations (such
as a partially dissected embryo or tissue slice) should be secured in
an appropriate fashion. The region of interest is exposed, and a mi-
crosyringe or micropipette containing BDA is lowered into the target
structure using stereotaxic methods or visual guidance.
2. For anterograde labeling using iontophoresis, micropipettes with tip
diameters of 10–50 µm are filled with 10–15% BDA10kDa in 0.01 M
sodium phosphate buffer (PB; buffers are at physiological pH, i.e.,
pH 7.2–7.4, unless otherwise noted). BDA10kDa can be injected ion-
tophoretically using 2–5 µA positive current pulses (7 s on and 7 s
off ) for a period of 30–60 min. Smaller micropipette (2–4 µm tip di-
ameters) and shorter iontophoretic sessions (2–5 min) favor smaller
injections, which might be more suitable for small animals, small brain
regions, or detailed studies of projection topography (Guirado et al.,
1999). During penetration and withdrawal of the pipette, the current
is reversed to prevent leakage of tracer along the penetration track.
3. Pressure injections can be made with either a micropipette (20–
50 µm tip diameter) or a microsyringe, using 2–15% BDA10kDa in
0.01 M PB for anterograde tracing and 10% BDA3kDa in 0.1 M sodium
citrate–HCl (pH 3.0) for retrograde labeling. For pressure injections
with a microsyringe, we use 0.01 µl steps per minute until the desired
amount (typically 0.05–0.20 µl per injection site) has been injected.
We and other investigators have typically pressure injected BDA3kDa
(Kaneko et al., 1996; Lei et al., 2004; Medina et al., 1997), but BDA3kDa
can be injected iontophoretically as well.
4. Note that it is also possible to place a crystal of BDA directly in or on
the desired site, or use fluorescent dextran amines (Nance and Burns,
1990; Pieribone and Aston-Jones, 1993; Schmued et al., 1990).
DEXTRAN AMINES 323
5. The region overlying the injection site is then closed/sutured and the
animal returned to normal housing. Semi-intact preparations are main-
tained using the standard procedures necessary for their viability (e.g.,
incubation in oxygenated saline solution or media) for a sufficient time
to allow for transport (Veenman et al., 1992).
2. Fixation
1. The postinjection survival period may be as brief as several hours if

transport over only a small distance is needed (e.g., peripheral nerve
labeling in a small specimen such as a chick embryo) (Veenman et al.,
1992) or 7–21 days for labeling over long CNS pathways in an adult
vertebrate (Brandt and Apkarian, 1992; Veenman et al., 1992).
2. For fixation of nervous tissue in intact animals for light microscopic
studies, the chest cavity is opened after deep anesthesia, heparinized
saline (12 mg heparin per ml physiological saline) injected into the
heart to prevent blood clotting, and the animal perfused transcar-
dially with 30–50 ml 6% dextran in 0.1 M sodium PB, followed by a
fixative consisting of freshly made 4% paraformaldehyde in 0.1 M PB,
to which we commonly add 0.1 M lysine and 0.01 M sodium periodate.
3. For small animals, tissues slices or partially dissected preparations,
immersion fixation is adequate.
4. For EM grade fixation, freshly made 2% glutaraldehyde in 0.1 M PB,
1.25% glutaraldehyde and 1% paraformaldehyde in 0.1 M PB, 0.5%
glutaraldehyde and 3% paraformaldehyde in 0.1 M PB, 0.6% glu-
taraldehyde, 3.5% paraformaldehyde, and 15% saturated picric acid
in PB, or 4% paraformaldehyde in 0.1 M acetate buffer (pH 4.5) fol-
lowed by 4% paraformaldehyde and 0.05% glutaraldehyde in 0.05 M
borate buffer (pH 9.5) are all effective for visualizing BDA-labeled ax-
ons or perikarya (Brandt and Apkarian, 1992; Lei et al., 2004; Reiner
et al., 2003; Veenman et al., 1992). Perfused tissues can be postfixed.
3. Sectioning
1. The tissue of interest is dissected free as necessary and cryoprotected

if it is to be sectioned frozen. We use cryoprotection in 0.1 M PB
containing 20% sucrose, 10% glycerol, and 0.02% sodium azide for
material to be sectioned frozen on a sliding microtome and 20%
sucrose PB for tissue to be sectioned frozen with a cryostat. Tissue
should be stored at 4◦ C until sectioned.
2. Sections are cut with either a sliding microtome, vibrating microtome
(i.e., Vibratome r
), or cryostat. Sliding microtome or vibrating mi-
crotome slices are collected as free-floating sections in 0.1 M PB and
stored at 4◦ C until further processing. Cryostat sections are mounted
onto gelatin-coated or Superfrost r
/Plus slides, as they are cut and
stored at 4◦ C until processed on the slide. Gelatin-coated slides are

prepared by dipping thoroughly washed slides in 2% gelatin in dis-
tilled water, draining them and then baking them at 50◦ C till dry.
3. Since BDA appears stable in fixed tissue (Brandt and Apkarian, 1992;
Veenman et al., 1992), sectioned or unsectioned tissue can be stored
for months at 4◦ C in 0.02% sodium azide and PB prior to processing.
4. BDA Visualization
1. For visualizing BDA, we employ the ABC procedure (Hsu et al., 1981),
using the Vectastain ABC Elite kit (Vector Laboratories). With the Elite
kit, one drop (50 µl) of avidin DH and one drop (50 µl) of biotinylated
HRP are mixed in 2.5–10 ml 0.1 M PB, at least half an hour prior to
use.
2. Sections are rinsed three times in 0.1 M PB and incubated in ABC
solution in a vial on a rotator or orbital shaker for 30–60 min at room
temperature, or overnight at 4◦ C (Brandt and Apkarian, 1992; Hsu and
Soban, 1982; Veenman et al., 1992; Wouterlood and Jorritsma-Byham,
1993).
3. Subsequently, the sections are rinsed several times in buffer, and
then the labeling visualized using a brown DAB reaction (Anderson
and Reiner, 1990; Hancock, 1986; Hsu and Soban, 1982; Zaborszky
and Heimer, 1989) or a metal-intensified DAB procedure for a dark
blue-black reaction product (Medina et al., 1997; Veenman et al., 1992),
the latter of which tends to be more sensitive and provide better con-
trast for viewing the BDA labeling.
4. The metal-intensified procedure that we have used involves incubat-
ing tissue for 10–15 min in a solution containing 0.05% DAB tetrahy-
drochloride and 0.04% nickel ammonium sulfate in 0.1 M sodium PB
(pH 7.2), followed by an additional 10–15 min of incubation with hy-
drogen peroxide added to a final concentration of 0.01% (Medina
et al., 1997).
5. For processing tissue fixed with a glutaraldehyde-containing fixative,
a permeabilization step (e.g., 1 h at room temperature in 0.03% Tri-
ton X-100 in 0.1 M PB) carried out prior to the ABC procedure may
help increase penetration of the ABC complex without damaging ul-
trastructure (Veenman et al., 1992; Wouterlood and Jorritsma-Byham,
1993).
6. Investigators can also visualize BDA with a fluorophore-conjugated
avidin or a fluorophore-conjugated antibiotin antiserum, although this
obviates the advantage (i.e., permanence) of using BDA in single-
labeling mapping studies. Finally, peroxidase immunolabeling can
be used to detect RDA or FDA, using antibodies against rhodamine
or fluorescein, respectively (Kaneko et al., 1996; Reiner et al., 2000,
2003).
DEXTRAN AMINES 325
5. Mounting and Viewing
1. Labeled free-floating and slide-mounted sections are rinsed, the free-

floating sections mounted onto gelatin-coated or Superfrost/Plus
slides, and slides then air-dried.
2. All DAB-labeled sections are next dehydrated through an ascending
alcohol series to xylene, coverslipped with Permount and examined.
3. DAB-labeled sections can also be counterstained with neutral red,
methyl green, or cresyl violet, depending on the color of the DAB
reaction product used prior to being coverslipped.
4. In the case of BDA visualized by a fluorescent label or with the use
of FDA or RDA, the sections should be rinsed in 0.1 M PB, mounted
on gelatin-coated or Superfrost/Plus slides, and coverslipped with 9:1
glycerin:0.05 M carbonate buffer, 9:1 glycerol:PB saline containing p-
phenylenediamine, or any of a number of commercially available fade-
retarding coverslipping media (Anderson and Reiner, 1990; Jiao et al.,
2000).
B. Step-by-Step BDA Protocol for LM Double-Labeling Studies
1. Injection and Fixation
1. For multiple-label anterograde or retrograde pathway-tracing studies,

all steps from the BDA injection up to the point of tissue processing
are the same as for BDA single labeling.
2. It should be noted that if the transport times of the two tracers differ
or if the distances from the projection target differ for the two injected
substances, the injections need to be carried out at separate times.
For example, assuming equal transport distances, markers transported
more slowly than BDA (e.g., PHA-L) need to be injected well before the
BDA, while markers transported more quickly than BDA (e.g., HRP)
need to be injected after the BDA.
3. Tissue fixation and sectioning are as for tissue labeled with dextran
amine alone, unless the second tracer constrains which fixative must be
used. For example, HRP retrograde tracing requires high glutaralde-
hyde concentration in the fixative. Conversely, most antigens to be
detected by immunolabeling call for little or no glutaraldehyde.
2. Visualization of BDA and Second Marker
1. For differentially visualizing the BDA and the second marker (e.g., a
pathway-tracing agent or an antigen detected by immunolabeling) at
the transmitted light level, two-color DAB procedures or DAB labeling
combined with VIP should be used (Alisky and Tolbert, 1994; Antal
et al., 1990; Hancock, 1986; Hsu and Soban, 1982; Lanciego et al., 1997,
1998a, b, 2000; Lanciego and Gimenez-Amaya, 1999; Medina et al.,
1997; Reiner et al., 1993; Veenman et al., 1992).
2. For combining BDA with immunohistochemical labeling, the BDA de-
tection can be carried out before or after the entire immunohisto-
chemical procedure (Lei et al., 2004; Reiner et al., 2004). For example,
immunohistochemical labeling of neurons can be performed using a
brown DAB reaction, while BDA-containing axons and terminals can
be labeled blue/black using a metal-intensified DAB reaction. Sim-
ilarly, neural structures labeled with one tracer (such as HRP) can
be visualized first with a blue/black metal–DAB reaction and BDA-
labeled axons visualized second with a brown DAB reaction (Veen-
man et al., 1992). Alternatively, both BDA and second neuroanatomical
tracer can be visualized by immunofluorescence using separate fluo-
rophores (or the dextran amines viewed directly if RDA or FDA is, e.g.,
used).
3. Other diverse pairs of markers are possible for distinct LM visualization
of BDA and a second tracer/marker, such as DAB and silver-intensified
immunogold (Anderson et al., 1991; Chan et al., 1990), DAB and BDHC
(Anderson et al., 1991; Levey et al., 1986), DAB and a glucose oxi-
dase reaction product (Piekut and Knigge, 1984), or DAB and an al-
kaline phosphatase reaction product (Falini et al., 1982). Moreover,
with the suitable combination of tracers and markers, LM triple and
even quadruple labelings are possible (Anderson and Reiner, 1990;
Kiss et al., 1988; Lanciego et al., 2000; Luo et al., 2001; Staines et al.,
1988; Wessendorf et al., 1990).
C. Step-by-Step BDA Protocol for EM Studies
1. Injection, Fixation, and Sectioning
1. Fibers and terminals that have been anterogradely labeled with

BDA10kDa can be visualized at the EM level, as can neurons or their col-
laterals that have been retrogradely labeled with BDA3kDa (Guirado
et al., 1999; Lei et al., 2004; Reiner et al., 2003; Veenman and Reiner,
1996; Wouterlood and Jorritsma-Byham, 1993).
2. For the use of BDA10kDa or BDA3kDa in ultrastructural studies, the
BDA injection and the procedures for visualization of BDA are the same
as for LM studies. The major difference is that a fixative suitable for
preservation of ultrastructural detail must be used. Wouterlood and
Jorritsma-Byham (1993) have used freshly made 4% paraformalde-
hyde, 0.1% glutaraldehyde, and 0.02% picric acid in 0.1 M PB, and
reported both good ultrastructural preservation and good BDA label-
ing. We have successfully used freshly made 0.6% glutaraldehyde, 3.5%
DEXTRAN AMINES 327
paraformaldehyde, and 15% saturated picric acid in 0.1 M sodium PB
(pH 7.3) (Lei et al., 2004; Reiner et al., 2003; Veenman and Reiner,
1996). The upper limit of glutaraldehyde concentration that will be
tolerated by BDA is uncertain, but available data indicate that up to
2.5% glutaraldehyde is consistent with good BDA labeling (Veenman
et al., 1992; Wouterlood and Jorritsma-Byham, 1993).
3. We routinely postfix our EM tissue overnight in the same fixative as
used for perfusion but without glutaraldehyde, and then section it at
50 µm on a vibrating microtome (Lei et al., 2004; Reiner et al., 2003).
2. Visualization of BDA
Penetration of the various reagents during the ABC labeling procedure

can be enhanced by freeze-thawing methods or brief 0.03% Triton X-100
treatment (Reiner et al., 1993; Wouterlood and Jorritsma-Byham, 1993). The
BDA-labeled tissue can then be osmicated, dehydrated, and plastic embed-
ded by standard procedures.
D. Step-by-Step BDA Protocol for EM Double-Labeling Studies
When combining BDA with another pathway tracer or immunolabeling at

the EM level, we recommend DAB labeling together with silver-intensified
immunogold labeling, BDHC labeling, VIP labeling, or silver-intensified
DAB labeling (Anderson et al., 1991, 1994; Chan et al., 1990; Groenewegen
and Wouterlood, 1990; Lei et al., 2004; Levey et al., 1986; Van Haeften and
Wouterlood, 2000). Generally, we carry out the BDA labeling, using a nickel-
enhanced blue-black DAB reaction product followed by the second labeling
(Lei et al., 2004). Note that with the suitable combination of tracers and
distinct markers, triple and even quadruple labelings at the EM level are
possible (Anderson et al., 1994; Anderson and Reiner, 1990; Groenewegen
and Wouterlood, 1990; Kiss et al., 1988; Staines et al., 1988; Van Haeften and
Wouterlood, 2000; Wessendorf et al., 1990; Wouterlood et al., 1993).
E. Expected Results
1. Injection Sites—LM
Iontophoretic injections of 10% BDA10kDa or BDA3kDa using mi-

cropipette with tip diameters of 20–50 µm yield small injection sites, typically
∼500 µm in diameter (Figs. 10.1A and 10.2A), in which dense neuropil la-
beling and Golgi-like neuronal labeling are observed. The injection sites
after pressure injections of 10–15% BDA are less well defined, with necrosis
possibly present at the injection core. Use of 2% BDA for pressure injections
may reduce such necrosis and yield better-defined injection sites (Naito and
Kita, 1994). Finally, BDA can spread along the injection track (Veenman
et al., 1995; Wouterlood and Jorritsma-Byham, 1993). Slow delivery from
a carefully cleaned pipette or syringe and slow withdrawal of the syringe
or micropipette should mitigate this problem (Brandt and Apkarian, 1992;
Veenman et al., 1992).
2. Anterograde Labeling—LM
Injections of BDA10kDa, by either iontophoresis or pressure, yield ex-

tensive anterograde labeling of axons and terminals (Figs. 10.1 and 10.2).
The BDA10kDa-labeled axons/terminals are labeled with great clarity and
detail (Figs. 10.1B,D and 10.2B). BDA is transported at least 15–20 mm over
a 1-week period. Longer survival times allow transport over even greater
distances, and longer survival times may enhance labeling over shorter dis-
tances, due to increased cumulative buildup of BDA over time. It is clear that
BDA transports relatively quickly, shows little degradation with time, and ap-
pears effective for a wide variety of projection systems, including those within
the CNS as well as those from the CNS to such peripheral targets as muscle
(Veenman et al., 1992).
3. Retrograde Labeling—LM
Retrograde labeling is often seen with BDA10kDa and is routine with

BDA3kDa (Figs. 10.1B,C, 10.3A–C, 10.4A,B, 10.5A,B, and 10.6A). For
BDA10kDa, the numbers of neurons retrogradely labeled and the loci of
the cell groups retrogradely labeled are variable even for injections of the
same target nucleus, although a small variable fraction of the retrogradely
labeled neurons shows a Golgi-like clarity and extent of labeling (Figs. 10.1C
and 10.6A). In general, the use of micropipette with tip diameter less than
40 µm may minimize the amount of retrograde labeling from iontophoretic
BDA10kDa injections, and retrograde labeling with pressure injections may
be reduced by using a lower concentration of BDA10kDa (e.g., 2%), which
still yields excellent anterograde labeling (Reiner et al., 1993). Even when
retrograde labeling occurs with BDA10kDa, retrogradely labeled neurons
are not necessarily present in all known sources of input to the injected
structure (Veenman et al., 1992; Wouterlood and Jorritsma-Byham, 1993).
BDA3kDa, by contrast, is a sensitive and dependable retrograde tracer.
4. Double Labeling—LM
Perikarya- or fibers/terminals-labeled brown with DAB using immunohis-

tochemical methods can readily be distinguished from BDA-labeled axons
DEXTRAN AMINES 329
stained blue-black with either nickel–DAB. Similarly, brown BDA-labeled
axons can also be distinguished from blue-black immunohistochemically
labeled perikarya and fibers in tissue double labeled for BDA and a second
marker. In general, both sequences for carrying out the double labeling
(i.e., immunohistochemistry then BDA, or BDA then immunohistochem-
istry) yield distinct two-color labeling, although carrying out the BDA/metal-
intensified DAB procedure second can result in darkening of the brown
DAB label. Similar results are obtained when combining BDA labeling with
another pathway labeling method (e.g., HRP or PHA-L visualized immuno-
histochemically). In general, it appears critical to the success of the two-color
procedure that the first DAB reaction be run to completion, to inactivate
the peroxidase and thereby avoiding color mixing during the second DAB-
based reaction. Lanciego and coworkers (Lanciego et al., 1997, 1998a, b,
2000; Lanciego and Gimenez-Amaya, 1999) have also shown the efficacy of
VIP in combination with DAB for two-color LM labeling.
5. Anterograde and Retrograde Labeling—EM
The characteristics of BDA labeling at the EM level depend on the marker

used to visualize the BDA. When DAB is the chromogen used, the distri-
bution and appearance of the DAB reaction product within BDA-labeled
structures is indistinguishable from that seen when DAB is used to visu-
alize PHA-L-labeled structures (Veenman and Reiner, 1996; Wouterlood
and Jorritsma-Byham, 1993). BDA-labeled axons, terminals, perikarya, den-
drites, and spines are uniformly filled with the flocculent DAB reaction prod-
uct. If other EM markers such as silver-intensified immunogold, BDHC, or
VIP are used to visualize the BDA or to visualize a second marker in com-
bination with BDA, the appearance of the labeling will be as in previous
studies using these markers (Anderson et al., 1991, 1994; Chan et al., 1990;
Groenewegen and Wouterlood, 1990; Levey et al., 1986; Van Haeften and
Wouterlood, 2000).
Acknowledgments. We thank Drs. C. L. Veenman, L. Medina, and Y. Jiao

for their contributions to the development of dextran amines as pathway-
tracing agents, and S. L. Cuthbertson for her excellent technical assistance
in preparing some of the material presented here. Our research is supported
by NS-19620 (AR), NS-28721 (AR), EY-05298 (AR), and NS-26386 (MGH).
REFERENCES
Albert, J. S., Yamamoto, N., Yoshimoto, M., Sawai, N., and Ito, H., 1999, Visual thalamote-
lencephalic pathways in the sturgeon Acipenser, a non-teleous actinopterygina fish, Brain
Behav. Evol. 53:156–172.
Alisky, J. M., and Tolbert, D. L., 1994, Differential labeling of converging afferent pathways using
biotinylated dextran amine and cholera toxin subunit B, J. Neurosci. Methods 52:143–148.
Anderson, K. D., Karle, E. J., and Reiner, A., 1991, Ultrastructural single- and double-label im-
munohistochemical studies of substance P-containing terminals and dopaminergic neu-
rons in the substantia nigra in pigeons, J. Comp. Neurol. 309:341–362.
Anderson, K. D., Karle, E. J., and Reiner, A., 1994, A pre-embedding triple-label electron
microscopic immunohistochemical method as applied to the study of multiple inputs to
defined nigral neurons, J. Histochem. Cytochem. 42:49–56.
Anderson, K. D., and Reiner, A., 1990, The extensive co-occurrence of substance P and dynor-
phin in striatal projection neurons: an evolutionarily conserved feature of basal ganglia
organization, J. Comp. Neurol. 295:339–369.
Antal, M., Freund, T. F., Somogyi, P., and McIlhinney, R. A. J., 1990, Simultaneous anterograde
labelling of two afferent pathways to the same target area with Phaseolus vulgaris leucoag-
glutinin and Phaseolus vulgaris leucoagglutinin conjugated to biotin or dinitrophenol, J.
Chem. Neuroanat. 3:1–9.
circuits in brain: opportunities and limitations, J. Neurosci. Methods 103:51–61.
Chan, J., Aoki, C., and Pickel, V. M., 1990, Optimization of differential immunogold-silver and
peroxidase labeling with maintenance of ultrastructure in brain sections before plastic
embedding, J. Neurosci. Methods 33:113–127.
Chang, H. T., 1991, Anterograde transport of Lucifer Yellow— dextran conjugate, Brain Res.
Bull. 26:813–816.
Chen, A., and Aston-Jones, G., 1998, Axonal collateral–collateral transport of tract trac-
ers in brain neurons: false anterograde labeling and useful tool, Neuroscience 82:1151–
1163.
Coolen, L. M., Jansen, H. T., Goodman, R. L., Wood, R. I., and Lehman, M. N., 1999, A new
method for simultaneous demonstration of anterograde and retrograde connections in
the brain: co-injections of biotinylated dextran amine and the beta subunit of cholera
toxin, J. Neurosci. Methods 91:1–8.
Cowan, R. L., and Wilson, C. J., 1994, Spontaneous firing patterns and axonal projections
of single corticostriatal neurons in the rat medial agranular cortex, J. Neurophysiol. 71:
17–32.
Cowan, W. M., Gottlieb, D. I., Hendrickson, A. E., Price, J. L., and Woolsey, T. A., 1972, The
autoradiographic demonstration of axonal connections in the central nervous system,
Brain Res. 37:21–51.
Davila, J. C., Andreu, M. J., Real, M. J., Puelles, L., and Guirado, S., 2002, Mesencephalic
and diencephalic afferent connections to the thalamic nucleus rotundus in the lizard,
Psammodromus algirus, Eur. J. Neurosci. 16:267–282.
Dolleman-van der Weel, M. J., Wouterlood, F. G., and Witter, M. P., 1994, Multiple antero-
grade tracing combining Phaseolus vulgaris leucoagglutinin with rhodamine- and biotin-
Edwards, S. B., 1972, The ascending and descending projections of the red nucleus in the
cat: an experimental study using an autoradiographic tracing method, Brain Res. 48:
45–63.
Falini, B., De Solas, I., Halverson, C., Parker, J. W., and Taylor, C. R., 1982, Double labeled-
antigen method for demonstration of intracellular antigens in paraffin-embedded tissue,
J. Histochem. Cytochem. 30:21–26.
Faulkner, B., Brown, T. H., and Evinger, C., 1997, Identification and characterization of rat
orbicularis oculi motoneurons using confocal laser scanning microscopy, Exp. Brain Res.
116:10–19.
Ferguson, I. A., Xian, C., Barati, E., and Rush, R. A., 2001, Comparison of wheat germ agglutinin-
horseradish peroxidase and biotinylated dextran for anterograde tracing of corticospinal
tract following spinal cord injury, J. Neurosci. Methods 109:81–89.
Fink, R. P., and Heimer, L., 1967, Two methods for selective silver impregnation of degenerating
axons and their synaptic endings in the central nervous system, Brain Res. 4:369–374.
DEXTRAN AMINES 331
Fritzsch, B., 1993, Fast axonal diffusion of 3000 molecular weight dextran amines, J. Neurosci.
Methods 50:95–103.
Fritzsch, B., and Wilm, C., 1990, Dextran amines in neuronal tracing, Trends Neurosci. 13:14.
Gerfen, C. R., Sawchenko, P. E., and Carlsen, J., 1989. The PHA-L anterograde axonal tracing
method, In: Heimer, L., and Zaborszky, L. (eds.), Neuroanatomical Tract-Tracing Methods II,
Recent Progress, New York: Plenum Publishing Corporation, pp. 19–47.
Glover, J. C., Petursdottir, G., and Jansen, K. S., 1986, Fluorescent dextran-amines used as axonal
tracers in the nervous system of the chicken embryo, J. Neurosci. Methods 18:243–254.
Gonzalo, N., Moreno, A., Erdozain, M. A., Garcia, P., Vazquez, A., Castle, M., and Lanciego, J.
L., 2001, A sequential protocol combining dual neuroanatomical tract-tracing with visual-
ization of local circuit neurons within the striatum, J. Neurosci. Methods 111:59–66.
Groenewegen, H. J., and Wouterlood, F. G., 1990, Light and electron microscopic tracing of
neuronal connections with Phaseolus vulgaris-leucoagglutinin (PHA-L), and combinations
with other neuroanatomical techniques, In: Bjorklund, A., Hokfelt, T., Wouterlood, F.
G., and van den Pol, A. N. (eds.), Handbook of Chemical Neuroanatomy. Volume 8. Analysis
of Neuronal Microcircuits and Synaptic Interactions, Amsterdam: Elsevier Science Publishers,
pp. 47–124.
Guirado, S., Real, M. A., Davila, J. C., and Medina, L., 1999, The nucleus accumbens in the
lizard Psammodromus algirus: chemoarchitecture and cortical afferent connections, J. Comp.
Neurol. 405:15–31.
Hancock, M. B., 1986, Two-color immunoperoxidase staining: visualization of anatomic rela-
tionships between immunoreactive neural elements, Am. J. Anat. 175:343–352.
Haugland, R. P., 1996, Handbook of Fluorescent Probes and Research Chemicals, Eugene, OR: Molec-
ular Probes Inc.
Hendrickson, A., and Edwards, S. B., 1978, The use of axonal transport for autoradiographic
tracing of pathways in the central nervous system, In: Thompson, R. F., and Robertson, R.
T. (eds.), Methods in Physiological Psychology. Volume 2. Neuroanatomical Research Techniques,
New York: Academic Press, pp. 241–290.
Hsu, S. M., Raine, L., and Fanger, H., 1981, Use of avidin–biotin–peroxidase complex (ABC)
in immunoperoxidase techniques. A comparison between ABC and unlabeled antibody
(PAP) procedures, J. Histochem. Cytochem. 29:577–580.
Hsu, S. M., and Soban, E., 1982, Color modification of diaminobenzidine (DAB) precipita-
tion by metallic ions and its application for double immunohistochemistry, J. Histochem.
Cytochem. 30:1079–1082.
Izzo, P. N., 1991, A note on the use of biocytin in anterograde tracing studies in the central ner-
vous system: applications at both light and electron microscopic level, J. Neurosci. Methods
36:155–166.
Jiang, X., Johnson, R. R., and Burkhalter, A., 1993, Visualization of dendritic morphology
of cortical projection neurons by retrograde axonal tracing, J. Neurosci. Methods 50:
45–60.
Jiao, Y., Medina, L., Veenman, C. L., Toledo, C., Puelles, L., and Reiner, A., 2000, Identification
of the anterior nucleus of the ansa lenticularis in birds as the homologue of the mammalian
subthalamic nucleus, J. Neurosci. 20:6998–7010.
Kaneko, T., Saeki, K., Lee, T., and Mizuno, N., 1996, Improved retrograde axonal trans-
port and subsequent visualization of tetramethylrhodamine (TMR)— dextran amine by
means of an acidic injection vehicle and antibodies against TMR, J. Neurosci. Methods 65:
157–165.
Kenigfest, N. B., Belekhova, M. G., Reperant, J., Rio, J. P., Vesselkin, N. P., and Ward, R., 2000,
Pretectal connections in turtles with special reference to the visual thalamic centers: a
hodological and gamma-aminobutyric acid-immunohistochemical study, J. Comp. Neurol.
426:31–50.
Kincaid, A. E., and Wilson, C. J., 1996, Corticostriatal innervation of the patch and matrix in
the rat neostriatum, J. Comp. Neurol. 374:578–592.
King, M. A., Louis, P. M., Hunter, B. E., and Walker, D. W., 1989, Biocytin: a versatile anterograde
neuroanatomical tract-tracing alternative, Brain Res. 497:361–367.
Kiss, A., Palkovits, M., and Skirboll, L. R., 1988, Light microscopic triple-colored immunohis-
tochemical staining on the same vibratome section using the avidin–biotin–peroxidase
complex technique, Histochemistry 88:353–356.
Kita, H., and Armstrong, W., 1991, A biocytin-containing compound N-(2-aminoethyl) bioti-
namide for intracellular labeling and neuronal tracing studies: comparison with biocytin,
J. Neurosci. Methods 37:141–150.
Lanciego, J. L., and Gimenez-Amaya, J. M., 1999, Notes on the combined use of V-VIP and
DAB peroxidase substrates for the detection of colocalizating antigens, Histochem. Cell Biol.
111:305–311.
Lanciego, J. L., Goede, P. H., Witter, M. P., and Wouterlood, F. G., 1997, Use of peroxidase
substrate Vector r VIP for multiple staining in light microscopy, J. Neurosci. Methods 74:1–7.
Lanciego, J. L., Luquin, M. R., Guillen, J., and Gimenez-Amaya, J. M., 1998a, Multiple neu-
roanatomical tracing in primates, Brain Res. Protoc. 2:323–332.
Lanciego, J. L., and Wouterlood, F. G., 1994, Dual anterograde axonal tracing with Phaseolus
vulgaris leucoagglutinin (PHA-L) and biotinylated dextran amine (BDA), Neurosci. Protoc.
94-050-06.
Lanciego, J. L., Wouterlood, F. G., Erro, E., Arribas, J., Gonzalo, N., Urra, X., Cervantes, X.,
and Gimenez-Amaya, J. M., 2000, Complex brain circuits studied via simultaneous and per-
manent detection of three transported neuroanatomical tracers in the same histological
section, J. Neurosci. Methods 103:127–135.
Lanciego, J. L., Wouterlood, F. G., Erro, E., and Gimenez-Amaya, J. M., 1998b, Multiple axonal
tracing: simultaneous detection of three tracers in the same section, Histochem. Cell Biol.
110:509–515.
Lanuza, E., Belekhova, M., Martinez-Marcos, A., Font, C., and Martinez-Garcia, F., 1998, Identi-
fication of the reptilian basolateral amygdala: an anatomical investigation of the afferents
to the posterior dorsal venricular ridge of the lizard Podarcis hispanica, Eur. J. Neurosci.
20:3517–3534.
Lapper, S. R., and Bolam, J. P., 1991, The anterograde and retrograde transport of neurobiotin
in the central nervous system of the rat: comparison with biocytin, J. Neurosci. Methods
39:163–174.
Lei, W. L., Jiao, Y., Del Mar, N., and Reiner, A., 2004, Evidence for differential cortical input
to direct pathway versus indirect pathway striatal projection neurons in rats, J. Neurosci.
24:8289–8299.
Levey, A. I., Bolam, J. P., Rye, D. B., Hallanger, A. E., Demuth, R. M., Mesulam, M. M., and
Wainer, B. H., 1986, A light and electron microscopic procedure for sequential double
antigen localization using diaminobenzidine and benzidine dihydrochloride, J. Histochem.
Cytochem. 34:1449–1457.
Liu, W. L., Behbehani, M. M., and Shipley, M. T., 1993, Intracellular filling in fixed brain slices
using miniruby, a fluorescent biocytin compound, Brain Res. 608:78–86.
Lopes-Correa, S. A., Grant, K., and Hoffmann, A., 1998, Afferent and efferent connections of
the dorsolateral telencephalon in an electrosensory teleost, Gymnotus carapo, Brain Behav.
Evol. 52:81–98.
Luo, P., Haines, A., and Dessem, D., 2001, Elucidation of neuronal circuitry: protocol(s) com-
bining intracellular labeling, neuroanatomical tracing and immunocytochemical method-
ologies, Brain Res. Protoc. 7:222–234.
Medina, L., and Reiner, A., 1997, The efferent projections of the dorsal and ventral pallidal
parts of the pigeon basal ganglia, studied with biotinylated dextran amine, Neuroscience
81:773–802.
Medina, L., Veenman, C. L., and Reiner, A., 1997, Evidence for a possible avian dorsal thalamic
region comparable to the mammalian ventral anterior, ventral lateral, and oral ventropos-
terolateral nuclei, J. Comp. Neurol. 384:86–108.
DEXTRAN AMINES 333
Naito, A., and Kita, H., 1994, The cortico-pallidal projection in the rat: an anterograde tracing
study with biotinylated dextran amine, Brain Res. 653:251–257.
Nance, D. M., and Burns, J., 1990, Fluorescent dextrans as sensitive anterograde neuroanatom-
ical tracers: applications and pitfalls, Brain Res. Bull. 25:139–145.
Naumann, T., Härtig, W., and Frotscher, M., 2000, Retrograde tracing with Fluoro-Gold: differ-
ent methods of tracer detection at ultrastructural level and neurodegenerative changes of
back-filled neurons in long-term studies. J. Neurosci. Methods 103:11–21.
Nauta, W. J. H., and Gygax, P. A., 1954, Silver impregnation of degenerating axons in the CNS.
A modified technique, Stain Technol. 29:91–93.
Novikov, L. N., 2001, Labeling of central projections of primary afferents in adult rats: a
comparison between biotinylated dextran amine, neurobiotin r and Phaseolus vulgaris-
leucoagglutinin, J. Neurosci. Methods 112:145–154.
Novikova, L., Novikov, L., and Kellerth, J. O., 1997, Persistent neuronal labeling by retrograde
Ojima, H., and Takayanagi, M., 2001, Use of two anterograde axon tracers to label distinct
cortical populations located in close proximity, J. Neurosci. Methods 104:177–182.
Peters, A., 1985, The visual cortex of the rat, In: Peters, A., and Jones, E. G. (eds.), Cerebral
Cortex. Volume 3. Visual Cortex, New York: Plenum Publishing Corporation, pp. 19–80.
Piekut, D. T., and Knigge, K. M., 1984, Relationship of alpha MSH-specific neurons to the
arcuate opiocortin neuronal system as determined by dual antigen immunocytochemical
procedures, Peptides 5:1089–1095.
Pieribone, V. A, and Aston-Jones, G., 1993, The iontophoretic application of Fluoro-Gold for
the study of afferents to deep brain nuclei. Brain Res. 607:47–53.
Pombal, M. A., Yanez, J., Marin, O., Gonzalez, A., and Anadon, R., 1999, Cholinergic and
GABAergic neuronal elements in the pineal organ of lampreys, and tract-tracing ob-
servations of differential connections of pinealofugal neurons, Cell Tissue Res. 295:
215–223.
terograde and rerograde neuronal tracer, Brain Res. 607:47–53.
Reiner, A., Jiao, Y., Del Mar, N., Laverghetta, A. V., and Lei, W. L., 2003, Differential morphology
of pyramidal-tract type and intratelencephalically-projecting type corticostriatal neurons
and their intrastriatal terminals in rats, J. Comp. Neurol. 457:420–440.
Reiner, A., Laverghetta, A. V., Meade, C. A., Cuthbertson, S. L., and Bottjer, S. W., 2004, An
immunohistochemical and pathway tracing study on the striatopallidal organization of
area X the in male zebra finch, J. Comp. Neurol. 469:239–261.
Reiner, A., Veenman, C. L., and Honig, M. G., 1993, Anterograde tracing using biotinylated
dextran amine, Neurosci. Protoc. 93-050-14.
Reiner, A., Veenman, C. L., Medina, L., Jiao, Y, Del Mar, N., and Honig, M. G., 2000, Pathway
Reiner, A., Zhang, D., and Eldred, W. D., 1996, Use of cholera toxin tracer reveals new details
of the central retinal projections in turtles, Brain Behav. Evol. 48:307–337.
Richmond, F. J, Gladdy, R., Creasy, J. L., Kitamura, S., Smits, E., and Thomson, D. B., 1995,
Efficacy of seven retrograde tracers, compared in multiple-labeling studies of feline mo-
toneurones, J. Neurosci. Methods 53:35–46.
Scalia, F., Galoyan, S. M., Eisner, S., Haris, E., and Su, W., 1997, Biotinylated dextran amine and
biocytin hydrochloride are useful tracers for the study of retinal projections in the frog, J.
Schmued, L. C., Beltramino, C., and Slikker, W., Jr., 1993, Intracranial injection of Fluoro-
Gold results in the degeneration of local but not retrogradely labeled neurons, Brain Res.
626:71–77.
Schmued, L. C., and Fallon, J. H., 1986, Fluoro-Gold: a new fluorescent retrograde axonal
tracer with numerous unique properties, Brain Res. 377:147–154.
Schmued, L., Kyriakidis, K., and Heimer, L., 1990, In vivo anterograde and retrograde axonal
transport of the fluorescent rhohamine-dextran-amine, Fluoro-Ruby, within the CNS, Brain
Res. 526:127–134.
Sesack, S. R., Miner, L. H., and Omelchenko, N., 2006, Pre-embedding immunoelectron mi-
croscopy: applications for studies of the nervous system, In: Zaborszky, L., Wouterlood, F.
G., and Lanciego, J. L. (eds.), Neuroanatomical Tract-Tracing Methods 3: Molecules–Neurons–
Systems, New York: Springer/Kluwer/Plenum Publishers, pp. xxx–xxx..
Shimizu, T., Cox, K., Karten, H. J., and Britto, L. R. G., 1994, Cholera toxin mapping of retinal
projections in pigeons (Columba livia), with emphasis on retinohypothalamic connections,
Vis. Neurosci. 11:441–446.
Shu, S. Y., and Peterson, G. M., 1988, Anterograde and retrograde axonal transport of Phaseolus
vulgaris leucoagglutinin (PHA-L) from the globus pallidus to the striatum of the rat, J.
Staines, W. A., Meister, B., Melander, T., Nagy, J. I., and Hokfelt, T., 1988, Three-color im-
munofluorescence histochemistry allowing triple labeling within a single section, J. His-
tochem. Cytochem. 36:145–151.
Striedter, G. F., 1994, The vocal control pathways in budgerigars differ from those in songbirds,
J. Comp. Neurol. 343:35–56.
Tellegen, A. J, Arends, J. J., and Dubbeldam, J. L., 2001, The vestibular nuclei and vesitibuloretic-
ular connections in the mallard (Anas platyrhynchos L.). An anterograde and retrograde
tracing study, Brain Behav. Evol. 58:205–217.
Todorova, N., and Rodziewicz, G. S., 1995, Biotin-dextran: fast retrograde tracing of sciatic
nerve motoneurons, J. Neurosci. Methods 61:145–150.
Torres, B., Perez-Perez, M. P., Herrero, L., Ligero, M., and Nunez-Abades, P. A., 2002, Neural
substrata underlying tectal eye movement codification in goldfish, Brain Res. Bull. 57:345–
348.
Van Den Pol, A. N., and Decavel, C., 1990, Synaptic interactions between chemically defined
neurons: dual ultrastructural immunocytochemical approaches, In: Bjorklund, A., Hok-
felt, T., Wouterlood, F. G., and Van Den Pol, A. N. (eds.), Handbook of Chemical Neuroanatomy.
Volume 8. Analysis of Neuronal Microcircuits and Synaptic Interactions, Amsterdam: Elsevier Sci-
ence Publishers, pp. 199–271.
Van Haeften, T., and Wouterlood, F. G., 2000, Neuroanatomical tracing at high resolution, J.
Veenman, C. L., and Reiner, A., 1996, Ultrastructural morphology of synapses formed by cor-
ticostriatal terminals in the avian striatum, Brain Res. 707:1–12.
Veenman, C. L., Reiner, A., and Honig, M. G., 1992, Biotinylated dextran amine as an antero-
grade tracer for single- and double-label studies, J. Neurosci. Methods 41:239–254.
Veenman, C. L., Wild, J. M., and Reiner, A., 1995, Organization of the avian “corticostriatal”
projection system: a retrograde and anterograde pathway tracing study in pigeons, J. Comp.
Neurol. 354:87–126.
Vercelli, A., Repici, M., Garbossa, D., and Grimaldi, A., 2000, Recent techniques for tracing
pathways in the central nervous system of developing and adult mammals, Brain Res. Bull.
51:11–28.
Wang, Y., Major, D. E., and Karten, H. J., 2004, Morphology and connections of nucleus isthmi
pars magnocellularis in chicks (Gallus domesticus), J. Comp. Neurol. 469:275–297.
Wessendorf, M. W., Appel, N. M., Molitor, T. W., and Elde, R. P., 1990, A method for immunoflu-
orescent demonstration of three co-existing neurotransmitters in rat brain and spinal cord,
using the fluorophores fluorescein, lissamine rhodamine and 7-amino-4-methylcoumarin-
3-acetic acid, J. Histochem. Cytochem. 38:1859–1877.
Wilson, C. J., 1987, Morphology and synaptic connections of crossed corticostriatal neurons in
the rat, J. Comp. Neurol. 263:567–580.
Woodson, W., Reiner, A., Anderson, K. D., and Karten, H. J., 1991, The distribution, laminar
location and morphology of tectal neurons projecting to the isthmo-optic nucleus and
the nucleus isthmi, pars parvocellularis in the pigeon (Columba livia) and chick (Gallus
domesticus): a retrograde labeling study, J. Comp. Neurol. 305:470–488.
DEXTRAN AMINES 335
biotinylated dextran amine: comparison with the tracer Phaseolus vulgaris leucoagglutinin
in preparations for electron microscopy, J. Neurosci. Methods 48:75–87.
Wouterlood, F. G., Pattiselanno, A., Jorritsm-Byham, B., Arts, M. P. M., and Meredith, G. E.,
1993, Connectional, immunocytochemical and ultrastructural characterization of neurons
injected intracellularly in fixed brain tissue, In: Meredith, G. E., and Arbuthnott, G. W.
(eds.), Morphological Investigations of Single Neurons in Vitro, New York: John Wiley and Sons,
pp. 47–169.
Wright, A. K., Norrie, L., Ingham, C. A., Hutton, A. M., Arbuthnott, G. W., 1999, Double
anterograde tracing of the outputs from adjacent “barrel columns” of rat somatosen-
sory cortex neostriatal projection patterns and terminal ultrastructure, Neuroscience
88:119–133.
Wright, A. K., Ramanthan, S., and Arbuthnott, G. W., 2001, Identification of the source of the
bilateral projection system from cortex to somatosensory neostriatum and an exploration
of its physiological actions, Neuroscience 103:87–96.
Xue, H. G., Yamamoto, N., Yoshimoto, M., Yang, C. Y., and Ito, H., 2001, Fiber connections of
the nucleus isthmian in the carp (Cyprinus carpio) and tilapia (Oreochromis niloticus), Brain
Behav. Evol. 58:185–204.
Zaborszky, L., and Heimer, L., 1989, Combination of tracer techniques, especially HRP and
PHA-L, with transmitter identification for correlated light and electron microscopic stud-
ies, In: Heimer, L., and Zaborszky, L. (eds.), Neuroanatomical Tract-Tracing Methods II, Recent
Progress, New York: Plenum Publishing Corporation, pp. 173–199.
11
Multiple Neuroanatomical
Tract-Tracing: Approaches for
Multiple Tract-Tracing
JOSÉ L. LANCIEGO and
FLORIS G. WOUTERLOOD
TRACT-TRACING METHODS: HISTORICAL PERSPECTIVE

Retrograde and Anterograde Tracers
Previous and Current Generations of Tracing Methods
Horseradish Peroxidase and its Conjugates
Cholera Toxin Subunit B
Fluorescent Dyes
Phaseolus vulgaris Leucoagglutinin and Biotinylated
Dextran Amine
Viruses
PARADIGMS AND SELECTION OF TRACERS FOR COMBINED USE
The Complexity of the Brain Requires Multiple Tracing
“Golden Rules”: Requirements for a Tracer to Be Successful in
a Multitracing Paradigm
Winning Tracers
Winning Combinations of Tracers
EVENTS AT THE INJECTION SITE
Uptake Mechanisms
BDA May Cause Track Labeling
The Effective Injection Site Shape, Size, and Delivery of Tracer
PHA-L
Cholera Toxin Subunit B
JOSÉ L. LANCIEGO • Neuromorphology-Tracing Lab, Neurosciences Division, Center

for Applied Medical Research (C.I.M.A.), University of Navarra Medical School, Pio XII Avenue,
31008 Pamplona, Spain FLORIS G. WOUTERLOOD • Department of Anatomy,
Vrije Universiteit Medical Center, Rm MF-G-136, P.O. Box 7057, 1007 MB Amsterdam, The
Netherlands
336
MULTIPLE NEUROANATOMICAL TRACT-TRACING 337
Fluoro-Gold
Survival Time
Deposition of an Anterograde and a Retrograde
Tracer in the Same Spot
VISUALIZATION METHODS: MULTIPLE IMMUNOPEROXIDASE FOR
BRIGHT FIELD MICROSCOPY
Troubleshooting
Results
APPENDIX
Step-By-Step Procedure
Technical Tips
REFERENCES
Abstract: Experimental neuroanatomical tracing techniques are fundamental to the

study of the structure of the central nervous system. In the last few decades, many new
methods for axonal tracing and cell labeling have been introduced. Neuroanatom-
ical tracing applied as an isolated method produces relatively straightforward an-
swers, for instance, whether there is connectivity from compartment Y in nucleus
A to layer X in area B. However, questions that deal with the intrinsic complexity
of brain circuits require the application of multiple-tracing paradigms in which two
or even three different tracers are combined in single histological sections. With
such paradigms we can handle questions like “are the fibers arriving in layer X of
area B in contact with neurons that project to compartment Z in nucleus C,” “do
these projection neurons receive as well innervation from area W,” and “what is the
neurochemical signature of these connectivity-identified neurons?” We illustrate this
approach with examples from our studies on pallidonigral connectivity in association
with nigrostriatal efferent neurons.
Analysis of the data acquired via a multiple-tracing approach provides more in-
sight into the organization of the brain than does the analysis of data from single
tracing, especially when it comes to network circuitry. Furthermore, by virtue of the
simultaneous visualization of projections in the same section, these multiple tech-
niques enable the precise determination of the degree of convergence or divergence
of particular projections to a particular terminal zone or to particular neurons (the
latter to be identified via retrograde tracing or via neurotransmitter immunocyto-
chemistry). An additional advantage of multitracer methods is that the experimental
animals can be most efficiently used and the number of used animals reduced.
In this chapter we will discuss in detail several existing protocols for the simulta-
neous detection of three different tracers, as well as methods in which we combine
two tracers and the immunocytochemical detection of a neuroactive substance. Em-
phasis will be placed on providing a step-by-step account of each procedure. We will
be dealing with peroxidase substrates and precipitates with different colors since
these precipitates are persistent without specific storage measures and because at
the end of the staining procedure the ensuing slides can be studied any time under
any routine microscope.
Keywords: cholera toxin, dextran amines, Fluoro-Gold, immunocytochemistry, per-

oxidase, triple staining
338 JOSÉ L. LANCIEGO and FLORIS G. WOUTERLOOD
I. TRACT-TRACING METHODS: HISTORICAL PERSPECTIVE
A. Retrograde and Anterograde Tracers
Modern neuroanatomical tracing is based on axonal flow, an inherent

physiological transport mechanism in neurons first reported in 1948
by Weiss and Hiscoe. Depending on the direction of transport, neu-
roanatomical tracers are divided into two main groups, i.e., retrograde and
anterograde tracers. Following injection into the brain, a retrograde tracer
is taken up by axon terminals, incorporated in transport vesicles, and then
transported back to the parent cell bodies. As a result of the delivery of a
retrograde tracer to a given brain area, the somata of neurons that project
to that area become labeled. However, in most cases the accumulation of
retrogradely transported tracer is restricted to the cell somata and the main,
thickest dendrites. Anterograde tracers, by contrast, are taken up by the cell
somata or dendrites, and they are transported from the cell bodies into the
axons, where they can finally be detected in the most distal arborizations
and terminal boutons. The successful application of a sensitive anterograde
tracer results in the staining of the cell somata and complete dendritic trees
of the labeled cells, and in addition in the staining of the entire axonal
configurations of the labeled neurons including their collateral branches
and appendages, their terminal fields, arborizations, rosettes, and terminal
boutons.
B. Previous and Current Generations of Tracing Methods
The previous generation of tracing methods, described in the first edi-

tion of Neuroanatomical Tract-Tracing Methods, included silver-degeneration
methods (de Olmos et al., 1981), Golgi silver impregnation (Blackstad, 1981;
Millhouse, 1981), and the autoradiographic method capitalizing on the up-
take (by neurons) of injected, radioactively labeled amino acids, their in-
corporation in proteins, and subsequent anterograde transport (Edwards
and Hendrickson, 1981). As these methods were fully developed at the time
when the first edition of Neuroanatomical Tract-Tracing Methods was published,
we will not further discuss them here. The current generation of tracing
methods includes the tracers discussed briefly below.
C. Horseradish Peroxidase and its Conjugates
The current generation of tracing methods has its origin in the pub-
lication by Kristensson and Olson in 1971, describing the uptake and
retrograde axonal transport of the glycoprotein enzyme horseradish per-
oxidase (HRP). Transported HRP is histochemically visualized using the
enzymatic electron transfer from the substrate diaminobenzidine (DAB;
Sigma) to hydrogen peroxide through which a brown, insoluble DAB poly-
mer is formed (Graham and Karnovsky, 1966; LaVail, 1975; see also Warr
et al., 1981). Improvements and refinements were added soon to the HRP
transport technique (LaVail and LaVail, 1972; Mesulam, 1976, 1978, 1982).
The enzyme was conjugated with the plant lectin wheat germ agglutinin
(WGA; WGA-HRP; Gonatas et al., 1979), generating a highly effective and
sensitive bidirectional tracer (i.e., transported both anterogradely and ret-
rogradely). Soon it was discovered that WGA-HRP could be transferred from
one neuron to another neuron at the synaptic junction (the so-called trans-
synaptic transport).
D. Cholera Toxin Subunit B
A decade after the development of the HRP method, the β subunit of

cholera toxin (CTB; List Biological Laboratories, Campbell, CA) was in-
troduced as a substance transported both anterogradely and retrogradely
(Trojanowski et al., 1981). Like HRP, CTB works highly effectively both in
the peripheral and in the central nervous systems (Stoeckel et al., 1977).
CTB has a general performance as a retrograde tracer that is superior to
that of HRP (Trojanowski et al., 1981, 1982; Wan et al., 1982). Visualization
of transported CTB can easily be combined with other immunocytochemi-
cal procedures, such as neuropeptide or neurotransmitter detection (Luppi
et al., 1987, 1990), and with other anterograde tracing techniques (Bruce
and Grofova, 1992; Vetter et al., 1993). Several procedures have been devel-
oped to further improve the performance of CTB in anterograde tracing
experiments (Angelucci et al., 1996).
E. Fluorescent Dyes
Parallel to the development of the HRP and CTB methods, retrograde

tracing techniques were refined utilizing uptake and transport of fluores-
cent dyes: Evans Blue, Fast Blue, Diamidino Yellow, Propidium Iodide, True
Blue, Nuclear Yellow, and Lucifer Yellow to mention just a few (for a review,
see Akintunde and Buxton, 1992; Kuypers and Huisman, 1984). Soon the
major advantages of fluorescence-based retrograde tracing were recognized
by neuroscientists: fast, highly productive, compatibility with chemical char-
acterization of neurons, and compatibility with anterograde and retrograde
neuroanatomical tracing (Skirboll et al., 1989). One of the advantages of
fluorescence methods was also that the study of axonal collateralization
became easier than with the silver methods. The rapid development in con-
focal laser scanning microscopy and in parallel that of new, bright, and stable
fluorochromes have been instrumental for a renaissance in fluorescence
tracing and chemical identification methods, beginning in the 1990s. Re-

liable detection of multiple markers in small structures, such as fibers and
axon terminals, has been made possible by this instrument (Wouterlood, this
volume).
The hydroxystilbamidine derivative known as Fluoro-Gold (FG; Fluo-
rochrome) was first introduced as a retrograde tracer by Schmued and
Fallon in 1986. This fluorescent dye has a peak excitation at 325 nm, with
a peak emission at 440 nm. FG is widely used in tracing neuronal connec-
tions, not only due to its excellent stable fluorescence but also because of the
simplicity of its application, its exclusive retrograde transport, the lack of up-
take by intact or damaged fibers traversing the injection site (especially when
performing iontophoretic delivery of the dye), and the elaborate filling of
neuronal somata and main dendrites (Akintunde and Buxton, 1992; Divac
and Mogensen, 1990; Novikova et al., 1997; Pieribone and Aston-Jones, 1988;
Schmued, 1994; Schmued and Heimer, 1990; Wessendorf, 1991). Since FG
in itself is not electron dense, it is without histological processing not im-
mediately suitable for ultrastructural visualization. Photoconversion of the
fluorescence (Balercia et al., 1992; Bentivoglio and Su, 1990), or an antibody
against FG, is used to aggregate electron-dense material at loci where FG has
been transported (Chang et al., 1990; Van Bockstaele et al., 1994). Hence,
FG today has the added advantage of being visible both under the fluores-
cence microscope and under the confocal laser-scanning microscope, while
it is compatible with biotinylated dextran amine (BDA; Molecular Probes)
tracing and a host of other immunofluorescence methods. Finally, FG al-
lows downstream processing for electron microscopic visualization (Köbbert
et al., 2000; Lanciego et al., 1997, 1998a, b; Lanciego and Giménez-Amaya,
1999).
Until WGA-HRP tracing became available as a reliable anterograde tracer
(Trojanowski et al., 1981) and was recognized as being compatible with other
tracing methods (Zaborszky et al., 1984), combined anterograde–retrograde
tracing methods hinged on anterograde degeneration as the only effective
tracing method applicable in a common histological laboratory (immuno-
cytochemistry combined with anterograde autoradiographic tracing being
the province of a handful of highly specialized laboratories). Since silver-
degeneration methods offer little compatibility with other light microscopic
tracing methods, combined methods including anterograde tracing had
been developed mainly for ultrastructural research (e.g., Blackstad, 1965,
1981; Somogyi et al., 1979; Wouterlood et al., 1985; Zaborszky and Cullinan,
1989).
F. Phaseolus vulgaris Leucoagglutinin and Biotinylated Dextran Amine
In the mid-1980s, the plant lectin Phaseolus vulgaris leucoagglutinin (PHA-

L) was introduced as an anterograde neuroanatomical tracer (Gerfen and
Sawchenko, 1984; Ter Horst et al., 1984). The unique properties of this
tracer include nearly exclusive anterograde transport and little uptake by
fibers of passage. Furthermore, upon labeling all exquisite details of the
labeled neurons are visible, and the tracer can be visualized in electron mi-
croscopy preparations (Groenewegen and Wouterlood, 1990; Wouterlood
et al., 1990; Wouterlood and Groenewegen, 1985, 1991). Finally, PHA-L
tracing can rather easily be combined with a number of other methods.
Gerfen and Sawchenko (1985) pioneered in this respect by combining PHA-
L tracing with tyrosine hydroxylase (TH) immunofluorescence. Once set,
this example was soon followed by others(Cullinan and Zaborszky, 1991;
Gaykema et al., 1991; Woolf et al., 1986; Wouterlood et al., 1987; Zaborszky
and Cullinan, 1989 ). Innovations were also introduced, for example, by
Antal et al. (1990) who demonstrated that conjugates of PHA-L with biotin
and dinitrophenol are taken up like native PHA-L and transported antero-
gradely. Their work opened the avenue for the study of convergence or
divergence to brain areas of axonal projections originating from different
sources. PHA-L tracing combined with intracellular injection of Lucifer Yel-
low in neurons in sections of fixed brain labeled by retrograde transport of a
fluorescent tracer was described by Wouterlood et al. (1992). Lanciego and
Wouterlood (1994) showed the feasibility of PHA-L tracing in one prepara-
tion together with retrograde tracing, neurotransmitter immunocytochem-
istry, and with anterograde tracing using BDA. Biotinylated compounds and
dextran amines had been introduced in the early 1990s (Fritzsch and Wilm,
1990; Schmued et al., 1990), yet the breakthrough in the form of BDA arrived
with the publication by Veenman et al. in 1992 (see chapter by Reiner and
Honig in this volume). Through the years, BDA has become more popular
than PHA-L as the tracer of choice, mainly because of simpler procedures,
more stable results, and better properties, for instance, if one deals with
electron microscopic issues.
G. Viruses
Viruses offer an instrument for tracing peripheral and central nervous

connectivity based on retrograde transport, multiplication by overtaking
metabolism, trans-synaptic transfer of new virions, transport, and so on.
The unique aspect of using virus as a tracer is its built-in multiplication. In
the chapter by Loewy et al. in this volume, the details of tracing using virus
are fully discussed.
II. PARADIGMS AND SELECTION OF TRACERS

FOR COMBINED USE
A. The Complexity of the Brain Requires Multiple Tracing
Complexity is an inherent feature of brain function. The anatomical study

of complex brain circuits requires the thoughtful design of tracing strategies.
Figure 11.1. The study of the pallidal innervation of different subtypes of nigral
dopaminergic neurons defined on the basis of their projection patterns as a typical
example illustrating a problem to be approached by conducting a multitracer
paradigm. (A) The goal: Analysis of the pallidal afferents to the substantia nigra
pars compacta and their relationship with nigral neurons projecting either to the
caudate nucleus or to the putamen. (B) The experimental paradigm: Injection of
the tracer BDA in the internal division of the globus pallidus (GPi), followed by the
delivery of the tracer CTB in the caudate nucleus and then the injection of FG in the
The scientific paradigm used here to highlight the multiple-tracing ap-
proach is basal ganglia connectivity (Figs. 11.1 and 11.2). Our working hy-
pothesis is that the activity of nigral dopaminergic neurons is modulated by
pallidal outflow. The anatomical substrate for such interference is connec-
tivity between neurons located in the internal part of the globus pallidus
and two subtypes of efferent neurons within the substantia nigra pars com-
pacta, defined on the basis of their different target area within the primate
caudate–putamen. One of the strategies to study these anatomical relation-
ships is the simultaneous application of different neuroanatomical tracers
at different loci in one and the same experiment (Fig. 11.1). Although
single-tracer experiments usually provide valuable data on the organization
of individual projections, they generally provide little information about
potential convergence, divergence, or the degree of overlap between the
circuits involved. A large number of such single-tracer experiments would
be required just to show the topography of the connections, yet provide lit-
tle information about the relationships of the incoming fibers with output
neurons. By contrast, in a single animal the combination of various exper-
iments as one multiple-tracing paradigm dramatically improves the view
where fibers innervating an area end with respect to the precise location
of neurons projecting from that area. Also, combination of several single-
tracing experiments into one multiple-tracer project markedly reduces the
number of required experimental animals. In a society in which ani-
mal experimentation is coming under ever increasing criticism, this is a
premium. Yet, most of all, more and better information with respect to
the relationships between the circuits of interest can be obtained with
multiple-tracing experiments than can ever be achieved with single-tracing
experiments.
←
Figure 11.1. (Cont.) putamen. The delivery of all the injected tracers is performed in a
single surgical session. (C) Tracer uptake and transport: BDA is taken up by dendrites
of neurons located within the area of deposit and transported anterogradely to the
striatum. Both CTB and FG are taken up by axon terminals arborizing within their
respective injection sites and transported retrogradely to their parent cell bodies.
(D) Expected results: Pallidal afferents to the substantia nigra pars compacta are
labeled in blue-black by using DAB-Ni as a chromogen. Nigral neurons projecting
to the caudate nucleus are labeled in brown with DAB substrate. Nigral neurons
innervating the putamen are labeled in purple as a V-VIP precipitate. Injection sites
became clearly defined within their respective targeted areas of deposit. The exper-
imental design included BDA injection into the internal part of the GPi, followed
first by CTB delivery in the ipsilateral caudate nucleus and then by injection of FG
into the putamen. After 2 weeks of survival time, the animal was perfused, the brain
extracted from the skull, and cryoprotected. Frozen coronal sections (40 µm thick)
were obtained and processed for the simultaneous visualization of labeled structures
(according to the step-by-step procedure delineated in the Appendix). At the end,
within our area of study (the primate substantia nigra), labeled structures became
apparent according to a three-color paradigm.
Figure 11.2. Examples illustrating different multitracer paradigms. (A, B) Dual retro-
grade tracing with FG and CTB, combined with BDA anterograde tracing. Photomi-
crographs of coronal sections through the primate substantia nigra. Projections from
the internal part of the globus pallidus innervating the substantia nigra compacta
are labeled with BDA (blue-black) and visualized by using the peroxidase substrate
DAB-Ni. Nigral neurons projecting to the caudate nucleus are retrogradely labeled
with CTB and visualized with DAB chromogen in brown (brown asterisk). Nigral
neurons innervating the putamen are retrogradely labeled with FG and stained in
purple by using V-VIP substrate (purple asterisk). Scale bar is 120 µm in (A) and
35 µm in (B). (C) Analysis of either convergence or divergence innervation arising
from the substantia innominata (SI) or from the pedunculopontine nucleus (PPN)
onto thalamic neurons giving rise to thalamostriatal projections. Photomicrograph
is taken from a coronal section through the rodent thalamus at the level of the in-
tralaminar nuclei. Projections from SI to the paracentral nucleus of the thalamus
are labeled with PHA-L and stained in brown with DAB (brown asterisk). Projections
from PPN to the central lateral nucleus of the thalamus are labeled with BDA and
stained in dark blue-black with DAB-Ni (black asterisk). Neurons giving rise to tha-
lamostriatal projections are retrogradely labeled after an FG deposit in the striatum.
These neurons are stained in dark-light purple color by using V-VIP peroxidase sub-
strate (purple arrowhead). Scale bar: 100 µm. (D) Combination of two tracers (BDA
and FG stained with DAB-Ni and V-VIP, respectively) together with the immuno-
cytochemical visualization of striatal giant cholinergic interneurons (brown-stained
with DAB). Scale bar: 40 µm. (E) Combination of two tracers (BDA and FG stained
with DAB and V-VIP, respectively) together with the histochemical visualization of
NADPH diaphorase striatal interneurons (blue-stained with nitroblue tetrazolium).
Scale bar: 45 µm.
Table 11.1. Best tracers to be selected for their combined use.
Tracer Anterograde Retrograde Comments
Cholera toxin β subunit + +++ Second-choice retrograde tracer
Commercial antibodies available
Bidirectional transport1
Pressure/iontophoretical delivery
Survival time from 4 days to 4 weeks
Taken up by fibers of passage
PHA-L +++ − Second-choice anterograde tracer
Commercial antibodies available
Capricious nature
Iontophoretical delivery
Survival time from 7 days to 3 months
Lack of transport by fibers of passage
Fluoro-gold − +++ First-choice retrograde tracer
Commercial antibody available
Direct visualization by epifluorescence
Survival time from 4 days to 1 year
Taken up by fibers of passage2
Biotinylated dextran +++ + First-choice anterograde tracer
amine No antibodies involved in visualization
Survival time from 4 days to 2 months3
Well suited to ultrastructural study
Taken up by fibers of passage4
1 A procedure improving the anterograde transport of CTB was reported by Angelucci et al.
(1996).
2 It is generally accepted that the iontophoretical delivery of FG minimizes the uptake by fibers
of passage (Divac and Mogensen, 1990; Pieribone and Aston-Jones, 1988; Schmued and Fallon,
1986; Schmued and Heimer, 1990).
3 Longer survival times were never tested by us. For more information, please see chapter by
Reiner and Honig in this volume.

4 See chapter by Reiner and Honig in this volume.
B. “Golden Rules”: Requirements for a Tracer to Be Successful in

a Multitracing Paradigm
Currently, we have at our disposal a wide variety of different tracers (Table

11.1), a huge array of antibodies to detect these tracers, and several possibil-
ities to visualize them. Hence, in the process of designing multiple-tracing
protocols the first question the investigator faces is that of the selection
of the proper combination of tracers, i.e., which tracers are best suited for
combined use, which order to maintain to successfully inject them, and how
to inject them. In order to answer these difficult questions, one should keep
in mind that the selected combination has to fulfill the following demands:
1. The first “golden rule” for multiple tracing is that the tracers involved
should preferably be transported strictly unidirectionally (either an-
terogradely or retrogradely).
2. Only tracers of a similar nature can be combined together, i.e., nonfluo-

rescent tracers cannot be combined with fluorescent tracers unless the
latter can be appropriately detected in conventional light microscopy,
for instance, after photoconversion or by detection via specific primary
antibodies.
3. The tracers must have compatible survival times in order to avoid repet-
itive surgical procedures, particularly given the increasing criticism of
these procedures by ethical committees. In this regard, tracers such
as HRP or biocytin are quickly transported and require only 1 or 2
days of survival time. However, such short range of survival times often
impedes combination of these tracers with other commonly used trac-
ers such as PHA-L, BDA, CTB, or FG, all of which require or tolerate
longer survival times, from a few days to several months.
4. Commercially specific antibodies for the immunodetection of each
individual tracer should be available, and these antibodies should not
cross-react with each other or in any way interfere with the detection
of any of the other markers under study.
5. It is desirable that the detection of the combined tracers can be per-
formed by both immunoperoxidase and immunofluorescence meth-
ods. When multiple immunoperoxidase detection is to be used, the
individual chromogens should produce colored precipitates that can
be unequivocally discriminated from each other (Table 11.2). Further-
more, the resulting precipitates should preferably be permanent and
compatible with existing cytoarchitectonic staining procedures. If im-
munofluorescence is the chosen method to detect multiple tracers, dif-
ferent fluorescence-coupled bridging antibodies should be available,
each being labeled with a fluorescent dye with characteristic excitation
and emission spectra clearly distinct from those of the other antibodies.
Nevertheless, if this rule cannot be properly satisfied, modern confo-
cal laser scanning microscopes are equipped with spectral detectors
capable of distinguishing emission profiles overlapping up to 100%
(one emission profile located within the other).
6. Finally, ultrastructural examination of labeled structures is always a
desirable option, and thus all peroxidase substrates used to visualize
tracers should be identifiable in the electron microscope by their own
ultrastructural, electron-dense texture.
C. Winning Tracers
Although nearly all currently used tracers fulfill at least some of these de-
manding criteria, some of these tracers more closely approach the ideal than
others. According to our experience, BDA is undoubtedly the best tracer
for anterograde tracing in multiple-labeling experiments. This is mainly due
to its broad spectrum of survival time (from a few days to several weeks),
its mostly anterograde transport especially when injected by iontophoresis,
Table 11.2. Commonly used peroxidase substrates.
Substrate Color Comments
Tetramethylbenzidine Dark blue Own electron-dense precipitate
Difficult to combine with other chromogens
Presumed carcinogenic
Diaminobenzidine Brown Incubation time of 10–40 min
Own electron-dense texture
Compatible with DAB-Ni, BHDC, V-VIP, and HGR
Nickel-enhanced DAB Black Strongest chromogen
Incubation time of 5–10 min
Compatible with DAB, BHDC, V-VIP, and HGR
Benzidine dihydrochloride Blue Own electron-dense texture
Compatible with DAB-Ni and DAB
Vector very intense purple Purple Incubation time of 3–5 min
Own electron-dense texture
Presumed noncarcinogenic
Soluble in ethanol
1-Naphthol/azur B Blue-green Not electron dense
Fading over time
HistoGreen Green Incubation time of 1–5 min
Presumed noncarcinogenic
Soluble in ethanol
So far, electron-dense texture not reported
Note. DAB, diaminobenzidine; BDHC, benzidine dihydrochloride; V-VIP, vector very intense
purple; DAB-Ni, nickel-enhanced diaminobenzidine; HGR, HistoGreen.
straightforward histochemical detection using the avidin–biotin peroxidase

(ABC) method without the need to use antibodies (this in turn simplifies
the choice of antibodies to be used to detect the other markers), the avail-
ability of a broad range of fluorochrome-labeled streptavidin conjugates for
the observation of transported BDA by confocal laser scanning, and suffi-
cient tissue penetration of the reagents when used in ultrastructural analysis
(Reiner and Honig, 2005, this volume; Wouterlood and Jorritsma-Byham,
1993). We consider PHA-L as the second option for anterograde tracing
in combined paradigms. This tracer also offers a broad range of survival
times (up to several months), and the axonal transport of this tracer is
verified to be only in the anterograde direction. Both immunocytochemical
and immunofluorescence detections have been used, and a good electron
microscopy correlate is available. Nevertheless, many investigators report
variable results using this tracer, and when the purpose is to conduct elec-
tron microscopy, the limited penetration of the anti-PHA-L antibodies into
tissue sections may offer a histotechnical challenge (Wouterlood et al., 1990).
For retrograde tracing, we feel that FG is undoubtedly the most efficient

choice. The feasibility of using FG in multiple-tracer paradigms is impres-
sive, given the advantages offered by several of its properties. The tracer can
be delivered either by pressure or via iontophoresis, it has a broad range
of survival times, and there is no anterograde transport. FG exquisitely fills
the neuronal soma and main dendrites, and uptake by fibers of passage is
minimal. Moreover, direct detection of transported FG is possible via fluores-
cence microscopy whereas indirect immunohistochemical or immunofluo-
rescence detection can be combined with protocols for neurotransmitter
immunodetection or other types of detection. FG can in addition be used
in ultrastructural analysis. Our second choice for retrograde tracing is CTB,
which shares several properties with FG, such as the wide range and flexi-
bility of survival times, different detection methods (immunocytochemistry
or immunofluorescence), and ultrastructural visualization. Although CTB
is nicely transported in the retrograde direction, this tracer has several dis-
advantages when compared to FG, such as (1) larger tracer deposits are
often required, (2) a moderate amount of anterograde transport is always
present, (3) the transported tracer is often only located within the cell soma
as a punctate deposit that does not enter the main dendrites, and (4) in our
hands, the best results are obtained only with CTB applied through pressure
injections. However, such injections always cover larger brain volumes than
covered by iontophoretic injections.
D. Winning Combinations of Tracers
In the design of a multitracer paradigm centered around the purpose of

defining the degree of convergence between projections arising from two
different brain regions to the efferent neurons in another area, our cur-
rently favored choice is to combine dual anterograde tracing using BDA
and PHA-L together with retrograde tract-tracing using FG (for more de-
tails, see Lanciego et al., 1998a; Fig. 11.2C). When the purpose of the de-
sign is to investigate whether two distinct subpopulations of efferent neu-
rons located within a brain area are targeted by afferents arising from a
different, third area, we feel that it is best to combine dual retrograde
tracing with FG and CTB, together with anterograde tracing using BDA
(Lanciego et al., 1998b, 2000; see also Fig. 11.2A,B). Additionally, several ef-
ficient triple-staining procedures that combine dual tracing using BDA and
FG with the simultaneous detection of neuroactive substances are available
(Gonzalo et al., 2001; Köbbert et al., 2000; Lanciego et al., 1997, 2000; see
also Fig. 11.2C, D). In our experiments, we ultimately selected BDA as the
anterograde tracer of choice for injecting the internal part of the globus
pallidus (Lanciego et al., 2004; Fig. 11.2A,B), and FG and CTB as retrograde
tracers. These two tracers were injected into the putamen and the caudate
nucleus, respectively.
III. EVENTS AT THE INJECTION SITE
The injection site deserves special attention: how big is the injection spot,
what is the effective injection spot, and do all neurons that lie within the
demarcation of the injection spot internalize and transport the tracer?
The procedure by which a tracer is deposited at its desired locus is as
follows:
1. opening of the skull and meninges, positioning of the tip of the pipette
exactly on top of the pial surface of the brain, and reading its Z -
coordinate;
2. lowering of the pipette to its final spatial position in the brain;
3. ejection of tracer substance from the pipette tip; and
4. retraction and subsequent closure of the wound.
Especially in Stage 3, tracer substance is ejected forcefully out of the mi-
cropipette or the injection needle, and it may take the way of the least
mechanical resistance, i.e., some tracer may leak into the space between the
pipette and the brain parenchyma, labeling cells here.
A. Uptake Mechanisms
Following its ejection from a pipette or a syringe, the tracer needs to pass
the cell membrane in order to be transported by the neuron. The uptake
mechanism of several tracers has been studied while the uptake of other
tracers yet needs to be elucidated. There is evidence that glycoprotein trac-
ers like HRP are taken up via a fluid-phase pinocytotic process (Gonatas
et al., 1979). The lectin conjugate of HRP binds to receptors on the exter-
nal faces of cell membranes, thus enhancing the uptake (Trojanowski and
Schmidt, 1984). PHA-L and BDA undergo receptor-mediated endocytosis
(Fritzsch, 1993; Groenewegen and Wouterlood, 1990). By contrast, (electri-
cally uncharged) FG may pass the cell membrane by simple diffusion and
then become trapped via a pH gradient in endosomes (Wessendorf, 1991).
The endosomes in turn are transported to the lysosome apparatus in the
cell body in an attempt to degrade the ingested FG metabolically. A debate
has been raging in the literature for the last 10 years as to whether cholera
toxin enters cells via small pinocytotic-type fluid-phase vesicular carriers
known as “caveolae” (Anderson and Edwards, 1993) or via clathrin-mediated
endocytotic vesicles (Kirkham et al., 2005). Whether the attenuated form of
the toxin (i.e., CTB) uses either one internalization pathway is unknown.
B. BDA May Cause Track Labeling
During the insertion and retraction stages of the stereotaxic pipette or

needle, tracer leakage may occur. BDA track labeling or uptake of the tracer
by neurons located on the trajectory of the pipette when it was lowered to its
final stereotaxic position in the brain or when it was retracted afterward have
been reported by Dolleman-van der Weel et al. (1995). In our experience,
track labeling can be suppressed by carefully cleaning the outer surface
of the pipette before placement. Only when the positioning is successful
and verified is the iontophoretic current applied to force the tracer out of
the pipette. After iontophoresis, we leave the pipette in place for 15 min
and then slowly retract, giving the brain parenchyma the chance to seal the
injection area. An effective additional measure to suppress track labeling is
to reverse the polarity of the iontophoretic current during the retraction of
the pipette. If in spite of these measures track labeling is still a problem, the
pipette or needle might be inserted into the brain at an angle so as to avoid
track labeling vulnerable areas.
C. The Effective Injection Site Shape, Size, and Delivery of Tracer
Injection sites usually have a teardrop-like shape, with the diameter of

the “teardrop” measured perpendicular to the axis of penetration of the
application pipette or the needle. In all cases, the effective injection area
is surrounded by an injection halo. The size and shape of an injection site
depend in part on the cytoarchitectonics of the injection region, e.g., the
presence of layers or bundles of myelinated fibers. In the case of BDA, diam-
eters of the injection sites in the striatum (at injection time) vary between
200 and 300 µm (iontophoretical application with a pipette with an inner
tip diameter of 20–30 µm) and up to 800 µm (single, large, pressure injec-
tion). Another issue that is directly related to the size of the injection site
is the duration of the iontophoretical injection. As a rule of thumb, longer
injection times result in larger deposits.
The effective site of injection of an anterograde tracer, e.g. BDA, is pos-
tulated here as that brain area located at the original injection coordinates
that contains an aggregation of cells that stain above background with the
detection method. Since this injection site is the spot visible after survival
and histochemical staining, the border of this injection spot may not re-
flect the extent of diffusion of tracer away from the tip of the application
pipette at or shortly after injection time. Yet, since we are appreciating la-
beling of fibers and terminals, the cell bodies to which these fibers belong
must be labeled as well. There is little reason to assume that nonlabeled
cell bodies give rise to labeled projections and, conversely, that label accu-
mulates during the survival time in the distal axons and axon terminals,
with no labeled cell bodies remaining in the original injection spots. In
a longitudinal study of the fate of the tracer PHA-L, quite the contrary
was seen. Label progressively disappeared from axon terminals, next from
fibers, and finally from the cell bodies in injection sites (Wouterlood et al.,
1990).
D. PHA-L
One characteristic of PHA-L is that with pipette tip diameters larger than
30 µm there is no ejection of tracer when the iontophoretic current is ap-
plied. Consequently, the size of the injection site is always limited to a colum-
nar or a teardrop-like space with a diameter of approximately 200 µm. If
larger injections are required, for instance when the purpose is to cover an
entire brain area, then multiple PHA-L injections have to be made in an
array-like manner.
E. Cholera Toxin Subunit B
CTB can be delivered either by iontophoresis or by pressure. We prefer in

our laboratory the pressure delivery method: 2% solution of CTB in sterile
phosphate buffer (PB). Injection volume is typically in the range of 0.1–
0.2 µl. The balance between the anterograde and the retrograde transport
of the tracer is apparently related to the survival time. Short survival times
(2–4 days) tend to increase the anterograde transport component, while
survival times longer than a week have a tendency to enhance the retrograde
transport component. The wide range of survival times for CTB, together
with its feasibility for being delivered by using different methods, make
CTB a very useful tracer for combined procedures as well as for coinjection
protocols (see below).
F. Fluoro-Gold
In the last few years FG has become the tracer of choice for retrograde trac-
ing purposes. This tracer is delivered either by pressure or by iontophoresis
(if smaller injections are required). FG has the widest survival time spec-
trum, ranging from just a few days to 1 year (Akintunde and Buxton, 1992;
Novikova et al., 1997; Pieribone and Aston-Jones, 1988; Schmued, 1994;
Schmued and Fallon, 1986; Schmued and Heimer, 1990). According to our
experience, the most critical issue when dealing with FG tracing is the vehi-
cle in which the tracer is dissolved. Best results using FG have been obtained
by dissolving the tracer as a 2–3% solution in 0.1 M cacodylate buffer at neu-
tral pH, as previously reported elsewhere (Schmued and Heimer, 1990).
If the vehicle is not cacodylate buffer (e.g., distilled water, PBs, acetate
buffer), then a nonhomogeneous solution easily forms that may precipitate
and seriously compromise axonal uptake. In this regard, the use of FG for
tract-tracing studies in primates is still somewhat controversial. It has been
reported that FG may produce variable results when used in nonhuman pri-
mates (Schmued, 1994). Taking advantage of the superior solubility of FG
in cacodylate buffer, we obtained satisfying retrograde labeling of both the
nigrostriatal and the thalamostriatal pathways after pressure-injecting FG

into the putamen of the primate Macaca fascicularis (Lanciego et al., 1998b;
see also Fig. 11.2A,B).
G. Survival Time
The survival time has to be adapted to the length of the pathways under
scrutiny. According to the “golden rules” for multiple-tracing paradigms, the
range of survival times of all the tracers involved should be sufficiently broad
to facilitate their combined use. Hence, when carrying out experiments in
rats, a survival time of 1 week is sufficient to produce high-quality staining of
all the injected tracers. In our hands, BDA yields good results after survival
times of up to 8 weeks in rodents (we have not tested longer survival times).
In primates, good results were obtained over periods of 2 weeks, although
Brandt and Apkarian (1992) have reported good BDA labeling in primates
after survival times of up to 7 weeks.
H. Deposition of an Anterograde and a Retrograde

Tracer in the Same Spot
If reciprocally connected networks are to be studied, it would be advan-

tageous to use tract-tracing paradigms including the combined and simul-
taneous delivery of two different tracers (anterograde and retrograde) for
disclosing the presence of reciprocal connections to a given brain area or nu-
cleus. Two strategies may be followed to achieve coinjection: mixing tracers
and injecting via two different pipettes or needles.
The injection of a mixture of FG and fluorescence-labeled dextrans re-
ported by Nance and Burns (1990) was the first attempt to study reciprocal
connections with one single injection (injection achieved by pressure). The
only limitation of this elegant approach was the fading of signal over time,
due to the fluorescent nature of both tracers. Later on, a different solution
was provided by Risold and Swanson (1995) by coinjecting FG and PHA-L.
In this approach, coinjections were achieved by using double-barreled glass
pipettes. Since distally the two barrels had different tip diameters, the sizes
of the injection sites were not comparable to each other. Probably the best
circumvention of these drawbacks was provided by Coolen and Wood (1998)
and by Coolen et al. (1999), by combining BDA and CTB. Both tracers share
many features such as wide range of survival time, dilution parameters, etc.
Coolen et al. (1999) coinjected BDA and CTB by using one pipette, which
was lowered to the desired Z -coordinate. Ejection of the tracers from the
pipette was achieved by pressure and iontophoresis (1:1 mixture of BDA
and CTB, final concentration of 5 and 0.25%, respectively). This approach
resulted in perfectly overlapping injection sites and in the availability of per-
manent detection through immunoperoxidase techniques. In our opinion
this is the best choice when studying brain circuits in which the involved
components exhibit reciprocal connections.
IV. VISUALIZATION METHODS: MULTIPLE

IMMUNOPEROXIDASE FOR BRIGHT FIELD MICROSCOPY
It has long been a desire in modern neuroanatomy to be able to simul-

taneously and permanently visualize up to three transported markers in a
single histological section in order to produce detailed mapping of brain
circuits. The first attempts to achieve this were undertaken long ago, before
the advent of strong fluorescent markers and fluorescent counterstains, and
the natural approach was to combine these methods with classical counter-
staining to determine the cytoarchitectonic context of the traced connectiv-
ity. Although technically extremely demanding, different approaches have
been proposed and the possibilities seem almost unlimited, restricted only
by the researcher’s own imagination and technical skills.
The first successful attempts to simultaneously detect three different trac-
ers were reported by Smith and Bolam (1991, 1992) using a protocol that
combined retrograde transport of WGA-HRP with anterograde transport of
PHA-L and biocytin. The tracers were detected with three different peroxi-
dase substrates, namely tetramethylbenzidine (TMB) to visualize the trans-
ported WGA-HRP, nickel-enhanced DAB (DAB-Ni) to detect PHA-L, and
finally, regular DAB to visualize biocytin. In its day, this technical tour de
force was an imaginative solution, although the clear discrimination of the
labeled elements was sometimes a difficult task. Subsequently, Dolleman-
van der Weel et al. (1995) introduced a sequential light microscopy proto-
col combining anterograde tracing with PHA-L, rhodamine dextran amine
(RDA), and BDA, stained respectively with a DAB-Ni (black precipitate),
DAB (brown reaction product), and 1-naphthol/azur B (blue-green color)
as the peroxidase substrates. Using this procedure, the labeled elements
could be unequivocally distinguished, except in brain areas where the three
tracers concurred “en massive.” The only, although minor, drawback of this
delicate procedure is that fading of 1-naphthol/azur B precipitate was no-
ticed at room temperature, although the stain could be restored by repeating
the incubation with the basic dye (Mauro et al., 1985).
When trying to perform ultrastructural studies using three markers,
the work published by Anderson et al. (1994) must be seen as a jewel
in the tract-tracing literature. These authors designed an elegant preem-
bedding triple-labeling method to clearly visualize the convergence of sub-
stance P- (SP-ir) and enkephalin-positive (ENK-ir) afferents onto tegmen-
tal dopaminergic neurons, the latter identified with a primary antibody
against TH-ir. The antigens of interest were visualized with three distinct
peroxidase substrates, resulting in three morphologically different electron-
dense textures, namely DAB (flocculent) for the SP-ir terminals, silver-
intensified immunogold (distinct particles) for the stain of ENK-ir terminals,
and benzidine hydrochloride (needle-shaped crystalline material) for TH

detection.
In general, in the process of the design of a triple-staining paradigm,
a main limitation run into is the relatively small number of peroxidase
substrates currently available (Table 11.2). This limitation does not always
permit a straightforward color discrimination of the labeled elements. Fur-
thermore, several peroxidase substrates used in light microscopical studies
do not provide precipitates with a texture that can be easily differentiated
in the electron microscope from “regular” DAB precipitate. In this regard,
the introduction by Zhou and Grofova (1995) of the peroxidase substrate
Vectorr
very intense purple (V-VIP) must be considered as an important
advance in multiple-staining strategies. The peroxidase substrate V-VIP has
two unique properties, notably a characteristic purple color that under the
light microscope is distinguishable with ease from other commonly used
chromogens (Fig. 11.2), and an electron-dense precipitate with a typical
dense granular appearance. Initially developed as an alternative to DAB
for double-labeling procedures, this new peroxidase substrate was rapidly
incorporated into multiple-staining paradigms aimed at simultaneously de-
tecting three different markers, specifically in those protocols in which three
chromogens were combined, i.e., DAB-Ni (black), DAB (brown), and V-VIP
(purple; Lanciego et al., 1997). The recent introduction of a new peroxidase
substrate named HistoGreen (Thomas and Lemmer, 2005) has enriched the
technical arsenal at our disposal for designing multiple colored detection
of immunoperoxidase protocols.
A. Troubleshooting
When performing triple-staining procedures, the researcher is faced with

three main problems that can reduce the quality of the final stain. The
critical issues are (i) the choice of the most appropriate chromogen for
each particular marker, (ii) background staining, and (iii) color mixing
phenomena.
1. Chromogen Selection and Chromogen Sequence
We are dealing with three different peroxidase substrates (DAB-Ni, DAB,

and V-VIP), each characterized by a different strength. The strongest
chromogen is DAB-Ni, while V-VIP is the weakest. If we consider that the
detection of BDA does not depend on antibodies, DAB-Ni is probably the
best choice to visualize the BDA stain since this tracer stain fibers nicely with
minimal background staining. As both DAB and V-VIP chromogens always
produce some light background, we consider these two substrates as the
most appropriate for cellular stains. Accordingly, we choose to use DAB as
the substrate for CTB stain and V-VIP for the detection of FG.
Another important issue is the sequence in which the chromogens are
applied during the staining procedure. By virtue of their differences in
strength, the best results are obtained by using the strongest substrate first
(DAB-Ni), then the DAB solution, while the weakest chromogen (V-VIP)
should be used last. Different sequences were tested during the develop-
ment of the three-color paradigm, always resulting in the appearance of
intolerable background staining and color mixing phenomena.
2. Background Staining
Avoiding or suppressing background stain is critically important when

attempting to obtain high-quality staining since background stain may se-
riously hamper the contrast between the different color reaction products
of the chromogens. The appearance of an intolerable degree of nonspe-
cific background staining often hinders the interpretation of the results
since it may prevent the correct visualization of the finer structures la-
beled with a particular antigen (Groenewegen and Wouterlood, 1990). It
is important to achieve a low level of background staining in triple-staining
methods. Background staining is often seen when one antigen requires the
prolonged incubation of its corresponding chromogen solution to obtain
adequate staining. To avoid this pitfall, we recommend careful and contin-
uous monitoring under the microscope of the progress of each reaction.
In this way, the incubations with the different peroxidase substrates can
be stopped in time before nonspecific background starts to build up. If
such care is not taken, the background stain obtained in the first reaction
will enhance the background stain obtained in the subsequent chromogen
incubations. Nonspecific binding of the antibodies may also increase back-
ground stain, a drawback that can be minimized by adding 2% bovine serum
albumin (BSA; Merck) to the incubation solution containing the primary
antibodies.
3. Color Mixing
The tendency of a peroxidase substrate to adopt the color of other chro-

mogens used previously is an undesired phenomenon that we refer to as
“color mixing.” Color mixing is often observed in circumstances where
(i) the reaction product is at a high concentration, such as at the injec-
tion sites, (ii) cross-reactivity occurs between the antisera, (iii) prolonged
incubations of the chromogen are necessary, and (iv) two or three of the
antigens detected colocalize within the same structure. The best way to
avoid color mixing consists of carefully selecting the antisera, preferably
using antibodies raised in different animal species. Furthermore, monitor-
ing the progress of the chromogen reactions at intervals by microscopy is
indispensable. When the antigens under study colocalize within the same
structure but are found in separate subcellular compartments, colocaliza-

tion is not generally a major problem (Lanciego and Giménez-Amaya, 1999).
B. Results
The reward of successful completion of a multitracing experiment is the

full display in multiple colors of the complexity of neuronal connectivity.
This is illustrated in Fig. 11.2, which is composed of photomicrographs of
sections of monkey brain in which the specific triple-tracing paradigm in-
cluded injection of BDA into the internal part of the GPi, CTB delivery in
the caudate nucleus, and the FG deposit in the putamen. Thus, in our sec-
tions, we observed in the substantia nigra pars compacta a subpopulation of
neurons retrogradely labeled with CTB (cell bodies containing brown pre-
cipitate). CTB-labeled neurons were distributed within cell clusters mainly
composed of cells with similar projection patterns. Indeed, nigrocaudate-
projecting neurons remained largely complementary to another cellular
group stained in purple with V-VIP precipitate. These purple-stained cells
are nigroputaminal neurons retrogradely labeled with FG. The cell bodies
and dendrites of both types of retrogradely labeled nigrostriatal neurons
were embedded in a dense terminal plexus of BDA-labeled fibers arising
from GPi (stained black with DAB-Ni substrate). Cell bodies of both types
of nigrostriatal retrogradely labeled neurons appeared within the terminal
plexus of pallidonigral projections, indicating that both the nigrocaudate
and the nigroputaminal projecting neurons receive pallidal innervation.
The secondary observation of the complementary distribution of the dif-
ferent subtypes of nigrostriatal neurons and the distribution of pallidal af-
ferents over both populations of projection neurons is the added value of
triple-tracing paradigms.
APPENDIX
In order to facilitate the use of this method by others, we provide a de-

tailed description below that deals with the simultaneous detection of the
anterograde transport of BDA in combination with the retrograde transport
of CTB and FG. Only minor adjustments mainly related to the selection of
antibodies are required when detecting different combinations of antigens.
A. Step-By-Step Procedure
1. Tracer Delivery, Survival Time, Perfusion, Cryoprotection,

and Sectioning
All injected tracers are delivered in a single surgical session. The antero-
grade tracer BDA is iontophoretically delivered as a 10% solution in 10 mM
PB, pH 7.25 (see section “Preparation of the BDA Solution for Injection”)
through a glass micropipette (inner tip diameter ranging from 20 to 35 µm),
using a positive-pulsed direct current (7 s on/off) for 3–10 min (depend-
ing on the desired size of the injection site). Next, a 2% solution of FG in
100 mM cacodylate buffer, pH 7.3, is injected using the same iontophoretic
parameters described above for BDA delivery. Finally, a total volume of 0.5–
5 µl of a 2% solution of CTB in 0.1 PB, pH 6.0, is pressure-injected into the
target area at a rate of 0.2 µl/min. For studies in primates, both BDA and
FG were also pressure-delivered.
Regarding tract-tracing with CTB, survival times of 1 week for rodents
and 2 weeks when dealing with primates always result in satisfactory bidirec-
tional labeling. It has been proposed that the anterograde component of
CTB transport is intensified with short survival times, while longer periods
may enhance its retrograde transport (Angelucci et al., 1996; Trojanowski
et al., 1981). However, in our hands, no significant variability of the antero-
grade/retrograde transport ratio of CTB was noticed when the survival times
ranged from 2 days to 4 weeks.
Tissue fixation is always performed by trans-cardiac perfusion after deeply
anesthetizing the animal with an overdose of 10% of chloral hydrate in
distilled water. Once anesthetized, a saline Ringer’s solution is used to
perform a brief vascular rinse. Immediately afterward, the perfusion is con-
tinued with either 500 ml (rats) or 3000 ml (primates) of a solution con-
taining 4% freshly depolymerized paraformaldehyde, 0.1% glutaraldehyde,
and 0.2% of saturated picric acid solution in 125 mM PB, neutral pH. In
the case of primates only, perfusion is continued with 1000 ml of a cry-
oprotective solution consisting of 10% glycerin and 1% dimethylsulfoxide
(DMSO) in 125 mM PB, neutral pH. The brain is then extracted from the
skull and stored in a cryoprotective solution composed of 20% glycerin and
2% DMSO in 125 mM PB, neutral pH (Rosene et al., 1986). Finally, frozen
tissue sections are obtained at a thickness of 40 µm on a sliding microtome
and collected in 125 mM PB, neutral pH. Each series of sections is trans-
ferred to cryoprotectant and stored in a freezer at −40˚C until they undergo
further histological processing. In section “Cryoprotective Solutions” (see
below) some advise regarding cryoprotection is offered.
2. Histological Processing
All procedures are carried out at room temperature, with gentle shaking
on an orbital rotator, unless otherwise stated.
1. Inactivate the endogenous peroxidase activity, for 40 min (see section
“Inactivation of the Endogenous Peroxidase Activity”).
2. Rinse 3 × 10 min in 50 mM Tris-buffered saline–Triton X (TBS-Tx),
pH 8.
3. Incubate in ABC solution for 90 min (see section “Preparation of the
ABC Solution”).
4. Rinse 2 × 10 min in 50 mM TBS-Tx, pH 8.

5. Rinse 2 × 10 min in 50 mM Tris-HCl, pH 8.
6. Stain with the first chromogen solution (DAB-Ni) for ∼5 min (see
section “Preparation of the DAB-Ni Solution”).
8. Incubate for 60 h at 4˚C in a cocktail solution of primary antibodies
containing 1:2000 goat anti-CTB (List Biological) and 1:2000 rabbit
anti-FG (Chemicon) in 50 mM TBS-Tx, pH 8 + 2% BSA.
10. Incubate for 2 h in a cocktail solution of bridge antibodies, containing
1:50 donkey anti-goat IgG (Nordic or Sigma) and 1:50 swine anti-
rabbit IgG (Dako) in 50 mM TBS-Tx, pH 8.
12. Incubate for 90 min in 1:600 goat-PAP (Nordic or Sigma) in 50 mM
TBS-Tx, pH 8.
15. Rinse 2 × 10 min in 50 mM Tris-HCl, pH 7.6.
16. Stain with the second chromogen solution (DAB) for approximately
10–30 min (see section “Preparation of the DAB Solution”).
20. Incubate for 90 min in 1:600 rabbit-PAP (Dako) in 50 mM TBS-Tx,
pH 8.
22. Rinse 1 × 10 min in 50 mM Tris/HCl, pH 8.
23. Rinse 2 × 10 min in 50 mM Tris/HCl, pH 7.6.
24. Stain with the third chromogen solution (V-VIP) for approximately
3–5 min (see section “Preparation of the V-VIP Solution”).
26. Mount sections in 2% gelatin (Merck) in 50 mM Tris-HCl, pH 7.6.
27. Once dried thoroughly, quickly dehydrate the sections with 2 ×
10 min rinses in 100% toluene (see section “On the Use of Toluene
for Dehydration Purposes”) and mount in DPX, Entellan, or similar
mounting media.
B. Technical Tips
1. Preparation of the BDA Solution for Injection
BDA (biotin, dextran 10 KD, lysine fixable, catalog number D1956) is

purchased from Molecular Probes Europe (Leiden, The Netherlands). For
either iontophoretic or pressure injections, the tracer is dissolved as 2–15%
in 10 mM PB, pH 7.25 (Reiner et al., 2000). Good results with BDA have also
been reported by pressure-injecting either a 10% solution in saline (Brandt
and Apkarian, 1992) or a 5% solution in distilled water (Rajakumar et al.,
1993).
2. Cryoprotective Solutions
Cryoprotection is achieved using a standard solution modified according

to Rosene et al. (1986). This cryoprotectant solution consists of 20% glycerin
and 2% DMSO in 125 mM PB at neutral pH. With this solution cryoprotec-
tion of a complete rat brain usually takes overnight at 4˚C, or that for cat and
monkey brain takes approximately 48 h at 4˚C. Subsequently, it is possible
to cut large brain sections without introducing freezing artifacts. In addi-
tion, this cryoprotective solution helps to preserve the ultrastructural details
fairly well (Fig. 11.2). Another advantage when compared to other com-
monly used cryoprotectants, such as 30% sucrose in PB, is that the buffered
combination of glycerin and DMSO produces a minimal amount of shrink-
age. Finally, in our experience frozen tissue immersed in this cryoprotectant
solution can be stored for up to 7 years at −85˚C without showing any signif-
icant loss of immunoreactivity when compared to freshly processed tissue
from the same animal.
3. Inactivation of the Endogenous Peroxidase Activity
Prior to immunocytochemical staining, we always inactivate the endoge-

nous peroxidase activity in the tissue to minimize background staining. This
procedure is especially advisable in cases of improperly perfused brains.
The inactivation solution is a mixture of methanol and hydrogen peroxide
(0.1 ml of H2 O2 + 0.9 ml of H2 O in 15 ml of 100% methanol). Sections
are incubated in the inactivation solution at a ratio of 1 min/µm section
thickness (i.e., 40 min when dealing with 40-µm-thick sections).
4. Preparation of the ABC Solution
The ABC complex is supplied by Vector Labs (Burlingame, CA), cata-

log number Standard PK-4000. No differences in the final staining quality
were observed when using the Standard ABC PK-4000 instead of the Vec-
tor ABC Elite kit. The ABC solution should be freshly prepared following
the manufacturer’s instructions. To prepare the incubation solution, add 8
µl of solution A to 1 ml of 50 mM TBS-Tx, pH 8, and shake gently. Then
add 8 µl of solution B, shake gently, and wait for 30 min before using the
solution.
5. Preparation of the DAB-Ni Solution
The DAB-Ni solution should be freshly prepared by dissolving 0.2 g of

nickel ammonium sulfate (Carlo Erba) and 7.5 mg of DAB in 50 ml of
50 mM Tris/HCl, pH 8. Immediately before use, 10 µl of hydrogen perox-
ide must be added. Incubation usually takes 5–10 min, resulting in a black
precipitate.
6. Preparation of the DAB Solution
The DAB solution is freshly prepared by dissolving 5 mg of DAB in 10 ml of

50 mM Tris-HCl, pH 7.6. Once dissolved, the resulting solution is filtered and
3.3 µl of hydrogen peroxide is added immediately prior to use. Incubation
takes 20–40 min, resulting in a brown precipitate.
7. Preparation of the V-VIP Solution
The V-VIP solution is freshly prepared in 50 mM Tris-HCl, pH 7.6, with

one drop from each vial (V-VIP substrate is presented as a kit containing
four vials) to prepare 3.5 ml of the working solution. Incubation takes 5–
10 min, resulting in a purple precipitate. Note that the final working so-
lution of V-VIP presented here is a twofold dilution of the original recipe
provided by the supplier. The use of a more diluted substrate results in a
more manageable chromogen, making it easier to monitor the progress of
the reaction, and therefore one can avoid typical drawbacks such as excessive
background staining or color mixing phenomena.
8. On the Use of Toluene for Dehydration Purposes
V-VIP chromogen is soluble in ethanol. Accordingly, when using standard

dehydration procedures in ascending series of ethanol followed by clearing
in xylene, V-VIP stain may disappear or be diluted. To circumvent this prob-
lem, we perform the dehydration and clearing procedures by placing the
slides with the thoroughly dried sections directly in 100% toluene (two rinses
of 5 min each), which yields the best results.
Acknowledgments. Most of the technical training received by José L.

Lanciego was conducted under the expert guidance of his mentors, Prof.
Francisco Collı́a at the University of Salamanca and Prof. Floris G. Wouter-
lood at the Amsterdam Vrije Universiteit Medical Center.
The data presented here are gathered from the work conducted by José
L. Lanciego and his collaborators. Among others, we would like to express
our gratitude to Drs. Elena Erro, José Arribas, Nancy Gonzalo, and Marı́a
Castle. We also acknowledge the expert assistance provided by my techni-
cians Ms. Elvira Roda and Ms. Ainhoa Moreno.
This work was partially funded by research grants from The Michael
J. Fox Foundation, Fondo de Investigaciones Sanitarias ref. FIS 01/0237,
Ministerio de Ciencia y Tecnologı́a ref. BFI2003-02033, FEDER, Departa-
mentos de Salud y Educación del Gobierno de Navarra, and by the “UTE
Project FIMA.”
REFERENCES
Akintunde, A., and Buxton, D. F., 1992, Quadruple labeling of brainstem neurons: a multiple
retrograde fluorescent tracer study of axonal collateralization, J. Neurosci. Methods 45:15–
22.
Anderson, C. R., and Edwards, S. L., 1993, Subunit b of cholera toxin labels interstitial cells of
Cajal in the gut of rat and mouse, Histochemistry 100:457–464.
Anderson, K. D., Karle, E. J., and Reiner, A., 1994, A pre-embedding triple-label electron
microscopic immunohistochemical method as applied to the study of multiple inputs to
defined tegmental neurons, J. Histochem. Cytochem. 42:49–56.
Angelucci, A., Clascá, F., and Sur, M., 1996, Anterograde axonal tracing with the sub-
unit B of cholera toxin: a highly sensitive immunohistochemical protocol for reveal-
ing the fine axonal morphology in adult and neonatal brains, J. Neurosci. Methods 65:
101–112.
Antal, M., Freund, T. F., Somogyi, P., and McIlhinney, R. A., 1990, Simultaneous anterograde
labelling of two afferent pathways to the same target area with Phaseolus vulgaris leucoag-
glutinin and Phaseolus vulgaris leucoagglutinin conjugated to biotin or dinitrophenol, J.
Chem. Neuroanat. 3:1–9.
Balercia, G., Cheng, S., and Bentivoglio, M., 1992, Electron microscopic analysis of fluorescent
neuronal labeling after photoconversion, J. Neurosci. Methods 45:87–98.
Bentivoglio, M., and Su, H. S., 1990, Photoconversion of fluorescent retrograde tracers, Neurosci.
Lett. 113:45–87.
Blackstad, T. W., 1965, Mapping of experimental axon degeneration by electron microscopy
of Golgi preparations, Z. Zellforsch. 67:819–834.
Blackstad, T. W., 1981, Tract tracing by electron microscopy of Golgi preparations, In: Heimer,
L., and RoBards, M. (eds.), Neuroanatomical Tract-Tracing Methods, New York: Plenum Press,
pp. 407–440.
Bruce, K., and Grofova, I., 1992, Notes on a light and electron microscopic double-labeling
methods combining anterograde tracing with Phaseolus vulgaris leucoagglutinin and ret-
rograde tracing with cholera toxin subunit β,J. Neurosci. Methods 45:23–33.
Chang, H. T., Kuo, H., Whittaker, J. A., and Cooper, N. G. F., 1990, Light and electron micro-
scopic analysis of projection neurons retrogradely labeled with Fluoro-Gold: notes on the
application of antibodies to Fluoro-Gold, J. Neurosci. Methods 35:31–37.
Coolen L. M., Jansen H. T., Goodman R. L., Wood R. I., and Lehman M. N., 1999, A new
method for simultaneous demonstration of anterograde and retrograde connections in
the brain: co-injections of biotinylated dextran amine and the beta subunit of cholera
toxin, J. Neurosci. Methods 91:1–8.
Coolen L. M., and Wood R. I., 1998, Reciprocal connections of the medial amygdaloid nu-
cleus in the Syrian hamster brain: simultaneous anterograde and retrograde tract tracing,
J. Comp. Neurol. 399:189–209.
Cullinan, W. E., and Zaborszky, L., 1991, Organization of ascending hypothalamic projections
to the rostral forebrain with special reference to the innervation of cholinergic projection
neurons, J. Comp. Neurol. 306:631–667.
de Olmos, J. S., Ebbesson, S. O. E., and Heimer, L., 1981, Silver methods for the impregnation
of degenerating axoplasm, In: Heimer, L., and Robards, M. (eds.), Neuroanatomical Tract-
Tracing Methods, New York: Plenum Press, pp. 117–170.
Divac, I., and Mogensen, J., 1990, Long-term retrograde labeling of neurons, Brain Res. 524:339–
341.
Dolleman-Van der Weel, M. J., Wouterlood, F. G., and Witter, M. P., 1995, Multiple antero-
grade tracing, combining Phaseolus vulgaris leucoagglutinin with rhodamine- and biotin-
Edwards, S. B., and Hendrickson, A., 1981, The autoradiographic tracing of axonal connections
in the central nervous system, In: Heimer, L., and RoBards, M. (eds.), Neuroanatomical
Tract-Tracing Methods, New York: Plenum Press, pp. 171–205.
Fritzsch, B., 1993, Fast axonal diffusion of 3000 molecular weight dextran amines, J. Neurosci.
Methods 50:95–103.
Fritzsch, B., and Wilm, C., 1990, Dextran amines in neuronal tracing, Trends Neurosci. 13:14.
Gaykema, R. P., van Weeghel R., Hersh, L. B., and Luiten, P. G., 1991, Prefrontal cortical
projections to the cholinergic neurons in the basal forebrain, J. Comp. Neurol. 303:563–
583.
Gerfen, C. R., and Sawchenko, P. E., 1985, A method for anterograde axonal tracing of chemi-
cally specified circuits in the central nervous system: combined Phaseolus vulgaris leucoag-
glutinin (PHA-L) tract tracing and immunohistochemistry, Brain Res. 343:144–150.
Gonatas, N. K., Harper, C., Mizutani, T., and Gonatas, J. O., 1979, Superior sensitivity of con-
jugates of horseradish peroxidase with wheat germ agglutinin for studies of retrograde
axonal transport, J. Histochem. Cytochem. 27:728–734.
Gonzalo, N., Moreno, A., Erdozain, M. A., Garcı́a, P., Vázquez, A., Castle, M., and Lanciego,
J. L., 2001, A sequential protocol combining dual neuroanatomical tract-tracing with the
visualization of local circuit neurons within the striatum, J. Neurosci. Methods 111:59–66.
Graham, R. C., and Karnovsky, M. J., 1966, The early stages of absorption of injected horseradish
peroxidase in the proximal tubules of mouse kidney: ultrastructural cytochemistry by a
new technique, J. Histochem. Cytochem. 14:291–302.
neuronal connections with Phaseolus vulgaris-leucoagglutinin (PHA-L) and combinations
with other neuroanatomical techniques, In: Wouterlood, F. G., Van den Pol, A., Björklund,
A., and Hökfelt, T. (eds.), Handbook of Chemical Neuroanatomy, Vol. 8, Amsterdam: Elsevier
Science Publishers, pp. 47–124.
Kirkham, M., Fujita, A., Chadda, R., Nixon, S. J., Kurzchalia, T. V., Sharma, D. K., Pagano
R. E., Hancock, J. F., Mayor, S., and Parton, R. G., 2005, Ultrastructural identification of
uncoated caveolin-independent early endocytic vehicles, J. Cell Biol. 168:465–476.
Köbbert, C., Apps, R., Bechmann, I., Lanciego, J. L., Mey, J., and Thanos, S., 2000, Current
concepts in neuroanatomical tracing, Prog. Neurobiol. 62:327–351.
Kristensson, K., and Olson, Y., 1971, Retrograde axonal transport of protein, Brain Res. 29:363–
365.
Kuypers, H. G. J. M., and Huisman, A. M., 1984, Fluorescent neuronal tracers, Adv. Cell. Neuro-
biol. 5:307–340.
Lanciego, J. L., and Giménez-Amaya, J. M., 1999, Notes on the combined use of V-VIP and
DAB peroxidase substrates for the detection of colocalising antigens, Histochem. Cell Biol.
111:305–311.
Lanciego, J. L., Goede, P. H., Witter, M. P., and Wouterlood, F. G., 1997, Use of peroxidase
substrate Vector r VIP for multiple staining in light microscopy, J. Neurosci. Methods 74:1–7.
Lanciego, J. L., Gonzalo, N., Castle, M., Sanchez-Escobar, C., Aymerich, M. S., and Obeso, J.
A., 2004, Thalamic innervation of striatal and subthalamic neurons projecting to the rat
entopeduncular nucleus, Eur. J. Neurosci. 19:1267–1277.
Lanciego, J. L., Luquin, M. R., Guillén, J., and Giménez-Amaya, J. M., 1998b, Multiple neu-
roanatomical tracing in primates, Brain Res. Protoc. 2:323–332.
Lanciego, J. L., and Wouterlood, F. G., 1994, Dual anterograde axonal tracing with Phaseolus
vulgaris leucoagglutinin (PHA-L) and biotinylated dextran amine (BDA), Neurosci. Protoc.
94-050-06.
Lanciego, J. L., Wouterlood, F. G., Erro, E., Arribas, J., Gonzalo, N., Urra, X., Cervantes, S., and
Giménez-Amaya, J. M., 2000, Complex brain circuits studied via simultaneous and per-
manent detection of three transported neuroanatomical tracers in the same histological
section, J. Neurosci. Methods 103:127–135.
Lanciego, J. L., Wouterlood, F. G., Erro, E., and Giménez-Amaya, J. M., 1998a, Multiple axonal
tracing: simultaneous detection of three tracers in the same histological section, Histochem.
Cell Biol. 110:509–515.
LaVail, J. H. 1975, The retrograde transport method, Fed. Proc. 34:1618–1624.
LaVail, J. H., and LaVail, M. M., 1972, Retrograde axonal transport in the central nervous
system, Science 176:1416–1417.
Luppi, P. H., Fort, P., and Jouvet, M., 1990, Iontophoretic application of unconjugated cholera
toxin β subunit (CTB) combined with immunohistochemistry of neurochemical sub-
stances: a method for transmitter identification of retrogradely labeled neurons, Brain
Res. 534:209–224.
Luppi, P. H., Sakai, K., Salvert, D., Fort, P., and Jouvet, M., 1987, Peptidergic hypothala-
mic afferents to the cat nucleus raphe pallidus as revealed by double immunostaining
technique using unconjugated cholera toxin as a retrograde tracer, Brain Res. 402:339–
345.
Mauro, A., Germano, I., Giaccone, G., Giordana, M. T., and Schiffer, D., 1985, 1-Naphthol ba-
sic dye (1-NBD), an alternative to diaminobenzidine (DAB) in immunoperoxidase tech-
niques, Histochemistry 83:97–102.
Mesulam, M. M., 1976, The blue reaction product in horseradish peroxidase neurohistochem-
istry, J. Histochem. Cytochem. 24:1273–1280.
Mesulam, M. M., 1978, Tetramethylbenzidine for horseradish peroxidase neurohistochemistry:
a non-carcinogenic blue reaction product with superior sensitivity for visualizing neural
afferents and efferents, J. Histochem. Cytochem. 26:106–117.
Mesulam, M. M., 1982, Principles of horseradish peroxidase neurochemistry and their appli-
cations for tracing neural pathways-axonal transport, enzyme histochemistry and light mi-
croscopic analysis, In: Mesulam, M. M. (ed.), Tracing Neural Connections with Horseradish
Peroxidase, New York: Wiley: IBRO Handbook Series: Methods in the Neurosciences,
pp. 1–551.
Millhouse, O. E., 1981, The Golgi methods, In: Heimer, L., and RoBards, M. (eds.), Neu-
roanatomical Tract-Tracing Methods, New York: Plenum Press, pp. 311–344.
Nance, D. M., and Burns, J., 1990, Fluorescent dextrans as sensitive anterograde neuroanatom-
ical tracers: applications and pitfalls, Brain Res. Bull. 25:139–145.
Novikova, L., Novikov, L., and Kellerth, J.-O., 1997, Persistent neuronal labeling by retrograde
Pieribone, V. A., and Aston-Jones, G., 1988, The iontophoretic application of Fluoro-Gold for
the study of afferents to deep brain nuclei, Brain Res. 475:259–271.
terograde and retrograde neuronal tracer, Brain Res. 607:47–53.
Risold P. Y., and Swanson L. W., 1995, Evidence for a hypothalamothalamocortical circuit
mediating pheromonal influences on eye and head movements, Proc. Natl. Acad. Sci. U. S.
A. 92:3902–3989.
Rosene, D. L., Roy, N. J., and Davis, B. J., 1986, A cryoprotection method that facilitates cut-
ting frozen sections of whole monkey brain for histological and histochemical processing
Schmued, L. C., 1994, Anterograde and retrograde neuroanatomical tract-tracing with fluo-
rescent compounds, Neurosci. Protoc. 94-050-02.
Schmued, L. C., and Fallon, J. H., 1986, Fluoro-Gold: a new fluorescent tracer with numerous
unique properties, Brain Res. 377:147–154.
Schmued, L. C., and Heimer, L., 1990, Iontophoretic injection of fluorogold and other fluo-
Schmued, L. C., Kyriakidis, K., and Heimer, L., 1990, In vivo anterograde and retrograde
axonal transport of the fluorescent rhodamine-dextran-amine, Fluoro-Ruby, within the
CNS, Brain Res. 526:127–134.
Skirboll, L. R., Thor, K., Helke, C., Hökfelt, T., Robertson, B., and Long, R., 1989, Use of
retrograde fluorescent tracers in combination with immunohistochemical methods, In:
Zaborszki, L., and Heimer, L., (eds.), Neuroanatomical Tract-Tracing Methods 2, New York:
Smith, Y., and Bolam, J. P., 1991, Convergence of synaptic inputs from the striatum and the
globus pallidus onto identified nigrocollicular cells in the rat: a double anterograde label-
ing study, Neuroscience 44:45–73.
Smith, Y., and Bolam, J. P., 1992, Combined approaches to experimental neuroanatomy: com-
bined tracing and immunocytochemical techniques for the study of neuronal microcir-
cuits, In: Bolam, J. P. (ed.), Experimental Neuroanatomy, A Practical Approach, Oxford: Oxford
University Press, pp. 239–266.
Somogyi, P., Hodgson, A. J., and Smith, A. D., 1979, An approach to tracing neuron networks in
the cerebral cortex and basal ganglia. Combination of Golgi staining, retrograde transport
of horseradish peroxidase and anterograde degeneration of synaptic boutons in the same
material, Neuroscience 4:1805–1852.
Stoeckel, K., Schwab, M. E., and Thoenen, H., 1977, Role of gangliosides in the uptake and
retrograde transport of cholera and tetanus toxin as compared to nerve growth factor and
wheat germ agglutinin, Brain Res. 132:273–285.
Ter Horst, G. J., Groenewegen, H. J., Karst, H., and Luiten, P. G. M., 1984, Phaseolus vulgaris leu-
coagglutinin immunohistochemistry. A comparison between autoradiographic and lectin
tracing of neuronal efferents, Brain Res. 307:379–383.
Thomas M. A., and Lemmer B., 2005, HistoGreen: a new alternative to 3.3 -diaminobenzidine-
tetrahydrochloride-dihydrate (DAB) as a peroxidase substrate in immunohistochemistry?
Brain Res. Protoc. 14:107–118.
Trojanowski, J. Q., Gonatas, J. O., and Gonatas, N. K., 1981, Conjugates of horseradish perox-
idase (HRP) with cholera toxin and wheat germ agglutinin are superior to free HRP as
orthograde transported markers, Brain Res. 223:381–385.
Trojanowski, J. Q., Gonatas, J. O., Steiber, A., and Gonatas, N. K., 1982, Horseradish peroxidase
(HRP) conjugates of cholera toxin and lectins are more sensitive retrograde transported
markers than free HRP, Brain Res. 231:33–50.
Trojanowski, J. Q., and Schmidt, M. L. 1984, Interneuronal transfer of axonally transported
proteins: studies with HRP and HRP conjugates of wheat germ agglutinin, cholera toxin
and the B subunit of cholera toxin, Brain Res. 311:366–369.
Van Bockstaele, E. J., Wright, A. M., Cesari, D. M., and Pickel, V., 1994, Immunolabeling of
retrogradely transported fluorogold. Sensitivity and application of ultrastructural analysis
of transmitter-specific mesolimbic circuitry, J. Neurosci. Methods 55:65–78.
Veenman, C. L., Reiner, A., and Honig, M. G., 1992, Biotinylated dextran amine as an antero-
grade tracer for single- and double-label studies, J. Neurosci. Methods 41:239–254.
Vetter, D. E., Saldaña, E., and Mugnaini, E., 1993, Input from the inferior colliculus to medial
olivocochlear neurons in the rat: a double label study with PHA-L and cholera toxin, Hear.
Res. 70:173–186.
Wan, X. C. S., Trojanowski, J. Q., and Gonatas, J. O., 1982, Cholera toxin and wheat germ agglu-
tinin conjugates as neuroanatomical probes: their uptake and clearance, transganglionic
and retrograde transport and sensitivity, Brain Res. 243:215–224.
Warr, W. B., de Olmos, J. S., and Heimer, L., 1981, Horseradish peroxidase: the basic procedure,
In: Heimer, L., and RoBards, M. (eds.), Neuroanatomical Tract-Tracing Methods, New York:
Weiss, P., and Hiscoe, H. B., 1948, Experiments on the mechanism of nerve growth, J. Exp. Zool.
107:315–396.
Wessendorf, M. W., 1991, Fluoro-Gold: composition and mechanism of uptake, Brain Res.
553:135–148.
Woolf, N. J., Hernit, M. C., and Butcher, L. L., 1986, Cholinergic and noncholinergic projections
from the rat basal forebrain revealed by combined choline acetyltransferase and Phaseolus
vulgaris-leucoagglutinin immunohistochemistry, Neurosci. Lett. 66:281–286.
Wouterlood, F. G., Bol, J. G. J. M., and Steinbusch, H. W. M., 1987, Double-label immunocyto-
chemistry: combination of anterograde neuroanatomical tracing with Phaseolus vulgaris-
leucoagglutinin and enzyme immunocytochemistry of target neurons, J. Histochem. Cy-
tochem. 35:817–823.
Wouterlood, F. G., Goede, P. H., Arts, M. P. M., and Groenewegen, H. J., 1992, Simultaneous
characterization of efferent and afferent connectivity, neuroactive substances and mor-
phology of neurons, J. Histochem. Cytochem. 40:457–465.
Wouterlood, F. G., Goede, P. H., and Groenewegen, H. J., 1990, The in situ detectability of the
neuroanatomical tracer Phaseolus vulgaris-leucoagglutinin, J. Chem. Neuroanat. 3:11–18.
Wouterlood, F. G., and Groenewegen, H. J., 1985, Neuroanatomical tracing by use of Phaseolus
vulgaris-leucoagglutinin (PHA-L): electron microscopy of PHA-L filled neuronal somata,
dendrites, axons and axon terminals, Brain Res. 326:188–191.
Wouterlood, F. G., and Groenewegen, H. J., 1991, The Phaseolus vulgaris-leucoagglutinin tracing
technique for the study of neuronal connections, Prog. Histochem. Cytochem. 22:1–78.
biotinylated dextran amine: comparison with the tracer PHA-L in preparations for electron
microscopy, J. Neurosci. Methods 48:75–87.
Wouterlood, F. G., Mugnaini, E., and Nederlof, J., 1985, Projection of olfactory bulb effer-
ents to layer I GABA-ergic neurons in the entorhinal area. Combination of anterograde
degeneration and immunoelectron microscopy in rat, Brain Res. 343:283–296.
Zaborszky, L., and Cullinan, W. E. 1989, Hypothalamic axons terminate on forebrain cholin-
ergic neurons: an ultrastructural double-labeling study using PHA-L tracing and ChAT
immunocytochemistry, Brain Res. 479:177–184.
Zaborszky, L., Leranth, C., and Heimer, L., 1984, Ultrastructural evidence of amygdalofugal
axons terminating on cholinergic cells of the rostral forebrain, Neurosci. Lett. 21:219–225.
12
Tract-Tracing in Developing
Systems and in Postmortem
Human Material Using
Carbocyanine Dyes
ZOLTÁN MOLNÁR, DANIEL BLAKEY,
IRINA BYSTRON, and
ROSALIND S. E. CARNEY
INTRODUCTION
OVERVIEW OF CLASSICAL AND MODERN TRACT-TRACING
TECHNIQUES
Classical Neuroanatomical Tract-Tracing Techniques
Modern Techniques Using Carbocyanine Dyes
Methodology of Carbocyanine Dye Tracing
Additional Techniques: Combination with Immunohistochemistry,
Photoconversion of Carbocyanine Dye-Labeled Material, and
Electron Microscopic Analysis
Technical Considerations and Limitations of Carbocyanine Dye
Tracing
CARBOCYANINE DYE TRACING IN PRIMATES, INCLUDING NORMAL
AND PATHOLOGICAL POSTMORTEM HUMAN SPECIMENS
Interspecies Comparison of Axonal Pathway Development
Carbocyanine Dye Tracing in Postmortem Human Tissue
EMERGING TECHNIQUES FOR NEUROANATOMICAL TRACT
IMAGING
ZOLTÁN MOLNÁR, DANIEL BLAKEY AND ROSALIND S. E. CARNEY

• Department of Human Anatomy and Genetics, University of Oxford, South Parks Road,
Oxford OX1 3QX, United Kingdom IRINA BYSTRON • University Laboratory of
Physiology, University of Oxford, Parks Road, Oxford OX1 3PT, United Kingdom; Department
of Morphology, Institute of Experimental Medicine, 12 Acad. Pavlova, St Petersburg 197376,
Russia ROSALIND S. E. CARNEY • Department of Neuroscience, Research Building
EP08, Georgetown University School of Medicine, Washington, DC 20057, USA
366
TRACT-TRACING IN DEVELOPING SYSTEMS 367
CONCLUDING REMARKS
APPENDIX
Protocol for DiI Placement and Histological Processing (Modified
from Molnár et al., 1998)
Protocol for Photoconversion (Modified from Lübke, 1993)
REFERENCES
Abstract: Rapid progress in neurobiology and genetics demands knowledge of fun-

damental aspects of brain development including the connectivity patterns within
developing and adult brains. The primary focus of this chapter is on neuroanatomical
tract-tracing using carbocyanine dyes which have several advantages over traditional
tracing methods. First utilized for in vitro studies, a major breakthrough in the late
1980s was the demonstration that carbocyanine dyes act as anterograde and retro-
grade tracers in fixed tissue, eliminating the need for diffusion of tracers in vivo.
Moreover, carbocyanine dyes are more efficacious than classical tracing methodolo-
gies especially during early stages of development, and consequently have been used
to reveal the spatiotemporal patterns of axonal development in different species.
Furthermore, the unique properties of the carbocyanine dye tracing method have
opened up new avenues for tracing connections in human postmortem specimens.
This is a key step in determining the precise connectivity of neural circuits in the
human brain, and subsequently to relate this knowledge to pathological cases.
The success of carbocyanine dyes as tracers, both in vitro and in fixed material,
is reflected in the flurry of publications throughout the 1990s and into the present.
However, there are relatively few systematic studies that have tested parameters to
optimize their use or to give practical advice to enhance their efficacy. This chapter
aims to bring together some of our experiences with the carbocyanine dye tracing
method drawn from our studies in mammalian, reptilian, and human and nonhu-
man primate specimens.
Keywords: carbocyanine dyes, cortex, development, photoconversion, primate,
thalamus
I. INTRODUCTION
Topographic patterns of cortico-cortical and cortico-subcortical axonal

projections have been elucidated in classical studies examining brain con-
nectivity using techniques considered time-consuming, laborious, and ex-
pensive. Such techniques often require demanding surgical procedures to
deliver tracers into living animals, which are subsequently sacrificed after
sufficient transport time to reveal the tracer in the target tissue (for overview
see Zaborszky and Heimer, 1989). These methods necessitate maintaining
a surgical facility and appropriate housing of animal colonies. The applica-
tion of carbocyanine dyes (Godement et al., 1987; Honig and Hume, 1986)
for neuroanatomical tract-tracing in fixed brains from embryonic and early
postnatal animals provided an opportunity to simplify this process. The use
of fixed brains gives the experimenter more precise temporal control, as
fixation provides a snapshot of the status of axonal development during
368 ZOLTÁN MOLNÁR et al.
embryogenesis. Spatial control is achieved by having access to a particular

region, various nuclei, or lamina under direct visual control in fixed brain
specimens, in contrast to earlier methods requiring stereotactic injections.
Next to the controllability of carbocyanine dye tracing, one of its biggest
advantages is the opportunity it provides to trace in human postmortem
brain tissue.
In this chapter, we discuss classical methods of neuroanatomical tract-
tracing coupled with the emergence of carbocyanine dyes as a powerful tool
in the field of developmental neurobiology. Whereas classical techniques
provided fundamental knowledge of connectivity within the adult brain,
the developmental timetable of these projections has been revised using
carbocyanine dye tracing. Carbocyanine dye tracing continues to flourish
in publications from investigative studies of rodent brain development and
particularly in transgenic models. This has enabled developmental neurobi-
ologists to form causal links between targeted gene ablation and abnormal
axonal pathfinding.
II. OVERVIEW OF CLASSICAL AND MODERN

TRACT-TRACING TECHNIQUES
A. Classical Neuroanatomical Tract-Tracing Techniques
The Golgi stain, used by neuroanatomists in the nineteenth century, pro-

vided a means of labeling neurons and was used to visualize their soma
and processes. This technique was extensively used in embryonic and early
postnatal animals and humans. However, the method was not conducive
to the study of axonal tracts, as individual axons could not be traced
from one brain region to the next. However, the eminent neuroanatomist
Ramon y Cajal was able to perform some axon tracing analysis in embryonic
brains using the Golgi staining techniques (Cajal, 1909). He also found that
the stain gave better results in the nervous system of young and develop-
ing animals compared to the mature brain. Indeed, Golgi impregnation in
fibers stops when the fiber acquires its myelin sheath (Wouterlood and Mug-
naini, 1984). In the mid-twentieth century, axon tracing was accompnished
by injuring neurons and analyzing their subsequent degeneration to follow
the trajectory of the neurites (Nauta and Gygax, 1951). Modern techniques
involve the injection of tracers into the brain, which are transported through
axons anterogradely (e.g., autoradiography of labeled amino acids or Phase-
olus vulgaris), and/or retrogradely (e.g., fluorescent dyes, rhodamine beads;
see Zaborszky and Heimer, 1989). Transneuronal transport can also be
achieved with wheat germ agglutinin–horseradish peroxidase.
B. Modern Techniques Using Carbocyanine Dyes
The neuroanatomical visualization of fiber projections has been

markedly enhanced by the use of the lipophilic carbocyanine dye series
Table 12.1. Specifications of the most commonly used carbocyanine dyes
(based on information provided by Molecular Probes).1
Absorption Fluorescence Molecular
Common Chemical wavelength emission Probes catalog
name name peak (nm) peak (nm) numbers
DiI 1,1 -Dioctadecyl-3,3,3 ,3 - 551 569 D-282, D-383,
tetramethylindocarbocyanine D-384, D-3886,
perchlorate D-3911, N-22880
Fast DiI 1,1 -Dilinoleyl-3,3,3 ,3 - 5512 5692 D-3899, D-7756
tetramethylindocarbocyanine,
4-chlorobenzenesulfonate
DiA 4-(4-(Dihexadecylamino)styryl)- 456 590 D-3883, D-3897,
N-methylpyridinium iodide D-7758
DiAsp 4-(4-(Didecylamino)styryl)-N- 4562 5902 D-291
methylpyridinium iodide
DiO 3,3 - 484 501 D-275, D-1125,
Dioctadecyloxacarbocyanine D-3898, N-22881
perchlorate
DiD 1,1 -Dioctadecyl-3,3,3 ,3 - 644 665 D-307, D-7757,
tetramethylindodicarbocyanine, N-22882
4-chlorobenzenesulfonate salt
1 http://probes.invitrogen.com/handbook/tables/0346.html.
2 Spectral data stated represent estimated values based on similar compounds.
(Honig and Hume, 1986). Initially these dyes were used to study
membrane fluidity by cell biologists. Subsequently, tracers such as DiI
(1,1 -dioctadecyl-3,3,3 ,3 -tetramethylindocarbocyanine perchlorate; all car-
bocyanine dyes listed are from Molecular Probes, Eugene, OR) and
DiO (3,3 -dioctadecyloxacarbocyanine perchlorate), DiD (1,1 -dioctadecyl-
3,3,3 ,3 -tetramethylindodicarbocyanine, 4-chlorobenzenesulfonate salt),
DiAsp (4-(4-(didecylamino)styryl)-N-methylpyridinium iodide), and DiA (4-
(4-(dihexadecylamino)styryl)-N-methylpyridinium iodide) were used to la-
bel living, cultured neurons, enabling neuronal interactions and cell migra-
tion to be examined in vitro (Honig and Hume, 1986). See Table 12.1 for
absorption and emission spectra for various carbocyanine dyes, listed with
catalog numbers from Molecular Probes.
Additional applications of carbocyanine dye tracing in vitro have been
studied incorporating time-lapse video microscopy to track axon growth
or cell migration within developing systems (for review see Fishell et al.,
1995). A forebrain explant from murine cortex showed that DiI-labeled
precursor cells can undergo lateral dispersion within the ventricular zone
prior to radial migration so as to reach the developing cortical plate (Fishell
et al., 1993). Carbocyanine dye labeling of cells of the lateral ganglionic em-
inence in organotypic slice preparations was used in one of the first studies
that showed tangential migration of interneurons to the dorsal cortex (De
Carlos et al., 1996). Time-lapse video microscopy was also used to describe
growth cone morphology of thalamic axons in vitro in thalamocortical slices
(Skaliora et al., 2000) or cocultures (Yamamoto et al., 1997).
The discovery by Godement et al. (1987) that the dye series is effective for
the antero- and retrograde labeling of fiber populations in fixed tissue was a
significant breakthrough. The dyes are thought to diffuse laterally through
the plasma membrane of the cell at a rate of ∼6 mm/day in living tissue
(Honig and Hume, 1989), with a slower diffusion rate in fixed tissue of
∼2 mm/day. The carbocyanine dye series contains several variants with dif-
ferent absorption/emission spectra permitting simultaneous use. DiI, which
is excited in the green range and fluoresces in a red/orange color under a
rhodamine filter, is often used in combination with DiA, which emits in a
green wavelength, for two-color imaging (Honig and Hume, 1989). DiI and
DiA can be detected with rhodamine and fluorescein optical filters, respec-
tively. Such multicolor labeling is useful for visualizing multiple pathways
concurrently, or for topographic labeling within a single axonal pathway
(see Agmon et al., 1996; Bicknese et al., 1994; Miller et al., 1993; Molnár
et al., 1998a). Figure 12.1 demonstrates several examples of the application
of DiI tracing in various systems.
The mechanism of tracing with carbocyanine dyes is based on their lipid
solubility. DiI becomes incorporated into the outer leaflet of the plasma
membrane and can laterally diffuse with negligible transfer between intact
cell membranes in living and fixed material (Godement et al., 1987; Honig
and Hume, 1986, 1989). There is some evidence that the dye molecules form
detergent-like micelles which are sparingly soluble in the cytoplasm and can
therefore “fill” the entire cell (Bruce et al., 1997). The direction of diffusion
can be either toward the cell body (retrograde) or toward the distal end of
the axon (anterograde). Simultaneous retrograde and anterograde labeling
can be either a desirable occurrence or an adverse occurrence, depending
on the objective of the experiment. The ability to label bidirectionally can
be useful to detect the source of fibers that project through the dye place-
ment site. For example, carbocyanine crystal placement into the internal
capsule of the embryonic forebrain reveals the origin of the early cortical
→
Figure 12.1. Examples for the application of carbocyanine dye tracing in various
systems. (A) Pyramidal neurons were backlabeled through their projections across
the corpus callosum in a postnatal rat brain (P3). A DiI crystal was placed in the
contralateral hemisphere and the brain was incubated for 3 weeks at 37◦ C. Numerous
layer 5 and layers 2/3 cells were labeled. At this early stage, they all possess an
apical dendrite with a terminal tuft reaching the marginal zone. (B) Two cerebral
cortical slices were taken at P3 and cocultured for 2 weeks. Then, the cultures were
fixed and DiI crystals were placed into one explant. This layer 5 pyramidal cell was
backlabeled with DiI indicating in vivo-like connectivity between the explants. (C)
Organized reciprocal connections cross the embryonic rat internal capsule in this
horizontal section of an E20 rat brain counterstained with bisbenzimide. Multiple
carbocyanine dye placements were made along an anteroposterior parasagittal line
into the convexity of the cortex (DiA, DiI, and DiAsp). Since thalamic fibers have
already arrived at the cortex prior to E16, each placement labeled a mixed bundle of
thalamocortical and corticofugal axons. Six weeks incubation at room temperature
Figure 12.1. (Cont.) was used to enable full anterograde and retrograde diffusion.
Three distinct bundles are clearly visible passing through the primitive internal
capsule without substantial mixing or crossing. (D) Carbocyanine dye labeling (DiI)
revealed the periphery-related patterning of thalamocortical axons in a tangential
section through the barrel cortex in the primary somatosensory cortex of a P3 rat
brain. A DiI crystal was placed into the ventrobasal complex and we used 4 weeks
of incubation at 37◦ C. The nonlabeled darker areas represent the emerging septa.
Scale bars: 50 µm (A); 20 µm (B); 100 µm (C); 250 µm (D).
and thalamic neurons that project through the internal capsule (McConnell
et al., 1989; Molnár and Cordery, 1999). As well as revealing the origin of
axons by retrograde transport, anterograde diffusion of the dye reveals the
target of the fibers of passage that incorporates the dye at the placement site.
This can be useful for investigating patterns of connectivity that are poorly
understood. According to our own experience, there is limited scope to ma-
nipulate the relative strength of the anterograde and retrograde transport
by reducing or increasing, respectively, the incubation periods.
C. Methodology of Carbocyanine Dye Tracing
This section details the standard methodology used to apply carbo-

cyanine dyes, either in solid form or as a solution, to fixed and in vitro
preparations, tissue processing, and analysis. The section “Additional
Techniques: Combination with Immunohistochemistry, Photoconversion
of Carbocyanine Dye-Labeled Material, and Electron Microscopic Analysis”
discusses other techniques that may be combined with dye tracing. Some
technical considerations for the suitability of carbocyanine dyes are
considered in the section “Technical Considerations and Limitations of
Carbocyanine Dye Tracing.”
1. Fixation and Storage of the Tissue
Chemical fixation using aldehyde fixatives is compatible with carbocya-

nine dye tracing. Generally, in younger embryos [< E16 (embryonic day
16) in mice, E17 in rat], the heads are removed and fixed overnight in
4% formaldehyde in phosphate buffer (PB) 0.1 M pH 7.4 (4% formalin),
whereas older embryos and postnatal animals require transcardial perfu-
sion prior to overnight immersion fixation. Thereafter, tissue may be stored
in phosphate-buffered saline with sodium azide (PB 0.1 M, 0.09% NaCl,
0.05% sodium azide, pH 7.4) to avoid microbial growth, and to prevent anti-
gen masking if immunohistochemistry is desired. Additional glutaraldehyde
during initial fixation may be used if the tissue is to be analyzed by electron
microscopy; however, fluorescence from the glutaraldehyde may produce
higher background and thus reduce the signal-to-noise ratio of the dye.
Carbocyanine dye tracing is incompatible with alcohol fixation and paraffin
embedding of tissue. It is also believed that freshly fixed brains produce bet-
ter results, but we have observed successful labeling in material that has had
extended storage in a fixative (4% formaldehyde). Nonetheless, a prolonged
postfixation period is not recommended; the resultant high precipitation
of membrane and cytoplasmic proteins can lead to slower diffusion of DiI
(Bruce et al., 1997), and polymerization of the formaldehyde releases al-
cohol moieties which may impair labeling. The carbocyanine dye diffusion
speed has not been systematically analyzed as a function of the length or the
strength of the fixation.
2. Tracer Delivery
There are several methodologies in use for the delivery of carbocyanine

dyes in living and fixed tissue:
(A) One of the most widely used methods is crystal placement. Small individual
crystals (∼100 µm in size) are picked up and inserted into the tissue
using platinum wires or insect pins. If the crystals are to be inserted
through a membrane rather than just into nervous tissue, preboring
a hole is usually desirable. The tissue, including cell membranes, is
damaged at the implantation site, but this is believed to facilitate
the dye uptake. This method is equally suitable for living and fixed
tissues (see McConnell et al., 1989; Molnár et al., 1998a). Removing
loose or sprinkled fragments of carbocyanine crystals, attached to
the surface of the specimen with a damp tissue or by pipetting off
with H2 O/buffer, avoids contamination of the label from the site of
attachment. Otherwise, tracer diffusion from aberrant crystals may
confound the interpretation of the experiment.
(B) Solution injection. Saturated solutions of the dyes may be made up in
95% ethanol, 5% DMSO. Picospritzers (Intracel, Royston, UK) are
used to inject nanoliter volumes of dye into the tissue (see Agmon
et al., 1996; Métin and Godement, 1996). Close to the site of the
injection, there is aspecific diffusion, which is partially due to the
alcohol solution. This method of delivery is also suitable for fixed
specimens in addition to living tissues in vivo and in vitro.
(C) Solution soaking can be used to label culture explants or grafts in living
tissue. To label the entire surface of the tissue, the explant may be
briefly immersed in a dye solution followed by rapid and extensive
rinsing. “Microcrystals” of dye will precipitate onto the surface of
the explant and neurite outgrowth can be studied within cocultured
structures, for example, explants of thalamus and cortex (Molnár and
Blakemore, 1991). This method is suitable for in vitro living tissue.
During short culture periods (up to 4–6 days in vitro) this method
labels neurites, but at later stages the dye becomes incorporated into
the cytosol and the fiber labeling gradually disappears. This method
will also reveal the movement of motile cells (such as macrophages)
within the in vitro preparation as well as migratory cells that incorpo-
rated the label at the dye placement site.
(D) Coating inert substrates. Tungsten or gold particles are placed onto
a slide and a drop of dye solution in methylene chloride is added.
The solvent rapidly evaporates, leaving a coating of dye on the par-
ticles. These particles can then be delivered into the tissue using a
“gene gun” (BioRad, Hercules CA). Single cells and low density cell
populations can be labeled in this manner (Gan et al., 2000). This
method is suitable for both living and fixed tissues. Alternatively,
nitrocellulose sheets may be soaked in the same manner, cut into
0.5-mm squares, and inserted into fixed tissue (Ma et al., 2002).
3. Targeting the Tracer to Specific Anatomical Locations
Successful targeting of specific nuclei or anatomical structures is aided by

using a high-magnification binocular light microscope. It is relatively easy
in fixed material to dissect away overlying tissue to access nuclei or particu-
lar anatomical layers directly. For example, to perform a crystal placement
in the lateral geniculate nucleus (LGN), the midbrain may be transected
coronally, rostral to the superior colliculus to gain access to the posterior
dorsal thalamus. Alternatively, to access the ventrobasal (VB) complex, the
brain is bisected longitudinally to separate the hemispheres, and the me-
dial part of the diencephalon is cut away (Molnár et al., 1998a). The use of
whole mount preparations, such as the intact brain, is advantageous for fixed
tissue tracing as the whole axonal projection should be maintained. It is also
possible to perform the tracer delivery in sectioned tissue (see sectioning in
“Sectioning of the Tissue”); however, this is advisable only if the connections
are already well understood, and can be preserved in the preparation. For
example, thalamocortical slice preparations can maintain connectivity be-
tween thalamus and cortex for the somatosensory or the auditory pathways,
yet a similar preparation is not possible for the visual pathway (Agmon and
Connors, 1991).
4. Incubation Periods
The temperature and length of the incubation are two main parameters
that can influence dye diffusion. These variables are specific to different
carbocyanine dyes. Additional techniques of enhancing diffusion in fixed
human specimens (Sparks et al., 2000) are discussed in the section “Car-
bocyanine Dye Tracing in Primates, Including Normal and Pathological
Postmortem Human Specimens” which reviews carbocyanine tracing in pri-
mates.
Incubation at 37◦ C is believed to enable a faster diffusion rate in fixed
tissue (see Table 12.2 from Molnár et al., 1998a), and carbocyanine dyes with
shorter fatty acid chain diffuse faster than tracers with longer chains (see Fast
Table 12.2. Incubation periods (weeks) needed for

carbocyanine dye (DiI) transport in embryonic and
early postnatal rats to label thalamocortical
projections (Molnár et al., 1998b).
Incubation period (weeks)
Age 22◦ C 37◦ C

E13–E14.5 2 1
E14.5–E16 3–6 2–4
E18–P0 6 4
P0–P8 6–10 4–8
DiI, in section “Altering Experimental Variables to Attempt Enhancement
of DiI Diffusion in Primate Brains”). For young embryonic specimens, room
temperature is preferred because the incubation period is longer and thus
easier to control, while background staining and transneuronal labeling
(discussed below) are less apparent. Different carbocyanine dyes can vary
in their optimal incubation periods in a given specimen. In mammalian
fixed brain tissue, as aforementioned, DiI (18 carbons “C18”) proves to be
the most reliable tracer, yielding the highest intensity of fluorescent labeling
and appropriate background staining. The incubation period required for
DiA or DiAsp seems to be shorter than that used for DiI. Therefore, for an
E16 tracing experiment, we routinely implant DiA crystals a week later than
DiI crystals (see Molnár et al., 1998a). This delayed implantation protocol
optimizes the transport period for both dyes.
DiA, DiD, and DiAsp (see Molecular Probes Handbook) are also reason-
able tracers in fixed embryonic and early postnatal rodent tissue, but the
background staining tends to be higher, despite a shorter incubation period
than that used for DiI. However, such parameters may strongly depend on
several factors, such as the length of storage in fixative, the age of the spec-
imen, and even the particular pathway studied. Table 12.2 gives a general
idea on the incubation periods needed for DiI labeling of the thalamocor-
tical pathway in embryonic and early postnatal rats (Molnár et al., 1998b).
DiD has also been used in combination with DiI in other studies (Catalano
and Shatz, 1998), although we have not found these tracers to be as effec-
tive. While DiO is an excellent tracer in living tissue, its transport is not as
impressive in fixed material.
5. Sectioning of the Tissue
Tracing with carbocyanine dyes limits the repertoire of the storage and sec-
tioning methods generally used for tissue processing. The specimen should
be stored without freezing in 4% formalin (0.1 M PBS with 0.05% NaN3 ).
Methods that require freezing (cryostat, sliding microtome) or wax embed-
ding are not compatible, and thus Vibratome sectioning is most commonly
used (Vibratome, Oxford Instruments or Vibroslicer, Leica VT1000S). Em-
bryonic and early postnatal sections are fragile; therefore, it is advisable to
embed the tissue in agarose (4% in PBS), which is heated and mixed until
the agar is fully dissolved, typically in a microwave oven. The agarose must be
allowed to cool prior to embedding the tissue; in most cases 50◦ C is appro-
priate. To further reduce any heat damage, we place the embedding molds
on crushed ice and fill with heated agarose. Once the correct temperature
has been reached, the brain is then placed into the agarose. For small em-
bryonic brains low melting agar (e.g., Type VII, Sigma A-4018) should be
used. After cutting it is not necessary to remove the agar surrounding the
section, which in fact aids mounting the section onto the slide and maintain
morphology.
6. Counterstaining, Mounting, and Analysis of Carbocyanine

Dye-Labeled Material
This section deals with counterstaining and mounting of carbocyanine

dye-labeled material for immediate analysis with epifluorescence and con-
focal microscopy. Further processing for photoconversion of DiI label into
a permanent product, electron microscopy, and immunohistochemistry is
addressed in the section “Additional Techniques: Combination with Im-
munohistochemistry, Photoconversion of Carbocyanine Dye-Labeled Mate-
rial, and Electron Microscopic Analysis.”
(A) Counterstaining. To reveal the major anatomical subdivisions and cy-

toarchitecture of various structures, fluorescent “Nissl” counterstains
can be used. We routinely use Bisbenzimide (10 min in 2.5 µg/ml
solution in PBS, protected from light), although acridine orange
(10 µg/ml in PBS, 10 min) or DAPI or Neuro Trace 500/525 green or
the Sytox counterstains are also excellent (see Molecular Probes Hand-
book for concentrations of other counterstains; www.probes.com). Ob-
viously, the fluorescence signal of the counterstain must not conflict
with that of the dye, and therefore the choice of counterstain is de-
pendent on the carbocyanine dye used; e.g., DiA is not compatible
with acridine orange.
(B) Mounting media. The recommended mounting media are PBS, vari-
ous versions of PBS–glycerol solutions (1:1, 3:1, or 1:9 mixture), or
Hydromount (National Diagnostics). Antiquenching agents (1% p-
phenylenediamine, 10% PBS, and 90% glycerol) (e.g., Johnson et al.,
1982) can prolong the signal for numerous other methods, but we
did not find this particularly beneficial for carbocyanine dyes (Molnár
and Carney, unpublished observations, 2003). Since the label is ro-
bust, the major problem is the deterioration of the label at the cut
surfaces and nonspecific signal. Degradation is inevitable after a few
weeks, and therefore it is advisable that the material is analyzed shortly
after sectioning (preferably within the first 3 days).
(C) Coverslipping and storage. During mounting, it is imperative to cover the
sections before they dry out, otherwise the label becomes blurry and
diffuse. Since the sections do not adhere to the slide firmly, lowering
the coverslips should be done with extra care. The coverslips are
sealed with Paraseal (Raymond A Lamb, London) or with nail varnish
and are stored, light-protected at 4◦ C.
(D) Epifluorescence analysis. DiI and DiA labeling are photographed using
TRTIC and FITC filters, respectively. For conventional photography,
highly sensitive films (e.g., 400 or 1600 ASA Kodak Ectachrome for
Slide) are recommended. For double and triple exposure, it is recom-
mended that the bisbenzimide counterstain or DiA label is exposed
for a shorter period than is DiI. Nowadays, digital cameras easily fa-
cilitate documentation. After single exposure, the individual frames
can be adjusted for brightness, contrast, and can be superimposed
on the counterstain label using image processing. Thus, combining
dye labeling with nuclear counterstaining provides an accurate view
of labeled axonal projections within the context of the anatomical
structure. When only one carbocyanine dye is used, e.g., DiI, pho-
tographs taken under TRITC filter illumination can be converted to
gray scale to create a high-contrast image (gray scale tone inversion
may further enhance the image). Multicolor dye tracing provides in-
formation on the topographic relationship between different axonal
tracts. In fiber systems that exhibit a high topographical organization,
alternating dye combinations, such as DiA–DiI–DiA, have been used
to reveal the trajectories of neighboring axonal tracts (Molnár et al.,
1998a).
(E) Confocal microscopy. Confocal microscopy can be used in selected re-
gions not only to reveal precise cellular and subcellular colocaliza-
tion, cell morphology, dendritic spines, and growth cone morphol-
ogy of labeled axons, but also the precise topographic relationship
between intimately associated axon tracts. Confocal microscopy of
the primitive internal capsule following multiple DiI and DiA crystal
placements in embryonic rat cortex and thalamus demonstrated the
spatial relationship between the developing fiber systems (Molnár
et al., 1998a). Labeling from multiple sites of cortex also revealed
that the spatial arrangement of adjacent corticofugal and thalamo-
cortical axonal tracts was maintained in the Snap25 knockout brain
(Molnár et al., 2002). The bright intensity of DiA labeling can lead
to “bleedthrough” in filter settings detecting DiI labeling, but DiI
label is not strong enough to cause the reverse effect. Therefore, it is
recommended that DiI and DiA labeling is scanned separately dur-
ing confocal microscopy to avoid false interpretation (see chapter by
Wouterlood in this volume).
D. Additional Techniques: Combination with Immunohistochemistry,

Photoconversion of Carbocyanine Dye-Labeled Material, and
Electron Microscopic Analysis
Carbocyanine dye labeling is relatively robust, but like other fluorophores

the signal will degrade over time and when subjected to certain requisite
conditions such as illumination during epifluorescence and especially con-
focal microscopy. Furthermore, unspecific diffusion of the label from the
cut surfaces after sectioning and coverslipping will increase the background
staining. Hence, carbocyanine dye labeling should be documented as soon
as possible with both epifluorescence and confocal microscopy. To avoid
inclusion of regional photobleaching, start documentation at lower magni-
fications and increase as needed. Although tracing with carbocyanine dyes
has limitations, it is possible to convert the label into a permanent product,
which can be further utilized in combination with immunohistochemistry

and in situ hybridization.
1. Carbocyanine Dye Tracing Combined with Immunohistochemistry
Carbocyanine dyes reveal elaborate neuronal morphology and patterns

of connectivity, however, it is also desirable to determine the neurochemi-
cal properties of such cells. Standard immunohistochemical protocols use
permeabilization agents (e.g., Triton X-100) to enable antibody penetration
into this tissue, which would disrupt the cell membrane and cause leakage
of the dye. Therefore, other axonal tracers such as fluorescent beads are
often combined with tracing studies and immunohistochemistry (Voelker
et al., 2004) especially from postnatal ages upward. Antibody labeling in
DiI-labeled specimens may be obtained by prolonged incubation with the
primary antibody and avoiding the use of permeabilization agents, although
only the uppermost ∼20 µm should be used for analysis because of the lim-
ited penetration of the antibody. However, the upper ∼5 µm of the tissue
may also be unstable because of the leakage of carbocyanine dye from trun-
cated axons; hence the thickness of the tissue with reliable double staining
might be limited. The success of DiI tracing and immunofluorescence will
largely depend on the antibody used. For example, DiI crystal placement
in the internal capsule of E14.5 mice backlabeled cortical pioneer neu-
rons, which expressed the Tbrain1 protein as shown by Tbr1 immunohisto-
chemistry visualized with the Alexa 594 fluorochrome (Hevner et al., 2002).
Cytoplasmic markers are likely to be more successful in combination with
carbocyanine dyes than nuclear markers, which would require deeper pen-
etration of the antibody. Indeed, immunofluorescence for a nuclear mitotic
marker, anti-phospho-histone H3 (1:500; Upstate) in murine cortex, with
omission of Triton X-100 from the immunohistochemical protocol, leads to
uneven staining with both strong- and weak-labeled mitotic cells (Carney,
2004). Since immunohistochemistry in carbocyanine dye-traced material is
performed without the addition of detergent, a considerable amount of
false negativity can be expected.
Simultaneous carbocyanine dye tracing and immunohistochemistry
would be extremely advantageous to couple aspects of axonal connectivity
and neurochemical properties. To this end, Molecular Probes (now owned
by Invitrogen) developed a sulfonated DiI derivative (D-7777), supposedly
compatible with permeabilization procedures. However, sulfonated DiI did
not diffuse from the dorsal thalamus when implanted as a paste or as liquid
(in DMSO) in fixed embryonic rat brains (Carney and Molnár, unpublished
observations, 2004), and hence does not represent a viable alternative as a
tracer at present.
Histological sections with carbocyanine labeling can be further used
for permanent immunohistochemistry for robust antigens, such as TAG1,
L1, calbindin, or calretinin, following initial documentation of the tracing
(López-Bendito et al., 2002).
2. Photoconversion of DiI Label to a Permanent Reaction Product
For the reasons outlined above, immunofluorescence is not always com-

patible with carbocyanine dyes. Therefore, protocols have been developed
that convert the delicate fluorescent label to a permanent diaminobenzidine
(DAB) product (Lübke, 1993; Sandell and Masland, 1988). These proto-
cols permit subsequent tissue permeabilization using detergents, which
dramatically increase the efficacy of many immunohistochemistry protocols.
Following sectioning of the tissue, the labeled material is impregnated
with DAB and exposed to light through an epifluorescence microscope
using appropriate filters for visualization of the dye, e.g., a rhodamine filter
set for DiI. This initiates a redox reaction between the dye and the DAB,
creating the characteristic brown precipitate in labeled cells and neurites.
The material can subsequently be used for a plethora of techniques includ-
ing permanent immunohistochemistry, immunofluorescence, and electron
microscopy (Fujimori et al., 1997; Liu et al., 1999; Métin and Godement,
1996; Papadopoulos and Dori, 1993).
However, there are drawbacks to using this method. Due to the limited
region illuminated by a medium/high power objective lens, larger areas
must be photoconverted in a series of exposures. The technique is time-
consuming, as each field may take up to 1 h (Blakey and Molnár, unpub-
lished observations, 2005) to achieve acceptable DAB label density. Photo-
conversion of DiI-labeled thalamocortical axons in an E17 mouse brain is
shown in Fig. 12.2. Photoconversion can be done with carbocyanine dyes
Figure 12.2. Photoconversion of DiI into a permanent DAB precipitate in labeled

E17 mouse thalamocortical axons. An E17 mouse head was fixed by immersion in
4% paraformaldehyde overnight. The brain was dissected from the head and a single
DiI crystal was inserted in the ventrobasal complex of the developing thalamus, thus
filling a population of putative somatosensory axons. The brain was sectioned on a
vibroslicer (Leica VT1000S) at a thickness of 75 µm. Slices were selected for axon
labeling, and processed according to the protocol described above. As can be ob-
served in panel A, only a limited proportion of the section can be photoconverted
at any one time. However, individual fibers are filled and can be observed under
higher magnification (Panel B). This material is now suitable for further processing.
Abbreviations: CP, cortical plate; CTX, cerebral cortex; GZ, germinal zone; IZ, inter-
mediate zone; MZ, marginal zone; STR, striatum; TH, thalamus. Scale bars: 300 µm
(A); 100 µm (B).
other than DiI, using the appropriate filters. Note that to be successful, pho-
toconversion should be performed while the staining intensity is still high,
preferably as soon as possible after sectioning of the tissue. Some authors
have also observed axon breakages following photoconversion, although
this is often attributable to dehydrating procedures used subsequently to
prepare the slides (Catalano et al., 1996). Importantly, notice in the photo-
conversion protocol in the appendix of this chapter, that DAB is a suspected
carcinogen; hence appropriate measures, such as use of a fume hood and
decontamination procedures, must be employed.
3. Use of DiI-Labeled Material for Electron Microscopy
Ultrastructural investigation of labeled tissue is possible following photo-

conversion into a DAB precipitate (Lübke, 1993; Métin et al., 2000). Time-
lapse video microscopy to study in vitro growth cone movement has been
combined with electron microscopy to gain knowledge about cell form, cy-
tology, and cell–cell interactions (Fishell et al., 1995). After photoconversion
of the DiI label, sections are incubated in 1% osmium tetroxide in 100 mM
PB for 30 min. Then the sections are rinsed with PB before uranyl magne-
sium acetate staining for 30 min (0.5% UrMgAc in 0.9% saline). Following
this standard dehydration, blocking and cutting protocols are employed.
Photoconversion followed by electron microscopy can allow the differen-
tiation of directly and transneuronally labeled cells (Bruce et al., 1997). The
two types of labeling exhibit different staining intensities under fluorescent
illumination, and after photoconversion the amount and localization of the
permanent product is also distinct. Directly stained tissue shows dense DAB
precipitation in the cytosol and throughout the membranes; however, when
it is labeled transneuronally only the membrane structures are stained. Al-
though EM can reveal additional information about the labeled material,
the electron-dense product can obscure the presynaptic structures.
E. Technical Considerations and Limitations of

Carbocyanine Dye Tracing
Despite the many advantages that carbocyanine dye tracing conferred

to investigative studies elucidating patterns of neuronal connectivity, cell
migration, etc., there are some limitations and drawbacks to this technique.
1. Age and Size Considerations for Suitability of Carbocyanine Dye

Tracing in Rodent Brains
It is believed that the age of the tissue alters the incubation period
needed because of increased myelination and the greater axonal distances.
It is noticeable that after the first postnatal week carbocyanine dye tracing
becomes less reliable in rodent brains, and is not recommended after P9
(Molnár and Blakemore, 1995). Higashi et al. (2002) observed that another
family of carbocyanine dyes (used as voltage-sensitive dyes) also had a bet-
ter uptake at embryonic and early postnatal ages. However, Spires et al.
(2004) achieved excellent labeling of callosal projections in rodent at P14,
and some limited transport was even observed at adulthood. Axonal myeli-
nation might be the most obvious cause for the decreased carbocyanine
dye transport in older brains, yet no systematic study has demonstrated the
link directly. The limitations of DiI tracing in large brains are a more evi-
dent restriction in the primate brain. Strategies to overcome this problem
are raised in the section “Carbocyanine Dye Tracing in Primates, Including
Normal and Pathological Postmortem Human Specimens.”
2. Transneuronal Labeling
DiI can diffuse transneuronally especially during incubation at high tem-

peratures, and in the case of thalamocortical pathway may label radial glia
(Godement et al., 1987). This phenomenon is distinct from the transneu-
ronal labeling achieved by wgHRP or tritiated proline [3 H], since these
require active transport and uptake mechanisms. The mechanism of DiI
transneuronal transport is not fully understood; however, one model pro-
poses that where cells are in very close contact, dye molecules are able to pas-
sively diffuse from one membrane to the next (Hofmann and Bleckmann,
1999). However, the biological nature of this contact is not clear. Some
studies have shown that removing calcium ions by the addition of EDTA to
fixation and storage buffers can prevent this from occurring and improve
the sharpness of the axon fills (Hofmann and Bleckmann, 1999).
We experienced transneuronal labeling of radial glia in some fixed em-
bryonic rhesus monkey brains (see section “Carbocyanine Dye Tracing in
Primates, Including Normal and Pathological Postmortem Human Speci-
mens”), which had undergone extended incubation to encourage diffusion
of DiI to the fullest extent. The transneuronal labeling that occurred after
dorsal thalamic crystal placement, predominantly occurred at the cortico-
striatal boundary, labeling the radial glial pallisade of this region (Carney,
2004).
III. CARBOCYANINE DYE TRACING IN PRIMATES,

INCLUDING NORMAL AND PATHOLOGICAL
POSTMORTEM HUMAN SPECIMENS
A. Interspecies Comparison of Axonal Pathway Development
In this part of the review, we discuss previously employed methods for

tracing in human brains and then give examples of our work, and that of
others, in carbocyanine dye tracing studies in the primate brain.
1. Classical Tracing Methods in Human Specimens
Some of the classical studies aimed at determining axonal connectivity

used the invasive technique of applying lesions to the brain, and map-
ping fibers which derived from, or transited through, the damaged region.
This approach can also be applied in human specimens with focal lesions.
After a period, the anterograde degeneration that followed lesioning was ex-
ploited to “map” the site of the lesion with the regions where degenerating
axons were present, thus revealing topographic connections between brain
regions. Degenerating axons could be visualized using silver stains by vari-
ous protocols including the Bielschowsky, Nauta–Gygax, and Fink–Heimer
methods, which are discussed next.
Silver impregnation of degenerating axons that has been attained in
postmortem human tissues (Grafe and Leonard, 1980). Miklossy et al.
(1991) further demonstrated the selective Nauta method applicable to
tracing myelinated and unmyelinated axons in the central nervous sys-
tem of human tissue, which was fixed up to 24 h after death, and after
long-term storage (successful staining was achieved in one human brain
stored in 10% formalin for 8 years). The time course of anterograde de-
generation varies according to species, age, and fiber size. Miklossy et al.
(1991) concluded that the optimal postsurvival time for use of the selec-
tive Nauta method following degeneration of nerve fibers was between
9 days and 5 months. A modification of the Nauta method was used in
a study that described the pattern of callosal afferents in areas 17, 18,
and 19 in human cortex (Clarke and Miklossy, 1990). Also, other meth-
ods that enabled visualization of myelin breakdown products have been
used to identify anterograde degeneration of axon tracts in humans, 5–20
months after the onset of neurological symptoms, and in tissue with pro-
longed storage in formalin (Miklossy and Van der Loos, 1991). Beach and
McGeer (1988) developed a method for using HRP in the human brain
with minimal postmortem delay. HRP crystals were briefly applied to the
white matter underlying the cortex, which was then washed off and the
tissue was incubated for 1–2 days. Frozen sectioning and revelation of the
HRP product using 3,3 -diaminobenzidine demonstrated retrogradely la-
beled pyramidal cells in layers 5 and 6 of the cortex (Beach and McGeer,
1988).
Dai et al. (1998a) utilized an in vitro anterograde tracing with neu-
robiotin to reveal the hypothalamic connections of the suprachiasmatic
nucleus in postmortem human brains. It is suggested that this tech-
nique is particularly useful for tracing in the adult brain when myeli-
nation has precluded the use of carbocyanine dyes. This method has
been used in a number of publications, regarding retinohypothalamic and
intrahypothalamic projections in the human brain (Dai et al., 1998b,c).
Haber (1988) described fiber tracing in postmortem human brains using
WGA-HRP.
B. Carbocyanine Dye Tracing in Postmortem Human Tissue
DiI has been effectively used to study prenatal development of human

visual (FitzGibbon, 1997; Hevner, 2000; Qu et al., 2002) and thalamocortical
pathways (Bystron et al., 2002, 2005), development of central vagal sensory
and motor connections (Cheng et al., 2004), connections of the nucleus of
the solitary tract in the medulla oblongata (Zec and Kinney, 2003); retinal
ganglion cells (Pavlidis et al., 2003), and intrinsic connectivity of auditory
areas in adult brain (Tardif and Clarke, 2001). DiI labeling of neural con-
nectivity in the embryonic human diencephalon is shown in Fig. 12.3.
Developing connectivity of the hippocampus was investigated in fixed
brain from fetal humans at 19–22 gestational weeks (19–22 GW). At 19 GW,
DiI tracing revealed neurons from layers 2 and 3 of entorhinal cortex that
project to the hippocampus and subiculum, and reciprocal connections to
the entorhinal cortex originated from pyramidal neurons in CA1 and the
subiculum (Hevner and Kinney, 1996). In addition to determining timeta-
bles of development of axonal pathways in the nonhuman and human pri-
mate brain, DiI tracing has been used to visualize cell types. During cortico-
genesis, early generated neurons accumulate subjacent to the pial surface
and form the primordial plexiform layer (PPL) as shown in both cats and
humans (Marin-Padilla, 1971, 1983). At 6 GW, the Cajal-Retzius (CR) cells
appear in the PPL and later settle near the pial surface after the appearance
of the cortical plate. The granule cells of the subpial granular layer (SGL)
initially accumulate around the CR cells, and subsequently the two cell types
become segregated while maintaining connectivity. The fate of polymorphic
Figure 12.3. Carbocyanine dye tracing in postmortem embryonic human tissue at

6 gestational weeks. A single carbocyanine dye crystal (DiI) placed in the border
between ventral and dorsal thalamus revealed early connectivity between the dorsal
thalamus, subthalamus, and hypothalamus. (A) The crystal placement is indicated by
the asterisk. (B) High-power view from a more posterior section of the same brain.
The carbocyanine dye labeling is apparent in an extensive neural network in the
dorsal thalamus. Abbreviations: Th, thalamus; STh, subthalamus; H, hypothalamus;
OC, optic cup; Mes, mesenchyme; PPL, primordial plexiform layer; VZ, ventricular
zone. Scale bars: 200 µm (A); 50 µm (B). The study was a part of collaborative project
with Prof Blakemore and Prof Otellin.
CR and SGL cells is sealed around 24 GW when cell degeneration occurs,

although a minority of CR cells exist throughout life (Meyer and Gonzalez-
Hernandez, 1993).
DiI labeling from the pial surface in newborn ferret cortex visualized
radial glia cells. Subsequently, DiI labeling was found in astrocytes identified
with glial fibrillary acidic protein (GFAP) staining, thus demonstrating that
radial glia transforms into astrocytes (Voigt, 1989). DiI/DiA labeling was
used to recognize transitional forms of the radial glial cell transformation
to astrocytes in the human brain (deAzevedo et al., 2003).
1. Carbocyanine Dye-Tracing in Human Pathological Specimens
Notwithstanding the importance of determining axonal connectivity in

fetal human subjects without developmental anomalies, carbocyanine dye
tracing can be combined with standard histology to examine specific patho-
logical specimens. Malformations of cortical development can result in a va-
riety of clinical manifestations including epilepsy. DiI tracing was combined
with immunohistochemistry in examining brain specimens of four human
infants with various etiology in epilepsy, displaying subcortical or periven-
tricular heterotopia, to assess the connectivity of the nodules (Hannan et al.,
1999). Crystal placement in the vicinity of nodules revealed short-distance
diffusion, but the fibers did not penetrate them. DiI crystal placement within
the nodules reveals interconnectivity between them in one case of hetero-
topia, suggesting that this altered connectivity may affect neuronal matura-
tion, the balance of inhibition and excitation, and ultimate epileptogenic
potential (Hannan et al., 1999).
2. Large Brain Size in Primates May Limit the Applicability of

Carbocyanine Dyes as Effective Tracers
Using DiI tracing in fixed embryonic rhesus monkey brains, we demon-

strated that thalamocortical axons were present in the appropriate spa-
tiotemporal position to exert the mitogenic influence on cortical precursors
(Carney, 2004; Carney et al., 2002, 2003), as suggested from previous in vivo
and in vitro results in primates and rodents, respectively (Dehay et al., 1989,
2001).
In our studies of embryonic rhesus monkey brains, although carbocyanine
dye tracing was effective for examining early interactions between thalamus
and cortex, there were some drawbacks that were sometimes encountered at
later stages of development with large brain sizes. We placed DiI crystal in the
dorsal thalamus, targeting the LGN at several embryonic ages. At E40, the
onset of corticogenesis, the brain is relatively small and anterograde labeling
of thalamocortical axons from DiI crystal placements revealed growth cones
in the cortex (Carney et al., 2002, 2003). However, at later ages, the lack
of apparent growth cones on some axons indicated that the DiI had not
diffused along the entire extent of the axon. In order to obtain the furthest
diffusion possible we had incubated the brains at 37◦ C for up to 1 year
following crystal placements of DiI in the LGN, to determine the timetable
of development of geniculocortical axons to area 17. Unfortunately, the
limited diffusion of DiI meant that only a minority of axons were labeled
along the most distal part of the axons in area 17. A study of geniculocortical
axon development in the human brain did not trace from the LGN but
rather placed DiI crystals in the optic radiations of tissue blocks of human
tissue at 20–22 GW (Hevner, 2000). On the other hand, as DiI diffuses
both anterogradely and retrogradely, DiI placements in the optic radiation
would be expected to retrogradely label corticofugal axons originating from
area 17 in addition to geniculocortical axons from the LGN, and so this
approach was not feasible in our study in monkey, which focused solely on
thalamocortical projections.
3. Altering Experimental Variables to Attempt Enhancement of DiI

Diffusion in Primate Brains
Undoubtedly the application of carbocyanine dye tracing in developmen-

tal neurobiology has yielded considerable advancements in the understand-
ing of the development of axonal pathways in the embryonic brain. The
importance of effective tracing in fetal human material including patholog-
ical specimens is clear, and modifications of the carbocyanine-dye-tracing
technique have been attempted to increase diffusion from the injection site.
Different incubation times were reported by several authors ranging from 6
to 15.5 months (Krassioukov et al., 1998; Tardif and Clarke, 2001; Zec et al.,
1997). However, our observations and data of Lukas et al. (1998) suggest that
incubation times of 5 months and more are unnecessarily long. Although
neuronal tracing studies in embryonic and fetal human tissue did not fo-
cus on exact correlations between incubation times and tracing distance,
such data were reported for postmortem material from adult donors. Car-
bocyanine crystals (DiI, DiA, DiO) were applied to the cervical spinal cord,
sciatic nerve, and branchial plexus in humans and guinea pigs. This study
was carried out to optimize tracing procedures and to reveal the validity of
the combination of postmortem tracing with immunocytochemistry. Incu-
bation in the dark at 37◦ C for 12–15 weeks proved optimal to achieve the
longest tracing distances (28.9 ± 2.2 mm). Short postmortem times before
fixation proved to be important. The concentration of tracers at the appli-
cation site, longer incubation times, and incubation temperatures higher
than 37◦ C did not result in longer tracing distances.
Using the Fast DiI (1,1-dioctadecy 1-3,3,3 ,3 -tetramethylinsocarbocyanine
perchlorate) version of the carbocyanine dye in human specimens still
requires long incubation periods for dye diffusion (Mufson et al., 1990;
Zec et al., 1997). Thus, neither prolonged incubation nor the use of a
“Fast” carbocyanine dye could reach optimal diffusion in larger, primate
brains (Carney and Molnár, unpublished observations, 2003), leaving the

fixation method as an alternative parameter to be altered. Sparks et al. (2000)
reasoned that aldehyde fixation that causes protein cross-linking may ac-
tually hamper DiI diffusion and therefore attempted a “delayed-fixation”
approach. Human brain specimens of a short postmortem interval of 3 h or
less were injected with Fast DiI and covered with paper towels soaked with
phosphate buffer and incubated for 36 h at 4◦ C, prior to postfixation with
4% PFA. The authors report that such delayed fixation enabled a diffusion
distance of 20–40 mm or more, which is at least three orders of magnitude
greater than that achieved using conventional fixation protocols (Sparks
et al., 2000).
IV. EMERGING TECHNIQUES FOR NEUROANATOMICAL

TRACT IMAGING
As scientific techniques progress, current popular techniques will be used

less and less. Just as carbocyanine dye tracing relegated the need for classical
axon tracing techniques, such as those using degeneration or electrophysi-
ological stimulation at embryonic and early postnatal stages, the future may
lie in magnetic resonance imaging (MRI) studies. White matter fiber tracts
of the cerebrum contain densely packed myelinated axons along which the
movement of water molecules occurs preferentially in directions that are
perpendicular to the longitudinal axis of the axon. A major advantage of
MRI studies is that it can be used in living animals and human subjects,
repeatedly, hence permitting follow-up assessments, which are particularly
information after brain injury. Pautler et al. (1998) showed that the olfactory
and visual tracts in mice could be imaged in vivo using an MRI-visible con-
trast agent, manganese chloride (MnCl2 ) solution, by topical application
of MnCl2 solution to the naris and intravitreal injection to target retinal
ganglion cells.
Enhancement of the MRI technique, diffusion tensor MRI (DT-MRI) per-
mits the visualization of white matter fiber tract in human subjects. Using
probabilistic diffusion tractography (Behrens et al., 2003b), the human tha-
lamus could be segmented on the basis of its connectivity to the cortex
(Behrens et al., 2003a). The resulting subdivisions correspond to groups of
nuclei and this approach has been used to produce a probabilistic atlas of
the human thalamus based on its cortical connectivity (Johansen-Berg et al.,
2004). At present, however, carbocyanine dye tracing provides an unparal-
leled method to study developing brain circuitry. As MRI matures, its utility
in research will become increasingly valuable.
V. CONCLUDING REMARKS
Carbocyanine dye tracing has several advantages over classical tracing

techniques, although the drawbacks discussed above should be considered
to determine the suitability of this method for the question at hand.
Throughout this chapter, we refer to several examples where carbocya-
nine dye tracing has been used in the field of developmental neurobiol-
ogy to enhance our understanding of the anatomical complexity of brain
development. Undoubtedly the most enticing feature of carbocyanine dyes
is the fact that they can be used in fixed tissue. This is especially impor-
tant for neuronal tract-tracing of developing systems during the embryonic
period. The use of fixed tissue eliminates the need for complex in utero
surgery. Furthermore, direct knowledge of aspects of human nervous sys-
tem development has come from carbocyanine dye tracing in fixed human
tissue. This is particularly useful in pathological specimens. Simultaneous
application of more than one carbocyanine dye has yielded significant ad-
vances in discovering the interactions between fiber systems and examining
their precise topography in the developing brain. However, all methods of
neuronal tract-tracing have their criticisms and carbocyanine dye tracing
is no exception. Carbocyanine dyes label all axons and cells at the implan-
tation site and cannot discriminate between anterograde and retrograde
labeling. Another concern is that despite long incubations periods, the lim-
ited diffusion distance of carbocyanine dyes means that the entire axon
may not be labeled in larger brains especially in later stages. Using this
technique places a time constraint for data analysis as the fluorescence
starts to fade a few days after the material is processed. Other methods
such as photoconversion provide a means of turning carbocyanine dye flu-
orescence into a permanent product in small areas of tissue, but this tech-
nique introduces elements where safety precautions are needed. Nonethe-
less, standard carbocyanine dye tracing is a nonhazardous technique, which
can be easily applied by researchers with no prior experience with this
method.
We hope that this review puts carbocyanine dye tracing into the context
of classical methodologies that were used to map axonal projections, but
which proved impractical in developing systems or in human postmortem
specimens. The advent of emerging imaging techniques, which have not
reached sufficient resolution, may in turn dominate future experimental
approaches. We hope that the true potential of the carbocyanine dye tracing
method will be exploited by laboratories from various backgrounds.
APPENDIX
A. Protocol for DiI Placement and Histological Processing

(Modified from Molnár et al., 1998)
1. Materials
4% formaldehyde in phosphate buffer (PB) 0.1 M pH 7.4

PBS with sodium azide (PB 0.1 M, 0.09% NaCl, 0.05% sodium azide, pH
7.4)
Bisbenzimide (10 µg/ml in PBS)

Carbocyanine dye crystals (see Table 12.1 for details)
Platinum wires or insect pins
Fine forceps
Binocular dissecting microscope
Low melting agar (Type VII, Sigma A-4018)
Vibratome (Oxford Instruments) or Vibroslicer (Leica, VT1000S)
Slides, coverslips
Paraseal (Raymond A Lamb, London) or nail varnish
EPI fluorescence microscope with appropriate filters
2. Method
1. Fix brains with 4% formaldehyde.

2. Insert carbocyanine dye crystals with the help of an insect pin or
platinum wire.
3. Keep brains at room temperature in PBS with sodium azide or at 4◦ C
depending on the required incubation period (see Table 12.2 for
examples).
4. Embed brains in agarose (4% in PBS) and prepare blocks according
to the desired orientation for sectioning.
5. Cut sections (50–200 µm) with Vibratome (Oxford Instruments) or
Vibroslicer (Leica, VT1000S). Collect serial sections in PBS solution.
6. Counterstain sections with bisbenzimide solution (10 min in
2.5 µg/ml solution in PBS), and then rinse them with PBS. Safety
note: Bisbenzimide is a suspected carcinogen; hence gloves and appro-
priate precautions when handling or disposing of solutions must be
used.
7. Transfer the sections to slides and coverslip with PBS solution without
drying the sections. Seal coverslips with Paraseal or nail varnish.
8. Examine and document results within 3 days using the appropriate
excitatory wavelength light (see details in Table 12.1 for filters). Store
slides at 4◦ C.
B. Protocol for Photoconversion (Modified from Lübke, 1993)
1. Materials
Tris-HCl buffer 0.1 M pH 8.2 plus 0.9% NaCl (TBS)

0.05% 3,3 diaminobenzidine tetrahydrochloride in TBS (TBS–DAB)
Cavity slides
Coverslips
EPI fluorescence microscope with appropriate filters depending on the
carbocyanine dye used
2. Method
1. Rinse (2 × 10 min) sectioned material in TBS in the dark at room

temperature.
2. Incubate with TBS–DAB for 30–60 min (dark, room temperature).
Safety note: DAB is a suspected carcinogen, and so the microscope
should be placed in a fume hood and decontamination procedures
should be followed after photoconversion.
3. Transfer a section to the cavity slide, add a drop of TBS–DAB, and
coverslip.
4. Expose to appropriate excitatory wavelength light through at least a
10× objective lens. (See details in Table 12.1 for filters.)
5. Change the TBS–DAB solution every 20 min.
6. Photoconvert until optimal staining is achieved, periodically stopping
the reaction to check the progress of the conversion using bright-field
illumination.
7. Rinse the section thoroughly with TBS.
The sections are then used for further protocols, such as immunohis-
tochemistry, prepared for electron microscopy, or mounted onto slides,
dehydrated through an alcohol series, and coverslipped under DePeX
(BDH, Poole, UK).
Acknowledgments. The study in our laboratory was supported by grants

from the Medical Research Council UK; Human Frontiers Science Program
(RGP0107/2001), The European Community (Grant QLRT-1999-30158),
Wellcome Trust (Grant 063974/B/01/Z), and Swiss National Science Foun-
dation (Grant 31-56032.98 to ZM). We thank Prof. Colin Blakemore and
Ms. Patricia Cordery for their continued support. Dr. Colette Dehay and
Dr. Henry Kennedy of INSERM U371, Lyon, France, kindly provided the
monkey specimens used for tracing as part of a collaborative project. The
DiI tracing study in the human brain was a part of collaborative project with
Prof. Vladimir Otellin, Institute of Experimental Medicine, St Petersburg,
Russia. R.S.E.C. was supported by an MRC studentship and the Goodger
Scholarship from the Oxford Medical Science Division in the University of
Oxford. D.B. was supported by a Wellcome Trust 4-year PhD scholarship. IB
was sponsored, in part, by the Royal Society/Russian Academy of Science
exchange program. We are grateful to Jamin De Proto for his help in editing
and for useful discussions.
REFERENCES
Agmon, A., and Connors, B. W., 1991, Thalamocortical responses of mouse somatosensory
(barrel) cortex in vitro, Neuroscience 41:365–379.
Agmon, A., Hollrigel, G., and O’Dowd, D. K., 1996, Functional GABAergic synaptic connection
in neonatal mouse barrel cortex, J. Neurosci. 16:4684–4695.
Beach, T. G., and McGeer, E. G., 1988, Retrograde filling of pyramidal neurons in postmortem
human cerebral cortex using horseradish peroxidase, J. Neurosci. Methods 23:187–193.
Behrens, T. E., Johansen-Berg, H., Woolrich, M. W., Smith, S. M., Wheeler-Kingshott, C. A.,
Boulby, P. A., Barker, G. J., Sillery, E. L., Sheehan, K., Ciccarelli, O., Thompson, A. J.,
Brady, J. M., and Matthews, P. M., 2003a, Noninvasive mapping of connections between
Behrens, T. E., Woolrich, M. W., Jenkinson, M., Johansen-Berg, H., Nunes, R. G., Clare, S.,
Matthews, P. M., Brady, J. M., and Smith, S. M., 2003b, Characterization and propagation
of uncertainty in diffusion-weighted MR imaging, Magn. Reson. Med. 50:1077–1088.
Bicknese, A. R., Sheppard, A. M., O’Leary, D. D., and Pearlman, A. L., 1994, Thalamocortical
axons extend along a chondroitin sulfate proteoglycan-enriched pathway coincident with
the neocortical subplate and distinct from the efferent path, J. Neurosci. 14: 3500–3510.
Bruce, L. L., Christensen, M. A., and Fritzsch, B., 1997, Electron microscopic differentiation of
directly and transneuronally transported DiI and applications for studies of synaptogenesis,
Bystron, I., Molnár, Z., Otellin, V., and Blakemore, C., 2005, Tangential networks of preco-
cious neurons and early axonal outgrowth in the embryonic human forebrain, J. Neurosci.
25:2781–2792.
Bystron, I., Otellin, V., Blakemore, C., and Molnár, Z., 2002, The early development of inter-
connections between thalamus and cortex in humans. FENS, Paris, France A039.3.
Cajal, S. R., 1909, Histologie du système nerveux de l’homme et des vertebres. Reprinted by
Consejo Superior de Investigaciones Cientifias, Insituto Ramón y Cajal, 1952–1955.
Carney, R. S. E., 2004, Thalamocortical development and cell proliferation in fetal primate and rodent
cortex, D. Phil Thesis, Department of Human Anatomy and Genetics, University of Oxford,
UK.
Carney, R. S. E., Molnár, Z., Giroud, P., Cortay, V., Berland, M., Kennedy, H., and Dehay, C.,
2002, Thalamocortical projections in the developing primate cortex. FENS, Paris, France
A007.06.
Carney, R. S. E., Molnár, Z., Giroud, P., Cortay, V., Berland, M., Kennedy, H., and Dehay, C.,
2003, Novel Features of Thalamocortical Development in the Fetal Primate Cortex, P3.09 ed., UK:
British Neuroscience Association.
Catalano, S. M., Robertson, R. T., and Killackey, H. P., 1996, Individual axon morphology
and thalamocortical topography in developing rat somatosensory cortex, J. Comp. Neurol.
367:36–53.
Cheng, G., Zhou, X., Qu, J., Ashwell, K. W., and Paxinos, G., 2004, Central vagal sensory and
motor connections: human embryonic and fetal development, Auton. Neurosci. 114:83–96.
Clarke, S., and Miklossy, J., 1990, Occipital cortex in man: organization of callosal connections,
related myelo- and cytoarchitecture, and putative boundaries of functional visual areas, J.
Comp. Neurol. 298:188–214.
Dai, J., Swaab, D. F., Van der Vliet, J., and Buijs, R. M., 1998a, Postmortem tracing reveals the
organization of hypothalamic projections of the suprachiasmatic nucleus in the human
brain, J. Comp. Neurol. 400:87–102.
Dai, J., Van Der Vliet, J., Swaab, D. F., and Buijs, R. M., 1998b, Postmortem anterograde tracing
of intrahypothalamic projections of the human dorsomedial nucleus of the hypothalamus,
J. Comp. Neurol. 401:16–33.
Dai, J., Van der Vliet, J., Swaab, D. F., and Buijs, R. M., 1998c, Human retinohypothalamic tract
as revealed by in vitro postmortem tracing, J. Comp. Neurol. 397:357–370.
deAzevedo, L. C., Fallet, C., Moura-Neto, V., Daumas-Duport, C., Hedin-Pereira, C., and Lent,
R., 2003, Cortical radial glial cells in human fetuses: depth-correlated transformation into
astrocytes, J. Neurobiol. 55:288–298.
De Carlos, J. A., Lopez-Mascaraque, L., and Valverde, F., 1996, Dynamics of cell migration from
the lateral ganglionic eminence in the rat, J. Neurosci. 16:6146–6156.
Dehay, C., Horsburgh, G., Berland, M., Killackey, H., and Kennedy, H., 1989, Maturation and
connectivity of the visual cortex in monkey is altered by prenatal removal of retinal input,
Nature 337:265–267.
Dehay, C., Savatier, P., Cortay, V., and Kennedy, H., 2001, Cell-cycle kinetics of neocortical
precursors are influenced by embryonic thalamic axons, J. Neurosci. 21:201–214.
Fishell, G., Blazeski, R., Godement, P., Rivas, R., Wang, L. C., and Mason, C. A., 1995, Optical
microscopy: III. Tracking fluorescently labeled neurons in developing brain, FASEB J.
9:324–334.
Fishell, G., Mason, C. A., and Hatten, M. E., 1993, Dispersion of neural progenitors within the
germinal zones of the forebrain, Nature 362:636–638.
FitzGibbon, T., 1997, The human fetal retinal nerve fiber layer and optic nerve head: a DiI and
DiA tracing study, Vis. Neurosci. 14:433–447.
Fujimori, K. E., Takauji, R., Yoshihara, Y., Tamada, A., Mori, K., and Tamamaki, N., 1997,
A procedure for in situ hybridization combined with retrograde labeling of neurons:
application to the study of cell adhesion molecule expression in Dil-labeled rat pyramidal
neurons, J. Histochem. Cytochem. 45:455–459.
Gan, W. B., Grutzendler, J., Wong, W. T., Wong, R. O., and Lichtman, J. W., 2000, Multi-
color “DiOlistic” labeling of the nervous system using lipophilic dye combinations, Neuron
27:219–225.
Godement, P., Vanselow, J., Thanos, S., and Bonhoeffer, F., 1987, A study in developing vi-
sual systems with a new method of staining neurones and their processes in fixed tissue,
Development 101:697–713.
Grafe, M. R., and Leonard, C. M., 1980, Successful silver impregnation of degenerating axons
after long survivals in the human brain, J. Neuropathol. Exp. Neurol. 39:555–574.
Haber, S., 1988, Tracing intrinsic fiber connections in postmortem human brain with WGA-
HRP, J. Neurosci. Methods 23:15–22.
Hannan, A. J., Servotte, S., Katsnelson, A., Sisodiya, S., Blakemore, C., Squier, M., and Molnár, Z.,
1999, Characterization of nodular neuronal heterotopia in children. Brain 122(Pt 2):219–
238.
Hevner, R. F., 2000, Development of connections in the human visual system during fetal
mid-gestation: a DiI-tracing study, J. Neuropathol. Exp. Neurol. 59:385–392.
Hevner, R. F., and Kinney, H. C., 1996, Reciprocal entorhinal-hippocampal connections estab-
lished by human fetal midgestation, J. Comp. Neurol. 372:384–394.
Hevner, R. F., Miyashita-Lin, E., and Rubenstein, J. L., 2002, Cortical and thalamic axon
pathfinding defects in Tbr1, Gbx2, and Pax6 mutant mice: evidence that cortical and
thalamic axons interact and guide each other, J. Comp. Neurol. 447:8–17.
Higashi, S., Molnár, Z., Kurotani, T., and Toyama, K., 2002, Prenatal development of neural
excitation in rat thalamocortical projections studied by optical recording, Neuroscience
115:1231–1246.
Hofmann, M. H., and Bleckmann, H., 1999, Effect of temperature and calcium on transneu-
ronal diffusion of DiI in fixed brain preparations, J. Neurosci. Methods 88:27–31.
Honig, M. G., and Hume, R. I., 1986, Fluorescent carbocyanine dyes allow living neurons of
identified origin to be studied in long-term cultures, J. Cell Biol. 103:171–187.
Honig, M. G., and Hume, R. I., 1989, Dil and diO: versatile fluorescent dyes for neuronal
labelling and pathway tracing, Trends Neurosci. 12:333–335, 340–331.
Johansen-Berg, H., Behrens, T. E., Robson, M. D., Drobnjak, I., Rushworth, M. F., Brady, J. M.,
Smith, S. M., Higham, D. J., and Matthews, P. M., 2004, Changes in connectivity profiles
define functionally distinct regions in human medial frontal cortex, Proc. Natl. Acad. Sci.
USA 101:13335–13340.
Johnson, G. D., Davidson, R. S., McNamee, K. C., Russell, G., Goodwin, D., and Holborow, E.
J., 1982, Fading of immunofluorescence during microscopy: a study of the phenomenon
and its remedy, J. Immunol. Methods 55:231–242.
Krassioukov, A. V., Bygrave, M. A., Puckett, W. R., Bunge, R. P., and Rogers, K. A., 1998, Human
sympathetic preganglionic neurons and motoneurons retrogradely labelled with DiI, J.
Auton. Nerv. Syst. 70:123–128.
Liu, Q., Sanborn, K. L., Cobb, N., Raymond, P. A., and Marrs, J. A., 1999, R-cadherin ex-
pression in the developing and adult zebrafish visual system, J. Comp. Neurol. 410:303–
319.
López-Bendito, G., Chan, C. H., Mallamaci, A., Parnavelas, J., and Molnár, Z., 2002, Role of
Emx2 in the development of the reciprocal connectivity between cortex and thalamus, J.
Comp. Neurol. 451:153–169.
Lübke, J., 1993, Photoconversion of diaminobenzidine with different fluorescent neuronal
markers into a light and electron microscopic dense reaction product, Microsc. Res. Tech.
24:2–14.
Lukas, J. R., Aigner, M., Denk, M., Heinzl, H., Burian, M., and Mayr, R., 1998, Carbocyanine
postmortem neuronal tracing. Influence of different parameters on tracing distance and
combination with immunocytochemistry, J. Histochem. Cytochem. 46:901–910.
Ma, L., Harada, T., Harada, C., Romero, M., Hebert, J. M., McConnell, S. K., and Parada,
L. F., 2002, Neurotrophin-3 is required for appropriate establishment of thalamocortical
connections, Neuron 36:623–634.
Marin-Padilla, M., 1971, Early prenatal ontogenesis of the cerebral cortex (neocortex) of the
cat (Felis domestica). A Golgi study: I. The primordial neocortical organization, Z. Anat.
Entwicklungsgesch 134:117–145.
Marin-Padilla, M., 1983, Structural organization of the human cerebral cortex prior to the
appearance of the cortical plate, Anat. Embryol. 168:21–40.
McConnell, S. K., Ghosh, A., and Shatz, C. J., 1989, Subplate neurons pioneer the first axon
pathway from the cerebral cortex, Science 245:978–982.
Métin, C., Denizot, J. P., and Ropert, N., 2000, Intermediate zone cells express calcium-
permeable AMPA receptors and establish close contact with growing axons, J. Neurosci.
20:696–708.
Métin, C., and Godement, P., 1996, The ganglionic eminence may be an intermediate target
for corticofugal and thalamocortical axons, J. Neurosci. 16:3219–3235.
Meyer, G., and Gonzalez-Hernandez, T., 1993, Developmental changes in layer 1 of the human
neocortex during prenatal life: a DiI-tracing and AChE and NADPH-d histochemistry
study, J. Comp. Neurol. 338:317–336.
Miklossy, J., Clarke, S., and Van der Loos, H., 1991, The long distance effects of brain lesions:
visualization of axonal pathways and their terminations in the human brain by the Nauta
method, J. Neuropathol. Exp. Neurol. 50:595–614.
Miklossy, J., and Van der Loos, H., 1991, The long-distance effects of brain lesions: visualization
Miller, B., Chou, L., and Finlay, B. L., 1993, The early development of thalamocortical and
corticothalamic projections, J. Comp. Neurol. 335:16–41.
Molnár, Z., Adams, R., and Blakemore, C., 1998a, Mechanisms underlying the early establish-
ment of thalamocortical connections in the rat, J. Neurosci. 18:5723–5745.
Molnár, Z., Adams, R., Goffinet, A. M., and Blakemore, C., 1998b, The role of the first postmi-
totic cortical cells in the development of thalamocortical innervation in the reeler mouse,
J. Neurosci. 18:5746–5765.
Molnár, Z., and Blakemore, C., 1991, Lack of regional specificity for connections formed be-
tween thalamus and cortex in coculture, Nature 351:475–477.
Molnár, Z., and Blakemore, C., 1995, How do thalamic axons find their way to the cortex?
Trends Neurosci. 18:389–397.
Molnár, Z., and Cordery, P., 1999, Connections between cells of the internal capsule, thalamus,
and cerebral cortex in embryonic rat, J. Comp. Neurol. 413:1–25.
Molnár, Z., López-Bendito, G., Small, J., Partridge, L. D., Blakemore, C., and Wilson, M. C.,
2002, Normal development of embryonic thalamocortical connectivity in the absence of
evoked synaptic activity, J. Neurosci. 22:10313–10323.
Mufson, E. J., Brady, D. R., and Kordower, J. H., 1990, Tracing neuronal connections in post-
mortem human hippocampal complex with the carbocyanine dye DiI, Neurobiol. Aging
11:649–653.
Nauta, W. J., and Gygax, P. A., 1951, Silver impregnation of degenerating axon terminals in the
central nervous system: (1) technic. (2) chemical notes, Stain Technol. 26:5–11.
Papadopoulos, G. C., and Dori, I., 1993, DiI labeling combined with conventional immunocy-
tochemical techniques for correlated light and electron microscopic studies, J. Neurosci.
Methods 46:251–258.
Pautler, R. G., Silva, A. C., and Koretsky, A. P., 1998, In vivo neuronal tract tracing using
manganese-enhanced magnetic resonance imaging, Magn. Reson. Med. 40: 740–748.
Pavlidis, M., Stupp, T., Naskar, R., Cengiz, C., and Thanos, S., 2003, Retinal ganglion cells resis-
tant to advanced glaucoma: a postmortem study of human retinas with the carbocyanine
dye DiI, Invest. Ophthalmol. Vis. Sci. 44:5196–5205.
Qu, J., Zhou, X., Zhang, L., Ni, H., Ashwell, K., and Lu, F., 2002, A preliminary study on
development of human visual system in fetus by DiI-tracing, Zhonghua Yan Ke Za Zhi 38:517–
519.
Sandell, J. H., and Masland, R. H., 1988, Photoconversion of some fluorescent markers to a
diaminobenzidine product, J. Histochem. Cytochem. 36:555–559.
Skaliora, I., Adams, R., and Blakemore, C., 2000, Morphology and growth patterns of developing
thalamocortical axons, J. Neurosci. 20:3650–3662.
Sparks, D. L., Lue, L. F., Martin, T. A., and Rogers, J., 2000, Neural tract tracing using Di-I:
a review and a new method to make fast Di-I faster in human brain, J. Neurosci. Methods
103:3–10.
Spires, T. L., Molnár, Z., Kind, P. C., Cordery, P. M., Upton, A. L., Blakemore, C., and
Hannan, A. J., 2005, Activity-dependent regulation of synapse and dendritic spine mor-
phology in developing barrel cortex requires phospholipase C-β1 signalling, Cereb. Cortex
15(4):385–393.
Tardif, E., and Clarke, S., 2001, Intrinsic connectivity of human auditory areas: a tracing study
with DiI, Eur. J. Neurosci. 13:1045–1050.
Voelker, C. C., Garin, N., Taylor, J. S., Gahwiler, B. H., Hornung, J. P., and Molnár, Z., 2004,
Selective neurofilament (SMI-32, FNP-7 and N200) expression in subpopulations of layer
5 pyramidal neurons in vivo and in vitro, Cereb. Cortex 14:1276–1286.
Voigt, T., 1989, Development of glial cells in the cerebral wall of ferrets: direct tracing of their
transformation from radial glia into astrocytes, J. Comp. Neurol. 289:74–88.
Wouterlood, F. G., and Mugnaini, E., 1984, Cartwheel neurons of the dorsal cochlear nucleus:
a Golgi-electron microscopic study in rat, J. Comp. Neurol. 227:136–157.
Yamamoto, N., Higashi, S., and Toyama, K., 1997, Stop and branch behaviors of geniculocortical
axons: a time-lapse study in organotypic cocultures, J. Neurosci. 17:3653–3663.
Zaborszky, L., and Heimer, L., 1989, Combinatios of Tracer Techniques, Especially HRP and PHA-L,
with Transmitter Identification for Correlated Light and Electron Microscopic Studies, New York:
Plenum Press.
Zec, N., Filiano, J. J., and Kinney, H. C., 1997, Anatomic relationships of the human arcuate
nucleus of the medulla: a DiI-labeling study, J. Neuropathol. Exp. Neurol. 56:509–522.
Zec, N., and Kinney, H. C., 2003, Anatomic relationships of the human nucleus of the solitary
tract in the medulla oblongata: a DiI labeling study, Auton. Neurosci. 105:131–144.
13
Combined Fluorescence
Methods to Determine
Synapses in the Light
Microscope: Multilabel
Confocal Laser Scanning
Microscopy
FLORIS G. WOUTERLOOD
INTRODUCTION
THEORETICAL CONSIDERATIONS
Can We See Synapses? Practical Implications of Optics Theory
Pushing the Envelope: Improvements in Resolution and Image
Key Instrument Parameters in Confocal Laser Scanning
Microscopy
METHODOLOGY
Introductory
Anterograde Neuroanatomical Tracing and Follow-Up
Controls
Fluorochromes and their Characteristics
Notorious: Crosstalk
Operating the Confocal Instrument: “Operator Awareness”
Postacquisition Image Processing and 3D Reconstruction
RESULTS
CONCLUSIONS AND FUTURE
APPENDIX
Surgery, Injection of BDA in the Rat
FLORIS G. WOUTERLOOD • Department of Anatomy, Vrije Universiteit Medical

Center, Rm MF-G-136, P.O. Box 7057, 1007 MB Amsterdam, The Netherlands
394
COMBINED FLUORESCENCE METHODS 395
Perfusion-Fixation, Sectioning, Storage
Resectioning Slices into Sections to Obtain Better Penetration of
Antibodies
Preparation of Thin Sections for Free-Floating Incubation
Triple-Fluorescence Staining Procedure
Troubleshooting
REFERENCES
Abstract: The dimensions of synapses are at or below the resolution limit of classical
light microscopy. Under optimal conditions, one can appreciate processes of pre- and
postsynaptic neurons that appose each other. Such appositions may be casual only
and as such not functional in terms of synaptic communication. As a consequence,
until quite recently, electron microscopy was the only means available to determine
whether identified neurons synapse with each other. Technological developments,
however, have created a middle ground between the strictly separated realms of light
and electron microscopy. In this chapter I present a triple-fluorescence approach
aimed at identifying the apposition of a presynaptic and a postsynaptic neuron,
and simultaneously pinpointing a highly specific synapse-associated marker. This
third marker identifies the presence of an active zone, necessary to distinguish ca-
sual appositions from functional synapses. Methods involved are neuroanatomical
tracing, immunofluorescence, confocal laser scanning, and postacquisition com-
puter processing followed by three-dimensional reconstruction and inspection. In
my contribution, I will review the theory and practice involved in triple-labeling
confocal fluorescence imaging. I begin by dealing with the dimensions of synapses
and the structures involved, and relate the physical limitations of light microscopy
to the problem of resolving synaptic structure. I then review the principles of image
formation in fluorescence microscopy, and present the conditions that must be ful-
filled in order to do sound multilabel confocal laser scanning: fluorochromes, lasers,
channels, channel separation, and procedures to recognize and suppress unwanted
phenomena such as crosstalk. In order to fully illustrate the points discussed, an ac-
tual triple visualization experiment will be described. Finally, I will emphasize several
important aspects of “operator awareness”, that is, the mind setting necessary to work
with an advanced optoelectronic instrument like a confocal microscope and its so-
phisticated software. An aware user senses when some part of the complicated chain
of processes is not producing what it is supposed to produce. If operator awareness
is absent, strange results may be obtained.
Keywords: anterograde tracing, crosstalk, deconvolution, emission, excitation, fluo-
rescence, neuron markers, synapses, three-dimensional reconstruction
I. INTRODUCTION
Synapses are at the very focus of neuronal functioning. While today the
term synapse has a descriptive, morphological meaning, physiologists in-
stead of neuroanatomists introduced the term long ago to underscore the
concept of a functional juxtaposition of two neurons exchanging electri-
cal nervous activity (Foster and Sherrington, 1897). In those days of the
belle époque, neuroanatomists lacked instruments with sufficient resolution
396 FLORIS G. WOUTERLOOD
to study synapses, and the dispute between supporters of the novel neu-
ron doctrine (Waldeyer, 1891) and those entertaining the earlier reticulum
doctrine propagated by Gerlach (1858) lingered on for 50 years. The argu-
ment was finally settled in favor of the neuron doctrine after the invention
of the electron microscope and the parallel development of appropriate
preparative histologic techniques. In the early 1950s, the morphological
correlate of Foster and Sherrington’s functional synapse was revealed by
Palade and Palay (1954, 1955). In the electron microscope, the ingredients
of a typical central nervous system (CNS) synapse consist of a presynaptic
axon terminal or bouton and a postsynaptic element that may be a dendritic
spine, dendritic shaft, cell body, or even an axon hillock or axon terminal
(Fig. 13.1A). Such a site where the outer membranes of two neurons are
closely together will be referred to in this chapter as juxtaposition or appo-
sition. It is evident that in an environment with a dense packing like that
in the CNS, not all appositions can be synapses. Appositions involved in
synapses display highly specialized areas with increased electron density:
active zones. After the arrival of an action potential at a presynaptic bou-
ton, synaptic vesicles docked at the active zone in this terminal fuse to the
presynaptic membrane and release their neurotransmitter content into the
synaptic cleft. The membrane postsynaptic to the active zone hosts post-
synaptic receptors. Neurotransmitter molecules initiate, via their specific
receptor, a chain of molecular events that finally generates an excitatory
or inhibitory postsynaptic action potential. The point further exploited in
this chapter is that the molecular machinery of the synapse includes unique
proteins located pre- or postsynaptically. Excitatory and inhibitory events at
synapses require completely different molecular machineries. As a conse-
quence, if it could be possible to visualize a presynaptic axon terminal and its
juxtaposed postsynaptic element, and to immunostain simultaneously some
of the unique proteins belonging to either the excitatory or the inhibitory
kind of molecular machinery (Fig. 13.1B), it might be possible to identify
the presence of a synapse in the light microscope and to determine its neu-
rochemical role at the same time. Translated into methodology terms, we
need a triple-labeling experiment. We have successfully applied an antibody
against ProSAP2/Shank3 as the “third marker” (Wouterlood et al., 2003).
ProSAP2/Shank3 is a postsynaptic scaffolding protein involved in position-
ing the N-methyl-d-aspartate (NMDA) receptor at the postsynaptic density
of excitatory synapses (Böckers et al., 1999, 2002). For inhibitory synapses,
the protein gephyrin, i.e., a scaffolding protein for the gamma-aminobutyric
acid A (GABAA ) receptor at the postsynaptic density, has been proposed as
“third marker” (Sassoë-Pognetto and Fritschy, 2000).
The identification of synapses and their possible neurochemical role was
until recently a scientific activity confined exclusively within the domain of
the electron microscope (Sesack et al., this volume); however, electron mi-
croscopy requires fairly large investments in terms of resources, personnel,
time, laboratory equipment, and instrument. Furthermore, as symbolized
in the inset in Fig. 13.1A, the electron microscope is a sampling instrument
Figure 13.1. (A) Ingredients of a synapse: presynaptic axon terminal, pre- and postsy-
naptic membrane (“synapse”), postsynaptic element, in this case a dendritic spine.
Synapses with marked asymmetry of the membrane specializations are thought to be
excitatory. The postsynaptic density contains the molecular scaffolding machinery
of the postsynaptic receptors. (B) Concept of a synapse in a light microscopical fluo-
rescence paradigm: labeling of the presynaptic element (marker #1), labeling of the
postsynaptic element (label #2, labels #1 and 2 may be neuroanatomical tracers or
immunocytochemical markers). Labeling of a synapse-associated protein uniquely
present in the postsynaptic density provides label #3. Inset: when fluorochromes
are applied, a sandwich of fluorochromes 1, 2, and 3 will show up in the imaging
system.
par excellence. Due to its enormous resolution and associated with this the
requirement of extremely thin sections, the electron microscope is not the
instrument of choice when the purpose of the investigation is to do three-
dimensional (3D) reconstruction of large numbers of samples or to see com-
plete neurons including their synapses. The modern confocal laser scanning
microscope (CLSM), supplemented with image deconvolution and 3D re-
construction, provides just enough resolution to detect synapses, as will be
argued in the following section.
II. THEORETICAL CONSIDERATIONS
A. Can We See Synapses? Practical Implications of Optics Theory
A typical CNS axon terminal is a three-dimensionally organized varicosity

with a diameter of 0.5–1.0 µm. The active zone of such a bouton resem-
bles a disk with a diameter of 0.2–0.3 µm and a thickness of approximately
50 nm (Peters et al., 1991). Can we see such small structures in an optical
microscope? The resolution of an optical system, or the smallest distance
at which two points can be seen as separate points, is given by Ernst Abbe’s
equation
r = 0.61 λ/NAobj ,
where the parameter r or lateral resolution distance is measured in the
plane perpendicular to the optical axis (Inoué, 1995). Note that there is a
wavelength component (λ) and a component related to the quality of the
objective lens (NA, numeric aperture). This means that the wavelength of
the light used is one of the factors that determine the resolution of the
microscope. Based on this formula, with a good 40× dry objective (NA =
0.7), two points seen with blue light (λ = 450 nm) should be at least 392 nm
apart in order to be seen as separate points. With red light (λ = 600 nm),
the minimum distance becomes 522 nm, or 0.5 µm. Note that these are min-
imum theoretical distances between mathematical points. Such theoretical
distances are always smaller than those practically attainable in tissue sec-
tions. With a high-quality oil immersion lens (NA = 1.30), the theoretical
minimum distances under blue light and red light illumination become 211
and 281 nm, respectively. These numbers illustrate clearly that a synapse is
a structure whose size lingers around and below the theoretical resolution
limit of a normal light microscope. Because of this constraint, we cannot see
under normal conditions with a light microscope whether single molecules
in or around the synaptic junction belong to the presynaptic or the post-
synaptic compartment. We may see clusters of molecules if such clusters
are large enough or when we surround them with sufficient label to cre-
ate aggregates of staining agent or precipitate in the order of 0.3–0.5 µm.
Without doubt, these figures underscore the demand for high-quality ob-
jective lenses (high NA) if the aim of the microscope is to look (using bright
or fluorescence light) at very small objects like the synapses between axon
terminals and postsynaptic elements.
The above situation is further aggravated for the light microscopist by the
fact that light is subject to diffraction. This is the way it is in nature. We have
to accept that the projected image from a bright, one-dimensional point
onto our eyes, a screen, or a detector is determined by laws of quantum
physics and is seen by us and by our instruments as blurred.
Image formation in an optical system is as follows. The wave front of the
light, or the photons if light is considered from a quantum physics point of
view, distributes in a statistical fashion onto a screen or a detector, with a
so-called primary projection maximum surrounded by primary projection
minima, secondary projection maxima, secondary projection minima, and
so on (Fig. 13.2A). This diffraction pattern is named an Airy distribution,
or point spread function (PSF), if one likes to consider light as a stream
of photons. The distance between the two primary projection minima in
a two-dimensional plot equals the parameter r of Abbe’s equation. Since
the projection of a stream of photons on a screen is a spot, it is better to
refer to the diameter of the disk whose center is the primary projection
maximum while the edge is the primary projection minimum. This spot
is called the Airy disk, and its diameter equals the parameter r of Abbe’s
equation. It is important to keep in mind that the shape of the diffraction
pattern depends on the wavelength of the light involved. Green light (λ =
500 nm) has a sharper and narrower distribution curve compared with red
light (λ = 600 nm) (Fig. 13.2B). As a rule of thumb, the higher the energy of
the electromagnetic waves, the sharper the peak of the Airy distribution and
the better the resolution. Blue light has a shorter wavelength and a higher
energy than red light.
Considering two points close to each other, the outcome (r ) of Abbe’s
equation in the previous section should be considered in terms of the dis-
tance between the peaks of two partially overlapping Airy distribution curves
rather than an absolute distance between two mathematically defined, one-
dimensional points. Two distributions of photons can still be distinguished
from each other down to a minimum distance. This minimum distance is
reached when the primary projection maximum of the first Airy distribution
coincides with the first projection minimum of the second Airy distribution
(Fig. 13.2C). This minimum distance, which equals the radius of the Airy
disk (1/2 r of Abbe’s equation), is called Rayleigh’s criterion (named af-
ter Lord Rayleigh who published numerous papers on light theory, e.g., in
1891, on the behavior of light cast through a pinhole). The consequence of
these physical laws is that a microscopist desperately trying to distinguish two
blurred structures from each other by switching to a higher power lens finds
that, beyond a certain magnification, this action does not further improve
the image.
In classical optical and fluorescence microscopy with its inherent ortho-
scopic view, the microscopist typically deals with information present in one
focal plane. Diffraction is likewise measured, and resolution is expressed
Figure 13.2. Basics of diffraction. (A) Light (photons) projected onto a screen dis-
tributes according to a diffraction pattern. The distance between the primary maxi-
mum and the first diffraction minimum is called one Airy disk radius. (B) Diffraction
is wavelength dependent. The diffraction pattern of light with high energy (short
wavelength, e.g., green light) shows a narrower peak than that of light with low energy
(long wavelength, e.g., red light). A point light source using green light produces a
smaller diffraction spot than that of a point light source using red light. An object
“seen” with green light appears therefore smaller than the same object “seen” with
red light. (C) Resolution according to Rayleigh’s criterion: the smallest distance at
which two separate points are still distinguishable as separate entities. The primary
maximum of the diffraction pattern of point X coincides with the first diffraction
minimum of point Y. This distance equals one Airy radius or half the diameter of the
Airy disk. As can be inferred from Panel B, resolution depends on the wavelength
of the used light.
in only one optical plane, the XY plane. This resolution is also referred to
as the “radial resolution.” In confocal laser scanning, one typically deals
with the distribution of information in a 3D tissue volume. Accordingly, the
microscopist has to take into account the axial resolution as well, that is,
resolution measured along the optical axis or Z axis. This “axial resolution”
is lower than that in the radial direction, since the mathematical expression
for axial resolution is as follows:
r = 2λή/(NAobj )2 ,
where ή is the refractive index of the mounting medium/immersion
medium. With blue light (λ = 450 nm) and a high-quality oil immersion
lens (NA = 1.30) and using oil immersion (ή = 1.5), the theoretical mini-
mum distance between two points in the Z direction at which these points
are still distinguishable as points is 799 nm. Radial resolution in an optical
system is 392 nm (see above), and therefore, it is approximately better than
axial resolution by factor 2.
B. Pushing the Envelope: Improvements in Resolution and Image
Two improvements in optics have helped to push the limit of resolution

a factor 1.4 down from the theoretically attainable values in a normal op-
tical microscope (Inoué, 1995; Sheppard and Choudhurry, 1977). A third
improvement has increased the detail seen by the observer’s eyes and has
made optical slicing possible. Note that these improvements belong to the
category “optical and mathematical tricks” since the underlying fundamen-
tal quantum physics cannot be changed.
The first of these improvements is the use of monochromatic light, while
the second improvement is postacquisition statistical processing of the sig-
nal, called deconvolution. Deconvolution can be considered a sort of revers-
ing the statistics of an Airy distribution. There comes into spotlight the third
improvement, which is the CLSM or the sublime instrument implementing
these improvements. A laser provides a spot illumination of the object with
monochromatic light. The confocal imaging system, whose centerpiece is a
pinhole in front of its detectors, blocks haze and other out-of-focus infor-
mation discomforting to the eye (Fig. 13.3). In-focus images generated by
the laser scanning instrument are bitmaps stored on computer hard disk.
Postacquisition statistical processing, i.e., deconvolution, “sharpens” the im-
age further in a scientifically valid way.
1. Illumination with Monochromatic Light
The essence of white light is that it is a mixture of light of various wave-

lengths. As argued above, each wavelength has its own Airy distribution.
Illumination of an object via a monochromatic illumination system produces
a better image than illumination with white light, since a monochromatic
A B
Figure 13.3. The essence of a confocal imaging system is a pinhole in front of the de-
tector. (A) Fluorescence emitted from labeled structures located in the focal plane
passes the pinhole and reaches the detector. (B) Emitted light from structures lo-
cated in all planes other than the focal plane is rejected by the pinhole and does not
reach the detector.
illumination system is dealing with only one Airy distribution

instead of dealing with many. Furthermore, a lens refracts each wavelength
in a slightly different way. The result is the color shift named “chromatic
aberration.”
Although lenses are usually color corrected to reduce chromatic aberra-
tion, the only way to reduce Airy-related blur would be to improve their
numeric aperture. However, this parameter is bound by an absolute limit
(NA = 1.4). The mixing of Airy distributions associated with different wave-
lengths combined with a touch of chromatic aberration results in increased
blur of the details in the resulting image.
Excitation of a fluorochrome with a monochromatic illumination system
avoids the conventional situation in which the object is illuminated with
a cocktail of different Airy distributions (at least on the excitation side of
the system). It also avoids chromatic aberration. The result is a markedly
improved quality of the obtained image. Conventional fluorescence micro-
scopes with their mercury or xenon lamps attempt to achieve via filtering
of “excitation lines” from the lamp’s light spectrum what a laser does by its
very nature. Note that these mercury or xenon excitation lines are always
narrow bands of wavelengths and by no means single discrete wavelengths.
The fact that in addition to being monochromatic, laser light is also coher-
ent (light waves are in sync) further contributes in a positive way to image
formation. To put it simply, the truly monochromatic and coherent light
of a laser produces a result superior to that obtained with a conventional
fluorescence microscope. A laser illumination system also produces much
better color separation in dual- or multifluorescence applications. The third
advantage of a laser is the extremely small beam of high-intensity monochro-
matic light that can be used to scan a specimen with pulses of light. Note
here that the light emitted by the fluorescent specimen is not monochro-
matic nor coherent. Filtering is necessary to narrow the bandwidth of the
emitted light, especially in dual- or multiple-fluorescence applications.
2. Pinhole: Better Resolution at the Cost of Illumination
The most important optical improvement, however, offered by a confocal

laser scanning instrument in comparison with a conventional fluorescence
microscope is gained by the application of a pinhole in front of the detector.
The pinhole is a device that allows only light emitted in a focal plane to pass,
whereas emitted light originating from planes above and below the focal
plane is rejected (Fig. 13.3; Minsky, 1957; Brakenhoff et al., 1979; Inuoé,
1995). Thus, a confocal instrument possesses an intrinsic mechanism by
which out-of-focus light (the major contributor to blur) does not reach the
detector. The image looks as if it is sharper (which it is, since all information
as well as blur over and under the focal plane is absent).
The diameter of the pinhole has its own effect on resolution, be-
cause Rayleigh’s criterion also holds for projection apertures. In formula,
Rayleigh’s criterion applied to a pinhole is expressed as follows:
θ = (1.22λ)/D ,
where θ is the angular separation, λ the wavelength of the used light, and
D the diameter of the pinhole. Most important in this respect is the back-
projected pinhole, that is, the calculated diameter of the real pinhole projected
back onto the fluorescence-emitting specimen. It is this back-projected pin-
hole that really matters and not the real size of the physical pinhole. Most
manufacturers of confocal instruments refer in their documentation to this
back-projected pinhole when they present data on their instrument’s “pin-
hole.” The formula implies that the smaller the pinhole, the better the
angular separation, or resolution. Pinholes in general and especially small
pinholes reject much light. In fact so much light is blocked by the pinhole
(more than 99.99%) that the few photons that manage to pass the pinhole
cannot be seen with the naked eye and have to be detected with an ex-
pensive and ultrasensitive electronic device: a photomultiplier. Since the
emitted light from a fluorescent specimen is a fraction of the light used
for excitation, it follows that a section with a fluorescent object needs to
be literally flooded with high-intensity light in order to generate enormous
number of photons of which only a fraction will ultimately reach the photo-
multipliers. A drastic measure like saturating a specimen with high-energy
excitation light cannot be taken without dire consequences. Bleaching of
the specimen is always a major source of concern. Apart from the application
of antifading agents, a solution to bleaching is offered by the two-photon
confocal microscope; however, the description of this complex instrument
is outside the scope of the present chapter.
3. Intermediate Step: Pixelizing the Image
A photomultiplier is a photon counter and it generates an analogous sig-

nal. This signal is digitized and , in conjunction with the scanning movement
of the laser beam, used to build up a bitmapped image of the structures of
interest. These bitmaps can be further processed with a computer. The pro-
jection pattern of an image onto the photomultiplier detector is converted
into discrete samples referred to as pixels. A pixel is a square area with a
finite size and with a finite intensity level. The light intensity measured in
this square area is a gray value, usually a number between 0 and 255 (8-bit
intensity sampling) or between 0 and 4095 (12-bit intensity sampling). The
size of the pixels compared with the size of the structures to be sampled
is important. At this stage of image recording, the sampling rate accord-
ing to Nyquist comes into the spotlight (Webb and Dorey, 1995). “Nyquist”
provides a criterion inasmuch how dense sampling must be in a confocal
instrument to satisfy Rayleigh’s criterion. The Nyquist sampling rate applied
to an Airy distribution implies that sampling must occur at a rate of at least
twice the frequency of a distribution curve. In practice, four samples across
the Airy disk of a projection diffraction spot originating from a single bright
point is the minimum according to Nyquist. The publication by Webb and
Dorey (1995) discusses the details of the process of converting a projection
image into pixels.
By the application of a pinhole alone, resolution is not pushed beyond the
theoretical limit (Inuoé, 1995). It is the contrast of √the signal that is being
improved and that provides the often-mentioned 2 better “resolution”
(factor 1.4; Inuoé, 1995). In real-world terms, the theoretical minimum
distance at 450 nm illumination (blue light) to distinguish two points in the
radial direction is 280 nm, and 570 nm along the Z axis.
4. One Step Beyond Classical Resolution: Deconvolution
A real improvement of the resolution of the optical system is achieved via

the combined use of a pinhole (see section “Intermediate Step: Pixelizing
the Image”) and an additional postacquisition data processing step called
deconvolution.
Deconvolution (also called image restoration, deblurring) is, broadly
speaking, the postacquisition computational processing of a blurry or noisy
image with the purpose to obtain the very best resolution with the highest
degree of statistical confidence (Bertero et al., 1990; Holmes et al., 1995;
Snyder et al., 1992; van der Voort and Strasters, 1995). As the generation of
an Airy distribution of projected light is considered to be a “convolution”
process, the reversal of this process is called “deconvolution.” The statis-
tical nature of the generation of an Airy pattern requires that “reversed”
statistical calculations be applied in the deconvolution process. Although
deconvolution can be performed on single images, this type of processing is
in neuroscience mostly applied to Z series of confocal images. The reason is
that most biological structures extend into three dimensions where, due to
the very construction of the instrument, the X and Y directions have a res-
olution (radial resolution) different from that along the optical axis (axial
resolution, Z direction). A Z series is a number of images taken in confocal
mode. A stepping motor lifts the stage a small, controllable step between
each subsequent image. A Z series is in fact a series of images each in the
narrow focal plane, with the object moving stepwise along the Z axis through
the focal plane. We will therefore deal with a limited and specialized appli-
cation of deconvolution, notably the deconvolution of fluorescence images
in Z series.
5. Three-Dimensional Shape of the PSF in a Laser Scanning Instrument
Basic to the theory behind deconvolution is the diffraction pattern of light

as outlined earlier. Photons emitted by a point source (the light emitted by
a molecule of fluorescent marker) distribute onto a plane or a detector
according to a PSF, similar to the Airy distribution of the light wave front
in a conventional microscope. The PSF for any given microscope is a com-
pound PSF influenced by all optical components: the PSFi . Even within one
instrument, each objective lens–intermediate lenses-microscope combina-
tion has its own particular PSFi , since the numeric aperture of the objective
lens is paramount. It is important to realize that the PSFi of a confocal
instrument changes every time a different objective lens is selected.
In a routine light microscope, diffraction in the Z direction is neglected.
By contrast, in 3D reconstructions from confocal images, knowledge of the
Z component of the PSFi is very important. One would expect that the axial
component of the PSFi of a confocal instrument is the same as the radial
component. This is not the case, however, due to the factor-2 lower resolu-
tion in the axial direction versus that in the radial direction and, surprisingly,
by the presence of the pinhole. The difference between the diffraction pat-
terns in the XY and Z directions (and thus differences between the radial
and the axial components of the PSFi ) can be understood intuitively as
follows. The optical axis of a microscope is aligned with the pinhole. The
optical axis “cuts through” the entire thickness of the section. As a conse-
quence, all photons emitted by structures in the section along the path of
the optical axis pass the pinhole, even the photons that have been generated
in planes above and below the focal plane. The consequence of this photon
behavior is that the “focal plane” in a confocal microscope is not a flat plane
but a deformed plane “rippled” according to the diffraction pattern of the
back-projected pinhole. In this plane, the very area where the objective lens
performs best, the blur in the Z direction, unfortunately, is at its highest
Figure 13.4. Image formation in a confocal microscope. Images of a 100-nm-diameter

multifluorescent latex microsphere. (A) View of a confocal Zseries in XY, XZ, and YZ
directions. In the lateral direction (XY plane), the microsphere shows its true shape
whereas in the axial direction (XZ and YZ) the microsphere appears elongated.
(B) 3D reconstruction of this microsphere. The axial distortion of the sphere is
caused by the different shape of the axial component of the point spread function
(PSFi ) of a confocal microscope compared with the radial component.
and the confocality at its poorest. Fortunately, during scanning, the laser
beam coincides only very shortly with the optical axis of the instrument. As
a consequence of the presence of the pinhole, the 3D shape of the PSFi
of a confocal instrument resembles an ellipsoid rather than a sphere, with
an axial or Z component definitely elongated compared with the radial or
XY component (Hiesinger et al., 2001; Shaw, 1995). This difference between
radial and axial diffraction can easily be demonstrated by means of scanning
very small fluorescent microspheres and by 3D reconstructing these spheres
(Fig. 13.4, without postacquisition processing). In the XY plane, all micro-
spheres appear spherical, while in the XZ and YZ planes they invariably look
like mini rugby balls. The effect of the different shapes of the PSFi measured
radially versus axially is that the axial resolution of a confocal instrument is
considerably lower (factor 2–2.5) than its radial resolution.
C. Key Instrument Parameters in Confocal Laser

Scanning Microscopy
Since the theoretical optical considerations that hold for a conventional

microscope are also valid for a confocal microscope, the key parameters to
obtain high resolution in the confocal microscope are the wavelength of
the light projecting through the optical system (that is, the emitted fluo-
rescent light and not the incident light), the quality of the objective lenses
(the higher the numeric aperture, the better), and the compound PSF of
the optical system. Along with these factors, equipment like powerful and
reliable lasers, good beamsplitters, excellent filter sets, and highly sensitive
photomultipliers are a necessity. The opinions of the manufacturers differ
with respect to the size and shape of the pinhole. The back-projected size of
the pinhole should match the diameter of the Airy disk belonging to the flu-
orescence light emitted by the specimen. The main problem here is that the
emitted fluorescence falls within a spectral emission band rather than being
a fixed wavelength like the laser light used for excitation. Which emission
wavelength to select? Conventional wisdom here is to use the wavelength at
which peak emission intensity occurs.
Manufacturers are not specific about the physical shape and size of the
pinholes implemented in their instruments. The shape of the pinhole is
often determined by construction-related mechanical considerations. It may
be a square rather than a circular aperture. Often the position of the pinhole
is fixed in the Z direction (also a compromise) and, when the operator
switches to an excitation laser with a different wavelength, the pinhole does
not change its diameter. Theoretically, the diameter of the pinhole and the
distance between the pinhole and the detector plane should vary according
to the wavelength, yet this seems not to deter manufacturers from designing
confocal instruments with fixed-position pinholes. Like many instruments,
an actual confocal instrument is a compromise between theory and the
practically attainable.
Since the PSFi determines the amount of divergence of photons on their
way from the fluorescent object to the detector, an advanced deconvolu-
tion program needs to know this PSFi to do its job properly. Each confocal
instrument has its own PSFi . Although this PSFi depends primarily on the
objective lens, as argued above, construction factors play a role as well. The
PSFi at a particular magnification can be approximated via measurements
on microspheres in the actual instrument and can be used in the com-
puter program to calculate with high-statistical likelihood the origin of the
photons.
The result of deconvolution calculations is a markedly improved image.
Several deconvolution algorithms exist of which we use the Huygens II pro-
fessional software (SVI, Hilversum, The Netherlands, http://www.svi.nl). Ver-
sions of Huygens II exist for Unix, Linux, Apple, and Windows platforms.
Huygens II rapidly deconvolves Z series of images. According to Kano et al.
(1996; images obtained in a two-photon confocal instrument), an improve-
ment in resolution by two times in the XY plane can be obtained as well as
an improvement by four times in the Z direction. This could amount to a
resolving power with blue light (450 nm) of structures as small as 140 nm
in the radial direction and 285 nm in the Z direction. Since the size of a
CNS axon terminal is in the 0.5–1.0 µm range, the resolution of combined
confocal scanning-deconvolution is therefore sufficient to study for instance
colocalization of markers in nerve fibers and axon terminals. An active zone
of a synapse (200–300 nm wide, 50 nm thick), if stained with a fluorescent
marker that produces enough emission to hit the detector, will be rendered
as a bright aggregate of fluorescence.
III. METHODOLOGY
A. Introductory
With the above physicooptical constraints and possibilities in mind,

we use the confocal microscope to acquire images at high resolution.
Through subsequent postacquisition processing, we improve the resolution
via deconvolution.
We will now present the methodology used in carrying out our triple
labeling experiment. Two rules of thumb apply throughout the entire pro-
cess. First, overall performance depends on the weakest link in the chain.
Second, preacquisition histology should be perfect. The ultimate goal is
to identify the presynaptic terminal via neuroanatomical tracing (marker
#1), to identify simultaneously the postsynaptic element (marker #2), and
a protein uniquely associated with the synapse (marker #3). The outcome
is a triple immunofluorescence protocol, which we will discuss in detail in
this section. This protocol is illustrated with the projection in the rat from
the presubiculum to parvalbumin interneurons in the entorhinal cortex
(Wouterlood et al., 2003).
In the introduction, we put forward that a synapse can be represented at
the light microscopic level by a presynaptic axon terminal, a postsynaptic
structure, and by molecules uniquely attached to a synapse as the intermedi-
ate marker. The challenge is to visualize this three-marker sandwich. Given
the small dimensions of these sandwiches, we need to go beyond the classical
limit of resolution to make them visible. We approach this suboptical resolu-
tion scale by using confocal laser scanning followed up with deconvolution
and 3D reconstruction. Since “breaking the resolution barrier” occurs at the
very end of a long and rather complicated chain of histochemical, physico-
optical, and digital procedures, one should continuously keep in mind that
the quality and reliability of the final 3D reconstructed image is completely
dependent on the quality of every manipulation of the tissue sections in
all the stages preceding the actual confocal laser scanning session and, of
course, on the parameters applied during the laser scanning and postac-
quisition computer processing. The weakest link somewhere in the chain
immediately affects the resolution at the end of the chain.
The multitude of factors influencing the end result of any confocal exper-
iment can be grouped into four major clusters. As many factors as possible
will be discussed while we proceed with the methodology:
1. Preacquisition histological procedures.

2. The confocal instrument itself: lenses, filters, detectors, and parame-
ters.
3. Human factors like the skill, competence, and awareness of the person
operating the confocal instrument.
4. Postacquisition image processing.
Sloppy histology, poor understanding of the basics, and incompetent op-
eration of the instrument and computers can easily ruin the results and
cannot be offset by the most sophisticated instruments, computers, and
postacquisition computer processing.
The ultimate goal of multiple fluorescence is to observe sandwiches of
aggregates of fluorochromes. This goal can be reached only when the brain
is very well fixed such that proteins have had no chance before, during, or
after fixation to diffuse out or move away from their original position. From
this starting point, it follows that all membranes in general, and pre- and
postsynaptic membranes in particular need to be in perfect shape. The same
prerequisite of a very well fixed brain also holds for studies with the confocal
instrument in which one wants to analyze colocalization of multiple markers
in small cellular compartments such as axon terminals. On the other hand,
fixation should not be too rigid because antibodies still must be able to
penetrate into the sections in order to bind to their favorite epitopes. The
fixation conditions are comparable to those described for preembedding
electron microscopy by Leranth and Pickel (1989) in the previous issue of
this book, except that in confocal laser scanning histochemistry, the use of
detergents in the incubation media is allowed as a measure to enhance the
penetration of the antibodies into the sections.
B. Anterograde Neuroanatomical Tracing and Follow-Up
The chapter by Lanciego in this book summarizes the pros and cons of
various neuroanatomical tracers. Our experiment aims to visualize synapses,
so we need to label the presynaptic axon terminal. This is best done via
anterograde neuroanatomical tracing. A versatile anterograde tracer is bi-
otinylated dextran amine (BDA; 10 kDa, Molecular Probes, Eugene, OR).
This tracer is stable, its application relatively easy, it labels all the processes
of neurons and their appendages throughout, and the detection is fairly
straightforward with streptavidin conjugated to a fluorochrome of choice.
BDA is also highly compatible with electron microscopy (Wouterlood and
Jorritsma-Byham, 1993). An alternative anterograde tracer is the lectin Phase-
olus vulgaris leucoagglutinin (Gerfen et al., 1989; Gerfen and Sawchenko,
1984; Groenewegen and Wouterlood, 1990; Lanciego, 2005, this volume;
Zaborszky and Cullinan, 1989; Zaborszky and Heimer, 1989). The procedu-
ral steps are listed in the Appendix.
C. Controls
Controls are extremely important in multilabel fluorescence staining.

Each stage of the entire procedure requires its specific controls:
1. Immunofluorescence controls to test the specificity of the immunos-
taining.
2. Confocal instrument controls to check in a multichannel configuration

the specificity of each channel and, separately, to check laser alignment.
The laser alignment check is especially important if the purpose is to
make 3D reconstruction of small structures presumed to lie very close
to each other (markers #1 and #2 in our paradigm), or if the purpose is
to study colocalization of markers in small structures (markers #2 and
#3 in our paradigm).
1. Immunofluorescence Specificity Controls
Inherent in immunocytochemistry is the requirement to conduct suffi-

cient control experiments to determine the specificity of the binding of the
primary antibody to its epitope, the binding of the proper secondary anti-
body to the proper primary antibody, to exclude cross-reactivity, and to check
whether nonspecific binding of antibodies to tissue components occurs (see
Leranth and Pickel, 1989). In brief, control incubations should be designed
to test that the primary antibodies react only with their specific epitopes and
with no other tissue component, and to check that each of the secondary
antibodies–fluorochrome conjugates reacts only with its corresponding pri-
mary antibody and not with the noncorresponding primary or secondary
antibody. Primary antibodies can be tested with absorption controls (see
Leranth and Pickel, 1989). Various controls with fluorochrome-tagged sec-
ondary antibodies are described by Wouterlood et al. (1998). It is important
to know a priori that the immunocytochemical reactions have been success-
fully completed since otherwise no firm conclusions can be drawn from the
images acquired in the confocal instrument. A thorough discussion of the
application of fluorochromes in a multilabel experimental environment is
provided by Wessendorf (1990), although this discussion stems from the
preconfocal era. However, the principles of sound immunofluorescence
practice as outlined by Wessendorf (1990) are still valid for confocal laser
scanning configurations.
2. Confocal Instrument Controls
Central in confocal laser scanning is the concept of a channel. A channel is

a specific configuration of the confocal instrument, which includes illumi-
nation with one of the available lasers and the corresponding specific filter,
mirror, and detector settings such that the instrument is optimized to detect
the associated fluorochrome and nothing else. This is especially important
in multifluorescence confocal laser scanning. Each fluorochrome should
be visible only in its own channel and not in channels set up to detect other
fluorochromes. Signal detected in a channel associated with a different flu-
orochrome is considered crosstalk (also known as bleeding through).
Configuration of the channels of the instrument depends on the char-
acteristics of the used fluorochromes. Conversely, the selection of proper
fluorochromes is based on the laser wavelengths and filters available in the
confocal instrument.
D. Fluorochromes and their Characteristics
The lasers used in confocal instruments produce spots of very high-

intensity illumination of the section. Fluorochromes designed for use in
such a harsh environment should therefore be particularly stable under
high-intensity illumination. The classical fluorochrome fluorescein isoth-
iocyanate is unsuitable in a laser illumination environment, since this dye
is bleached away in a matter of seconds. Dyes with improved resistance to
bleaching are the carbocyanine dyes (Amersham) and the Alexa FluorTM
dyes (Molecular Probes). Bleaching can be suppressed by the addition of
an antifading agent to the mounting medium (see Longin et al., 1993; Ono
et al., 2001).
In addition to resistance to bleaching, a fluorochrome for use in a CLSM
should meet the following demands.
1. Excitation Spectrum
The shape of the excitation curve of a fluorochrome (excitation inten-

sity plotted against wavelength) should be smooth, narrow, and steep, with
an excitation maximum close to or coinciding with the wavelength of the
assigned laser light of the confocal instrument. As an example, the fluo-
rochrome Alexa FluorTM 488 (excitation maximum at 491 nm; Table 13.1)
will produce fluorescence with the highest intensity and quantum efficiency
when illuminated with a 488 nm laser. Alexa FluorTM 594 (excitation max-
imum of 590 nm) should be used in conjunction with a 594 nm laser. It
Table 13.1. Fluorochromes, their excitation peaks and the laser wavelength with
which we excite these dyes in our confocal instruments, and potential of excitation
crosstalk.
Illumination with laser Excitation crosstalk
Fluorochrome Excitation peak (nm) wavelength(s) (nm) with laser wavelength
Cy2TM 489 488 —
Cy3TM 554 543, 568 —
Cy5TM 649 633, 647 —
Alexa FluorTM 488 491 488 —
Texas Red 595 594 568
Alexa FluorTM 633 632 (shoulder at 580) 633, 647 568, 594
Alexa FluorTM 647 650 (shoulder at 580) 633, 647 568, 594
Excitation peaks as provided by the manufacturers of the respective fluorochromes.

makes little sense to view fluorescence by, say, Alexa FluorTM 594 through
illumination with a laser that produces 488 nm or even 543 nm laser light.
One may expect in those cases a low-quantum efficiency (a low intensity of
the fluorescence related to the intensity of the excitation light, i.e., little
bang for the buck and much bleaching), since the peak absorption of this
Alexa dye is too far off the fixed excitation wavelengths supplied by the 488
and 543 nm laser light.
2. Emission Spectrum
Likewise, the curve of the fluorochrome showing the intensity of the emit-
ted light plotted against the wavelength should be similarly shaped as its
(ideal) excitation curve: smooth, narrow, and steep. Especially, the “tail” of
the curve lingering toward the “red” end of the light spectrum should either
be absent or else be as low and flat as possible. If the emission curve has an
above-background spectral tail, it may cause emission crosstalk in double or
multiple laser scanning, that is, inappropriate signal in channels configured
toward the “red” end of the spectrum.
3. Resistance of Fluorochromes to Bleaching is Important

for 3D Reconstruction
Bleaching of fluorescence signal (also called quenching or fading) may

easily occur because of the very intense illumination of the fluorochrome
by its assigned laser. 3D reconstructions are made on the basis of Z series of
images. Imagine what happens if the fluorochrome offers little resistance to
bleaching. The basic fact here is that the illumination part of most laser scan-
ning microscopes is not confocal at all: the laser excites all fluorochrome
molecules throughout the entire thickness of the section and all these ex-
cited molecules emit fluorescence; only photons emitted from the focal plane are
detected. In a Z series, the region of interest (ROI) of the section will be ex-
posed to a particular amount of laser light every time an image is acquired.
If a Z series consists of n images, the ROI is exposed n times to the high-
intensity laser light. In case of weak fluorescence, the operator may decide
to scan each Z plane twice and average the result (the option frame averaging
or Kalman filtering). In such a scenario, the ROI is exposed 2n times to the
intense laser light. At the high magnification used in our experiments (63×
immersion), all the laser light is concentrated by the objective lens onto a
very small ROI. Thus, bleaching occurs faster at high magnification than at
low magnification. In a Z series subject to bleaching, the first frame of the
series will show a complete range of gray values. The second frame of the
series may look less bright and crisp, while some low-intensity gray values
will have disappeared. This reduction of brightness and contrast together
with loss of low-gray values will progressively occur with the continuation of
the Z series until at some point the acquired frame will be entirely black
(no intensity left, everything bleached). When this occurs, completion of
the Z sectioning becomes senseless and the subsequent 3D reconstruction
becomes senseless as well. Thus, the operator of the confocal instrument
always needs to be aware that bleaching is a potential danger. Countermea-
sures against bleaching are numerous. An experienced instrument operator
knows to balance these countermeasures:
1. Addition of an antifading agent to the embedding medium prior to
mounting (see Longin et al., 1993; Ono et al., 2001). There are several
good antifading agents on the market, e.g., VectashieldTM Mounting
Medium (Vector Laboratories, Burlingame, CA; www. vectorlabs.com)
and ProLong (Molecular Probes). One can also apply one’s own home-
made additive (Platt and Michael, 1983).
2. Reduction of the laser intensity and increase in the gain (sensitivity)
of the photomultipliers.
3. Reduction of the number of Z images.
4. Increase of the Z stepping increment.
5. Faster scanning (produces more noise, though).
6. Reduction of the image bitmap size.
7. No averaging of images (Kalman filtering off ).
8. Increase in the scanning frequency.
E. Notorious: Crosstalk
One of the phenomena that may interfere in a negative way with the re-
sults in multilabel fluorescence studies is crosstalk. Crosstalk (also called
“bleeding through”) is, generally speaking, the observation of inappropri-
ate fluorescence signal in a channel configured for imaging another fluo-
rochrome. Crosstalk classically occurs always in a “higher” channel, that is, a
channel configured around a laser–fluorochrome combination with longer
wavelengths. We distinguish two types of crosstalk: emission crosstalk and
excitation crosstalk.
1. Two Types of Crosstalk
Emission crosstalk is the excitation of a fluorochrome by its associ-

ated laser, and the inadvertent occurrence of some emission of this fluo-
rochrome in the next, higher channel configured for a longer wavelength
fluorochrome/laser combination. When emission crosstalk occurs, the ap-
propriate channel shows a nice image while a faint copy of the image occurs
in the inappropriate, “higher” channel. An example is excitation by a 488 nm
laser of the fluorochrome Alexa FluorTM 488, producing some emission in a
channel configured around the combination of 543 nm laser/Alexa FluorTM
546 (Figs. 13.5B and 13.6).
Figure 13.5. Diagram explaining channel separation and crosstalk in a confocal laser
scanning instrument. Dashed line = excitation curve, solid line = emission curve.
(A) Situation with ideally separated channels. For each channel, the laser excitation,
fluorochrome excitation, and emission are strictly confined to the assigned wave-
length frequency band. There is no interference between neighboring channels.
(B) Emission crosstalk in channel #2 (shaded area): emission of the fluorochrome
in channel #1 overflows in channel #2. This type of crosstalk can be avoided by
sequential scanning. (C) Excitation crosstalk: the laser in channel #1 excites fluo-
rochrome 1 but also fluorochrome 2 (emission of fluorochrome 2 in channels #1 and
#2 is shown shaded), since the excitation curve of fluorochrome 2 extends into the
wavelength frequency band of channel #1. This type of crosstalk cannot be avoided
by sequential scanning. The signal produced in channel #1 by fluorochrome 2 has
to be removed by postacquisition computer processing (so-called linear unmixing).
Excitation crosstalk is the effect in a double- (or triple-) fluorescence

experiment that a particular laser excites next to “its own” associated fluo-
rochrome also a second fluorochrome, e.g., one that belongs to the next,
“higher” channel. Inadvertent signal of that next-channel fluorochrome is
Figure 13.6. Practical examples of emission crosstalk, excitation crosstalk, and sig-
nal leakage. Section of rat hippocampal field CA1 immunostained with antibodies
against calretinin (cell 1, Alexa Fluor 594) and parvalbumin (cells 2 and 3; Alexa
Fluor 633). These markers were selected because CA1 cells express either calretinin
or parvalbumin and never both markers. Channel #1: 594 nm laser, emission band-
pass filter setting of 605–628 nm. Channel #2: 633 nm laser, emission longpass filter
setting of 643–750 nm. The detector sensitivity for both channels had been opti-
mized for its corresponding signal and was not further changed. Image pair A and
B: Situation with only the laser in channel #2 switched on. In both channels, a ghost
of the calretinin cell 1 is visible. In channel #1, this signal leakage effect is probably
due to internal reflections or by incomplete cutoff by the bandpass filter assigned
to channel #1. In channel #2, the ghost is caused by excitation and emission of the
594 fluorochrome by the 633 nm laser. The image pair in C and D was recorded
with both lasers switched on. In C, ghost images of the parvalbumin cells 2 and 3 are
visible caused by excitation crosstalk: the 594 nm laser excites the Alexa FluorTM 633,
and signal is picked up in channel #1. The ghost of the calretinin cell 1 in channel
#2 is caused by emission crosstalk or by straightaway excitation of Alexa FluorTM 594
in channel #2. All images at the same magnification.
produced in both channels. A faint copy of the image in the second channel
is produced in the first channel, in addition to the image appropriate for the
first channel. This “excitation” crosstalk signal resembles emission crosstalk,
yet has a completely different cause (Fig. 13.6). Excitation crosstalk is much
more difficult to recognize and avoid than emission crosstalk. An example
is the excitation of Alexa FluorTM 633 (emission in both 594 and 633 nm
channels) by a 594 nm laser used to excite the fluorochrome Alexa FluorTM
594 in the 594 nm channel (Figs. 13.5C and 13.6).
In addition to crosstalk phenomena, there may be internal reflections in
the confocal instrument and incomplete cutoff of bandpass filters producing
ghost images into a lower channel (“signal leakage”), e.g., signal of a “red”
fluorochrome into a “green” channel (Fig. 13.6A).
2. Procedure to Determine Emission Crosstalk
Although it is possible to determine this type of crosstalk in double- or

triple-stained sections, this check is best done with single-stained sections:
for each channel, a section stained only with the fluorochrome assigned
to that particular channel. In our laboratory, we have within reach these
single-stained control sections for each of the laser excitation wavelengths
with which our confocal instrument is equipped. In this check, we submit to
the test three single-stained sections, each associated with its own channel
of a three-channel setup.
1. Use the three single-stained sections to configure, one after another,
three channels such that on the display screen a nice and brightness-
contrast balanced image appears for each individual channel. Save
these settings.
2. Insert the single-stained section associated with the first channel in the
microscope.
3. Turn the laser intensities for the second and third channels back to
zero but do not change the detector sensitivities for these channels.
4. Illuminate the section subsequently with the laser belonging to the first
channel only (e.g., 488 nm).
5. Images remaining in the second and third channels represent emission
crosstalk.
6. Repeat this procedure for the next combination of channels.
A countermeasure against emission crosstalk is to reduce the intensity of
the laser in the first channel such that the crosstalk image in the second
channel is no longer visible (of course, the laser for the second channel
should be turned down temporarily to see the effect). Adjust, if necessary,
the detector sensitivity for the first channel. Repeat the procedure for the
next combination of channels: two and three (do not increase at this stage
the sensitivity setting of channel 2, since the current setting of that channel
has been determined in order to avoid crosstalk from channel 1). Next, con-
tinue multichannel laser scanning with these settings for laser intensities and
detector sensitivities. If these measures do not help, then sequential scan-
ning might offer a solution. Emission crosstalk can be avoided completely
by resorting to a sequential scanning procedure.
In a multichannel configuration, crosstalk usually occurs in “higher” chan-
nels, for instance, crosstalk showing up in a 543 nm channel when the laser
in the other channel is a 488 nm laser. This (emission-type) crosstalk oc-
curs because the emission spectrum of any fluorochrome is always shifted to
longer wavelengths compared with the excitation spectrum (so-called Stokes
shift) and never to shorter wavelengths (which is impossible according to the
second law of thermodynamics). The “sneaky” feature of excitation crosstalk
is that it occurs in a channel configured around a shorter laser wavelength
(a “lower” channel) than its appropriate channel. Given Stokes shift, this
sounds paradoxical at first sight. There is, however, no conflict with the sec-
ond law of thermodynamics at this point. The essence of this type of crosstalk
is that the involved fluorochrome is excited by the laser belonging to the in-
appropriate, “lower” channel and that it produces its normal, Stokes-shifted
fluorescence signal in both the inappropriate channel and the appropriate
channels.
Excitation crosstalk is much harder to detect than emission crosstalk.
Also, because most microscope operators trained as they are in recogniz-
ing Stokes shift and associated emission into a higher channel do not ex-
pect crosstalk in a “lower” channel, they are inclined to be aware of and
test only for emission crosstalk and not for excitation crosstalk. Excitation
crosstalk can be considerable when, for example, the fluorochrome Alexa
FluorTM 594 (excitation 594 nm laser) is combined with Alexa FluorTM 633
(excitation 633 nm laser). The excitation curve of the latter fluorochrome
possesses a shoulder that makes the dye sensitive to excitation at 594 nm.
As a consequence, Alexa FluorTM 633 produces signal simultaneously in
both the 594 and 633 nm channels when the section is illuminated with
594 nm laser light (Figs. 13.5C and 13.6). Similarly, excitation crosstalk oc-
curs for instance at 568 nm laser illumination with the combination Cy3TM
(568 nm excitation maximum)–Alexa FluorTM 633 (633 nm excitation max-
imum). It occurs also at 568 nm laser illumination with the combination
Cy3TM (568 nm excitation maximum)–Alexa FluorTM 647 (633 nm excitation
maximum).
3. Procedure to Determine Excitation Crosstalk
This check is done with double- or triple-stained sections, with backup

of single-stained sections. The test cannot be done with markers that are
colocalized. Prior to testing, the single-stained sections should be tested to be
sure that the individual fluorochromes do not produce emission crosstalk.
In this example, it is assumed that excitation crosstalk occurs in two adjacent

channels:
1. Configure with a double-stained section two adjacent channels such

that on the display screen a nice, brightness–contrast balanced image
appears for each channel. Save these settings.
2. Replace the double-stained section with a single-stained section con-
taining only the fluorochrome assigned to the second channel.
3. If now in the first channel an image is detected, this is due to excitation
of the second channel’s fluorochrome in the first channel. Increasing
and decreasing the laser intensity for the first channel increases and
decreases the amount of this excitation crosstalk, respectively.
4. Repeat these steps for the next combination of channels.
Note that this check is meant only to determine whether excitation

crosstalk exists in the specimen. The elimination of excitation crosstalk is
quite another story since, besides the fact that excitation crosstalk is hard
to detect, one cannot filter to eliminate this type of crosstalk. Scanning in
sequential mode offers no solution either since both fluorochromes are
excited and produce emission whenever the first laser illuminates the spec-
imen. The only way to get rid completely of excitation crosstalk is to replace
the “offending” fluorochrome with a completely different fluorochrome.
A way to intentionally reduce excitation crosstalk is via postacquisition im-
age processing by a program called “dye separation” or “linear unmixing,”
provided that the intensity of the crosstalk signal is modest.
There exist several laser-fluorochromes combinations, which are rela-
tively safe with respect to excitation crosstalk. Of course, these combinations
should match the available lasers and filters of the confocal instrument. The
bottomline is that with any combination of fluorochromes, the spectral ex-
citation curves must be as much separated from each other as possible (as
in Fig. 13.5A).
Examples of such “safe” combinations are the following:
1. Double-fluorescence labeling
1st laser Combined 2nd laser

(nm) 1st fluorochrome with (nm) 2nd fluorochrome
488 Cy2TM 568 Cy3TM

488 Alexa Fluor 488 568 Alexa FluorTM 568
488 Cy2TM 594 Alexa FluorTM 594
488 Cy2TM 594 Cy5TM
488 Cy2 TM 647 Cy5TM
488 Alexa FluorTM 488 647 Alexa FluorTM 594
488 Alexa FluorTM 488 647 Cy5TM
2. Triple-fluorescence labeling
1st laser 1st Combined 2nd laser 2nd Combined 3rd laser 3rd
(nm) fluorochrome with (nm) fluorochrome with (nm) fluorochrome
488 Cy2TM or Alexa 543 Alexa Fluor 594 Alexa

FluorTM 488 546 FluorTM 594
488 Cy2TM or Alexa 543 Alexa Fluor 647 Alexa FluorTM
FluorTM 488 546 633 or 647
488 Cy2TM or Alexa 543 Alexa Fluor 647 Cy5TM
FluorTM 488 546
4. Practical Advice to Guard Against False Results Caused by Crosstalk
1. Always be aware of crosstalk.

2. Keep single-stained sections at hand.
When a particular set of scanning parameters for multifluorescence scan-
ning has been determined such as laser intensities, detector sensitivities,
filter selection, etc., first scan single-stained sections with these parameters
and check whether an image is produced in the inappropriate “higher” or
“lower” channels.
F. Operating the Confocal Instrument: “Operator Awareness”
A CLSM is a complicated optical and digital instrument, and a thorough

understanding of what one is doing and what is happening (in physicoop-
tical terms and in terms of instrument handling and software) is necessary.
The reason is that by simply turning the controls of the instrument always
some image can be produced in which the information contained may range
from doubtful to completely worthless. The operator should always keep in
mind that images obtained should represent the real world as close as possi-
ble and that the imaging should not be disturbed by some interference such
as false positivity (e.g., channel crosstalk, sensitivity too high) or false nega-
tivity (e.g., out of focus, incorrect filter selection, sensitivity too low, fading,
insufficient penetration of fluorescent marker). As the software in new con-
focal instruments is increasingly being equipped with all sorts of automatic
functions, chances are on the rise that an unaware or inexperienced opera-
tor working “on autopilot” may program the detectors to acquire invalid or
noninformation instead of a valid series of images.
In an instrument equipped with separate lasers, the perfect alignment
of these lasers is a matter of concern. In a multiuser environment, instru-
ment awareness of the operator is necessary at this point. Alignment and/or
chromatic aberration can be tested with multifluorescent latex microspheres
(e.g., the TetraSpeck r
microspheres kit, Molecular Probes) or with sections
containing small structures multilabeled on purpose (Wouterlood et al.,
1998). Also, for the purpose of calibration, we have control brain sections
at hand containing axons labeled anterogradely with the neuroanatomical

tracer BDA and incubated with a cocktail of streptavidin conjugates: Alexa
FluorTM 488, 546, and 633. By virtue of this triple-color cocktail, the same
labeled fiber fluoresces under illumination with the 488, 543, or 633 nm
lasers. In the confocal instrument, we scan in three channels and we overlay
the acquired images. If misalignment or chromatic aberration occurs, this
will reveal itself as “pixel shift,” which is the slight deviation of the image
in one channel compared with that acquired in a different channel. Note
that in addition to radial pixel shift (in the XY plane) also axial pixel shift
may occur (in the XZ and YZ planes). We use our triple-stained calibration
sections to investigate both radial and axial pixel shifts (Fig. 13.7; see below
for further details on pixel shift and image mismatch). The advantage of
our calibration sections is that the labeled fibers are embedded in brain
tissue. The brain parenchyma surrounding the fluorescence-marked struc-
tures is by no means isotropic. This anisotropism may cause distortion of
the image. In this respect, our calibration sections are more realistic than
slides containing multifluorescent latex microspheres embedded in an ul-
trahomogeneous, isotropic mounting medium.
G. Postacquisition Image Processing and 3D Reconstruction
Image processing software is available from a range of companies. Sev-

eral manufacturers of confocal instruments offer a dye separation (linear
unmixing) package to improve spectral separation if necessary. We pre-
fer perfect signal separation at acquisition time, though, before relying on
postacquisition dye separation. We consider the latter a measure of last
resort.
Dye separation attempts, and often succeeds, in removing emission
crosstalk from the acquired images. These programs use the spectral emis-
sion characteristics of the fluorochromes. As mentioned earlier, a nasty char-
acteristic of the fluorochrome Alexa FluorTM 633, and the same holds also
for Cy5TM , is that its excitation spectrum has a shoulder at 568 nm. This im-
plies that these fluorochromes always produce fluorescence when a 568 nm
laser is used (excitation crosstalk in the 568 nm channel, cf. Fig. 13.6) and
that these fluorochromes under 568 nm laser illumination produce signal
as well in the 633 nm channel, even when the 633 nm laser is switched off.
When it is impossible in such and similar cases to use different fluorochrome
combinations, then postacquisition dye separation is indicated as a helpful
tool to remove the unwanted results of excitation crosstalk.
1. Deconvolution
As argued in the theoretical part of this chapter, the quantum physics of

image formation predicts that any image recorded with an optical instru-
ment is always blurred to some degree. The amount of blur depends on the
Figure 13.7. Pixel shift and image mismatch. Imaging of a BDA-labeled fiber stained
with a cocktail of three fluorochromes. Image series not deconvoluted. (A) Com-
posite XY, XZ, and YZ view of the image series. In the color image, shift of green
and red pixels is seen in the axial direction; there is no shift in the lateral direction.
(B) Enlarged portion of (A) 3D reconstructed (two of the three channels, XY view).
There is no lateral shift of both images indicating that radial alignment of the lasers
is perfect. (C, D) Single-channel 3D reconstructions of the image in the 488 and
633 nm channels (3D reconstruction turned 90◦ ; XZ view). (E) Merge of C and D
showing that in the axial direction, there is image mismatch. In this case, mismatch
amounts to ∼100 nm.
objective lense i.e. it is instrument specific information about how blur in a

particular optical instrument is produced. This characteristics can be used to
calculate from a blurred image via a statistical approach an image that resem-
bles the original object with the highest degree of confidence. This reversal
of the process of image formation is called deconvolution, “deblurring,” or
“image restoration” (Bertero et al., 1990; Snyder et al., 1992; van der Voort
and Strasters, 1995). Deconvolution is the final step breaking the resolution
limit barrier. Alternatively, deconvolution can be applied when an optical
A B
C D
Figure 13.8. Deconvolution of a Z series of images of a BDA-labeled fiber. (A) Com-

posite image of the Z series immediately after acquisition. (B) Same image series,
deconvoluted with Huygens II software. (C) Detail of B. (D) 3D reconstruction of
the detail in C.
signal is particularly blurry or noisy. Several deconvolution algorithms exist:

blind deconvolution, iterative deconvolution according to Miller–Tikhonov,
and iterative maximum likelihood estimate (MLE) deconvolution. MLE is
specifically advised for the deconvolution of Z series of confocal images.
The core of MLE consists of a statistical calculation that takes into account
the PSFi of the confocal instrument. The characteristics of the objective
lenses and the other optical parts of the confocal instrument used for image
acquisition, as well as the refractive indices of the tissue, immersion and
embedding media, contribute to this PSFi . The computer program uses the
PSFi to calculate, for each pixel of each acquired image in a Z series, the
statistical likelihood of the exact origin of the photons emitted by the flu-
orescent specimen. The result is a statistically reliable, improved version of
the Z image series (Fig. 13.8). The deconvolution program used in our lab-
oratory can, however, only process the Z series of images belonging to one
channel at a time, and so we have to run the program as many times as there
are channels. After deconvolution, the Z series of the respective channels
can be merged into a final multicolor image in which all immunostained
structures are visible, color-coded according to their specific microscope
channel. The increase in resolution can be as much as two times in the ra-
dial direction (XY) and four times in the axial direction (Z ) (Kano et al.,
1996; this holds for two-photon confocal images).
Note that a multicolor 3D reconstruction is in fact an overlay of as many
separate 3D reconstructions, one for every channel in the confocal instru-
ment. For instance, a three-channel confocal imaging session produces
three Z series of images. Each of these series is deconvoluted on its own and
then merged into one final 3D reconstruction. The color code assigned to
each single-channel reconstruction is conserved. Most operators of confocal
instruments adhere to the convention to render the 3D reconstruction in
the first channel in green, in the second channel in red, and that in the third
channel in blue. The color of the fluorescence emitted by the fluorochrome
is used as the associated color in the 3D reconstruction. This convention has
of course nothing to do with the real colors since all confocal images are
basically 8- or 12-bit gray scale bitmapped images.
2. Correction for Image Mismatch
Because in a multilabel experiment the images acquired in each channel

are 3D, reconstructed independently from those acquired in the other chan-
nels, small deviations of the relative positions of reconstructed objects could
go unnoticed. This so-called image mismatch is a source of both false positiv-
ity and false negativity when it comes to the detection of sandwiches of three
(independently reconstructed) markers (Wouterlood et al., 2002). Here,
the intentionally triple-stained, single tracer containing sections comes back
into focus (Fig. 13.7). The 3D reconstructions in each channel of the fibers in
these sections should exactly match. If, for example, the 3D reconstructed
image in one channel for some reason does not match with those in the
two other channels, this is evidence that somewhere in the chain of image
acquisition and processing something has gone wrong. The cause may be
laser misalignment, deviation of the scanning mirrors, chromatic aberra-
tion, operator unawareness, and so on.
In each confocal image acquisition session we scan, for the purpose of
instrument calibration, always a preparation containing intentionally triple-
labeled fibers. The 3D reconstructed images of these fibers are used to detect
image mismatch. If necessary, we can correct image mismatch by shifting the
3D reconstructions in each channel a few pixels in the appropriate X, Y , or Z
direction. The result of this exercise is that in 3D reconstruction, the perfect
overlay image of the triple-stained fibers appears like a structure painted with
three layers of paint of different color. The amount of necessary shift is the
correction factor that is next applied to the Z series of the images belonging
to the scientific experiment. One assumption underlying this correction is
that the parameters of the confocal instrument causing the misalignment

are constant during the entire image recording session. Another assumption
is that these parameters are similar for both the calibration section and the
sections belonging to the scientific experiments.
Most confocal instruments are used in a multiuser environment. It is
relevant in such an environment to conduct every now and then a check of
the instruments to see that the alignment of the lasers and other components
has not changed over time. It should also be noted that mismatch may occur
in a confocal instrument just after switching on (a “cold” instrument), and
then decrease until it becomes stable (in our instrument, a plateau is reached
30 min after powering on) (Wouterlood et al., 2002).
3. Multichannel 3D Reconstruction
There are several competing software packages on the market capable

of doing the job of calculating and rendering 3D reconstructions. In our
laboratory, we have installed two packages: FluVRTM (SVI) and AmiraTM
(www.amiravis.com), which both run on the same Silicon Graphics work-
station as the deconvolution software does. FluVRTM is a so-called volume-
rendering program, whereas AmiraTM is a surface-rendering program. Ver-
sions of both programs are available running under the Linux and Microsoft
Windows r
operating systems.
r Volume rendering: In this type of 3D rendering, a Z series of images is
considered as a rectangular box filled with layers of cubes. Each layer
corresponds with one frame of the image series. Each of the cubes (a
voxel) has an assigned gray value. The program considers the gray value
of each voxel as a measure for the absorption and emission of light by
that voxel, and it starts a simulation in which it casts light from a virtual
light source onto the space filled with voxels. Next, the program cal-
culates the amount of light absorbed and the amount of fluorescence
emitted by each voxel (depending on their gray value). A number of
parameters such as distance from the light source, direction of the light,
light intensity, transparency of voxels for excitation light, transparency
of voxels for emission light, angle of inspection (camera position), and
even reflection from the background can be modified interactively. The
result of the calculation is a simulated 3D image of the fluorescence of
all voxels of the Z series. Since all voxels including those that are “trans-
parent” are involved, this type of rendering produces scientifically most
valid 3D reconstructions. Since volume rendering includes calculations
on all voxels of the entire Z series, it is time and memory resource con-
suming. Real-time rotation of 3D structures reconstructed via volume
rendering is not possible on our computer. Multichannel volume ren-
dering is possible (up to 32 channels, SVI; personal communication).
r Surface rendering: In this type of 3D rendering, a Z series of images is
also considered as a rectangular box filled with layers of grayish cubes.
However, the operator selects a particular gray value (a threshold; in 8-
bit images, there are 256 possible gray values). The program calculates
lines connecting all voxels in the voxel space expressing the threshold
gray value and draws on screen a wireframe consisting of these isolines
(expressed as a grid of triangles). The software thus connects voxels in
all the three directions. The surfaces of the wireframes are covered with
a colored texture of choice. All further calculations are done with the
coordinates of the corners of the triangles forming the wireframe. This
saves a tremendous amount of processor resources. Surface rendering
allows real-time rotation and zooming. These features are also available
in the PC/Windows version of the program we use (AmiraTM , a high-
end graphics board in the PC is recommended). Surface rendering
has its analog in 3D computer games, where the objects or characters
are based on wireframes clad with textures. Note that, apart from the
use of RGB color images as textures, the number of colors that can be
assigned to these animated characters is virtually infinite. The number
of channels that can be rendered in surface rendering is therefore also
virtually infinite.
IV. RESULTS
After surface 3D rendering of structures containing the three markers—

presynaptic marker BDA (fluorochrome Alexa FluorTM 546, laser 543 nm),
postsynaptic marker parvalbumin (fluorochrome Alexa FluorTM 594, laser
594 nm), and synapse-associated marker ProSAP2/Shank3 (fluorochrome
Alexa FluorTM 488, laser 488)—we noted consistent spatial separation
of the reconstructed immunofluorescent material. Most reconstructed
ProSAP2/Shank3 aggregates appeared exterior to BDA-labeled fibers.
ProSAP2/Shank3 aggregates frequently resided in the interior of cell
bodies and dendrites of parvalbumin-stained cell bodies and dendrites.
In accord with the expectations (localization in the postsynaptic den-
sity), ProSAP2/Shank3 material mostly had a peripheral localization in
parvalbumin-labeled structures, that is, just below the surface envelopes
of parvalbumin-stained structures. There were no small ProSAP2/Shank3
aggregates seen deep in the interiors of parvalbumin-expressing structures
or in BDA-labeled fibers and axon terminals.
In several cases where varicosities on BDA-labeled fibers appeared
to be apposed to parvalbumin immunofluorescent dendrites, we noted
ProSAP2/Shank3 immunofluorescent material sandwiched in between,
immediately subjacent to the surface envelope of the target neuron, or
immediately next to the apposition. We regard such a sandwich as the
confocal analogon of a terminal bouton, forming a synapse with the
parvalbumin-containing cell body. An example of such an apposition with
aggregation of ProSAP2/Shank3 immunofluorescent material in a position
inside a parvalbumin cell body facing a varicosity on a BDA-labeled fiber is
shown in Fig. 13.9.
Figure 13.9. Result of triple channel confocal laser scanning, deconvolution, and 3D
reconstruction. (A) Three-channel image generated by the confocal instrument of a
Z series: ProSAP2/Shank3 (channel #1, 488 nm, label 3), BDA (channel #2, 543 nm,
label 1), and parvalbumin (channel #3, 633 nm, label 2). (B) Channels #2 and
#3 in overlay projection at higher magnification. BDA-labeled fibers stand out; the
parvalbumin labeling is weak. (C) 3D reconstruction with Amira after deconvolution,
channels merged. (D) Detail of the 3D reconstructions, showing a sandwich of the
three markers indicative for a synaptic contact between the BDA-labeled fiber and the
parvalbumin neuron. Label-2 structures (parvalbumin) rendered with a transparent
texture.
V. CONCLUSIONS AND FUTURE
The above results of observable sandwiches of fluorescence material as-

sociated with an anterograde tracer, a postsynaptic marker, and a marker
unique for a synapse lead to two conclusions. First, in this way at least appo-
sitions between processes of neurons can be made visible with great detail.
Second, the application of the third marker provides the evidence in favor
of the existence of a synapse. It is, in particular, the sandwiching that proves
that a synapse occurs between the labeled presynaptic axon terminal on
the one hand and the structure containing the postsynaptic marker on the
other. Resolution indeed seems to be sufficient to distinguish structures
with sizes in the order of 0.2–0.3 µm and to see whether there is apposition
and/or colocalization at this edge of the theoretically possible resolution.
In a separate series of experiments, we have conducted double staining, i.e.,
we labeled presynaptic fibers and axon terminals with BDA and we labeled
presumed postsynaptic dendrites via filling through pericellular application
of NeurobiotinTM . We reconstructed numerous appositions of BDA-labeled
axon terminals and dendrites containing Neurobiotin. In parallel electron
microscopy experiments, we lesioned the source area in the brain and we
studied the area with Neurobiotin-filled dendrites at high magnification un-
der the electron microscope. We indeed noted synaptic contacts between
degenerating axon terminals and containing dendrites (Wouterlood et al.,
2004).
At present available in the form of antibody against ProSAP2/Shank3 is
only a marker for NMDA-regulated synapses, i.e., glutamatergic, excitatory
synapses. Several antibodies have been developed against components of
GABAergic, inhibitory synapses, of which the GABAA receptor scaffolding
protein gephyrin (Sassoë-Pognetto and Fritschy, 2000) may be a candidate
for labeling experiments similar to those presented in this chapter. In that
case, we could in the future map on CNS neurons, the relative numbers of
excitatory and inhibitory synapses with unprecedented speed.
VI. ADVANTAGES AND LIMITATIONS
The advantages and simultaneously the disadvantages of the confocal ap-

proach as outlined in this chapter are closely associated with the ability of the
combination of confocal instrument and postacquisition processing to im-
prove resolution of an optical system down to and slightly over its very limit
as governed by the physical laws of optics and diffraction of light. Resolution
offered by conventional double-label epifluorescence microscopy is limited
due to the great depth of focus (fluorescence emitted by structures through-
out an entire section reaches the eye) and, in association with this, the low
resolving power. In fact, resolution offered by conventional fluorescence mi-
croscopy is barely sufficient to determine with confidence colocalization of
markers in neuronal cell bodies. Determination of colocalization of markers
in fibers is out of question. Resolving power is markedly improved in the
CLSM. Combination of confocal microscopy with postacquisition deconvo-
lution further improves the total resolution to a degree that colocalization
of markers in fibers and axon terminals can be determined. Furthermore,
with the addition of 3D reconstruction the observer can see rapidly and de-
cisively whether axon terminals of a particular origin or chemical signature
appose presumed postsynaptic structures, for instance processes belonging
to a particular chemical category of neurons. The 3D computer reconstruc-

tion is essential here because it enables us to look at the apposing structures
at any desired angle. By rotating and inspecting the 3D reconstruct, falsepos-
itive apposition can be distinguished from true apposition because at angles
of inspection different than the conventional, fixed orthoscopic look, spa-
tial separation of the involved structures is rapidly detectable. In addition
to this, the ProSAP2/Shank3 marker indicates whether there is a synaptic
interface between the presynaptic terminal and the postsynaptic structure.
Compared with double- and triple-labeling electron microscopic analysis,
this means an enormous advantage in speed of analysis as well as in the
number of synapses that can be inspected and counted in a given amount
of time.
One disadvantage of the present approach is that we need more, better,
and stable synapse markers, for instance a reliable marker associated with
specific proteins present at the interface in inhibitory synapses. We assume
that in the future, these markers will become available as the molecular struc-
ture of the postsynaptic density becomes better understood. A second and
more fundamental disadvantage is that determination of synapses requires
resolution at a level, which is at the edge of resolution attainable with opti-
cal systems. Note that the optical resolutions calculated in this chapter are
ideal resolutions. In the extremely heterogeneous environment offered by
brain tissue, the practical attainable resolution is always lower. Furthermore,
the physical law of diffraction predicts that at the magnification necessary
to pinpoint synapses, all stained “structures” will appear as blurry distribu-
tions of photons rather than the crisp images we are familiar with at low
magnification. At a synapse, the pre- and postsynaptic markers are by na-
ture extremely close. This, and the fluorescence associated with the synaptic
marker sandwiched in between, causes overlapping distributions of photons
in the detectors of the confocal microscope. This may cause confusion in
many interpreters. If a crisp image is really necessary, then the electron mi-
croscope with its superior resolution (measured in tenths of nanometers
instead of hundreds) is the instrument of choice. After all, electrons cannot
be beaten as vehicles for the imaging of nanostructures.
APPENDIX
A. Surgery, Injection of BDA in the Rat
1. Anesthetize the rat deeply with an intraperitoneal injection of a mix-

ture of four parts Ketaset (ketamine; 1% solution; Ket, Aesco, Boxtel,
The Netherlands) mixed with three parts Rompun (xylazine; 2% solu-
tion, Bayer, Brussels, Belgium) (1 ml/kg body weight of this mixture).
2. Mount the animal in a stereotaxic frame.
3. Expose the skull and anesthetize the periost with lidocaine (10% spray;
Astra Pharmaceutica BV, Zoetermeer, The Netherlands).
4. Drill an opening in the skull at the desired X–Y coordinates and open
the meninges.
5. Lower the tip of a borosilicate glass micropipette filled with tracer
(BDA, 5% in 10 mM phosphate buffer, pH 7.25, pipette tip diameter
10–20 mm) to the desired rostrocaudal, lateral, and vertical coordinate.
6. Apply to the micropipette a positive pulsed 5 mA DC current (7 s on/7 s
off) for 10–15 min. Leave the pipette in situ for 10 min after delivery
of the tracer.
7. Retract the pipette, close the wound, and allow the animal to recover.
The survival period postsurgery is usually 1 week.
B. Perfusion-Fixation, Sectioning, Storage
1. Inject an overdose of sodium pentobarbital (Nembutal, Ceva, Paris,

France; intraperitoneally, 60 mg/kg body weight).
2. Perfuse transcardially, first with 100 ml of physiological saline solution
of 38◦ C, pH 6.9, immediately followed by 1000 ml of 4% freshly de-
polymerized paraformaldehyde and 0.1% glutaraldehyde in 125 mM
phosphate buffer, pH 7.4 (room temperature). We use a perfusion
system driven by compressed air, which delivers perfusion fluids at a
constant, controllable hydrostatic pressure (Jonkers et al., 1984). The
thoracic aorta is clamped to ensure that all fixative is directed at the
upper part of the body.
3. Immediately after perfusion carefully remove the brain from the skull.
Cut 120-µm-thick slices with a vibrating microtome and collect these in
chilled 125 mM phosphate buffer, pH 7.4 (vials kept on melting ice).
4. Infiltrate the slices with a cryoprotectant consisting of 20% glycerin
and 2% DMSO in 125 mM phosphate buffer, pH 7.4 (Rosene et al.,
1986).
5. Transfer the slices to storage vials and place these in a freezer (−20 or
−40◦ C) for later use.
6. Resection the slices into sections prior to immunohistochemistry for
the purpose of obtaining better penetration of antibodies.
C. Resectioning Slices into Sections to Obtain Better

Penetration of Antibodies
The third label in our paradigm (see Fig. 13.1) identifies a component
uniquely associated with the synapse. We screened for this purpose many an-
tibodies directed at protein components of receptor scaffolding molecules
located in or at the postsynaptic density of the synapse. Although these an-
tibodies work well in cell cultures, the vast majority of them suffer from
insufficient penetration into brain sections of regular thickness (25–40 µm)
or even worse, they simply do not penetrate at all. An antibody that works
in our free-floating section incubation environment was raised against the
Figure 13.10. Insufficient penetration of antibodies into a section detected via Z con-
focal scanning. The marker is vesicular glutamate transporter 1, which is present in
glutamatergic axon terminals. Image taken in stratum radiatum of CA1, hippocam-
pus, 63× immersion lens NA 1.3, electronic zoom 8×. (A) Top and side views of
a Z image stack (26 frames). Inspection of the series in simultaneous XY, XZ, and
YZ rendering reveals intense staining of aggregates of immunofluorescence at the
upper and lower surfaces of the section, while a band in its core is dark, with weak or
low immunofluorescence (arrows). (B) 3D reconstruction of this Z series in XY and
YZ view. The (artifactual) absence of staining in the core of the section (arrows) is
even more dramatic after reconstruction.
protein PsoSAP2/Shank3 by Böckers et al. (1999). This protein is interpreted

as an anchoring protein of the NMDA receptor at excitatory synapses, and
it is located in the postsynaptic density (Böckers et al., 2002). However, in
pilot incubations, the penetration of this antibody appeared to be insuffi-
cient. The degree of penetration of an antibody can easily be measured in
a confocal microscope by means of a Z scan of the complete section, from
the upper to the lower surface (Fig. 13.10). A condition is that the marker
is known to be distributed homogeneously throughout the area of interest.
If in this Z scan, the distribution of the immunofluorescence occurs only
at the outer surfaces of the section, or when the intensity of staining shows
a gradient with good staining in a small superficial band at the upper and
lower surfaces of the section and poor or no staining in between (as is vis-
ible in Fig. 13.10), then it is likely that penetration has been inadequate.
Another phenomenon may lead to similar insufficient staining, notably a
very high concentration of epitope in the section. The abundancy of epitope
may lead to premature exhaustion of the antibody solution such that at a
certain point no antibody is available to bind with epitope in the center of
the section. Thus, a gradient type of immunostaining is produced similar to
the poor-penetration type. The poor penetration phenomenon is different
from bleaching caused by laser illumination since insufficient penetration
shows up at both the upper and lower surfaces of a section, whereas bleach-
ing produces a local loss of immunofluorescence throughout the section’s
thickness.
Several measures can be introduced to improve the penetration of anti-
bodies or, for that matter, to attempt avoiding gradient types of immunos-
taining:
1. Prolong the incubation.

2. Incubate at a higher temperature.
3. Increase the concentration of antibody.
4. Reduce of the number of sections per incubation well.
5. Use less and/or smaller sections (if the epitope is present in such excess
that the antiserum is fast exhausted, e.g., in case of neurotransmitters,
transporters, postsynaptic density proteins, etc.).
6. Use thinner sections or sections cut on a cryostat.
7. Treat the sections with microwaves prior to or during incubation.
8. Add (excess) detergent.
9. Freeze-thaw the sections prior to incubation. This freeze-thawing is
done by immersing the sections in a bath of isopentane that, in turn, is
subsequently rapidly cooled down by liquid nitrogen (details in Wouter-
lood et al., 1993).
As cryostat sections usually need on-slide incubation with antibodies, we

resorted in the case of ProSAP2/Shank3 to the solution of cutting the
thinnest sections possible. In order to do so, we started with 120-µm-thick
slices. Slices with such a thickness (or thicker slices, e.g., 150 µm) are easy to
cut with a vibrating microtome as well as easy to handle, manage, and store
in a cryoprotection solution in a freezer. When needed, a 120-µm slice can
be recovered from the freezer and resectioned according to the following
procedure.
D. Preparation of Thin Sections for Free-Floating Incubation
1. Drip 30% sucrose on the cold stage on a freezing microtome until a

mound is formed.
2. Flatten this mound by moving the knife of the microtome over it.
3. Take the 120 µm slice from its cryoprotectant and dip fast in 30%
sucrose.
4. Place the 120 µm slice on a (gloved) finger and put it on the flat surface
of the mound.
5. Allow to equilibrate, trim if necessary, and cut 10–15-µm-thick sections.
Collect these sections in wells of a 24-well plate for further processing.
E. Triple-Fluorescence Staining Procedure
Continuous gentle agitation on a rocking plateau is always provided dur-

ing the incubation to prevent the contents inside the wells from settling
onto the bottom.
In between all incubation steps, the sections are thoroughly rinsed with
incubation buffer: 50 mM Tris/HCl buffer with 0.875% sodium chloride,
0.5% Triton X-100, pH 8.0 (TBS-TX). We use excess of antibody solution.
Steps are as follows:
1. Preincubate 1 h at room temperature with 5% normal goat serum.
2. Incubate for at least 48 h at 4◦ C with a cocktail of primary antibodies:
mouse anti-parvalbumin (Sigma, St. Louis, MO; 1:500, marker #2) and
guinea pig anti-ProSAP2/Shank3 (1:500; marker #3; antibody kindly
supplied by Dr. Tobias Böckers, University of Freiburg, Germany).
3. Incubate for at least 24 h at 4◦ C with a cocktail consisting of strep-
tavidin conjugated to the fluorochrome Alexa FluorTM 546 (1:200,
marker #1), goat anti-guinea pig IgG conjugated to Alexa FluorTM 488
(1:100), and goat anti-mouse IgG conjugated to Alexa FluorTM 594
(1:200).
4. Rinse with Tris buffer (6.06 g/l aqua dest, pH 7.4). Mount in Tris
buffer with gelatin (0.2 g/100 ml Tris buffer, pH 7.4).
5. Dry and coverslip with DPX (Fluka Chemie AG, Buchs, Switzerland).
After coverslipping, slides are always stored in a freezer at −20◦ C. This
cold storage is intended to reduce fading of the fluorochromes over
time. Sections containing fluorescence prepared as long as 8 years
ago in our laboratory and stored at −20◦ C still contain sufficient flu-
orescence to be of good use.
6. Shrinkage of the tissue can be a problem if conservation of the 3D
shape is paramount. Since a section adheres to a solid glass surface,
drying will cause shrinkage mostly in the Z direction and to a lesser
degree in the XY direction. Deformation of shape will be inevitable.
In addition to shrinkage in the Z direction comes the reduced res-
olution in the axial direction in the confocal instrument. Shrinkage
by drying and mounting in DPX can reduce the thickness of a sec-
tion 60–75% compared with its original “wet” thickness. A measure to
reduce shrinkage and deformation is to mount and embed directly
in AquamountTM (Gurr; BDH, Poole, UK), or to apply measures dis-
cussed by Bacallao et al. (1995).
F. Troubleshooting
Insufficient penetration, crosstalk, and bleaching are the biggest prob-

lems encountered in multichannel confocal laser scanning. Insufficient
penetration can be solved by several measures as listed in section “Introduc-
tory.” Emission and excitation crosstalk can be excluded with some combi-
nations of laser excitation wavelengths and fluorochromes (see section “Flu-
orochromes and Their Characteristics”). If excitation crosstalk cannot be
avoided, then postacquisition dye separation can be attempted. Bleaching
can be suppressed by the application of antifading agents to the preparations
and, at acquisition time, by being very conservative with the intensity set-
ting of the lasers used for the excitation of the fluorochromes (see section
“Controls”). Some fluorochromes resist bleaching much more than other
fluorochromes. Remember that excessive bleaching in a Z series of images
reveals itself as a one-way gradient over the frames of decreasing crispness,
with structures becoming vague and finally merging with the background
noise. The latter situation is devastating since surface-rendered 3D recon-
struction is based on connecting voxels in the image series with correspond-
ing gray levels. Thus, with bleached sections, 3D rendering is unreliable
at least. Anticipating and preventing bleaching is therefore of vital impor-
tance. Several software solutions exist that recognize and allow correction
for bleaching in Z series. The Huygens II software used by us is equipped
with such an option.
Acknowledgments. The know-how discussed in this chapter has been col-

lected in several years of intense collaboration with a number of persons.
I am much indebted to Peter Goede, Luciënne Baks-te Bulte, and Mariska
Vonck for their extremely skilful immunohistochemical assistance. Various
confocal instruments were at our disposal by the kind collaboration of sev-
eral institutions and persons: Jan van Minnen, Faculty of Biology, Vrije Uni-
versity (Zeiss LSM 410), Jeroen Beliën, Department of Pathology, VUMC
(Leica TCS-SP), and Wolfgang Härtig and Jens Grosche, Paul-Flechsig Insti-
tute of Neuroscience, Leipzig, Germany (Zeiss LSM-510). The continuous
care by Nico Blijleven for the hardware and software with which we did the
deconvolutions and 3D reconstructions is much appreciated. Several grad-
uate students contributed as well to improvements in scanning method-
ology: Helen Pothuizen, Bas Jasperse, Ivo van den Elskamp, Michel van
den Oever, Cathrin Canto, Malika Dahmaza, Robert Schuit, and Johanna
Ramirez-Reatiga. I especially thank Amber Boekel for looking very critically
at the procedures.
REFERENCES
Bacallao, R., Kiai, K., and Jesaites, L., 1995, Guiding principles of specimen preparation for
confocal fluorescence microscopy, In: Pawley, J. B. (ed.), Handbook of Biological Confocal
Microscopy, New York: Plenum Press, pp. 311–325.
Bertero, M., Boccacci, P., Brakenhoff, G. J., Malfanti, F., and van der Voort, H. T. M., 1990, Three-
dimensional image restoration and super-resolution in fluorescence confocal microscopy,
J. Microsc. 157:3–20.
Böckers, T. M., Bockmann, J., Kreutz, M. R., and Gundelfinger, E. D., 2002, ProSAP/Shank
proteins—a family of higher order organizing molecules of the postsynaptic density with
an emerging role in human neurological disease, J. Neurochem. 81:903–910.
Böckers, T. M., Kreutz, M. R., Winter, C., Zuschratter, W., Smalla, K. H., Sanmarti-Vila, L., Wex,
H., Langnaese, K., Bockmann, J., Garner, C. C., and Gundelfinger, E. D., 1999, Proline-
rich synapse-associated protein-1/cortactin binding protein 1 (ProSAP1/CortBP1) is a
PDZ-domain protein highly enriched in the postsynaptic density, J. Neurosci. 9:6506–6518.
Brakenhoff, G. J., Blom, P., and Barends, P., 1979, Confocal scanning light microscopy with
high aperture immersion lenses. J. Microsc. 117:219–232.
Foster, M., and Sherrington, C. S., 1897, A Text Book of Physiology, Part III: The Central Nervous
System, 7th ed., London: Macmillan.
Gerfen, C. R., and Sawchenko, P. E., 1984, A method for anterograde axonal tracing of
chemically specified circuits in the central nervous system: combined Phaseolus vulgaris-
leucoagglutinin (PHA-L) tract tracing and immunohistochemistry, Brain Res. 343:144–150.
Gerfen, C. R., Sawchenko, P. E., and Carlsen, J., 1989, The PHA-L anterograde axonal tracing
method, In: Heimer, L., and Zaborszky, L. (eds.), Neuroanatomical Tract-Tracing Methods 2,
Recent Progress, New York: Plenum Press, pp. 19–48.
Gerlach, J., 1858, Mikroskopische Studien aus den Gebiete der menschlichen Morphologie, Erlangen,
Germany: Enke.
neuronal connections with Phaseolus vulgaris-leucoagglutinin (PHA-L), and combinations
with other neuroanatomical techniques, In: Björklund, A., Hökfelt, T., Wouterlood, F. G.,
and van den Pol, A. N. (eds.), Analysis of Neuronal Microcircuits and Synaptic Interactions.
Handbook of Chemical Neuroanatomy, Vol. 8, Amsterdam: Elsevier Biomedical Press, pp. 47–
124.
Hiesinger, P. R., Scholz, M., Meinertzhagen, I. A., Fischbach, K. -F., and Obermayer, K., 2001, Vi-
sualization fo synaptic markers in the optic neuropil of Drosophila using a new constrained
deconvolution method, J. Comp. Neurol. 429:277–288.
Holmes, T. J., Bhattacharyya, S., Cooper, J. A., Hanzel, D., Szarowski, D. H., and Turner, J. N.,
1995, Light microscopic images reconstructed by maximum likelihood deconvolution,
In: Pawley, J. B. (ed.), Handbook of Biological Confocal Microscopy, New York: Plenum Press,
pp. 389–402.
Inoué, S., 1995, Foundations of confocal scanned imaging in light microscopy. In: Pawley, J. B.
(ed.), Handbook of Biological Confocal Microscopy, New York: Plenum Press, pp. 1–14.
Jonkers, B., Sterk, J., and Wouterlood, F. G., 1984, Transcardial perfusion fixation of the CNS
by means of a compressed-air driven device, J. Neurosci. Methods 12:141–149.
Kano, H., van der Voort, H. T. M., Schrader, M., van Kempen, G. M. P., and Hell, S. W.,
1996, Avalanche photodiode detection with object scanning and image restoration pro-
vides 2–4 fold resolution increase in two-photon fluorescence microscopy, Bioimaging 4:
187–197.
Leranth, C., and Pickel, V. M., 1989, Electron microscopic preembedding double immunos-
taining methods, In: Heimer, L., and Zaborszky, L. (eds.), Neuroanatomical Tract-Tracing
Methods 2, Recent Progress, New York: Plenum Press, pp. 129–172.
Longin, A., Souchier, C., Ffrench, M., and Bryon, P. A., 1993, Comparison of anti-fading agents
used in fluorescence microscopy: image analysis and laser confocal microscopy study, J.
Histochem. Cytochem. 41:1833–1840.
Minsky, M., 1957, US patent no. 301467, Microscopy Apparatus.
Ono, M., Murakami, T., Kudo, A., Isshiki, M., Sawada, H., and Segawa, A., 2001, Quantitative
comparison of anti-fading mounting media for confocal laser scanning microscopy, J.
Histochem. Cytochem. 49:305–311.
Palade, G. E., and Palay, S. L., 1954, Electron microscope observations of interneuronal and
neuromuscular synapses, Anat Rec. 118:335–336.
Palay, S. L., and Palade, G. E., 1955, The fine structure of neurons, J. Biophys. Biochem. Cytol.
1:69–88.
Peters, A., Palay, S. L., and de Webster, F. H., 1991, The Fine Structure of the Nervous System: Neurons
and Their Supporting Cells, 2nd ed, Oxford: Oxford University Press, 494 pp.
Platt, J. L., and Michael, A. F., 1983, Retardation of fading and enhancement of intensity of
immunofluorescence by p-phenylendiamine, J. Histochem. Cytochem. 31:840–842.
Rayleigh, L., and Strutt, J. W., 1891, On pin-hole photography, Philos. Mag. 11:87–99.
Rosene, D. L., Roy, N. J., and Davis, B. J., 1986, A cryoprotection method that facilitates cutting
frozen sections of whole monkey brains for histological and histochemical processing
Sassoë-Pognetto, M., and Fritschy, J. -M., 2000, Gephyrin, a major postsynaptic protein of
GABAergic synapses, Eur. J. Neurosci. 12:2205–2210.
Shaw, P. J., 1995, Comparison of wide-field/deconvolution and confocal microscopy for 3D
imaging, In: Pawley, J. B. (ed.), Handbook of Biological Confocal Microscopy, New York: Plenum
Press, pp. 373–387.
Sheppard, C. J. R., and Choudhurry, A., 1977, Image formation in the scanning microscope,
Opt. Acta 24:1051–1073.
Snyder, D. L., Schulz, T. J., and O’Sullivan, J. A., 1992, Deblurring subject to nonnegativity
constraints, IEEE Trans. Sign. Proc. 40:1143–1150.
van der Voort, H. T. M., and Strasters, K. C., 1995, Restoration of confocal images for quantitative
image analysis, J. Microsc. 158:43–45.
Waldeyer, F., 1891, Über einige neuere Forschungen im Gebiete der Anatomie des Central-
nervensystems, Dtsch. Med. Wochenschr. 17:1213–1218, 1244–1246, 1267–1269, 1287–1289,
1331–1332, 1352–1356.
Webb, R. H., and Dorey, C. K., 1995, The pixilated image, In: Pawley, J. B. (ed.), Handbook of
Biological Confocal Microscopy, New York: Plenum Press, pp. 55–67.
Wessendorf, M. W., 1990, Characterization and use of multi-color fluorescence microscopic
techniques, In: Björklund, A., Hökfelt, T., Wouterlood, F. G., and van den Pol, A. N.
(eds.), Handbook of Chemical Neuroanatomy: Analysis of Neuronal Microcircuits and Synaptic
Interactions, Vol. 8. Amsterdam: Elsevier, pp. 1–46.
Wouterlood, F. G., Böckers, T., and Witter, M. P., 2003, Synaptic contacts between identified
neurons visualized in the confocal laser scanning microscope. Neuroanatomical tracing
combined with immunofluorescence detection of postsynaptic density proteins and target
neuron-markers, J. Neurosci. Methods 128:129–142.
Wouterlood, F. G., and Jorritsma-Byham, B., 1993, The anterograde tracer biotinylated dextran-
amine: comparison with the tracer Phaseolus vulgaris-leucoagglutnin in preparations for
electron microscopy, J. Neurosci. Methods 48:75–88.
Wouterlood, F. G., Pattiselanno, A., Jorritsma-Byham, B., Arts M. P. M., and Meredith G. E.,
1993, Connectional, immunocytochemical and ultrastructural characterization of neurons
injected intracellularly in fixed brain tissue, In: Meredith, G. E., and Arbuthnott, G. W.
(eds.), Morphological Investigations of Single Neurons In Vitro. IBRO Handbook Series “Methods
in the Neurosciences, No. 16, Chichester, UK: Wiley and Sons, pp. 47–74.
Wouterlood, F. G., van Denderen, J. C. M., Blijleven, N., van Minnen, J., and Härtig, W., 1998,
Two-laser dual-immunofluorescence confocal laserscanning microscopy using Cy2- and
Cy5-conjugated secondary antibodies: unequivocal detection of co-localization of neu-
ronal markers, Brain Res. Protoc. 2:149–159.
Wouterlood, F. G., van Haeften, T., Blijleven, N., Perez-Templado, P., and Perez-Templado, E.,
2002, Double-label confocal laserscanning microscopy, image restoration and real-time
3D-reconstruction to study axons in the CNS and their contacts with target neurons, Appl.
Immunohistochem. Mol. Morphol. 10:85–102.
Wouterlood, F. G., van Haeften, T., Eijkhoudt, M., Baks-te-Bulte, L., Goede, P. H., and Witter,
M. P., 2004, Input from the presubiculum to dendrites of layer-V neurons of the medial
entorhinal cortex of the rat, Brain Res. 1013:1–12.
Zaborszky, L., and Cullinan, W. E., 1989, Hypothalamic axons terminate on forebrain cholin-
ergic neurons: an ultrastructural double-labeling study using PHA-L tracing and ChAT
immunocytochemistry, Brain Res. 479:177–184.
Zaborszky, L., and Heimer, L., 1989, Combination of tracer technqiues, especially HRP and
PHA-L with transmitter identification for correlated light and electron microscopic studies,
In: Heimer, L., and Zaborszky, L. (eds.), Neuroanatomical Tract-Tracing Methods 2, Recent
Progress, New York: Plenum Press, pp. 49–96.
14
Advances in Understanding
Cortical Function Through
Combined Voltage-Sensitive
Dye Imaging, Whole-Cell
Recordings, and Analysis of
Cellular Morphology
CARL C. H. PETERSEN
INTRODUCTION
VOLTAGE-SENSITIVE DYE IMAGING
WHOLE-CELL PATCH-CLAMP RECORDING
AN APPLICATION: ANALYZING THE SENSORY RESPONSE IN
RODENT BARREL CORTEX
APPENDIX
Commercial Sources of Voltage-Sensitive Dye, Camera,
and Imaging Software
Staining Neocortex with Voltage-Sensitive Dye
Voltage-Sensitive Dye Imaging
Whole-Cell Recording
Anatomical Analysis of Neuronal Structure and Position
REFERENCES
Abstract: Voltage-sensitive dyes can be used to image cortical network function with
millisecond temporal resolution and with a horizontal spatial resolution of approxi-
mately 50 µm. This imaging technique can be combined with whole-cell patch-clamp
CARL C. H. PETERSEN • Laboratory of Sensory Processing, Brain Mind Institute,

SV-BMI-LSENS AAB 105, Ecole Polytechnique Federale de Lausanne, CH-1015 Lausanne,
Switzerland
436
VOLTAGE-SENSITIVE DYE IMAGING AND WHOLE-CELL RECORDINGS 437
measurement of membrane potential followed by the anatomical analysis of neu-
ronal morphology. Together, such experiments reveal the relationship of activity
recorded in individual identified neurons with the spatiotemporally resolved en-
semble dynamics of a cortical region. Application of these techniques to the rodent
barrel cortex has advanced our understanding of the synaptic mechanisms underly-
ing sensory responses to simple whisker stimuli.
Keywords: axonal and dendritic morphology, sensory processing, voltage-sensitive
dye imaging, whole-cell patch-clamp recording
I. INTRODUCTION
Understanding the organizing principles and functions of the neocortex

is likely to lead to insight concerning the mechanisms of sensory perception
and behavior. Such complex mental processes are likely to result from the in-
teractions of many neurons distributed throughout cortical and subcortical
areas. A variety of different approaches have been taken to analyze large-
scale brain activity. Electrophysiological techniques were among the first to
be applied to analyze brain function. Recent technological advances in elec-
trophysiology allow simultaneous extracellular recordings of action poten-
tial activity from hundreds of individual neurons located in multiple brain
areas in awake behaving animals (Buzsaki, 2004; Nicolelis et al., 2003). Such
measurements are, however, limited to an analysis of suprathreshold spik-
ing activity of neurons and give no information relating to the underlying
mechanisms leading to the action potentials. More detailed measurements
of neuronal function have been made using intracellular recordings either
using sharp microelectrodes or by the patch-clamp technique (Hamill et al.,
1981; Neher, 1992; Sakmann, 1992). These intracellular electrophysiologi-
cal techniques can record the neuronal membrane potential capturing both
the subthreshold and the suprathreshold responses of neurons. However,
rather few cells (usually only one or two cells) can be recorded simultane-
ously due to technical difficulties. Electrophysiological measurements offer
excellent temporal resolution, but are limited in their ability to map the
spatial extent of brain activity (even with over a hundred extracellular elec-
trodes).
Imaging methods have been developed to give spatial information re-
garding brain activity. Spectacular images of brain function have come from
functional magnetic resonance imaging; however, the temporal resolution
of this technique is limited. Most current functional magnetic resonance
imaging techniques focus on the blood oxygenation level-dependent signal
and thus temporal resolution is limited not only by the measuring appa-
ratus but also by the delayed coupling of neural activity to hemodynamic
changes.
An obvious wish would be to directly image the electrical activity of
neurons in the brain. Recent advances in both imaging technology and
novel voltage-sensitive dyes now offer the opportunity for high temporal
438 CARL C. H. PETERSEN
and spatial resolution recording of neocortical activity. Below, I describe a

combination of a voltage-sensitive dye imaging technique with whole-cell
recordings and postrecording anatomical analysis of synaptic circuits. The
application of this combination to the rat somatosensory barrel cortex has
advanced our understanding of cortical function.
II. VOLTAGE-SENSITIVE DYE IMAGING
Dye molecules that insert into the plasma membrane and change their op-
tical absorption and/or emission properties, dependent upon the electrical
field across the membrane, can be considered as voltage-sensitive dyes. The
first optical measurements of action potentials were made on invertebrate
preparations such as the squid giant axon and neurons of the leech (Salzberg
et al., 1973; Tasaki et al., 1968). The first vertebrate in vivo voltage-sensitive
dyes measurements visualized the spatiotemporal dynamics of sensory pro-
cessing (Grinvald et al., 1984). Since these pioneering steps, many further
compounds have been successfully tested as voltage-sensitive dyes. Of equal
importance, camera technology has advanced dramatically.
Voltage-sensitive dyes typically show a linear, approximately 10% change
in fluorescence for 100 mV change in membrane potential. Despite the
small amplitude of the signals, the spatiotemporal dynamics of the mem-
brane potential of individual neurons can be imaged in vitro in individual
mammalian neurons by loading voltage-sensitive dyes intracellularly with
the whole-cell patch-clamp technique (Antic et al., 1999; Djurisic et al., 2004;
Zecevic, 1996). Significant signal-to-noise improvements may be realized
from excitation at the red spectral edge (Kuhn et al., 2004).
In vivo imaging is hindered both by the small signal amplitude and by
the heart-beat-pulsation-related artifacts, resulting primarily from changes
in the blood oxygenation level. Shoham et al. (1999) made dramatic
advances with in vivo imaging by using novel blue dyes, which are ex-
cited at long wavelengths where hemoglobin has little absorption. One
of these dyes RH1691 (Fig. 14.1A) can be excited with 630-nm red light
(Fig. 14.1B) at which wavelength it shows an increase in fluorescence
(>665 nm) upon depolarization (Fig. 14.1C). RH1691 has proven extremely
useful for high-resolution in vivo imaging of cortical function (Grinvald and
Hildesheim, 2004) with low toxicity to the imaged neurons (Petersen et al.,
2003a).
III. WHOLE-CELL PATCH-CLAMP RECORDING
The patch-clamp technique developed by Bert Sakmann and Erwin

Neher provided the first direct recordings of single ion channels (Neher
and Sakmann, 1976). Further development of different recording config-
urations together with the remarkable mechanical stability and low noise
A
RH 1691
SO3H
OCH3
O OH
N N
Cl N
S S
CH3
B hv
hv
630nm >665nm
RH 1691
out
cell
membrane
in
C ex em
hyperpolarized
depolarized
450 550 650 750nm
Figure 14.1. Part A shows the chemical structure of voltage-sensitive dye RH1691.
After extracellular application, the dye molecules insert into the plasma membrane
and upon excitation with 630-nm light they emit fluorescence that can be measured
after 665-nm long-pass filters, as schematically indicated in Part B. A schematic draw-
ing of the spectral shifts induced by changes in membrane potential is shown in
Part C. During depolarization there is an increased absorption at 630 nm and an
increase in fluorescence can be detected.
of the “gigaseal” (Hamill et al., 1981) has extended the usefulness of this
electrophysiological recording technique to virtually all biological prepa-
rations. The whole-cell patch-clamp configuration has proven particularly
useful for recording both in acute brain slices and in vivo in the intact living
animal (Fig. 14.2A,B). This is possible because the tip of the glass electrode
patch-pipette can be kept clean by applying positive pressure to its inside
with a resultant continuous outflow of intracellular solution. During the
“search” for a cell to be recorded (Fig. 14.2C) this flux of liquid prevents de-
bris from sticking to the patch-pipette and also “cleans” the cell membrane
A in vitro brain slice B in vivo

wc
40x
obj.
wc
C
search cell-attached whole-cell
VP
IP
Figure 14.2. Whole-cell (WC) patch-clamp recordings can be made in vitro from
neurons in brain slices (A) or in vivo from the intact living animal (B). Part C
schematically shows various configurations of the patch-clamp technique. In the
search mode, the pipette is under positive pressure, which keeps the tip of the
electrode clean as it penetrates brain tissue. Large square shape currents (IP ) are
evoked by square-shaped voltage command pulses (VP ) to the patch-clamp ampli-
fier. When a cell is encountered current flow is decreased and gentle suction is
applied. This allows formation of the gigaseal. This cell-attached configuration is
characterized by very little current flow during voltage pulses, indicating a high re-
sistance. Further gentle suction breaks down the membrane patch inside the tip of
the electrode while maintaining the tight electrical seal of the membrane in con-
tact with the glass pipette. This whole-cell configuration is characterized by large
capacitative current transients in response to voltage pulses. The whole-cell config-
uration allows exchange of molecules from the patch-pipette with the inside of the
cell.
priming it for the gigaseal. The cell to be recorded can be either visualized
by various microscopy techniques (Margrie et al., 2003; Stuart et al., 1993) or
electrically sensed by the increased resistance encountered as the pipette
hits a cell (Blanton et al., 1989). Upon encountering a target cell, the posi-
tive pressure is reversed by gentle sucking and an electrically tight gigaseal is
formed. The gigaseal has resistance of >1 G and has remarkable mechan-
ical stability. This recording configuration is termed cell attached and can
be used to record single channels in the patch membrane. The whole-cell
recording configuration can be entered by further suction and can be mon-
itored by the large increase in capacitance associated with the whole-cell
membrane area.
The whole-cell configuration can remain stable for hours even in vivo
despite the small movements of the brain (caused, for example, by
heart-beat-pulsation), allowing high-quality measurements of membrane
potential. The ease and stability with which whole-cell recordings can be
obtained allow this experimental technique to be combined with other tech-
niques such as voltage-sensitive dye imaging (Fig. 14.3).
Figure 14.3. Part A shows a bright field photograph of a brain slice of the rat so-
matosensory cortex. The layer 4 barrels are outlined in cyan. One layer 4 neuron
was recorded and later reconstructed with dendrites shown in black and axon in
green. One layer 2/3 neuron was recorded and drawn with dendrites in red and
axon in blue. A photograph of the stained neurons is shown in Part B. While the
two neurons were recorded by the whole-cell patch-clamp technique, large-scale
network activity was evoked by an extracellular stimulus delivered by a third elec-
trode. This stimulation electrode was placed in layer 4 and the ensemble response
was visualized with millisecond resolution with voltage-sensitive dye. Part C shows
the voltage-sensitive dye image captured 12 ms after stimulation. The image shows
columnar excitation. Part D shows a reconstruction of the somatodendritic com-
partment a layer 2/3 neuron recorded in vivo (viewed in a plane normal to the pial
surface and along the row). During the whole-cell recording the cortical dynamics
were imaged with voltage-sensitive dye. Part E shows the quantification of the C3
whisker-evoked voltage-sensitive dye response quantified over a 200 × 200 µm re-
gion centered on the soma of the recorded neuron. The time course of the optical
response closely matches the time course of the changes in membrane potential of
the recorded neuron. Part F indicated the lateral locations of the recorded neuron
relative to the layer 4 barrels (cyan) and the evoked voltage-sensitive dye response
recorded 15 ms after stimulation. (Reprinted in modified form with permission from
Petersen and Sakmann, 2001 ( c 2001 Society for Neuroscience), and Petersen et al.,
2003a ( c 2003 Society for Neuroscience)).
In addition to measuring the electrophysiology of neurons, the whole-cell

technique allows the introduction of small molecules into the cell cytoplasm.
By including biocytin (biotinyl lysine) in the solution in the patch pipette,
this label diffuses into intracellular milieu entering both axonal and den-
dritic compartments. After recording of the neuron, the brain can be fixed
and sectioned. The location of the biocytin molecules can be revealed by
the specific binding of the biotin motif (part of the biocytin molecule) to
avidin conjugated to peroxidase, which in a reaction with diaminobenzidine
forms a dark deposit. This allows visualization of the neuronal structure un-
der light microscopy, and computer-aided tracing of dendritic and axonal
compartments in three dimensions (Fig. 14.3).
IV. AN APPLICATION: ANALYZING THE SENSORY

RESPONSE IN RODENT BARREL CORTEX
The rodent primary somatosensory cortex has aroused much interest

in neurobiology because of its high degree of anatomical organization
(Petersen, 2003). The mystacial vibrissae representation in this cortical area
is segregated into discrete units termed barrels present in layer 4, which
can be visualized in living brain slices (Petersen and Sakmann, 2000) as well
as by numerous staining techniques. The layout of the barrels across the
cortical map is identical to the layout of the whiskers on the snout of the
rodent. This suggests that each of these barrels is intimately involved in pro-
cessing the information from its corresponding whisker (Woolsey and Van
der Loos, 1970).
The first level of neocortical processing begins with the layer 4 barrel
neurons, which are directly connected to thalamic VPM neurons through
glutamatergic synapses (Agmon and Connors, 1991). Both glutamatergic
excitatory (spiny stellate and star pyramidal neurons) and diverse classes of
GABAergic neurons in layer 4 receive direct VPM input (Bruno and Simons,
2002; Porter et al., 2001). Excitatory layer 4 neurons within the same barrel
are strongly connected with approximately every third pair of neurons being
synaptically connected, but there is very little synaptic connectivity between
neighboring barrels (Feldmeyer et al., 1999; Petersen and Sakmann, 2000,
2001; Schubert et al., 2003; Shepherd et al., 2003). As seen in Fig. 14.4A,B,
this pattern of physiologically measured synaptic connectivity likely results
from the highly polarized dendritic and axonal arbors of these neurons,
which rarely enter neighboring layer 4 barrels (Lübke et al., 2000; Petersen
and Sakmann, 2000). Each layer 4 barrel, therefore, is an independent and
irreducible unit consisting of a few thousand neurons which process infor-
mation relating primarily to its isomorphic whisker. The ability to define the
neuronal network in terms of both the number of participating neurons and
the normal physiological input (since the location of the barrel in the sen-
sory map can be established) makes the barrel cortex an ideal starting point
for quantitative modeling of neocortical networks (Petersen, 2002).
A C 0ms 1.2ms
L2/3
L2/3
L4
100µm 2.4ms 3.6ms 4.8ms
L4
L5
B 6ms 12ms 25.2ms
L2/3
L4
200µm 0.1 49.8ms 99.6ms 199.8ms
0.05
L2/3
∆I/I0 (%)
0
L4
-0.05
Figure 14.4. Parts A and B show the superposition of many excitatory neurons re-
constructed from in vitro brain slice recordings and normalized according to the
barrel width. Dendrites and cell bodies of layer 4 neurons shown in black are largely
confined to the layer 4 barrel. The axons of the layer 4 neurons shown in green are
laterally confined to the width of the layer 4 barrel but project heavily to both layers
2/3 and 4. The dendrites and cell bodies of the layer 2/3 pyramidal neurons are
indicated in red and their axons in blue. The layer 2/3 axon spreads far laterally.
Part C shows the functional activation of a barrel column evoked by extracellular
stimulation of the layer 4 barrel and measured by voltage-sensitive dye imaging. In
addition to the large stimulation electrode (green) in the layer 4 barrel, there is also
a whole-cell recording pipette in layer 4 (red) and another in layer 2/3 (blue). The
images demonstrate remarkably tight columnar activation throughout the duration
of the response. (Reprinted in modified form with permission from Petersen and
Sakmann, 2001 ( c 2001 Society for Neuroscience)).
The excitatory layer 4 neurons project most densely into layer 2/3 with
the horizontal axonal field spreading little wider than the underlying layer
4 barrel, thus defining anatomically a neocortical column (Fig. 14.4A, B).
Excitatory synaptic connections from layer 4 to layer 2/3 pyramidal neu-
rons occur frequently but have smaller efficacies and smaller NMDA re-
ceptor components than the layer 4 to layer 4 synapses (Feldmeyer et al.,
2002). The flow of excitation is strictly feed forward since there are no
reciprocal excitatory connections from layer 2/3 to layer 4. Layer 2/3 pyra-
midal neurons synapse with their neighboring layer 2/3 pyramidal neu-
rons, layer 5/6 pyramidal neurons (Reyes and Sakmann, 1999) and project
to other cortical areas including contralateral somatosensory cortex, mo-
tor cortex, and secondary somatosensory cortex. Within the local circuits
the axonal fields of layer 2/3 pyramidal neurons do not respect barrel col-
umn boundaries (Fig. 14.4A,B) extending far into the neighboring barrel
columns.
To probe how this neuronal network operates when many neurons are
excited, we imaged the membrane potential with voltage-sensitive dye (Fig.
14.4C). Stimuli were delivered to a single layer 4 barrel causing local exci-
tation and spread of activity to the supragranular layer in a columnar fash-
ion (Petersen and Sakmann, 2001). This was the first demonstration of a
functional neocortical column at the subthreshold synaptic level, which
matches the anatomically defined extent of the layer 4 axons.
The activity of the excitatory neuronal network is likely to be strongly
regulated by the many diverse types of cortical GABAergic interneurons
(Gupta et al., 2000). When GABAergic inhibition is blocked in vitro, synaptic
excitation can spread horizontally in layer 2/3 presumably through local
excitatory synapses (Petersen and Sakmann, 2001).
During behavior, the whiskers usually operate in concert as a sensory
organ. Therefore, the exchange of information related to the individual
whiskers is likely to play a prominent part in cortical processing. One role
for the barrel cortex is then to distribute the information related to the
movement of a single whisker and compare this with information relating
to movements of other whiskers. Such a process may occur in a defined
spatial and temporal integrative process in the cortex.
The distributed nature of sensory signals originating from single brief
sensory stimuli has been highlighted by combined in vivo voltage-sensitive
dye imaging and whole-cell recordings (Fig. 14.5). This direct measurement
of how cortical activity evoked by a single whisker is spatiotemporally dis-
tributed across the barrel cortex (Petersen et al., 2003a) correlates well with
measurements of receptive field properties of individual neurons analyzed
by sequentially deflecting many whiskers (Armstrong-James et al., 1992;
Brecht et al., 2003; Brecht and Sakmann, 2002; Moore and Nelson, 1998;
Simons, 1978; Zhu and Connors, 1999). The earliest sensory response
occurs ∼8 ms following whisker deflection and is localized to the direct tar-
gets of the VPM input, the layer 4 barrel neurons and a fraction of neurons
in mid-layer 5/6. In the next milliseconds, excitation propagates into layer
2/3 in a columnar fashion. Thus a functional neocortical column, bounded
laterally by the layer 4 barrel structure, is depolarized 10–12 ms after whisker
deflection (Fig. 14.5C). In the following milliseconds both infragranular
neurons and neurons in neighboring barrel columns become excited, ap-
parently mainly through local cortical synaptic circuits. Excitation spreads
preferentially along the row orientation of the barrel cortex, for example
deflection of the D2 whisker evokes first a response in the D2 barrel column
and over the next milliseconds the largest responses are found in D1 and D3
neighboring barrel columns with smaller responses in the C2 or E2 columns.
This oriented spread of excitation may serve a useful physiological function.
The whisking behavior involves rapid whisker movements oriented largely in
a plane along the rows. Thus during the forward motion of the whiskers, the
D3 whisker will pass through a point in space a few milliseconds before the
D2 whisker, which in turn will be followed by the D1 whisker moving through
the identical spatial location. Thus whiskers lying in the same row will often
B C 10ms
A Tangential Row 500µm
Arc
D 20ms
B1 B2 B3 B4
β
Row
C1 C2 C3 C4
L1 γ
D1 D2 D3 D4
L2/3
δ
E1 E2 E3
L4
E 50ms
500µm
Arc
L1
L2/3
L4
-0.1 0 0.1
∆F/F0 (%)
Figure 14.5. Part A shows the projection in three orthogonal directions of three
dimensionally reconstructed axons (in blue) and dendrites (in red) of layer 2/3
pyramidal neurons recorded in vivo. The right-hand column shows the 10 and 50%
contours of the length density of axon (blue) and dendrite (red) computed from the
superimposed, gaussian smoothed, normalized computer-aided three-dimensional
reconstructions. The axons extend preferentially in the row direction of the barrel
cortex organization. Part B shows the blood vessels at the surface of the somatosen-
sory cortex. Part C shows the voltage-sensitive dye signals recorded in response to
deflection of the D2 whisker. The earliest signals occur ∼10 ms following whisker
deflection and are localized to the homologous barrel column. In the next tens of
milliseconds the signal propagates over a large cortical area preferentially in the
row direction. After the functional imaging, DiI was injected into the location of the
epicenter of the response. The DiI was allowed to diffuse and later the brain was
sectioned tangentially to locate the layer 4 barrels (viewed under transillumination
without staining in Part D). DiI fluorescence in layer 4 was found in the D2 barrel
(E) in agreement with the location of the response to the D2 whisker deflection.
(Reprinted in modified form with permission from Petersen et al., 2003a ( c 2003
Society for Neuroscience)).
sample the same point in space within milliseconds of each other. In order
for the animal to process this information relating to individual whiskers
distributed across the neocortical barrel field, it is likely to be important that
this single-whisker-related information is rapidly exchanged along the rows
of the barrel cortex. The rapid spread of the sensory response may mediate
this integrative process, with propagation velocities along the row being
twice as fast as along the orthogonal arc direction (Petersen et al., 2003a).
This spread of excitation may be mediated by local excitatory synaptic
connections in layer 2/3 since their axons are preferentially oriented along
the rows of the barrel cortex (Fig. 14.5A). The combined methodologies
of single-cell recording, anatomical reconstruction, and ensemble imaging

are therefore beginning to describe the synaptic events underlying sensory
processing.
V. SUMMARY OF ADVANTAGES AND LIMITATIONS
Recent advances in imaging technology and new voltage-sensitive dyes

now allow high-resolution imaging of cortical dynamics (Petersen et al.,
2003b). However, the current in vivo technique does not allow signals
from individual neurons to be resolved within the stained neocortical net-
work, but only ensemble activity. To overcome this limitation will likely
require future generations of much improved voltage-sensitive fluorescent
proteins (Ataka and Pieribone, 2002; Cacciatore et al., 1999; Sakai et al., 2001;
Siegel and Isacoff, 1997). However, calcium as a measure of neuronal activ-
ity can be readily imaged with single-cell resolution within networks stained
with cell permeant calcium-sensitive dyes (Peterlin et al., 2000; Stosiek et al.,
2003; also see the chapter of Goldberg et al., 2005, in this volume).
Since voltage-sensitive dye imaging is technically straightforward, this
imaging technique can be readily combined with whole-cell patch-clamp
recordings. Thus simultaneous measurements of single neuron and spa-
tiotemporally resolved ensemble membrane potential measurements can
be made. Together with anatomical analysis we can therefore begin to re-
construct the synaptic pathways, underlying simple sensory responses in the
neocortex.
APPENDIX
A. Commercial Sources of Voltage-Sensitive Dye, Camera,

and Imaging Software
1. Commercial Source of Voltage-Sensitive Dye RH1691
Optical Imaging Inc., PO Box 1262, Mountainside, NJ 07092-1262

(http://www.opt-imaging.com).
2. Commercial Sources of Cameras and Imaging Software
Imager 3001: Optical Imaging Inc., PO Box 1262, Mountainside, NJ 07092-

1262 (http://www.opt-imaging.com).
NeuroCCD/NeuroPDA: RedShirtImaging LLC, 2 Stoneleigh Road, Fairfield,
CT 06825 (http://www.redshirtimaging.com/).
MiCAM: SciMedia Ltd, 4 Executive Circle, Suite 170, Irvine, CA 92614
(http://www.scimedia.com).
B. Staining Neocortex with Voltage-Sensitive Dye
1. It is important to carry out animal experiments in accordance with

local legislation. Anesthetize animal (e.g., juvenile 200 g Wistar rat
injected intraperitoneally with urethane at 1.75 mg/g). Maintain the
body temperature at 37◦ C.
2. Remove or reflect the skin covering the skull.
3. Carefully scrape the bone clean. Further cleaning of the bone can be
performed, if necessary, with 1% hydrogen peroxide.
4. Apply a thin layer of cyanoacrylate glue to the surrounding region
of the bone, away from where the craniotomy will be performed
(this helps dental cement adhesion). Glue a metal head-plate to the
skull with dental cement. The plate must be tangential to the region
of cortex to be imaged and the hole in the plate must be positioned
over this region.
5. When the dental cement has hardened, immobilize the skull and min-
imize movement of the brain by fixing the head-plate firmly between
metal posts.
6. Perform a craniotomy of desired size by drilling within the hole of
the head-plate. Be careful not to damage the underlying brain tissue.
7. Finally, remove the dura, leaving the pia of the underlying cortex
exposed.
8. Dissolve the voltage-sensitive dye RH1691 (Shoham et al., 1999) to
0.1–1 mg/ml in Ringer’s solution: 135 mM NaCl, 5 mM KCl, 5 mM
HEPES, 1.8 mM CaCl2 , and 1 mM MgCl2 .
9. Apply a small quantity (∼250 µl) of this dye solution to the craniotomy.
Seal this chamber with a glass coverslip and very slight pressure is
applied to prevent brain edema.
10. Leave the cortex to stain for 1–2 h. During this period, the dye will
diffuse into the superficial layers of the neocortex. At the end of the
staining period, remove unbound dye by washing the cortex exten-
sively with Ringer’s solution. The cortex should now have a pale blue
color.
11. Cover the craniotomy with 1% agarose, dissolved in Ringer’s solution,
and place a glass coverslip on top. The glass coverslip should not be
just large enough to cover the width of the craniotomy, but little more.
This will allow access for whole-cell recordings to be made.
C. Voltage-Sensitive Dye Imaging
1. Illuminate with ∼530-nm green light and record the blood vessel pat-
tern on the cortical surface using the camera.
2. Move the focal plane 300 µm into the cortex and excite the voltage-
sensitive dye with epifluorescent light at 630 nm. Emitted light is long-
pass filtered (>665 nm), forming the voltage-sensitive dye signal, which
should be recorded at frame rates faster than 100 Hz. Heart-beat-

related signals form the largest artifacts. The timing of these known
artifacts can be recorded via an electrocardiogram. The artifacts can
then be removed by computer processing to improve the resolution of
the collected signals.
D. Whole-Cell Recording
1. Fill whole-cell pipettes with intrapipette solution: 135 mM potassium

gluconate; 4 mM KCl; 10 mM HEPES; 10 mM phosphocreatine;
4 mM MgATP; and 0.3 mM Na3 GTP; pH 7.2 adjusted with KOH
and include 3 mg/ml of biocytin (to allow staining of the recorded
neurons).
2. Pipettes should have a resistance of ∼5 M. Monitor the tip resistance
by applying brief voltage steps of 5 mV in the voltage-clamp mode
while measuring the current flow on an oscilloscope. Apply a positive
pressure of ∼200 mbar on the pipette.
3. Slowly advance the electrode through the agarose under the coverslip
and into the cortex. When the tip is close to the chosen recording site,
reduce the positive pressure to ∼30 mbar.
4. Advance the pipette in 2 µm steps until the tip resistance suddenly
increases (indicating contact with a cell membrane). Release the
pressure in the pipette and apply light suction until a gigaseal is
formed.
5. Establish whole-cell recording configuration by rupturing the mem-
brane in the pipette. This can be achieved by applying either brief
suction pulses or slowly increasing the suction pressure.
6. After collecting data, slowly retract the whole-cell recording pipette
while monitoring whole-cell capacitance transients with 5 mV voltage
steps. During the retraction the excised patch configuration should be
established. This insures that the neuron remains intact and viable for
later anatomical staining.
E. Anatomical Analysis of Neuronal Structure and Position
1. Supplement anesthetic (e.g., by an additional intraperitoneal injection

of 1 mg/g of urethane). Perfuse the animal transcardially with ice-
cold 0.1 M phosphate buffer (pH ∼7.3) and then with ∼50 ml of 4%
formaldehyde. Remove the brain from the skull and postfix overnight
at 4◦ C.
2. Section the cortex tangentially with a vibratome at 100 µm and stain for
biocytin using ABC kit (Vectastain Laboratories). The angle of mount-
ing the brain for tangential sectioning is difficult to determine exactly,
but is helped by the slightly blue color of the brain in the craniotomy
remaining from the voltage-sensitive dye staining.
3. The blood vessels initially imaged with green illumination now pro-
vide the link between the location of the neuronal processes visual-
ized by the anatomical stain and the functional voltage-sensitive dye
images. The blood vessels, barrel patterns, and the axonal and den-
dritic processes can be traced in three dimensions, using an image-
combining computerized microscopy system (e.g., Neurolucida, Mi-
croBrightField, Inc. Williston, VT). For further details please refer in
this volume to the chapter by Ascoli and Scorcioni and also the chapter
by Bjaalie and Leergard.
4. In addition, other fluorescent labels can be introduced into the brain
during the in vivo experiment for additional position information.
Pipettes filled with DiI (1 mg/ml dissolved in dimethylformamide)
can be inserted into the brain and DiI ejected through a brief pres-
sure pulse. The DiI diffuses into the nearby tissue labeling axon and
dendrites.
Acknowledgments. I would like to thank the generous support of the

Swiss National Science Foundation 3100A0-103832 “Synaptic mechanisms
of sensory perception and associative learning” and the Leenaards Foun-
dation “Neurobiology of addiction: interaction of barrel cortex and ventral
tegmental area.”
REFERENCES
Agmon, A., and Connors, B. W., 1991, Thalamocortical responses of mouse somatosensory
(barrel) cortex in vitro, Neuroscience 41:365–379.
Antic, S., Major, G., and Zecevic, D., 1999, Fast optical recordings of membrane potential
changes from dendrites of pyramidal neurons, J. Neurophys. 82:1615–1621.
Armstrong-James, M., Fox, K., and Das-Gupta, A., 1992, Flow of excitation within rat barrel
cortex on striking a single vibrissa, J. Neurophysiol. 68:1345–1358.
Ataka, K., and Pieribone, V. A., 2002, A genetically targetable fluorescent probe of channel
gating with rapid kinetics, Biophys. J. 82:509–516.
Blanton, M. G., Lo Turco, J. J., and Kriegstein, A. R., 1989, Whole-cell recording from neurons
in slices of reptilian and mammalian cerebral cortex, J. Neurosci. Methods 30:203–210.
Brecht, M., Roth, A., and Sakmann, B., 2003, Dynamic receptive fields of reconstructed pyra-
midal cells in layers 3 and 2 of rat somatosensory barrel cortex, J. Physiol. 553:243–
265.
Brecht, M., and Sakmann, B., 2002, Dynamic representation of whisker deflection by synaptic
potentials in spiny stellate and pyramidal cells in the barrels and septa of layer 4 rat
somatosensory cortex, J. Physiol. 543:49–70.
Bruno, R. M., and Simons, D. J., 2002, Feedforward mechanisms of excitatory and inhibitory
cortical receptive fields, J. Neurosci. 22:10966–10975.
Buzsaki, G., 2004, Large-scale recording of neuronal ensembles, Nat. Neurosci. 7:446–451.
Cacciatore, T. W., Brodfuehrer, P. D., Gonzalez, J. E., Jiang, T., Adams, S. R., Tsien, R. Y.,
Kristan, W. B., and Kleinfeld, D., 1999, Identification of neural circuits by imaging coherent
electrical activity with FRET-based dyes, Neuron 23:449–459.
Djurisic, M., Antic, S., Chen, W. R., and Zecevic, D., 2004, Voltage imaging from dendrites of
mitral cells: EPSP attenuation and spike trigger zones, J. Neurosci. 24:6703–6714.
Feldmeyer, D., Egger, V., Lubke, J., and Sakmann, B., 1999, Reliable synaptic connections
between pairs of excitatory layer 4 neurons within a single “barrel” of developing rat
somatosensory cortex, J. Physiol. 521:169–190.
Feldmeyer, D., Silver, R. A., Lübke, J., and Sakmann, B., 2002, Synaptic connections between
layer 4 spiny neurone-layer 2/3 pyramidal cell pairs in juvenile rat barrel cortex: phys-
iology and anatomy of interlaminar signaling within a cortical column, J. Physiol. 538:
803–822.
Grinvald, A., Anglister, L., Freeman, J. A., Hildesheim, R., and Manker, A., 1984, Real-time
optical imaging of naturally evoked electrical activity in intact frog brain, Nature 308:848–
850.
Grinvald, A., and Hildesheim, R., 2004, VSDI: a new era in functional imaging of cortical
dynamics, Nat. Rev. Neurosci. 5:874–885.
Gupta, A., Wang, Y., and Markram, H., 2000, Organizing principles for a diversity of GABAergic
interneurons and synapses in the neocortex, Science 287:273–278.
Hamill, O. P., Marty, A., Neheri, E., Sakmann, B., and Sigworth, F. J., 1981, Improved patch-
clamp techniques for high-resolution current recording from cells and cell-free membrane
patches, Pflügers Arch. 391:85–100.
Kuhn, B., Fromherz, P., and Denk, W., 2004, High sensitivity of stark-shift voltage-sensing dyes
by one- or two-photon excitation near the red spectral edge, Biophys. J. 87:631–639.
Lübke, J., Egger, V., Sakmann, B., and Feldmeyer, D., 2000, Columnar organization of dendrites
and axons of single and synaptically coupled excitatory spiny neurons in layer 4 of the rat
barrel cortex, J. Neurosci. 20:5300–5311.
Margrie, T. W., Meyer, A. H., Caputi, A., Monyer, H., Hasan, M. T., Schaefer, A. T., Denk, W., and
Brecht, M., 2003, Targeted whole-cell recordings in the mammalian brain in vivo, Neuron
39:911–918.
Moore, C. I., and Nelson, S. B., 1998, Spatio-temporal subthreshold receptive fields in
the vibrissa representation of rat primary somatosensory cortex, J. Neurophysiol. 80:
2882–2892.
Neher, E., 1992, Ion channels for communication between and within cells, Science 256:498–502.
Neher, E., and Sakmann, B., 1976, Single-channel currents recorded from membrane of den-
ervated frog muscle fibres, Nature 260:799–802.
Nicolelis, M. A., Dimitrov, D., Carmena, J. M., Crist, R., Lehew, G., Kralik, J. D., and Wise S. P.,
2003, Chronic, multisite, multielectrode recordings in macaque monkeys, Proc. Natl. Acad.
Sci. U.S.A. 100:11041–11046.
Peterlin, Z. A., Kozloski, J., Mao, B. Q., Tsiola, A., and Yuste, R., 2000, Optical probing of
neuronal circuits with calcium indicators, Proc. Natl. Acad. Sci. U.S.A. 97:3619–3624.
Petersen, C. C. H., 2002, Short-term plasticity within the excitatory neuronal network of layer
4 barrel cortex, J. Neurophysiol. 87:2904–2914.
Petersen, C. C. H., 2003, The barrel cortex—integrating molecular, cellular and systems phys-
iology, Pflügers Arch. 447:126–134.
Petersen, C. C. H., Grinvald, A., and Sakmann, B., 2003a, Spatiotemporal dynamics of sensory
responses in layer 2/3 of rat barrel cortex measured in vivo by voltage-sensitive dye imaging
combined with whole-cell voltage recordings and anatomical reconstructions, J. Neurosci.
23:1298–1309.
Petersen, C. C. H., Hahn, T. T. G., Mehta, M., Grinvald, A., and Sakmann, B., 2003b, Interaction
of sensory responses with spontaneous depolarization in layer 2/3 barrel cortex, Proc. Natl.
Acad. Sci. U.S.A. 100:13638–13643.
Petersen, C. C. H., and Sakmann, B., 2000, The excitatory neuronal network of layer 4 barrel
cortex, J. Neurosci. 20:7579–7586.
Petersen, C. C. H., and Sakmann, B., 2001, Functionally independent columns of rat so-
matosensory barrel cortex revealed with voltage-sensitive dye imaging, J. Neurosci. 21:
8435–8446.
Porter, J. T., Johnson, C. K., and Agmon, A., 2001, Diverse types of interneurons generate
thalamus-evoked feedforward inhibition in the mouse barrel cortex, J. Neurosci. 21:2699–
2710.
Reyes, A., and Sakmann, B., 1999, Developmental switch in the short-term modification of
unitary EPSPs evoked in layer 2/3 and layer 5 pyramidal neurons of rat neocortex, J.
Neurosci. 19:3827–3835.
Sakai, R., Repunte-Canonigo, V., Raj, C. D., and Knopfel, T., 2001, Design and characterization
of a DNA-encoded, voltage-sensitive fluorescent protein, Eur. J. Neurosci. 13:2314–2318.
Sakmann, B., 1992, Elementary steps in synaptic transmission revealed by currents through
single ion channels, Science 256:503–512.
Salzberg, B. M., Davila, H. V., and Cohen, L. B., 1973, Optical recording of impulses in individual
neurons of an invertebrate central nervous system, Nature 246:508–509.
Schubert, D., Kotter, R., Zilles, K., Luhmann, H. J., and Staiger, J. F., 2003, Cell type-specific
circuits of cortical layer 4 spiny neurons, J. Neurosci. 23:2961–2970.
Shepherd, G. M., Pologruto, T. A., and Svoboda, K., 2003, Circuit analysis of experience-
dependent plasticity in the developing rat barrel cortex, Neuron 38:277–289.
Shoham, D., Glaser, D. E., Arieli, A., Kenet, T., Wijnbergen, C., Toledo, Y., Hildesheim, R., and
Grinvald, A., 1999, Imaging cortical dynamics at high spatial and temporal resolution with
novel blue voltage-sensitive dyes, Neuron 24:791–802.
Siegel, M. S., and Isacoff, E. Y., 1997, A genetically encoded optical probe of membrane voltage,
Neuron 19:735–741.
Simons, D. J., 1978, Response properties of vibrissa units in rat SI somatosensory neocortex, J.
Stosiek, C., Garaschuk, O., Holthoff, K., and Konnerth, A., 2003, In vivo two-photon calcium
imaging of neuronal networks, Proc. Natl. Acad. Sci. U.S.A. 100:7319–7324.
Stuart, G. J., Dodt, H. U., and Sakmann, B., 1993, Patch-clamp recordings from the soma
and dendrites of neurons in brain slices using infrared video microscopy, Pflügers Arch.
423:511–518.
Tasaki, I., Watanabe, A., Sandlin, R., and Carnay, L., 1968, Changes in fluorescence, turbidity
and birefringence associated with nerve excitation, Proc. Natl. Acad. Sci. U.S.A. 61:883–888.
Woolsey, T. A., and Van der Loos, H., 1970, The structural organization of layer 4 in the
somatosensory region (SI) of the mouse cerebral cortex: the description of a cortical field
composed of discrete cytoarchitectonic units, Brain Res. 17:205–242.
Zecevic, D., 1996, Multiple spike-initiation zones in single neurons revealed by voltage-sensitive
dyes, Nature 381:322–325.
Zhu, J. J., and Connors, B. W., 1999, Intrinsic firing patterns and whiskers-evoked synaptic
responses of neurons in the rat barrel cortex, J. Neurophysiol. 81:1171–1183.
15
From Dendrites to Networks:
Optically Probing the Living
Brain Slice and Using
Principal Component
Analysis to Characterize
Neuronal Morphology
JESSE H. GOLDBERG, FARID HAMZEI-SICHANI,
JASON MacLEAN, GABOR TAMAS, ROCHELLE
URBAN, and RAFAEL YUSTE
INTRODUCTION
TWO-PHOTON CALCIUM IMAGING IN DENDRITES
Selection of the Calcium Indicator
Targeting the Stimulation Electrode to the Dendrite of Interest
Synaptic Activation of Single Synapses
ULTRASTRUCTURAL IDENTIFICATION OF IMAGED DENDRITIC
SEGMENTS
IMAGING NEURONAL ENSEMBLES
“Bulk” Loading Acute Cortical Slices with Calcium Indicator
Imaging Ensemble Activity
QUANTITATIVE CLASSIFICATION OF NEURONAL MORPHOLOGY
USING PRINCIPAL COMPONENT ANALYSIS
JESSE H. GOLDBERG • McGovern Institute for Brain and Cognitive Sciences, Massachus-
etts Institute of Technology, Cambridge, MA JASON MACLEAN AND ROCHELLE
URBAN • Department of Biological Sciences, Columbia University, New York, NY 10027
RAFAEL YUSTE • Howard Hughes Medical Institute, Department of Biological Sciences,
Columbia University New York, NY 10027 FARID HAMZEI-SICHANI • Department
of Physiology and Pharmacology, State University of New York, Downstate Medical Center,
Brooklyn, NY GABOR TAMAS • Department of Comparative Physiology, University
of Szeged, Kozepfasor 52, Szeged H-6726, Hungary
452
OPTICALLY PROBING THE LIVING BRAIN SLICE 453
Fundamental Concepts of Principal Component Analysis
Example of Application
Two-Photon Imaging of Dendritic Calcium Microdomains
Imaging Neuronal Ensembles
Principal Component Analysis Versus Factor Analysis
APPENDIX
Two-Photon Imaging of Dendritic Calcium Microdomains
Electron Microscopic Reconstruction of Imaged Dendritic
Domains
Imaging Neuronal Ensembles
REFERENCES
Abstract: Recently, advances in optical imaging of the living brain slice preparation
have permitted neuronal circuitry to be examined at multiple levels, ranging from
individual synaptic contacts on dendrites to whole populations of neurons in a net-
work. In this chapter, we describe three techniques that, together, enable a powerful
dissection of neuronal circuits across multiple space scales. We describe methods
for (1) combining whole-cell recording with two-photon calcium imaging and elec-
tron microscopic reconstruction to examine the functions of individual synapses
and dendrites during synaptic stimulation, (2) imaging hundreds of neurons in the
brain slice simultaneously to examine the spatiotemporal dynamics of activity in
living neuronal networks, and (3) performing an unbiased, quantitative analysis of
neuronal morphology that is increasingly necessary in light of the multiparametric
structural diversity of distinct neuronal subclasses.
Keywords: cluster analysis, dendrite, imaging, microdomain, network, principal com-
ponent analysis, two-photon calcium
I. INTRODUCTION
Brain slices have proven to be an excellent system to study the structure

and function of cortical networks (Dingledine et al., 1980). Early studies of
pharmacologically disinhibited cortex showed that the in vitro circuit was
functionally intact and capable of generating epileptiform activity (Prince
and Connors, 1984), and plasticity studies with extracellular recording elec-
trodes showed that synapses could undergo activity-dependent plasticity in
vitro (Alger and Teyler, 1976; Sarvey et al., 1989), with potentially great
relevance to fundamental questions of learning and memory. Later, the
advent of intracellular recording in brain slices revealed that distinct neu-
ronal populations exhibited unique electrical behaviors (Connors and Gut-
nick, 1990), and, combined with post-hoc anatomical reconstructions and
immunohistochemistry, have enabled the detailed study of hippocampal
and neocortical networks, composed of excitatory pyramidal neurons as
well as a great diversity of GABAergic interneurons (Freund and Buzsaki,
1996; Markram et al., 2004; McBain and Fisahn, 2001; Somogyi et al., 1998).
454 JESSE H. GOLDBERG et al.
Further, the advent of simultaneous intracellular recordings from connected

pairs of neurons revealed that connections between neurons are highly spe-
cific with respect to short-term synaptic dynamics, electrical coupling, and
even the timing that governs synaptic plasticity (Gibson et al., 1999; Gupta
et al., 2000; Markram et al., 1997; Tamas et al., 2000).
A major limitation of the electrophysiological approach to the brain slice
preparation is inability to examine many spatially distinct components of
cortical computation. This is important at the levels of both the individual
neuron, where synaptic integration occurs in geometrically complex den-
drites (Goldberg et al., 2002; Poirazi and Mel, 2001; Yuste et al., 1994), and the
network at large, where ensembles of coactive neurons can be distributed
unpredictably in space (Cossart et al., 2003). Optical imaging techniques
are ideal for approaching these issues. In this chapter, we describe three
methods used to analyze the neuronal circuit at multiple levels. We describe
(1) combining whole-cell recording with two-photon calcium imaging and
electron microscopic reconstruction to examine the structure and function
of dendrites, (2) imaging hundreds of neurons in the brain slice simultane-
ously to examine the spatiotemporal dynamics of activity in living neuronal
networks, and (3) performing an unbiased, quantitative analysis of neuronal
morphology that is increasingly necessary in light of the multiparametric di-
versity of neuronal subclasses.
II. TWO-PHOTON CALCIUM IMAGING IN DENDRITES
Two-photon calcium imaging has revolutionized the study of dendrites,

allowing calcium signals to be examined in small structures such as dendritic
spines (Yuste and Denk, 1995) and aspiny dendritic shafts (Goldberg et al.,
2003c). In this section, we will discuss combining intracellular recording
with two-photon calcium imaging during synaptic activation. This combined
approach is a powerful method for examining the two main functions of den-
drites: chemical compartmentalization and synaptic integration. In these ex-
periments, neurons from mouse neocortex are patched in whole-cell mode
at physiologic temperature (37◦ C) and loaded via the patch pipette with a
calcium-dependent fluorophore, i.e., a calcium indicator. Electrical synap-
tic activity is monitored through the recording electrode, and dendrites are
visualized via two-photon excitation of the calcium indicator. Next, an elec-
trode to be used for synaptic stimulation is placed immediately adjacent
(ideally < 20 µm) to a specific dendritic region of interest (ROI). Electrical
shocks delivered through this electrode excite local presynaptic axons that
activate the dendrite of interest, and the resulting changes in intracellular
calcium concentration are monitored by changes in the fluorescence of the
calcium indicator.
To perform these experiments, we use a custom-made modified confocal
microscope (described in detail in Majewska et al., 2000; Nikolenko et al.,
2003), although turn-key commercial multiphoton microscopy systems are
now readily available. We will discuss the three major considerations in per-
forming these experiments: calcium indicator selection, targeting the stim-
ulation electrode to the dendritic ROI, and synaptic activation of single
synapses.
A. Selection of the Calcium Indicator
Calcium indicators are fluorophores whose fluorescence is depen-

dent on calcium concentration. Important characteristics to consider
when selecting an indicator are its resting fluorescence, its calcium-
dependent change in fluorescence, and the kinetics of its chemical re-
action with calcium. 1,2-Bis(o-aminophenoxy)ethane-N,N,N ,N -tetraacetic
acid (BAPTA)-based high-affinity calcium indicators have fast Kon (∼108
M/s) and are ideal for monitoring fast changes in calcium concentration,
such as during synaptic activation and action potential backpropagation. In
these experiments, the potassium salt form of the indicator is combined with
standard internal recording solution and loaded into the cell during whole-
cell recording. Typically, the dendritic tree is fully loaded approximately 20
min after gaining access to the cell.
Many different calcium indicators are available (Molecular Probes:
http://www.probes.com), and are well suited to different experimental ques-
tions. We have used the potassium salt forms of the high-affinity indicators
calcium green-1 (CG-1), Oregon green BAPTA-1 (OGB-1), and Fluo-4, and
all three are excited by a mode-locked laser at 800 nm. A major advan-
tage of CG-1 and OGB-1 is their high-resting fluorescence, which means
that they can be used at lower concentrations (50–150 µM), reducing both
exogenous buffer capacity and experimental misrepresentation of free cal-
cium signals (see below). However, in part due to their high fluorescence at
rest, their increase in intensity on binding calcium is compromised. Thus,
Fluo-4 is better for detecting small or heavily buffered signals, such as cal-
cium influx resulting from a single action potential or microdomains in
GABAergic interneurons (Goldberg et al., 2003c). The disadvantage of Fluo-
4 is that it is dim at rest, necessitating higher concentrations (100–400
µM) of the dye to visualize the dendritic tree. This issue can be circum-
vented by loading the intracellular electrode with the calcium-insensitive dye
Alexa594 in tandem with Fluo-4 during two-channel acquisition (Sabatini
et al., 2002).
In two-channel acquisition, two separate photomultiplier tubes (PMTs)
are used to collect red and green light. In these experiments, both calcium-
insensitive Alexa594 (25–100 µM) and calcium-sensitive indicator (see
above) are loaded into the recording pipette. A dichroic mirror separates
red and green emission light (e.g., 565 DCXR from Chroma Technology
Corp., Brattleboro, VT) to separate the calcium-insensitive fluorescence
from Alexa594, and the calcium-sensitive fluorescence from the Fluo-4. At
an angle of 45◦ , this filter transmits ∼80% of light between ∼580 and 880 nm
(Red Channel: infrared light from laser to sample and red Alexa594 fluo-
rescence back to internal PMT) and reflects light below 530 nm (Green
Channel: emission from calcium indicator) to external PMT. We place an
additional band pass 510/40 (transmits between 490 and 530 nm) filter in
front of the external PMT to reduce scattered red light. Thus, morphol-
ogy and calcium signal are represented on the red and green channels,
respectively. The advantage of using a separate PMT for each signal is that
low levels of calcium indicator can be used, reducing the experimental per-
turbation of calcium dynamics (see section “Summary of Advantages and
Limitations”).
B. Targeting the Stimulation Electrode to the Dendrite of Interest
Stimulating and detecting calcium signals in dendrites at individual

synapses is in part a matter of chance, since the stimulation electrode must
overlie an axon that happens to target the dendritic segment being imaged
(Fig. 15.1). For this reason, the stimulation electrode has to be frequently
moved to several locations even within the small vicinity of the dendrite of in-
terest, but without compromising the quality of the electrical recording, i.e.,
without causing even slight movement of the brain slice. Thus, in order to
reliably activate single or a limited number of synapses that converge onto a
dendritic ROI, the stimulation electrode must be carefully constructed, con-
trolled, and mounted on a stable micromanipulator. We use a borosilicate
glass patch pipette (6–10 M) bent at the tip by 60–80◦ with a microforge
(Narishige, Japan). Bending the tip allows the electrode to enter the slice
perpendicularly, reducing tissue deformity and allowing the electrode to
Figure 15.1. Two-photon imaging of calcium microdomains in aspiny dendrites. Top,

interneuron from a slice of primary visual cortex filled with 100 µM Fluo-4. Stim-
ulation electrode S was placed via visual guidance near dendrite of interest. Scale
bar: 20 µm. Middle: dendritic segment as indicated by white box. Scale bar: 1 µm.
Bottom: line scan placed in parallel through dendritic segment from above, and the
calcium signal evoked by a single shock delivered through the stimulation electrode.
Image is an average of four trials. Vertical scale: 400 ms; horizontal scale: 1 µm as
above. (Reprinted, with permission, from Goldberg et al., 2003c).
be repeatedly placed in the slice without compromising the quality of the
whole-cell recording. Once constructed, the electrode is filled with a fluores-
cently labeled fluorophore [we use 20–200 µM Alexa488-dextran in standard
artificial cerebrospinal fluid (ACSF)]. The dextran form of Alexa works well
because it is not ejected from the electrode during electrical stimulation.
This fluorescent labeling of the stimulation electrode is essential in order to
visualize it during the experiment, allowing for precise control over where
it is placed.
Before patching the neuron of interest, the stimulation electrode should
be mounted on the micromanipulator, placed with its curved tip oriented
toward the slice, adjacent to the patch electrode, to ensure that the objective
can focus on the tip without physically touching its proximal segment.
Once the neuron is patched and the dendritic tree is filled with indicator,
a region of the dendrite oriented in parallel to the position of the line scan
is chosen (Fig. 15.1). With either spiny or aspiny dendrites, this approach
maximizes the surface area of dendrite shaft (or number of spines) imaged
and increases the likelihood of overlying an activated synapse.
With the dendritic segment of interest placed at the center of the imaging
field, the objective is brought out of the slice until the fluorescently labeled
tip of the stimulation electrode is in focus. The electrode is moved in the x-
and y -axes until it is near the center, above the dendrite of interest, such that
as it is gently lowered into the slice it will fall immediately next (< 15 µm)
to it. Control of the stimulation electrode is perhaps the most crucial and
experience-dependent part of the experiment.
C. Synaptic Activation of Single Synapses
The hallmark of a single synaptic calcium signal is the observation of

successive failures and successes of a local calcium signal on repeated tri-
als, reflecting the stochasticity of synaptic transmission (Goldberg et al.,
2003c; Yuste and Denk, 1995). Once the stimulation electrode is in position,
adjacent to the dendrite of interest, it delivers single shocks (100–200 µs)
in voltage mode (0.1–2 V) using a stimulus isolation unit (A.M.P.I Iso-Flex:
http://www.ampi.co.il/isoflex.html). First, start with a small voltage setting
(0.1 V) and monitor the excitatory postsynaptic potentials and currents
(EPSP/C) response to observe if the neuron is being activated. Gradu-
ally ramp up the stimulus strength until calcium signals and subthresh-
old EPSP/Cs are observed. Frequently, no calcium signals will be observed
until suprathreshold activation is achieved, reflecting strong activation of
synapses off of the imaged dendrite of interest and global calcium accumu-
lations secondary to action potential backpropagation.
If this occurs, or if repeated shocks fail to induce single synaptic calcium
signals, the electrode must be lifted out of the slice, repositioned, and again
dropped next to the dendrite of interest in its new position. This process
may need to be repeated several times during an experiment.
Lastly, it is important to control direct activation of the dendrite by the

stimulation electrode, as opposed to activation of axons that synapse on the
dendritic segment under investigation. This may occur if the electrode is too
close to the dendrite of interest (< 5 µm) and may result in global calcium
signals and undershooting action potentials that may reflect the dendritic
spiking. One way to control this artifact is by blocking synaptic transmission,
such as with bath application of 20 µM DNQX, 50 µM d-APV, and 1 µM
bicucculine, at the end of an experiment to ensure that the postsynaptic
potential and the calcium signal are blocked. If a fast calcium signal persists,
then either the stimulation strength was too high or the electrode was too
close to the dendrite, and the data must be discarded.
III. ULTRASTRUCTURAL IDENTIFICATION OF IMAGED

DENDRITIC SEGMENTS
During whole-cell recording, 0.3% biocytin is loaded into the recording

pipette in addition to the calcium indicator. Standard procedures for the
visualization of biocytin (Buhl et al., 1994; Horikawa and Armstrong, 1988;
Tamas et al., 1997) can be used for the light and electron microscopic as-
sessment of dendritic segments imaged previously with two-photon lasers
(Fig. 15.2). Importantly, our experience shows that successful detection of
biocytin is enhanced by restricting the number of scans (< 20) through a
particular dendritic segment, presumably due to the photosensitivity of bio-
cytin. The step-by-step protocol for electron microscopic reconstruction of
imaged dendritic domains is presented in Appendix.
IV. IMAGING NEURONAL ENSEMBLES
The understanding of neuronal circuits has been, and will continue to be,
greatly advanced by the simultaneous imaging of hundreds of neurons in
the brain slice allowing for the examination of the spatiotemporal dynam-
ics of activity in neuronal networks. In this section we describe the “bulk”
loading of brain slices with acetoxy-methyl (AM) ester calcium indicators,
in contrast to the calcium indicator salts described above, in order to image
activity in large populations of neurons simultaneously. In our experience,
voltage-sensitive dyes are still insufficient for the detection of activity in an
individual neuron within a population of neurons, such as a cortical slice,
due to their nonspecific staining pattern and poor spatial resolution (Yuste
et al., 1997). Calcium indicators (Tsien, 1989) that can be bulk-loaded into
brain slices using their AM ester derivatives act as very good, albeit indi-
rect, measures of action potential generation (Smetters et al., 1999; Yuste
and Katz, 1991) and provide single-cell resolution. Calcium entry via
calcium channels resulting from the depolarization indicative of an action
potential is sufficient to be imaged. These dyes still provide the best means of
Figure 15.2. Ultrastructural reconstructions of imaged dendrites. (Aa) Reconstruc-

tion of biocytin-filled layer V fast spiking (FS) cell from a 300-µm-thick visual cortical
mouse brain slice. Dendrites, orange; axon, black. Red dendritic segment repre-
sents ROI in (Ab). (Ab) Two-photon z projection of imaged ROI of an FS cell filled
with 100-µM Fluo-4 (left) and corresponding region from the cell reconstruction
(right). Boxes indicate the dendritic segment selected for line scan imaging and for
expectation-maximization (EM) reconstruction. Scale bar: 20 µM. (Ac) Top: hori-
zontal dendrite of interest with the cartoon of the serial EM reconstruction overlay
at the precisely realigned section. Bottom: line scan through dendrite reveals the
evoked single synaptic calcium signal. Note how its position appeared aligned to the
synapse, as indicated by the red arrow in the cartoon. (Ad) Top: cartoon detail of
the serial reconstruction. Dendrite, d, is labeled in green, and terminals, t, in white.
The terminal labeled by “t” corresponds to the terminal of interest in Ac. Bottom:
the electron micrograph focusing on the site aligned to the microdomain; arrows
indicate synapses. (Reprinted, with permission, from Goldberg et al. 2003c). See
Appendix for step-by-step protocol of how imaged dendrites were reconstructed.
imaging activity in large populations of neurons when single-cell resolution

is desirable.
Since first used (Yuste and Katz, 1991), this method has been successfully
applied throughout the central nervous system to study neuronal ensemble
activity. Reports include studies on neocortex (Cossart et al., 2003; Mao
et al., 2001), hippocampus (Tanaka et al., 2002), cerebellum (Ghozland et al.,
2002), striatum (Mao and Wang, 2003), and spinal cord (Voitenko et al.,
1999), utilizing acute slices, cultured slices, or cultured dissociated neurons.
Recently, the technique has also been applied in vivo to mouse, rat, and cat
neocortex (Ohki et al., 2005; Stosiek et al., 2003).
The limitations of the technique are due to the properties of the dyes
themselves. Calcium indicators, being charged molecules, do not easily cross
the cell membrane and need, therefore, to be microinjected. To circum-
vent this problem, AM ester derivatives of the indicators were synthesized
(Tsien, 1981). The AM esters mask negative charges, making the indicator
molecules more lipophilic and membrane-permeant, thus allowing them to
enter the cell. Once inside the cell, cytoplasmic esterases hydrolyze the acetyl
ester linkage, releasing formaldehyde and free indicator, which then accu-
mulates intracellularly as it is once again charged. However, the dependence
on intracellular enzymatic cleavage makes this process cell dependent. This
can result in differential loading efficiency in different neurons. In addition,
the increased hydrophobicity of the AM ester derivatives of the indicators
can cause problems in delivering sufficient amounts to their targets. This
problem becomes significant in adult preparations, where the slice paint-
ing method appears to be the best loading strategy. Finally, while the time
constant (i.e, rate) for the onset of quenching indicator fluorescence in
neurons is rapid, the offset is proportionally slow due to saturation of the
dye. Thus, while the calcium indicators are excellent measures for the onset
of activity, they do not provide adequate temporal resolution to detect sin-
gle action potentials during a burst of action potentials. This is in contrast
to the voltage-sensitive dyes, which provide rapid onset and offset of sig-
nal. The trade-off between the two methods then is spatial versus temporal
resolution, with the calcium indicators providing a far superior spatial reso-
lution and the voltage-sensitive dyes providing greater temporal resolution,
especially when examining the offset of activity.
A. “Bulk” Loading Acute Cortical Slices with Calcium Indicator
Different AM calcium indicators, under similar experimental conditions,

have different loading efficacies. Further, indicators also vary in their effec-
tiveness as detectors of action potentials, determined by the Kd and dynamic
range of the dye, and importantly, in their affinity for neurons versus glia.
The majority of experimental procedures in our laboratory use Fura-2, AM,
which provides sufficient sensitivity to reliably detect, at minimum, two ac-
tion potentials and is by far the best of the AM dyes at targeting neurons
for loading. However, the methods described here can be used for any of
the commercially available AM calcium indicators. Further, although the
methodologies described below are for mouse brain slices, bulk loading
of calcium indicators into neurons from rat and cat has been successfully
applied by other groups (Ohki et al., 2005).
Both juvenile and adult acute mouse cortical slices (PND10–PND30) can
be loaded with the calcium indicators. Prepare the Fura-2, AM by dissolv-
ing 50 µg Fura-2, AM (Molecular Probes) in 15–48 µl DMSO and 2 µl of
Pluronic F-127 (Molecular Probes) for a final concentration of 1–3.3 mM.
The solubility of calcium AM indicators is rather poor. Thus, it is necessary
to vortex the solution for 10–15 min prior to use. Place slices in loading
chamber (a petri dish 35 × 10 mm) containing 2.5 ml of oxygenated ACSF.
Pipette 5–10 µl of Fura-2, AM solution on top of each slice, resulting in a
high initial concentration of Fura-2, AM. The concentration decreases as
the Fura-2, AM diffuses away from the site of application, resulting in a final
concentration in the entire chamber of approximately 10–20 µM. Alterna-
tively, simply pipette the entire volume of the prepared Fura-2, AM at the far
side of the loading chamber and allow the dye to passively diffuse through-
out the chamber lading the slices. Once again this is the preferred method
especially if the slices are from a younger animal. Load the slices in the
dark for 20–30 min at 35–37◦ C with 95% O2 /5% CO2 lightly ventilated into
the chamber. As a rule of thumb, the loading time should be 10 min, plus
as many minutes as the age of the animal in postnatal days. Remove slices
from the loading chamber and place into incubation chamber containing
oxygenated ACSF to allow them to rest (for more details see Sakmann and
Neher, 1983). Imaging may begin after a 30-min recovery period.
B. Imaging Ensemble Activity
Two-photon imaging of calcium indicators allows highly sensitive detec-

tion of changes in neuronal calcium concentration with relatively little
bleaching and photodamage. However, the simultaneous imaging of large
populations of neurons is necessarily slow. Epifluorescent imaging of cal-
cium indicators is sufficient to detect changes in neuronal calcium concen-
tration and has the advantage that, using a fast camera, the detection of
changes can be of higher temporal resolution than in the two-photon sys-
tem. However, fluorescent imaging of bulk-loaded slices is subject to rapid
photobleaching. Finally in our experience, spinning disk confocals, together
with fast cameras, can be used to image thousands of neurons simultane-
ously, without significant photobleaching and with good signal-to-noise ra-
tios, over long periods of time. In all three cases, cell activity, as indicated by
fluorescence change, can be detected automatically, using custom written
software (MatLab) allowing one to evaluate the spatiotemporal pattern of
activity in the slice and in the neuronal ensembles. The software analyzes
images of neurons, which are filtered and thresholded—round contiguous
areas of brightness are recognized as neuronal somata. Activity is detected
as a decrease (recall that fura decreases its fluorescence on binding calcium)
in neuronal somata brightness that is greater than the surrounding area by
some arbitrary number (determined using calibration experiments on each
individual experimental setup) of standard deviations above the average
(noise) fluctuation level.
For example, Fig. 15.3 illustrates the two-photon imaging of Fura-2, and
the utilization of Fura-2, as an indicator of neuronal activity. Figure 15.3A
illustrates thousands of loaded neurons in a neocortical slice imaged us-
ing a two-photon microscope. Figure 15.3B reveals the correlation between
Figure 15.3. Calcium imaging of neuronal activity. (A) Two-photon image of 300-
µm-thick transverse section of mouse somatosensory neocortex loaded with Fura-2,
AM. Scale bar, 50 µm. Note how hundreds of neurons can be visualized. (Cour-
tesy of Brendon O. Watson.) (B) Correspondence between action potentials and
somatic fluorescence change. (i) Whole-cell recording of a burst of action poten-
tials in response to a single intracellular depolarizing current step (150 pA, 200 ms).
(ii) Normalized fluorescence change for the recorded neuron imaged with a cooled
CCD camera (Micromax, Princeton Instruments). Recording pipette contained
25 µm of fura pentapotassium salt (comparable to the intracellular concentration of
fura following bulk loading). While individual action potentials cannot be resolved
at such a high firing frequency, onset of neuronal activity is accurately detected.
(C) This fluorescent image can then be analyzed to detect spontaneously coactive
neurons, as indicated by the filled contours.
neuronal activity and a change in fluorescence as detected by epifluores-
cence. The fluorescence of Fura-2, decreases in the presence of calcium,
thus the negative fluorescent change in Fig. 15.3B (ii). As illustrated, even
one-photon imaging of Fura-2, is sufficient to detect and resolve action po-
tential generation in a single neuron within a field of hundreds or a thousand
loaded cells. Because of the rapid time course for the onset of its response,
Fura-2, lends itself to the elucidation of network dynamics when simultane-
ously imaging large populations of neurons. Figure 15.3C shows an example
of automatically detected ensemble activity. The contours, corresponding to
the somata of imaged neurons, which are filled indicate a group of neurons,
which were simultaneously active.
V. QUANTITATIVE CLASSIFICATION OF NEURONAL

MORPHOLOGY USING PRINCIPAL COMPONENT ANALYSIS
The goal in this section is to introduce principal component analysis

(PCA) in simple terms and provide a practical guide for its use in the anal-
ysis of morphometric data. It is not intended to provide a rigorous intro-
duction to PCA nor a comprehensive review of its many applications in
data analysis. There are excellent publications addressing these two issues
( Jolliffe, 2002).
The concept of PCA was introduced by Pearson (1901) and later indepen-
dently developed by Hotelling (1933). PCA is a statistical method for trans-
forming a set of data with presumably large number of correlated variables
to a more parsimonious set, yet preserving as much original information as
possible. Given a set of n cases with p measured variables (attributes), PCA
is aimed at reconstructing a set of k uncorrelated variables, where k p (k
is much smaller than p). These k uncorrelated variables are called principal
components (PCs). PCs can be derived using the covariance or the correla-
tion matrix of data. In this section, for reasons that will be mentioned later,
the derivation of PCs is solely based on the correlation matrix.
PCA offers several advantages with respect to the analysis of high-
dimensional data where numerous attributes need to be considered
simultaneously. The aim in the analysis of the neuronal morphometric data
using PCA is to describe the variance structure of the data in terms of a
few morphological indices in order to facilitate detection of clusters and/or
outliers.
A. Fundamental Concepts of Principal Component Analysis
If a large number of morphometric variables are measured for a set of n

cases, and if these measured variables are correlated with each other, then it
is possible to derive a limited number of uncorrelated variables named prin-
cipal components that could faithfully represent the data. The mathematical
operations underlying PCA ensure that PCs represent the maximum vari-
ance of the original data.
A PC is a linear combination of the original variables x1 , x2 , x3 , . . . , x p
in the form y i = ai1 x1 + ai2 x2 + ai3 x3 + · · · + ai p x p , where the coefficients
ai1 , ai2 , ai3 , . . . , ai p are derived through the mathematical operations that
maximize the variance of each PC y i . Hence, the first PC accounts for the
largest proportion of variance, the second PC for a smaller proportion, etc.
Therefore, it is possible to represent a significant proportion of the original
variance in terms of only a few PCs.
From a geometrical point of view, in a multidimensional variable space
occupied by cases, a PC is the best fit line through the swarm of points.
That is, the sum of squared distances from all the points to that line is at
its minimum. In this sense, the first two PCs define the best fit 2D plane to
the swarm of points in the original p-dimensional space. Similarly, the first
three PCs define the best-fit 3D hyperplane.
Using this geometrical interpretation, one can imagine how PCA can
reduce the dimensions necessary to represent the data. The upper bound
on the number of PCs is the number of original variables. The number of
PCs k can take any integer value from 1 to p; however, it is usually enough
to choose k much smaller than p and still be confident that the structure
of data remains intact. The ability to reduce dimensionality comes from
the interrelatedness of the original variables. If the original variables are
uncorrelated with each other, PCA will not be able to reduce the number
of dimensions. Hence, one needs to examine cross correlations of variables
before applying this method.
B. Example of Application
Let us apply the method of PCA to a set of neuronal morphometric

data acquired through the NeuroExplorer software (MicroBrightField Inc.,
Williston, VT). The data contain 82 morphometric parameters (both raw
and derived) from 67 neocortical neurons, filled with biocytin and re-
constructed in three dimensions using Neurolucida. The morphological
parameters include dendritic, somatic, and axonal variables. The correla-
tion matrix was then computed, resulting in an 82 × 82 matrix (partly shown
in Table 15.1). The correlation matrix shows both positive and negative val-
ues of correlation among variables. Thus, it is appropriate to apply PCA to
this type of data hoping that one can reduce the 82-dimensional variable-
space to a smaller and more manageable number of dimensions.
Table 15.2 shows some of the first eigenvectors and eigenvalues of the
correlation matrix. The variance of each PC is indicated by its associated
eigenvalues. For example, the third PC y 3 = 0.030× (somatic perimeter)
+(−0.023) × (somatic area) + 0.240 × (total axonal node) + · · · + a p x p .
The third PC accounts for a variance of 6.740. The total variance ac-
counted for by all the original variables is 82, the same as the number
Table 15.1. Correlation matrix of morphometric data (only the first 12 variables shown).
SP SA ANT TAL TAA ALTA ND TDN TDL ALD TAD DLTA
Somatic perimeter 1.000 0.790 0.149 0.328 0.295 0.057 0.164 0.118 0.420 0.323 0.455 −0.092
Somatic area 1.000 0.206 0.461 0.439 −0.026 0.283 0.128 0.446 0.290 0.611 −0.297
Total axonal node 1.000 0.803 0.770 −0.083 0.163 −0.010 −0.008 −0.110 0.176 −0.214
Total axonal length 1.000 0.943 −0.024 0.256 −0.087 0.193 −0.001 0.340 −0.148
Tile area of axon 1.000 −0.346 0.246 −0.086 0.131 −0.037 0.315 −0.229
Axonal length/tile area 1.000 −0.018 −0.001 0.139 0.081 −0.036 0.307
Number of dendrites 1.000 0.385 0.418 −0.173 0.376 −0.029
Total dendritic node 1.000 0.665 0.528 0.477 0.134
Total dendritic length 1.000 0.789 0.774 0.202
Average length of dendrites 1.000 0.600 0.207
Tile area of dendrites 1.000 −0.335
Dendritic length/tile area 1.000
Table 15.2. Eigenvectors and eigenvalues of the correlation matrix (only 12 shown).
a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 a11 a12
Somatic perimeter −0.005 0.203 0.030 −0.053 −0.014 −0.167 −0.182 0.020 0.028 0.084 −0.056 −0.142
Somatic area −0.050 0.202 −0.023 0.089 0.061 −0.058 −0.121 0.053 −0.036 0.103 −0.136 −0.053
Total axonal node −0.155 0.044 0.240 0.106 0.027 −0.011 −0.081 0.082 −0.066 −0.005 0.075 −0.070
Total axonal length −0.116 0.129 0.177 0.114 0.053 0.046 −0.190 −0.018 −0.093 0.180 −0.030 −0.030
Tile area of axon −0.130 0.107 0.148 0.107 0.037 0.062 −0.236 −0.013 −0.079 0.195 0.008 0.060
Axonal length/tile area 0.078 0.048 0.060 −0.015 0.046 −0.044 0.179 −0.039 −0.025 −0.056 −0.097 −0.284
Number of dendrites −0.029 0.051 −0.022 0.228 −0.036 −0.132 −0.029 −0.035 0.113 0.151 −0.093 −0.104
Total dendritic node 0.044 0.072 −0.029 0.215 −0.191 −0.207 0.124 0.067 0.026 −0.056 0.112 0.145
Total dendritic length 0.062 0.224 −0.001 0.160 −0.072 −0.182 0.069 −0.066 −0.046 0.055 0.025 0.152
Average length of dendrites 0.078 0.194 −0.002 0.055 −0.071 −0.124 0.078 −0.022 −0.098 −0.077 0.083 0.261
Tile area of dendrites −0.028 0.247 −0.048 0.159 −0.063 −0.071 0.020 0.085 0.005 0.031 −0.081 0.067
Dendritic length/tile area 0.111 −0.067 0.076 0.010 0.005 −0.149 0.069 −0.250 0.001 0.103 0.117 0.095
OPTICALLY PROBING THE LIVING BRAIN SLICE
Variance (eigenvalue) 15.173 9.348 6.740 5.842 4.741 4.292 3.461 3.228 2.846 2.535 2.167 1.971
465
of variables, since the variance of each standardized variable is equal to

1. One can then compute the quality of representation by assessing how
much of the original variance is accounted by the selected PCs. For exam-
ple, sum variance of the first 12 PCs can be computed by adding up the first
12 eigenvalues, 15.173 + 9.348 + 6.740 + 5.842 + 4.741 + 4.292 + 3.461 +
3.228 + 2.846 + 2.535 + 2.167 + 1.971 = 62.344. Thus, the number of di-
mensions can be reduced from 82 to 12, yet keeping 62.344/82 or about
76% of the total variance of the data set.
Correlations of original variables with PCs are given in Table 15.3. These
correlations are computed by multiplying each eigenvector column by the
square root of its eigenvalue. These correlations are sometimes called load-
ings, a term borrowed from factor analysis (FA). Variables that highly cor-
relate with a PC determine much of the “character” of that component and
can be useful in interpreting PCs.
1. How Many Principal Components Should Be Retained?
There is no definite answer to this question. The number of PCs sufficient

for an adequate description of the data depends on both the structure of the
data and the goal of the analysis. However, guidelines, mostly based on intu-
ition, have been suggested in choosing a subset of PCs. Since each individual
variable in its standardized form contributes 1 to the total variance, it seems
reasonable to exclude PCs that account for less than 1 unit of variance, i.e.,
have an eigenvalue less than 1 (Kaiser, 1960). However, Kaiser’s criterion
may not always hold and could lead to exclusion of some important PCs
having low eigenvalue. In order to address this problem, Jolliffe (1972) has
suggested a cutoff value of 0.7 for the variance of PCs. Based on simulation
studies, PCs of sample data with variance greater than 0.7 correspond to
population PCs with variance greater than 1 and thus should be retained.
An alternative graphical method for choosing PCs is called a “scree” plot
(Cattell, 1966). Variance of each PC is plotted against the ordinal number
of each PC and those PCs that make the steeper part of the plot above the
first inflection point are selected. The problem with the scree plot is that,
depending on the data structure, there may not be any clear inflection point
to use as a guide. If the sole purpose of PCA is to account for a significant
portion of the total variance, one can arbitrarily choose a value such as 90%
and include enough PCs to achieve this level of significance. All the above
methods should be taken only as a guide, since there can be significant
differences in the number of PCs retained by one method versus the others.
2. Reduction of Dimensionality by Choosing a Subset

of Original Variables
For reasons that are outside the scope of this chapter, one may be inter-
ested in selecting a subset of the original variables rather than their linear
Table 15.3. Correlation of variables with individual PCs.
PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9 PC10 PC11 PC12
Somatic perimeter −0.021 0.621 0.078 −0.128 −0.031 −0.347 −0.340 0.036 0.048 0.133 −0.083 −0.199
Somatic area −0.195 0.619 −0.059 0.215 0.132 −0.120 −0.225 0.095 −0.060 0.164 −0.201 −0.075
Total axonal node −0.602 0.135 0.622 0.256 0.060 −0.023 −0.151 0.148 −0.111 −0.008 0.111 −0.098
Total axonal length −0.453 0.394 0.461 0.276 0.115 0.095 −0.353 −0.032 −0.157 0.287 −0.045 −0.042
Tile area of axon −0.508 0.328 0.384 0.259 0.081 0.128 −0.438 −0.023 −0.133 0.310 0.012 0.084
Axonal length/tile area 0.303 0.148 0.155 −0.036 0.099 −0.092 0.333 −0.069 −0.042 −0.089 −0.142 −0.399
Number of dendrites −0.111 0.157 −0.058 0.551 −0.078 −0.274 −0.055 −0.063 0.190 0.240 −0.136 −0.147
Total dendritic node 0.171 0.219 −0.076 0.520 −0.416 −0.430 0.231 0.120 0.045 −0.089 0.165 0.203
Total dendritic length 0.241 0.684 −0.004 0.386 −0.157 −0.376 0.129 −0.119 −0.078 0.087 0.036 0.213
Average length of dendrites 0.303 0.593 −0.006 0.132 −0.155 −0.257 0.145 −0.040 −0.166 −0.123 0.122 0.367
Tile area of dendrites −0.110 0.754 −0.124 0.384 −0.137 −0.146 0.038 0.153 0.008 0.049 −0.119 0.094
Dendritic length/tile area 0.431 −0.205 0.197 0.025 0.011 −0.308 0.128 −0.449 0.002 0.165 0.173 0.133
OPTICALLY PROBING THE LIVING BRAIN SLICE
467
combination in the form of PCs ( Jolliffe, 1972). One strategy is to retain

one variable from each retained PC. The variable with the highest loading
or correlation with the PC is retained. Conversely, one can exclude variables
that are highly correlated with the excluded PCs and keep the remaining
variables.
3. Detection of Clusters and Outliers
High dimensionality of the data precludes most of the common measures

used for data visualization. PCA provides a parsimonious description of the
data with a limited number of dimensions allowing questions to be asked
regarding the existence of clusters, their topology, and the nature of out-
liers. This is an especially important issue in classification since methods
of cluster analysis will always produce clusters whether or not these clus-
ters “naturally” exist. Similarly, PCA facilitates detection of outlying and/or
influential cases so that they can be eliminated before cluster analysis is
performed. The choice of the clustering algorithm makes a significant dif-
ference in the composition of the clusters. Certain clustering algorithms
(e.g., Ward’s method) will detect mostly spherical clusters while others
might be biased toward the detection of more oblong clusters (e.g., single
linkage). PCA can verify the existence of clusters and provide clues about
their topological features so that appropriate clustering algorithms can be
used.
As a logical step after performing PCA, cluster analysis was used to de-
fine clusters of morphologically reconstructed neurons. Multiple clustering
algorithms (K-Means, Hierarchical Agglomerative) produced similar clus-
ter structures. An instance of cluster analysis utilizing Ward’s method and
squared Euclidean distances is shown in Fig. 15.4. The majority of neu-
rons containing neuropeptide Y or paravalbumin are clustered together.
The parameters used for clustering were strictly limited to morphological
measurements from these neurons.
VI. SUMMARY OF ADVANTAGES AND LIMITATIONS
A. Two-Photon Imaging of Dendritic Calcium Microdomains
Advantages of two-photon excitation include reduced phototoxicity,

higher spatial resolution, and deeper penetration into the sample (for more
discussion see Denk and Svoboda, 1997; Denk et al., 1990). These advantages
are especially important for imaging small structures, such as dendrites and
spines, which frequently lie deep in tissue and are sensitive to photodamage.
In addition to calcium-dependent fluorophores, voltage-sensitive dyes
(Antic et al., 1997) and sodium-dependent indicators (Rose et al., 1999;
Rose and Konnerth, 2001) have been used to functionally image dendrites.
Figure 15.4. Dendrogram showing the five-cluster structure of the 67 morphologically

reconstructed neurons. Neurons were assigned to clusters such that the sum of
squared deviations from the mean values of parameters for each cluster is minimized.
Squared Euclidean distances were used as a measure of dissimilarity among neurons
and their clusters. Thus, the greater the similarity between neurons, the smaller the
resulting Euclidian distance.
Calcium imaging offers several advantages. First, a large diversity of calcium

indicators are available, and they offer good signal-to-noise ratio. Second,
action potential backpropagation and subsequent opening of voltage-gated
calcium channels appear to be a general function of dendrites. Thus, cal-
cium imaging can be used as an indirect measure of dendritic voltage prop-
agation (Goldberg et al., 2003a; Spruston et al., 1995; Waters et al., 2005).
Third, calcium is of particular interest because of its pluripotent role as a
second messenger. As the major ion that both follows electrical potentials
and binds proteins, it connects the electrical and organic worlds in excitable
tissue. Thus, the spatial extent of a calcium signal offers key insights into the
potential range of action of a given stimulus (Euler et al., 2002; Goldberg and
Yuste, 2005; Sabatini et al., 2002). However, there are important limitations
of calcium imaging in dendrites, and they must be understood to accurately
interpret calcium fluorescent signals. Experimentally observed calcium
fluorescence changes do not reflect native calcium signals; rather, they rep-
resent the chemical reaction between calcium and the indicator. Because
high-affinity indicators bind virtually every ion that enters the cytoplasm, cal-
cium extrusion, diffusion, and interactions with endogenous proteins are
disturbed. Thus, fluorescent calcium signals differ significantly from “true”
free calcium signals that would exist in the absence of indicator (for ex-
cellent discussion of these issues see Helmchen, 1999). First, because the
unbinding process between calcium and its indicator is slow, calcium flu-
orescence transients are significantly slower than the free calcium signals
they represent (Helmchen, 2002; Neher, 1998). Second, by binding cal-
cium, indicators inhibit extrusion pumps, which in dendrites can act on fast
time scales to control both the amplitude and the spatial range of calcium
signals. Thus, slower calcium influxes, e.g., by N-methyl-d-asparate recep-
tors, are overrepresented in amplitude, time, and space relative to faster
influxes, such as through voltage-gated calcium channels during action po-
tential backpropagation (Goldberg et al., 2003a; Sabatini et al., 2002). Third,
calcium diffusion in dendrites is controlled by endogenous buffers, many
of which are fixed (Allbritton et al., 1992). The presence of highly mobile
calcium indicator can thus distribute calcium over unphysiologically long
ranges, such as through spine necks or along dendritic shafts. Lastly, calcium-
dependent processes such as long-term potentiation (LTP) and long-term
depression (LTD) may be disturbed by indicators that by binding calcium
interfere with its normal signaling cascades. Thus, one must be careful to
examine the functions of calcium in the presence of indicator.
B. Imaging Neuronal Ensembles
The major advantage of bulk loading calcium imaging is the excellent spa-
tial resolution. In contrast to network imaging using voltage-sensitive dyes
(Petersen and Sakmann, 2001), our technique has the ability to resolve sin-
gle neurons. This is fundamental to understanding the spatial organization
of coactive neural ensembles (Cossart et al., 2003, 2005; Ikegaya et al., 2004).
However, the disadvantage of this methodology is the relatively poor tem-
poral resolution, particularly if the cessation of activity is an important vari-
able. Calcium indicators have a relatively slow decay time constant and as a
result the dyes are best utilized as indicators of the onset of activity rather
than the offset. In this regard, voltage-sensitive dye imaging has improved
temporal resolution and is better suited to questions focusing on the timing
of network activity over the participation of individual cells (Petersen and
Sakmann, 2001; see chapter by Carl C. H. Petersen in this volume).
C. Principal Component Analysis Versus Factor Analysis
Although PCA and FA can both reduce the dimensionality of a data set,
they achieve this goal in ways that are fundamentally different. PCA assumes
no explicit statistical model underlying the observed variance of the original
data. The goal is maximal representation of the total variance by successive
PCs. However, in FA, according to the underlying statistical model, the orig-
inal variance is decomposed (partitioned) to common (shared among fac-
tors) and unique (accounted for by a single specific factor) components.
Unique variances are not of interest in FA. In contrast, FA describes the com-
mon variance (buried in the off-diagonal elements of the correlation matrix)
in terms of underlying factors. PCA and FA are techniques to explore dif-
ferent aspects of the correlation matrix, with PCA focusing on the diagonal
elements of this matrix and FA focusing on the off-diagonal elements.
We will conclude this section with a word of caution. PCA, in the context
presented in this chapter, provides a tool for descriptive data analysis of
a sample. Therefore, much of the analysis deals with correlation matrices.
Making inferences about a population, using PCA analysis of a population
sample, requires analysis of covariance matrices and multivariate normality
of the data. The former introduces significant problems in the analysis of
morphometric data and the latter condition is usually not satisfied.
APPENDIX
A. Two-Photon Imaging of Dendritic Calcium Microdomains
1. Making the stimulation electrode is time consuming and should be

done before brain slices are prepared. Pull a borosilicate glass pipette
to 6–10 M, and bend the tip 60–80◦ using a microforge. Make sure that
the tip remains patent and that the segment distal to the curve is not
longer than the working distance of the objective to be used in the
experiment. Fill the pipette with standard ACSF and a fluorophore of
choice (we use 20–200 µM Alexa488-dextran). Mount the electrode
on the micromanipulator, and image the tip of the electrode to ensure
that it can be visualized.
2. Prepare brain slices (we make slices 300 µm in mice aged 13–17 days)
and intracellular electrodes (borosilicate glass pipettes pulled to 6–
10 M). Fill the intracellular electrode with standard internal pipette
solution and a calcium-dependent fluorophore (we use Fluo-4 at a con-
centration of 50–400 µM), and bring both the intracellular electrode
and stimulation electrode into the field, in focus just above the brain
slice. Patch the neuron of interest with the intracellular electrode and
hold the cell for 20 min to allow time for the fluorophore to diffuse
into the dendritic tree.
3. Next, bring the stimulation electrode to the immediate vicinity of the
ROI. Since moving the stimluation electrode in the slice can compro-
mise the quality of the intracellular recording, all x- and y -axes position-
ing of the stimulation electrode should be done above the slice. To do
this, visualize the dendritic tree, and zoom into a ROI. For imaging mi-
crodomains on aspiny dendrites, it is important to find a dendritic seg-
ment that runs parallel to the orientation of the line scanner. Maintain
the x and y positions of the scanner at the ROI, zoom out, and come out
of the slice in the z-axis until the fluorescently labeled tip of the stimula-
tion electrode is in focus. Position the stimulation electrode, currently
out of the slice, at the x and y coordinates corresponding to the ROI.
4. Once the stimulation electrode is in focus above the slice at x and y
coordinates above the ROI, refocus on the dendrite and gently lower
the stimulation electrode until it is in view, adjacent to dendrite to be
imaged (see stimulation electrode, labeled S, in Fig. 15.1). It should
be clear that the advantage of taking the time to bend the electrode
tip in Step 1 is that bringing the electrode into the slice does not
deform the tissue or interfere with the intracellular recording.
5. Set the stimulation strength. Deliver brief (100 µs) single shocks
through the stimulation electrode, and adjust the stimulation strength
so that there are subthreshold EPSPs recorded by the intracellular
electrode.
6. Next, image the ROI during these single shocks. Reliable stimulation
of a single synapse on the imaged dendritic segment is difficult
and in part a matter of chance. Activation of a single synapse on
the imaged dendrite should result in all-or-none localized calcium
signals on successive trials, reflecting the stochasticity of synaptic
transmission. Achieving this result depends on perfectly adjusting the
stimulation strength. If it is too low, there may be no calcium signal
on the dendrite, even though EPSPs are recorded at the soma. This
may occur if the axons triggered by the stimulation electrode do not
synapse on the imaged dendritic segment. If this occurs, increase
the stimulation intensity, staying within the subthreshold range, until
calcium signals are observed. Alternatively, if stimulation strength
is too high, nonlocalized signals may be observed along the entire
dendritic region. This may occur if stimulation triggers multiple
synapses converging onto the branch (Goldberg et al., 2003b) and/or
local dendritic spikes (Holthoff et al., 2004; Schiller et al., 2000).
7. If activation of a single synapse cannot be achieved with the electrode
in its present position, it is advised to rapidly change the position of
the stimulation electrode. Take the stimulation electrode out of the
slice, readjust in x and y coordinates its position, and lower it back into
the slice near the ROI, as in Step 4. In our experience, the stimulation
electrode often had to be repositioned as many as five times during
a single experiment to isolate single synapses and observe calcium
microdomains.
B. Electron Microscopic Reconstruction of

Imaged Dendritic Domains
1. All experiments were carried out according to the NIH Guide for
Care and Use of Laboratory Animals (NIH publication no. 86-23,
revised 1987). After the in situ phase of the experiment, the 300-µm-
thick mouse brain slices are put between two Millipore filters to avoid
deformations and fixed in 2.5% paraformaldehyde, 1.25% glutaralde-
hyde, and 15% saturated picric acid in 0.1 M phosphate buffer (PB,
pH 7.4) for 12–36 h at 4◦ C.
2. After rinsing in 0.1 M PB (2 × 10 min at room temperature), slices
are incubated in 10 and 20% sucrose in PB (30 min in each) and
freeze-thawed in liquid nitrogen.
3. The slices are embedded in gelatine (10% in distilled water) and then
resectioned at a thickness of 60 µm in 0.1 M PB to ensure proper
transparency during light microscopic reconstruction.
4. The biocytin filled cells are visualized by the avidin-biotinylated
horseradish peroxidase method with diaminobenzidine as chro-
mogen.
5. Sections are postfixed with 1% (w/v) OsO4 in 0.1 M PB and
block stained in 1% (w/v) uranyl acetate in distilled water, dehy-
drated, and embedded into epoxy resin (Durcupan, Fluka) on glass
slides.
6. Previously two-photon imaged areas were light microscopically pho-
tographed at serial (5 µm steps) focal depths on the permanent prepa-
rations.
7. Three-dimensional light microscopic reconstructions are carried out
using Neurolucida and NeuroExplorer (MicroBrightField Inc.) with
oil immersion objective at 1250× magnification.
8. A region containing the imaged cell is reembedded and a complete
series of ultrathin sections is made from the soma and the imaged
dendrite of the cell.
9. The imaged dendrite is identified protruding from the soma in the
electron microscope and then the dendrite is followed on the serial
ultrathin sections up to the imaged portion. The alignment of light
microscopic and ultrastructural images is based on the distance from
the soma, on the undulation/course of the dendrite, on the position
of characteristic changes in dendritic diameter, or on the relative
location of dendritic beads or spine-like structures. It is essential to
observe all ultrathin sections in the area of interest by tilting the
goniometer of the electron microscope ±75◦ for checking synapses
cut at oblique planes.
10. Three-dimensional reconstructions of dendritic segments can be
performed with Neurolucida and analyzed in NeuroExplorer soft-
ware. Series of 50–120 ultrathin sections containing the ROI
are photographed. Contours of the profiles of dendrites, synap-
tic terminals, and synaptic junctions are traced using a digitiz-
ing tablet on each photograph. Image alignment is based on
minimally five points identifiable on neighboring sections using
standard functions of the Neurolucida/NeuroExplorer software
package.
C. Imaging Neuronal Ensembles
1. Prepare the Fura-2, AM by dissolving 50 µg Fura-2, AM (Molecular

Probes) in 15–48 µl DMSO and 2 µl of Pluronic F-127 (Molecular
Probes) for a final concentration of 1–3.3 mM.
2. Vortex the solution for 10–15 min prior to use.
3. Place slices in loading chamber (a petri dish 35 mm × 10 mm) contain-

ing 2.5 ml of oxygenated ACSF. Pipette 5–10 µl of Fura-2, AM solution
on top of each slice, resulting in a high initial concentration of Fura-2,
AM. Alternatively, pipette the entire volume of the prepared Fura-2,
AM at the far side of the loading chamber and allow the dye to pas-
sively diffuse throughout the chamber lading the slices. Once again
this is the preferred method especially if the slices are from a younger
animal.
4. Load the slices (we use 300-µm-thick cortical slices from mice aged 13–
30 days) in the dark for 20–30 min at 35–37◦ C with 95% O2 /5% CO2
lightly ventilated into the chamber. As a rule of thumb, the loading
time should be 10 min plus as many minutes as the age of the animal
in postnatal days.
5. Remove slices from the loading chamber and place into incubation
chamber containing oxygenated ACSF to allow them to rest.
6. Regardless of the method, imaging can begin 30 min following loading
procedure.
REFERENCES
Alger, B. E., and Teyler, T. J., 1976, Long-term and short-term plasticity in the CA1, CA3, and
dentate regions of the rat hippocampal slice, Brain Res. 110:463–480.
Allbritton, N. L., Meyer, T., and Stryer, L., 1992, Range of messenger action of calcium ion and
inositol 1,4,5-trisphosphate, Science 258:1812–1815.
Antic, S., Major, G., Chen, W. R., Wuskel, J., Loew, L., and Zecevic, D., 1997, Fast voltage-sensitive
dye recording of membrane potential changes at multiple sites on an individual nerve cell
in the rat cortical slice, Biol. Bull. 193:261.
Buhl, E. H., Halasy, K., and Somogyi, P., 1994, Diverse sources of hippocampal unitary inhibitory
postsynaptic potentials and the number of synaptic release sites, Nature 368:823–828.
Cattell, R., 1966, The scree test for the number of factors, Multivariate Behav. Res. 2:245–276.
Connors, B. W., and Gutnick, M. J., 1990, Intrinsic firing patterns of diverse neocortical neurons,
Trends Neurosci. 13:99–104.
Cossart, R., Aronov, D., and Yuste, R., 2003, Attractor dynamics of network UP states in neo-
cortex, Nature 423:283–289.
Cossart, R., Ikegaya, Y., and Yuste, R., 2005, Calcium imaging of cortical networks dynamics,
Cell Calcium 37:451–457.
Denk, W., and Svoboda, K., 1997, Photon upmanship: why multiphoton imaging is more than
a gimmick, Neuron 18:351–357.
Denk, W., Strickler, J. H., and Webb, W. W., 1990, Two-photon laser scanning fluorescence
microscopy, Science 248:73–76.
Dingledine, R., Dodd, J., and Kelly, J. S., 1980, The in vitro brain slice as a useful neurophysio-
logical preparation for intracellular recording, J. Neurosci. Methods 2:323–362.
Euler, T., Detwiler, P. B., and Denk, W., 2002, Directionally selective calcium signals in dendrites
of starburst amacrine cells, Nature 418:845–852.
Freund, T. F., and Buzsaki, G., 1996, Interneurons of the hippocampus, Hippocampus 6:347–470.
Ghozland, S., Aguado, F., Espinosa-Parrilla, J. F., Soriano, E., and Maldonado, R., 2002, Spon-
taneous network activity of cerebellar granule neurons: impairment by in vivo chronic
cannabinoid administration, Eur. J. Neurosci. 16:641–651.
Gibson, J. R., Beierlein, M., and Connors, B. W., 1999, Two networks of electrically coupled
inhibitory neurons in neocortex, Nature 402:75–79.
Goldberg, J., Holthoff, K., and Yuste, R., 2002, A problem with Hebb and local spikes, Trends
Neurosci. 25:433.
Goldberg, J. H., Tamas, G., Aronov, D., and Yuste, R., 2003c, Calcium microdomains in aspiny
dendrites, Neuron 40:807–821.
Goldberg, J. H., Tamas, G., and Yuste, R., 2003a, Ca2+ imaging of mouse neocortical interneu-
rone dendrites: Ia-type K+ channels control action potential backpropagation, J. Physiol.
551:49–65.
Goldberg, J. H., and Yuste, R., 2005, Space matters: local and global dendritic Ca2+ compart-
mentalization in cortical interneurons, Trends Neurosci. 28:158–167.
Goldberg, J. H., Yuste, R., and Tamas, G., 2003b, Ca2+ imaging of mouse neocortical in-
terneurone dendrites: contribution of Ca2+-permeable AMPA and NMDA receptors to
subthreshold Ca2+ dynamics, J. Physiol. 551:67–78.
interneurons and synapses in the neocortex, Science 287:273–278.
Helmchen, F., 1999, Dendrites as biochemical compartments, In: Stuart, G., Spruston, N., and
Hausser, M. (eds.), Dendrites, Oxford: Oxford University Press, pp. 161–192.
Helmchen, F., 2002, Raising the speed limit—fast Ca(2+) handling in dendritic spines, Trends
Neurosci. 25:438–441 (discussion 441).
Holthoff, K., Kovalchuk, Y., Yuste, R., and Konnerth, A., 2004, Single-shock LTD by local den-
dritic spikes in pyramidal neurons of mouse visual cortex, J. Physiol. 560:27–36.
Horikawa, K., and Armstrong, W. E., 1988, A versatile means of intracellular labeling: injection
of biocytin and its detection with avidin conjugates, J. Neurosci. Methods 25:1–11.
Ikegaya, Y., Aaron, G., Cossart, R., Aronov, D., Lampl, I., Ferster, D., and Yuste, R., 2004, Synfire
chains and cortical songs: temporal modules of cortical activity, Science 304:559–564.
Jolliffe, I. T., 1972, Discarding variables in a principal component analyis. I: artificial data, Appl.
Stat. 21:160–173.
Jolliffe, I. T., 2002, Principal component analysis, Springer Series in Statistics, 2nd ed., New York:
Springer-Verlag.
Kaiser, H. F., 1960, The application of electronic computers to factor analysis, Educ. Psychol.
Meas. 20:141–151.
Majewska, A., Yiu, G., and Yuste, R., 2000, A custom-made two-photon microscope and decon-
volution system, Pflügers Arch.—Eur. J. Physiol. 441:398–409.
Mao, B. Q., Hamzei-Sichani, F., Aronov, D., Froemke, R. C., and Yuste, R., 2001, Dynamics of
spontaneous activity in neocortical slices, Neuron 32:883–898.
Mao, L., and Wang, J. Q., 2003, Group I metabotropic glutamate receptor-mediated calcium
signaling and immediate early gene expression in cultured rat striatal neurons, Eur. J.
Neurosci. 17:741–750.
Markram, H., Luebke, J., Frotscher, M., and Sakmann, B., 1997, Regulation of synaptic efficacy
by coincidence of postsynaptic APs and EPSPs, Science 275:213–215.
Markram, H., Toledo-Rodriguez, M., Wang, Y., Gupta, A., Silberberg, G., and Wu, C., 2004,
Interneurons of the neocortical inhibitory system, Nat. Rev. Neurosci. 5:793–807.
McBain, C. J., and Fisahn, A., 2001, Interneurons unbound, Nat. Rev. Neurosci. 2:11–23.
Neher, E., 1998, Usefulness and limitations of linear approximations to the understanding of
Ca++ signals, Cell Calcium 24:345–375.
Nikolenko, V., Nemet, B., and Yuste, R., 2003, A two-photon and second-harmonic microscope,
Methods 30:3–15.
Ohki, K., Chung, S., Ch’ng, Y. H., Kara, P., and Reid, R. C., 2005, Functional imaging
with cellular resolution reveals precise microarchitecture in visual cortex, Nature 433:
597–603.
Petersen, C. C., and Sakmann, B., 2001, Functionally independent columns of rat somatosen-
sory barrel cortex revealed with voltage-sensitive dye imaging, J. Neurosci. 21:8435–8446.
Poirazi, P., and Mel, B. W., 2001, Impact of active dendrites and structural plasticity on the
memory capacity of neural tissue, Neuron 29:779–796.
Prince, D. A., and Connors, B. W., 1984, Mechanisms of epileptogenesis in cortical structures,
Ann. Neurol. 16:S59–S64.
Rose, C. R., and Konnerth, A., 2001, NMDA receptor-mediated Na+ signals in spines and
dendrites, J. Neurosci. 21:4207–4214.
Rose, C., Kovalchuk, Y., Eilers, J., and Konnerth, A., 1999, Two-photon Na+ imaging in spines
and fine dendrites of central neurons, Pflugers Arch. 439:201–207.
Sabatini, B. L., Oertner, T. G., and Svoboda, K., 2002, The life cycle of Ca(2+) ions in dendritic
spines. Neuron 33:439–452.
Sakmann, B., and Neher, E., 1983, Single Channel Recording, New York: Plenum Press.
Sarvey, J. M., Burgard, E. C., and Decker, G., 1989, Long-term potentiation: studies in the
hippocampal slice, J. Neurosci. Methods 28:109–124.
Schiller, J., Major, G., Koester, H. J., and Schiller, Y., 2000, NMDA spikes in basal dendrites of
cortical pyramidal neurons, Nature 404:285–289.
Smetters, D. K., Majewska, A., and Yuste, R., 1999, Detecting action potentials in neuronal
populations with calcium imaging, Methods 18:215–221.
Somogyi, P., Tamas, G., Lujan, R., and Buhl, E., 1998, Salient features of synaptic organisation
in the cerebral cortex, Brain Res. Brain Res. Rev. 26:113–135.
Spruston, N., Schiller, Y., Stuart, G., and Sakmann, B., 1995, Activity-dependent action potential
invasion and calcium influx into hippocampal CA1 dendrites, Science 286:297–300.
Stosiek, C., Garaschuk, O., Holthoff, K., and Konnerth, A., 2003, In vivo two-photon calcium
imaging of neuronal networks, Proc. Natl. Acad. Sci. U. S. A. 100:7319–7324.
Tamas, G., Buhl, E. H., Lorincz, A., and Somogyi, P., 2000, Proximally targeted GABAergic
synapses and gap junctions synchronize cortical interneurons, Nat. Neurosci. 3:366–371.
Tamas, G., Buhl, E. H., and Somogyi, P., 1997, Massive autaptic self-innervation of GABAergic
neurons in cat visual cortex, J. Neurosci. 17:6352–6364.
Tanaka, E., Uchikado, H., Niiyama, S., Uematsu, K., and Higashi, H., 2002, Extrusion of intracel-
lular calcium ion after in vitro ischemia in the rat hippocampal CA1 region, J. Neurophysiol.
88:879–887.
Tsien, R. Y., 1981, A nondisruptive technique for loading calcium buffers and indicators into
cells, Nature 290:527–528.
Tsien, R. Y., 1989, Fluorescent probes of cell signaling, Ann. Rev. Neurosci. 12:227–253.
Voitenko, N. V., Kostyuk, E. P., Kruglikov, I. A., and Kostyuk, P. G., 1999, Changes in calcium
signaling in dorsal horn neurons in rats with streptozotocin-induced diabetes, Neuroscience
94:887–890.
Waters, J., Schaefer, A., and Sakmann, B., 2005, Backpropagating action potentials in neurones:
measurement, mechanisms, and potential functions, Prog. Biophys. Mol. Biol. 87:145–170.
Yuste, R., and Denk, W., 1995, Dendritic spines as basic units of synaptic integration, Nature
375:682–684.
Yuste, R., Gutnick, M. J., Saar, D., Delaney, K. D., and Tank, D. W., 1994, Calcium accumula-
tions in dendrites from neocortical neurons: an apical band and evidence for functional
compartments, Neuron 13:23–43.
Yuste, R., and Katz, L. C., 1991, Control of postsynaptic Ca2+ influx in developing neocortex
by excitatory and inhibitory neurotransmitters, Neuron 6:333–344.
Yuste, R., Tank, D. W., and Kleinfeld, D., 1997, Functional study of the rat cortical microcircuitry
with voltage-sensitive dye imaging of neocortical slices, Cereb. Cortex 6/7:546–558.
16
Stereology of Neural
Connections: An Overview
CARLOS AVENDAÑO
ASSESSING NEURAL CONNECTIONS

TARGET PARAMETERS OF NEURAL CONNECTIONS
Structures Identified in Studies of Connections
Parameters
MORPHOMETRY AND STEREOLOGY
ON BIASES AND ARTIFACTS
Biases in Counting and Measuring
Observation Artifacts
Deformation Artifacts
CONNECTIONS AS DEFINED BY NEURONAL NUMBERS
Tools
Practical Tips
CONNECTIONS AS DEFINED BY AXON NUMBERS
Axon Numbers in 2D
Preterminal and Terminal Axonal Fields
CONNECTIONS AS DEFINED BY AXONAL LENGTH
Navigating the Isotropy Reefs
A Caveat: Axons are Irregular Tubes, Not Lines
CONNECTIONS AS DEFINED BY SYNAPSE NUMBER
What Is a Synapse?
Unbiased Counting
Of Pincushions and Haystacks: Sampling Designing
Densities and Total Numbers
A PRÉCIS ON ACCURACY, PRECISION, AND EFFICIENCY
ADVANTAGES, LIMITATIONS, AND PERSPECTIVES
APPENDIX: A CASE STUDY
Design and Equipment
Procedure
REFERENCES
CARLOS AVENDAÑO • Department of Anatomy, Histology and Neuroscience,

Autónoma University of Madrid, Medical School, 28029 Madrid, Spain
477
478 CARLOS AVENDAÑO
Abstract: Stereological methods have made a quiet revolution in the quantification

of neural structures. By combining stochastic geometry and statistics, they have given
neuroanatomists invaluable tools to estimate with accuracy and precision quantita-
tive information in three dimensions, from data obtained on flat images, sections,
and slabs of central and peripheral nervous tissues. This chapter reviews and dis-
cusses the application of a variety of stereological methods to quantify geometric
parameters relevant for analyzing neural connections, such as neuron, axon, or
synaptic number, axonal length, and volume of regions of interest. Special attention
is given to histological and observation artifacts that may hamper the efficacy of oth-
erwise unbiased and efficient sampling and measuring methods. Current strengths
and future developments of stereology in this field are briefly reviewed as well. The
chapter ends with a worked example of stereology applied to the quantitative analysis
of connections in a connectional study using retrograde tracers.
Keywords: axon number, axonal length, morphometry, neuron number, synapse

number
I. ASSESSING NEURAL CONNECTIONS
Cell-to-cell communication is a fundamental property of multicellular

organisms. It takes place by the information-laden exchange of molecu-
lar signals between cells which abut, or are close to each other (autocrine
and paracrine signals), and also by the effects of blood-borne molecules
originating in distant cells (endocrine signals). A special kind of paracrine
signaling in a broad sense (sometimes called neurocrine) is represented by
the chemical neurotransmission in the nervous system. Neurons achieve lo-
cal contacts with other neurons, which may be situated at remote places, by
means of their long neurites (dendrites and axons) and a variety of synaptic
specializations.
The generation of networks of local intercellular contacts at a distance
from the cell body is a hallmark of the nervous system. Through extensive
dendritic and axonal arborizations, individual neurons define ample do-
mains of influence. The geometry of axons and dendrites, together with
the distribution of ion channels, defines fundamental functional properties
of the neuron. Basic geometric parameters, such as shaft caliber, tapering,
or branching, initially shown to determine action potential propagation in
axons (Goldstein and Rall, 1974), also shape the centrifugal and centripetal
propagation of postsynaptic and action potentials in dendrites (Poirazi and
Mel, 2001; Schaefer et al., 2003; Spruston et al., 1999). Spines and dendritic
segments are also important in defining biochemical compartments in neu-
rons, and these are dependent on their size, length, and shape (Helmchen,
1999; Korkotian and Segal, 2000). Different patterns in the density and
spatial distribution of axonal or dendritic branches create a wide spectrum
of wiring configurations, which, already described in early Golgi studies,
are currently the subject of intensive experimental and theoretical research
(Chklovskii et al., 2002; Mitchison, 1992; Stepanyants et al., 2004; Van Pelt
et al., 2001).
STEREOLOGY OF NEURAL CONNECTIONS 479
In organisms other than simple invertebrates, however, brainwork is
mainly expressed at the population, rather than at the single-cell level. Neu-
rons aggregate in spatially discrete sets (nuclei, modules, columns, and so
on), which display common anatomical (in terms of form and connections)
and functional (coherent activation and firing, substrate for representa-
tional mapping) properties. Moreover, if the geometric properties of single
neurons and local axonal circuits are hard to tackle in quantitative terms,
difficulties pile up when circuits established by neuronal populations are
the target. So far, most of our current understanding of neural circuits rests
largely on qualitative descriptions, and when quantitative morphological as-
sessments are made, they are most often expressed in subjective and relative
terms (heavy, dense, light, sparse, scattered, etc.), according to imprecise
ordinal scales (Stevens, 1951). This situation, however, is changing for the
better with the increasing awareness that sound quantitative neuroanatomy
is a requisite to understanding the brain (Braitenberg and Schüz, 1998;
Evans et al., 2004).
II. TARGET PARAMETERS OF NEURAL CONNECTIONS
A. Structures Identified in Studies of Connections
Before the introduction of axoplasmic tracers and of methods to visual-

ize specific molecules in neurons and neurites, most anatomical studies of
neural connections had to rely on the Golgi and the myelin or axonal degen-
eration methods. Under optimal conditions, the Golgi method allows for the
recognition of complete cell bodies and dendritic arbors with their spines
and, at least, large stretches of unmyelinated axonal branches (Valverde,
1970). Camera lucida or digital imaging reconstructions provide data in
two and three dimensions (2D and 3D), respectively, which can be used to
parametrize and analyze geometrical and topological features of the various
components of the neuron (Hillman, 1979; Uylings et al., 1986a,b). Axonal
degeneration studies, such as the Nauta and the Fink–Heimer methods, rely
on the staining of abnormal (dying) and incomplete axonal profiles after
focal lesions. While being very sensitive, only rough quantitative estimates
of en passant and terminal connections can be obtained with these meth-
ods, which are more adequate for describing connections in qualitative and
topographical terms.
The knowledge of neural connections made a dramatic leap forward
with the introduction of tracing techniques, in the early 1970s, that ex-
ploited the axoplasmic transport of exogenous tracers, such as tritiated
amino acids, horseradish peroxidase (HRP; free or conjugated to other
molecules), lectins, and fluorescent molecules (references in Heimer and
Zaborszky, 1989; see also chapters by Reiner and Lanciego, this volume).
In addition to their biocompatibility, most of these tracers enable intense
and, in some cases, fairly complete labeling of populations of cell bodies,
dendrites, spines, and/or axonal trees and synaptic boutons to be obtained.

Also, HRP or smaller molecules (such as Lucifer yellow, biocytin, or neuro-
biotin) can be administered to individual neurons, by intra- or juxtacellular
injections, yielding in optimal conditions Golgi-like staining of the entire
neuron. Additionally, different axoplasmic tracers can be combined among
them, and/or with immunocytochemical labeling of specific molecules, to
furnish informative connectional and neurochemical multiple labeling of
neurons at single or population level (e.g., Vinkenoog et al., 2005; see also
Sesack et al., this volume). Immunocytochemistry alone may provide ex-
cellent labeling of specific subsets of neurons or their neurites in selected
regions, making them amenable to being measured and quantified. Finally,
ultrastructural studies of neural connections are mostly used to obtain mor-
phological and neurochemical data of synapses, spines, boutons, and other
components of neurons labeled with tracers or immunocytochemistry prior
to, or after embedding in resin (Fig. 16.1).
B. Parameters
Neural connections may be defined by a large number of parameters, and

an appropriate choice among them should be made when designing a stere-
ological (or, in general, any morphometric) study. On biological grounds,
it is convenient to distinguish measures referred to individual neurons, as
a whole (a cell soma, whole dendritic or axonal tree of a neuron, spines
in a cell) or in part (length, surface area, number of spines, synapses, bou-
tons, etc., referred to a dendritic or axonal subtree), from those referred
to sets of structures belonging to neuronal populations (groups of neuron
bodies, dendritic bundles, axonal networks, nerves or fiber tracts, synapses
in a territory, etc.).
Different types of parameters may be defined on geometrical and statis-
tical grounds: First-order parameters refer to primary geometrical measures,
such as number, length, surface, and volume; they may also include relative
measures, or ratios, such as numerical, length, surface, or volume densities.
These parameters are direct descriptors of the target neurons or neurites,
and will be the main target of this review (Table 16.1). Second-order measures,
in a morphometric/stereological context, refer to statistical descriptions of
the 3D spatial distribution of the first-order parameters. They inform about
→
Figure 16.1. Diagram illustrating the main neural components that may be defined by
different methods of labeling single neurons and neuronal populations. The soma
and dendritic arbor and, in optimal conditions, the full axonal tree can be labeled by
the Golgi stain, and by targeting selected neurons with exogenous tracers or revealing
specific molecules (top). Connections from neuron groups are usually studied by
anterograde or retrograde labeling with tracers. And the global population of neural
components in a region is also accessible to specific labeling and/or ultrastructural
study (bottom).
Figure 16.1.
Table 16.1. A selection of parameters of interest to stereologically assess neural

connections, and some pertinent references
Parameter Selected references
NUMBER
Absolute numbers (N)
Neuron bodies sending axons to a labeled target Bermejo et al. (2003) and Negredo
et al. (2004)
Neuron bodies labeled by a transneuronal Glatzer et al. (2003)
anterograde marker, such as viruses
Neuron bodies in a certain region, or expressing a West and Slomianka (1998) Donovan
certain molecular marker et al. (2002), and Avendoño et al.
(2005)
Neuron bodies that effectively take up an —
anterograde tracer
Axons in a peripheral nerve, a central tract, or Larsen et al. (1998b) and Marner and
white matter Pakkenberg (2003)
Terminal or preterminal axons in a target territory —
Presynaptic boutons (en passant or terminal) in a Calhoun et al. (1996)
target territory
Synapses in a target territory Geinisman et al. (1996) and Tang
Relative numbers (densities) et al. (2001)
NL , number of spines, boutons, synapses, etc., per Rusakov and Stewart (1995)
dendritic/axonal segment length
N A , number of cross-sectioned axons per unit area Partadiredja et al. (2003)
in a section of a nerve, tract, or territory
NS , number of spines or synapses per unit area of Mayhew (1979b) and Rusakov
dendritic/axonal/somatic membrane surface et al. (1998)
NV , numerical density of cell bodies (or axons, O’Malley et al. (2000) and Andersen
spines, boutons, synapses, etc.) per unit volume and Pakkenberg (2003)
of a predetermined reference space
Nu , numerical density of spines, synapses, etc. per Turner and Greenough (1985) and
neuron Kleim et al. (1997)
LENGTH (path length)
Absolute (L)
Total dendritic or axonal length in a certain Howard et al. (1992), Tang and
territory Nyengaard (1997), and Mouton
Relative (densities) et al. (2002)
L V , length of neurite branches per unit volume of Stocks et al. (1996) and Larsen
a predetermined reference space (nucleus, et al. (2004)
white matter sector, cortical layer, etc.)
L u , length of axonal or dendritic arborizations Rønn et al. (2000)
(complete or partial) per neuron
SURFACE
Absolute (S)
Total membrane surface of dendrites, spines, Peterson and Jones (1993)
axons, synaptic specializations, etc. in a given
territory
Relative (densities)
S S , membrane surface fraction (of a dendrite, —
spine, axon, etc.) occupied by synaptic
specializations
Table 16.1. (Cont.)
Parameter References
SV , extension of membrane surface (of dendrites, Machı́n-Cedrés (2004)
spines, axons, synaptic specializations, etc.) per
unit volume of a predetermined reference
space (neuropil, dendrites, axons, etc.)
CROSS-SECTIONAL AREA
Absolute (A) or relative (area fraction, A A )
Total or fractional area occupied by nerve or tract Larsen (1998) and Schiønning et al.
components (myelin, myelinated or (1998)
unmyelinated axons, blood vessels, connective
tissue) in a transverse section
Mean and distribution (an )
For cross-sectional areas of a population of axons Larsen (1998)
in a nerve or tract
VOLUME
Absolute (V )
Volume of a nucleus, cortical layer, cell aggregate, Blasco et al. (1999) and Bermejo et al.
injection site, projection territory, etc. (2003)
Relative (volume fraction, VV )
Relative occupancy of a predetermined reference Jones (1999), Machı́n-Cedrés (2004)
space by a defined neuronal (or vascular, glial,
etc.) component
Mean and distribution (vn )
For a population of cell bodies, presynaptic Avendaño and Dykes (1996) and
boutons, spines, etc. Lagares and Avendaño (1999)
the spatial architecture of the target neurons or neurites, either among

themselves, or with respect to other structures, such as glial cells, blood ves-
sels, etc. Second-order stereology is much less developed, and only a few
comments will be made in section “Advantages, Limitations, and Perspec-
tives.” Neural shape may also be quantitated in topological terms, that is, by
defining only the connectivity pattern of the (dendritic or axonal) tree ar-
chitecture of a neuron. Topological measures include nodes or vertices (root,
branch points, terminal tips) and segments (root, intermediate, terminal),
and ignore metrical variables. Excellent reviews on the topological analysis
of dendrites are available (Uylings and Van Pelt, 2002; Uylings et al., 1989;
Verwer et al., 1992).
It is important to determine which is the minimum set of parameters
that provide a satisfactory structural description of a neuron, seeking com-
pleteness, but avoiding redundancy. But it has to be realized that such a
description only reaches its full neurobiological sense in a populational,
and therefore statistical, sense (Ascoli, 2002). Seeking better estimators of
shape and its variation and complexity within and between cell classes, met-
rical, topological, and/or spatial parameters have been combined into more
sophisticated analytical methods. These provide measures which range from
simple relations of the same parameter at various locations, as is the case
of taper (ratio of a segment thickness between its end and its beginning)
or tortuosity (ratio of trace length to Euclidean length for a given neurite
segment), to fractal analysis, or to multivariate analysis of configurations of
points in artificial shape spaces. No further comments will be made on these
issues, but a growing body of literature is available to the interested reader
(Adams et al., 2004; Ascoli, 2002; Jelinek and Fernandez, 1998; Uylings and
Van Pelt, 2002).
III. MORPHOMETRY AND STEREOLOGY
In its widest acceptance, morphometry has been defined as “Quantita-

tive morphology; the measurement of structures by any method, including
stereology” (Weibel, 1979). In practical terms, however, morphometric ap-
proaches in neuroscience have focused on the refinement of quantitative
morphological descriptions and topological analyses of restricted structures,
such as single neurons or limited axonal or dendritic fields. Stereology, in
turn, has carved itself an expanding niche in the quantitation of large pop-
ulations of cells and neurites.
Stereology combines stochastic geometry and statistics and is directly
aimed at estimating values in a higher dimension from data obtained at a
lower dimension on a properly sampled material (Cruz-Orive, 2002b). This
is the case in most neurohistological and neuroimaging studies, which are
based on sections or slices (physical or virtual) of parts of the central or pe-
ripheral nervous system. The need to apply correct sampling strategies due
to the complexity and the large number of neural structures may be justified
also for single cells. Figure 16.2 shows two examples of exceedingly com-
plex axonal or dendritic arbors which, while amenable to costly individual
complete reconstructions, would justify applying more efficient statistical
sampling to estimate some morphometric parameters.
Heretofore, measurements of geometric parameters relevant for ana-
lyzing neural connections are carried out in most cases on sections of
postmortem-processed tissue. Moreover, they are often performed on ana-
logical or digital graphic reconstructions of images obtained from that tis-
sue, in particular for dendritic and local axonal arborizations. Confocal
images from labeled neurons in in vitro slices make an exception, but only
in the case of small neurons with their whole axonal and dendritic trees
in the slice. In this case the neurons, while partially deafferented, are oth-
erwise intact; in any other case, we will be dealing with directly lesioned
cells undergoing acute axotomy and dendrotomy to various degrees. A
promising, but just incipient, development is the application of two-photon
microscopy to the quantitative analysis of dendrites in vivo (Broser et al.,
2004).
Tissue processing introduces scores of structural deformations, which may
be the cause of serious biases, if they are not properly controlled, or at
1 mm
Figure 16.2. Examples of complex dendritic and axonal arbors of single neurons.
(A) Profile drawn with camera lucida of a Golgi-stained human Purkinje cell (Cajal,
1904). (B) Pyramidal neuron in the rat hippocampus, filled intracellularly with HRP;
the whole axonal arbor was stained and drawn with camera lucida. (Redrawn from
Tamamaki et al., 1984, with permission from Elsevier).
least taken into account (see section “Deformation Artifacts”). Structural

deformations (or “morphological noise”; Horcholle-Bossavit et al., 2000)
also plague 3D reconstructions of dendritic or local axonal arbors made
with the help of imaging software. They are due to histological, optical,
and operator-linked distortions, worsened by the need to merge serial sec-
tions, which are unlikely to display identical distortions. Considerable differ-
ences in many parameters have been reported to occur between laboratories
when reconstructing neurons of the same class, or, what is worse, between
reconstructions of the same neuron carried out by different operators, or
by the same operator working on two different systems (Scorcioni et al.,
2004).
Many stereological applications can be carried out entirely without any
special equipment, other than a microscope with good immersion and dry
optics, and a high-resolution microcator, or any alternative gadget destined to
measure displacements of the stage in the z-axis (see section “Deformation
Artifacts”). It cannot be denied, though, that stereology has already made
an irreversible liaison with computer-assisted microscopy, arousing the in-
terest of high-tech research-oriented companies to develop and improve
interactive systems (Glaser and Glaser, 2000). This includes stereological
tools, which simply could not be implemented without computer help (see
section “Connections as Defined by Axonal Length”). These systems focus
on providing software control for (1) video acquisition of live and still micro-
scopic images, (2) creating and overlaying test grids on them, (3) moving the
microscope stage, (4) collecting data on spreadsheets, and (5) performing
some computations online (for local stereological estimators, such as the
nucleator or the rotator). Yet, the systems do not “make decisions” on the
sampling design, choice of probe(s), and recognition of objects and targets,
which are left to the operator. The computer-assisted stereological toolbox
(CAST; Visiopharm, Hørsholm, Denmark) and the StereoInvestigator (Mi-
crobrigthfield, Williston, VT) systems are probably the most complete and
widely used, but several others1 provide reasonable sets of the most popular
stereological tools.
In the past years useful monographs on stereology have appeared that pro-
vide theoretical and practical information of great help for users, without
delving too deeply into abstruse mathematical proofs (e.g., Evans et al., 2004;
Howard and Reed, 2005; Mouton, 2002). It is very recommendable, how-
ever, to add specialized practical training to reading. Intensive courses and
workshops are available around the world, some of them sponsored by the In-
ternational Society for Stereology (ISS; http://www.stereologysociety.org),
which are very useful to boost beginners’ acquaintance with the intricacies
of stereology.
IV. ON BIASES AND ARTIFACTS
A. Biases in Counting and Measuring
An unbiased method is an accurate method, that is, a method that yields

quantities whose mean coincides with the true value (see section “A Précis
on Accuracy, Precision, and Efficiency”). Any systematic error (bias) in the
method will likely lead to an estimation bias. But a clear distinction must
be made between counting/measuring methods, and their application to
real experimental material. The former have to prove only unbiasedness via
statistical/geometrical arguments, and therefore their acceptance cannot
await certification by “validation” against empirical “gold standards,” or by
showing “reproducibility” by various laboratories (Cruz-Orive, 1994, 2002a).
Hence, possible sources of bias with respect to the method have to be sought
in sampling design or mathematical errors, or in a wrongful choice of the
geometric probes.
When an unbiased method is applied to anything closer to real life than a
model data set, however, new possible sources of estimation bias can emerge.
1
Stereologer (Systems Planning and Analysis, Inc., Alexandria, VA); Explora Nova Stereol-
ogy (Explora Nova, La Rochelle, France); Digital Stereology (Kinetic Imaging, Liverpool, UK);
MCID Basic Stereology System (Imaging Research Inc., St. Catharines, Ontario, Canada); Stere-
ology Toolkit (BIOQUANT Image Analysis Corporation, Nashville, TN).
They are based on the ubiquitous presence of two types of artifacts. Ob-
servation artifacts are typically found in transmission and fluorescence mi-
croscopy on histological sections, but may also be present in electron mi-
croscopy (EM) images, as well as in virtual sections obtained by imaging
techniques (such as PET, MR, or CT scans). Deformation artifacts originate in
any dimension-changing deformation inflicted on the tissue by the various
manipulations it suffers, from the processes of organ death and fixation to
the staining and coverslipping of the sections. The following explanation of
these artifacts is largely inspired on Cruz-Orive (1990) and Dorph-Petersen
et al. (2001).
B. Observation Artifacts
Concerning observation, the general caveat is that objects to be counted

or measured have to be unambiguously identified. This means, in first
place, that they should be observed with sufficient resolving power, i.e.,
with a convenient spatial resolution of the imaging system. The use of in-
adequate magnifications, nonmatching refractive indexes of optical phases
between the specimen and the objective, and objectives/condensers with
low numerical aperture may fail to provide images above a crisp detection
threshold.
However, even if the imaging system is adequate, further problems can
derive from the nature of the sections and the staining of the target objects.
Figure 16.3 summarizes these problems.
1. Signal Weakness
Lack of information due to incomplete staining in transmission mi-

croscopy, or to fading of light emission by a fluorophore. The former, in
varying degrees, is a common result when immunolabeling thick sections,
but is not restricted to it. Molecular size of the reagents, their solubility and
ability to penetrate the tissue, and reaction time and temperature are fac-
tors that condition the completeness of the staining. Fading of fluorescence
occurs naturally when a fluorophore is subjected to illumination with an
appropriate excitation wavelength. An excellent review of the strengths and
weaknesses of the use of fluorescent tracers for cell counting may be found
in Peterson (2004); see also Negredo et al. (2004).
2. Overprojection
Excess of information in an image: when the section has a definite

thickness, the objects of interest are opaque, and the surrounding tis-
sue/embedding medium is translucent or transparent.
Figure 16.3. Observation artifacts in transmission microscopy on tissue sections. SW,

signal weakness produced by poor stain penetration and/or fading in sections
stained free floating (left) or after mounting (right). OP, overprojection, or Holmes
effect. UP, underprojection. Ov, overlapping, Tr, truncation, physical/mechanical
(with loss of material, 1–2), and optical/observational (3–5). Membranes or fila-
ments cut tangentially are likely to be overlooked (3), or misidentified (e.g., as be-
longing to separate objects, 4). Fractions of spheroidal particles may be overlooked
if cut at shallow angles (5). Larger particle fractions, on the other hand, may appear
as entire particles (6, 7), because of overprojection. The four bottom diagrams are
somewhat modified after Cruz-Orive (1990), with permission from the author.
3. Underprojection
Lack of information in an image: when the section has a definite thick-

ness, the objects of interest are transparent, and the surrounding tis-
sue/embedding medium is opaque.
4. Overlapping
Loss of information when objects overlap and mask each other in a section
that is substantially thicker than the objects’ size.
5. Truncation
Lack of information due to the physical loss of object fragments from

the section surface (physical or mechanical truncation), or to the inabil-
ity to detect object fragments present in the section, but which are cut at
shallow angles (optical or observational truncation). Both are typical for
particles; the latter is also common for membranes and flattened struc-
tures (particularly in ultrastructural studies of subcellular membranes and
synapses).
Any of these artifacts, if moderate, can be handled by appropriate em-
bedding, cutting, sampling, and choice of stereological procedures. Severe
truncation or overlapping, and other causes of marked signal loss, however,
cannot in general be corrected and should be avoided.
C. Deformation Artifacts
Stained tissue sections, the basic material on which measurements are

to be made, are the end product of submitting a piece of previously living
tissue to perfusion and fixation, dehydration, embedding, cutting, stain-
ing, mounting, and coverslipping. Obviously, a large number of interacting
artifacts may be introduced by these procedures, and different types of de-
formation may result (Dorph-Petersen et al., 2001; Fig. 16.4). Confidence
that the estimates obtained represent real numbers and values would thus
depend on the certainty that the dimensional distortions introduced by his-
tological processing can be controlled.
1. From Fixation to Embedding
As soon as the tissue becomes anoxic, a cascade of autolytic effects sets

out that leads to structural changes, first seen at the ultrastructural level.
The rapid administration of fixatives is intended to preserve the structure
(and the enzymatic reactivity, antigenicity, etc.), but it cannot entirely ful-
fill the objective. After an early intracellular swelling, the ensuing shrink-
age/swelling balance is influenced by the osmolarity and ionic composition
of the fixative solution, and its penetration and fixation rate. Moreover, not
all tissue components, including membranes and subcellular organelles,
display the same degree of deformation for a given osmolarity of the fix-
ative. Hence, even though a good choice of the fixative and the fixation
Figure 16.4. (A) Different types of deformation of a mock block of brain tissue with
two layers and some heterogeneous discrete components in one of them. To the left
is shown an isotropic shrinkage (affecting all layers and components equally in 3D).
To the right there are four examples of anisotropic shrinkage of the whole block,
which affects the z-, but not the x- and y-axes. Shrinkage may affect homogeneously
all components of the block in a uniform or nonuniform manner. Or it may be
differential, affecting some components and not others. (B) Different meanings of
thickness (t), shown on a diagram of the process of cutting and mounting a tissue
section. tm , thickness set at the microtome, or “block advance”; ts , thickness of the
section after cutting and in the collecting solution; tf , final thickness measured on
the completely processed section. This figure is inspired from Figs. 3 and 4 in Dorph-
Petersen et al. (2001).
procedure may keep the overall volume of a brain or nerve block close to
its premortem condition, it does not guarantee that cell nuclei and somata,
dendrites, blood vessels, myelin, or extracellular space have not differen-
tially shrunk or swelled. Specific details for different structures are lacking
to a large extent, but good reviews are available that should help gauge the
problem (Hayat, 1981, 2000).
Additional deformations are introduced by embedding procedures.
When embedding entails dehydration and defatting, as in paraffin-,
celloidin-, and epoxy resin-embedding, the net result is different degrees
of shrinkage. For vibratome sectioning, fresh or fixed, but otherwise non-
manipulated tissue is used, and therefore no further changes should be
expected at this stage. In contrast, tissue processing for cryosectioning en-
tails some additional shrinkage, because of the partial dehydration the tis-
sue undergoes during cryoprotection in hyperosmolar sucrose or glycerol
solutions. Water-soluble embedding media, such as glycol methacrylate,
in general, yield moderate shrinkage, which is compensated by swelling
at subsequent steps of processing of the sections (Gerrits et al., 1992).
Again, whether deformation is homogeneous or differential (see below)
is largely untested, but it is likely that the use of plastic, celloidin, or
resin media results in a quite homogeneous matrix (Dorph-Petersen et al.,
2001).
2. From Sectioning to Coverslipping
The knife advance produces an orthogonal (rotary and sliding micro-

tomes) or zigzagging (vibratome) strain on the tissue block. This inevitably
causes three kinds of mechanical effects on the resulting section: compres-
sion, creasing, and grazing. These effects can be substantially minimized
when the chosen cutting thickness is the most adequate to the embedding
medium, the knife’s edge is sharp and unblemished, the clearance angle
(that between the knife edge bevel and the tissue block) is set at the most
effective minimum, and the cutting angle (angle of edge bevel) is shallow
enough to minimize compression, but not too much, to avoid vibrating and
causing chatter in the section (Ellis, 2000; Hayat, 2000).
Compression may be particularly severe when cutting very thin paraf-
fin sections, so that the “height” of the section (parallel to the knife ad-
vance) may become over 50% shorter than the original “height” of the
block (Dempster, 1943). Creasing consisting of regular undulations of the
cutting surface is often found in vibratome sections. In general, irregulari-
ties of the surface are significant in vibratome and frozen sections, and less
marked in sections from embedded material. Grazing, if not too coarse, is
difficult to detect and is responsible for truncation of object fragments at
the surface of the sections.
When sections are not mounted directly on glass slides (cryostat), they are
collected in buffer (frozen, vibratome) or ethanol (celloidin), or floated on
water (paraffin, resins) before mounting. As long as buffer osmolarity is con-
trolled, no dimensional changes are likely to occur to frozen or vibratome
sections. Celloidin sections are also stable in 70% ethanol. Paraffin sections
and semi- and ultrathin resin sections, however, stretch on a warm water
bath (or droplet), which compensates to a varying degree the compression
and creasing underwent during cutting. Methacrylate sections display no-
table stretching on the water bath and in the mounting step, which may
fully compensate the shrinkage suffered during the dehydration step of
histoprocessing (Gerrits et al., 1987).
Mounted sections adhere firmly to the supporting slide and no longer

change their planar (x–y axes) dimensions. Their height, or thickness
(z-axis), however, may decrease over 50% for unembedded, hydrated sec-
tions that are allowed to dry on the slide (frozen, vibratome), while it re-
mains essentially unchanged in celloidin- or methacrylate-embedded sec-
tions (Avendaño and Dykes, 1996; Bermejo et al., 2003; Dorph-Petersen et
al., 2001). Differential vertical compression, resulting in changes in the par-
ticle distribution along the z-axis, has been recently reported (Gardella et al.,
2003). Overall, deformations along the z-axis are a source of considerable
practical difficulties, and deserve specific consideration.
3. The Annoying t
Control of the reference space is a must for any direct stereological esti-
mator of volume, as well as for most ratio estimators (number, length, and
surface densities, as well as volume fractions). Since, as seen above, the tissue
may undergo dimensional changes in any of the three axes: (1) volume esti-
mations have to be corrected for any deformations taking place between the
condition established as a reference baseline (living individual, fixed organ,
etc.) and the processing stage at which the Cavalieri estimator of volume is
applied (organ slabs, mounted sections, etc.) and (2) the estimation of the
reference volume for ratio estimations, such as NV or L V , should ideally be
made at the same sampling stage at which objects (cell bodies, axons, etc.)
are counted with disectors; if V is estimated at an earlier stage, it is necessary
to check whether additional deformations occurred between stages. When
numbers are estimated using a fractionator sampling design (Gundersen
et al., 1988; Howard and Reed, 2005), the size and shape of the reference
space are irrelevant. However, deformations along the z-axis (thickness) of
the section are still relevant for this estimator, because the latest fraction
that enters the equation in the estimator [see Eqs. (16.4) and (16.5) below]
includes section thickness values.
When dealing with section thickness, it is important not to mix up con-
cepts:
tm : thickness set at the microtome. Equivalent to “block advance,” it can
be calibrated if doubts exist over the precision of the cutting device
(Avendaño and Dykes, 1996; Dorph-Petersen et al., 2001; Sterio, 1984).
ts : thickness of the section in the collecting—or floating—solution. In
addition to already showing local unevenness, it may differ from tm be-
cause of compression during cutting, swelling by hydration or heating,
etc.
tf : final thickness measured at any specific location on the completely
processed section. It is expected to vary in different positions because
of the unevenness of the cutting surfaces.
t f : mean final thickness over individual measurements taken at locations
properly sampled throughout the region of interest in the section.
The existence of many methods to estimate tf , particularly for ultrathin
sections (De Groot, 1988), indicates that none gathers sufficient simplicity,
efficiency, accuracy, and general applicability to be proposed as a standard.
For stereological measurements, however, ultrathin (below 100 nm) and
semithin sections (0.1 − 2.5 µm) are generally used for applying physical
disectors. Hence, it is tm that matters, which is easier to gauge.
There is a compelling need to warrant unbiased and precise estimates of
tf on thick sections, however, given its importance in estimators based on the
optical disector and fractionator. The difficulties it entails, and the frequent
failure to adequately report on the mode and precision with which it was
estimated, have served recently as a basis for some criticisms of stereological
applications (Guillery, 2002). It is widely accepted that the most practical
way to estimate tf is by focusing through the section depth with high mag-
nification immersion lenses of high numerical aperture (at least 60× and
1.2 n.a.), and recording the positions of the upper and lower surfaces with
a sensitive distance encoder (see Uylings et al., 1986a,b for equations on
optical depth due to both optics and accommodation). The optoelectronic
microcators are excellent z-position recorders, with a resolution as good as
±0.25 µm, but they are costly and not readily available to all laboratories.
However, ingenious and inexpensive alternatives exist (Howard and Reed,
2005; Korkmaz and Tümkaya, 1997), with nearly as good a precision.
The introduction of observer biases when taking tf readings is a real risk.
The section surfaces may be irregular, fuzzy (in particular the bottom sur-
face), indistinct, or just invisible. This is the case especially when working un-
der dark field or fluorescence microscopy in spots without signal or, worse,
where a strong signal arises from just the inner part of the section. All of
these problems may be tackled by procedures that increase contrast at the
surfaces (using semidark field illumination or differential interference con-
trast), but the specific application in each case takes learning and shared
practice (Bermejo et al., 2003; Dorph-Petersen et al., 2001; Korkmaz and
Tümkaya, 1997; Peterson, 2004). In an optical fractionator design, more-
over, the ratio of the disector height (h) over t f is the final sampling frac-
tion. This is valid only when the variability of tf within and between sections
is very low, which seldom happens. If many objects are present in a few lo-
cations/sections that are thicker, and fewer in many locations/sections that
are thinner, the mean fraction will tend to be larger and the object number
estimated smaller than if the opposite situation is true. It has then been
proposed that the contribution to t¯f of the tf measured at each spot, or at
least that of the mean value of tf for each section, be weighted by the relative
contribution of objects counted ( Q − ) at the corresponding spot/section
(Bermejo et al., 2003; Dorph-Petersen et al., 2001). The latter correction is
given by
s n −
s1 t i q
t Q− = s n − i (16.1)
s1 Q
where t¯Q − is the mean value over the sections of the mean thickness per
section, t¯i weighted by the neuron count per section, q i− .
V. CONNECTIONS AS DEFINED BY NEURONAL NUMBERS
In studies of neural connections, neuron numbers usually refer to those

cell bodies that are retrogradely labeled by axonal tracers, or are labeled
through transneuronal anterograde transport by neurotropic viruses. Indi-
rectly, a measure of connectivity is also given by the total number of pro-
jection neurons located in a regional domain whose target is known (e.g.,
granular cells of the dentate gyrus, parvocellular layers of the lateral genic-
ulate nucleus, lateral cerebellar nucleus, etc.), or local circuit neurons in a
defined territory (e.g., inhibitory interneurons in thalamic nuclei, granular
cells of the olfactory bulb, bipolar cells of the retina, etc.). These neurons
may have been labeled by a variety of histochemical or immunocytochem-
ical procedures. Moreover, when using anterograde tracer injections, it is
also desirable to grasp the magnitude of the cell population that took up
the tracer by counting strongly labeled neuronal bodies.
A. Tools
1. Equality Guaranteed
Counting cells has to rely on a well-established stereological principle: all

cells have to have the same probability of being counted. In order to comply
with that principle, it is necessary to fulfill three requirements:
1. Counting in 3D demands that an unbiased 3D sampling probe is “in-
serted” in the target tissue. This can be accomplished in either of these
two ways: by defining a known area in two registered serial sections of
practical zero thickness separated a known distance, or by defining a
known area at a certain focal plane in a thick section, and “moving”
downward the focal plane a known distance along the z-axis. In both
cases the 2D area is delineated by the already classical unbiased count-
ing frame (Gundersen, 1977; Sterio, 1984), and the distance between
the top and the bottom planes incorporates the third dimension (dis-
ector height h) to create a 3D unbiased “box” or “brick.” If appropriate
disector height and counting rules are respected, these probes are the
basis of the physical and optical disector methods (Howard and Reed,
2005; Fig. 16.5).
2. A single feature of a cell has to be identified as the “counting unit,”
so that each cell is counted only once, regardless of its size or shape.
The theory goes that “cell tops” (i.e., their first appearance as a recog-
nizable target) fit the requirement, and so it is for nuclei or nucleoli
Figure 16.5. (A) The physical disector. An unbiased counting frame is overlaid on
matching regions of two thin sections, separated by distance h, showing some stained
and unstained cell body profiles. The frame consists of a solid line marking the
exclusion territory and a dashed line bounding the acceptance area. A guard area
is left around the frames so that all candidate profiles can be seen in their entirety.
The upper disector is used as a reference (inclusion) plane (R-p) and the lower as a
look-up (exclusion) plane (LU-p) when performing the disector counting from top
to bottom; planes are reversed when performing the counting the other way. (B) The
optical disector (slightly modified from Williams and Rakic, 1988, with permission
from Wiley-Liss, Inc.). A virtual counting “box” is created inside a thick section by
scanning a fraction of its thickness (h over tf ) with an unbiased frame, leaving guard
zones around it. Darkened particles are excluded from the count for invading the
forbidden space. As for the physical disector, the sampling volume is defined as
v(dis) = xy h.
in a physical disector setting. The use of the cell body surface for this
purpose, while possible, is often inconvenient, because of its irregular-
ities and the ambiguous interface between body and dendritic bases.
When using an optical disector it is better to rely on nucleolar or nuclear
“optical equators” (planes where the nuclear membrane or the nucle-
olar surface is at its sharpest focus), because of the difficulty inherent
to define membranes or surfaces cut at nearly tangential angles at the
top (or bottom) of the body, nucleus, or nucleolus.
3. Since each disector explores only a tiny part of the target region, it is
necessary to ensure that the whole region is surveyed with disectors
distributed in a statistically correct manner. Whilst a simple or uniform
random (UR) sampling may be used, it is more efficient to follow a
systematic random sampling (SRS) regime (Gundersen and Jensen,
1987; Gundersen et al., 1999).
2. Densities Vs. Total Numbers
It may occur that only a fragment of the whole target is accessible to study
(brain biopsy, incomplete collection of a nucleus or region), or that the
target boundaries are imprecise (some “reticular,” transitional, or otherwise
inconspicuously bounded cortical or subcortical areas). In such cases, it is
possible to estimate only the real neuronal densities (number of cell bodies
per unit of tissue volume, NV ) as

−
Q
N̂V = (16.2)
v(dis)
where Q − is the sum of cells counted, and v(dis) is the total volume of
tissue explored by the disectors.
If the whole target region (or reference space) is available, and its vol-
ume can be determined, then the total neuron number may be indirectly
estimated:
N̂ = N̂V · Vref (16.3)
It is evident that these estimates of NV or N are sensitive to tissue defor-

mations. NV increases or decreases when net shrinkage or swelling occurs,
respectively. And the estimate of N given by Eq. (16.3) may vary if Vref is
determined at a different stage of tissue processing. This may occur, for
example, when only fragments of a region of interest are available to per-
form the number estimation, while volume estimates of the whole region
were obtained in the living individual by imaging techniques. In this case,
it is necessary to introduce a correction that accounts for any differential
deformation that may have taken place (see section “Densities and Total
Numbers”).
When the whole region is accessible, it is possible to estimate directly
the total number of cells by the fractionator, a procedure that is efficient,
robust, and insensitive to tissue deformations (Gundersen, 1986). It can
be associated with the optical disector method in what was called optical
fractionator (West et al., 1991), and has become a widespread tool to count
neurons. Recent developments have improved its efficiency (Gundersen,
2002b). The rationale behind this procedure is simple: if a known fraction
of the target region is adequately sampled, and the cell count in that fraction
is Q − , then

N̂ = Q − · f T−1 (16.4)
where f T−1 is the inverse of the total, or final, fraction taken, which results
from computing the fractions taken at the various steps of sampling. A typical
paradigm consists of sampling a fraction of the serial sections, which contain
the target territory ( f s ), the areal fraction of the structure that is covered by
the sampling frames in the sections ( f a ), and the linear fraction of section
thickness covered by the height of the disectors ( f h ):
1 1 1
f T−1 = (16.5)
fs fa fh
The fractionator does not need any absolute measurements (of reference
space, stage movement, magnification, etc.), and therefore is unaffected
by dimensional changes of the tissue during processing. However, the last
fraction f h requires fair estimates of the final section thickness, and hence it
is necessary to keep the annoying t on a tight rein (see section “Deformation
Artifact”).
B. Practical Tips
As a general recommendation, it is necessary to keep track of all data that

are relevant to understand—and possibly replicate—the sampling protocol,
and the values entered in the different estimators. Otherwise, errors may
easily slip in, and skepticism may understandably grow in confused readers.
The whole target region should be defined unambiguously. If this is not
possible (e.g., because of blurred borders or partial unavailability), the best
possible sampling with disectors should be used instead of the fractionator.
A good design for the optical disector (or optical fractionator) must be
efficient and not excessively labor-intensive. This is not always the case for
the physical disector.
Vertical shrinkage may be considerable (up to 65%) when using vibratome
or frozen sections. If these sections are obtained at a nominal thickness below
30 µm, it is possible that there is no room left for the optical disector height
to be at least 10 µm. Thinner disectors risk being too “noisy,” because of the
limited focal resolution (never better than 0.5 µm).
The importance of measuring t with accuracy and a maximum precision
cannot be sufficiently emphasized.
Staining must be complete, but, if too intense, it may create serious prob-
lems of overlapping.
A good selection of the counting units increases accuracy and efficiency.
As commented above, cell bodies are not as good options as nucleoli or
nuclei.
Fluorescent tracers of cell bodies and neurites are an excellent, and in-
creasingly used option to label neurons and connections with chemical
specificity. However, stereological applications to fluorescent material (and
other dark-field applications) have particular difficulties, which cannot be
ignored.
Confocal microscopy provides excellent focal resolution and noise elimi-
nation. However, the technique has its own constraints (apart from the costly
equipment), and at present stereological applications in this field are just
starting. While the future is certainly promising, conventional fluorescence
optics with a good selection of filters together with Nomarski optics could
yield nearly as good results (Peterson, 2004).
Stereology has been seldom applied to quantitate the neuron groups that
are directly affected by the local deposits of tracer. However, such informa-
tion could be of great value to assess the degree of divergence and conver-
gence of afferent and efferent connections, evaluate the effective injection
site (EIS), or study tracer clearance or fading in the injected area over time
(see Appendix).
VI. CONNECTIONS AS DEFINED BY AXON NUMBERS
Two different scenarios pose quite different challenges to the task of

counting axons. The easier situation is determining the number of axons
in a peripheral nerve (or a central tract) cut more or less orthogonally to
its major axis. The second, more complex, scenario is the estimation of the
number of axons of a given kind that penetrates or distributes in a target
territory.
A. Axon Numbers in 2D
Counting axons of peripheral nerves or central tracts is efficiently accom-

plished by applying a 2D fractionator sampling scheme to properly stained
thin cross sections of the target structure (Mayhew, 1988, 1990). Thick sec-
tions could also in principle be used, but the fibers usually display wavy
or slanted trajectories, which cause severe problems of overprojection and
overlapping. What follows is summarized in Fig. 16.6, and borrows largely
from the excellent review by Larsen (1998).
The procedure is particularly simple when only myelinated axons are
to be counted. A 1-µm-thick section of the resin-embedded nerve or tract
is stained with toluidine blue and coverslipped. The whole cross-sectional
area containing fibers is sampled at high magnification, using a 100× oil-
immersion lens. An unbiased counting frame of a known area a(frame)
is systematically advanced in a meander path at preestablished dx and dy
steps from a random starting point. If Q is the number of myelinated
axon cross sections included in all the frames, then the total number of
axonsin the nerve is estimated using the fractionator equation [Eq. (16.4)]
N̂ = Q · f T−1 , where f T is now the only fraction performed, namely the
ratio a(frame)/dxdy .
When unmyelinated axons are also a target, it is necessary to work on
electron micrographs at a final magnification of at least 8000×. If the whole
nerve or tract section fits the supporting metal grid, a complete sampling
may be performed and the total N may be estimated by the fractionator
equation as above (Fig. 16.6B). If the nerve is too large, however, it is possible
that only a part of it is accessible. An alternative then exists, which is to
estimate the density of axon profiles per area,

Q
N̂A = (16.6)
a(frame)
from which the total axon number may be estimated, if the whole cross-
sectional area of the nerve is available from semithin sections (and correct-
ing for distortions in the ultrathin sections).
Most of these stereological analyses can be carried out capturing still video
images (from semithin sections), or digitizing EM negatives, and merging
them with customized test systems using any commercially available graphic
Figure 16.6. Counting axonal profiles in cross sections of a small nerve or fascicle
by the 2D fractionator. (A) A virtual grid is overlaid on a toluidine blue-stained
1-µm-thick cross section of a distal branch of a cat median nerve. The grid may be
created by a “meander” X-Y displacement of the microscope stage at dxdy steps.
Within the upper left grid square a counting frame is randomly located; the rest are
systematically placed at dxdy steps. At highest optical magnification practically all
profiles of myelinated axons can be identified. (B) Unmyelinated axons can only be
counted on EM images (slightly modified from Larsen, 1998, with permission from
Elsevier). Sampling includes an additional intermediate step to locate small fields of
vision where the EM pictures will be made. Each micrograph is examined with four
sampling frames, to increase efficiency and reduce the impact of axon clusters on
sample variance.
software. However, the efficiency improves greatly by using a stereological

package (see section “Morphometry and Stereology”) to control the X–Y
stage displacements and generate the frames and other test systems. Sam-
pling on the EM can also be expedited by using digital capture of the images
instead of conventional photographic film.
B. Preterminal and Terminal Axonal Fields
A quantitative description of the innervation of a territory would seem aw-

fully incomplete without estimating the number of incoming axons. How-
ever, this parameter has been largely neglected, probably because of the
technical difficulties involved. A quantitative approach to axonal arbors has
been practically restricted to metric and topological analyses of the whole
or part of Golgi or intracellularly stained axons of individual cells, recon-

structed in 2D (with camera lucida) or 3D (with digital imaging systems).
Similar analyses have been more widely applied to quantitate dendritic ar-
bors. Very recently, however, semiautomated vectorization for axonal recon-
structions presents a promising alternative to axonal topology (see Ascoli
and Scorcioni, this volume).
Stereology is a convenient, and often necessary, alternative to complete
reconstructions, when dealing not only with axonal arbors from cell popu-
lations, but also with very complex arbors from single neurons (Fig. 16.2).
Two parameters may be of particular interest: the number of somehow la-
beled preterminal and terminal axons in a defined target territory, and the
number of varicosities (terminal or en passant) displayed by those axons.
1. Axons
Counting axons leads us away from metric analysis and takes us to topol-
ogy. The number of axon segments is a function of the number of branching
points (or nodes) and is independent of axonal length, thickness, or tortuos-
ity. Most nodes correspond to axonal bifurcations, whereby three segments
meet in the node (nodes with valence = 3); more rarely they mark trifurca-
tions (valence = 4). Axon branches are not supposed to form closed loops,
but rather give off collaterals in several intermediate steps until making ter-
minal segments. Their analysis is then simpler than when fibrous or tubular
structures arise and end in an interconnected meshwork. Such is the case of
the microvessels in a brain region, which have been counted by a procedure
that applies the disector method to a topologically defined microvessel unit
(Løkkegaard et al., 2001).
A simplified version of that procedure, suited to count axons in thick
sections, would be given by the estimator

Pn (nval − 1)
ŴV = (16.7)
v(dis)
where WV is the density of axon segments per unit volume of the target
region, Pn is the number of nodes counted in the volume sampled with the
disectors [v(dis)], and nval is the valence of each node counted. If tri- or
multifurcations are very rare, the numerator may be simplified to Pn . If
the total volume of the target region is available or can be estimated (e.g.,
by the Cavalieri method), then the total number of axons in that region may
be computed as
Ŵ = ŴV · Vref (16.8)
When applying this procedure, some caveats are in order
1. Neither root segments, nor axon branches that divide no further
within the disector space (Fig. 16.7) are included in the count. The
Figure 16.7. Diagrams to illustrate the stereological procedure to count axons.

(A) In a simplified, 2D setting, two axonal profiles are shown. One is unbranched
(gray), and the other (black) is highly ramified. The former has no nodes, and there-
fore is not counted at all. The second one shows 12 nodes, 10 bifurcations, and 2
trifurcations. If Eq. (16.7), with a simple adaptation to 2D, is applied, the estimation
of axonal segments yields 26, which is the real number of segments, except for the
root segment on the lower left corner. (B) A network of varicose axons of various
types is sampled by an optical disector box. Axon portions inside the box are shown
in dark shades; those outside are in light gray, and transects with the box walls are
hatched. Five nodes (bifurcations; 1–5) are present within the inclusion volume of
the sampling box, yielding a count of 10 segments [Eq. (16.7)]. In fact, there are
13 segments, but 3 of them correspond to (relative) “root” segments entering the
box. Undercounting of root segments is only apparent: statistically they have been
counted at their origin in another node, as shown by the fiber marked by an arrow.
An additional axon winding on the upper part of the box does not have any node
and therefore is not counted.
exclusion of root segments introduces a meaningless bias, since it is

local branches what we look for. The omission of undivided branches
is unbiasedly compensated by appropriate sampling within the target
territory. As is usual in many stereological settings, about 200 events
(nodes) counted in the same number of, or somewhat fewer, disectors
systematically distributed should yield a satisfactory precision in the
estimate.
2. Nodes have to be identified and defined unambiguously. This requires
in first place that the staining has to be reasonably complete, that is, con-
tinuous along the full axon branches, or, if granular or discontinuous,
it should leave no doubt as to the continuity of the stained segment
fragments (Fig. 16.7B). Very thin axons can, in any case, fall below the
optical resolution (about 0.3 µm). Moreover, the nodes have a volume,
and therefore should be defined by their smallest possible single fea-
ture, such as the cusp of the smallest angle formed by the meeting
segments.
3. Although the estimator is based on a topological feature of axons,
it does not provide many interesting topological measures of axon
branches, such as the centrifugal order of each segment or node,

or the partition mode of each axonal subtree (Uylings and Van Pelt,
2002).
4. As for all density estimators, tissue deformations have to be taken
into account, if values are to be referred to the in vivo condition,
or comparisons between subjects or groups of subjects are to be
made.
2. Boutons
Terminal axon segments end in a swelling (terminal bouton), which is

filled with synaptic vesicles, and display a synaptic membrane specialization.
Similar swellings (varicosities, boutons en passant) often occur along the
axon stem and collaterals, and may bear synaptic membrane specializations,
depending on the cell of origin and the neurochemical content of the axon.
Boutons vary in size from the barely visible smallest cholinergic varicosities in
the cerebral cortex (0.3 µm in diameter; Chédotal et al., 1994) to the several-
micrometer-wide swellings in neurosecretory fibers in the hypothalamus, or
the mossy fiber terminals in the hippocampus.
The density of axonal swellings in a target territory is a relevant measure
of the local synaptic density. The total population of synaptic boutons can
only be quantitated using EM, which gives enough resolution to identify all
boutons, but introduces a number of difficulties (see section “Connections
as Defined by Synapse Number”). When specific axonal populations are la-
beled, however, it is possible to estimate total numbers or densities by light
microscopy, using the optical disector or the optical fractionator methods
[see Eqs. (16.2)–(16.5) in section “Tools”]. Although still seldom applied,
the stereological estimation of bouton numbers may provide a valuable,
and anatomically precise, indirect estimation of synaptic contacts in a de-
fined region (Calhoun et al., 1996). Still, it is unjustified to presume that a
one-to-one correlation exists between boutons and actual synaptic contacts
(Mayhew, 1979b; see section “What is a Synapse”).
VII. CONNECTIONS AS DEFINED BY AXONAL LENGTH
The length of a linear feature per unit volume of a reference space (length
density, L V ) may be easily estimated by the classical stereological estimator
L̂ V = 2 · Q A (16.9)
where Q A is the number of random intersections of the feature per unit
area of a 2D probe (a bounded plane, virtual or physical) introduced in the
space (Howard and Reed, 2005).
Despite its well-proven theoretical simplicity, the practical implementa-
tion of the estimator poses several difficulties. In first place, the encounters
of the feature with the plane must fulfill not only translational (or posi-
tional) randomness, but also directional randomness, or isotropy. Secondly,
real fibers (axons, dendrites, etc.) may be regarded as linear features of
nonzero thickness, which are usually studied in sections of nonzero thick-
ness. As we will see below, problems posed by length estimation differ when
section thickness is than fiber thickness, as in transmission EM studies,
and when it is ≥, typically for fibers in thick sections, and also applicable to
very thin axons or dendrites in semithin sections.
A. Navigating the Isotropy Reefs
Axon branches may take any direction. However, in any given territory,
not all directions are present with the same probability. That axons dis-
tribute anisotropically is a highly probable and prudent generalization. To
guarantee randomness, therefore, it is necessary to proceed along either of
two lines: (1) to generate sampling probes that are themselves isotropic, and
to distribute them within the target space in a random pattern and (2) to
randomize the tissue, by rotating tissue fragments along one or more axes.
If, in order to increase efficiency, sampling is made systematic (or uniform),
then the requisites for an isotropic uniform random (IUR) test system and
sample, respectively, are fulfilled. These procedures, which will be summa-
rized below, are clearly explained in recent stereological reviews (Calhoun
and Mouton, 2001; Howard and Reed, 2005).
1. Isotropy in 2D and 3D
Isotropy in 2D consists in guaranteeing that all directions in the plane

(represented by intersections of a line on a circumference) are equally
probable. An IUR design in 2D using a set of parallel test lines spaced a
distance d is easily achieved by randomly choosing an orientation angle ω
such that 0◦ ≤ ω ≤ 80◦ and positioning the set randomly on the target ob-
ject. When multiple objects are measured, several orientations (z = 3 − 4 is
usually appropriate) may be systematically applied: An orientation angle ω
is randomly selected in the interval 0–(180/z)◦ ; the next orientations are set
at ω + 180/z)◦ , ω + 2(180/z)◦ , etc. Alternatively, a prime number such as 37
may be chosen as the interval; the first orientation is randomly selected and
then in the range 0–37◦ . This approach has the advantage that, in practice,
all orientations used are different.
Isotropy in 3D requires ensuring that all orientations in the 3D space
(represented by intersections on the surface of a sphere of straight lines,
or axes, rotating about its center) are equally likely to happen. It depends
on the randomness of two angles: the longitude (the azimuthal coordinate
φ, in the range 0–360◦ , or 0–2π), and the latitude or its complementary,
the co-latitude (the polar coordinate θ , in the range 0–π/2). While φ may
be chosen with a standard random number table, θ can not. The reason is
that, on the sphere’s surface, spherical quadrilaterals of the same angular
dimensions, dφ× dθ, have larger areas the closer they are to the equator.
Thence, if axes are directed to the angles of all quadrilaterals having the
same angular dimensions, intersecting spots on the surface will crowd to-
ward the poles. To avoid this bias, it is necessary to weigh the co-latitude
with its sine; thus, the probability of a line (orientation) hitting any of the
mentioned quadrilaterals is (Cruz-Orive, 2002b)
1
P = · dφ · dθ · sin θ (16.10)
2π
which means that the probability is 0 in an exact North Pole direction (when
θ = 0), it increases toward the equator, and then decreases to become again
0 in the South Pole direction.
2. Randomizing Tissue
There are two well-known procedures to generate IUR sections from a

tissue block. The orientator (Mattfeldt et al., 1990) is applicable when the
block is large enough to permit that fragments or slabs of the block are
subjected to successive stages of sectioning to ensure randomness in φ and θ .
when the target object is divided into small fragments, they can be placed in
separate spherical molds and embedded in resin. Once hardened, each resin
ball is allowed to roll free and is then reembedded in a standard rectangular
mold. This is the original isector method (Nyengaard and Gundersen, 1992),
which more recently was adapted to preembed several tissue samples in
spheres of agar, which in turn are embedded in methacrylate (Løkkegaard
et al., 2001).
In an IUR design, sections have to guarantee randomness in the three
axes. It is possible, however, to ensure full isotropy if all sections are parallel
to a fixed, arbitrary “vertical” axis. This vertical uniform random (VUR) de-
sign, developed by Baddeley et al. (1986; see also Howard and Reed, 2005),
is of particular interest for research fields that, like neuroanatomy, rely heav-
ily on serial sections. In this design all sections are made perpendicular to
the “horizontal” plane and parallel to a randomly chosen orientation in
that plane. It is important to realize that VUR sections are not isotropic,
because they lack isotropy along the vertical axis (VA). As will be further
commented below, the missing isotropy is provided by the cycloids, a special
type of test lines whose length contains a continuous set of orientations that
are proportional to sin θ .
Randomizing tissue samples, particularly in the IUR designs, has the in-
convenience of having to break apart the original structure. This results in a
marked loss of the anatomical references that are of great help in studies of
neural connections. An exception may be found in the study of brain biop-
sies of cortical or subcortical territories, when overall data (e.g., the length
of all axons in the white matter; Tang and Nyengaard, 1997) are sought. The
problem also exists in VUR settings, because for length (but not number)
estimations, serial sections cannot be made at an anatomically convenient,
preestablished orientation. Rather, the plane of sectioning of each speci-
men has to be oriented randomly in a 0–360◦ interval to avoid bias. This has
been a strong drive behind new developments of fully isotropic probes that
can be used in any type of sections.
3. A Panoply of Isotropic Probes
Depending on the type of tissue and sections, density of target axons, and
randomness design chosen, five different methods can be applied. As in the
case of estimating number or volume, it is better to count dimensionless
events (such as “hits” or intersections) than actually measure in 1D, 2D,
or 3D. To ensure that such events are really dimensionless, the dimensions
of the probe and the target feature must sum 3 (Howard and Reed, 2005;
West, 1993). Therefore, all the probes used to estimate length are 2D (flat or
curved bounded planes; Fig. 16.8). While some of these tests may be imple-
mented manually on photomicrographs or live microscopic images, all bene-
fit greatly from (and some actually require) the use of stereological software.
1. Thin IUR sections: an unbiased frame is applied in a UR pattern to

the sections, using very high optical (or low EM) magnification (Fig.
16.8A). The length density of the axons is estimated as

Q(ax)
L̂ V = 2 · (16.11)
a(frame)
where Q(ax) is the total number of axon profiles counted in the

sampling frames, and a(frame) is the total area covered by the frames
hitting the target region. The total axon length may be estimated if the
volume of the target region is known. This method has been applied
to estimate length density and total length of myelinated fibers in the
white matter of human brains (Marner and Pakkenberg, 2003; Tang
and Nyengaard, 1997).
2. Thick VUR sections: a cycloid, with its longer axis parallel to the VA of
the section, is projected through the section depth. All intersections
of target axons with the cycloid are recorded. For efficiency, an array
of cycloids with associated points is recommended (Fig. 16.8B). Since
axons are likely to span a number of sections, all sections must have
the same thickness. The length density is estimated as (Gokhale, 1990;
see also Batra et al., 1995 and Stocks et al., 1996)

Ic
L̂ V = 2 (16.12)
Pcl c t¯s
Figure 16.8. Stereological test systems for estimating axonal length. (A) Myelinated
axons in systematically sampled IUR semithin sections of the rat cerebral cortex
are counted with unbiased frames. Gray levels in the exclusion zones and an ample
guard area have been dimmed. (B) An array of cycloids with points associated (one
for four cycloids) is used to scan a “box” of tissue within a vertical thick section that
contains a network of axons. The major axes of the cycloids are parallel to the vertical
axis (va), which has to be identified in all sections. (C) Sampling of a thick section
using a single isotropic virtual plane. The plane is computer generated at a random
(isotropic in 3D) orientation. The line marking the intersection of the plane with the
upper surface of the sampling box “moves” along the plane (lm), as the focal plane
changes along the z-axis. Five intersections (black profiles) of axons with the virtual
plane are counted. (D) Sampling a thick section with a virtual isotropic sphere. The
sphere may be computer generated, or manually constructed as a series of concentric
circumferences representing transects of the sphere surface at different focal depths
inside the sampling box. Seven axon intersections are produced on the surface of
the sphere. The focusing direction (fd) is indicated in (B–D).
where the numerator represents the total number of intersections of

the cycloids with the projected axons, and the denominator is the total
surface area of the cycloid used: Pc is the number of cycloids (from
the count of the associated points), l c is the individual cycloid length,
and ts is the cycloid depth, represented by the mean section thickness.
3. Total vertical projections (TVP; Cruz-Orive and Howard, 1991): it is possi-
ble to estimate the total length of the branches when the whole arbor
is present in a thick section by projecting it on a plane. The projection
must be repeated a few times after UR rotations of the arbor. The total
length of the branches is a function of the number of intersections of
the projected profiles with a grid of cycloids. Although applications to
axonal arbors are lacking, this system was used to estimate the total
dendritic length of Golgi-stained substantia nigra neurons (Howard et
al., 1992). Computer-based 3D reconstructions of axons or dendrites in
confocal microscopy may be good candidates to apply TVP to estimate
length in selected brain regions.
4. Virtual isotropic planes: parallel virtual planes at random orientations are
generated by software and projected, under high magnification, inside
thick sections obtained at any convenient arbitrary orientation. As the
image is focused along the z-axis of the physical section, a “moving”
line is visualized sliding along the virtual plane (Fig. 16.8C). In order
to explore the same area with each random orientation, only the inter-
sections of several equidistant parallel virtual planes within a “sampling
box” of constant volume are used (Larsen et al., 1998a). An unbiased
estimator of length density is then

Ivp
L̂ V = 2 · (16.13)
a(vp)
where Ivp is the total number of axonal transects with the virtual
planes, and a(vp) is the sum of the sampling plane areas. The length
of regenerating axons in spinal cord lesions has been recently estimated
using this procedure (Larsen et al., 2004). Virtual planes and the above
estimator are implemented in the CAST software.
5. Virtual isotropic sphere: a translucent, computer-generated virtual sphere
(or hemisphere) is UR placed across the target region in thick sections
under high magnification. The sphere is contained within an unbiased
box whose dimensions are greater than or equal to the sphere diameter
(Fig. 16.8D). Since the surface of the sphere is an isotropic curved plane
that contains all possible orientations, the tissue sections may have been
cut at any arbitrary orientation. Length density is estimated as

Iss
L̂ V = 2 · (v/a) · (16.14)
v(dis)
where (v/a) is the ratio of the sampling box volume to the sphere’s
surface area, Iss is the total number of axonal transects with the
sphere’s surface, and v(dis) is the total volume of disectors used.
As for the virtual plane estimator, the total length may be estimated
here if the volume of the reference space is known, or a fractionator
scheme is applied. Length density and total length of cholinergic
fibers in the neocortex and the hippocampus have been estimated
using this procedure (Calhoun and Mouton, 2001; Calhoun et al.,
2004). Although virtual spheres can be generated manually (Mouton
et al., 2002), the task is greatly facilitated by stereological software.
Virtual spheres (or “space balls”) have been recently implemented in
the StereoInvestigator software.
B. A Caveat: Axons are Irregular Tubes, Not Lines
If isotropy conditions are properly met, simple linear features can be

unbiasedly measured by any of the procedures described in the previous
sections. When the features have a cross-sectional area, as for tubes, their
axis (or spine) is hidden by the tube walls, and tube intersections with a
plane become 2D profiles. This generates a number of problems, whose
practical importance differs depending on whether the tubes have a stable
thickness or present irregularities (swellings and constrictions), the degree
of branching of the tubes, the frequency of free endings, and the type of
section and probes used (Gundersen, 2002a; Larsen et al., 2004).
1. When the arbor is located in a section of thickness much greater than

the average tube (axon) diameter and is projected on a plane (see TVP
design, previous section), in principle the number of intersections of
the tube (axon) walls with the test lines is a close match of (twice) the
number of intersections with the invisible axis (Fig. 16.9A). However,
severe problems of overprojection and overlap may occur, leading to
uncontrollable underestimations. Moreover, the thicker the section,
the more difficult to combine full focal depth with enough magnifica-
tion, both conditions being necessary to identify all profiles.
2. If, on the contrary, section thickness is less than or equal to the axon di-
ameter, it is not possible to derive estimates of the real axon length from
Figure 16.9. Problems in measuring length of tubular structures. (A) A tubular struc-
ture seen under TVP in a thick section (a) or in a vertical section of thickness less
than or equal to the mean tubular diameter. In (a) the total length (M) is a function
of the number of intersections (I ) of the cycloids with the projected tube boundaries
(B). In (b), however, there does not exist a simple relation between the tube profiles
and their imaginary (invisible) axes, and therefore length estimations based on the
intersections with the cycloids are biased (reproduced from Gundersen, 2002a, with
permission from Blackwell Publ. Ltd.). (B) Thin sections of tubular structures yield
a number of 2D profiles, which may unbiasedly represent intersections with the tube
axis. For smooth tubes, the axial bending and curves are not a problem, because the
number of random transects with the axis matches the number of whole profiles
(a). Branching leads to underestimates, and terminal free endings to (negligible)
overestimates (a). Nonsmooth tubes produce more severe degrees of overestimation
(b). This is particularly the case for markedly beaded axons or dendrites. (Modified
from Gundersen, 2002a and Mouton et al., 2002).
the intersections observed between axon profiles with test lines. The
reasons are multiple (Fig. 16.9B): curved trajectories are not a problem,
but branching leads to underestimation, free ends to a slight overesti-
mation, and irregularities in the fiber thickness lead to overestimation.
3. Scanning through thick sections at high magnification and with the
finest possible depth of focus, using optical disectors (or confocal
microscopy), eliminates several problems inherent to the TVP design.
Its combination with appropriate test systems is, so far, the most
reasonable approach to estimate length.
In sum, there is no simple solution to measure the length of tubular
structures, such as axons, in a completely unbiased, efficient, and precise
manner. The problem is still open to new theoretical and practical develop-
ments (Gundersen, 2002a; Mouton et al., 2002).
VIII. CONNECTIONS AS DEFINED BY SYNAPSE NUMBER
The number of synapses that a neuron, or a system of neurons, makes

in a target territory is undoubtedly a fundamental measure of neural con-
nectivity. Since the late 1960s, transmission EM has been used to assess
synapse numbers and density. But researchers soon realized that synapses
were geometrically complex structures, and that inferences about numbers
did not smoothly derive from linear and planar measures taken on EM
images.
The simplest scenario would feature synapses as disk-shaped, flat or
slightly convex, unambiguously bounded structures of similar (and not too
small) size. In such a case, accurate number estimates could be given by
classical formulas, such as the widely used NV = N A /d̄, where d̄ is the mean
trace length of the synaptic profiles (Colonnier and Beaulieu, 1985; De-
Felipe et al., 1999). Yet, overprojection and truncation, which may occur
even in this ideal situation (and be significantly greater in others), have led
to the introduction of a variety of correction factors, with mixed results as
to the improvement of the estimates (Calverley et al., 1988; Colonnier and
Beaulieu, 1985; Mayhew, 1979b; Zaborszky et al., 1975).
Certainly, accepting assumption-based counting methods, which are
hardly extendable to all conditions of the feature counted, does not help
to meet the challenge posed by the practical problems inherent to synapse
morphometry. In fact, the realization of these problems cannot but highlight
the importance of using unbiased stereological methods, without ignoring
the difficulties in their implementation.
A. What Is a Synapse?
Defining a synapse as a morphological entity that can be used as an un-

ambiguous counting unit entails difficulties, which have been recognized
some time ago (Mayhew, 1979a). For the chemical synapses, different com-
ponents have been used for this purpose, each with specific advantages and
limitations.
1. The Presynaptic Boutons Themselves
Vesicle-filled swellings along or at the end of axonal branches gener-

ally qualify as presynaptic entities, but, are they? Boutons identified by im-
munostaining of molecules associated with synaptic vesicles have been used
as counting units with light microscopy (Calhoun et al., 1996; Silver and
Stryker, 1999). Yet, many vesicle-containing varicosities lack synaptic mem-
brane specializations, particularly in cholinergic and bioaminergic axonal
systems (Chédotal et al., 1994; Descarries et al., 1991). While representing
other kind of (nonjunctional) synapse, they cannot be equated to classical
(or junctional) synapses. Some would set stricter criteria for the latter, re-
quiring that a few vesicles contact the axolemma. In addition, a significant
number of boutons display paramembranous densities apposed to more
than one postsynaptic element (Jones et al., 1997; Sorra and Harris, 1993).
Finally, synaptic boutons may fail to show presynaptic densities along a series
of ultrathin sections, and this effect is size dependent.
2. The Synaptic Cleft
In principle, this space could be considered the anatomical kernel of

a classical synapse, both symmetrical and asymmetrical. It presents, how-
ever, a major practical disadvantage because of its small transverse diame-
ter (20–40 nm) and relative electron lucency. Since average ultrathin sec-
tions are at least 60 nm thick, and the flanking membranes are electron
dense, the synaptic cleft is subject to severe problems of underprojection and
truncation.
3. The Pre- and Postsynaptic Membrane Specializations
As pointed out by Mayhew (1979a), there was an early shift of empha-

sis toward accepting the whole apposition complex, or synaptic plate (the
paramembranous densities plus the synaptic cleft), as the best unit for
counting classical synapses. The membrane densities can be stained by
the standard osmium-uranyl-lead procedure or the specific ethanolic phos-
photungstic acid method; with either method they become electron dense.
Moreover, the transverse thickness of the complex (or synaptic height) is
larger than the average section thickness. Yet, recognition problems that
arise from the size and morphology of the complex may significantly bias
the estimates.
B. Unbiased Counting
Given the dimensions of the synapses, the physical disector (section “Con-
nections as Defined by Neuronal Numbers”) is the only stereological method
capable of yielding unbiased estimates of synapse number, using the synap-
tic plate as the counting unit. Even so, counting synapses is no free ride,
and faces a number of theoretical and practical problems, which have been
recently discussed in detail by Tang et al. (2001).
1. Strict and educated criteria have to be applied to recognize synaptic

profiles in 2D, so that they can be defined independently in each sec-
tion of a disector pair (Coggeshall, 1999). The difficulties involved
are substantial, as shown by comparing the results obtained with the
disector and with a serial analysis of many sequential sections (stack
analysis; De Groot and Bierman, 1983; Tang et al., 2001). Any part of
the synaptic plate included in the sampling area of the disector frame
should qualify for the count. Observational truncation (section “Ob-
servation Artifacts”) is a common problem, when tangential sections of
the synaptic plate periphery offer limited information. If very rigorous
criteria to recognize a synapse as such are set, truncation may account
for nearly 25% underestimation (Tang et al., 2001).
2. A few serial sections (3–7) are collected in a formvar- or pioloform-
covered single-slot EM grid. Section thickness has to be measured on
each section of every disector pair (t1 , t2 ). Small’s minimal fold method
(De Groot, 1988; Small, 1968) is recommended for its relative easiness
and not needing additional equipment. It is customary to use adjacent
sections for a disector pair, but alternate sections may be used as well, on
the condition that the thickness of the intermediate section (tint ) is also
known. It is highly improbable that any synapse would fit entirely (and
thus invalidate the procedure) in the missing section. Disector height
h(dis) would equal t¯ = (t1 + t2 )/2 if adjacent sections are used in one
direction, and (t1 + t2 ) if (as recommended for efficiency) counting is
done both ways. When alternate sections are used, and both ways are
used to count, h(dis) = t1 + t2 + 2tint .
3. Apart from substantial size differences, synapses may adopt complex
shapes. Varying sizes or curvatures should not affect disector counting,
unless so pronounced that a single synapse would appear as two or more
disjoint profiles in two or more adjacent sections. But complexity usu-
ally refers to irregularities in the 2D geometry of the plate produced
by simple and complex fenestrations in the paramembranous densi-
ties. Complete fenestration results in a segmented or double/multiple
synapse. If marked changes in curvature are added, W-shaped and
other bizarre 3D configurations can be obtained (Fig. 16.10). In order
to avoid the overcounting bias due to mistaking a horseshoe-shaped or
incompletely perforated for more than one synapse, Tang et al. (2001)
incorporated a topological rule: if two disjoint plate profiles in one
A B C
Figure 16.10. The shape of synaptic contacts in the same region may change sig-
nificantly under different natural or experimental conditions. (A) shows a com-
mon field of the molecular layer of the dentate gyrus in a control rat. Most synap-
tic profiles correspond to flat (straight arrows) or convex (curved arrows) sim-
ple axospinous synapses. In (B) and (C) there appear two highly complex spines
with fenestrated synaptic plates (thin arrows) from the same region, several weeks
after partial denervation by transection of the perforant fibers. Calibration bar:
0.5 µm.
section were matched by a single plate in the paired section, it was

a proof that the latter bridged the former; all were part of the same
synapse, and therefore no synapse should be counted.
4. A new procedure has been proposed to avoid the limitations inherent
to the physical disector. It consists in estimating NV or NL on unbiased
3D bricks projected within computer-assisted volume reconstructions of
a test region, made from large stacks of serial ultrathin sections (Fiala,
2005; Fiala and Harris, 2001).
C. Of Pincushions and Haystacks: Sampling Designing
As a general rule, uniform (or systematic) random sampling is recom-

mended for counting objects of any kind in stereology, and this includes
synapses. However, the vast dimensional “distance” that exists between
synapses and the reference structure (a whole brain, the striatum, a tha-
lamic nucleus, etc.) makes a proper sampling an unusually demanding task.
And it is more so when we want to sample, say, less than a trillionth (<10−12 )
of the target region. This was the case in Tang et al.’s study of synapse num-
ber in human brain neocortex, but huge disproportions between sample
size and reference space are likely to occur in most studies of regions larger
than a very small brain nucleus. True, sample size would be of little relevance
when the distribution of target objects is rather uniform. But since this is
not usually the case for synapses, the problem remains, being a cause of
concern (DeFelipe et al., 1999). The following comments may help to ease
that concern:
1. Using the cerebral cortex as an example, there is a much higher synapse

density in the “thin” neuropil than in the remaining tissue, where cell
bodies, blood vessels, large dendritic shafts, and myelinated axons con-
centrate (DeFelipe et al., 1999). Yet, if the target objects are asymmetric
synapses in general, some will be present in practically all sampled fields
at a convenient magnification (30,000–40,000×), except in those fully
hitting blood vessels or neuron bodies. Moreover, blood vessels and
neuron bodies only account for a minimum of 3% in layer I and a max-
imum of 17% in layer IV of the rat parietal cortex (Machı́n-Cedrés,
2004); therefore, their impact on the efficiency of the procedure is
small.
2. It is certainly harder to find pins in a haystack than in a pincushion;
in fact, if there is only one pin and the haystack is bulky, any sampling
design will very likely return 0 pins found. This does not invalidate
an unbiased counting procedure; it will simply orient toward a more
directed search, that is, to increase efficiency by sensibly narrowing
the defined reference space. If data for specific types of synapse are
sought, then specific practical solutions have to be applied. Good ex-
amples could be the symmetrical synapses around the initial segment of
pyramidal cell axons, the axosomatic contacts, the synaptic contacts on
a defined subpopulation of neurons, etc. These cases may show a very
low overall density of events, with clustering in specific regions. In such
cases, it is advisable to define—and count or measure—new convenient
reference spaces, such as the number, volume, and/or membrane sur-
face of initial segments, neuronal bodies, or labeled cells in a given
region. They would be used as subindices for density estimations, and
sampling would be directed only to the targeted space.
3. Two widespread misconceptions emerge from the true observation
that the goal of many quantitative studies is just to make comparisons
(Guillery and Herrup, 1997). One is that absolute numbers are of little
interest. While difficult to disprove, it may be argued that the interest
will rise exponentially when accurate data become available; moreover,
real rather than artificial data greatly benefit biological simulations and
modeling. The second is the assumption that biased or inaccurate pro-
cedures may not be too harmful, as long as the same are applied to
the experimental conditions that are being compared. This is wrong,
and potentially very harmful, because it does not take into account
that the reference space or even the tissue deformation due to fixation
may vary between experimental (or natural, such as aging) conditions
(Haug et al., 1984), or that the size and shape of the objects/synapses
may vary substantially in the same region under different conditions
(Fig. 16.10).
D. Densities and Total Numbers
While it is possible in principle to apply a fractionator sampling scheme to

synapse counting, in practice it would be extremely inefficient to section ex-
haustively the (proportionately very large) sampled pieces, and keep track of
the fraction used at every step (however, see Nyengaard et al., 2005). Hence,
it is more convenient to indirectly estimate the total number, from the esti-
mation of the numerical density of synapses, and the volume of the reference
space (section “Connections as Defined by Neuronal Numbers”; Geinisman
et al., 1996; Tang et al., 2001). Since the latter has to be estimated indepen-
dent of the former, it is important to control the dimensional changes that
occur between the processing stages.
Volume shrinkage between the fixation step and the ultrathin sections
may be determined by areal measurements in two steps:
1. From vibratome slices to resin-embedded blocks and semithin sections:
the area of the target region, or a well-defined part of it, is estimated
by point counting on the surface of the vibratome sections that will be
used for further processing. Adjacent sections, mounted and stained,
can be used instead, when no further X–Y dimensional changes are
expected. Because shrinkage is not necessarily isotropic in 3D, it is
recommended that vibratome sections are already isotropic (Tang et
al., 2001). When this is not possible (e.g., when the whole target region
is serially cut in the vibratome), two tissue pieces from nearby regions
are cut at two orthogonal angles, and a section from each is processed
in parallel to the remaining sections. When the section or tissue block
is embedded in resin, the area of the target region (or defined part)
is again measured by point counting directly on the embedded piece
or on the first semithin section taken from it. If doubts remain about
the existence of additional areal shrinkage in the cut section, this may
be easily estimated by comparing the total surface area of the section
with the cut surface of the block.
2. From the semithin section to the ultrathin section: the area of an ultra-
thin section is estimated by point counting on a low-power EM printout
of the section at a known magnification. Frequent calibration with a
grating replica is recommended.
This procedure should be repeated for a few blocks or vibratome sections,
especially when the tissue is markedly heterogeneous in terms of white mat-
ter, gray matter, and blood vessels content. The final areal shrinkage would
be estimated as

Ablock Aut
Shr(a) = (16.15)
Avib Ast
where areas, represented by point counts, refer to those obtained on vi-
bratome sections (vib), the tissue block in the resin (block), and the semithin
(st) and ultrathin (ut) sections. If heterogeneous samples of similar size are
used, the estimation of Shr(a) must be averaged over them. If blocks differ
in volume, the contribution of each block to estimate shrinkage must be
weighted accordingly.
The final volume shrinkage, as a percentage, may then be estimated as
(slightly modified from Tang et al., 2001)
3

Shr(%vol) = 100 1 + Shr(a) − 1 (16.16)
IX. A PRÉCIS ON ACCURACY, PRECISION, AND EFFICIENCY
An estimator of a parameter is accurate when the means of the estimates

obtained concentrate about the true value, i.e., when its mean error (bias) is
0 for any fixed sample size. When the sampling and counting protocols are
unbiased, estimates are accurate. If they are biased, the estimates will diverge
consistently (systematic error) from the true value, regardless of the sample
size. Precision is a measure of the degree of concentration of the individual
estimates about their own mean. It will be high when the objects counted or
measured are homogeneously distributed, and when sampling intensity is
adequate. Since a very imprecise estimate, though unbiased, is of little value,
the precision of an estimator must be determined, at least approximately. Fi-
nally, the efficiency of a method is a measure of the precision it offers per unit
cost, computing as cost the animals used and the time (and money) spent in
all procedures involved, from tissue processing to data collection and anal-
ysis. A practical glossary on these issues may be found in Cruz-Orive (2004).
The variability observed among individual estimates is a function of the
natural, or biological, intersubject variance, increased in the imprecision, or
“noise” inherent to the sampling/counting method (error variance). Variance
can be expressed in relative terms as the squared coefficient of variation
[CV2 = (SD/mean)2 ] of the variable R studied (R = number, length, etc.),
and a reasonable approximation to the error variance is given by the squared
coefficient of error of the estimates of R for each subject (CE2 ), averaged
over all subjects. This may be expressed as (Gundersen, 1986; Howard and
Reed, 2005)
OCV2 ( R̂) = CV2 (R) + mean[C E 2 ( R̂)] (16.17)
The CE is a good measure of the precision of individual estimates, and
the evaluation of its contribution to the total, observed variance should
help improve the sampling design. Since sampling in stereology is usually
done at multiple levels (subjects, blocks, or sections containing the target
region, areas sampled by sampling frames, disector volumes, and counting
points), it may be interesting to identify which of these levels contributes
more to the final variance. This can help design better strategies to increase
precision, keeping efficiency high. However, from practical and theoreti-
cal reasons, it is generally accepted that the critical level is that just below
the level that contributes most to the total variance. Certainly, we expect
that the observed variance represents as faithfully as possible the natural
or intersubject variance. Therefore, sampling should be increased at the
block/section level until the error variance attributed to the sampling and
counting procedure is reduced to levels considered acceptable. A popular
recommendation is that the contribution of the error variance (expressed
as CE2 ) does not exceed 50% of the OCV2 (Gundersen, 1986). This “rule”
is, however, rather arbitrary and not backed by any theory. It is possible, on
the other hand, to estimate at the outset a “tolerable” mean CE2 due to the
stereological method from Eq. (16.17), by fixing the error levels α and β
and the true relative difference of means between groups we want to detect,
and making a guess about the number of subjects that may be reasonably
used to detect that difference, as well as of the expected biological variance
of the target quantity among subjects (see Cruz-Orive, 2004 for a complete
explanation).
In measuring error variance, however, there is another catch. This error
cannot be determined exactly under systematic sampling—the one com-
monly used in stereology—as it is done for independent sampling (CE =
SEM/mean). The values of geometrical properties (areal size, number of ob-
ject profiles present) obtained on adjacent sections collected in a systematic
series are not independent, and their variance is a function of two indepen-
dent factors, the variance of the points or intersections counted for each
section, and the variance due to the systematic sampling of sections, which
depends on how—not on how much—the individual counts vary from one
section to the next. The estimation of CE becomes even trickier when using
ratio estimators, such as NV or L V , and the derived two-stage estimators,
like N̂ = N̂V · V̂ref . In such cases the overall CE is proportional to the sum
of the individual CEs, decreased by a factor proportional to the covariance
between them (Howard and Reed, 2005). On the other hand, there are sta-
tistical limitations that forbid such computations, under certain sampling
conditions (Pakkenberg and Gundersen, 1997).
An update on the methods to estimate the error variance in differ-
ent stereological designs may be found in recent papers by Gundersen,
Cruz-Orive, and their colleagues (Cruz-Orive, 1999, 2004; Cruz-Orive et al.,
2004; Garcı́a-Fiñana and Cruz-Orive, 2000; Geinisman et al., 1996; Gunder-
sen et al., 1999). It must be remarked that making a good sampling design
greatly benefits from analyzing variance and evaluating precision in a pilot
study. This is an onerous, but inescapable part of the stereological practice
for greenhorn quantitative neuroanatomists—another good reason for ac-
quiring direct experience in some of the practical courses mentioned at the
beginning of the chapter.
X. ADVANTAGES, LIMITATIONS, AND PERSPECTIVES
Shortly after the introduction of the disector (Sterio, 1984), the neural tis-
sue became a cherished target for the application of the “new” stereological
tools (Braendgaard and Gundersen, 1986; Pakkenberg and Gundersen,
1988). In less than two decades, stereology has become firmly established
as a required set of methods in neuroscience. Its solid theoretical founda-
tions opened up a new period of high expectations concerning the use of
unbiased counting and measuring methods. In some notable cases, these
expectations were formulated as requirements (Saper, 1996). However, and
more importantly, “neurostereology” is smoothly spreading mainly because
growing numbers of neuroscientists are convinced of its value. Whatever
reluctant attitudes may remain, they are likely to fade with the dissemina-
tion of and the delving into the advantages and limitations of the methods
proposed (Cruz-Orive, 2002a).
Before proceeding, it is convenient to stress that counting and measuring
represent an added value in neuroanatomy, but can never replace obser-
vation and description. As a particular advice for students, the qualitative
analysis of the tissue is not only necessary to evaluate the results of the stain-
ing method used, but is invaluable to design and refine whatever quantitative
method is to be applied.
Regional volume measurements and cell counting were, and probably
will be for a long time, the first objectives of “neurostereology.” The field of
application is boundless, from basic numerical descriptions of neural struc-
tures in different species, to development and aging, neuropathology, and
a variety of neural changes secondary to physiological, pharmacological, or
traumatic manipulations. Cell counting, however, was carried out by a variety
of nonstereological methods since the late 19th century. Understandably, it
is around cell counting where concern has grown among researchers and
journal editors alike over which are the “best” methods, and how they should
be reported and discussed in a publication. If properly applied by trained
operators, the stereological methods fulfill the first requirement for any
“good” method: accuracy, regardless of the type of counting objects. But
the conditional above points to the importance of a good design, a good
execution, and a good control of artifacts. Also for neural quantification,
the devil is in the details.
Length and surface measurements of brain structures were essentially
neglected before the advent of the new, design-based, stereological methods.
While barely starting today, their study will likely expand considerably in the
next few years.
New developments are expected in the implementation of stereological
applications to confocal microscopy (Howell et al., 2002; Kubı́nova et al.,
2004), neuroimaging in experimental and clinical settings (Roberts et al.,
2000), and, in general, all computer-assisted procedures directed to grab,
digitize, and manipulate images (see Bjaalie and Leergaard and Nadasdy et
al., this volume). It is advisable, however, to not be enticed by the automatic
analysis trap. At present, any quantification of neural tissue needs a consid-
erable amount of direct personal interaction with images. In fact, computer
assistance in stereology boosts efficiency mainly by quickly bringing to the
operator’s attention those and only those spots on which he or she must
make decisions.
The multiple spatial relationships among various cell or subcellular pop-

ulations are currently the focus of a variety of computational approaches
(Nadasdy and Zaborszky, 2001; Zaborszky et al., 2002). Second-order stere-
ology, whose main objective is the quantitative analysis of the spatial arrange-
ments of objects (Cruz-Orive, 1989), deserves therefore a separate mention.
Neural components are not scrambled into a chaotic meshwork. Rather, they
display exceedingly complex and plastic spatial configurations, which reveal
genetically conditioned developmental events, and give support to brain
functions. Certainly, the possibility to tackle in quantitative terms some of
the architectonic intricacies of the brain is appealing. Second-order meth-
ods, however, are theoretically complex and have not been fully developed
for 3D distributions. Not surprisingly, they have not been applied yet to the
nervous system. The interested reader may enjoy getting updated on this
promising field by consulting Mayhew (1999), Mattfeldt and Stoyan (2000),
and Krasnoperov and Stoyan (2004).
The stereological estimators of geometrical parameters are fairly well de-
veloped, and their unbiasedness is guaranteed when sampling is correct.
However, the estimators proposed to predict variance and error are only
approximations, good enough under some conditions, but still under devel-
opment in others (Cruz-Orive, 1999). Further refinements are foreseeable.
Finally, the very existence of powerful and unbiased methods to count and
measure is renewing the interest in investigating and controlling the arti-
facts derived from histological processing and observation. The possibility
of making direct and accurate quantitative inferences from processed tissue
to in vivo conditions will then no longer be a bridge too far.
APPENDIX: A CASE STUDY
A single deposit of a nonfluorescent retrograde tracer (e.g., HRP or bi-

otinylated dextran amine) has been made in a thalamic nucleus (TN) of a
rat. The main quantity of interest is the total number of retrogradely labeled
neurons in the neocortex (Cx). As additional measures of convergence, the
volume of and the neuron number in the region receiving the tracer will
also be estimated.
A. Design and Equipment
A single or separate brain blocks containing the entire extent of TN and

Cx are serially cut at 50 µm in the freezing microtome. Alternate sections
are processed for, at least, revealing the tracer and Nissl staining. Eventually,
additional series are collected for other staining purposes. This, however, is
contingent upon the spatial extension of the labeled targets in TN and Cx
along the cutting direction. A minimum of seven sections (and not more
than 10–12) containing the target(s) is required.
The material is well suited to apply the Cavalieri estimator (to estimate
the volume of tissue affected by the injection) and the optical fractionator
(to estimate total neuronal number). The former may be applied without
any special equipment: images of the injection site may be projected (on
paper, with a table projector, or a camera lucida; or on a video/computer
screen, with a video camera coupled to the microscope) or photographed.
The test system consists of a quadratic point grid drawn either on paper or
on a transparent film, depending on whether the image is to be projected
on the grid, or the grid is to be laid over the image.
The optical fractionator requires, at the very least, a high numerical aper-
ture, high-power (60×–100×) immersion lens, and a high-resolution micro-
cator (see section “Deformation Artifacts”). Manual control of the X–Y stage
displacements can also be greatly helped by digital encoders (e.g., MD2 or
MD3 Microscope Digitizers, AccuStage, Shoreview, MN) at a moderate cost.
It is very recommendable, however, to use an integral stereological system
(see section “Morphometry and Stereology”; Footnote 1). The test system
consists of an unbiased sampling frame (Fig. 16.5).
B. Procedure
A quick glance of the material shows that the injection site spans less than
1 mm along the direction of the block advance, while labeled neurons are
found along an 8 mm long stretch of Cx. Therefore, sampling intensities
should vary for TN and Cx: every second section will be used for volume and
number estimation in TN, and every 20th will be used for number estimation
in Cx.
1. Total Number of Labeled Neurons in Cx
The cell nucleus is chosen as the counting unit. The first section that will
be used for counting is randomly chosen among those reacted for the tracer
within the first 1000 µm of the Cx block. Subsequent sections will be picked
orderly at 1000 µm intervals. A pilot evaluation of the area sampling intensity
may then be performed on a section containing roughly an intermediate
number of labeled neurons, compared to those sections holding the largest
numbers and the fewest.
a. Pilot Study
At low power (using a 2.5× or 4× dry lens), the Cx area that includes
labeled cells is quickly outlined, taking care not to leave outside any labeled
cell, but sparing obviously unlabeled regions. A fraction of that area is then
sampled by systematically displacing the stage at a fixed X, Y distance, using
a 60× or 100× oil-immersion lens. A sampling frame that leaves sufficient

guard space around it (at least equal to the largest nuclear diameter) is
laid over the center of every field sampled. Expecting at least 55% vertical
shrinkage, and therefore a final mean section thickness of around 22 µm,
the disector height is set at 14 µm, so that 3 µm is left for the upper and
the rest for the lower guard areas. All labeled neurons whose nucleus’ equa-
torial plane comes into focus within the permitted frame boundaries are
counted, and readings of section thickness are taken at least every second
field. Once finished, we have to focus on the following items, in order to
tailor adequately the design for the full study:
1. Ideally, Q − −
(pilot) ≈ 25 so that Q (full) ≈ 200 (given that about eight sec-
tions will be sampled). If notably fewer or more neurons are counted,
sampling intensity should be either increased or decreased, by modi-
fying the X–Y sampling intervals or the size of the sampling frame, or
both.
2. A certain variability of tissue thickness is expected both within and be-
tween sections. If moderate (namely, a relative variability CV < 15%),
the frequency of thickness reading may be reduced (say, to one every
4–5 fields), and the contribution of the mean thickness of each sec-
tion to the final mean thickness is weighted according to Eq. (16.1)
(section “The Annoying t”). Large variations in thickness very likely
point to nonoptimal quality of the sections, making it mandatory to
measure thickness at every sampled field and weigh individually each
measurement according to Dorph-Petersen et al. (2001).
3. For practical purposes, it is recommendable that the average number
of cell counts per sampling field be in the range 0.5–5. Lower mean
densities are unavoidable, however, when few neurons are labeled, or
when the distribution of the labeled population is very uneven. Higher
densities should be avoided by decreasing the frame size.
b. Full Study
Once the final sampling design is established, it is applied to all systemati-

cally sampled sections. The most complete stereological systems incorporate
all data into datasheets, from which all needed computations are easily per-
formed. If no such systems are available, data have to be fed manually to a
custom-made spreadsheet. An example of an ad hoc organized datasheet is
shown in Table 16.2. Procedures to estimate precision by computing the CE
of the estimates are also included in the table.
2. Volume of TN Affected by the Injected Tracer
The determination of the “EIS,” that is, the territory from which the
tracer has been taken up by neurons, which eventually are recognized as
Table 16.2. Example of datasheet organization with itemized data
Volume of
injection site Neuron number in
Labeled neurons in Cx in TN injected spot in TN
Case Section
no. no. Q−
(section) Mean tsection Pi(section) Q−
(section) Mean tsection
1
2
3
..
..
n
Q−(Cx) = Mean t Q − V = Q−(TN) = Mean t Q −
A= A= A=4
B= B= B=
C = C = C =
Nug(Q − ) = Nug(V ) = Nug(Q − ) ==
Var(N)SRS ≈ Var(V )SRS ≈ Var(N)SRS ≈
CE(Q − ) ≈ CE(V ) ≈ CE(Q − ) ≈
estN = estV = estN=
For number estimates,
A = Q i− Q i− (i = 1, . . . , n)
B = Q i− Q i+1
−
C = Q i− Q i+2
−
For volume estimates, substitute point counts ( Pi ) for Q i−
Nug(Q − ) = Q −
Nug(V ) = 0.0724(ba −1/2 )(n Pi )−1/2
where b is the mean perimeter length of injected spot profiles and a is the mean cross-sectional
area of injected spot.
[For approximately spheroidal injections, the “shape” factor (ba−1/2 ) is around 4.5.]
VarSRS ≈ α[3(A − Nug) + C − 4B]
The value of α ranges between 1/240 and 1/12, depending on whether the values (point
counts, cell numbers) are uniform or change smoothly over consecutive sections, or display
marked irregularities (cf. Garcı́a-Fiñana and Cruz-Orive, 2000).
CE(Q − ) ≈ [(Nug + Var SRS)]1/2 ( Q − )−1
Mean t Q − : compute using Eq. (16.1); see section “The annoying t.”
retrogradely labeled, is a complex problem. The EIS varies for different trac-
ers and survival times (Ahlsen, 1981; Conde, 1987; Gerfen and Sawchenko,
1984; Mesulam, 1982; Shook et al., 1984) and may not correspond to the
tracer deposit, which is apparent on microscopic examination, or “virtual
injection site” (VIS; Mesulam, 1982). It is therefore convenient to indi-
cate the criteria used to define the injection site. A commonly accepted
criterion is the presence of clear-cut, uniform deposits of reaction product

in the neuropil and cell bodies, but this can be extended to the surround-
ing territory displaying an overall heavy staining of cell bodies within a more
weakly and irregularly labeled neuropil.
The cross-sectional area of the VIS is determined by point counting on
consecutive sections (when deposits are small) or a systematic sample of
sections. In general, between 5 and 10 sections should suffice to yield a
precise volume estimate applying the Cavalieri estimator (Table 16.2).
3. Number of TN Neurons Present in the Injected Area
Small tracer deposits usually allow distinguishing cell bodies within it.
If this is the case, their number may be estimated by applying the optical
fractionator to the same sections used to determine the injection size. How-
ever, regardless of the injection size, several problems may hamper such an
approach: (1) heavily stained cell bodies (and neuropil) usually produce no-
ticeable, and often severe, overlapping artifacts in thick sections; (2) heavy
overall labeling may also make it hard to distinguish between glial cells and
small neurons; and (3) tissue damage often occurs in the injection core,
leading to a complete local loss of signal.
An alternative strategy consists in estimating the number of neurons in
a mirror “VIS” of the contralateral hemisphere, generated by overlaying
mirror images of the cross-sectional profiles of the injection site on roughly
symmetrical spots of the contralateral TN. While natural side differences,
or asymmetries due to oblique sectioning may result in a tolerable degree of
inaccuracy, notable tissue deformations due to local hemorrhage or tracer
toxicity may, however, rule out this strategy.
Once the entire VIS is defined, the total neuron number enclosed in it is
estimated by the optical fractionator method, following a similar procedure
to that shown above (section “Total Number of Labeled Neurons in Cx”;
Table 16.2).
Acknowledgments. I thank Prof. Luis Cruz-Orive for reading the final

draft of this chapter and contributing valuable suggestions. This work was
supported in part by Project IST-2001-34892 (ROSANA) from the E.U.
REFERENCES
Adams, D. C., Slice, D. E., and Rohlf, F. J., 2004, Geometric morphometrics: ten years of progress
following the ‘revolution’, Ital. J. Zool. 71:5–16.
Ahlsen, G., 1981, Retrograde labelling of retinogeniculate neurones in the cat by HRP uptake
from the diffuse injection zone, Brain Res. 223:374–380.
Andersen, B. B., and Pakkenberg, B., 2003, Stereological quantitation in cerebella from people
with schizophrenia, Br. J. Psychiatry 182:354–361.
Ascoli, G. A., 2002, Neuroanatomical algorithms for dendritic modelling, Network 13:247–260.
Avendaño, C., and Dykes, R. W., 1996, Quantitative analysis of anatomical changes in the
cuneate nucleus following forelimb denervation: a stereological morphometric study in
adult cats, J. Comp. Neurol. 370:491–500.
Avendaño, C., Machin, R., Bermejo, P. E., and Logares, A., 2005, Neuron numbers in the sensory
trigeminal nuclei of the rat: A GABA- and glycine-immunocytochemical and stereological
analysis, J. Comp. Neurol. 493:538–553.
Baddeley, A. J., Gundersen, H. J. G., and Cruz-Orive, L. M., 1986, Estimation of surface area
from vertical sections, J. Microsc. 142:259–276.
Batra, S., König, M. F., and Cruz-Orive, L. M., 1995, Unbiased estimation of capillary length
from vertical slices, J. Microsc. 178:152–159.
Bermejo, P. E., Jiménez, C. E., Torres, C. V., and Avendaño, C., 2003, Quantitative stereological
evaluation of the gracile and cuneate nuclei and their projection neurons in the rat, J.
Comp. Neurol. 463:419–433.
Blasco, B., Avendaño, C., and Cavada, C., 1999, A stereological analysis of the lateral geniculate
nucleus in adult Macaca nemestrina monkeys, Vis. Neurosci. 16:933–941.
Braendgaard, H., and Gundersen, H. J. G., 1986, The impact of recent stereological advances
on quantitative studies of the nervous system, J. Neurosci. Methods 18:39–78.
Braitenberg, V., and Schüz, A., 1998, Cortex: Statistics and Geometry of Neuronal Connectivity, Berlin:
Springer-Verlag.
Broser, P. J., Schulte, R., Lang, S., Roth, A., Helmchen, F., Waters, J., Sakmann, B., and Wittum,
G., 2004, Nonlinear anisotropic diffusion filtering of three-dimensional image data from
two-photon microscopy, J. Biomed. Opt. 9:1253–1264.
Cajal, S. R. 1904, Textura del Sistema Nervioso del Hombre y los Vertebrados, Madrid: N. Moya.
Calhoun, M. E., Jucker, M., Martin, L. J., Thinakaran, G., Price, D. L., and Mouton, P. R., 1996,
Comparative evaluation of synaptophysin-based methods for quantification of synapses, J.
Neurocytol. 25:821–828.
Calhoun, M. E., Mao, Y., Roberts, J. A., and Rapp, P. R., 2004, Reduction in hippocampal
cholinergic innervation is unrelated to recognition memory impairment in aged rhesus
monkeys, J. Comp. Neurol. 475:238–246.
Calhoun, M. E., and Mouton, P. R., 2001, Length measurement: new developments in neu-
rostereology and 3D imagery: erratum, J. Chem. Neuroanat. 21:257–265.
Calverley, R. K., Bedi, K. S., and Jones, D. G., 1988, Estimation of the numerical density of
synapses in rat neocortex. Comparison of the “disector” with an “unfolding” method, J.
Chédotal, A., Umbriaco, D., Descarries, L., Hartman, B. K., and Hamel, E., 1994, Light and
electron microscopic immunocytochemical analysis of the neurovascular relationships of
choline acetyltransferase and vasoactive intestinal polypeptide nerve terminals in the rat
cerebral cortex, J. Comp. Neurol. 343:57–71.
Chklovskii, D. B., Schikorski, T., and Stevens, C. F., 2002, Wiring optimization in cortical circuits,
Neuron 34:341–347.
Coggeshall, R. E., 1999, Assaying structural changes after nerve damage, an essay on quantitative
morphology, Pain (Suppl. 6):S21–S25.
Colonnier, M., and Beaulieu, C., 1985, An empirical assessment of stereological formulae ap-
plied to the counting of synaptic disks in the cerebral cortex, J. Comp. Neurol. 231:175–179.
Conde, F., 1987, Further studies on the use of the fluorescent tracers fast blue and diamidino
yellow: effective uptake area and cellular storage sites, J. Neurosci. Methods 21:31–43.
Cruz-Orive, L. M., 1989, Second-order stereology: estimation of second moment volume mea-
sures, Acta Stereol. 8:641–646.
Cruz-Orive, L. M., 1990, Observation artifacts in stereology, Notes for the Stereology Course of the
ISS at GB-Martin Mere (Unpublished).
Cruz-Orive, L. M., 1994, Toward a more objective biology, Neurobiol. Aging 15:377–378.
Cruz-Orive, L. M., 1999, Precision of Cavalieri sections and slices with local errors, J. Microsc.
193:182–198.
Cruz-Orive, L. M., 2002a, Land mensuration without pontification: a tale of Ancient Greece, Unpub-
lished Work.
Cruz-Orive, L. M., 2002b, Stereology: meeting point of integral geometry, probability, and
statistics. In memory of Professor Luis A. Santaló (1911–2001), Math. Notae 41:49–98.
Cruz-Orive, L. M., 2004, Precision of the fractionator from Cavalieri designs, J. Microsc. 213:205–
211.
Cruz-Orive, L. M., and Howard, C. V., 1991, Estimating the length of a bounded curve in three
dimensions using total vertical projections, J. Microsc. 163:101–113.
Cruz-Orive, L. M., Insausti, A. M., Insausti, R., and Crespo, D., 2004, A case study from neuro-
science involving stereology and multivariate analysis, In: Evans, S. M., Janson, A. M., and
Nyengaard, J. R. (eds.), Quantitative Methods in Neuroscience, Oxford, UK: Oxford University
Press.
DeFelipe, J., Marco, P., Busturia, I., and Merchán-Pérez, A., 1999, Estimation of the num-
ber of synapses in the cerebral cortex: methodological considerations, Cereb. Cortex 9:
722–732.
De Groot, D. M. G., 1988, Comparison of methods for the estimation of the thickness of
ultrathin tissue sections, J. Microsc. 151:23–42.
De Groot, D. M., and Bierman, E. P., 1983, The complex-shaped “perforated” synapse, a prob-
lem in quantitative stereology of the brain, J. Microsc. 131:355–360.
Dempster, W. T., 1943, Paraffin compression due to the rotary microtome, Stain Technol. 18:13–
26.
Descarries, L., Séguéla, P., and Watkins, K. C., 1991, Nonjunctional relationships of monoamine
axon terminals in the cerebral cortex of adult rat, In: Fuxe, K., and Agnati, L. F. (eds.),
Volume Transmission in the Brain: Novel Mechanisms for Neural Transmission, New York: Raven
Press, pp. 53–62.
Donovan, S. L., Mamounas, L. A., Andrews, A. M., Blue, M. E., and McCasland, J. S., 2002,
GAP-43 is critical for normal development of the serotonergic innervation in forebrain, J.
Neurosci. 22:3543–3552.
Dorph-Petersen, K. A., Nyengaard, J. R., and Gundersen, H. J. G., 2001, Tissue shrinkage and
unbiased stereological estimation of particle number and size, J. Microsc. 204:232–246.
Ellis, R. C., 2000, The microtome: function and design, http://home.primus.com.au/royellis/
microt/microt.htm, 1–13 (+15 Figures and 2 Tables).
Evans, S. M., Janson, A. M., and Nyengaard, J. R., 2004, Quantitative Methods in Neuroscience,
Oxford, UK: Oxford University Press.
Fiala, J. C., 2005, Reconstruct: a free editor for serial section microscopy, J. Microsc. 218: 52–61.
Fiala, J. C., and Harris, K. M., 2001, Extending unbiased stereology of brain ultrastructure to
three-dimensional volumes, J. Am. Med. Inf. Assoc. 8:1–16.
Garcı́a-Fiñana, M., and Cruz-Orive, L. M., 2000, Fractional trend of the variance in Cavalieri
sampling, Image Anal. Stereol. 19:71–79.
Gardella, D., Hatton, W. J., Rind, H. B., Glenn, G. D., and Von Bartheld, C. S., 2003, Differential
tissue shrinkage and compression in the z-axis: implications for optical disector counting
in vibratome-, plastic- and cryosections, J. Neurosci. Methods 124:45–59.
Geinisman, Y., Gundersen, H. J. G., Van der Zee, E., and West, M. J., 1996, Unbiased stereo-
logical estimation of the total number of synapses in a brain region, J. Neurocytol. 25:805–
819.
Gerrits, P. O., Horobin, R. W., and Stokroos, I., 1992, The effects of glycol methacrylate
as a dehydrating agent on the dimensional changes of liver tissue, J. Microsc. 165:273–
280.
Gerrits, P. O., van Leeuwen, M. B. M., Boon, M. E., and Kok, L. P., 1987, Floating on a water bath
and mounting glycol methacrylate and hydroxypropyl methacrylate sections influence
final dimensions, J. Microsc. 145:107–113.
Glaser, J. R., and Glaser, E. M., 2000, Stereology, morphometry, and mapping: the whole is
greater than the sum of its parts, J. Chem. Neuroanat. 20:115–126.
Glatzer, N. R., Hasney, C. P., Bhaskaran, M. D., and Smith, B. N., 2003, Synaptic and mor-
phologic properties in vitro of premotor rat nucleus tractus solitarius neurons labeled
transneuronally from the stomach, J. Comp. Neurol. 464:525–539.
Gokhale, A. M., 1990, Unbiased estimation of curve length in 3D using vertical slices, J. Microsc.
159:133–141.
Goldstein, S. S., and Rall, W., 1974, Changes of action potential shape and velocity for changing
core conductor geometry, Biophys. J. 14:731–757.
Guillery, R. W., 2002, On counting and counting errors, J. Comp. Neurol. 447:1–7.
Guillery, R. W., and Herrup, K., 1997, Quantification without pontification: choosing a method
for counting objects in sectioned tissues, J. Comp. Neurol. 386:2–7.
Gundersen, H. J. G., 1977, Notes on the estimation of the numerical density of arbitrary profiles.
The edge effect, J. Microsc. 111:219–222.
Gundersen, H. J., 1986, Stereology of arbitrary particles. A review of unbiased number and size
estimators and the presentation of some new ones, in memory of William R. Thompson,
J. Microsc. 143:3–45.
Gundersen, H. J., 2002a, Stereological estimation of tubular length, J. Microsc. 207:155–
160.
Gundersen, H. J. G., 2002b, The smooth fractionator, J. Microsc. 207:191–210.
Gundersen, H. J. G., Bagger, P., Bendtsen, T. F., Evans, S. M., Korbo, L., Marcussen, N.,
Møller, A., Nielsen, K., Nyengaard, J. R., Pakkenberg, B., Sørensen, F. B., Vesterby, A.,
and West, M. J., 1988, The new stereological tools: disector, fractionator, nucleator and
point sampled intercepts and their use in pathological research and diagnosis, APMIS 96:
857–881.
Gundersen, H. J. G., and Jensen, E. B., 1987, The efficiency of systematic sampling in stereology
and its prediction, J. Microsc. 147:229–263.
Gundersen, H. J. G., Jensen, E. B. V., Kiêu, K., and Nielsen, J., 1999, The efficiency of systematic
sampling in stereology—reconsidered, J. Microsc. 193:199–211.
Haug, H., Kuhl, S., Mecke, E., Sass, N. L., and Wasner, K., 1984, The significance of morpho-
metric procedures in the investigation of age changes in cytoarchitectonic structures of
human brain, J. Hirnforsch. 25:353–374.
Hayat, M. A., 1981, Fixation for Electron Microscopy, New York: Academic Press.
Hayat, M. A., 2000, Principles and Techniques of Electron Microscopy. Biological Applications, Cam-
bridge: Cambridge University Press.
Heimer, L., and Zaborszky, L., 1989, Neuroanatomical Tract-Tracing Methods 2. Recent Progress, New
York: Plenum Press.
Helmchen, F., 1999, Dendrites as biochemical compartments, In: Stuart, G., Spruston, N., and
Häusser, M. (eds.), Dendrites, Oxford: Oxford University Press, pp. 161–192.
Hillman, D. E., 1979, Neuronal shape parameters and substructures as a basis of neuronal
form, In: Schmitt, F. O., and Worden, F. G. (eds.), The Neurosciences: Fourth Study Program,
Cambridge, MA: MIT Press, pp. 477–498.
Horcholle-Bossavit, G., Gogan, P., Ivanov, Y., Korogod, S., and Tyc-Dumont, S., 2000, The prob-
lem of the morphological noise in reconstructed dendritic arborizations, J. Neurosci. Meth-
ods 95:83–93.
Howard, C. V., Cruz-Orive, L. M., and Yaegashi, H., 1992, Estimating neuron dendritic length
in 3D from total vertical projections and from vertical slices, Acta Neurol. Scand. Suppl.
137:14–19.
Howard, C. V., and Reed, M. G., 2005, Unbiased Stereology. 2nd ed., Oxford: BIOS Scientific
Publishers.
Howell, K., Hopkins, N., and Mcloughlin, P., 2002, Combined confocal microscopy and stere-
ology: a highly efficient and unbiased approach to quantitative structural measurement
in tissues, Exp. Physiol. 87:747–756.
Jelinek, H. F., and Fernandez, E., 1998, Neurons and fractals: how reliable and useful are
calculations of fractal dimensions? J. Neurosci. Methods 81:9–18.
Jones, T. A., 1999, Multiple synapse formation in the motor cortex opposite unilateral sensori-
motor cortex lesions in adult rats, J. Comp. Neurol. 414:57–66.
Jones, T. A., Klintsova, A. Y., Kilman, V. L., Sirevaag, A. M., and Greenough, W. T., 1997,
Induction of multiple synapses by experience in the visual cortex of adult rats, Neurobiol.
Learn. Mem. 68:13–20.
Kleim, J. A., Vij, K., Ballard, D. H., and Greenough, W. T., 1997, Learning-dependent synaptic
modifications in the cerebellar cortex of the adult rat persist for at least four weeks, J.
Neurosci. 17:717–721.
Korkmaz, A., and Tümkaya, L., 1997, Estimation of the section thickness and optical disector
height with a simple calibration method, J. Microsc. 187:104–109.
Korkotian, E., and Segal, M., 2000, Structure–function relations in dendritic spines: is size
important? Hippocampus 10:587–595.
Krasnoperov, R. A., and Stoyan, D., 2004, Second-order stereology of spatial fibre systems, J.
Microsc. 216:156–164.
Kubı́nova, L., Janacek, J., Karen, P., Radochova, B., Difato, F., and Krekule, I., 2004, Confocal
stereology and image analysis: methods for estimating geometrical characteristics of cells
and tissues from three-dimensional confocal images, Physiol. Res. 53:S47–S55.
Lagares, A., and Avendaño, C., 1999, An efficient method to estimate cell number and volume
in multiple dorsal root ganglia, Acta Stereol. 18:185–195.
Larsen, J. O., 1998, Stereology of nerve cross sections, J. Neurosci. Methods 85:107–118.
Larsen, J. O., Gundersen, H. J., and Nielsen, J., 1998a, Global spatial sampling with isotropic
virtual planes: estimators of length density and total length in thick, arbitrarily orientated
sections, J. Microsc. 191:238–248.
Larsen, J. O., Thomsen, M., Haugland, M., and Sinkjaer, T., 1998b, Degeneration and regenera-
tion in rabbit peripheral nerve with long-term nerve cuff electrode implant: a stereological
study of myelinated and unmyelinated axons, Acta Neuropathol. 96:365–378.
Larsen, J. O., Von Euler, M., and Janson, A. M., 2004, Virtual test systems for estimation of
orientation-dependent parameters in thick, arbitrarily orientated sections exemplified by
length quantification of regenerating axons in spinal cord lesions using isotropic, virtual
planes, In: Evans, S. M., Janson, A. M., and Nyengaard, J. R. (eds.), Quantitative Methods in
Neuroscience, Oxford: Oxford University Press, pp. 264–284.
Løkkegaard, A., Nyengaard, J. R., and West, M. J., 2001, Stereological estimates of number
and length of capillaries in subdivisions of the human hippocampal region, Hippocampus
11:726–740.
Machı́n-Cedrés, R., 2004, Respuesta de la corteza sensorial a cambios crónicos en la estimulación fun-
cional: Un estudio estereológico ultraestructural en la rata, Doctoral Thesis, Autonoma University
of Madrid.
Marner, L., and Pakkenberg, B., 2003, Total length of nerve fibers in prefrontal and global
white matter of chronic schizophrenics, J. Psychiatr. Res. 37:539–547.
Mattfeldt, T., Mall, G., Gharehbaghi, H., and Møller, P., 1990, Estimation of surface area and
length with the orientator, J. Microsc. 159:301–317.
Mattfeldt, T., and Stoyan, D., 2000, Improved estimation of the pair correlation function of
random sets, J. Microsc. 200:158–173.
Mayhew, T. M., 1979a, Basic stereological relationships for quantitative microscopical
anatomy—a simple systematic approach, J. Anat. 129:95–105.
Mayhew, T. M., 1979b, Stereological approach to the study of synapse morphometry with
particular regard to estimating number in a volume and on a surface, J. Neurocytol. 8:
121–138.
Mayhew, T. M., 1988, An efficient sampling scheme for estimating fibre number from nerve
cross sections: the fractionator, J. Anat. 157:127–134.
Mayhew, T. M., 1990, Efficient and unbiased sampling of nerve fibers for estimating fiber
number and size, In: Conn, P. M. (ed.), Quantitative and Qualitative Microscopy—Methods in
Neurosciences, Vol. 3, New York: Academic Press, pp. 172–187.
Mayhew, T. M., 1999, Second-order stereology and ultrastructural examination of the spatial
arrangements of tissue compartments within glomeruli of normal and diabetic kidneys, J.
Microsc. 195:87–95.
Mesulam, M.-M., 1982, Principles of horseradish peroxidase neurohistochemistry and their ap-
plications for tracing neural pathways, In: Mesulam, M.-M. (ed.), Tracing Neural Connections
with Horseradish Peroxidase, New York: John Wiley and Sons, pp. 1–151.
Mitchison, G., 1992, Axonal trees and cortical architecture, Trends Neurosci. 15:122–126.
Mouton, P. R., 2002, Principles and Practices of Unbiased Stereology: An Introduction for Bioscientists,
Baltimore, MD: Johns Hopkins University Press.
Mouton, P. R., Gokhale, A. M., Ward, N. L., and West, M. J., 2002, Stereological length estimation
using spherical probes, J. Microsc. 206:54–64.
Nadasdy, Z., and Zaborszky, L., 2001, Visualization of density relations in large-scale neural
networks, Anat. Embryol. 204:303–317.
Negredo, P., Castro, J., Lago, N., Navarro, X., and Avendaño, C., 2004, Differential growth of
axons from sensory and motor neurons through a regenerative electrode: a stereological,
retrograde tracer, and functional study in the rat, Neuroscience 128:605–615.
Nyengaard, J. R., and Gundersen, H. J. G., 1992, The isector: a simple and direct method for
generating isotropic, uniform random sections from small specimens, J. Microsc. 165:427–
431.
Nyengaard, J. R., Witgen, B. M., Grady, S. M., Anderson, B. B., and Gunderson, H. J. G.,
2005, Estimation of synapse number using electron microscopy and a new fractionator
principle with varying sampling fractions, Abstract/Viewer Intinerary planner, Soc. Neurosci.
Prog. no. 454.14.
O’Malley, A., O’Connell, C., Murphy, K. J., and Regan, C. M., 2000, Transient spine density in-
creases in the mid-molecular layer of hippocampal dentate gyrus accompany consolidation
of a spatial learning task in the rodent, Neuroscience 99:229–232.
Pakkenberg, B., and Gundersen, H. J., 1988, Total number of neurons and glial cells in human
brain nuclei estimated by the disector and the fractionator, J. Microsc. 150:1–20.
Pakkenberg, B., and Gundersen, H. J. G., 1997, Neocortical neuron number in humans: effect
of sex and age, J. Comp. Neurol. 384:312–320.
Partadiredja, G., Miller, R., and Oorschot, D. E., 2003, The number, size, and type of axons in
rat subcortical white matter on left and right sides: a stereological, ultrastructural study, J.
Neurocytol. 32:1165–1179.
Peterson, D. A., 2004, The use of fluorescent probes in cell-counting procedures, In: Evans,
S. M., Janson, A. M., and Nyengaard, J. R. (eds.), Quantitative Methods in Neuroscience, New
York: Oxford University Press, pp. 85–114.
Peterson, D. A., and Jones, D. G., 1993, Determination of neuronal number and process sur-
face area in organotypic cultures: a stereological approach, J. Neurosci. Methods 46:107–
120.
Poirazi, P., and Mel, B. W., 2001, Impact of active dendrites and structural plasticity on the
memory capacity of neural tissue, Neuron 29:779–796.
Roberts, N., Puddephat, M. J., and McNulty, V., 2000, The benefit of stereology for quantitative
radiology, Br. J. Radiol. 73:679–697.
Rønn, L. C., Ralets, I., Hartz, B. P., Bech, M., Berezin, A., Berezin, V., Møller, A., and Bock, E.,
2000, A simple procedure for quantification of neurite outgrowth based on stereological
principles, J. Neurosci. Methods 100:25–32.
Rusakov, D. A., Harrison, E., and Stewart, M. G., 1998, Synapses in hippocampus occupy only
1–2% of cell membranes and are spaced less than half-micron apart: a quantitative ultra-
structural analysis with discussion of physiological implications, Neuropharmacology 37:513–
521.
Rusakov, D. A., and Stewart, M. G., 1995, Quantification of dendritic spine populations using
image analysis and a tilting disector, J. Neurosci. Methods 60:11–21.
Saper, C. B., 1996, Any way you cut it: a new journal policy for the use of unbiased counting
methods, J. Comp. Neurol. 364:5.
Schaefer, A. T., Larkum, M. E., Sakmann, B., and Roth, A., 2003, Coincidence detection in
pyramidal neurons is tuned by their dendritic branching pattern, J. Neurophysiol. 89:3143–
3154.
Schiønning, J. D., Larsen, J. O., Tandrup, T., and Braendgaard, H., 1998, Selective degener-
ation of dorsal root ganglia and dorsal nerve roots in methyl mercury-intoxicated rats: a
stereological study, Acta Neuropathol. (Berl.) 96:191–201.
oratories, J. Comp. Neurol. 473:177–193.
Shook, B. L., Abramson, B. P., and Chalupa, L. M., 1984, An analysis of the transport of WGA-
HRP in the cat’s visual system, J. Neurosci. Methods 11:65–77.
Silver, M. A., and Stryker, M. P., 1999, Synaptic density in geniculocortical afferents remains
constant after monocular deprivation in the cat, J. Neurosci. 19:10829–10842.
Small, J. V., 1968, Measurement of section thickness, In: 4th European Regional Conference on
Electron Microscopy, Rome, pp. 609–610 (Abstract).
Sorra, K. E., and Harris, K. M., 1993, Occurrence and three-dimensional structure of multiple
synapses between individual radiatum axons and their target pyramidal cells in hippocam-
pal area CA1, J. Neurosci. 13:3736–3748.
Spruston, N., Stuart, G., and Häusser, M., 1999, Dendritic integration, In: Stuart, G., Spruston,
N., and Häusser, M. (eds.), Dendrites, New York: Oxford University Press, pp. 231–260.
wiring, Neuron 43:251–259.
Sterio, D. C., 1984, The unbiased estimation of number and sizes of arbitrary particles using
the disector, J. Microsc. 134:127–136.
Stevens, S. S., 1951, Mathematics, measurement, and psychophysics, In: Stevens, S. S. (ed.),
Handbook of Experimental Psychology, New York: John Wiley.
Stocks, E. A., McArthur, J. C., Griffen, J. W., and Mouton, P. R., 1996, An unbiased method for
estimation of total epidermal nerve fibre length, J. Neurocytol. 25:637–644.
Tamamaki, N., Watanabe, K., and Nojyo, Y., 1984, A whole image of the hippocampal pyramidal
neuron revealed by intracellular pressure-injection of horseradish peroxidase, Brain Res.
307:336–340.
Tang, Y., and Nyengaard, J. R., 1997, A stereological method for estimating the total length and
size of myelin fibers in human brain white matter, J. Neurosci. Methods 73:193–200.
Tang, Y., Nyengaard, J. R., De-Groot, D. M. G., and Gundersen, H. J. G., 2001, Total regional
and global number of synapses in the human brain neocortex, Synapse 41:258–273.
Turner, A. M., and Greenough, W. T., 1985, Differential rearing effects on rat visual cortex
synapses: I. Synaptic and neuronal density and synapses per neuron, Brain Res. 329:195–
203.
Uylings, H. B., Ruiz-Marcos, A., and Van Pelt, J., 1986a, The metric analysis of three-dimensional
dendritic tree patterns: a methodological review, J. Neurosci. Methods 18:127–151.
Uylings, H. B. M., Van Eden, C. G., and Hofman, M. A., 1986b, Morphometry of size/volume
variables and comparison of their bivariate relations in the nervous system under different
conditions, J. Neurosci. Methods 18:19–37.
Uylings, H. B., and Van Pelt, J., 2002, Measures for quantifying dendritic arborizations, Network
13:397–414.
Uylings, H. B., Van Pelt, J., and Verwer, R. W. H., 1989, Topological analysis of individual
neurons, In: Capowski, J. J. (ed.), Computer Techniques in Neuroanatomy, New York: Plenum
Press, pp. 215–239.
Valverde, F., 1970, The Golgi method. A tool for comparative structural analyses, In: Nauta, W.
J. H., and Ebbeson, S. O. E. (eds.), Contemporary Research Methods in Neuroanatomy, Berlin:
Springer, pp. 12–31.
Van Pelt, J., Van Ooyen, A., and Uylings, H. B. M., 2001, The need for integrating neuronal
morphology databases and computational environments in exploring neuronal structure
and function, Anat. Embryol. 204:255–265.
Verwer, R. W. H., Van Pelt, J., and Uylings, H. B. M., 1992, An introduction to topological analysis
of neurones, In: Stewart, M. G. (ed.), Quantitative Methods in Neuroanatomy, Chichester,
England: John Wiley & Sons, pp. 295–323.
Vinkenoog, M., van den Oever, M. C., Uylings, H. B., and Wouterlood, F. G., 2005, Random
or selective neuroanatomical connectivity. Study of the distribution of fibers over two
populations of identified interneurons in cerebral cortex, Brain Res. Brain Res. Protoc.
14:67–76.
Weibel, E. R., 1979, Stereological Methods. Vol. I. Practical Methods for Biological Morphometry, New
York: Academic Press.
West, M. J., 1993, New stereological methods for counting neurons, Neurobiol. Aging 14:275–285.
West, M. J., and Slomianka, L., 1998, Total number of neurons in the layers of the human
entorhinal cortex, Hippocampus 8:69–82.
West, M. J., Slomianka, L., and Gundersen, H. J. G., 1991, Unbiased stereological estimation of
the total number of neurons in the subdivisions of the rat hippocampus using the optical
fractionator, Anat. Rec. 231:482–497.
Williams, R. W., and Rakic, P., 1988, Three-dimensional counting: an accurate and direct
method to estimate numbers of cells in sectioned material, J. Comp. Neurol. 278:344–352.
Zaborszky, L., Csordas, A., Buhl, D. L., Duque, A., Somogyi, J., and Nadasdy, Z., 2002, Com-
putational anatomical analysis of the basal forebrain corticopetal system, In: Ascoli, A.
(ed.), Computational Neuroanatomy: Principles and Methods, Totowa, NJ: Humana Press, pp.
171–197.
Zaborszky, L., Leranth, C., Makara, G. B., and Palkovits, M., 1975, Quantitative studies on the
supraoptic nucleus in the rat: II. Afferent fiber connections, Exp. Brain Res. 22:525–540.
17
Three-Dimensional
Computerized Reconstruction
from Serial Sections: Cell
Populations, Regions, and
Whole Brain
JAN G. BJAALIE and TRYGVE B. LEERGAARD
INTRODUCTION
EXAMPLE DATA
PREPARATORY STEPS: TISSUE BLOCKING AND HISTOLOGY
Tissue Block Size
Tissue Block Imaging
Orientation of Section Plane
Fiducials and Block-Face Imaging
Estimates of Shrinkage in Overall Tissue and in the Section Plane
DATA ACQUISITION: USER-GUIDED AND AUTOMATIC PROCEDURES
Image-Combining Computerized Microscopy
Image Analysis
Data Acquisition Exemplified
ALIGNMENT OF DATA FROM SECTIONS: BUILDING THE
RECONSTRUCTION
Fiducials and Block-Face Images
Landmarks and Boundaries
Template Sections, Atlases, and Tomography Data
Example Alignment
COORDINATE SYSTEMS
VISUALIZATION AND ANALYSES
Primary Viewing of the Reconstruction
JAN G. BJAALIE AND TRYGVE B. LEERGAARD • Neural Systems and Graphics

Computing Laboratory, Centre for Molecular Biology and Neuroscience, and Department
of Anatomy, University of Oslo, N-0317 Oslo, Norway
530
3-D RECONSTRUCTION FROM SERIAL SECTIONS 531
Slicing
Surface Modeling of Labeling Patterns
Analysis of Spatial Overlap
Density Gradient Analysis
Stereo-Imaging
WHOLE BRAIN ATLASING
APPENDIX
Application of Local Coordinate System for the Rat Pontine Nuclei
Translation Between Local and Global Coordinates
REFERENCES
Abstract: Three-dimensional (3D) reconstruction of data collected from serial sec-

tions brings together information that has been separated by sectioning. From 3D
reconstructions, shapes of structures and complicated distributions of tissue ele-
ments can be analyzed to an extent not possible by inspection of individual sections.
This chapter reviews preparatory steps that are needed to produce high-quality se-
ries of sections, procedures that are typically used to acquire data from the sections,
alignment of the section data, use of global and local coordinate systems for com-
parison of data across experiments, and useful tools (visualization and others) for
extracting information from 3D reconstructions. The examples shown are based on
axonal tracing data from the rat cerebro–cerebellar system, made available via The
Rodent Brain Workbench (www.rbwb.org).
Keywords: brain atlas, coordinate system, histology, section alignment, three dimen-
sional, visualization
I. INTRODUCTION
The brain consists of numerous areas and regions that are intermin-
gled and interconnected in a complex fashion. Serial sectioning followed
by the use of various microscopic techniques is frequently used to study
shapes, dimensions, and spatial distributions of objects at multiple levels of
granularity. At the light microscopic level, the objects studied range from
parts of cells, such as dendritic or axonal elements, via cell populations to
regions, areas, and whole brains. A problem with the sectioning, however,
is that information that is inherently three dimensional (3D) is separated.
Furthermore, the orientation of the section plane influences the visualiza-
tion, analysis, and interpretations of the data collected from the sections.
Since many neurobiological problems rely on accurate mapping of infor-
mation in space, 3D reconstruction of selected features from serial sections
will often represent an advantage for data presentation, as well as for the
implementation of analysis that will reveal essential features in the material.
It is therefore expected that the use of 3D reconstruction will lead to a more
complete understanding or spatial organization, influencing the transition
from data to knowledge.
532 JAN G. BJAALIE and TRYGVE B. LEERGAARD
Computerized 3D reconstruction of data from serial sections has been em-

ployed for more than three decades (for references, see Bjaalie, 1991). In
this method, data acquired from a stack of consecutive sections are brought
together in registration. By use of surface- or volume-based rendering, fea-
tures present in the 3D reconstruction can be visualized in a much more
complete fashion than made possible by inspection of single sections only.
An additional advantage posed by the 3D format is the opportunity, relying
on well-designed software, to perform manipulations and analyses that could
not have been performed at the single section level. Important examples
include re-slicing at any chosen angle, which may reveal spatial relationships
that are difficult to detect in the original plane of sectioning, and density
measurements, used to identify important aspects of spatial distribution pat-
terns. Many categories of quantitative analyses may be performed directly
on single sections. But with 3D reconstruction from series of sections, such
analyses tend to be more complete and accurate. Finally, the use of comput-
erized 3D reconstruction offers opportunities for efficient comparisons of
data from different experimental animals of the same species. The 3D digital
data, if assembled carefully and according to standardized procedures, may
be fitted to common coordinate systems and reused in novel combinations.
In this chapter, we will deal with 3D reconstruction of geometric repre-
sentations, essentially point and line coded data. In our context, point data
represent distributions of defined populations of cell bodies or axonal ter-
minal fields. Line data may represent boundaries of territories identified
by a specific label, the boundaries of areas or nuclei, or the pial or ven-
tricular surfaces of the brain. Volume rendering, based on data present in
a full 3D array of volume elements (voxels), is not dealt with (for consid-
erations of such methods, see, e.g., Pommert et al., 2002; Senft, 2002). We
will describe some of the basic requirements for 3D reconstruction at the
level of the biological material, before briefly discussing data acquisition
procedures, and alignment of the acquired point and line coded data into
a 3D reconstruction. Further, we will outline principle steps of visualization
and analysis, using examples based on software that is generally available for
the neuroscience community. Our experiences related to the use of local
and global coordinate systems will be integrated in the descriptions. The
examples given and illustrations shown are taken from system level investi-
gations of neuronal populations and brain regions, more specifically from
our investigations of the rat cerebro–cerebellar system.
II. EXAMPLE DATA
To illustrate and exemplify the different stages involved in 3D reconstruc-

tion of data collected from serial sections, with subsequent visualization and
analysis, we have mostly used published data that are available from the
database application Functional Anatomy of the Cerebro–Cerebellar System in Rat
(FACCS), which is a part of the Rodent Brain Workbench (www.rbwb.org; see
also Bjaalie et al., 2005). The cerebro–cerebellar system is monosynaptically
interrupted in the pontine nuclei, a major brain stem nuclear complex. The
system originates in large parts of the cerebral cortex, reaches most parts
of the cerebellum, and is one of the largest projection systems in the brain
(Brodal, 1982; Brodal and Bjaalie, 1992; Ruigrok, 2004). The data available
in FACCS are axonal tracing data. Axonal tracers were injected in the cere-
bral cortex and/or cerebellum and the distribution of the ensuing labeling
in the pontine nuclei (anterogradely labeled cerebro–pontine axonal termi-
nal fields, and retrogradely labeled ponto–cerebellar neuronal cell bodies)
was mapped in 3D. The focus of the present examples and illustrations is on
the cerebro–pontine projection, with data primarily from Leergaard et al.
(2000a, b, 2004). The overall aims of these studies were to discover principles
of topographical organization and overall structural design of the cerebro–
ponto–cerebellar system, including the transformations of the maps from
one part of the system to the next. At a more specific level, the aims were to
elucidate organizations reminiscent of developmental principles, and orga-
nizations that may influence the nature of the information transfer in the
system (for review, see Leergaard, 2003).
III. PREPARATORY STEPS: TISSUE BLOCKING

AND HISTOLOGY
Brain material that is sectioned and submitted to 3D reconstruction may

be produced under a variety of conditions and by use of diverse experi-
mental procedures. The procedures include, e.g., perturbations imposed
on the experimental animals, surgical manipulation with introduction of
dyes or tracers, tissue fixation and extraction, and cutting of sections and
application of histological procedures. In the context of 3D reconstruction,
it is particularly important to produce a material of serial sections that is as
complete as possible and without distortions in the individual sections. The
quality of the sections relies on multiple factors, most of which are specific to
the sort of tissue and type of investigation, and should be optimized within
given limits. Considerations pertaining to the use of chemical tissue fixation,
embedding, choice of microtome and section thickness, section sampling,
and subsequent histological processing are well described elsewhere (see,
e.g., Maunsbach and Afzelius, 1999; Romeis, 1968; Woods and Ellis, 1994).
In the experimental material illustrated in Figs. 17.1–17.9 (data from Brevik
et al., 2001; Leergaard et al., 2000a, b, 2004; Lillehaug et al., 2002), brains were
fixed by paraformaldehyde perfusion, cryoprotected with buffered sucrose,
and sectioned at 50 µm on a freezing microtome. To minimize distortions,
microtomy was performed carefully at stable temperatures. For some of
the material (from the cerebellum, see Brevik et al., 2001; Lillehaug et al.,
2002), tissue blocks were embedded in gelatin prior to sectioning. Series
of sections that were incomplete or contained damaged sections were not
used for reconstruction.
In this section, we will deal with the most fundamental and invariant
preparatory steps, relevant in the context of 3D reconstruction. These steps
include the dividing of the tissue into blocks and the subsequent handling
of the tissue blocks, orientation of plane of sectioning, use of fiducials and
block-face images, and estimates of shrinkage. General advice concerning
preparation of histological material may be found in standard textbooks on
histology and microscopic techniques (see, e.g., Romeis, 1968; Woods and
Ellis, 1994).
A. Tissue Block Size
High-quality histological material is usually easier to produce from small

tissue blocks. The smaller the tissue block, the smaller are the chances of
having distortions in the sections. The investigator should therefore con-
sider the possibility of cutting sections from several smaller tissue blocks
rather than from one large block. If different regions of the brain are stud-
ied in separate tissue blocks, relationships between regions may nevertheless
be studied. Thus, the application of local coordinate systems (outlined in
section “Coordinate Systems,” and in the Appendix) will not only facilitate
comparisons across animal analysis, but also allow transfer of the data from
the local coordinate systems onto common global coordinates.
B. Tissue Block Imaging
Prior to the sectioning, it is often useful to capture images of the tissue

from standardized angles (Fig. 17.1A, B), before and after tissue blocking or
tissue embedding. Such images may serve to extract positional information
for a series of sections, as well as to measure tissue shrinkage due to the his-
tological processing. Images of the tissue blocks may also serve to document
positional information about, e.g., axonal tracer injection sites (Fig. 17.1D)
that may be transferred to standard diagrams (Fig. 17.1E) and compared
with microscope images (Fig. 17.4B).
C. Orientation of Section Plane
For most types of investigation, exact determination of the orientation of

the section plane in relation to the whole brain and the tissue block facilitates
subsequent registration and assembly of sections into a 3D reconstruction.
The plane of sectioning is usually determined with a cut applied directly
to the tissue, defining the block face to be placed on the microtome tissue
platform, or by using a mold and an embedding medium (Paxinos and
Watson, 1998, 2005; Swanson, 1999). Several strategies and tools may be
used to ensure a defined and reproducible section plane. Section planes
are usually chosen in relation to recognizable anatomical structures (e.g.,
midline of the brain, ventral surface of the brain, or others), or to reproduce
Figure 17.1. Preparation of paraformaldehyde fixed rodent brains for microtomy.

(A) and (B) show isolated mouse brains photographed from standard dorsal (A)
and lateral (B, C) angles. Various measurements from the images are used for verifi-
cation of the obtained section angle, for assigning spatial coordinates to individual
sections, and for adjustment of shape and proportion of subsequent 3D reconstruc-
tions. (C) shows a custom-made device for flexible definition of reproducible coro-
nal/transverse section planes in rodent brains (developed by T. B. Leergaard and
K. A. Rekdahl, University of Oslo, Norway). Isolated or embedded rodent brains are
placed on a pivotable mount and oriented to align selected anatomical landmarks
to a grid (red arrows), before a razor blade is inserted in a vertical slit (dashed yellow
line) to apply a coronal cut to the brain in a guillotine fashion. The depicted mouse
brain (C) is oriented to reproduce the coronal section plane used in the stereotaxic
atlas of Paxinos and Franklin (2001). (D) shows an isolated rat brain in a view from
dorsal. A focal injection of a fluorescent neuronal tracer is visible in the parietal
cerebral cortex. (E) shows the position and extent of the tracer injection shown in
(D), mapped onto a standard diagram of the rat cerebral cortex. Line spacing in the
grids in (C) and (E) is 1 mm. (C) and (D) are from the database application FACCS
(www.rbwb.org; see also Bjaalie et al., 2005).
standardized angles (e.g., as used in published brain atlases). For example,

use of commercially available brain blockers (such as blockers for rodent
brains provided by David Kopf Instruments, Tujunga, CA) facilitates the
reproduction of section planes used in stereotaxic rat and mouse brain
atlases (Paxinos and Franklin, 2001; Paxinos and Watson, 1998, 2005). A
similar customized device is shown in Fig. 17.1C. The angle of sectioning is
typically chosen to reproduce those of previous investigations, or of standard
atlases.
D. Fiducials and Block-Face Imaging
Fiducial markers are introduced artificially in the tissue as references for

the alignment of sections. Classical fiducials are drilled holes, oriented or-
thogonally to the section plane, or notches in the surface of the tissue or the
embedding medium. The use of fiducials in the tissue itself, if placed close to
regions of interest, is not recommended, as tissue destructions imposed by
the fiducial markers may be substantial. Other approaches can be employed
to compensate for lack of fiducials during section alignment (see below). To
facilitate section alignment, some investigators capture images of the tissue
block face (of every section) during sectioning (Toga and Banerjee, 1993;
see also, Ewald et al., 2002).
E. Estimates of Shrinkage in Overall Tissue and in the Section Plane
Overall tissue shrinkage (or enlargement) relative to the in vivo brain

is usually not taken into consideration when preparing a 3D reconstruc-
tion. Shrinkage in individual sections, as a result of histological processing,
needs to be taken into consideration to ensure a correctly proportioned 3D
reconstruction. Correct spatial proportions are obtained by subsequent size
adjustment of the data derived from the sections.
IV. DATA ACQUISITION: USER-GUIDED AND

AUTOMATIC PROCEDURES
To acquire simplified geometric representations (point and line coded

data) of selected objects present in the sections, methods ranging from
completely manual to nearly fully automatic may be employed. Data need to
be recorded at sufficient resolution (using appropriate magnification) and
accuracy across regions of interest ranging from small divisions of a brain nu-
cleus (a few hundred micrometers) to entire sections (several centimeters).
A basic approach is to use camera lucida drawings that are subsequently dig-
itized (reviewed in Bjaalie, 1991). More sophisticated methods, briefly de-
scribed below, include a semiautomatic method based on image-combining
computerized microscopy (ICCM) and a more automated method combin-
ing digital camera technology with the use of image analysis software.
A. Image-Combining Computerized Microscopy
This method is currently available, e.g., in the Neurolucida system (Micro-

BrightField Inc., Williston, VT), or in customized versions (Leergaard and
Bjaalie, 1995). The principle of ICCM was first introduced by Glaser and van
der Loos (Glaser et al., 1979, 1983; Glaser and Glaser, 1990). The method is
based on the mixing of a computer graphical drawing area with the image of
the specimen. Movement of the microscope stage (controlled by stepping
motors via the computer) is accompanied by a translation of the graphical
image. The user thus obtains direct feedback during the data entry pro-
cedure. ICCM systems are usually optimized for high-resolution recording
of numerous x, y (or x, y , z) coordinates across large brain regions. The
precision of the data recording depends on the quality of the microscope
optics, the stepping motors, and the resolution of the graphic feedback im-
age (Glaser et al., 1983). Carefully calibrated customized systems based on
high-precision Märzhäuser microscope stages (Märzhäuser Wetzlar Gmbh,
Wetzlar, Germany) have been reported to be able to reproduce spatial lo-
cations with a precision of ±2 µm (Leergaard and Bjaalie, 1995). Higher
precision levels have been reported for commercially available systems.
B. Image Analysis
More automated data acquisition methods have been employed, moti-

vated by the need for procedures that are less dependent on human pattern
recognition abilities and more suitable for high-throughput analysis. Thus,
image analysis techniques have been applied to detect cytoarchitectonic
boundaries (Amunts and Zilles, 2001; Grefkes et al., 2001; Rademacher et al.,
2001; Schleicher and Zilles, 1990; Schmitt et al., 2003; Wree et al., 1982), to
delineate the boundaries of labeled axonal plexuses (Lillehaug et al., 2002;
Schwarz and Möck, 2001; Schwarz and Their, 1995), and to identify tracer-
labeled neurons and terminal fields of tracer-labeled axons (Lillehaug et
al., 2002). In principle, these methods apply filtering techniques to digital
grayscale images. Elements of interest are identified with use of threshold
values creating binarized images (Fig. 17.2; see also Lillehaug et al., 2002).
Various procedures, included in most image analysis software packages, are
in turn used to distinguish between objects of interest and artifacts or noise.
The end result of such image analysis based methods may, e.g., be lists of
point coordinates representing the centers of identified objects, such as
labeled cell bodies (Fig. 17.2; see also Lillehaug et al., 2002).
C. Data Acquisition Exemplified
In our investigations of the cerebro–pontine system illustrated here, we

used image-combining microscopy as outlined above to digitize contour
lines for a number of structures (the ventral surface of the pons, the
outlines of the pontine gray, the contours of the corticobulbar and corti-
cospinal fiber tracts, the midline of the brain, and the outline of the fourth
ventricle; see Figs. 17.3A and 17.4E, F) as a reference for the alignment
Figure 17.2. Automatic methods for data acquisition from serial sections through rat
pontine nuclei (modified from Lillehaug et al., 2002, with permission). Automatic
digitization of labeled fiber distributions: (A) Labeled cerebro–pontine axonal plexuses
are visible as dense and loose plexuses in an image of a transverse section through
the pontine nuclei. The labeled fibers, appearing as black or shades of gray, are
identified from the digitized gray-scale image by choosing a threshold value from
minimum 0 (black) to a certain maximum level (dark gray), and represented as
white (B). A binary image (C) is created from a chosen threshold value. By super-
imposing a grid onto the binary image (D), it is possible to obtain x, y coordinates
from squares containing more than a given number of white pixels, which in turn
may be represented as black dots (G). Automatic detection of labeled cell distributions:
Digital gray-scale image of transverse section through the pontine nuclei (E), show-
ing pontocerebellar neurons labeled following injection of the fluorescent tracer
rhodamine conjugated dextran amine into the cerebellar crus IIa. The digital image
is first converted to a binary black and white image by applying a threshold value,
and binary operators are applied to remove artifacts, fill closed gaps, and separate
particles. (F) Unique colors are assigned to different area ranges, and spatial coor-
dinates (x, y ) are automatically assigned to the center of gravity of each particle,
and may be represented by black dots in a graphic plot (H). Comparison of the
original “raw image” with the analyzed image shows that most labeled neurons are
recognized correctly. Some cells, however, were not counted as separate objects due
to the merging of overlapping cells. Weakly labeled cells, cellular fragments, and
small artifacts (pink) that gave rise to particle sizes below the defined size (area)
range for labeled cells were excluded from the plot in (H). Scale bars: 200 µm.
of the sections. The density and distribution of anterogradely labeled ax-

onal plexuses within the pontine nuclei were digitized semiquantitatively as
points (Figs. 17.4C, E and 17.7; Leergaard et al., 1995, 2000a, b, 2004). In
areas with low densities of labeling, point coordinates were placed at regu-
lar intervals along the length of single axons. In areas with dense labeling,
a rough correspondence was sought between the density of labeling and
the number of digitized points. When different investigators recorded data
Figure 17.3. Assembly and visualization of a 3D reconstruction of the rat brain stem
and cerebellum (modified from Brevik et al., 2001). (A) Series of digitized trans-
verse brain sections are aligned interactively according to multiple anatomical land-
marks. (B) The external boundary of the brain stem, digitized from a sagittal section
through a different rat brain (bold contour line), is used as a template for fine adjust-
ment of the dorsoventral alignment of the transverse sections. (C) Oblique lateral
view of the complete aligned 3D reconstruction. Real-time rotation with inspection of
the 3D reconstruction from multiple angles of view was used to facilitate alignment.
(D) The outer boundaries of the cerebellum, the descending peduncle (corticobul-
bar and corticospinal tracts), the fourth ventricle, and precerebellar nuclei (pontine
nuclei, trigeminal nuclei, reticulotegmental nucleus of the pons/RtTg, lateral retic-
ular nucleus/LRt, and the inferior olive) are represented as solid surfaces.
from the same sections, some variation in the density of digitized points
was observed, but spatial distribution patterns remained nearly identical.
This approach allowed subsequent analyses of spatial distributions, density
gradients, and overlap patterns (Figs. 17.5–17.9). A similar logic as above
Figure 17.4. Methodological procedures from cerebral cortical injection of axonal

tracer to assembly of 3D reconstruction of the rat pontine nuclei (modified from
Bjaalie et al., 2005). (A) Drawing of the right cerebral hemisphere of the rat with
a cartoon representation of the somatotopic map of body surface representations
found in the primary somatosensory cortex (SI, redrawn from Welker, 1971). The
image in (B) shows two typical tracer injection sites (Fluoro-ruby, FR, with a bright
appearance, and biotinylated dextran amine, BDA, seen as a black area) in a
horizontal section through SI (data from Leergaard et al., 2000a, case D46). The
injected axonal tracers give rise to labeled axons in the pontine nuclei. The image in
(C) is taken from a transverse section through the pontine nuclei and shows typical
axonal clusters labeled with FR (bright) and BDA (black). The positions and sizes
of the FR (red) and BDA (black) injection sites are indicated in a drawing of the
SI whisker barrels (D), which represent the mystacial vibrissae arrangement on the
contralateral mystacial pad (redrawn from Fabri and Burton, 1991). (E) Digitized
transverse section [same section as in (C)] frame indicates position of the image
shown in (C) in which labeled axonal plexuses are semiquantitatively represented
applies to investigations focusing on distribution of cell bodies, labeled with
axonal tracing techniques or any other technique that would lead to la-
beling of a population of cells, or to studies of shapes and dimensions of
brain areas and regions (for references based on the same analytical proce-
dures as outlined here but with other data categories, see Brevik et al. 2001;
Lillehaug et al., 2002; Nikundiwe et al., 1994; Vassbø et al., 1999). For land-
marks and boundaries, usually recorded as open or closed contour lines,
the spacing of points and relationships of start and end points for each
boundary through the stack of sections should be as consistent as possible
to facilitate later surface modeling (see section “Primary Viewing of the
Reconstruction”).
V. ALIGNMENT OF DATA FROM SECTIONS: BUILDING

THE RECONSTRUCTION
Since the data acquisition process is performed independently for each

section, the spatial relationships between the geometric representations
(points and lines) collected from the serial sections are disrupted. Data
from each section level thus need to be registered to recover the original
relative position prior to the sectioning. Depending on the software em-
ployed, manual, automatic, or combined methods may be used. Here we will
only deal with the fundamental approaches independent of the particular
implementation. Under most circumstances, the best end result is produced
by the combined use of several approaches, as outlined below.
A. Fiducials and Block-Face Images
Given that the use of fiducial markers may be problematic, as discussed

above (section “Orientation of Section Plane”), here we will only acknowl-
edge that these markers represent a possible aid during alignment and
emphasize that they must be used with care, in particular to avoid tissue
damage. Block-face imaging has been reported to be of considerable aid
←
Figure 17.4. (Cont.) as dots (FR, red dots, BDA, black dots), together with lines
representing selected anatomical boundaries. (F) Series of digitized sections are
assembled to a 3D reconstruction by interactive alignment of multiple anatomical
landmarks (arrows) while inspecting the growing 3D reconstruction (black sections)
from multiple angles of view on the computer screen. (G) The complete 3D recon-
struction with different anatomic structures displayed in different colors, in a view
from ventral. (H) Visualization of the same reconstruction as in (G), with bound-
aries of the descending fiber tracts shown as solid surfaces and boundaries of the
pontine nuclei and ventral brain stem surface as transparent surfaces. Scale bar in
(C): 100 µm. Scale bar in (D): 1 mm. Abbreviations: A, anterior; M, medial; ped,
peduncle; pn, pontine nuclei; tfp, transverse fibers of the pons.
during alignment of sections from large tissue blocks or whole brain (Toga
and Banerjee, 1993; see also, Ewald et al., 2002). With this method, the cut
surface, or face, of the tissue block is digitally imaged prior to sectioning. An
image is collected for each section level, and since the camera is fixed, the
relative positions of all block-face images (and hence, sections) are thereby
collected. To detect possible accidental movement of the camera during
the sectioning procedure, calibration via other markers, such as a grid, is
recommended.
B. Landmarks and Boundaries
The primary focus during data acquisition is usually to record data de-
scribing the region of interest, specific labeling patterns, or other features of
relevance for the biological problem in question. In addition, it is essential
to also record landmarks and boundaries for the purpose of alignment of
data from the sections. It is particularly important to record distinguished
landmarks close to the region of interest, but more remote landmarks may
also be of value for the initial global orientation of the section drawings (see
section “Example alignment”). Pial or ventricular surfaces, distinct bound-
aries of fiber tracts, the midline of the brain, and the most prominent ar-
chitectonic boundaries represent examples of tissue elements that may pro-
vide useful guidance for the subsequent alignment of data from a series of
sections.
C. Template Sections, Atlases, and Tomography Data
Template sections with an orientation perpendicular to the section plane

used to generate the 3D reconstruction may be useful as a reference during
the alignment procedure (Fig. 17.3B; Brevik et al., 2001). The perpendicular
section template serves as an aid to reconstruct the correct curvature and
slope of the reconstructed brain region. Use of whole model templates, such
as 3D digital atlases (Fig. 17.10; Leergaard et al., 2003) or tomographic im-
ages of the intact specimen (or a similar specimen), also represents powerful
strategies for the alignment of section data. A general discussion of the use
of reference brains for registration is provided by Amunts and Zilles (in this
volume).
D. Example Alignment
In the example reconstructions from the rat brain stem illustrated here,
we used multiple landmarks and boundaries, as well as template sections.
Thus, to build our brain stem reconstructions from series of transverse sec-
tions (Figs. 17.3A, B and 17.4E–G), we normally use the brain midline, the
ventral surface of the brain stem (with particular emphasis on the groove
of the basilary artery), and the floor of the fourth ventricle as global and
initial landmarks, before conducting a more fine adjustment based on local
landmarks, such as the external boundaries of the pontine nuclei and the
descending corticobulbar and corticospinal pathways (peduncles). Irregu-
larities in the surface, recognizable across sections, were also detected and
used during the alignment, since these were often systematically repeated
at the same locations through several adjacent sections. Furthermore, we
used structures known to have a perpendicular orientation relative to the
section plane: in the case of our brain stem reconstructions, the descending
corticobulbar and corticospinal tracts (Figs. 17.3 and 17.4) were particularly
useful. The overall straight course of these structures was faithfully recon-
structed. Finally, to help perform an even finer alignment of the transverse
series of sections, a series of sagittal sections through the brain stem of an-
other brain (from an animal of the same strain, gender, and age as the
reconstructed brain) were translated to the coordinate system of the recon-
struction and used as a template to adjust the dorsoventral location of the
sections.
VI. COORDINATE SYSTEMS
Implicitly, any 3D reconstruction consisting of x, y , z coordinate data

occurs in a coordinate system. Coordinate systems determined by trivial fac-
tors, such as values recorded from a digitizing tablet or ranges of movement
of a motorized microscope stage, are of limited value. Coordinate systems
that refer to an established standard can, however, be of considerable use.
Thus, transfer of all data collected in a given study to the same standardized
coordinate system facilitates interindividual comparisons and extraction of
organizational principles based on multiple experimental animals (for a
general discussion, see Bjaalie, 2002).
3D reconstructions covering an entire brain require, needless to say,
the use of a global coordinate system. Such coordinate systems are usu-
ally skull based and well defined in standard atlases for a given species
(see, e.g., Paxinos and Franklin, 2001; Paxinos and Watson, 1998, 2005).
But skull coordinates are not easily used, since standard sectioning proce-
dures require that the brain is removed from the skull. Local coordinate
systems, however, rely entirely on the use of landmarks and boundaries
intrinsic to the brain and are therefore more easily implemented. It is
recommended that 3D reconstructions covering only limited parts of the
brain employ a standardized, local coordinate system, which in turn are
mapped onto a global coordinate system defined in a standard atlas (see
Appendix).
With the use of a local coordinate system, comparison of data collected
across experimental animals is facilitated. If a small brain region is stud-
ied, and a defined rodent strain is chosen (to limit variations in shapes of
the region studied), simple procedures can be employed to transfer data

from multiple experimental animals to the same coordinate system. In each
animal, the same landmarks and boundaries will have to be identified in
order to establish the local coordinate system. Once this operation has
been carried out, only adjustments for size differences (scaling) and ori-
entation (rotation) are needed to overlay the data from different animals.
In principle, the procedure is independent of section angle. In practice,
it is considerably more convenient to perform the data transfer if the sec-
tion angle is standardized (see section “Orientation of Section Plane”). This
is primarily achieved by accurate determination of the section angle prior
to sectioning (Fig. 17.1C) and secondarily by measurements from the sec-
tions. To this end, high-magnification images of the intact tissue block are
useful.
For our investigations of the rat pontine nuclei, we have implemented a
local coordinate system (Fig. 17.11; Brevik et al., 2001, see also Leergaard
et al., 2000a, b, 2004), suitable for detailed analysis of experimental label-
ing patterns and presentation of neural tracing data in standardized dia-
grams (Figs. 17.5A, B, 17.6, and 17.9). This standardized coordinate system
(detailed in section “Application of Local Coordinate System for the Rat
Pontine Nuclei”) consists of a bounding box oriented in parallel to the de-
scending fiber tract at the level of the pontine nuclei, a plane corresponding
to the brain stem midline, and boundaries corresponding to the extreme
lateral, rostral, caudal, and ventral limits of the pontine nuclei. The dorsal
boundary is defined indirectly, using the ventral surface of the peduncles
as an intermediate landmark. It is applied individually to each 3D recon-
struction, and subsequently size adjusted to fit the standardized coordinate
system. In this way, point coordinate data from multiple animals are com-
bined within the same spatial framework. Based on this coordinate system,
we have made diagrams showing the outlines of the pontine nuclei in stan-
dard angles of view (Figs. 17.5, 17.6, and 17.9). These diagrams facilitate
data presentation and comparison of results from different animals. By reg-
istering the local coordinate system to a global (skull-based) coordinate sys-
tem, the local pontine positional values are readily translated into global
(sterotaxic) values (see section “Translation Between Local and Global
Coordinates”).
VII. VISUALIZATION AND ANALYSES
In the present account, most of the descriptions are based on the program
Micro3D, used in original research reports dealing with tract-tracing stud-
ies on insects (Berg et al., 1998, 2005), chicken embryo (Diaz et al., 2003),
newborn rat (Leergaard et al., 1995), adult rat (Leergaard et al., 2000a, b,
2003, 2004; Zaborszky et al., 2002), cat (Bajo et al., 1999; Malmierca et al.,
1998), and monkey (Vassbø et al., 1999). The Micro3D software has recently
been made available for general use by the neuroscience community in a
Figure 17.5. Standardized local coordinate system applied to 3D reconstruction of

the rat pontine nuclei and used as basis for combined presentation and analysis of
axonal tracing data obtained from different experimental animals [(C–I) are from
Bjaalie et al., 2005, with permission]. (A) shows a computer-generated 3D recon-
struction of the pontine nuclei (dark blue solid surfaces) and cerebral peduncles
(bright blue surfaces) in view from ventral. In (B), only the right pontine nuclei are
shown in an oblique ventromedial view. The local pontine coordinate system is ap-
plied by size adjusting a cuboid bounding box to fit the defined nuclear boundaries
(for further details, see section “Application of Local Coordinate System for the Rat
Pontine Nuclei”). (C–I) show the spatial distribution of cerebro–pontine projections
originating from eight different cortical sites in the primary somatosensory cortex.
The eight injection sites were taken from six animals that had been submitted to
tracing experiments in which the distribution of labeled cerebro–pontine axons was
3D reconstructed and transferred to the local pontine coordinate system (data from
Leergaard et al., 2000a, b). (C) The size and location of the eight tracer injection
sites within the primary somatosensory cortex are shown in the drawing of the right
cerebral hemisphere (modified from Welker, 1971). (D) Colored dots, representing
labeled cerebro–pontine axons, are shown together in a simplified frame, represent-
ing the local coordinate system. (E–I) Spatial distribution patterns and topographic
organization examined in consecutive 100-µm-thick slices oriented in parallel with
the ventral surface of the pons, from ventral (E) to dorsal (I) at locations indicated in
(D). The data sets were combined and analyzed using tools available in the database
application FACCS (www.rbwb.org; Bjaalie et al., 2005).
Figure 17.6. Dot maps showing the pontine labeling resulting from 54 individual
tracer injections into major SI body representations in rat, color coded according to
cortical site of origin (A) [from Bjaalie et al. (2005), with permission]. The drawing
in (A), with a cartoon representation of the SI body map, is modified from Welker
(1971). The dot maps are shown as projections viewed from ventral (B) or as 50-µm-
thick sagittal (C–E) or transverse (F–H) slices. The pontine projections arising from
SI whisker representations (purple dots, 40 cases) surround upper lip representa-
tions (red dots) externally and are largely colocated with projections from SI trunk
representations (green dots, B, C, F). In general, pontine clusters receiving pro-
jections from different SI body parts have complementary shapes and close spatial
relationships. In most slices through the models, patterns of interdigitating labeled
clusters are visible. While the neighboring relationships of the 2D SI map are largely
preserved in the pontine map, the 3D nature of this map allows introduction of new
neighboring relationships.
version running under Linux (Bjaalie et al., 2006; download available from
www.nesys.uio.no and from the Rodent Brain Workbench, www.rbwb.org).
Some illustrations (Figs. 17.5D–I and 17.6B–H) are produced with a suite
of tools based on Java (Sun Microsystems Inc., Santa Clara, CA) made avail-
able via the database application FACCS (www.rbwb.org). The description
below is generic only and does not deal with technical details, algorithms,
or comparisons of software solutions. Several other comparable tools and
approaches are available (see, e.g., Lohmann et al., 1998; Maurin et al., 1999;
Van Essen et al., 2001).
A. Primary Viewing of the Reconstruction
Rotation and zooming, performed in real time, with or without perspec-

tive viewing, represent primary manipulations that most users will perform
immediately on a 3D reconstruction. The visualization of the geometry of
points and lines in the 3D reconstruction relies on the chosen rendering
parameters, such as point size, line thickness, and different colors for each
object category (an object category in the present context is, e.g., different
populations of labeled cells and different categories of boundaries of brain
regions). Rendering of individual data points as spheres of variable diameter
may also be useful.
A fundamental visualization approach that helps perceive the shapes and
spatial relationships in the data is surface modeling based on contour line
data, describing boundaries of brain regions or external (pial) or internal
(ventricular) surfaces of the brain. Surface modeling from stacks of contour
lines can be performed using various approaches, in principle, in two steps:
first, a geometry of polygons is created and second, shading is applied to
each polygon, resulting in the “resynthesis” of a surface based on the con-
tour line data. The surface can typically be viewed as solid or transparent.
The final surface may be either rough (with the surface tightly coupled
to the original coordinate data, Fig. 17.5A, B), or smoothened (Figs. 17.3D
and 17.4H). Methods range from the construction of a geometry of triangles
directly based on the coordinates recorded along the contour lines (exem-
plified in Toga, 1990) to more complex parametric methods (Bjaalie et al.,
1997a; Geiger, 1993) and implicit methods (Shen et al., 2004). Examples of
point, line, and surface representations are given in Figs. 17.3–17.6, with the
transition from stacks of contour lines to surfaces are exemplified in Fig.
17.4G, H.
Most data sets need to be edited and optimized to achieve an optimal
surface modeling. Depending on the method employed, requirements may
include the ordering of the coordinates describing each contour line, use
of regular or irregular distances between points recorded along the contour
lines, and use of few or many points along the contour lines. Requirements
for the chosen surface modeling method should be taken into considera-
tion before data acquisition is performed. The possibilities for editing the
sequence and density of point coordinates in contour line data differ among
software solutions. In general, independent of the method chosen, a good
end result often follows from finding the right balance between the density
of points used to represent the contour lines, and the spacing between the
contour lines.
For combined visualization of points, lines, and surfaces, it is important
to find a balance between object and background colors, and surface trans-
parencies (Fig. 17.4H). Contour lines may be used to emphasize shape
when surfaces are made smooth and transparent, and are especially use-
ful when multiple layers of transparent surfaces are rendered (Fig. 17.4H).
When data points representing labeled structures such as cells or axonal
terminal fields, distributed through the thickness of a section, have been
recorded without assigning individual z-values within the section (i.e., all
data points from a given section are assigned the same z-value), random-
ized distribution of the data points within the thickness of the section
may be used to avoid an artificial spacing of the data points when viewed
“on edge.”
In Fig. 17.4G, H, we exemplify the primary viewing of a reconstruction of

the pontine nuclei from rat. The data shown represent the external brain
surface at the level of the pons, nuclear boundaries, and landmarks, as well
as patterns of axonal terminal fields resulting from dual tracer injections
placed in specific, electrophysiologically, and architectonically defined di-
visions of the barrel cortex. The data show that the two injection sites gen-
erate topographically distinct labeling patterns (for further considerations,
see Leergaard et al., 2000a).
B. Slicing
Visualization and analysis based on a single plane of sectioning may lead

to incomplete interpretations of spatial relationships. 3D reconstructions
partly solve this problem by bringing together the information that was sep-
arated by the sectioning. But subdivision of the 3D reconstruction is often
required to perform a detailed analysis and to extract essential informa-
tion. Reslicing of the reconstruction at chosen thicknesses and orientations
therefore represent a key analytical tool (Figs. 17.5 and 17.6). Such dynamic
reslicing, independent of the original section plane, introduces a new di-
mension to the analysis of brain topography. In our example analysis from
the rat pontine nuclei, the distribution and shape of the labeling patterns
of cerebro–pontine terminal fields are analyzed to advantage with reslicing
at variable angles (Fig. 17.6). When superimposing data from many ex-
periments in the same local coordinate system for the pontine nuclei (see
sections “Coordinate Systems” and “Application of Local Coordinate Sys-
tem for the Rat Pontine Nuclei”), reslicing turns out to be critical to reveal
spatial relationships in the data. For somatosensory cerebro–pontine termi-
nal fields, such an analysis has demonstrated the presence of concentrically
organized shells or lamellae with spatial relationships in the cortical map
being largely preserved in the pontine map (reviewed in Leergaard, 2003).
Careful inspection and digital reslicing of the reconstructions from multi-
ple angles of view have therefore been instrumental to fully understand the
3D distribution and shape of the labeling patterns (Leergaard et al., 2000b,
2004; for another example, see Malmierca et al., 1998).
C. Surface Modeling of Labeling Patterns
Spatial relationships among clouds of points representing patterns of la-

beling may also be shown to advantage with use of surface modeling. Such
surface modeling, based on point data representing labeled structures such
as cells and axonal terminal fields (and not contour line data representing
distinct boundaries), may be achieved by manually digitizing a set of con-
tour lines surrounding the dense regions of labeling (see, e.g., Leergaard
et al., 2000a, b; Malmierca et al., 1998). This simple approach may work well
for dense clusters of labeling. Automatic generation of a geometric repre-
sentation surrounding the point clouds is obviously more attractive and has
been shown for analysis of tract-tracing data in, e.g., Bjaalie et al. (1997a), as
well as in a recent analysis of the rat pontine nuclei (Leergaard et al., 2004).
In the most recent example, a method based on scalar fields (generated by
binning the point coordinates and estimating the relative density for each
bin) was used. For visualization, the isosurfaces were extracted by marching
cube-like algorithm (J.O. Nygaard, S. Gaure, C. Pettersen, H. Avlesen, and
J.G. Bjaalie, in preparation; see also, Nadasdy and Zaborszky, 2001). In the
example shown in Fig. 17.8C, D, isodensity surfaces were used to model
point clouds representing axonal terminal fields originating in precisely
defined parts of the primary motor and primary somatosensory cortices. In-
jection site details and labeling patterns as they occur in single sections are
shown in Fig. 17.7. This analysis demonstrated to advantage the shapes of
the terminal fields of labeling and aspects of the neighboring relationships
(Leergaard et al., 2004).
D. Analysis of Spatial Overlap
A useful approach for extracting quantitative information about the distri-

bution of labeling deals with the visualization and analysis of spatial overlap
between pairs of data. The problem raised is whether two populations of
points are segregated, overlapping, or randomly distributed with respect to
each other. If there is overlap, the overlap in turn needs to be quantitatively
expressed.
The terms “segregation” and “overlap” are commonly used to describe
spatial relationships, but they are not trivial to define, and different analytical
models should be employed for different purposes (Bjaalie and Diggle,
1990; Bjaalie et al., 1991; Diggle, 1986).
A typical problem confronted in investigations of system level connec-
tivity in the brain is to characterize whether two populations of labeled
neuronal elements are overlapping or segregated. To the extent that they
are overlapping, interaction is more likely to occur also at a functional level.
To estimate the amount of overlap in a material containing labeled axonal
terminal fields, many investigators apply a grid to each section and then
count the number of events (points of different categories representing
labeling) in each bin in the grid before computing statistics from the data
(Alloway et al., 1999; He et al., 1993; Leergaard et al., 2000a, 2004). Fig-
ure 17.7E, F demonstrates such an analysis at the single section level for
axonal tracing data. Figure 17.8E–H shows a similar analysis implemented
at the level of complete 3D reconstructions. It shows the 3D distribution
of bins expressing double labeling, defined as the presence of data points
of both labeling categories at a defined threshold (Leergaard et al., 2004).
The results of such analyses are inherently related to the choice of bin size,
and in case of axonal terminal fields, also to the data acquisition method
Figure 17.7. Digitization and 2D overlap analysis of two populations of cerebro–

pontine axons labeled after dual injections of axonal tracers into homotopic whisker
representations in the right primary motor (MI) and somatosensory (SI) cortices in
rat (modified from Leergaard et al., 2004). (A and B) Images of tangential sections
through cortical layer V, showing a biotinylated dextran amine injection (BDA) in
MI (A), and a Fluoro-ruby (FR) injection in SI (B). (C) The localization and extent
of the injection sites are mapped onto a diagram of the rat sensorimotor cortex (for
details, see Leergaard et al., 2004). (D) Axonal plexuses, labeled with BDA (black)
and FR (bright), are observed in a transverse section through the pontine nuclei.
(E) Raw data plot of the section shown in (D). Anatomical landmarks and bound-
aries are represented by lines and the axonal labeling by black (BDA) and red (FR)
dots. (F) Overlap analysis of the same section as shown in (E). The dot map is subdi-
vided into 35-µm2 bins, and the numbers of FR and BDA dots are counted in each
bin. Bins containing at least one or more FR or BDA dots are colored red or black,
respectively; those containing at least two of each type are colored green. The analy-
sis shows that the axons arising from homotopic whisker representations in MI and
SI are partly overlapping.
and the densities of points used to code a specific labeling intensity. The
method exemplified here allows robust relative comparisons among cases,
as well as 3D visualization of the shape and location of the estimated zones of
overlap.
Figure 17.8. Computer-generated visualization (stereo-pairs showing the 3D distri-

bution, shape, and relationship within the ipsilateral pontine nuclei) of cerebro-
pontine projections labeled after axonal tracer injections into homotopic whisker
representations in primary motor (MI), primary somatosensory (SI), and secondary
somatosensory (SII) cortices. Left column shows dual injections into MI (blue) and
SI (red). Right column shows dual injections into SI (blue) and SII (red) [modified
from Leergaard et al. (2004), with permission]. The position and extent of the in-
jections are indicated in the outline drawings of the cortical maps. To see 3D images
(A–H), the viewer must cross the eye axis to let the stereo-pair of images merge.
(A, B) Dot maps showing the distribution of cerebro–pontine axons labeled with
biotinylated dextran amine (BDA, blue) and Fluoro-ruby (FR, red). (C, D) Isoden-
sity surfaces surrounding the clusters of FR and BDA labeling, shown in view from
ventral. (E, F) 3D distribution of the results of overlap analysis (for details, see Fig.
17.7), with red (FR) and blue (BDA), and yellow (FR and BDA) bins. The shape and
distribution of labeled clusters originating from SI and SII are similar (B, D), while
labeled clusters originating from MI have a more rostral and ventral distribution (A,
C). The dual injections in SI and SII produce more overlap (F, H) than the dual
injections in SI and MI (E, G).
E. Density Gradient Analysis
Another approach for extracting quantitative information about the dis-

tribution of the labeling is density gradient analysis. This method has been
employed in our investigations of pontine nuclei organization (Fig. 17.9;
Leergaard et al., 2000b, 2004; for similar analysis of other brain regions, see
Malmierca et al., 1998; Vassbø et al., 1999). The analysis shown in Fig. 17.9 is
based on 2D “collapsed” projections of the 3D point data representing the
labeling. The analysis may be repeated for different angles of view. A square
grid is superimposed on the 2D map, and a gray or color level is assigned to
Figure 17.9. Dot maps (A, B) and density maps (C, D) showing the distribution of
whisker-related cerebro–pontine projections from MI (A, C) and SI (B, D) [modi-
fied from Leergaard et al. (2004), with permission]. Data from 14 different tracing
experiments (for details, see Leergaard et al., 2004) have been accumulated in the
same local pontine coordinate system and are shown in a view from ventral. Only
the right half of the pontine coordinate system is shown. Dots represent the spatial
distribution of labeled cerebro–pontine axons (A, B). The color gradient in (C)
and (D) shows the highest densities in red and the lowest in violet. Densities < 5%
of the maximum value are not shown. Overall, the distribution of high-density re-
gions of labeling is different. Together, the SI and MI high-density regions form
complementary parts of a ring-like volume. Scale bar: 500 µm.
each square, corresponding to the density of points within a user-defined
radius centered on the square. Thereby, a grayscale (or color) coded den-
sity map is constructed. With the use of small grid size and short radius, it
is possible to demonstrate changes in densities across short distances. The
example shown in Fig. 17.9 uses data accumulated from several experiments
and shows the overall differences in projection patterns of primary motor
and primary somatosenory projections to the pontine nuclei. The density
gradient analyses reveal that the high-density regions of the labeled pro-
jections are differently distributed, and that they together appear to form
complementary parts of a ring-like volume (see also Leergaard et al., 2004).
F. Stereo-Imaging
The classical (printed) journals have several limitations for visualization

of 3D reconstructions. Perception of depth in 3D reconstructions, displayed
in journal figures, may be achieved with the use of stereo-images. The stereo-
scopic effect is realized by the use of image pairs that have approximately 8◦
different vertical rotation. The viewer crosses the eye axis to merge the pair of
images into a 3D image. A series of stereo images of point clouds, surface-
modeled point clouds, and visualizations of spatial overlaps are shown in
Fig. 17.8.
VIII. WHOLE BRAIN ATLASING
The development of 3D digital whole brain atlas systems represents an

emerging opportunity for making more efficient use of data recorded from
histological sections. Recent developments in the fields of informatics, imag-
ing, and computerized microscopy make it technically feasible and scien-
tifically attractive to create such atlas systems. The general motivation be-
hind digital brain atlasing is to assign localization to data from the brain
and to combine data of multiple categories and from different individu-
als in the same information structure. This will allow investigators to aver-
age comparable observations, describe variability, reuse data in new com-
binations, and overall to extract knowledge based on large data sets. Many
data categories may be included in such atlas systems, including axonal
tract-tracing data describing organization of projections and connections,
as exemplified in the illustrations in the present chapter. At the level of the
mouse brain, extensive efforts are currently undertaken to build whole brain
atlas systems for gene expression data. Examples include the Allen Brain
Atlas (Pennisi, 2003) and The Mouse Brain Atlas (MacKenzie-Graham et al.,
2003).
Local coordinate systems, recommended above to be used for 3D recon-
structions of any region of the brain (smaller than the whole brain), can
be translated to global coordinates employed in whole brain digital atlas
systems. An example translation of a local coordinate system onto global

coordinates is shown in section “Translation Between Local and Global Co-
ordinates.” Hence, once 3D reconstructions are prepared in a local coordi-
nate system, the data contained in the reconstruction can in principle be
translated to global coordinates and uploaded in databases that are accessed
by whole brain atlas applications.
Here wee exemplify the potential use of a whole brain 3D digital atlas with
an analysis of in vivo tracing of axonal trajectories and terminal fields using
manganese-enhanced MRI (Leergaard et al., 2003). Figure 17.10 illustrates
a conceptual implementation of a 3D digital atlas template consisting of
contour line data for the brain surface and selected internal structures, cre-
ated by digitizing selected structures from a standard stereotaxic rat brain
atlas (Swanson, 1999; Fig. 17.10A, B). This 3D template is used as a matrix
for combining and segmenting MRI image volumes. The co-registration
of the MR images with the atlas was achieved by matching multiple well-
known anatomical structures, applying the same principles as described
above for the alignment of data from serial sections. Since the identifica-
tion of anatomical landmarks in MR images is variably difficult, depending
on the employed scanning resolution and parameters, the use of a 3D atlas
approach was particularly helpful for assessing the boundaries of several
brain structures (Fig. 17.10I, J). In addition to aiding anatomical parcella-
tion, the 3D atlasing approach was in this example fundamental for the
demonstration of topographical organization in the corticothalamic sys-
tem, based on in vivo tracing of pathways using manganese-enhanced MRI
(Fig. 17.10C–J).
IX. SUMMARY OF ADVANTAGES AND LIMITATIONS
By assembling computerized 3D reconstructions from serial sections, com-

plex spatial relationships may be visualized and analyzed in a more complete
and versatile fashion than with the use of single sections only. A 3D recon-
struction may be submitted to interactive inspection from any angle of view
and sophisticated visualization and analysis including, as outlined above,
digital re-slicing, surface modeling, analysis of spatial overlaps, density gra-
dient analysis, and analysis of other complex spatial relationships between
distributions recorded in the reconstruction. Altogether, the 3D reconstruc-
tion approach facilitates perception of complex architectonical features.
Three-dimensional reconstructed data, as exemplified in the present chap-
ter, are furthermore well suited for quantitative analysis.
A fundamental aspect of computerized 3D reconstruction is the opportu-
nity to compare data from different experimental animals. This can in its sim-
plest form be performed by (1) establishing local, standardized coordinate
systems for the brain region of interest and (2) applying such coordinate
systems to the 3D reconstructed data. This approach implies that data from
different animals are size-adjusted and rotated to fit the chosen coordinate
Figure 17.10. Digital rat brain atlasing applied to facilitate localization of manganese
(Mn2+ )-enhanced MRI signals, used to trace corticothalamic pathways [modified
from Leergaard et al. (2003), with permission]. (A, B) 3D atlas reconstructed from
serial section drawings available in the Swanson atlas of the rat brain (Swanson, 1999).
In (A), the external surface of the brain is shown as a solid surface. Stereotaxic atlas
coordinates are shown as a bounding box with selected reference lines. The position
and orientation of the 200-µm-thick coronal and sagittal slices shown in (C–J) are
indicated by solid blue boxes. In (B), the external surface is made transparent, and
the boundaries of the caudate–putamen complex (pink) and the thalamus (green)
are seen. (C–F) show coronal (C, E) and sagittal (D, F) T1 weighted MRI slices,
obtained with 390 µm3 isotropic voxels, using a 3 T human clinical MRI scanner,
about 10 h after focal injections of manganese chloride (MnCl2 ) were made into
medial (C, D) and lateral (E, F) parts of the somatosensory cortex in two adult rats,
respectively. Different patterns of specific Mn2+ signal enhancement are detectable
in cortico–cortical and cortico–subcortical pathways in the two animals. (G–J) By
superimposing the MR volumes from the two cases onto the 3D digital atlas, it was
possible to demonstrate topographically different labeling patterns. In G and H,
yellow represents labeling in the thalamus following medical injection (also shown
in C, D), and red labeling following lateral injection (also shown in E, F). Atlas
regions are shown in (I–J) with different colors for different regions (bright gray,
white matter; pink, caudate–putamen; orange, globus pallidum; green, thalamus).
system. More advanced warping, in particular needed for brain regions of

highly variable shapes, requires the use of more advanced nonlinear trans-
formations (see, e.g., Toga and Thompson, 1999). In our investigations of
neural populations and sensory projection systems, use of 3D reconstruc-
tions, visualizations, and analysis has been critical for attaining a fuller un-
derstanding of topographic organization and complex spatial relationships
among functionally defined neuronal elements (see, e.g., Bajo et al., 1999;
Bjaalie et al., 1997b, 2005; Leergaard et al., 1995, 2000a, b, 2004; Malmierca
et al., 1998; Nikundiwe et al., 1994). Combined use of tomographic image
data and geometric 3D reconstructions (Leergaard et al., 2003) illustrates
future possibilities for the construction of multidimensional 3D digital brain
atlases, in which multiple modalities of brain data may be brought together
and localized within a 3D atlas framework.
The technical and scientific limitations of 3D geometric reconstructions
are found at several levels. First, the quality of the ensuing 3D models never
exceeds the quality of the original tissue material and is further critically
determined by the procedures used for data acquisition and subsequent
alignment, as well as by the accuracy with which section angle, section po-
sition, and tissue shrinkage are monitored. The examples included in the
present chapter are taken from small animal brains displaying limited in-
dividual variation. The level of complexity is considerably higher in studies
of, e.g., primate or human cortical structures, and with the use of material
with more artifacts and section distortions. Second, the 3D reconstruction
method is overall time consuming, both at the level of tissue preparation
and data acquisition. Robotic tissue processing and automated data acqui-
sition/microscopy approaches (Herzig et al., 2001; Visel et al., 2004) may
further help to rapidly acquire complete high-resolution images of entire
sections at higher speeds. Automated procedures for acquiring the essen-
tial information from such images are being used (see, e.g., Lillehaug et al.,
2002). The 3D reconstruction process may also be further improved with
the aid of tomographic templates. Technical improvement in resolution
and quality of data obtained by tomographic imaging method will also pre-
sumably allow generation of data more suitable for detailed analysis. These
contemporary developments offer new opportunities and will most likely
influence the field substantially over the years to come.
APPENDIX
The descriptions in this section are modified from information posted

in the database application FACCS (www.rbwb.org; see also Bjaalie et al.,
2005).
A. Application of Local Coordinate System for the

Rat Pontine Nuclei
1. Purpose
The purpose is to facilitate interindividual comparison of data originating

from different experimental animals.
Figure 17.11. Local coordinate system for the rat pontine nuclei [from the database
application FACCS (www.rbwb.org; Bjaalie et al., 2005)]. (A) Image of the ventral
surface of the rat brain with the standardized local coordinate system for the pontine
nuclei. (B) The standardized local coordinate system shown in relation to the pon-
tine nuclei (blue surfaces) and descending fiber tracts (dark gray surfaces). (C) The
standardized local coordinate system shown in isolation.
2. Definition
The local coordinate system is a cuboid, referred to below as a bounding

box, with anatomically defined orientation, center, and extent. In order to
apply the coordinate system, using software tools such as those exemplified
by Bjaalie et al. (2006), the following anatomical boundaries have to be
identified from a stack of histological sections through the pontine nuclei
and in the 3D reconstruction based on such sections (Fig. 17.12):
r ventral surface of corticobulbar and corticospinal fiber tracts (pedun-

cle)
r rostralmost extent of pontine gray substance
r caudalmost extent of pontine gray substance
r ventralmost extent of pontine gray substance
r lateralmost extent of pontine gray substance
r midline of the brain.
The coordinate system is applied individually to the right and left pontine
nuclei. A bilateral coordinate system is obtained by subsequently scaling the
two sides so that all edges are aligned.
Figure 17.12. The boundaries of the rat pontine nuclei gray matter (dotted lines)
determined in images of transverse sections through the pontine nuclei, stained for
cresyl violet [from Brevik et al. (2001), with permission]. Small cell groups belonging
to the pontine nuclei are also found inside, and along the dorsal aspect, of the de-
scending fiber tracts. Abbreviations: IP, interpeduncular nuclei; ml, medial lemniscus;
ped, peduncle (corticobulbar and corticospinal tracts); Pn, pontine nuclei; RtTg,
reticulotegmental nucleus of the pons. Scale bar: 200 µm.
Orientation: The bounding box is oriented so that its surfaces are parallel
or perpendicular to the midline of the brain and the long axis of the
brain stem at the level of the pons (identified by the ventral surface of
corticobulbar and corticospinal fiber tracts).
Extent toward rostral, caudal, ventral, and lateral: The extent of the bounding
box in these directions is defined by the maximum extent of the pontine
gray substance in the same directions.
Center point: The center point in each half of the bilateral coordinate
system is placed at the ventral surface of the peduncle, halfway from
rostral to caudal and halfway from the midline to the lateral end of the
pontine nuclei.
Extent toward dorsal: The dorsal boundary of the pontine nuclei (pon-
tine neurons located dorsal to the fiber tracts) is less distinct than
the other boundaries. The dorsal extent of the bounding box, rel-
ative to the center point (see above), is therefore defined as the
distance from the center point to the ventral boundary of the bounding
box.
Origin: The origin of the local pontine coordinate system is placed at the
intersection of the midline and a line connecting the left and right
center points. All data coordinates made available through the FACCS
database are related to this origin defined for a bilateral coordinate
system. Note that the origin of the relative coordinates used in our
presentation diagrams in several of the related publications for practical
reasons is differently defined.
Standard size: Following application of the local pontine coordinate system
to an individual animal, the coordinate system and related data may be
size adjusted to a standard of choice in order to facilitate interindi-
vidual comparison. We have defined a standard based on the average
size measured from the experimental animals included in the publica-
tion by Leergaard et al. (2000b). The bounding box is size adjusted to
2000 × 2000 × 1200 µm for each side (rostrocaudal × mediolateral ×
ventrodorsal distances), i.e., 2000 × 4000 × 1200 µm for both sides of
the pontine nuclei.
3. Technical Implementation
a. Application to a 3D Reconstruction
A 3D reconstruction consists of a stack of digitized sections. Each digitized

section contains point and line coded data representing labeling (points),
boundaries, and landmarks (lines). The local coordinate system is applied
to a 3D reconstruction as follows:
r Orientation: A cuboid bounding box is oriented along the long axis of

the brain stem at the level of the pons.
r Center point: The center point is found at the ventral surface of the
peduncle by measuring or slicing the reconstruction of the pontine
nuclei halfway from rostral to caudal, and halfway from the midline to
the lateral end.
r Extent: The boundaries of the box are adjusted to fit the (histologically
defined) rostral, caudal, lateral, medial, and ventral limits of the pontine
nuclei. The distance from the center point to the dorsal boundary is
set to be equal to the distance from the center point to the ventral
boundary.
r Standard size: Affine transformations (i.e., combinations of linear scal-
ing, translation, and rotation) are used to bring individual coordinate
systems (together with all related data) in register with a chosen stan-
dard coordinate system.
r Application to 2D section images: The local coordinate system may be ap-
plied to pontine nuclei section images if the position and angle of the
sections are known and the above listed anatomical boundaries are
identified (i.e., maximum rostrocaudal and mediolateral extent of pon-
tine gray substance, and distance from center point to the maximum
ventral extent of pontine gray substance). The coordinate system is ap-
plied using the above-described criteria.
b. Application to 2D Images
The local coordinate system may be applied to pontine nuclei section

images if the position and angle of the sections are known and the above
listed anatomical boundaries are identified (i.e., maximum rostrocaudal
Bregma
A
−4
Interaural
line
B Bregma
0 –4 –8 –12 –16
C right
8 4
midline
0 4
left
8
8 8
4 4
0 0
6 4 0 –4 –6
'Sagittal view' from left. 'Coronal' view from rostral.
Figure 17.13. The standardized local coordinate system for the rat pontine nuclei
(red bounding box) shown in relation to a 3D reconstruction of the brain stem
(data from Brevik et al., 2001), and the skull-based coordinate system of Paxinos
and Watson (1998, 2005) in oblique lateral (A), sagittal (B), and coronal (C) views
[modified from Bjaalie and Leergaard (2005)]. The external boundaries of the
brain stem are shown as a transparent gray surface and other structures as solid
surfaces (gray, cerebellum; yellow, pontine nuclei; green, descending corticobulbar
and corticospinal tract; dark blue, trigeminal nuclei; red, lateral reticular nucleus;
bright blue, aquaduct/fourth ventricle).
and mediolateral extent of pontine gray substance, and distance from cen-
ter point to the maximum ventral extent of pontine gray substance). The
coordinate is then applied using the criteria above for the application to a
3D reconstruction.
B. Translation Between Local and Global Coordinates
1. Purpose
The purpose is to translate data from the standardized local coordinate

system for the pontine nuclei to a coordinate system for the whole brain
(here exemplified by the skull-based stereotaxic coordinates of Paxinos and
Watson, 1998).
2. Definitions
For the definition of standardized local coordinate system for the pontine
nuclei, see above. For definition of skull-based stereotaxic coordinates, see
Paxinos and Watson (1998, 2005; see also Swanson, 1999; Fig. 17.13).
3. Technical Description
Mediolateral levels are identical in the two coordinate systems and

can thus be directly superimposed. The translation of rostrocaudal and
dorsoventral coordinates from one coordinate system to the other is de-
fined in Fig. 17.14, where the internal coordinate system for the pontine
nuclei is applied to a 2D section image (see Appendix A.3.b Application to 2-
D images ; above) from the Paxinos and Watson (1998) stereotaxic atlas of the
rat brain, allowing translation of rostrocaudal and dorsoventral coordinates
between the two coordinate systems.
Figure 17.14. The local pontine nuclei coordinate system applied to the Paxinos
and Watson (1998) stereotaxic atlas of the rat brain, at lateral level of 1.40 mm
[modified from Brevik et al. (2001), with permission]. The dashed lines and blue
numbers represent the local pontine coordinate system. The black numbers on the
surrounding frame indicate the skull-based stereotaxic coordinates of Paxinos and
Watson (1998). The x-axis shows rostrocaudal position in millimeters relative to
the interaural line, while the y -axis indicates dorsoventral position relative to the
interaural line.
Acknowledgments. This work was supported by grants from The Research

Council of Norway and EC grant QLG3-CT 2001-002256.
REFERENCES
Alloway, K. D., Crist, J., Mutic, J. J., and Roy, S. A., 1999, Corticostriatal projections from rat barrel
cortex have an anisotropic organization that correlates with vibrissal whisking behavior, J.
Neurosci. 19:10908–10922.
Amunts, K., and Zilles, K., 2001, Advances in cytoarchitectonic mapping of the human cerebral
cortex, Neuroimaging Clin. N. Am. 11:151–169, vii.
Bajo, V. M., Merchán, M. A., Malmierca, M. S., Nodal, F. R., and Bjaalie, J. G., 1999, Topographic
organization of the dorsal nucleus of the lateral lemniscus in the cat. J. Comp. Neurol.
407:349–366.
Berg, B. G., Almaas, T. J., Bjaalie, J. G., and Mustaparta, H., 1998, The macroglomerular complex
of the antennal lobe in the tobacco budworm moth Heliothis virescens: specified subdivision
in four compartments according to information about biologically significant compounds,
J. Comp. Physiol. A 183:669–682.
Berg, B. G., Almaas, T. J., Bjaalie, J. G., and Mustaparta, H., 2005, Projections of male specific
receptor neurons in the antennal lobe of the Oriental tobacco budworm moth, Helicoverpa
assulta. A unique glomerular organisation among related species, J. Comp. Neurol. 486:209–
220.
Bjaalie, J. G., 1991, Three-dimensional computer reconstructions in neuroanatomy: basic prin-
ciples and methods for quantitative analysis, In: Stewart, M. G. (ed.), Quantitative Methods
in Neuroanatomy, Chichester: John Whiley & Sons Inc., pp. 249–293.
Bjaalie, J. G., 2002, Localization in the brain: new solutions emerging, Nat. Neurosci. Rev. 3:322–
325.
Bjaalie, J. G., Daehlen, M., and Stensby, T. V., 1997a, Surface modelling from biomedical data,
In: Daehlen, M., and Tveito, A. (eds.), Numerical Methods and Software Tools in Industrial
Mathematics, Boston: Birkhauser, pp. 9–26.
Bjaalie, J. G., and Diggle, P. J., 1990, Statistical analysis of corticopontine neuron distribution
in visual areas 17, 18, and 19 of the cat, J. Comp. Neurol. 295:15–32.
Bjaalie, J. G., Diggle, P. J., Nikundiwe, A., Karagülle, T., and Brodal, P., 1991, Spatial segregation
between populations of ponto–cerebellar neurons: statistical analysis of multivariate spatial
interactions, Anat. Rec. 231:510–523.
Bjaalie, J. G., and Leergaard, T. B., 2005, Three-dimensional visualization and analysis of wiring
patterns in the brain: experiments, tools, models, and databases, In: Koslow, S. H., and
Subramaniam, S., (eds.), Databasing the Brain: From Data to Knowledge, Chichester: John
Wiley & Sons Inc., pp. 349–368.
Bjaalie, J. G., Leergaard, T. B., and Pettersen, C., 2006, Micro3D: computer program for three-
dimensional reconstruction, visualization, and analysis of neuronal populations and brain
regions, Int. J. Neurosci., in press.
Bjaalie, J. G., Leergaard, T. B., Lillehaug, S., Odeh, F., Moene, I., Kjode, J. O., and Darin, D.,
2005, Database and tools for analysis of topographic organization and map transformations
in major projection systems of the brain, Neuroscience, 136:618–696.
Bjaalie, J. G., Sudbø, J., and Brodal, P., 1997b, Corticopontine terminal fibres form small scale
clusters and large scale lamellae in the cat, Neuroreport 8:1651–1655.
Brevik, A., Leergaard, T. B., Svanevik, M., and Bjaalie, J. G., 2001, Three-dimensional comput-
erised atlas of the rat brain stem precerebellar system: approaches for mapping, visualiza-
tion, and comparison of spatial distribution data, Anat. Embryol. 204:319–332.
Brodal, P., 1982, The cerebropontocerebellar pathway: salient features of its organization, In:
Chan Palay, V., and Palay, S. (eds.), The Cerebellum—New Vistas. Exp. Brain Res., Suppl. 6,
Berlin and Heidelberg: Springer-Verlag, pp. 108–132.
Brodal, P., and Bjaalie, J. G., 1992, Organization of the pontine nuclei, Neurosci. Res. 13:83–
118.
Diaz, C., Glover, J. C., Puelles, L., and Bjaalie, J. G., 2003, The relationship between hodolog-
ical and cytoarchitectonic organization in the vestibular complex of the 11-day chicken
embryo, J. Comp. Neurol. 457:87–105.
Diggle, P. J., 1986, Displaced amacrine cells in the retina of a rabbit: analysis of a bivariate
spatial point pattern, J. Neurosci. Methods 18:115–125.
Ewald, A. J., McBride, H., Reddington, M., Fraser, S. E., and Kerschmann, R., 2002, Surface
imaging microscopy, an automated method for visualizing whole embryo samples in three
dimensions at high resolution, Dev. Dyn. 225:369–375.
Geiger, B., 1993, Three-dimensional modelling of human organs and its application to diagnosis and
surgical planning, Institut National de Recherche en Informatique et en Automatique,
Sophia-Antipolis, France, Report 2105.
Glaser, E. M., Gissler, M., and Van der Loos, H., 1979, An interactive camera lucida computer-
microscope, Soc. Neurosci. Abstr. 5:1697.
for image combining microscopy, Comput. Med. Imaging Graph 14:307–317.
Glaser, E. M., Tagamets, M., McMullen, N. T., and Van der Loos, H., 1983, The image-combining
computer microscope—an interactive instrument for morphometry of the nervous system,
Grefkes, C., Geyer, S., Schormann, T., Roland, P., and Zilles, K., 2001, Human somatosensory
area 2: observer-independent cytoarchitectonic mapping, interindividual variability, and
population map, Neuroimage 14:617–631.
He, S. Q., Dum, R. P., and Strick, P. L., 1993, Topographic organization of corticospinal pro-
jections from the frontal lobe: motor areas on the lateral surface of the hemisphere, J.
Neurosci. 13:952–980.
Herzig, U., Cadenas, C., Sieckmann, F., Sierralta, W., Thaller, C., Visel, A., and Eichele, G.,
2001, Development of high-throughput tools to unravel the complexity of gene expression
patterns in the mammalian brain, Novartis Found. Symp. 239:129–146.
Leergaard, T. B., 2003, Clustered and laminar topographic patterns in rat cerebro–pontine
pathways, Anat. Embryol. 206:149–162.
Leergaard, T. B., Alloway, K. D., Mutic, J. J., and Bjaalie, J. G., 2000a, Three-dimensional topog-
raphy of corticopontine projections from rat barrel cortex: correlations with corticostriatal
organization, J. Neurosci. 20:8474–8484.
Leergaard, T. B., Alloway, K. D., Pham, T. A., Bolstad, I., Hoffer, Z. S., Pettersen, C., and
Bjaalie, J. G., 2004, Three-dimensional topography of corticopontine projections from rat
sensorimotor cortex: comparisons with corticostriatal projections reveal diverse integrative
organization, J. Comp. Neurol. 478:306–322.
Leergaard, T. B., and Bjaalie, J. G., 1995, Semi-automatic data acquisition for quantitative neu-
roanatomy. MicroTrace—computer programme for recording of the spatial distribution
of neuronal populations, Neurosci. Res. 22:231–243.
Leergaard, T. B., Bjaalie, J. G., Devor, A., Wald, L. L., and Dale, A. M., 2003, In vivo tracing of
major rat brain pathways using manganese-enhanced magnetic resonance imaging and
three-dimensional digital atlasing, NeuroImage 20:1591–1600.
Leergaard, T. B., Lakke, E. A., and Bjaalie, J. G., 1995, Topographical organization in the early
postnatal corticopontine projection: a carbocyanine dye and 3-D computer reconstruction
study in the rat, J. Comp. Neurol. 361:77–94.
Leergaard, T. B., Lyngstad, K. A., Thompson, J. H., Taeymans, S., Vos, B. P., De Schutter, E.,
Bower, J. M., and Bjaalie, J. G., 2000b, Rat somatosensory cerebropontocerebellar path-
ways: spatial relationships of the somatotopic map of the primary somatosensory cortex
are preserved in a three-dimensional clustered pontine map,J. Comp. Neurol. 422:246–
266.
Lillehaug, S., Oyan, D., Leergaard, T. B., and Bjaalie, J. G., 2002, Comparison of semi-automatic
and automatic data acquisition methods for studying three-dimensional distributions of
large neuronal populations and axonal plexuses, Network 13:343–356.
Lohmann, K., Gundelfinger, E. D., Scheich, H., Grimm, R., Tischmeyer, W., Richter, K., and
Hess, A., 1998, BrainView: a computer program for reconstruction and interactive visual-
ization of 3D data sets, J. Neurosci. Methods 84:143–154.
MacKenzie-Graham, A., Jones, E. S., Shattuck, D. W., Dinov, I. D., Bota, M., and Toga, A. W.,
2003, The informatics of a C57BL/6J mouse brain atlas, Neuroinformatics 1:397–410.
Malmierca, M. S., Leergaard, T. B., Bajo, V. M., Bjaalie, J. G., and Merchan, M. A., 1998,
Anatomic evidence of a three-dimensional mosaic pattern of tonotopic organization in
the ventral complex of the lateral lemniscus in Cat, J. Neurosci. 18:10603–10618.
Maunsbach, A. B., and Afzelius, B. A., 1999, Biomedical Electron Microscopy, San Diego: Academic
Press.
Maurin, Y., Banrezes, B., Menetrey, A., Mailly, P., and Deniau, J. M., 1999, Three-dimensional
distribution of nigrostriatal neurons in the rat: relation to the topography of striatonigral
projections, Neuroscience 91:891–909.
networks, Anat. Embryol. 204:303–317.
Nikundiwe, A. M., Bjaalie, J. G., and Brodal, P., 1994, Lamellar organization of pontocerebellar
neuronal populations. A multi-tracer and 3-D computer reconstruction study in the cat,
Eur. J. Neurosci. 6:173–186.
Paxinos, G., and Franklin, K. B. J., 2001, The Mouse Brain in Stereotaxic Coordinates, San Diego:
Academic.
Paxinos, G., and Watson, C., 1998, The Rat Brain in Stereotaxic Coordinates, San Diego: Academic
Press.
Paxinos, G., and Watson, C., 2005, The Rat Brain in Stereotaxic Coordinates, Burlington, MA:
Academic Press.
Pennisi, E., 2003, Neuroscience. Mapping the brain’s genes: a Microsoft dividend, Science
301:1642.
Pommert, A., Tiede, U., and Höhne, K. H., 2002, Volume visualization, In: Toga, A. W., and
Mazziotta, J. C. (eds.), Brain Mapping. The Methods, San Diego: Academic Press, pp. 707–
723.
Rademacher, J., Burgel, U., Geyer, S., Schormann, T., Schleicher, A., Freund, H. J., and Zilles,
K., 2001, Variability and asymmetry in the human precentral motor system. A cytoarchi-
tectonic and myeloarchitectonic brain mapping study. Brain 124:2232–2258.
Romeis, B., 1968, Mikroskopische Technik, München: R. Oldenbourg Verlag.
Ruigrok, T. J. H., 2004, Precerebellar nuclei and red nucleus, In: Paxinos, G. (ed.), The Rat
Nervous System, San Diego, CA: Elsevier Academic Press, pp. 167–204.
Schleicher, A., and Zilles, K., 1990, A quantitative approach to cytoarchitectonics: analysis of
structural inhomogeneities in nervous tissue using an image analyzer, J. Microsc. 157(Pt
3):367–381.
Schmitt, O., Homke, L., and Dumbgen, L., 2003, Detection of cortical transition regions uti-
lizing statistical analyses of excess masses, Neuroimage 19:42–63.
Schwarz, C., and Möck, M., 2001, Spatial arrangement of cerebro–pontine terminals, J. Comp.
Neurol. 435:418–432.
Schwarz, C., and Their, P., 1995, Modular organization of the pontine nuclei: dendritic fields
of identified pontine projection neurons in the rat respect the borders of cortical afferent
fields, J. Neurosci. 15:3475–3489.
Senft, S. L., 2002, Axonal navigation through voxel substrates: a strategy for reconstructing
brain circuitry, In: Ascoli, G. (ed.), Computational Neuroanatomy: Principles and Methods,
Totowa, NJ: Humana Press, pp. 245–270.
Shen, C., O’Brien, J. F., and Shewchuk, J. R., 2004, Interpolating and approximating implicit
surfaces from polygon soup, In: Proceedings of the ACM SIGGRAPH, Los Angeles, CA: ACM
Press.
Swanson, L. W., 1999, Brain Maps: Structure of the Rat Brain, Amsterdam: Elsevier.
Toga, A. W., 1990, Three-dimensional reconstruction, In: Toga, A. W. (ed.), Three-Dimensional
Imaging, New York: Raven Press, pp. 251–272.
Toga, A. W., and Banerjee, P. K., 1993, Registration revisited, J. Neurosci. Methods. 48:1–13.
Toga, A. W., and Thompson, P., 1999, An introduction to brain warping, In: Toga, A. W. (ed.),
Brain Warping, San Diego: Academic Press, pp. 1–26.
Van Essen, D. C., Drury, H. A., Dickson, J., Harwell, J., Hanlon, D., and Anderson, C. H., 2001,
An integrated software suite for surface-based analyses of cerebral cortex, J. Am. Med.
Inform. Assoc. 8:443–459.
Vassbø, K., Nicotra, G., Wiberg, M., and Bjaalie, J. G., 1999, Monkey somatosensory cerebro-
cerebellar pathways: uneven densities of corticopontine neurons in different body repre-
sentations of areas 3b, 1, and 2, J. Comp. Neurol. 406:109–128.
Visel, A, Thaller, C., and Eichele, G., 2004, GenePaint.org: an atlas of gene expression patterns
in the mouse embryo, Nucleic Acids Res. 32(Database issue): D552–D556.
Welker, C., 1971, Microelectrode delineation of fine grain somatotopic organization of (SmI)
cerebral neocortex in albino rat, Brain Res. 26:259–275.
Woods, A. E., and Ellis, R. C., 1994, Laboratory Histopahtology: A Complete Reference, Orlando, FL:
Churchill Livingstone.
Wree, A., Schleicher, A., and Zilles, K., 1982, Estimation of volume fractions in nervous tissue
with an image analyzer, J. Neurosci. Methods.6:29–43.
tional anatomical analysis of the basal forebrain corticopetal system, In: Ascoli, G. (ed.),
Computational Neuroanatomy: Principles and Methods, Totowa, NJ: Humana Press, pp. 171–
197.
18
Atlases of the Human Brain:
Tools for Functional
Neuroimaging
KATRIN AMUNTS and KARL ZILLES
INTRODUCTION
CLASSICAL BRAIN ATLASES
Brodmann’s Map
Other Historical Maps
THREE-DIMENSIONAL ANATOMICAL BRAIN ATLASES
The Stereotaxic Atlas of Talairach and Tournoux
Anatomical Maps and Atlases Based on Other Modalities
Fiber Tract Mapping
The Human Brain Atlas of the International Consortium of
Human Brain Mapping as an Example of a Multimodal Brain
Imaging Database
INTERSUBJECT VARIABILITY IN BRAIN ANATOMY
Variability in Brain Macroscopy
Microstructural Variability
Cortical Areas and Macroanatomical Landmarks
Image Registration Techniques to Eliminate Macroanatomical
Variability
STANDARD REFERENCE BRAINS
Individual Reference Brain
Average Brain
Alternative Approaches Based on Volume Data
Surface-Based Atlases and Flat Maps
MICROSTRUCTURAL LOCALIZATION OF NEURAL FUNCTIONS
KATRIN AMUNTS • Institute of Medicine, Research Center Jülich, D-52441 Jülich,

Germany; Department of Psychiatry and Psychotherapy, RWTH Aachen University, Pauwelsstr.
30, D-52074 Aachen, Germany KARL ZILLES • Institute of Medicine, Research Center
Jülich, D-52441 Jülich, Germany; C. and O. Vogt Institute for Brain Research, Heinrich Heine
University Düsseldorf, Universitätsstr. 1, D-40225 Düsseldorf, Germany; Brain Imaging Center
West, Research Center Jülich, D-52441 Jülich, Germany
566
HUMAN BRAIN ATLASES 567
SUMMARY OF ADVANTAGES/LIMITATIONS OF
CYTOARCHITECTONIC PROBABILISTIC MAPS
Advantages
Limitations
APPENDIX
Method of the Generation of Cytoarchitectonic Probabilistic Maps
Useful Links and Atlas Tools
REFERENCES
Abstract: Human brain atlases are frequently used tools for the analysis of data from
functional imaging and neurophysiology studies. This chapter briefly reviews histor-
ical, two- and three-dimensional printed and electronic atlas systems. It emphasizes
several key aspects of such atlases: spatial relationships of macro- and microstruc-
tures, their intersubject variability, definition of reference brains and spatial refer-
ence systems, linear and nonlinear registration procedures of data sets of individual
brains to reference brains, and multimodal comparisons of structural and functional
data in stereotaxic space. The Appendix outlines the method of generation of prob-
abilistic cytoarchitectonic maps, and provides addresses of some of the electronic
human brain atlases and software.
Keywords: cerebral cortex, cytoarchitecture, human brain mapping, neuroanatomy,

probability maps
I. INTRODUCTION
Modern functional neuroimaging and neurophysiological techniques

such as functional magnetic resonance tomography (fMRI), positron emis-
sion tomography (PET), receptor PET (rPET), single photon emission
computed tomography (SPECT), and magnetoencephalography (MEG)
provide detailed topographical information on human brain activity.
Brain activity is hereby estimated via the blood-oxygen-level-dependent
(BOLD) signal, blood flow and metabolism, distribution of receptor-binding
sites, and neuronal signaling, respectively. The spatial resolution of these
methods is in the millimeter range. In most cases, high-resolution anatom-
ical MR images of the same brain are coregistered to enable topographical
interpretation of imaging data. The spatial resolution of anatomical MRI en-
ables a localization of the detected brain activities at a macroscopical level,
e.g., gyri, sulci, and subcortical nuclei (e.g., basal ganglia, thalamus).
The size, precise location, shape, and extent of macroanatomical land-
marks, however, differ significantly between the subjects as well as between
the hemispheres (Duvernoy, 1991; Ono et al., 1990; Paus et al., 1996; Toga
and Thompson, 2003). Brain macroscopy is not a stable feature over the
whole life span, but changes during development and aging as well as in
neurological and psychiatric disorders (Sowell et al., 2003; Thompson et al.,
1998; Thompson and Toga, 2002). In order to compare anatomical and func-
tional imaging data between different subjects (e.g., in group studies), it is
568 KATRIN AMUNTS and KARL ZILLES
necessary to transform the data into a common reference space. This space
can be represented by an individual brain or an “average” brain. A com-
mon reference space also opens the possibility to integrate data concerning
the microscopical structure of postmortem brains such as cytoarchitecture.
Postmortem brains can be analyzed with maximal spatial resolution in the
micrometer range. In contrast to the macroanatomical pattern of the brain
with its highly variable sulci and gyri, microstructural parcellations, e.g.,
cytoarchitectonic areas, seem to be more closely related to brain function
(Amunts et al., 2004; Binkofski et al., 2000; Bodegard et al., 2000a; Geyer et al.,
1996, 2001; Horwitz et al., 2003; Larsson et al., 1999, 2002; Mazziotta et al.,
2001a, b; Naito et al., 1999, 2000; Roland and Zilles, 1994, 1998; Uylings et al.,
2000; Young et al., 2003; Zilles et al., 1995).
Cytoarchitectonic parcellations represent one aspect of cortical (and sub-
cortical) organization. These parcellations, however, can result in quite dif-
ferent maps, with differences in the number of areas, nomenclature, and
hierarchies. The famous Brodmann map represents only one (well-known)
of many (less well-known) maps. There is no general agreement about a
“gold standard” of a cytoarchitectonic map. An extreme example of a cy-
toarchitectonic map is that of Bailey and von Bonin, who asked whether
there is any objective basis for a detailed cytoarchitectonic map at all. They
came to the final conclusion that “. . . vast areas are as closely similar in struc-
ture as to make any attempt at subdivisions unprofitable, if not impossible.”
As a consequence, their cytoarchitectonic map is based only on a parcella-
tion into a few main types of cortical regions (Bailey and von Bonin, 1951).
The other extreme can be seen in a map of Gerhardt who distinguished
between “true” borders separating two different areas, gradual borders and
likely borders within one area (Gerhardt, 1940). This approach resulted in a
vast number of areas of the parietal cortex, which has not been reproduced
until now (Zilles and Palomero-Gallagher, 2001). It seems to be clear, at least,
that the problem of a reproducible and observer-independent definition of
areal borders plays a key role (see also section “Method of the Generation
of Cytoarchitectonic Probabilistic Maps”).
Would it be, theoretically, sufficient to know the true cytoarchitectonic
parcellation of the brain in order to disclose the relationship between brain
structure and function? Is a cytoarchitectonic area the same as a cortical
area? For some cortical regions, in particular primary sensory and motor
area, this assumption seems to hold true. For many other areas, including
higher associative ones, a correlation between a certain function and cy-
toarchitecture has not yet demonstrated. One aspect of this failure may be
related to the nonadequate characterization of a brain function at the level
of a cortical area. “Language” is clearly not a function of one cortical area,
but includes different aspects, e.g., phonological and semantic processing,
prosody, syntax, etc. involving different brain regions.
What, however, is a cortical area? This question provides an important
argument for multimodal architectonic mapping, since not all subparcel-
lations of the cerebral cortex constitute a cortical area. For example, the
subdivision of areas V1 and V2 into blob and interblob regions (Livingstone
and Hubel, 1987; Roe and Tso, 1995; Wong-Riley et al., 1993) reflects differ-
ences in color and orientation selectivity (V1) and receptive field properties
(V2), but these subdivisions do not constitute cortical areas. Additional ex-
amples are the somatotopy of the motor and somatosensory cortex, the
tonotopical organization of the auditory cortex, each of which represents
a functional segregation without representing an architectonic entity. The
isolated analysis of only one aspect of cortical organization, without con-
sideration of other mapping techniques, would lead to an over-parcellation
of the cerebral cortex. Instead, we propose a multimodal approach which
avoids this problem by providing an overview of the different hierarchical
levels (e.g., cytoarchitectonic or receptor architectonic families of cortical
areas) of the cortical organization.
The close relationship between brain function, connectivity, and architec-
ture has been demonstrated in combined histological and electrophysiolog-
ical experiments in monkeys (Luppino et al., 1991; Matelli et al., 1991; Tanji
and Kurata, 1989). It is, therefore, possible to relate functional activations
to microstructurally defined parcellations such as cytoarchitectonic areas of
the cerebral cortex. Moreover, it has been shown that maps based on dif-
ferent histological and histochemical techniques frequently show a perfect
spatial coincidence of areal borders; thus, corroborating the position of an
areal border by multimodal imaging including receptor autoradiography
(Zilles et al., 2002a, b). Moreover, since a single receptor may not reveal all
borders demonstrated by other markers, this finding can be used to define
a family of neurochemically related areas by studying the regional pattern
of one transmitter receptor and comparing its distribution with the maps
revealed by other receptors or cytoarchitecture (see section “Microstruc-
tural Localization of Neural Functions”). We think that such a multimodal
concept of cortical mapping improves and supplements classical cytoarchi-
tectonic analysis.
An important perspective for a functionally relevant architectonical par-
cellation of the cortex arises from the integration of architectonic maps
with recent PET, fMRI, and MEG studies in a common spatial reference sys-
tem (see section “Microstructural Localization of Neural Functions”) and
databases (see sections “The Human Brain Atlas of the International Consor-
tium of Human Brain Mapping as an Example of a Multimodal Brain Imag-
ing Database,” “Individual Reference Brain,” and “Surface-Based Atlases and
Flat Maps”). Thus, the analysis and evaluation of structural–functional rela-
tionship is a major goal of microstructural brain atlases. We here focus on
such approaches based on microscopical data.
II. CLASSICAL BRAIN ATLASES
A. Brodmann’s Map
One of the most widely used brain atlases is the cytoarchitectonic map
of Brodmann (1909). Brodmann published descriptions of cortical areas
based on cytoarchitectonic studies of the human brain. The book has been
translated into English, and thus, became accessible to a broader readership
(Brodmann, 1994). Brodmann subdivided the cortical surface into more
than 50 areas. The areas were formally numbered in the sequence of their
appearance in horizontal sections. This atlas did not consider subcortical
nuclei and fiber tracts. He did not study the functions of the cortical areas,
but provided some functional interpretations on the basis of the knowledge
available at the beginning of the 20th century. Although Brodmann showed
histological sections with labeled borders between cortical areas in his book
and original articles, his final map is a schematic drawing of the left lateral
and right medial surfaces of a “typical” brain. He was aware that this map was
a simplification of his observations and noticed that “. . . a schematic drawing
can reflect only the major spatial relationships, and therefore, precise topo-
graphical associations (i.e., between sulci and areal borders; remark of the
authors) cannot be considered in general or only in a distorted manner;
this is true in particular for all those cortical regions which have borders in
the neighborhood of sulci and those regions which are located in the depth
of such a cortical region” (Brodmann, 1908).
Therefore, it is not possible to conclude from Brodmann’s schematic
drawing whether an actual areal border is located on the walls of a sulcus, or
coincides with the bottom of the sulcus (Fig. 18.1). Even if one would accept
this uncertainty, the identification of areal borders by sulcal landmarks can
be very problematic, since many sulci are highly variable, and a small sulcus
shown in Brodmann’s map may not be present in another individual brain.
Figure 18.1. Historical cortical maps of (A) Brodmann (1909), (B) the Russian school
(Sarkisov et al., 1949), (C) Cecile and Oskar Vogt (1919), and (D) Bailey and von
Bonin (1951). Note the different shapes and sulcal patterns, as well as the different
number of areas.
B. Other Historical Maps
Although various complete cytoarchitectonic and myeloarchitectonic

maps of the human cerebral cortex have also been published by other
authors (e.g., Bailey and von Bonin, 1951; Campbell, 1905; Cecile and Oskar
Vogt, 1919; Elliot Smith, 1907; Kleist, 1934); the Russian school (Sarkisov
et al., 1949); and von Economo and Koskinas, 1925, these maps did not
achieve a comparably wide distribution and level of acceptance as did Brod-
mann’s map. The comparison between these maps reveals differences with
respect to (i) the underlying method of parcellation (e.g., cytoarchitecture,
myeloarchitecture, unstained sections), (ii) the pattern of sulci and gyri of
the particular brain used for the atlas, (iii) the location of areal borders with
respect to sulci and gyri, and (iv) the number of the areas. An extreme posi-
tion was held by Bailey and von Bonin (1951), who identified only four main
cytoarchitectonic types in the isocortex. The Vogts, in contrast, found ∼150
different myeloarchitectonic fields. They and their fellows defined dozens
of transitional forms (Gerhardt, 1940; Riegele, 1931). In addition to differ-
ences between individual brains, one important reason of this discordances
is presumably related to differing and highly observer-dependent criteria
used for the delineation of the areas and the definition of their borders.
This problem has only been mastered during the last years when statistical
tools of cytoarchitectonic analysis have been introduced for an observer-
independent definition of cortical borders (Annese et al., 2004; Schleicher
et al., 1998, 1999, 2000; Schmitt et al., 2003).
Starting with the development of techniques for functional imaging of the
living human brain, another disadvantage of these classical printed, two-
dimensional atlases became evident: They were not compatible with the
three-dimensional imaging data. Moreover, the drawings of brains in the
different historical maps showed differences in orientation, shape and size,
and often they did not disclose all aspects of the brain surface in an optimal
way (e.g., the insula, or the auditory cortex on the dorsal surface of the
superior temporal gyrus). These problems, and the need of neurosurgeons
to identify in space brain lesions, epileptic foci or targets for stereotaxic
surgery, stimulated the development of a stereotaxic atlas systems, e.g., the
atlas of Talairach and Tournoux (1988). Although originally published as a
book, this atlas provides stereotaxic coordinates of cortical areas.
III. THREE-DIMENSIONAL ANATOMICAL BRAIN ATLASES
A. The Stereotaxic Atlas of Talairach and Tournoux
The atlas of Talairach and Tournoux (1988) provides two choices of defin-
ing the spatial position: (i) x-, y -, and z-coordinates in millimeter represent-
ing the distances from a line connecting the anterior and the posterior com-
missures (AC–PC line), the vertical line through the anterior commissure,
and the midline between the hemispheres, and (ii) the concept of propor-
tionality realized as a proportional grid system, where the dimensions of
the grid system vary with the dimensions of the major axes of individual
brains. Thus, each point of this atlas brain is defined by three coordinates
(x, y , and z). The atlas brain is a single human postmortem brain, which
has been sectioned sagittaly. From this series of sagittal sections, two further
series of sections were reconstructed in the frontal and horizontal planes
by point-to-point projections. The borders of fiber tracts, subcortical nuclei,
and cortical areas were traced by comparison of macroanatomical features
of the atlas brain with previous descriptions, e.g., by Brodmann (1909) for
the cortical areas, and by Walker (1938), for the thalamus as well as by di-
rect observations in the actual postmortem atlas brain (e.g., olfactory tract).
It is notable, that the delineations of the cortical areas are not based on
histological analysis of the atlas brain, but inferred as the authors empha-
size by “the transfer of the cartography of Brodmann usually pictured in
two-dimensional projections,” [which] “sometimes possesses uncertainties”
(Talairach and Tournoux, 1988).
Although originally created for neurosurgery, radiology, and neurology,
the atlas rapidly gained an increasing importance for functional imag-
ing studies such as PET and fMRI, since it is assumed that its applica-
tion enables the comparison of brains and groups of brains within and
between studies (Fox et al., 1985; Friston, 1997). The atlas of Talairach
and Tournoux is available in a computerized version, the “Talairach Dae-
mon” (Lancaster et al., 2001). It is also implemented as an anatomical
reference system in various software packages, e.g., the widely used SPM
software www.fil.ion.ucl.ac.uk/spm and the International Consortium of
Human Brain Mapping (ICBM) viewer http://www.bic.mni.mcgill.ca/cgi/
icbm view/ (Fig. 18.2).
As compared to the classical architectonic maps, the atlas of Talairach and
Tournoux
1. offers three-dimensional information on the topography of cortical

areas, subcortical nuclei, and fiber tracts,
2. introduces a stereotaxic orientation using the AC–PC line as a reliable
and easy-to-find anatomical landmark,
3. proposes a proportional grid system in order to compensate intersub-
ject differences in brain shape and size, and
4. covers the whole cortical surface, several subcortical nuclei, and fiber
tracts.
At the same time, however, several important problems remained

unsolved:
1. The atlas is not based on microstructural analyses of the atlas brain,

but on a vaguely defined method of transfer of microscopical (archi-
tectonic) information about cortical areas from the classical, schematic
drawings of Brodmann to the atlas brain based on macroanatomical
Figure 18.2. ICBM viewer (http://www.bic.mni.mcgill.ca/cgi/icbm view/). Upper

row: average of 27 T1-weighted scans of an individual brain (“MNI single-subject
template”); lower row: from Talairach and Tournoux (1988).
similarities between brains. Such an approach, however, assumes an in-

variable correspondence between macroanatomical and microscopical
features, which is true for the borders of only a few cortical areas (see
below), but which cannot be accepted for the majority of other regions
(e.g., the so-called higher associative areas of the parietal lobe).
2. The delineations of cortical areas are based on Brodmann’s map.
During the last years, however, it has been shown (see below) that
Brodmann’s map provides an oversimplified parcellation in some brain
regions, or is insufficient in other regions. This is particularly impor-
tant, e.g., in the extrastriate, visual region, and the intraparietal sulcus,
where functional imaging studies and observations in nonhuman pri-
mates have already shown a lack of correspondence with the structural
parcellations performed by Brodmann.
3. Borders between cortical areas are not indicated.
4. Intersubject variability in brain size, shape, and sulcal pattern as well
as in microstructure is not considered (see below).
5. Interhemispheric asymmetry is not taken into account. Only the cor-
ticospinal tract “is represented on the right and the left in a different
manner” in correspondence to the variations of the central sulcus in
the two hemispheres (Talairach and Tournoux, 1988). The remaining
structural asymmetries are not considered.
6. The atlas is based on a single postmortem brain, which shows
macroanatomical deformation in comparison to in vivo brains. It ap-
pears, for example, more flat than most of the in vivo brains.
B. Anatomical Maps and Atlases Based on Other Modalities
Although the above-mentioned atlases use cytoarchitectonic information

in nearly all cases, brain mapping is not limited to this modality. Maps can
been created on the basis of the myeloarchitecture (Elliot Smith, 1907;
Mai et al., 2004; Vogt and Vogt, 1919), regional and laminar distribution
of receptor binding sites of different neurotransmitters such as the glu-
tamatergic, GABAergic, cholinergic, adrenergic, dopaminergic receptors
(Zilles et al., 1995, 1988b, 2002a, b, 2003; Zilles and Schleicher, 1993, 1995),
immunohistochemical markers (Bidmon et al., 1997; Geyer et al., 2000a;
Hayes and Lewis, 1992; Majocha et al., 1985; Zilles et al., 1991a), or gene ex-
pression, e.g., the Allen Brain Atlas (http://www.brainatlas.org/default.asp
and http://www.brain-map.org/index.jsp). The Allen Brain Atlas aims to
create a detailed, cellular-resolution, genome-wide map of gene expression
in the mouse brain. Gene expression tomography and voxelation in the
mouse and the human brain have been suggested in order to provide a
high-throughput approach to map regional gene expression patterns in the
brain (Singh and Smith, 2003).
The atlas of the human brain of Mai et al. (2004) combines a macroscopic
part with a microscopic (myeloarchitectonic) part. The latter comprises 69
serial sections of one hemisphere. The macroscopic part is based on three
brains, which were scanned using MRI before histological processing. In
contrast to many other maps, this atlas shows a detailed parcellation of
subcortical nuclei.
C. Fiber Tract Mapping
The development of diffusion tensor imaging and fiber tractography en-

able the mapping of fiber tracts in the living human brain (Barker, 2001;
Basser and Jones, 2002; Coenen et al., 2001; Conturo et al., 1999; Ito et al.,
2002; Jones et al., 2002; Krings et al., 2001; Le Bihan et al., 2001; Mori and
van Zijl, 2002; Parker et al., 2002; Poupon et al., 2000; Powell et al., 2004;
Turner et al., 1991). Diffusivity is known to depend upon the orientation of
the principal axes of fiber tracts (for a review see, e.g., Basser and Jones,
2002). Fiber tract trajectories can be generated from a fluid velocity field.
Several fiber tracts of the brain stem and spinal cord (Assaf et al., 2000; Clark
et al., 1999; Wheeler-Kingshott et al., 2002) have been traced. This method
has also been successfully applied in neurological and psychiatric disorders,
e.g., in multiple sclerosis (Assaf et al., 2002) and schizophrenia (Ardekani
et al., 2003; Foong et al., 2002; Iwasawa et al., 1997). Artifacts may arise, how-
ever, when discrete, coarsely sampled, noisy, voxel-averaged direction field
data are used. Further problems may occur if fiber tracking is applied to in-
coherently organized fiber tract (Basser and Jones, 2002) or if fibers merge,
branch, or cross each other (Le Bihan et al., 2001). It should be kept in
mind that all presently available variations of this technique do not reveal
anatomical connectivity in a strict sense, since this goal would require the
demonstration of synaptic connectivity at an electron-microscopical level in
combination with axonal tracing.
Although in vivo tracing studies have frequently been performed in ex-
perimental animals, this approach is not possible in the living human brain.
Tracing studies in adult postmortem, pathologically altered brains have been
reported using various staining techniques for degenerating nerve fibers
(Clarke, 1994; Clarke and Miklossy, 1990; Di Virgilio and Clarke, 1997;
Miklossy and van der Loos, 1991; Mufson et al., 1990; Wiesendanger et al.,
2004) or polarized light microscopy (Axer and Keyserlingk, 2000). Map-
ping of fiber tracts has also been performed in immature brains, where the
early myelinating fiber tracts can be easily distinguished from the unstained,
later myelinating fiber tracts (Flechsig, 1920; Yakovlev and Lecours, 1967).
All these studies have contributed important data on the topography of
fiber tracts and their maturation during ontogeny. They did not, however,
provide stereotaxic information concerning the course and extent of fiber
tracts. Moreover, they did not consider intersubject variability and inter-
hemispheric asymmetry, which are important aspects of brain atlases (see
below).
Using a modified myelin stain, fiber tracts have been mapped in his-
tological sections of 10 adult human brains, and probabilistic, three-
dimensional maps of the optic radiation (Bürgel et al., 1999), corticospinal
tract (Rademacher et al., 2001a), and the auditory system (Rademacher
et al., 2002) have been published Bürgel et al., 2005 (Fig. 18.3). Compared
to tracing studies in pathologically altered brains or in fetal brains, these
Figure 18.3. Thirty percent probability map of the optic radiation (red) and lateral
geniculate body (blue). The fiber tract and the geniculate body have been traced
in histological serial sections (thickness 20 µm) stained using the modified myelin
stain of Heidenhain–Wölcke (Bürgel et al., 1997, 1999) of 10 human postmortem
brains and registered in the stereotaxic MNI reference space (Bürgel et al., 2005).
maps display the normal and adult conditions. They took advantage of the
high microscopical resolution, which enabled to trace fibers even if fibers
abruptly change their direction, cross, or touch other fiber tracts (“kissing
fibers”). The probabilistic fiber maps can also not reveal synaptic connec-
tivity. These maps can be used, however, for evaluation of data from MR
tractography as an independent measure.
An interesting extension of tractography has recently been developed.
Thalamic subdivisions were classified according to the cortical region with
which they show the highest connection probability (Behrens et al., 2003;
Johansen-Berg et al., 2005). A further application led to the identification
of the connectivity of SMA and pre-SMA (Johansen-Berg et al., 2004). This
approach enabled the generation of population-based maps. The proposed
parcellations are comparable to those found by invasive studies in the non-
human primate brain. Parcellations of cortical areas and subcortical nu-
clei based on MR tractography is complementary to cyto-, receptor-, and
myeloarchitectonic mapping, and add new information about fiber tracts
and connectivity.
D. The Human Brain Atlas of the International Consortium of

Human Brain Mapping as an Example of a Multimodal
Brain Imaging Database
The primary goal of the International Consortium of Human Brain Map-

ping (ICBM) project (http://www.loni.ucla.edu/ICBM/) is the continuing
development of a probabilistic reference system for the human brain
(Mazziotta et al., 1995, 2001a). It comprises four core sites: University of
California, Los Angeles (UCLA), Montreal Neurologic Institute (MNI), Uni-
versity of Texas at San Antonio, and Institute of Medicine, Research Center
Jülich/Heinrich Heine University Düsseldorf. In addition, data acquisition
sites in Asia and Europe contribute to this international consortium. Various
modalities ranging from 3D tomographic images of in vivo and postmortem
brains to histological preparations for cyto- or myeloarchitectonic observa-
tions, quantitative in vitro receptor autoradiography, and functional imaging
using PET and fMRI have been included both from normal controls, differ-
ent developmental stages, and pathologically altered brains (Toga, 2003).
The atlas is based on the MNI template brain as standard reference space.
It includes various data sets from ∼7000 individuals.
IV. INTERSUBJECT VARIABILITY IN BRAIN ANATOMY
Human brains show an enormous intersubject variability of their anatomy.

Variability comprises macroanatomical features (size, shape, sulcal pattern),
the microstructure (e.g., cyto-, receptor-, and myeloarchitecture), and the
relationship between microstructurally defined cortical areas and macro-
scopical landmarks.
Figure 18.4. Three-dimensional reconstruction of the surface of three right-handed

males (age between 30 and 40 years). Note the differences in the sulcal pattern
between the brains.
A. Variability in Brain Macroscopy
Differences in brain shape between different ethnic populations or in

absolute brain size within each gender as well as between female and male
brains have been frequently described (Steinmetz et al., 1995; Zilles et al.,
1997, 2001). Differences in shape between both hemispheres have been
shown, e.g., for the occipital lobe (Falk et al., 1991; LeMay and Kido, 1978;
Zilles et al., 1996) and the central region (Amunts et al., 2000a).
Three-dimensional reconstructions of human brains using MRI at a reso-
lution of ∼1 mm × 1 mm × 1 mm voxel size demonstrate the variability in
sulcal patterns (Fig. 18.4). The degree of gyrification also differs consider-
ably between subjects (Zilles et al., 1988a). Although the large primary sulci,
which develop early during ontogeny, are relatively constant in their pres-
ence and shape, the smaller secondary and tertiary sulci are highly variable.
Sulcal variability (Thompson et al., 1996) includes
1. the presence of a sulcus, e.g., the diagonal sulcus is present in ∼50%

of the hemispheres (Amunts et al., 1999);
2. the number of segments of a sulcus, e.g., the precentral sulcus can be
split into two to four segments (Ono et al., 1990);
3. the number and course of side branches and connections, e.g., the
central sulcus can show up to four side branches and connections to
either the postcentral or precentral sulci (Ono et al., 1990);
4. the spatial location of sulci (Juch et al., 2005);
5. shape of the endings, the width of sulci, and the depths (Van Essen,
2005).
The macroanatomical variability can be a confounding factor in many

neuroimaging studies, e.g., group studies using fMRI or PET, if the aim
is an analysis of corresponding brain regions. Various image registra-
tion tools (see below) can be applied in order to measure or eliminate
Figure 18.5. Intersubject differences between the brains of 24 healthy normal volun-
teers and an individual reference brain (all right-handed males). The 24 T1-weighted
MR data sets have been warped to the individual target brain. As a first step, a linear
affine transformation (scaling, rotation, translation) has been performed, which also
normalizes the spatial orientation of all individual brains. As a next step, a nonlin-
ear elastic deformation (“warping”) has been applied. For each brain, deformation
fields were calculated. The vectors of each of these fields indicate the direction and
length of the elastic deformation (not the absolute distance between corresponding
voxels of the original and the target brains) in every single voxel (Pieperhoff et al.,
2005). The 24 individual deformation fields were then averaged, and the averaged
vector lengths were color coded. Red corresponds to a vector length of 5 mm, orange
of 4 mm, yellow of 3 mm, green of 2 mm, and blue of 1 mm. Large differences in
sulcal pattern and regional brain shape between the target brain and the sample
correspond to large vector lengths, that is, the target brain differs maximally from
the sample in those regions which are shown in red, and less in regions shown in
green and blue. The figure demonstrates larger differences in shape in the occipital
pole and lobe between the target brain and the actual sample as compared to the
frontal lobe (Pieperhoff and Amunts, unpublished observations).
macroanatomical differences between individual brains. Image registration

is, therefore, a necessary prerequisite for brain atlases which take the in-
terindividual variability into account. Registration tools include (i) linear,
affine transformation which removes and normalizes differences in shape
and size, but retains left–right asymmetries as well as the individual sulcal
pattern, and (ii) nonlinear transformations. After an ideal, nonlinear trans-
formation of an MR data set of an individual brain to the reference brain of
an atlas, brains are identical in size, shape, sulcal pattern, and asymmetry.
The information about the macrostructural, intersubject variability in a
sample or the difference between an individual brain and the reference
brain can be stored for further analysis and applications as “deformation
field” (Fig. 18.5).
B. Microstructural Variability
The giant pyramidal cells (Betz-cells) in Brodmann’s area 4 (primary

motor cortex) may vary between 60 and 120 µm in height and 30 and 60 µm
in width (von Economo and Koskinas, 1925). This is one of the numerous
examples of intersubject variability in microstructure at the cellular level.
Here, we will focus on the microstructural variability at the level of corti-
cal areas (e.g., size) and their laminar pattern (cytoarchitecture). Whereas
some brains show a clearly visible laminar pattern with considerable differ-
ences in cell packing density between the layers of a cytoarchitectonically
defined area, a much more blurred laminar pattern with minor differences
in cell packing density between the layers can be found in the same cor-
tical area of other brains. These cytoarchitectonic differences can only be
analyzed using quantitative microscopical techniques, e.g., by registration
of the volume fraction of cell bodies (gray level index, GLI; Schleicher and
Zilles, 1990) vertical to the cortical surface between this surface and the
cortex/white matter border. The results can be visualized as a profile curve
which describes quantitatively the specific cytoarchitecture of an actual area
in a single brain. The shapes of the profiles of the same area but from
different brains can be parameterized as feature vectors, and the intersub-
ject variability of these vectors can be analyzed using multivariate statistics.
How large is the difference in cytoarchitecture between two different areas
in one brain as compared to differences in cytoarchitecture of the same area
between different brains? For example, it has been found that cytoarchitec-
tonic differences in area 44 (one part of Broca’s speech region) between 10
postmortem brains are greater than the mean difference between area 44
and 45 (other part of Broca’s speech region) in the same sample of brains
(Amunts et al., 1999).
Another aspect of intersubject variability is represented by the consid-
erable variability in absolute and relative (with respect to the whole brain
volume) size of a cytoarchitectonic area (Filimonoff, 1932; Kononova, 1935;
Stensaas et al., 1974). This type of variability has a major impact on the con-
struction of any brain atlas designed as a tool for the analysis of functional
imaging data or stereotaxic neurosurgery, since the degree of variability dif-
fers between brain regions: whereas areas 44 and 45 of Broca’s region vary
in their absolute volumes up to a factor of 10 (Amunts et al., 1999), the
hippocampus and the amygdala vary “only” by a factor of 2 (Amunts et al.,
2005). Thus, the analysis of the intersubject variability is a major task for
future brain atlases.
C. Cortical Areas and Macroanatomical Landmarks
The spatial positions of a cortical area and macroanatomical landmarks

vary independently between brains (Zilles et al., 1997). A relatively strong
association between macroanatomical landmarks and cortical areas has been
found for primary cortical areas (Rademacher et al., 1993). The primary
motor cortex (Brodmann’s area 4) is always located in the anterior bank
of the central sulcus, and the somatosensory cortex is consistently found in
the posterior bank of this sulcus. The calcarine sulcus indicates the location
of the primary visual cortex (areas V1 or Brodmann’s area 17), and the
primary auditory cortex is found on Heschl’s gyrus (Fig. 18.1). However,

the extent of area 4 in anterior direction (to premotor area 6) cannot be
defined by any macroanatomical landmark. The same is true for the dorsal
and ventral borders of area 17 (to the extrastriate cortical areas). For the
majority of cortical areas, a sufficiently precise spatial relationship between
the macroanatomical landmarks and the border of cortical areas cannot be
established.
These findings do not imply a complete lack of spatial correlation between
the location of a cortical area and gyri or sulci. In some cortical regions, gyri
can be good indicators of the approximate location of a cortical area. For
example, major parts of area 44 are consistently found on the free surface
of the opercular part of the inferior frontal gyrus, and area 45 occupies
usually the triangular part of this gyrus. However, when moving from the
free surface of the triangular part into the depth of the horizontal branch
of the Sylvian fissure and further to the surface of the orbital part of the
inferior frontal gyrus, area 45 may reach the bottom of the sulcus and the free
surface in some hemispheres, but not in others (Amunts et al., 1999, 2004).
Consequently, the variability in the extent of one and the same area is low
at the free surface of the opercular part, but high in the sulci. Additionally,
these sulci may vary between brains (see also section “Variability in Brain
Macroscopy”).
It can be concluded, therefore, that any brain atlas based exclusively
on macroanatomical landmarks for the identification of the position and
extent of cortical areas may cause significant misinterpretations, e.g., of
the cytoarchitectonic identity of activated cortical regions found in func-
tional neuroimaging studies. There are two possible solutions of this pro-
blem:
1. Cyto- or myeloarchitectonic parcellations of the same living brain

which has been studied by functional neuroimaging. This is only fea-
sible if the parcellation can be performed by ultrahigh-resolution
anatomical MR imaging. First attempts have been made already for
selected regions of the cortex (Clark et al., 1992; Eickhoff et al., 2005d;
Fatterpekar et al., 2002; Kruggel et al., 2003; Walters et al., 2003).
2. A probabilistic approach, which defines for each voxel of a reference
space (or brain) the probability with which an actual cortical area,
subcortical nucleus, or fiber tract is present in this voxel. As a con-
sequence, it will not be possible to define unambiguously a certain
voxel at a given stereotaxic location as the representation of one single
anatomical structure (Fig. 18.6). Any voxel may have probabilities dif-
ferent from zero to belong to several cortical areas. This may not be a
confounding factor in regions covered by a large single cortical area. In
other regions, e.g., at the meeting point of several areas such as in the
central operculum (primary and secondary motor cortex, primary and
secondary somatosensory cortex, insula), a single voxel may represent
several cortical areas with relatively similar probabilities.
Figure 18.6. Cytoarchitectonic probabilistic maps of area 44 (left) and 45 (right).

Orientation according to the AC–PC line. Location at z = 15 (i.e., 15 mm above the
intercommissural line). Left hemisphere is left in the image. The probability is color
coded. Yellow corresponds to an overlap in 7 out of 10 brains, orange to an overlap
of 8 brains.
D. Image Registration Techniques to Eliminate

Macroanatomical Variability
Image registration techniques can be used to transform one MR data set

into another. Widely used synonyms of this procedure are “warping,” “image
transformation,” “deformation,” “spatial normalization,” and “alignment.”
The data sets may come from one and the same subject (which has been
scanned at several different occasions, e.g., during aging), or from different
subjects. We here focus on registration tools which are used to transform
MR data sets of different human brains to a common reference brain.
In this latter case, registration minimizes the macroanatomical differences
between brains, but does not eliminate microstructural differences. A ref-
erence brain can be either an individual brain (Roland and Zilles, 1994),
a digital brain phantom (Collins et al., 1998), or an average brain (Collins
et al., 1994; Evans et al., 1992, 1993).
Registration tools can be classified by the degrees of freedom, i.e., the
number of free parameters, which define the registration and must be cal-
culated for each individual brain. Here, we provide a few examples of fre-
quently applied registration tools.
1. Affine Transformation
The rigid-body transformations are a subset of affine transformations.

They have six degrees of freedom: three angles of rotation and three com-
ponents of translation. Rigid-body transformation may change the location
of a brain data set in space. This is necessary, for example, in order to align
different brains to the AC–PC line. This type of transformation does not
change the shape and size of the brains.
Other affine transformations have 12 degrees of freedom. In addition
to the changes which can be performed by the rigid-body transformation,
these transformations enable linear changes in brain size and shape due
to scaling and sheering. They are frequently used as the initial step for
a subsequent nonlinear transformation (see section “Nonlinear, Nonrigid
Transformations”).
2. Transformations of Medium Complexity
These transformations have several hundred up to several thousand

degrees of freedom. Examples of this type of registration tools have
been implemented in SPM (http://www.fil.ion.ucl.ac.uk/spm) and AIR
(http://air.bmap.ucla.edu). The registration tool of SPM calculates a spa-
tial transformation as linear combination of a given set of basic func-
tions (Ashburner and Friston, 1999). Hereby, the coefficients of these lin-
ear combinations have to be determined. The tool implemented in AIR
defines the transformation of a voxel using higher polynomials (Woods
et al., 1998).
These tools are frequently used for the spatial normalization of functional
imaging data. They do not lead to a precise match of the anatomical MR data
of different brains. Such a precise match is often not required due to the
relatively low spatial resolution of functional imaging data. Moreover, SPM
offers an average reference brain as a target brain for this transformation.
Such average brain shows much less anatomical detail than a single subject
brain.
3. Nonlinear, Nonrigid Transformations
Registration tools of this group are based on equations developed by

scientists working on the physics of elastic bodies or fluids. These methods
model the brain as an elastic body or a viscose fluid (Bajcsy and Kovacic,
1989; Christensen et al., 1994; Gee et al., 1993). A distance measure has
to be defined in order to estimate the quality of the match between the
transformed data set and the reference brain. The sum of the differences
in gray values between the reference brain and the transformed brain is an
example of such a measure. Frequently, nonlinear registrations are stepwise
procedures using sequentially different levels of spatial resolution (Henn
and Witsch, 2004b).
The registration tool based on a model which is known from linear
elasticity physics requires the repeated solution of the Navier–Lamé equa-
tion which belongs to the second-order partial differential equations. The
mathematical procedure is quite demanding, if we have to handle a three-
dimensional data set with 256 voxels in each direction (as it is usually the case
in an anatomical MR data set of a human brain), because the resulting system
of equations has 24 million unknowns. The system can efficiently be solved
using multiscale and multigrid approaches (Henn and Witsch, 2004a).
In general, registration tools based on elastic and fluid models are time-
consuming (e.g., in the range of 3.5 h per brain on a recent Linux system).
Furthermore, their application usually requires several steps of prealign-
Figure 18.7. Three-dimensional reconstruction of a MR data set of a postmortem

brain (left) with areas 44 (red) and 45 (yellow) before (left) and after (middle)
nonlinear, elastic registration to a reference brain. Right: sagittal section of the
brain after registration, in which the length and direction of the vectors of the
transformation are visualized by color-coded arrows. Red tones correspond to large
deformations, blue tones to small deformation (Amunts et al., 2004).
ment, e.g., an affine, linear registration, and a gray-level normalization,

which is further a quite critical step. Transformations of this kind, however,
can result in a precise match at the single-voxel level. They have successfully
been applied in order to warp both in vivo structural MR data sets and 3D
reconstructed histological data sets (Fig. 18.7).
An inherent problem of nonlinear transformations is to distinguish a suf-
ficient from an insufficient match. Theoretically, criteria such as the gray-
value differences (see above) may be applicable. The question remains,
however, whether homologue structures in different brains, e.g., sulci, have
been matched. This seems to be no major problem for the large and con-
sistently occurring primary sulci. It is a problem, however, when small side
branches, highly variable small sulci (e.g., diagonal sulcus), or correspond-
ing sulci with a different number of segments (e.g., the precentral sulcus of
two brains with two and four segments, respectively) have to be warped to a
reference brain with differing sulcal morphology. Moreover, the homology
of smaller sulci is not always clear. The comparison of textbooks and atlases
shows that different authors have classified the sulci of the brain in a differ-
ent way e.g., compare the atlas of Duvernoy (1991) with that of Ono et al.
(1990) with respect to the intraparietal and the lower postcentral sulci, or
the occipital sulci). Observer-independent tools for the definition of sulci
may help to define the sulci of an individual brain, but they cannot resolve
principal morphological differences and the problem of sulcal homology.
Anatomist/Brain VISA (http://anatomist/info/) is a software which sup-
ports segmentation, 3D reconstruction, and identification of sulci (Mangin
et al., 1995; Rivière et al., 2002).
V. STANDARD REFERENCE BRAINS
The strategy for selecting the spatial reference system (e.g., standard ref-
erence brain) of an atlas depends on the actual answers to several major
questions. Is it necessary to select an individual reference brain with the
most “prototypical” anatomy? Or, is it better to generate a mean brain by

averaging over a large population of brains? How to consider well-known
differences in brain shape and size between sexes, groups of handedness,
normal and pathologically altered brains? Do these differences require a
standard brain for each of these groups separately? Or, is it irrelevant which
type of standard reference brain has been chosen since all individual brains,
irrespective of their actual macroanatomy, will be registered to this brain?
Here, we can only discuss some aspects of the different strategies. Presently,
no final answer to all the above questions is available.
A. Individual Reference Brain
How can we select an individual brain with macroanatomical features

which are most similar to those of all other brains of a large sample? The
size of such a brain can be precisely calculated by averaging the sizes (e.g.,
volumes) of all brains of the sample. The location and sizes of the subcor-
tical nuclei (e.g., thalamus, striatum, red nucleus) and ventricles may also
be used for identifying the most prototypical brain. The problem becomes
extremely difficult, however, if we want to identify the most prototypical
cortical ribbon in a single brain. This is due to the extremely variable gyri-
fication pattern. Moreover, the degree of gyrification varies considerably
between brains (Armstrong et al., 1991). Thus, some cortical regions are
located at the free surface, other regions are hidden in the depths of the
sulci. The latter portion, however, includes two-third of the whole cortical
surface (Zilles et al., 1988a). Consequently, it is not possible to find homo-
logue positions on the folded cortical surface in different brains on the basis
of macroscopical information alone. It is also difficult if not impossible to
identify a “mean” brain, if the intraparietal sulcus in one brain is represented
by one large furrow whereas two separate segments represent this sulcus in
another brain.
Roland and Zilles (1994) used 21 MRI data sets obtained with a 3D FLASH
sequence at a 1.5 T Siemens Magnetom from healthy 20 to 30 years old,
right-handed, male volunteers for the selection of a reference brain. The
MR data sets were interactively peeled (removing meninges, skull, and ves-
sels from brain tissue), and then scaled in coronal, sagittal, and horizontal
directions until the deviations in brain volume among the 21 brains were
minimal. The standard brain was selected as the brain which deviates the
least in size and hemispheric shape from the 20 other brains. All possible
combinations of two brains were run until a minimum deviant brain was
found. This brain serves as the reference brain of the European Human
Brain Database [ECHBD (Roland and Zilles, 1996) and the Neurogenera-
tor (http://www.neurogenerator.org/about.htm; Roland et al., 2001)].
An advantage of this approach is that formal criteria based on a simi-
larity measure have been applied in order to define this reference brain.
The brain, however, is typical in some regions, but less typical in others. For
example, the occipital lobes are extraordinarily asymmetric. The left occipi-
tal pole protrudes into the space frequently covered by the right hemisphere
(Amunts et al., 2000b). Consequently, medially located activations in the left
visual cortex may appear as activations right of the midsagittal plane when
using this brain as a reference for the topographical interpretation of func-
tional imaging data.
Another individual reference brain is the T1-weighted “single-subject MNI
brain” of the Montreal Neurological Institute (Collins et al., 1994; Evans et al.,
1993; Holmes et al., 1998), which is widely used as a reference brain in func-
tional imaging studies, e.g., in SPM (www.fil.ion.ucl.ac.uk/spm) and the
ICBM viewer (http://www.bic.mni.mcgill.ca/cgi/icbm view/). This brain
has been selected by chance and is not supposed to be a minimum de-
viant brain of a larger sample of brains. Since the data set of this single
brain represents an average of 27 scans, the contrast and the signal-to-noise
ratio of the data set are excellent (Figs. 18.2 and 18.8). Moreover, it has
the advantage that it is coregistered to the “average brain” of the MNI (see
section “Average Brain”). Functional imaging studies have been registered
to both the “single-subject MNI brain” and the “average brain” during the
last years (Brett et al., 2002).
B. Average Brain
The Montreal Neurological Institute created a series of brain templates

which are averages over larger samples of normal brains. The ICBM has
adopted these templates as spatial reference systems (see section “The Hu-
man Brain Atlas of the International Consortium of Human Brain Map-
ping as an Example of a Multimodal Brain Imaging Database”). One of the
first templates was the MNI305 (Evans et al., 1993). The current standard is
known as the MNI152, which is based on 152 scans (Fig. 18.8). This template
is available together with several commonly used functional imaging analysis
Figure 18.8. Average of 305 T1-weighted volumes (MNI305, left), average of 152 T1-
weighted volumes (MNI152, center), and average of 27 T1-weighted scans of the sin-
gle MNI brain (right) at z = 0 (MNI space). The single-subject brain volume is coreg-
istered with the average volumes (http://www.bic.mni.mcgill.ca/cgi/icbm view/).
packages, including SPM and FSL (http://www.fmrib.ox.ac.uk/fsl/). These

two templates have been supplemented by templates from children aged
between 4 and 18 years, 12 and 18 years, and 4 and 11 years, which are all
based on a large number of scans. Although the MNI templates are sup-
posed to be spatially oriented according to the Talairach atlas, they do not
have precisely the same orientation, size, and shape. At the level of the in-
terhemispheric fissure, the anterior commissure does not correspond to the
origin x = 0, y = 0, and z = 0 in the Talairach atlas, but is shifted by 4 mm
in rostrocaudal and dorsoventral directions. For more detailed information
see Brett et al. (2002).
C. Alternative Approaches Based on Volume Data
It is also feasible to use any structural MR data set of an individual brain,

which belongs to the sample of an fMRI group study, as reference brain (in
analogy to the individual reference brain, see section “Individual Reference
Brain”). This approach has the advantage of a correspondence of the data
quality (resolution, contrast, scanner properties) between the “reference”
brain and the individual brains. The results, however, cannot directly be
compared with other studies using different reference brains.
Another possibility is the construction of a “synthetic” brain. An example
of this approach is the digital brain phantom as proposed by Collins et al.
(1998). The advantage of phantoms is the precise knowledge of its physical
features. They can be used in order to test and to evaluate image analysis
algorithms (e.g., tissue segmentation, correction of motion artifacts, partial
volume effects, and many other methodical problems). A major problem
of this approach is, however, the inevitable simplification of the real brain
anatomy. The digital phantom of Collins et al. is based on the high-resolution
single subject MNI template (see above). After preprocessing, the data set
was classified with respect to different tissue types (gray and white mat-
ter, cerebrospinal fluid, fat, and background). Using this data set, different
imaging modalities (MR, PET) can be simulated (Collins et al., 1998).
Reference brains can be generated or selected for a special group of
subjects. For example, if the brain shape of right-handed males has to be
compared with that of right-handed females, it may be problematic to reg-
ister both groups, e.g., to a right-handed male reference brain. The same
is true when brains of patients have to be compared with brains of healthy
subjects. The standard deviations will be different between both groups,
and the results may be confounded. If the selected reference brain is a post-
mortem brain, it may provide detailed, microstructural information, but the
selected brain may be distorted due to histological processing and artifacts.
In vivo brains, in contrast, are free of such artifacts, but do not contain
microstructural information.
The selection of an adequate reference brain depends on the neurosci-
entific question. Human brain anatomy and function vary across age and
gender, and may depend on the individual health status and many other fac-
tors. In conclusion, there is no gold standard for the selection of a reference
brain.
D. Surface-Based Atlases and Flat Maps
A strategy completely different from that of 3D reference brains

based on volume data is the mapping of structural and functional
data to a geometrically well-defined, two-dimensional surface (ellipsoidal
or planar). Flat maps (planar maps) enable a simultaneous visualiza-
tion of regions on the cortical surface as well as those buried in the
depths of sulci while respecting surface topology (Van Essen et al.,
1998). Flat maps are part of widely distributed analysis tools of func-
tional imaging data, e.g., developed by David Van Essen’s laboratory
(http://brainmap.wustl.edu/vanessen.html), Rainer Goebel’s laboratory
(http://www.brainvoyager.de/), Lohmann, von Cramon, 2000; Dale et al.,
1999, Fischl et al. 1999 (http://surfer.nmr.mgh.harvard.edu).
When flat maps are constructed, artificial cuts of the cortical surface must
be introduced in order to enable the “flattening” of the complete cortical
surface. The surface of a hemisphere is frequently cut along the bottom
of sulci such as the calcarine and the cingulate sulci. The visual cortex is
a good example of the advantages of flat maps. fMRI studies exploring
the retinotopy of visual areas have shown that the occipital lobe can be
subdivided into the striate and several extrastriate areas (Dougherty et al.,
2003; Sereno et al., 1995; Shipp et al., 1995; Tootell et al., 1997). The flat
maps demonstrate at the first glance that the early extrastriate areas (e.g.,
V2, V3v, VP, V4v, V5), which are hidden to some degree in the depths of
sulci, surround the primary visual cortex in a belt-like arrangement. The
registration of sulci, gyri, or cytoarchitectonically defined areas to ellipsoidal
surfaces rather than planar (flat) maps may circumvent at least some of the
problems caused by the artificial cuts necessary for the construction of flat
maps.
Both the volume and the surface-based atlases can be seen as complemen-
tary approaches. The data can be transferred from one into the other, and
vice versa. The population-average, landmark-, and surface-based (PALS)
atlas of human cerebral cortex includes both volume and surface represen-
tations of the cortex (Van Essen, 2005). The environment includes a human
brain database (http://sumsdb.wustl.edu:8081/sums/), tools for surface re-
construction (SurFit), visualization and analysis of surface maps (via Caret
software), and other tools (Van Essen et al., 2001). It enables the analysis
of brain data sets both in a population-based reference system (Van Essen,
2005) and in an individual brain, the MNI single-subject brain (Van Essen,
2002, 2004; see also sections “Individual Reference Brain” and “Average
Brain.”). First attempts have been done to integrate our cytoarchitectonic
probabilistic maps in this atlas (Fig. 18.9).
Figure 18.9. Probabilistic cytoarchitectonic map of area 44 (Amunts et al., 1999, 2004)
in coronal slice (overlaid on average sMRI for PALS-B12) plus lateral view and flat
map of the PALS-B12 surface (Van Essen, 2005).
A major problem of the surface-based mapping is the construction of

accurate surface reconstructions from 3D structural MRI data. For exam-
ple, if the MR images have an isotropic resolution of 1 mm, and the
cortical thickness varies between 2 and 4 mm, many voxels at the brain
surface and the cortex/white matter border represent both (partial vol-
ume effect) parts of the cortex and parts of the pial surface plus cere-
brospinal fluid and white matter, respectively. Voxels, which cover a sul-
cus and the adjoining gray matter on either side of the sulcus, “average”
signals from the cerebrospinal fluid, the pial surface, and the cortical rib-
bon. Different strategies have been developed to handle such problems
(http://white.stanford.edu/html/teo/mri/mri.html; http://v1/wustl.edu/
software.html).
VI. MICROSTRUCTURAL LOCALIZATION

OF NEURAL FUNCTIONS
A major goal of neuroimaging studies is to localize neural mechanisms,

and hereby to understand structural–functional relationships. The studies,
e.g., of Penfield (Penfield and Rasmussen, 1950) or of Ojemann (Ojemann
and Mateer, 1979), aimed at this goal by using neurophysiological map-
ping techniques during neurosurgery. Such approaches are major steps
forward to reveal functional mechanisms, but the anatomical localization
of neuronal activities is necessarily restricted to the level of macroscopical
landmarks. The scientifically most interesting microstructural–functional
relationships can only be studied, however, by the registration of functional
imaging data, macroscopical landmarks, and most importantly cyto-, myelo-,
and/or chemoarchitectonic data in an identical spatial reference system,
i.e., a multimodal brain atlas (Mazziotta et al., 2001a).
Therefore, multimodal atlas systems have been designed which combine
functional findings such as fMRI, PET, MEG data with anatomical findings
such as in vivo structural MRI data and postmortem cyto-, myelo-, and re-
ceptor architectonic data from different brains. Meanwhile, scientists at the
Karolinska Institute in Stockholm (Per Roland and coworkers) and at the
Research Center Jülich and the C. and O. Vogt Brain Research Institute,
University of Düsseldorf (Karl Zilles, Katrin Amunts, Axel Schleicher and
coworkers) have performed the combined microstructural–functional com-
parisons of this type in the human brain (Amunts et al., 2004; Binkofski et al.,
2000, 2002; Bodegard et al., 2000a,b; Ehrsson et al., 2000; Eickhoff et al.,
2005a; Geyer et al., 2001; Herath et al., 2001; Horwitz et al., 2003; Indefrey
et al., 2001; Kötter et al., 2001; Larsson et al., 1999, 2002; Naito et al., 1999,
2000; Roland et al., 2001; Roland and Zilles, 1994, 1996, 1998; Stöckel et al.,
2004; Wohlschläger et al., 2005; Young et al., 2003, 2004; Zilles et al., 1995).
Presently, ∼40 cytoarchitectonically defined cortical areas, 9 fiber tracts,
and 16 nuclei/subnuclei have been mapped (Amunts et al., 1999, 2000b,
2003, 2005; Bürgel et al., 1999; Choi et al., 2005; Eickhoff et al., 2005b; Geyer,
2004; Geyer et al., 1996, 1999, 2000b; Grefkes et al., 2001; Morosan et al.,
2001; Rademacher et al., 2001a, b, 2002; Zilles et al., 2002a, 2003, 2004; Zilles
and Palomero-Gallagher, 2001).
The combination of microstructural maps with functional imaging data
has not only improved the precision of the localization of the functional
activations, but led to new discoveries, e.g., subdivisions of the human pri-
mary motor cortex, Brodmann’s area 4. Area 4 has been subdivided into an
anterior and a posterior part, both subserving different functions (Geyer
et al., 1996). Functional maps show a much more detailed segregation of
the regional organization of the human cerebral cortex than was expected
from classical architectonic maps. This stimulated new architectonic studies.
For example, human area VIP (Ventral IntraParietal area) in the intrapari-
etal sulcus has been detected in a fMRI study, employing different moving
stimuli (Bremmer et al., 2001). This area cannot be found in Brodmann’s
classical map, but stimulated by the functional imaging study, the cytoarchi-
tectonic correlate of VIP has now been delineated in histological sections
of postmortem brains (Choi et al., 2005). On the other hand, cytoarchitec-
tonic data can generate working hypotheses for functional imaging studies.
For example, the cytoarchitectonic mapping of the secondary somatosen-
sory cortex has revealed a subdivision into four distinct cytoarchitectonic
areas, for which no correlate can be found in Brodmann’s map (Amunts
et al., 2003; Eickhoff et al., 2005b). Functional imaging has elucidated the
functional correlates of these architectonical subdivisions in more detail
(Eickhoff et al., 2005a).
Recently, a SPM toolbox (Fig. 18.10) has been developed (http://www.fz-
juelich.de/ime/index.php), which is MATLAB based, integrated into
the SPM2 software package (www.fil.ion.ucl.ac.uk/spm), and enables an
integration of probabilistic cytoarchitectonic maps and results of functional
imaging studies (Eickhoff et al., 2005c). The toolbox includes the func-
tionality for the construction of summary maps comprising probability of
several cortical areas. This is enabled by an algorithmic definition of the
most probable assignment of each voxel to one of these areas. Its main fea-
ture is to provide several measures defining the degree of correspondence
Figure 18.10. Combination of microanatomical and functional data using the novel
SPM toolbox (Eickhoff et al., 2005c).
between architectonic areas and functional foci. The software together with
the presently existing probability maps is available as open source software
to the neuroimaging community. This new toolbox provides an easy-to-use
tool for the integrated analysis of functional and microanatomical data in a
common reference space.
The toolbox contains a summary map, which is computed for all hitherto
published probabilistic maps. It defines the most likely cytoarchitectonic
area for each voxel (“maximum probability map,” MPM), and hereby en-
ables the definition of nonoverlapping volumes of interest for each area
(Fig. 18.11).
Figure 18.11. Surface rendering of the T1-weighted MNI reference brain and the
MPMs of all presently published cortical areas, which were defined by observer-
independent, quantitative cytoarchitectonic mapping. Only the surface extent of
the different areas is shown. (A) Lateral view of the left hemisphere, (B) dorsal view,
and (C) lateral view of the right hemisphere.
Ten years of experience with microstructural brain mapping suggests
that the classical architectonic maps of the human brain must be modi-
fied. New areas have to be defined, and other cytoarchitectonic areas, e.g.,
Brodmann’s area 19 of the extrastriate cortex, have to be replaced by more
detailed parcellation schemes. Not only the number and location of archi-
tectonic areas have to be considerably changed, but also the exclusively de-
scriptive organizational principles have to be enriched by more functional
aspects. Hypotheses have been suggested which propose hierarchies of cor-
tical areas, based on their structural or functional connectivity (Kötter et al.,
2001; Passingham et al., 2002; Stephan et al., 2000; Van Essen, 1985).
A novel promising, more functionally related chemoarchitectonic ap-
proach to brain mapping has recently been suggested. This approach is
based on the receptor architecture of the cerebral cortex, i.e., the analy-
sis of the regional and laminar distribution patterns of numerous recep-
tor binding sites of various neurotransmitter systems. The receptor bind-
ing sites can be visualized by quantitative in vitro receptor autoradiography
(Zilles, 1991; Zilles et al., 1991b, 1995, 2002a; Zilles and Palomero-Gallagher,
2001). It has been demonstrated that the balance between the total bind-
ing values (averaged over all cortical layers of a cortical area) of various
different receptors is a characteristic feature of a cortical area. This bal-
ance is like a “receptor fingerprint” of a cortical unit (Zilles et al., 2002a).
Moreover, it has been shown that areas of similar function show similar re-
ceptor fingerprints and differ from those with other properties (Zilles et
al., 2002a). Therefore, mathematical measures of similarity may help to de-
fine neurochemical and functional hierarchies and relationships of cortical
areas.
VII. SUMMARY OF ADVANTAGES/LIMITATIONS OF

CYTOARCHITECTONIC PROBABILISTIC MAPS
A. Advantages
1. Microstructural reference for data from imaging and neurophysio-

logical studies of the living human brain for structural–functional
analyses.
2. Link between different modalities of data (receptor architecture,
myeloarchitecture, fMRI, MEG, PET, etc.) as well as between post-
mortem and in vivo data.
3. Maps can be coregistered to any individual MR data set of a brain, e.g.,
brains of patients.
4. Quantification of interhemispheric differences and intersubject vari-
ability of cortical areas (“probabilistic” maps).
5. Open system—new cortical areas can be considered (cortical areas,
nuclei, and fiber tracts).
6. High-spatial resolution.
B. Limitations
1. The maps provide a probabilistic, but not an unambiguous anatomical

classification of an individual in vivo MR data set (due to intersubject
variability).
2. The maps do not cover the whole brain (“work in progress”).
3. They represent a single aspect of microstructural organization, i.e.,
cytoarchitecture.
4. No maps available for the developing human brain and for the patho-
logically altered brain.
APPENDIX
A. Method of the Generation of Cytoarchitectonic Probabilistic Maps
1. The procedure is described in detail in Amunts et al. (2000b) and

Zilles et al. (2002b).
2. Tissue: whole human postmortem brains fixed in buffered forma-
lin (4%) or Bodian mixture (formalin, glacial acetic acid, and etha-
nol).
3. MR scan of the fixed postmortem brain, e.g., T1-weighted 3D FLASH
sequence, flip angle 40◦ , repetition time of TR = 40 ms, echo time
of TE = 5 ms for each image; spatial resolution was 1 mm × 1 mm ×
1.17 mm, 8-bit gray-level resolution (“MR data set”).
4. Embedding in paraffin.
5. Serial histological sectioning at a microtone for large sections in
coronal, sagittal, or horizontal plane (6000–8000 sections, thickness
20 µm).
6. Each 60th blockface of the paraffin block (distance between images:
1.2 mm) is digitized via a CCD camera (“blockface images”).
7. Each 60th section is stained for cell bodies (Merker, 1983). Neighbor-
ing sections are stained for myelin (Bürgel et al., 1997; Gallyas, 1979)
or Heidenhain–Wölcke (Burck, 1973).
8. Digitalization of the stained sections; ∼130 per brain in coronal sec-
tions (“histological images”).
9. Three-dimensional reconstruction of the histological images using
the blockface images and the MR data set (Schormann et al., 1995).
10. Definition of borders of cortical areas using an algorithmic approach
based on the
r measurement of the GLI as an indicator of the cell packing density
in cortical regions (Schleicher and Zilles, 1990; Wree et al., 1982);
r generation of a sequence of GLI profiles, covering the cortical re-
gion of interest and reflecting the laminar distribution of the GLI
from the surface of the cortex to the white matter border;
r extraction of features defining the shape of the GLI profiles
(Schleicher et al., 1999);
r calculation of multivariate distance measures for the quantification
of differences between neighboring groups of profiles;
r definition of borders of cortical areas at those positions, at which
the distance measure reaches a local maximum.
11. Definition of borders of subcortical nuclei and fiber tracts in histo-
logical sections.
12. Tracing (interactively) of borders of cortical areas, subcortical nuclei,
and fiber tracts into the 3D reconstructed histological volume.
13. Registration of the complete volume data to standard reference
space [e.g., ECHBD (Roland and Zilles, 1994) and MNI template
(Evans et al., 1993; Mohlberg et al., 2003)] using different steps of
preprocessing, linear, affine, and nonlinear, elastic transformations
(Amunts et al., 2004; Schormann and Zilles, 1998).
14. Registration of the delineated areas, nuclei, and fibers (=anatomical
structures) to the standard reference space.
15. Superimposition of anatomical structures of 10 postmortem brains in
the common reference space.
16. Probability maps (population maps, probabilistic maps) are gener-
ated which describe, for each voxel of the reference space, the rela-
tive frequency with which an anatomical structure was present in the
sample of 10 brains. The frequency is color coded.
B. Useful Links and Atlas Tools (in alphabetical order)
1. AIR http://air.bmap.ucla.edu
2. Anatomist/Brain VISA http://anatomist/info/
3. Allan Atlas for gene expression (mouse brain) http://www.
brainatlas.org/default.asp and http://www.brain-map.org/index.jsp
4. Brain Voyager http://www.brainvoyager.de/
5. Caret and surface-based atlas http://sumsdb.wustl.edu:8081/sums/
6. Cytoarchitectonic probabilistic maps http://www.fz-juelich.de/ime/
index.php
7. Extraction of brain surfaces http://white.stanford.edu/html/teo/
mri/mri.html; http://v1/wustl.edu/software.html
8. Free surfer http://surfer.nmr.mgh.harvard.edu
9. FSL http://www.fmrib.ox.ac.uk/fsl/
10. ICBM http://www.loni.ucla.edu/ICBM/
11. Neurogenerator http://www.neurogenerator.org/about.htm
12. SPM www.fil.ion.ucl.ac.uk/spm/
13. Talairach Daemon http://ric.uthscsa.edu/projects/
talairachdaemon.html
14. Web site of the MNI, ICBM viewer http://www.bic.mni.mcgill.ca/
cgi/icbm view/
REFERENCES
Amunts, K., Eickhoff, S., and Zilles, K., 2003, Multimodal mapping of human cerebral cortex—
individual variability, In: Ng, V., Barker, G. J., and Hendler, T. (eds.), Psychiatric Neuroimag-
ing, Amsterdam, Berlin, Oxford, Tokyo, Washington, DC: IOS press, pp. 16–20.
Amunts, K., Jäncke, L., Mohlberg, H., Steinmetz, H., and Zilles, K., 2000a, Interhemispheric
asymmetry of the human motor cortex related to handedness and gender, Neuropsychologia
38:304–312.
Amunts, K., Kedo, O., Kindler, M., Pieperhott, P., Mohlberg, H., Shah, N. J., Habell, U., and
Zilles, K., 2005, Cytoarchitectonic mapping of the amygdala, hippocampal region and
entorhinal cortex: intersubject variability and probability maps. Anat. Embryol. (Berl.)
200:343–352.
Amunts, K., Malikovic, A., Mohlberg, H., Schormann, T., and Zilles, K., 2000b, Brodmann’s
areas 17 and 18 brought into stereotaxic space—where and how variable? Neuroimage
11:66–84.
Amunts, K., Schleicher, A., Bürgel, U., Mohlberg, H., Uylings, H. B. M., and Zilles, K., 1999,
Broca’s region revisited: cytoarchitecture and intersubject variability, J. Comp. Neurol.
412:319–341.
Amunts, K., Weiss, P. H., Mohlberg, H., Pieperhoff, P., Gurd, J., Shah, J. N., Marshall, C. J., Fink,
G. R., and Zilles, K., 2004, Analysis of the neural mechanisms underlying verbal fluency in
cytoarchitectonically defined stereotaxic space—the role of Brodmann’s areas 44 and 45,
Neuroimage 22:42–56.
Annese, J., Pitiot, A., Dinov, I. D., and Toga, A. W., 2004, A myelo-architectonic method for the
structural classification of cortical areas, Neuroimage 21:15–26.
Ardekani, B. A., Nierenberg, J., Hoptman, N. J., Javitt, D. C., and Lim, K. O., 2003,
MRI study of white matter diffusion anisotropy in schizophrenia, Neuroreport 14:2025–
2029.
Armstrong, E., Curtis, M., Buxhoeveden, D. P., Gregoe, C., Zilles, K., Casanova, M. F., and
McCarthy, W. F., 1991, Cortical gyrification in the rhesus monkey: a test of the mechanical
folding hypothesis, Cereb. Cortex 1:426–432.
Ashburner, J., and Friston, K. J., 1999, Nonlinear spatial normalization using basic functions,
Hum. Brain Mapp. 7:254–266.
Assaf, Y., Ben-Bashat, D., Chapman, J., Peled, S., Biton, I. E., Kafri, M., Segev, Y., Hendler, T.,
Korczyn, A. D., Graif, M., and Cohen, Y., 2002, High b-value q-space analyzed diffusion-
weighted MRI: application to multiple sclerosis, Magn. Reson. Med. 47: 115–126.
Assaf, Y., Mayk, A., and Cohen, Y., 2000, Displacement imaging of spinal cord using q-space
diffusion-weighted MRI, Magn. Reson. Med. 44:713–722.
Axer, H., and Keyserlingk, D. G., 2000, Mapping of fibre orientation in human internal capsule
by means of polarized light and confocal scanning laser microscopy, J. Neurosci. Methods
94:165–175.
Bailey, P., and von Bonin, G., 1951, The Isocortex of Man, Urbana: University of Illinois Press.
Bajcsy, R., and Kovacic, S., 1989, Multiresolution elastic matching, Comput. Vis. Graph. Image
Process 46:1–21.
Barker, G. J., 2001, Diffusion-weighted imaging of the spinal cord and optic nerve, J. Neurol.
Sci. 186:S45–S49.
Basser, P. J., and Jones, D. K., 2002, Diffusion-tensor MRI: theory, experimental design and data
analysis—a technical review, NMR Biomed. 15:456–467.
Behrens, T. E., Johansen-Berg, H., Woolrich, M. W., Smith, S. M., Wheeler-Kingshot, C. A.,
Boulby, P. A., Barker, G. J., Sillery, E. L., Sheehan, K., Ciccarelli, O., Thompson, A. J.,
Brady, J. M., and Matthews, P. M., 2003, Non-invasive mapping of connections between
Bidmon, H. J., Wu, J. Y., Godecke, A., Schleicher, A., Mayer, B., and Zilles, K., 1997, Nitric oxide
synthase-expressing neurons are area-specifically distributed within the cerebral cortex of
the rat, Neuroscience 81:321–330.
Binkofski, F., Amunts, K., Stephan, K. M., Posse, S., Schormann, T., Freund, H. -J., Zilles, K., and
Seitz, R. J., 2000, Broca’s region subserves imagery of motion: a combined cytoarchitectonic
and fMRI study, Hum. Brain Mapp. 11:273–285.
Binkofski, F., Fink, G. R., Geyer, S., Buccino, G., Gruber, O., Shah, J. N., Taylor, J. G., Seitz,
R. J., Zilles, K., and Freund, H. J., 2002, Neural activity in human primary motor cortex
areas 4a and 4p of human motor cortex is modulated differentially by attention to action,
J. Neurophysiol. 88:519.
Bodegard, A., Geyer, S., Naito, E., Zilles, K., and Roland, P. E., 2000a, Somatosensory areas in
men activated by moving stimuli: cytoarchitectonic mapping and PET, Neuroreport 11:187–
191.
Bodegard, A., Ledberg, A., Geyer, S., Naito, E., Zilles, K., and Roland, P. E., 2000b, Object
shape differences reflected by somatosensory cortical activation in human, J. Neurosci. 20
RC 51:1–5.
Bremmer, F., Schlack, A., Shah, N. J., Zafiris, O., Zilles, K., and Fink, G. R., 2001, Polymodal
motion processing in posterior parietal and premotor cortex: a human fMRI study strongly
implies equivalencies between humans and monkeys, Neuron 29:287–296.
Brett, M., Johnsrude, I. S., and Owen, A. M., 2002, The problem of functional localization in
the human brain, Nat. Rev. Neurosci. 3:243–249.
Brodmann, K., 1908, Beiträge zur histologischen Lokalisation der Großhirnrinde. VI: die Cor-
texgliederung des Menschen, J. Psychol. Neurol. X:231–246.
Brodmann, K., 1909, Vergleichende Lokalisationslehre der Großhirnrinde in ihren Prinzipien dargestellt
auf Grund des Zellenbaues, Leipzig: Barth JA.
Brodmann, K., 1994, Brodmann’s Localization in the Cerebral Cortex, London: Smith Gordon.
Burck, H. -C., 1973, Histologische Technik, Stuttgart: Thieme-Verlag.
Bürgel, U., Amunts, K., Hoemke, L., Mohlberg, H., Gilsbach, J. M., Zilles, K., 2005, White matter
fiber tracts of the humans brain: Three-dimensional mapping of microscopic resolution,
topography and intersubject variability, Neuroimage Epub ahead of print.
Bürgel, U., Mecklenburg, I., Blohm, U., and Zilles, K., 1997, Histological visualization of long
fiber tracts in the white matter of adult human brains, J. Brain Res. 3:393–404.
Bürgel, U., Schormann, T., Schleicher, A., and Zilles, K., 1999, Mapping of histologically
identified long fiber tracts in human cerebral hemispheres to the MRI volume of a ref-
erence brain: position and spatial variability of the optic radiation, Neuroimage 10:489–
499.
Campbell, A. W., 1905, Histological Studies on the Localisation of Cerebral Function, Cambridge:
Cambridge University Press.
Choi, H.-J., Zilles, K., Mohlberg, H., Schleicher, A., Fink, G. R., and Amunts, K., 2005, Cytoar-
chitectonic mapping of the anterior ventral bank of the human intraparietal sulcus, J.
Comp. Neurol., in press.
Christensen, G. E., Rabbitt, R. D., and Miller, M. I., 1994, 3-D brain mapping using deformable
neuroanatomy, Phys. Med. Biol. 39:609–618.
Clark, C. A., Barker, G. J., and Tofts, P. S., 1999, Magnetic resonance diffusion imaging of the
human cervical spinal cord in vivo, Magn. Reson. Med. 41:1269–1273.
Clark, V. P., Courchesne, E., and Grafe, M., 1992, In vivo myeloarchitectonic analysis of hu-
man striate and extrastriate cortex using magnetic resonance imaging, Cereb. Cortex 2:
424.
Clarke, S., 1994, Modular organization of human extrastriate visual cortex: evidence from
cytochrome oxidase pattern in normal and macular degeneration cases, Eur. J. Neurosci.
6:725–736.
Clarke, S., and Miklossy, J., 1990, Occipital cortex in man: organization of callosal connections,
related cyto- and myeloarchitecture, and putative boundaries of functional visual areas, J.
Comp. Neurol. 298:188–214.
Coenen, V. A., Krings, T., Mayfrank, L., Polin, R. S., Reinges, M. H. T., Thron, A., and Gils-
bach, J. M., 2001, Three-dimensional visualization of the pyramidal tract in a neuronaviga-
tion system during brain tumor surgery: first experiences and technical note, Neurosurgery
49:86–93.
Collins, D. L., Neelin, P., Peters, T. M., and Evans, A. C., 1994, Automatic 3D intersubject
registration of MR volumetric data in standardized Talairach space, J. Comput. Assis. Tomogr.
18:192–205.
Collins, D. L., Zijdenbos, A., Kollokian, V., Sled, J. G., Kabani, N. J., Holmes, C. J., and Evans,
A. C., 1998, Design and construction of a realistic digital brain phantom, IEEE Trans. Med.
Imaging 17:463–468.
Conturo, T. E., Lori, N. F., Cull, T. S., Akbudak, E., Snyder, A. Z., Shimony, J. S., McKinstry, R.
C., Burton, H., and Raichle, M. E., 1999, Tracking neuronal fiber pathways in the living
human brain, Proc. Natl. Acad. Sci. 96:10422–10427.
Dale, A. H., Fischl, B., and Sereno, H. I., 1999, Cortical surface based analysis. I: segmentation
and surface reconstruction, Neuroimage 9:179–194.
Di Virgilio, G., and Clarke, S., 1997, Direct interhemispheric visual input to human speech
areas, Hum. Brain Mapp. 5:347–354.
Dougherty, R. F., Koch, V. M., Brewer, A. A., Fischer, B., Modersitzky, J., and Wandell, B. A.,
2003, Visual field representations and locations of visual areas V1/2/3 in human visual
cortex, J. Vis. 3:586–598.
Duvernoy, H., 1991, The Human Brain. Surface, Three-Dimensional Sectional Anatomy and MRI,
Wien, New York: Springer.
Ehrsson, H. H., Naito, E., Geyer, S., Amunts, K., Zilles, K., Forssberg, H., and Roland, P. E.,
2000, Simultaneous movements of upper and lower limbs are coordinated by motor rep-
resentations that are shared by both limbs: a PET study, Eur. J. Neurosci. 12:3385–3398.
Eickhoff, S., Amunts, K., Mohlberg, H., and Zilles, K., 2005a, The human parietal operculum:
II. Stereotaxic maps and correlation with functional imaging results, Cereb. Cortex, Epub
ahead of print.
Eickhoff, S., Schleicher, A., Zilles, K., and Amunts, K., 2005b, The human parietal operculum:
I. Cytoarchitectonic mapping of subdivisions, Cereb. Cortex, Epub ahead of print.
Eickhoff, S., Stephan, K. E., Mohlberg, H., Grefkes, C., Fink, G. R., Amunts, K., and Zilles,
K., 2005c, A new SPM toolbox for combining probabilistic cytoarchitectonic maps and
functional imaging data, Neuroimage 25:1325–1335.
Eickhoff, S., Walters, N., Schleicher, A., Egan, G. F., Zilles, K., Watson, J. D. G., and Amunts,
K., 2005d, High-resolution MRI reflects myelo- and cytoarchitecture of human cerebral
cortex, Hum. Brain Mapp. 24:206–215.
Elliot Smith, G., 1907, A new topographical survey of the human cerebral cortex, being an
account of the distribution of the anatomically distinct cortical areas and their relationship
to the cerebral sulci, J. Anat. 41:237–254.
Evans, A. C., Collins, D. L., Mills, S. R., Brown, E. D., Kelly, R. L., and Peters, T. M., 1993, 3D
statistical neuroanatomical models from 305 MRI volumes, In: Proceedings of the IEEE-NSS-
MI Symposium, pp. 1813–1817.
Evans, A. C., Collins, D. L., and Milner, B., 1992, An MRI-based stereotactic atlas from 250
young normal subjects, Soc. Neurosci. Abstr. 18:408.
Falk, D., Hildeboldt, C., Cheverud, J., Kohn, L. A. P., Figiel, G., and Vannier, M., 1991, Human
cortical asymmetries determined with 3D MR technology, J. Neurosci. Methods 39:185–191.
Fatterpekar, G. M., Naidich, T. P., Delamn, B. N., Aguinaldo, J. G., Gultekin, S. H., Sherwood,
C. C., Hof, P. R., Drayer, B. P., and Fayad, Z. A., 2002, Cytoarchitecture of the human
cerebral cortex: MR microscopy of excised specimens at 9.4 Tesla, Am. J. Neuroradiol. 23:
1313–1321.
Filimonoff, I. N., 1932, Über die Variabilität der Großhirnrindenstruktur. Mitteilung II—Regio
occipitalis beim erwachsenen Menschen, J. Psychol. Neurol. 44:2–96.
Fischl, B., Sereno, M. I., and Dale, A. M., 1999, Cortical surface based analysis. II: inflation,
flatterring, and a surface based coordinate system, Neuroimage 9:195–207.
Flechsig, P., 1920, Anatomie des menschlichen Gehirns und Rückenmarks auf myelogenetischer Grund-
lage, Leipzig: Thieme.
Foong, J., Symms, M. R., Barker, G. J., Maier, M., Miller, D. G. H., and Ron, M. A., 2002, Investi-
gating regional white matter in schizophrenia using diffusion tensor imaging, Neuroreport
13:333–336.
Fox, P. T., Perlmutter, J. S., and Raichle, M., 1985, A stereotactic method of localization for
positron emission tomography, J. Comput. Assis. Tomogr. 9:141–153.
Friston, K. J., 1997, Testing for anatomically specified regional effects, Hum. Brain Mapp. 5:133–
136.
Gallyas, F., 1979, Silver staining of myelin by means of physical development, Neurol. Res. 1:203–
209.
Gee, J. C., Reivich, M., and Bajcsy, R., 1993, Elastically deforming an atlas to match anatomical
brain images, J. Comput. Assis. Tomogr. 17:225–236.
Gerhardt, E., 1940, Die Cytoarchitektonik des Isocortex parietalis beim Menschen, J. Psychol.
Neurol. 49:367–419.
Geyer, S., 2004, The Microstructural Border Between the Motor and the Cognitive Domain in the Human
Cerebral Cortex, Berlin, Heidelberg: Springer.
Geyer, S., Ledberg, A., Schleicher, A., Kinomura, S., Schormann, T., Bürgel, U., Klingberg, T.,
Larsson, J., Zilles, K., and Roland, P. E., 1996, Two different areas within the primary motor
cortex of man, Nature 382:805–807.
Geyer, S., Matelli, M., Luppino, G., and Zilles, K., 2000a, Functional neuroanatomy of the
primate isocortical motor system, Anat. Embryol. 202:443–474.
Geyer, S., Schleicher, A., Schormann, T., Mohlberg, H., Bodegard, A., Roland, P. E., and Zilles,
K., 2001, Integration of microstructural and functional aspects of human somatosensory
areas 3a, 3b, and 1 on the basis of a computerized brain atlas, Anat. Embryol. 204:351–
366.
Geyer, S., Schleicher, A., and Zilles, K., 1999, Areas 3a, 3b, and 1 of human primary so-
matosensory cortex: I. Microstructural organisation and interindividual variability, Neu-
roimage 10:63–83.
Geyer, S., Schormann, T., Mohlberg, H., and Zilles, K., 2000b, Areas 3a, 3b, and 1 of human
primary somatosensory cortex: II. Spatial normalization to standard anatomical space,
Neuroimage 11:684–696.
Grefkes, C., Geyer, S., Schormann, T., Roland, P., and Zilles, K., 2001, Human somatosensory
area 2: observer-independent cytoarchitectonic mapping, interindividual variability, and
population map, Neuroimage 14:617–632.
Hayes, T. L., and Lewis, D. A., 1992, Nonphosphorylated neurofilament protein and calbindin
immunoreactivity in layer III pyramidal neurons of human neocortex, Cereb. Cortex 2:56–
67.
Henn, S., and Witsch, K., 2004a, A multigrid approach for minimizing a nonlinear functional
for digital image matching, Computing 64:339–348.
Henn, S., and Witsch, K., 2004b, Iterative multigrid regularization techniques for image match-
ing, SIAM J. Sci. Comput. 23:1077–1093.
Herath, P., Klingberg, T., Young, Y., Amunts, K., and Roland, P., 2001, Neural correlates of dual
task interference can be dissociated from those of divided attention: an fMRI study, Cereb.
Cortex 11:796–805.
Holmes, C. J., Hoge, R., Collins, L., Woods, R., Toga, A. W., and Evans, A. C., 1998, Enhancement
of MR images using registration for signal averaging, J. Comput. Assis. Tomogr. 22:324–333.
Horwitz, B., Amunts, K., Bhattacharyya, R., Patkin, D., Jeffries, J., Zilles, K., and Braun, A. R.,
2003, Activation of Broca’s area during the production of spoken and signed language: a
combined cytoarchitectonic mapping and PET analysis, Neuropsychologia 41:1868–1876.
Indefrey, P., Brown, C. M., Hellwig, F., Amunts, K., Herzog, H., Seitz, R. J., and Hagoort, P.,
2001, A neural correlate of syntactic encoding during speech production, Proc. Natl. Acad.
Sci. 98:5933–5936.
Ito, R., Mori, S., and Melhem, E. R., 2002, Diffusion tensor brain imaging and tractography,
Neuroimaging Clin. N. Am. 12:1–19.
Iwasawa, T., Matoba, H., Ogi, A., Kurihara, H., Saito, K., Yoshida, T., Matsubara, S., and Nozaku,
A., 1997, Diffusion-weighted imaging of the human optic nerve: a new approach to evaluate
optic neuritis in multiple sclerosis, Magn. Reson. Med. 38:484–491.
Johansen-Berg, H., Behrens, T. E. J., Robson, M. D., Drobnjak, I., Rushworth, M. F. S., Brady, J.
M., Smith, S. M., Higham, D. J., and Matthews, P. M., 2004, Changes in connectivity profiles
define functionally distinct regions in human medial frontal cortex, Proc. Natl. Acad. Sci.
101:13335–13340.
Johansen-Berg, H., Behrens, T. E. J., Sillery, E., Ciccarelli, O., Thompson, A. J., Smith, S. M.,
and Matthews, P. M., 2005, Functional–anatomical validation and individual variation of
diffusion tractography-based segmentation of the human thalamus, Cereb. Cortex 15:31–39.
Jones, D. K., Williams, S. C., Gasston, D., Horsefiled, M. A., Simmons, A., and Howard, R., 2002,
Isotropic resolution diffusion tensor imaging with whole brain acquisition in a clinically
acceptable time, Hum. Brain Mapp. 15:216–230.
Juch, H., Zimine, I., Seghier, M. L., Lazeyras, F., and Fasel, J. H. D., 2005, Anatomical variabil-
ity of the lateral frontal lobe surface: implication for intersubject variability in language
neuroimaging, Neuroimage 24:504–514.
Kleist, K., 1934, Gehirnpathologie, Leipzig: Barth.
Kononova, E. P., 1935, Structural variability of the cortex cerebri. Inferior frontal gyrus in adults
(Russian), In: Sarkisov, S. A., and Filimonoff, I. N. (eds.), Annals of the Brain Research Institute.
Vol. I., Moscow-Leningrad: State Press for Biological and Medical Literature, pp. 49–118.
Kötter, R., Stephan, K. E., Palomero-Gallagher, N., Geyer, S., Schleicher, A., and Zilles, K., 2001,
Multimodal characterisation of cortical areas by multivariate analyses of receptor binding
and connectivity data, Anat. Embryol. 204:333–350.
Krings, T., Coenen, V. A., Axer, H., Möller-Hartmann, W., Mayfrank, L., Weidemann, J.,
Kränzlein, H., Gilsbach, J. M., and Thron, A., 2001, In vivo 3D visualization of pyrami-
dal tracts using anisotropic diffusion weighted magnetic resonance imaging, Neurosci. Lett.
307:192–196.
Kruggel, F., Brückner, M. K., Arendt, T., Wiggins, C. J., and von Cramon, D. Y., 2003, Analyzing
the neocortical fine-structure, Med. Image Anal. 7:251–264.
Lancaster, J. L., Woldorff, M. G., Parsons, L. M., Liotti, M., Freitas, C. S., Rainey, L., Kochunov,
P. V., Nickerson, D., Mikiten, S. A., and Fox, P. T., 2001, Automated Talairach atlas labels
for functional brain mapping, Hum. Brain Mapp. 10:120–131.
Larsson, J., Amunts, K., Gulyas, B., Malikovic, A., Zilles, K., and Roland, P. E., 1999, Neuronal
correlates of real and illusory contour perception: functional anatomy with PET, Eur. J.
Neurosci. 11:4024–4036.
Larsson, J., Amunts, K., Gulyas, B., Malikovic, A., Zilles, K., and Roland, P. E., 2002, Perceptual
segregation of overlapping shapes activates posterior extrastriate visual cortex in man, Exp.
Brain Res. 143:1–10.
Le Bihan, D., Mangin, J.-F., Poupon, C., Clark, C. A., Pappata, S., Molko, N., and Chabriat,
H., 2001, Diffusion tensor imaging: concepts and applications, J. Magn. Reson. Imaging
13:534–546.
LeMay, M., and Kido, D. K., 1978, Asymmetries of the cerebral hemispheres on computed
tomograms, J. Comput. Assis. Tomogr. 2:471–476.
Livingstone, M. S., and Hubel, D. H., 1987, Connections between layer 4B of area 17 and the
thick cytochrome oxidase stripes of area 18 in the squirrel monkey, J. Neurosci. 7:3371–3377.
Lohmann, G., and von Cramon, D. Y., 2000, Automated Labelling of the human cortical surface
using sulcal bassins. Med. Image Anal. 4:179–188.
Luppino, G., Matelli, M., Camarda, R. M., Gallese, V., and Rizzolatti, G., 1991, Multiple rep-
resentations of body movements in mesial area 6 and the adjacent cingulate cortex: an
intracortical microstimulation study in the macaque monkey, J. Comp. Neurol. 311: 463–482.
Mai, J. K., Assheuer, J., and Paxinos, G., 2004, Atlas of the Human Brain, San Diego, London:
Elsevier.
Majocha, R. E., Marotta, C. A., and Benes, F. M., 1985, Immunostaining of neurofilament
protein in human postmortem cortex: a sensitive and specific approach to the pattern
analysis of human cortical cytoarchitecture, Can. J. Biochem. Cell Biol. 63:577–584.
Mangin, J.-F., Frouin, V., Bloch, I., Régis, J., and Lopez-Krahe, J., 1995, From 3D magnetic
resonance images to structural representations of the cortex topography using topology
preserving deformations, J. Math. Imaging Vis. 5:297–318.
Matelli, M., Luppino, G., and Rizzolatti, G., 1991, Architecture of superior and mesial area 6
and the adjacent cingulate cortex in the macaque monkey, J. Comp. Neurol. 311: 445–462.
Mazziotta, J. C., Toga, A. W., Evans, A. C., Fox, P. T., and Lancaster, J. L., 1995, A probabilistic
atlas of the human brain: theory and rationale for its development, Neuroimage 2:89–
101.
Mazziotta, J., Toga, A., Evans, A., Fox, M., Lancaster, J., Zilles, K., Woods, R., Paus, T., Simpson,
G., Pike, B., Holmes, C., Collins, L., Thompson, P., Macdonald, D., Iacoboni, M., Schor-
mann, T., Amunts, K., Palomero-Gallagher, N., Geyer, S., Parsons, L., Narr, K., Kabani,
N., LeGoualher, G., Boomsma, D., Cannon, T., Kawashima, R., and Mazoyer, B., 2001a,
A probabilistic atlas and reference system for the human brain: international consortium
for brain mapping (ICBM), Phil. Trans. R. Soc. Lond. B 356:1293–1322.
Mazziotta, J., Toga, A., Evans, A., Fox, P., Lancaster, J., Zilles, K., Woods, R., Paus, T., Simpson, G.,
Pike, B., Holmes, C., Collins, L., Thompson, P., Macdonald, D., Iacoboni, M., Schormann,
T., Amunts, K., Palomero-Gallagher, N., Geyer, S., Parsons, L., Narr, K., Kabani, N., Le
Goualher, G., Feidler, J., Smith, K., Boomsma, D., Pol, H. H., Cannon, T., Kawashima, R.,
and Mazoyer, B., 2001b, A four-dimensional probabilistic atlas of the human brain, J. Am.
Med. Inform. Assoc. 8:401–430.
Merker, B., 1983, Silver staining of cell bodies by means of physical development, J. Neurosci.
Methods 9:235–241.
Miklossy, J., and van der Loos, H., 1991, The long-distance effects of brain lesions: visualization
Mohlberg, H., Lerch, J., Amunts, K., Evans, A. C., and Zilles, K., 2003, Probabilistic cytoarchitec-
tonic maps transformed into MNI space, In: Proceedings of the Ninth International Conference
of Neuroimage CD Rom on Functional Mapping of the Human Brain, New York, 2003.
Mori, S., and van Zijl, P. C., 2002, Fibre tracking: principles and strategies—a technical review,
NMR Biomed. 15:468–480.
Morosan, P., Rademacher, J., Schleicher, A., Amunts, K., Schormann, T., and Zilles, K., 2001,
Human primary auditory cortex: cytoarchitectonic subdivisions and mapping into a spatial
reference system, Neuroimage 13:684–701.
Mufson, E. J., Brady, D. R., and Kordower, J. H., 1990, Tracing neuronal connections in post-
mortem human hippocampal complex with the carbocyanine dye DiI, Neurobiol. Aging
11:649–653.
Naito, E., Ehrsson, H. H., Geyer, S., Zilles, K., and Roland, P. E., 1999, Illusory arm movements
activate cortical motor areas: a positron emission tomography study, J. Neurosci. 19:6134–
6144.
Naito, E., Kinomura, S., Geyer, S., Kawashima, R., Roland, P. E., and Zilles, K., 2000, Fast reaction
to different sensory modalities activate common fields in the motor areas, but the anterior
cingulate cortex is involved in the speed of reaction, J. Neurophysiol. 83:1701–1709.
Ojemann, G., and Mateer, C., 1979, Human language cortex: localization of memory, syntax,
and sequential motor-phoneme identification systems, Science 205:1401–1403.
Ono, M., Kubik, S., and Abernathey, C. D., 1990, Atlas of the Cerebral Sulci, Stuttgart, New York:
Thieme.
Parker, G. J. M., Stephan, K. E., Barker, G. J., Rowe, J. B., MacManus, D. G., Wheeler-Kingshot,
C. A. M., Ciccarelli, O., Passingham, R. E., Spinks, R. L., Lemon, R. N., and Turner, R.,
2002, Initial demonstration of in vivo tracing of axonal projections in the macaque brain
and comparison with the human brain using diffusion tensor imaging and fast marching
tractography, Neuroimage 15:797–809.
Passingham, R. E., Stephan, K. E., and Kötter, R., 2002, The anatomical basis of functional
localization in the cortex, Nature 3:606–616.
Paus, T., Tomaiolo, F., Otaky, N., Macdonald, D., Petrides, M., Atlas, J., Morris, R., and Evans,
A. C., 1996, Human cingulate and paracingulate sulci: pattern, variability, asymmetry, and
probabilistic maps, Cereb. Cortex 6:207–214.
Penfield, W., and Rasmussen, T., 1950, The Cerebral Cortex of Man, New York: Macmillan.
Pieperhoff, P., Mohlberg, H., and Amunts, K., 2005, Überlagerungen anatomischer und funk-
tioneller Daten. In: Schneider, F., and Fink, G. R. (eds.), Funktionelle Kernspintomographie
in Psychiatrie und Neurologie, Berlin: Springer.
Poupon, C., Clark, C. A., Frouin, V., Régis, J., Bloch, I., LeBihan, D., and Mangin, J.-F., 2000,
Regularization of diffusion-based direction maps for the tracking of brain white matter
fascicles, Neuroimage 12:184–195.
Powell, H. W. R., Guye, M., Parker, G. J. M., Symms, M. R., Boulby, P., Koepp, M. J., Barker, G.
J., and Duncan, J. S., 2004, Noninvasive in vivo demonstration of the connections of the
human parahippocampal gyrus, Neuroimage 22:740–747.
Rademacher, J., Bürgel, U., Geyer, S., Schormann, T., Schleicher, A., Freund, H.-J., and Zilles,
K., 2001a, Variability and asymmetry in the human precentral motor system. A cytoarchi-
tectonic and myeloarchitectonic brain mapping study, Brain 124:2232–2258.
Rademacher, J., Bürgel, U., and Zilles, K., 2002, Stereotaxic localization, intersubject variability,
and interhemispheric differences of the human auditory thalamocortical system, Neuroim-
age 17:142–160.
Rademacher, J., Caviness, J., Steinmetz, H., and Galaburda, A. M., 1993, Topographical varia-
tion of the human primary cortices: implications for neuroimaging, brain mapping, and
neurobiology, Cereb. Cortex 3:313–329.
Rademacher, J., Morosan, P., Schormann, T., Schleicher, A., Werner, C., Freund, H. -J., and
Zilles, K., 2001b, Probabilistic mapping and volume measurement of human primary au-
ditory cortex, Neuroimage 13:669–683.
Riegele, L., 1931, Die Cytoarchitektonik der Felder der Broca’schen Region, J. Psychol. Neurol.
42:496–514.
Rivière, D., Mangin, J. -F., Papadopoulos-Orfanos, D., Martinez, J. M., Frouin, V., and Régis, J.,
2002, Automatic recognition of cortical sulci of the human brain using a congregation of
neural networks, Med. Image Anal. 6:77–92.
Roe, A. W., and Tso, D. Y., 1995, Visual topography in primate v2: multiple representations
across functional stripes, J. Neurosci. 15:3689–3715.
Roland, P. E., Svensson, P., Lindeberg, T., Risch, T., Baumann, P., Dehmel, A., Fredriksson,
J., Halldorson, H., Forsberg, L., Young, J., and Zilles, K., 2001, A database generator for
human brain imaging, Trends Neurosci. 24:562–564.
Roland, P. E., and Zilles, K., 1994, Brain atlases—a new research tool, Trends Neurosci. 17:458–
467.
Roland, P. E., and Zilles, K., 1996, The developing European computerized human brain
database for all imaging modalities, Neuroimage 4:S39–S47.
Roland, P. E., and Zilles, K., 1998, Structural divisions and functional fields in the human
cerebral cortex, Brain Res. Rev. 26:87–105.
Sarkisov, S. A., Filimonoff, I. N., and Preobrashenskaya, N. S., 1949, Cytoarchitecture of the Human
Cortex Cerebri (Russ.), Moscow: Medgiz.
Schleicher, A., Amunts, K., Geyer, S., Kowalski, T., Schormann, T., Palomero-Gallagher, N., and
Zilles, K., 2000, A stereological approach to human cortical architecture: identification
and delineation of cortical areas, J. Chem. Neuroanat. 20:31–47.
Schleicher, A., Amunts, K., Geyer, S., Kowalski, T., and Zilles, K., 1998, An observer-independent
cytoarchitectonic mapping of the human cortex using a stereological approach, Acta Stereol.
17:75–82.
Schleicher, A., Amunts, K., Geyer, S., Morosan, P., and Zilles, K., 1999, Observer-independent
method for microstructural parcellation of cerebral cortex: a quantitative approach to
cytoarchitectonics, Neuroimage 9:165–177.
Schleicher, A., and Zilles, K., 1990, A quantitative approach to cytoarchitectonics: analysis of str-
uctural inhomogeneities in nervous tissue using an image analyser, J. Microsc. 157:367–
381.
Schmitt, O., Hömke, L., and Dümbgen, L., 2003, Detection of cortical transition regions uti-
lizing statistical analyses of excess masses, Neuroimage 19:42–63.
Schormann, T., Dabringhaus, A., and Zilles, K., 1995, Statistics of deformations in histology
and improved alignment with MRI, IEEE Trans. Med. Imaging 14:25–35.
Schormann, T., and Zilles, K., 1998, Three-dimensional linear and nonlinear transformations:
an integration of light microscopical and MRI data, Hum. Brain Mapp. 6:339–347.
Sereno, M. I., Dale, A. M., Reppas, J. B., Kwong, K. K., Belliveau, J. W., Brady, T. I., Rosen, B.
R., and Tootell, R. B. H., 1995, Borders of multiple visual areas in humans revealed by
functional magnetic resonance imaging, Science 268:889–893.
Shipp, S., Watson, J. D. G., Frackowiak, R. S. J., and Zeki, S., 1995, Retinotopic maps in human
prestriate visual cortex: the demarcation of areas V2 and V3, Neuroimage 2:125–132.
Singh, R. P., and Smith, D. J., 2003, Genome scale mapping of brain gene expression, Biol.
Psychiatry 53:1069–1074.
Sowell, E. R., Peterson, B. S., Thompson, P. M., Welcome, S. E., Henkenius, A. L., and Toga,
A. W., 2003, Mapping cortical change across the human life span, Nat. Neurosci. 6:309–
315.
Steinmetz, H., Staiger, J. F., Schlaug, G., Huang, Y., and Jäncke, L., 1995, Corpus callosum and
brain volume in women and men, Neuroreport 6:1002–1004.
Stensaas, S. S., Eddington, D. K., and Dobelle, W. H., 1974, The topography and variability of
the primary visual cortex in man, J. Neurosurg. 40:755.
Stephan, K. E., Zilles, K., and Kötter, R., 2000, Coordinate-independent mapping of structural
and functional data by objective relational transformation (ORT), Phil. Trans. R. Soc. Lond.
B 355:37–54.
Stöckel, M. C., Weder, B., Binkofski, F., Choi, H. -J., Amunts, K., Pieperhoff, P., Shah, N. J.,
and Seitz, R., 2004, Left and right parietal lobule in tactile object discrimination, Eur. J.
Neurosci. 19:1067–1072.
Talairach, J., and Tournoux, P., 1988, Coplanar Stereotaxic Atlas of the Human Brain, Stuttgart:
Thieme.
Tanji, J., and Kurata, K., 1989, Changing concepts of motor areas of the cerebral cortex, Brain
Dev. 11:374–377.
Thompson, P. M., Moussai, J., Zohoori, S., Goldkorn, A., Khan, A. A., Mega, M. S., Small, G.
W., Cummings, J. L., and Toga, A. W., 1998, Cortical variability and asymmetry in normal
aging and Alzheimer’s disease, Cereb. Cortex 8:492–509.
Thompson, P. M., Schwartz, C., Lin, R. T., Khan, A. A., and Toga, A. W., 1996, Three-dimensional
statistical analysis of sulcal variability in the human brain, J. Neurosci. 16:4261–4274.
Thompson, P. M., and Toga, A. W., 2002, A framework for computational anatomy, Comput. Vis.
Sci. 5:13–34.
Toga, A., 2003, Brain atlases for the 21st century, In: Ng, V., Barker, G. J., and Hendler, T. (eds.),
Psychiatric Neuroimaging, Ohmsha: IOS Press, pp. 9–15.
Toga, A. W., and Thompson, P. M., 2003, Mapping brain asymmetry, Nat. Rev. Neurosci. 4: 37–48.
Tootell, R. B. H., Mendola, J. D., Hadjikhani, N. K., Ledden, P. J., Liu, A. K., Reppas, J. B., Sereno,
M. I., and Dale, A. M., 1997, Functional analysis of V3A and related areas in human visual
cortex, J. Neurosci. 17:7060–7078.
Turner, R., Le Bihan, D., and Chesnick, A. S., 1991, Echo-planar imaging of diffusion and
perfusion, Magn. Reson. Med. 19:247–253.
Uylings, H. B. M., Sanz Arigita, E., de Vos, K., Smeets, W. J. A. J., Pool, C. W., Amunts, K.,
Rajkowska, G., and Zilles, K., 2000, The importance of a human 3D database and atlas for
studies of prefrontal and thalamic functions, In: Uylings, H. B. M., van Eden, C. G., de
Bruin, J. P. C., Feenstra, M. G. P., and Pennartz, C. M. A. (eds.), Progress in Brain Research,
Vol. 126, Amsterdam: Elsevier Science BV, pp. 357–368.
von Economo, C., and Koskinas, G. N., 1925, Die Cytoarchitektonik der Hirnrinde des erwachsenen
Menschen, Berlin: Springer.
Van Essen, D. C., 1985, Functional organization of primate visual cortex, In: Peters, A., and
Jones, E. G. (eds.), Cerebral Cortex, London: Plenum Press, pp. 259–329.
Van Essen, D. C., 2002, Surface-based atlases of cerebellar cortex in human, macaque, and
mouse. Ann. Ny. Acad. Sci. 978:468–479.
Van Essen, D. C., 2004, Surface-based approaches to spatial localization and registration in
primate cerebral cortex, Neuroimage 23:s97–s107.
Van Essen, D. C., 2005, A population-average, landmark- and surface-based (PALS) atlas of
human cerebral cortex, Neuroimage 28:635–662.
Van Essen, D. C., Drury, H. A., Dickson, J., Harwell, J., Hanlon, D., and Anderson, C. H., 2001,
An integrated software suite for surface-based analyses of cerebral cortex, J. Am. Med.
Inform. Assoc. 8:443–459.
Van Essen, D. C., Drury, H. A., Joshi, S., and Miller, M. I., 1998, Functional and structural
mapping of human cerebral cortex: solutions are in the surfaces, Proc. Natl. Acad. Sci.
95:788–795.
Vogt, C., and Vogt, O., 1919, Allgemeinere Ergebnisse unserer Hirnforschung, J. Psychol. Neurol.
25:292–398.
Walker, A. E., 1938, The Primate Thalamus, Chicago: University of Chicago Press.
Walters, N., Egan, G. F., Kean, M., Jenkinson, M., Kril, J. J., and Watson, J. D. G., 2003, In
vivo identification of human cortical areas using high resolution MRI: an approach to
structure–function correlation, Proc. Natl. Acad. Sci. 100:2981–2986.
Wheeler-Kingshott, C. A. M., Hickman, S. J., Parker, G. J. M., Ciccarelli, O., Symms, M. R.,
Miller, D. H., and Barker, G. J., 2002, Investigating cervical spinal cord structure using
axial diffusion tensor imaging, Neuroimage 16:93–102.
Wiesendanger, E., Clarke, S., Kraftsik, R., and Tardif, E., 2004, Topography of cortico-striatal
connections in man: anatomical evidence for parallel organization, Eur. J. Neurosci.
20:1915–1922.
Wohlschläger, A. M., Specht, K., Lie, C.-H., Mohlberg, H., Bente, K., Pietrzyk, U., Stöcker,
T., Zilles, K., Amunts, K., and Fink, G. R., 2005, Linking retinotopic fMRI mapping and
anatomical probability maps of human occipital areas V1 and V2, Neuroimage 15:73–82.
Wong-Riley, M. T. T., Hevner, R. F., Cutlan, R., Earnest, M., Egan, R., Frost, J., and Nguyen, T.,
1993, Cytochrome oxidase in the human visual cortex: distribution in the developing and
the adults brain, Vis. Neurosci. 10:41–58.
Woods, R. P., Grafton, S. T., Watson, J. D. G., Sicotte, N. L., and Mazziotta, J. C., 1998, Automated
image registration: II. Intersubject validation of linear and nonlinear tools, J. Comput. Assis.
Tomogr. 22:153–165.
Wree, A., Schleicher, A., and Zilles, K., 1982, Estimation of volume fractions in nervous tissue
with an image analyzer, J. Neurosci. Methods 6:29–43.
Yakovlev, P. I., and Lecours, A. R., 1967, The myelogenetic cycles of regional maturation of the
brain, In: Yakovlev, P. I. and Lecours, A. R. (eds.), Regional Development of the Brain in Early
Life, Oxford: Blackwell Scientific Publishing, pp. 3–70.
Young, J. P., Geyer, S., Grefkes, C., Amunts, K., Morosan, P., Zilles, K., and Roland, P. E., 2003,
Regional cerebral blood flow correlations of somatosensory Areas 3a, 3b, 1, and 2 in
humans during rest: a PET and cytoarchitectural study, Hum. Brain Mapp. 19: 183–196.
Young, J. P., Herath, P., Eickhoff, S., Choi, H.-J., Grefkes, C., Zilles, K., and Roland, P. E., 2004,
Somatotopy and attentional modulation of the human parietal and opercular regions, J.
Neurosci. 24:5391–5399.
Zilles, K., 1991, Codistribution of receptors in the human cerebral cortex, In: Mendelsohn, F.
A. O., and Paxinos, G. E. (eds.), Receptors in the Human Nervous System, San Diego: Academic
Press, pp. 165–206.
Zilles, K., Armstrong, E., Schleicher, A., and Kretschmann, H. J., 1988a, The human pattern of
gyrification in the cerebral cortex, Anat. Embryol. 179:173–179.
Zilles, K., Dabringhaus, A., Geyer, S., Amunts, K., Qü, M., Schleicher, A., Gilissen, E., Schlaug,
G., Seitz, R., and Steinmetz, H., 1996, Structural asymmetries in the human forebrain and
the forebrain of non-human primates and rats, Neurosci. Behav. Rev. 20:593–605.
Zilles, K., Eickhoff, S., and Palomero-Gallagher, N., 2003, The human parietal cortex: a novel
approach to its architectonical mapping, Adv. Neurol. 93:1–20.
Zilles, K., Hajos, F., Kalman, M., and Schleicher, A., 1991a, Mapping of glial fibrillary acidic
protein-immunoreactivity in the rat forebrain and mesencephalon by computerized image
analysis, J. Comp. Neurol. 308:340–355.
Zilles, K., Kawashima, R., Dabringhaus, A., Fukoda, H., and Schormann, T., 2001, Hemispheric
shape of European and Japanese brains: 3-D MRI analysis of intersubject variability, eth-
nical, and gender differences, Neuroimage 13:262–271.
Zilles, K., and Palomero-Gallagher, N., 2001, Cyto-, myelo-, and receptor architectonics of the
human parietal cortex, Neuroimage 14:S8–S20.
Zilles, K., Palomero-Gallagher, N., Grefkes, C., Scheperjans, F., Boy, C., Amunts, K., and Schle-
icher, A., 2002a, Architectonics of the human cerebral cortex and transmitter receptor fin-
gerprints: reconciling functional neuroanatomy and neurochemistry, Eur. Neuropsychophar-
macol. 12:587–599.
Zilles, K., Palomero-Gallagher, N., and Schleicher, A., 2004, Transmitter receptors and func-
tional anatomy of the cerebral cortex, J. Anat. 205:417–432.
Zilles, K., Schlaug, G., Matelli, M., Luppino, G., Schleicher, A., Qü, M., Dabringhaus, A., Seitz, R.,
and Roland, P. E., 1995, Mapping of human and macaque sensorimotor areas by integrating
architectonic, transmitter receptor, MRI and PET data, J. Anat. 187:515–537.
Zilles, K., and Schleicher, A., 1993, Cyto- and myeloarchitecture of human visual cortex and
the periodical GABA-A receptor distribution, In: Gulyas, B., Ottoson, D., and Roland,
P. E. (eds.), Functional Organization of the Human Visual Cortex, Oxford: Pergamon Press,
pp. 111–120.
Zilles, K., Schleicher, A., 1995, Correlative imaging of transmitter receptor distributions in
human cortex, In: Stumpf, W., and Solomon, H. (eds.), Autoradiography and Correlative
Imaging, San Diego: Academic Press, pp. 277–307.
Zilles, K., Schleicher, A., Langemann, C., Amunts, K., Morosan, P., Palomero-Gallagher, N.,
Schormann, T., Mohlberg, H., Bürgel, U., Steinmetz, H., Schlaug, G., and Roland, P. E.,
1997, A quantitative analysis of sulci in the human cerebral cortex: development, regional
heterogeneity, gender difference, asymmetry, intersubject variability and cortical architec-
ture, Hum. Brain Mapp. 5:218–221.
Zilles, K., Schleicher, A., Palomero-Gallagher, N., and Amunts, K., 2002b, Quantitative analysis
of cyto- and receptor architecture of the human brain, In: Mazziotta, J. C., and Toga, A.
(eds.), Brain Mapping: The Methods, Amsterdam: Elsevier, pp. 573–602.
Zilles, K., Schleicher, A., Rath, M., and Bauer, A., 1988b, Quantitative receptor autoradiography
in the human brain. Methodical aspects, Histochemistry 90:129–137.
Zilles, K., Werner, L., Qü, M., Schleicher, A., and Gross, G., 1991b, Quantitative autoradiography
of 11 different transmitter binding sites in the basal forebrain region of the rat—evidence
of heterogeneity in distribution patterns, Neuroscience 42:473–481.
19
Neuron and Network
Modeling
GIORGIO A. ASCOLI and
RUGGERO SCORCIONI
INTRODUCTION
DIGITAL MORPHOMETRY OF SINGLE NEURONS
Computer Acquisition
Digital Files
Dendritic Modeling
AXONAL CONNECTIVITY IN THE ELECTRONIC AGE
Semiautomated Vectorization
Derivation of Connectivity
Models of Axonal Navigation
BOTTOM-UP NETWORK MODELING
System-Level Boundaries and Virtual Stereology
Anatomically Realistic Neural Networks
PHYSIOLOGICAL RELEVANCE
Influence of Morphology on Neuronal Electrophysiology
Network Dynamics
Design Principles
CONCLUSIONS AND FUTURE PERSPECTIVES
APPENDIX
Simple Extraction of Morphometric Parameters with L-Measure
A More Complex Example
REFERENCES
Abstract: Computer technology constitutes a formidable asset in the acquisition,

manipulation, analysis, and modeling of neuroanatomical data. Single-cell arboriza-
tions can be digitally represented as a large number of connected cylinders. In
this form, neuronal structure is amenable to three-dimensional (3D) rendering,
extensive quantitative characterization, and computational modeling of biophysics,
GIORGIO A. ASCOLI • Krasnow Institute for Advanced Study and Psychology Depart-
ment, George Mason University, 4400 University Drive, MS2A1 Fairfax, VA 22030-4444
RUGGERO SCORCIONI • Krasnow Institute for Advanced Study, George Mason Uni-
versity, 4400 University Drive, MS2A1 Fairfax, VA 22030-4444
604
NEURON AND NETWORK MODELING 605
electrophysiology, outgrowth, network connectivity, and dynamics. This chapter de-
scribes the state of the art in neuron and network modeling, with particular emphasis
on the methods to acquire, analyze, and synthesize neuroanatomical data. Several
commercial and freeware systems are available to reconstruct neuronal morphology
in digital format, from a variety of preparations, either directly from the microscope
or off-line from captured images. The resulting, increasing amount of digital data
(and meta-data) can be archived and publicly distributed to maximize scientific im-
pact. This database enables continuing efforts in modeling dendritic branching of
neurons throughout the central nervous system, including cortex, cerebellum, and
spinal cord. The experimental acquisition of complete axonal projections from sin-
gle neurons poses additional challenges, which are only recently being overcome.
The combination of dendritic and axonal reconstructions (or models), together
with the surface and volumetric representation of the surrounding tissue, allows
the computational derivation of synaptic connectivity. Taken together, such models
constitute a powerful substrate for the implementation of large-scale, anatomically
realistic neural networks. These advances can be instrumental for the cross-scale
elucidation of the relationship between structure, activity, and function in the brain.
Keywords: algorithm, axon, computer, connectivity, dendrite, reconstruction, simu-
lation
I. INTRODUCTION
The mammalian brain is often referred to as the “most complex object

in the universe.” Indeed, the sheer number of cells and their connections
must be compounded with their exquisite organization, from the intricacy of
dendritic and axonal branching, to the specificity of the interactions among
neuronal classes. Facing such mighty complexity, neuroscientists have tra-
ditionally reverted to two levels of analysis. At the system level, descriptions
are typically qualitative, with interactions among functional components
simply tagged as “present” or “absent” (or perhaps “strong” and “weak”).
In contrast, quantitative characterization accompanies the reductionist ap-
proach to investigate ever more “elementary” components, from individual
cells to spines, synaptic densities, single receptors, their subunits, and indi-
vidual amino acids. Can the rigorous biophysical knowledge of cellular and
subcellular processes be synthesized at the network level?
Computer technology constitutes a formidable asset in the acquisition,
manipulation, analysis, and modeling of neuroanatomical data. Models have
played a fundamental role in most fields of science, including several sub-
disciplines within neurobiology. Neuroanatomy has somehow lagged be-
hind, as its cross-scale complexity prevented the intuitive development of
abstract theories. This deadlock can be now solved by adopting hardware
and software tools to render neuronal and network structures quantita-
tively accessible to our understanding. This chapter describes the state of
the art in neuron and network modeling, with particular emphasis on the
methods to acquire, analyze, and synthesize neuroanatomical data in digital
format.
606 GIORGIO A. ASCOLI and RUGGERO SCORCIONI
II. DIGITAL MORPHOMETRY OF SINGLE NEURONS
A. Computer Acquisition
Typical neuroanatomical experiments result in chemically processed tis-

sue mounted on a microscope slide. The corresponding observable micro-
scope image can only be further manipulated in a limited way. A key step
toward the flexible, quantitative, and extensive analysis and modeling of
these data consists of their computer acquisition or digitization. The result-
ing digital files represent data in numerical (machine-readable) format.
Among the essential elements of digital neuroanatomy (as of much of
neuroscience) are individual neurons. Ramon y Cajal pioneered the use of
camera lucida, or drawing tube, a system of mirrors mounted between the
microscope oculars and the stage, which allows the precise hand-tracing of
the specimen. With this method, still in use in many neuroscience labora-
tories, neuronal structure is captured on paper as a pencil drawing of its
two-dimensional (2D) projection.
Dendritic and axonal trees can be described in digital form as a series of in-
terconnected cylinders, each characterized by the three spatial coordinates
of the end point, the diameter, and the identity of the cylinder they are
attached to in the path to the soma. Several systems, alternative to camera
lucida, have been developed to acquire neuronal morphology directly in
digital form. The most widely adopted commercial system is MicroBright-
Field’s Neurolucida (www.microbrightfield.com). The Neurolucida setup
includes a computer–microscope interface, a motorized stage, and a com-
plete software suite (Glaser and Glaser, 1990). Similarly to the camera lucida
system, the user sees the computer’s monitor overlaid on the microscopic
image. However, instead of drawing with a pencil on paper, the user virtu-
ally draws the structure of interest with mouse clicks, and the digital file is
created in the computer memory in real time. Fine regulation of the focus
is logged as depth, yielding precise spatial information in all three dimen-
sions. The adjustable size of the mouse cursor determines the diameter of
the structure being traced. In addition, extended structures can be followed
continuously, thanks to the joystick-controlled horizontal movement of the
stage, virtually paralleled by corresponding moves on the drawing screen.
Neurolucida also allows the reconstruction of arborizations across multiple
serial sections (Fig. 19.1A).
An alternative to the online reconstruction of neurons at the microscope
is constituted by the semiautomated acquisition of a stack of (possibly tiled)
digital images, serially ordered by their depth (or focal plane). Digital re-
construction of neuronal morphology can then be carried out off-line. This
system allows lengthy reconstructions to be completed with minimal op-
eration of the (usually expensive) microscope, including from perishable
preparations such as fluorescence stains for confocal microscopy. In ad-
dition, image stacks can be postprocessed, e.g., by contrast optimization,
filtering, and deconvolution.
Figure 19.1. Computer-assisted digital reconstructions. (A) Screenshot from Micro-
BrightField’s Neurolucida r software for 3D neuron reconstruction. This image
shows neuron reconstruction superimposed on a live image from the microscope.
Additional windows provide accessory tools that assist with the neuron tracing are also
shown. Neuron reconstruction by Robin Price. (Courtesy of MicroBrightField, Inc.)
(B) Neuron morpho ImageJ plug-in screenshot. The top window contains a cerebel-
lar climbing fiber image from a Z-stack captured by optical microscopy. A portion of
this image is enlarged in the bottom window, showing the semimanual tracing oper-
ation. Each reconstructed tracing point (red lines) is represented as one row in the
inset window. Column entries for each point correspond to a progressive numerical
identity, type, x, y , and z coordinates, radius, and identity of the parent segment.
An additional powerful software for the off-line digital reconstruction of

neuronal morphology from image stacks is Neuron morpho (Fig. 19.1B),
a plug-in of the NIH-distributed imaging program ImageJ. Both ImageJ
(http://rsb.info.nih.gov/ij) and Neuron morpho (www.maths.soton.ac.uk/
staff/D’Alessandro/morpho) are freely available and run on all JAVA-
compatible platforms (including Windows, Linux, and MacOS). Another,
less widely used, system for digital reconstruction uses polynomial interpola-
tion to join 2D reconstructions from serial images (Wolf et al., 1995). Several
ongoing projects are also attempting to automate the digital reconstruction
process by pattern recognition (e.g., He et al., 2003; Rodriguez et al., 2003), an
extremely difficult but ultimately commanding step of progress in this field.
B. Digital Files
Digital neuronal morphologies can be displayed and inspected in

“pseudo-3D,” including angle views different from that originally imaged un-
der the microscope. Neurolucida offers its own rendering program, called
NeuroExplorer (Fig. 19.2). An extremely popular neuronal visualization
and editing software tool is Cvapp (Cannon et al., 1998), a freeware, JAVA-
based program that can be run both locally or through a web browser
(www.compneuro.org/CDROM/nmorph).
A major advantage of digital representation of neuronal structure is that
virtually any geometrical feature captured by the cylinder-based description
can be measured and statistically analyzed quickly, reliably, and precisely (see
also Appendix). Over 50 morphometric functions can be extracted from
single or multiple neuromorphological files with L-Measure (Scorcioni and
Ascoli, 2001), another JAVA program freely available both for download and
web-based usage (Fig. 19.2; www.krasnow.gmu.edu/L-Neuron).
Digital morphologies can be also used to implement anatomically
realistic simulations of neuronal biophysics and electrophysiology, e.g.,
with the popular NEURON environment (Hines and Carnevale, 2001;
www.neuron.yale.edu). A large collection of such models is available for
download and use (Migliore et al., 2003; http://senselab.med.yale.edu).
The computer acquisition of digital morphology is considerably labor in-
tensive (∼1 week-person per neuron). Thus, researchers willing to share
reconstructions with peers provide an invaluable service to the scientific
community (Gardner et al., 2003). Several electronic collections of neu-
ronal morphology are available for a variety of cell classes (reviewed in
Ascoli, 2002a; Turner et al., 2002). Although almost each archive of neu-
ronal reconstructions comes with its own unique file format (Ascoli et al.,
2001a), these can be easily interconverted using tools such as Cvapp and
L-Measure. Nevertheless, particular care must be taken in considering the
lab-idiosyncratic morphological characteristics, which can derive from spe-
cific experimental conditions and protocols, hardware and software setups,
and individual operators’ bias (Scorcioni et al., 2004).
Figure 19.2. Electronic tools for rendering and analyzing neuronal morphol-
ogy. (A) Cvapp display of a neuron from Markram’s neocortical database
(http://microcircuit.epfl.ch; LBC cell C300301B1 from layer 4). (B) Screenshot (in-
set) of the L-Measure web-based graphical user interface (see also Appendix). (C)
Screenshot from MicroBrightField’s NeuroExplorer (TM) software showing results
of quantitative morphological analyses and an interactive 3D graphical representa-
tion of a reconstructed neuron. (Courtesy of MicroBrightField, Inc.)
C. Dendritic Modeling
A single neuron can be represented in digital form by tens of thousands of

3D coordinates of its branching neurites. This structure can be modeled by
designing algorithms to generate synthetic neurons in virtual reality (Ascoli,
1999). The natural variability of neuronal anatomy within a given morpho-
logical class can be captured by stochastic simulations, in which nonidentical
virtual cells are generated in different runs of the model (provided that the
seed of random number generation is changed). If the parameters of the
algorithm have a straightforward geometric meaning, their statistical distri-
butions can be extracted directly from the populations of real neurons to
be modeled (Ascoli and Krichmar, 2000).
A seminal example of this approach is Burke’s diameter-based model of
dendrogram geometry (reviewed in Burke and Marks, 2002). In this algo-
rithm, dendrites sequentially elongate by a unitary length step, each time
sampling diameter dependent, experimentally derived, bifurcation and ter-
mination probabilities. If a bifurcation is sampled, two daughter segments
are attached at the next steps. If a termination is sampled, the growth of the
given branch stops. If neither a bifurcation nor a termination is sampled,
another dendritic segment is attached and the process repeats. Originally
developed to describe spinal motoneurons, variations of this model have
been successfully applied to cerebellar Purkinje cells (Ascoli et al., 2001b)
and hippocampal pyramidal cells (Donohue et al., 2002) as well (Fig. 19.3).
Models of dendritic morphology exclusively based on branch diameter
are generally under constrained. In other words, the simulated neurons
tend to display greater variability than observed in the real cells. In a recent
advancement, hidden variables (specifically, path distance and the number of
terminal tips, or degree) were exploited to address this issue. All model param-
eters were made dependent on the local values of the hidden variables, which
were updated at every step of the algorithm. The resulting hidden Markov
model successfully captured all relevant properties of dendrogram geometry
in hippocampal pyramidal cells (Samsonovich and Ascoli, in press).
An additional element in neuromorphological modeling is the spatial em-
bedding of dendrograms, i.e., the 3D orientation of dendrites. Dendrites can
be described as “pointing” in a given absolute direction, or in an orienta-
tion relative to the origin of their internal coordinates, i.e., the soma (Ascoli,
1999). A Bayesian method was recently introduced to measure the relative
contribution of these various components of tropism from experimental
data (Samsonovich and Ascoli, 2003). In all principal cell classes of the rat
hippocampus, it was found that the major (and only statistically significant)
component of systematic growth was away from the soma. Thus, the heavily
polarized shape of hippocampal pyramidal and granule cells may be solely
produced by the local orientation of the stems from the soma, which may
be genetically determined. As a result, a simple two-parameter model (only
specifying the amount of “push” away from the soma, and that of random
deflection) can surprisingly capture the emergent shape of these cell classes
Figure 19.3. Simulated dendritic morphologies. (A) Diameter-based model of a spinal

motoneuron. Scale bar: 1000 µm. (B) Diameter-based model of a cerebellar Purkinje
cell. Scale bar: 50 µm. (C) Hidden Markov model of a hippocampal CA3 pyrami-
dal neuron. Scale bar: 100 µm. (D) Hidden Markov model of a hippocampal CA1
pyramidal neuron. Scale bar: 100 µm. (E) Globally constrained model of a dentate
granule cell. Scale bar: 100 µm.
(Fig. 19.3). Thus, the hidden Markov model of dendrograms and the algo-
rithm of dendritic orientation together constitute a remarkably complete
description of neuronal morphology, which was recently also applied to
dentate granule cells (Samsonovich and Ascoli, in press).
An alternative model of granule cell morphology was developed based on
global constraints such as the position of terminations along the principal
component of the dendritic field (Winslow et al., 1999). While the resulting
shape coarsely reproduces the structure of real neurons (Fig. 19.3), algo-
rithms of this type cannot be taken (even metaphorically) as mechanistic
models of development, because real growing branches have access only to
locally expressed and stored signals, and not to global information regarding
the whole tree or the distal surrounding environment.
Nevertheless, algorithms based on the overall distribution of branching
probability against the number of bifurcations, even if a global termination
is externally imposed to the whole tree, can still be taken as descriptive models
of development, if they capture the temporal dynamics of neuronal growth

(Van Ooyen and Van Pelt, 2002). From this point of view, even local models
based on dendritic diameter must be considered “hidden,” since the param-
eter distributions are measured from adult shapes, and kept constant during
virtual growth. An extensive review of computational models of neuronal
outgrowth, with a discussion of the strengths, weaknesses, and biological
plausibility, has been recently published (Donohue and Ascoli, 2004).
III. AXONAL CONNECTIVITY IN THE ELECTRONIC AGE
A. Semiautomated Vectorization
Computer-assisted digital reconstructions of dendritic trees, albeit time-

consuming, have now become standard routine in modern cellular neu-
roanatomy laboratories. The axonal arborizations of projection neurons,
however, are simply too huge to be digitized in the same fashion. Only a
small number of such reconstructions have been successfully completed in
what amounts to a truly heroic effort of dedicated individuals. Camera lu-
cida tracings are still widely adopted to obtain a permanent graphic record
of axonal projections from single neurons. How can these data be converted
in digital form for improved analysis and modeling?
Pencil drawings can be computer-acquired with a high-resolution scanner,
and a segment representation of the tracings can be obtained with freely
available software (e.g., www.wintopo.com). Alternatively, camera lucida pro-
jections can be directly acquired in electronic form by using computer-
interfaced tablets (e.g., Gras and Killman, 1983). In both cases, the result
is a set of digitized but disjoined segments (for a technical comparison, see
Ewart et al., 1989).
Using a tablet-acquired data set from Tamamaki et al. (1988) as a test bed,
we have recently developed an algorithm to fully reconnect axonal arboriza-
tions in the same format as typically obtained with the techniques described
in section “Computer Acquisition” (Fig. 19.4). This implementation (which
can be also applied to scanned-in camera lucida paper-and-pencil drawings)
simply follows a nearest-neighbor strategy, taking into account the average
spread of axonal branches in the thickness of the serial sections (Scorcioni
and Ascoli, in press).
Using this semiautomated vectorization procedure, we reconstructed
eight complete axonal morphologies from individual neurons, including
at least one for each of the principal cell classes of the hippocampal forma-
tion: entorhinal cortex layer II stellate cells, dentate gyrus granule cells, CA3,
CA2, and CA1 pyramidal cells, and subicular neurons (Fig. 19.5). While the
amount of effort required to hand-trace these large arborizations on paper
or tablets is still quite considerable, the semiautomated image processing
now makes it feasible to obtain larger collections of digital axonal data from
each neuronal class.
Figure 19.4. Reconstruction of axonal trees from manual tracings. (A) Raw vector-
ized data. Note various disconnected segments (arrows). (B) Algorithmically con-
nected arborization. Long thin arrows indicate branches that have been joined by
an additional segment. Short thick arrows indicate “true” gaps that should remain
disconnected in the final reconstruction.
Figure 19.5. Montage of complete ipsilateral projections of one axon from each of
the principal cell classes of the hippocampal formation. (A) Lateral view. 1 (purple):
subicular pyramidal neuron; 2 (light blue): enthorhinal layer II spiny stellate cell;
3 (dark blue): CA3 pyramidal cell; 4 (red): CA2 pyramidal cell; 5–7 (light-to-dark
green): three (distal, medial, and proximal) CA1 pyramidal cells. (B) Horizontal
view. (C) Coronal view. (Raw data provided by Dr. N. Tamamaki.)
B. Derivation of Connectivity
A powerful application of digital neuronal reconstructions is constituted

by the computational derivation of the potential for connectivity among
cell classes. In particular, when axonal and dendritic arborizations share
the same anatomical space, it is possible to identify a minimum interaction
distance within which synapses could be established. A similar (and compu-
tationally equivalent) approach consists of defining a “sphere of influence”
around neurites, and analyzing their overlaps. In either case, given an “in-
put” and “output” arborization, it is possible to mathematically derive the
number and spatial distribution of potential synapses (e.g., Kalisman et al.,
2003).
The relevant geometric parameters for this analysis (e.g., interaction dis-
tance) can be estimated from experimental data, such as the size of dendritic
spines, the interbouton distance on axons, and the length of growth cones.
It is important to stress that this approach yields an estimate of the potential
for synaptic connectivity, rather than a direct number of synapses. This po-
tential can be regarded as an upper limit of the number of synapses, or as
the combinatorial pool of possible synapses, a subset of which is expressed at
any given time. In light of the anatomical plasticity of synaptic connections
(which are formed and eliminated continuously in at least some regions
of the cortex), this measure can be physiologically relevant in the study of
the cellular and network bases of learning and memory (Stepanyants et al.,
2004).
Potential connectivity is affected both by the intrinsic shape of afferent
and efferent cells and by their spatial distribution and orientation. Thus, this
type of analysis yields results that are cell-class specific, and can be used to
compare different types of neurons within an anatomical region, inferring
their possible functional roles (Stepanyants et al., 2004). This approach has
also been applied to elucidate the information processing in entire sensory
pathways of model systems down to synaptic level ( Jacobs and Pittendrigh,
2002). Alternatively, it is possible to estimate parameters of the spatial dis-
tribution of specific morphologies to ensure effective connectivity of the
network (Costa and Manoel, 2003).
Information of system-level connectivity among brain regions is currently
being collated in electronic databases, such as the Brain Architecture Man-
agement System (Bota and Arbib, 2004; http://brancusi.usc.edu/bkms).
The advances in computational neuroanatomy described in this and pre-
vious sections of this chapter will soon make it possible to create web-based
archives of neuronal connectivity at the cellular level (Ascoli and Atkeson,
in press).
C. Models of Axonal Navigation
Projecting neurons navigate long distances toward their target before ex-
pressing full arborizations in the neuropil (see, e.g., Fig. 19.5). Therefore,
computational models of axonal anatomy must include algorithmic descrip-
tions of pathfinding in addition to intrinsic structural determinants. Much
is known about the molecular correlates of axonal navigation (e.g., Dono-
hue and Ascoli, 2004). Nonetheless, the theoretical understanding of these
complex phenomena is still incomplete, and relatively little information has
been so far integrated in computational models.
Senft and Ascoli (1999) proposed a phenomenological model in which
axons navigated toward groups of neurons (or glia), turning and possibly
bifurcating depending on their local orientation relative to their target,
until they arrived within a given distance. At this point, axons started es-
tablishing synapses, again turning and bifurcating as necessary to optimally
interact with their local postsynaptic counterparts. After making synaptic

contact, the axonal branch in this model was temporarily inhibited from
further synapsing. This “refractory period” was a critical parameter of the
algorithm, which could discriminate among diverse axonal morphologies
such as perforant pathways, sprouting neurites, and climbing fibers (long-,
medium-, and short-lasting inhibition, respectively).
Earlier computational models concentrated on a mechanistic descrip-
tion of the biophysical processes underlying axonal movement, such as
filopodial dynamics (Buettner, 1995). Several studies have focused on the
mathematical description of chemical and cellular gradients as the main
guiding cue for axons (e.g., Goodhill, 1998). Increasing attention in com-
putational neuroanatomy is also being paid to the effect of competition
(both for external targets and for internal metabolic resources) among ax-
ons (Van Ooyen and Van Pelt, 2002). A recent model integrated gradient
navigation, axon–axon interaction, and the further influence of patterned
activity (Yates et al., 2004). Other relevant efforts include the attempt to
describe neuritic navigation and connectivity in 2D with a cell automata for-
malism (Segev and Ben-Jacob, 2000), and the introduction of cell fate mech-
anisms in the computational description of axonal pathfinding (Eglen and
Willshaw, 2002).
Notably, the relative scarcity of complete axonal reconstructions in digital
format prevents a rigorous statistical comparison of the simulated axonal
morphologies with the corresponding experimental data. In this sense, a
wealth of useful data may become available when the resolution of Diffusion
Tensor Imaging reaches the scale of individual axonal bundles (Mori, 2002).
IV. BOTTOM-UP NETWORK MODELING
A. System-Level Boundaries and Virtual Stereology
Axonal projections (and in some cases, dendritic trees as well) are typi-
cally affected by system-level geometric constrains, such as the shape of the
afferent and efferent nuclei and regions. Thus, in order to fully characterize
neuritic shape and neuropil connectivity, it is important to include in the
model a digital representation of the relevant tissue and layer boundaries.
These data can be acquired from neuroanatomical preparation in ways sim-
ilar to those described in “Computer Acquisition.” Suitable raw data include
high-resolution ex vivo microscopic magnetic resonance imaging, or µ-MRI
(see, e.g., Lester et al., 2002), cytostructural boundaries traced from intra-
cellular filling experiments (Fig. 19.6), or classic histochemical preparation
such as Nissl or myelin stains.
From serially traced system-level boundaries, it is possible to compute
continuous surfaces (rendered, e.g., as tiled triangles) and the correspond-
ing volumes (list of internal voxels). Both representations carry impor-
tant information. Surfaces often determine the orientation of axonal and
Figure 19.6. Cytostructural boundaries of the rat hippocampus traced from serial
sections. (A) Dorsolateral view. Dentate gyrus and Ammon’s horn are clearly visible
in both the left and right hippocampi. (B) Mediocaudal view. One of the hippocampi
is approximately displayed along its transversal axis, and the other one along its
longitudinal axis. (Raw data provided by Dr. N. Tamamaki.)
dendritic arborization, while cell bodies, synapses, and branch coordinates

can be virtually positioned in the appropriate volumes. Following this strat-
egy, a large number of reconstructed (or simulated) neurons can be as-
sembled in 3D (Scorcioni et al., 2002) to “recreate” regional anatomy from
cellular-level information (Fig. 19.7).
Figure 19.7. Large-scale model of the rat dentate gyrus showing one thousand granule
cells (green) over a surface reconstruction of the cellular layer (blue), and the axon
from a single spiny stellate cell (purple) projecting from layer II of the entorhinal
cortex. (A) Dorsomedial view. (B) Detail on one of the dentate blade endings. (C)
Detail from within the hilus. The volume is mostly empty as the granule cell axons
(mossy fibers) are not included in the visualization.
Large-scale neuroanatomical models such as those displayed in Fig. 19.7

can be used to compute synaptic connectivity (see section “Derivation
of Connectivity”), as well as to impose global constraints to models of
neuronal morphology (see “Dendritic Modeling” and “Models of Axonal
Navigation”). “Virtual slices” at arbitrary planes of orientation can be
explored to foster intuition and guide electrophysiological and anatomi-
cal experiments. In addition, basic stereological properties, such as spa-
tial occupancy and density of various subcellular components (e.g., den-
drites and axons), can be derived for each layer and position in the
virtual tissue. This approach has been applied to evaluate optimal mo-
toneuron packing in the spinal cord (Burke and Marks, 2002), to quan-
titatively analyze the basal forebrain corticopetal system (Zaborszky et al.,
2002), and the somatosensory cerebro–cerebellar and ascending auditory
pathways (Leergaard and Bjaalie, 2002). An important application of this
line of study will consist of the inclusion of glia and blood vessels in
considering the relationship between structure and function in neural
systems.
B. Anatomically Realistic Neural Networks
Biophysical models of single-cell electrophysiology now routinely include

a faithful description of neuronal morphology and account for the result-
ing functional compartmentalization (e.g., Lazarewicz et al., 2002a; Migliore
et al., 2003). In contrast, most artificial neural networks have grown increas-
ingly abstract, and retain almost none of the anatomical characteristics of the
brain regions they are supposed to represent. Recent efforts, however, have
concentrated on the development of anatomically realistic neural network
models.
Small networks can be assembled “by hand” out of individual cell models,
within the framework of existing modeling environments, such as NEURON
(e.g., www.physiol.ucl.ac.uk/research/silver a/neuroConstruct). Even with
massively parallel supercomputers, however, simulation of activity dynamics
at the level of subcellular electrophysiological mechanisms can be carried
out only for a limited number of neurons. Nevertheless, simple, compu-
tationally efficient formalisms exist to capture essential neuronal dynamics
(Izhikevich, in press), which can be in principle applied to real-scale network
models. The problem remains to automatically and efficiently assemble a
realistic anatomical network construct.
The cellular-level anatomy of the CA1 area of a hippocampal slice was sim-
ulated with the powerful ArborVitae software (Senft and Ascoli, 1999). Cell
bodies for a variety of morphological classes were distributed in layers, sub-
sequently warped to reproduce the natural folds of the rodent archicortex.
Dendrites were then virtually grown in 3D using approaches as described
in “Dendritic Modeling.” Finally, axons were made to navigate toward and
connect with targets specified according to the known wiring diagram of
area CA1. The whole simulation could be run, displayed, and saved in a
limited amount of time.
Recently, ArborVitae was augmented with the ability to “read in” digital
representations of system-level experimental data of layer surfaces and re-
gional volumes. Thus, cells can be distributed according to their precise
spatial location, while axons’ navigation can be simulated through a “real”
voxel substrate (Senft, 2002). Figure 19.8 shows an example of this appli-
cation to the simulation of a corticothalamic projection within an imaged
human brain.
Figure 19.8. An ArborVitae model of human corticothalamic projections. (A)

Reverse-contrast display of axons departing from a cortical sulcus and navigating
through the white matter toward the thalamus. (B) Zoom-in on the gray matter,
with several cortical neurons visible. Unique colors were assigned to each individual
cell and its respective processes. (Courtesy of Dr. S. L. Senft.)
The results of ArborVitae simulations can be used to model (offline) the
activity dynamics of neuronal populations. The results can then be reloaded
for interactive display and analysis. In fact, once the specific connectivity
among all cellular classes is obtained from anatomical data (see section
“Derivation of Connectivity”), neural network dynamics can be run without
the need to explicitly simulate and display the details of network structure
(e.g., Ascoli and Atkeson, in press). Three-dimensional arrangements of
neurons are however essential to reproduce system-level properties cap-
tured by imaging techniques, such as EEG and fMRI, as well as functional
interactions with other complex components (e.g., glia).
V. PHYSIOLOGICAL RELEVANCE
A. Influence of Morphology on Neuronal Electrophysiology
Paradoxical as it may sound to the readers of this book, many theoretical

neuroscientists question the need to consider anatomy in computational
models of the nervous system. In fact, many early computational models
of brain function, from the (sub)cellular to the system level, essentially
approximated neuroanatomy away. Recent mounting evidence (especially
from modeling studies), nevertheless, indicates that both neuronal mor-
phology and network connectivity play a critical role in shaping activity (and
thus, presumably, function).
At the cellular level, there is widespread consensus that the integrative
properties of dendrites in most neuronal classes are sculpted by their active
membrane properties (e.g., Migliore et al., 2003). Voltage-dependent chan-
nels, however, are not always uniformly distributed throughout the dendritic
trees, creating complex interactions that can only be investigated with nu-
merical simulations. For example, the peculiar bursting activity of CA3 pyra-
midal cells can be obtained with two distinct channel distributions, but the
corresponding subcellular mechanisms are drastically different (Lazarewicz
et al., 2002b).
If the distribution of channels is maintained uniform throughout the
dendritic trees, nearly all of the known neocortical spiking patterns can sim-
ply derive from the different morphologies of the various neuronal classes
(Mainen and Sejnowski, 1996). Specifically, the ability of action potentials
to forward- and back propagate in the neuronal dendrites of various mor-
phological classes is dramatically variable due to the geometrical difference
alone (Vetter et al., 2001). The topological structure of the trees also influ-
ences firing patterns (Van Ooyen et al., 2002).
Even the natural variability among individual neurons within a morpho-
logical class can heavily affect spiking dynamics. When simulated with the
same plausible distribution of membrane properties, a set of morpholog-
ically accurate CA3 pyramidal cells were shown to fire both regularly and
irregularly, with a wide frequency range between 1 and 100 Hz (Krichmar
et al., 2002). Such a morphological control of electrophysiological behavior

was robust with respect to the distribution of active channel and the sim-
ulation protocols (reviewed in Krichmar and Nasuto, 2002). Similarly, in
a combined experimental and computational study, Schaefer et al. (2003)
showed that temporal integration in neocortical pyramidal cells is affected
by the proximal branching pattern in apical trees.
As these biophysical mechanisms relating structure and activity at the sin-
gle neuron level are uncovered, it is important to critically consider the
corresponding subtlety in the representation of digital anatomy in elec-
trophysiological simulations (Lazarewicz et al., 2002a). Depending on both
experimental and simulation protocols, morphologies of the same class, re-
constructed in different laboratories, can yield more disparate firing prop-
erties than can morphologies of different classes, reconstructed in the same
laboratory (Scorcioni et al., 2004).
B. Network Dynamics
Since morphology affects the intrinsic excitability of individual neurons,

it can be expected to influence network dynamics as well. There are, how-
ever, multiple additional avenues of interaction between neuroanatomy and
network activity. For example, the spatial location of the neurons and the
axonal path length can determine the pattern of onset response latencies
(e.g., Kotter et al., 2002).
The typically laminated arrangement of fiber tracts in the cortex also
determines the synaptic position in the dendritic layers. This in turn corre-
lates with the electrotonic distance of the input signal from the soma. Even
an oversimplified model of passive integration can illustrate that specify-
ing this elementary level of anatomical information deeply changes nearly
all dynamical aspects of a recurrent network (Ascoli, 2003). Similar conclu-
sions can be drawn in less orderly and more abstract Hopfield-type networks
(Costa et al., 2003).
When multiple cell classes are considered within a subregion, the specific
pattern of their connectivity also powerfully modulate the input–output
function of neural network (Ascoli and Atkeson, in press). Thus, the func-
tion of a given subclass, and the robustness of its contribution to network
activity, can be inferred individually for each neuronal class of the subregion.
In this perspective, biological neural networks can be viewed as assemblies
of functional motifs. The internal anatomy of each motif determines its spe-
cific function. Likewise, the anatomy of the motif assembly determines the
overall network function.
It should be noted that neuronal structure correlates with and in fact de-
termines all of the above characteristics (time delays, electrotonic distances,
synaptic connectivity), as well as the intrinsic firing properties of each mor-
phology class (and individual cell). Thus, the effects of several of these
characteristics are likely to be strongly correlated in biological networks. It
is parsimonious to hypothesize that various levels of biophysical organiza-
tion, such as distribution of dendritic channels, branching pattern, layer
position of synapses, and class-specific interconnectivity, coevolved to ro-
bustly express the desired network functions. This coevolution may also
guarantee a certain degree of homeostatic balance in the resulting activity
dynamics.
Given the intricacy of these interactions, computational modeling is ex-
tremely useful to separate (at least in silico) and quantitatively examine the
contributions of each anatomical property to network dynamics. Anatomi-
cally realistic neural network models carry the potential to similarly investi-
gate several other mechanisms of interaction between structure and activity.
These include, but are not limited to, ephaptic interactions, intrinsic (or
extrinsic) electric field modulation of neuronal firing, chemical inhomo-
geneity in the extracellular medium, glia buffering, and control. Finally,
these effects should be expected to be compounded with the natural in-
terindividual variability of neural connectivity, and the related functional
and structural effects of (and on) synaptic plasticity.
C. Design Principles
A complementary approach to understanding the physiological relevance

of neuron and network anatomy consists of the analysis of the possible
principles underlying their structural design. For example, dendritic trees
can be observed to increase their space occupancy (i.e., elongate and sprout
additional branching) in response to deafferentiation, due, e.g., to lesioning
of the presynaptic cell population (Shetty and Turner, 1999). This experi-
mental observation could be interpreted by postulating that the principle
behind, or goal of (some of the characteristics of), the shape of dendrites is
the homeostatic formation of a given number of synapses. This hypothesis
can be further tested by altering the number of synapses with a different
extrinsic manipulation.
A popular assumption is that biological shape has evolved to optimize
the expression of its intended function while minimizing the metabolic or
structural cost, such as the total wiring length of input and output cables
(Cherniak et al., 2002). In particular, both the axonal branching pattern
(Mitchison, 1992) and the volumetric ratio between axon and dendrites
(Chklovskii et al., 2002) appear to be close to optimal in the mammalian
neocortex. The clustered organization of cortical connections reflect a key
topological characteristic of small-world networks (Hilgetag and Kaiser, in
press), resulting in highly efficient yield of functional connectivity despite a
limited physical connectivity.
The same design principles can be investigated at larger scales. For ex-
ample, the spatial placement of macroscopic components of the nervous
system (functional subregions in the cortex, or ganglia in invertebrates) can
be also compatible with optimal wiring and connectivity (e.g., Young and
Scannell, 1996). Similarly, the intricate spatial pattern of ocular columns

may correspond to the optimization of the trade-off between coverage and
continuity (Carreira-Perpinan and Goodhill, 2002).
VI. CONCLUSIONS AND FUTURE PERSPECTIVES
A central tenet of neuroinformatics is the digital representation and

archiving of (in principle) all relevant information about the nervous system
(Ascoli et al., 2003). This goal may constitute a powerful basis for the creation
of a large-scale, low-level model of brain structure, activity, and function.
Neuroanatomy is leading the pack of success stories in neuroinformatics
(Ascoli, 2002b). What is the rationale for envisioning a structural model of
the brain, down to the detail of dendritic morphology (Samsonovich and
Ascoli, 2002)?
Virtual experiments can be carried out quickly, reliably, safely, and inex-
pensively. In silico investigations can go beyond the boundary of wet lab
technical and physical limits (e.g., simultaneously recording from millions
of neurons). They allow the exploration of a large number of promising
questions, and optimal experimental conditions (only the best of which
to be implemented in a “real” experiment). Virtual experiments can also
examine the theoretical effect of each model parameter separately by pre-
cisely reproducing all other initial conditions. Finally, they limit the use of
ethically charged invasive procedures. A detailed, large-scale model of the
mammalian brain will also foster scientific education both at the basic and
advanced levels.
A simple estimation of the computational power necessary to handle the
structure and activity of billions of neurons and trillions of synapses may lead
to the conclusion that a truly realistic model of the brain at the cellular level
is implausible in any foreseeable future. However, such a model could be
dynamically computed piecewise with a multiscale strategy. When virtually
recording the overall activity of the cortex, no detail may be necessary about
dendritic spines in the cerebellum. Certainly, powerful computational and
statistical techniques will be required to exploit such a large-scale model, yet
such challenges in silico look less insurmountable than those faced when
envisioning the corresponding experiments in real brains.
Devil’s advocates will maintain that neuroscience is still too far from the
reach of such a grand goal. While this argument is difficult to disagree with, it
is relieving to look at the temporal growth of the GeneBank database. From
the start of the Human Genome Project (1986), it took 6 years to establish
acceptable guidelines, and almost 10 years to complete the yeast genome.
Yet only 3 years after that, an entire human chromosome was mapped. It
was only one additional year (2000) before all 23 human chromosomes were
completed. The clearly exponential graph now looks dramatically flat until
the incept of the quite recent boom. The creation of a realistic, large-scale
human brain model may follow a similar pattern.
Although the computational power of hardware and software is increas-
ing at an exponential rate, so is the amount of experimental data col-
lected in biomedical sciences in general, and neuroscience in particular.
In principle, much of these data need to be incorporated in the realis-
tic model. Neuroinformatics started when experimental neuroscience was
already a mature field. Is modeling catching up, or is the gap between rel-
evant published data and corresponding computational simulations ever
widening?
The answer to this question depends on what is meant by “relevant,”
i.e., what level of modeling computational neuroscientists are designing.
This, in turn, is defined by the type of scientific explanation being sought,
e.g., behavior in terms of genes, network rhythms in terms of neuronal
spiking properties, synaptic strength in terms of calcium buffering. In this
perspective, computational models become a constructive definition of our
quantitative understanding of the structure, activity, and function of the
nervous system. The “ultimate race” between experiments and models, then,
is along the fine line dividing data and knowledge.
APPENDIX
A. Simple Extraction of Morphometric Parameters with L-Measure
Common measurements extracted from neuronal morphology include

total neuronal length and minimum, average, and maximum dendritic di-
ameter. As a first basic example, a step-by-step guide is given, showing how
to extract these basic parameters with L-Measure. Only requirements are an
Internet connection and a JAVA-enabled browser.
1. Connect to the online version of L-Measure at www.krasnow.gmu
.edu/L-Neuron (click “L-Measure” in the left column, then click “On-
line Version”). A security window will appear to ask for access permis-
sion, click “Yes.”
2. From the panel “Function,” select “Length” in the top left box, and
then click “Add.” A new measurement “Length” will appear in the list
of functions to be measured on the right.
3. From the same panel, add function “Diameter.”
4. From the panel “Input,” open and add the neurons you wish to
analyze and measure. You can freely download electronic neuronal
sample from www.krasnow.gmu.edu/L-Neuron (click “Morphology
Database”).
5. From the panel “Go” click the “Go” button. L-Measure will list all ex-
tracted measurements in the bottom panel. For each selected function,
L-Measure displays six values:
a. Total sum: In case of “Length” this is the total sum of all segment
lengths, which represents the first desired measurement.
b. Compartments included: It lists how many compartments were in-

cluded in this measurement.
c. Compartments excluded.
d. Minimum value: It reports the minimum value for the specified
function. In the diameter row, this represents the second desired
measurement.
e. Average value: It reports the average value across all segments (the
third desired measurement).
f. Maximum value: In the diameter row, this represents the fourth
desired measurement.
g. Standard deviation.
B. A More Complex Example
This section illustrates a more complex example in which a Sholl dia-

gram of length vs. Euclidean distance is generated from basal dendrites
with a sphere radial increment of 50 µm. For this section, a pyramidal
cell in SWC file format is required, which can be freely downloaded from
www.krasnow.gmu.edu/L-Neuron (click “Morphology Database”).
1. Connect to the online version of L-Measure at www.krasnow.gmu
.edu/L-Neuron (click “L-Measure” in the left column).
2. From the “Specificity” panel, select the function “Type” and insert
the value “3” followed by the “=” radio button. Then click the “Add”
button (type = 3 in SWC files identifies basal dendrites).
3. From the top left box in the “Function” panel, select “Length.”
4. From the bottom left box, select “Euclidean Distance.”
5. Select “Width of Bins” from the bottom radio button, and insert a
value of 50.
6. Click the “Add” button. A new function named “Length vs. Euclidean
Distance” will be added to the right top panel.
7-A. In the “Input” panel, add the SWC neuronal reconstruction file.
8. In the “Go” panel, click the “Go” button. The bottom box will show
the resulting measurement table.
To obtain a graphical representation of the Sholl diagram, additional steps
are required together with a spreadsheet-like software, such as Microsoft
Excel. To obtain an Excel compatible output, insert the following extra step
into the above sequence:
7-B. In the “Output” panel, click the “Save As” button, choose a directory,
and write “example.xls.”
9. Double-click the produced “.xls” file. Microsoft Excel will automati-
cally recognize and open the selected file.
10. Within Excel, select the first two rows.
11. Click the “Chart Wizard” button.
12. Select “XY scatterplot.”
13. Press “finish.”
Acknowledgments. The authors are grateful to Drs. Stephen L. Senft and

Nobuaki Tamamaki, and to MicroBrightField, Inc., for supplying material
used in some of this chapter’s illustrations. Support was provided by R01
grants NS39600 ( jointly funded by NINDS, NIMH, and NSF under the Hu-
man Brain Project) and AG025633 as part of the NSF/NIH Collaborative
Research in Computational Neuroscience Program.
REFERENCES
Ascoli, G. A., 1999, Progress and perspectives in computational neuroanatomy, Anat. Rec.
257(6):195–207.
Ascoli, G. A., 2002a, Neuroanatomical algorithms for dendritic modelling, Network 13(3):247–
260.
Ascoli, G. A., 2002b, Computing the brain and the computing brain, In: Ascoli, G. A. (ed.),
Computational Neuroanatomy: Principles and Methods, Totowa, NJ: Humana Press, pp. 3–26.
Ascoli, G. A., 2003, Passive dendritic integration heavily affects spiking dynamics of recurrent
networks, Neural Netw. 16:657–663.
Ascoli, G. A., and Atkeson, J. C., 2005, Incorporating anatomically realistic cellular-level con-
nectivity in neural network models of the rat hippocampus, Biosystems. 79:173–181.
Ascoli, G. A., De Schutter, E., and Kennedy, D. N., 2003, An information science infrastructure
for neuroscience, Neuroinformatics 1(1):1–2.
Ascoli, G. A., and Krichmar, J. L., 2000, L-Neuron: a modeling tool for the efficient genera-
tion and parsimonious description of dendritic morphology, Neurocomputing 32–33:1003–
1011.
Ascoli, G. A., Krichmar, J. L., Nasuto, S. J., and Senft, S. L., 2001a, Generation, descrip-
tion, and storage of dendritic morphology data, Philos. Trans. R. Soc. Lond. B Biol. Sci.
356(1412):1131–1145.
Ascoli, G. A., Krichmar, J. L., Scorcioni, R., Nasuto, S. J., and Senft, S. L., 2001b, Computer
generation and quantitative morphometric analysis of virtual neurons, Anat. Embryol.
204(4):283–301.
Bota, M., and Arbib, M. A., 2004, Integrating databases and expert systems for the analysis of
brain structures: connections, similarities, and homologies, Neuroinformatics 2(1):19–58.
Buettner, H. M., 1995, Computer simulation of nerve growth cone filopodial dynamics for
visualization and analysis, Cell Motil. Cytoskeleton 32(3):187–204.
Burke, R. E., and Marks, W. B., 2002, Some approaches to quantitative dendritic morphology,
In: Ascoli, G. A. (ed.), Computational Neuroanatomy: Principles and Methods, Totowa, NJ:
Humana Press, pp. 27–48.
Cannon, R. C., Turner, D. A., Pyapali, G. K., and Wheal, H. V., 1998, An online archive of
reconstructed hippocampal neurons, J. Neurosci. Methods 84(1–2):49–54.
Carreira-Perpinan, M. A., and Goodhill, G. J., 2002, Development of columnar structures in
visual cortex, In: Ascoli, G. A. (ed.), Computational Neuroanatomy: Principles and Methods,
Cherniak, C., Mokhtarzada, Z., and Nodelman, U., 2002, Optimal-wiring models of neu-
roanatomy, In: Ascoli, G. A. (ed.), Computational Neuroanatomy: Principles and Methods,
Chklovskii, D. B., Schikorski, T., and Stevens, C. F., 2002, Wiring optimization in cortical circuits,
Neuron 34(3):341–347.
Costa Lda, F., and Manoel, E. T., 2003, A percolation approach to neural morphometry and
connectivity, Neuroinformatics 1(1):65–80.
Costa Lda, F., Barbosa, M. S., Coupez, V., and Stauffer, D., 2003, Morphological Hopfield
networks, Brain Mind 4:91–105.
Donohue, D. E., and Ascoli, G. A., 2005, Models of neuronal outgrowth, In: Koslow, S. H., and
Subramaniam, S. (eds.), Databasing the Brain: From Data to Knowledge, Wiley, New York, NY,
pp. 303–323.
Donohue, D. E., Scorcioni, R., and Ascoli, G. A., 2002, Generation and description of neu-
ronal morphology using L-Neuron: a case study, In: Ascoli, G. A. (ed.), Computational
Neuroanatomy: Principles and Methods, Totowa, NJ: Humana Press, pp. 49–70.
Eglen, S. J., and Willshaw, D. J., 2002, Influence of cell fate mechanisms upon retinal mosaic
formation: a modelling study, Development 129(23):5399–5408.
Ewart, D. P., Kuzon, W. M., Jr., Fish, J. S., and McKee, N. H., 1989, Nerve fibre morphometry:
a comparison of techniques, J. Neurosci. Methods 29(2):143–150.
Gardner, D., Toga, A. W., Ascoli, G. A., Beatty, J. T., Brinkley, J. F., Dale, A. M., Fox, P. T., Gardner,
E. P., George, J. S., Goddard, N., Harris, K. M., Herskovits, E. H., Hines, M. L., Jacobs, G.
A., Jacobs, R. E., Jones, E. G., Kennedy, D. N., Kimberg, D. Y., Mazziotta, J. C., Miller, P. L.,
Mori, S., Mountain, D. C., Reiss, A. L., Rosen, G. D., Rottenberg, D. A., Shepherd, G. M.,
Smalheiser, N. R., Smith, K. P., Strachan, T., Van Essen, D. C., Williams, R. W., and Wong,
S. T., 2003, Towards effective and rewarding data sharing, Neuroinformatics 1(3):289–295.
for image combining microscopy, Comput. Med. Imaging Graph. 14(5):307–317.
Goodhill, G. J., 1998, Mathematical guidance for axons, Trends Neurosci. 21(6):226–231.
Gras, H., and Killmann, F., 1983, NEUREC—a program package for 3D-reconstruction from
serial sections using a microcomputer, Comput. Programs Biomed. 17(1–2):145–155.
He, W., Hamilton, T. A., Cohen, A. R., Holmes, T. J., Pace, C., Szarowski, D. H., Turner, J. N.,
and Roysam, B., 2003, Automated three-dimensional tracing of neurons in confocal and
brightfield images, Microsc. Microanal. 9(4):296–310.
Hilgetag, C. C., and Kaiser, M., 2004, Clustered organisation of cortical connectivity, Neuroin-
formatics 2:353–360.
Hines, M. L., and Carnevale, N. T., 2001, NEURON: a tool for neuroscientists, Neuroscientist
7(2):123–135.
Izhikevich, E. M., 2004, Which model to use for cortical spiking neurons? IEEE Trans. Neural
Netw. 15:1063–1070.
Jacobs, G. A., and Pittendrigh, C. S., 2002, Predicting emergent properties of neuronal en-
sembles using a database of individual neurons, In: Ascoli, G. A. (ed.), Computational
Neuroanatomy: Principles and Methods, Totowa, NJ: Humana Press, pp. 151–170.
Kalisman, N., Silberberg, G., and Markram, H., 2003, Deriving physical connectivity from neu-
ronal morphology, Biol. Cybern. 88(3):210–218.
Kotter, R., Nielsen, P., Dyhrfjeld-Johnsen, J., Sommer, F. T., and Northoff, G., 2002, Multi-level
neuron and network modeling in computational neuroanatomy, In: Ascoli, G. A. (ed.),
Computational Neuroanatomy: Principles and Methods, Totowa, NJ: Humana Press, 359–382.
Krichmar, J. L., and Nasuto, S. J., 2002, The relationship between neuronal shape and neuronal
activity, In: Ascoli, G. A. (ed.), Computational Neuroanatomy: Principles and Methods, Totowa,
NJ: Humana Press, pp. 105–126.
Krichmar, J. L., Nasuto, S. J., Scorcioni, R., Washington, S. D., and Ascoli. G. A., 2002, Effects of
dendritic morphology on CA3 pyramidal cell electrophysiology: a simulation study, Brain
Res. 941(1–2):11–28.
Lazarewicz, M. T., Boer-Iwema, S., and Ascoli, G. A., 2002a, Practical aspects in anatomically
accurate simulations of neuronal electrophysiology, In: Ascoli, G. A. (ed.), Computational
Neuroanatomy: Principles and Methods, Totowa, NJ: Humana Press, pp, 127–148.
Lazarewicz, M. T., Migliore, M., and Ascoli, G. A., 2002b, A new bursting model of CA3 pyramidal
cell physiology suggests multiple locations for spike initiation, Biosystems 67:129–137.
Leergaard, T. B., and Bjaalie, J. G., 2002, Architecture of sensory map transformations: axonal
tracing in combination with 3-d reconstruction, geometric modeling, and quantitative
analyses, In: Ascoli, G. A. (ed.), Computational Neuroanatomy: Principles and Methods, Totowa,
NJ: Humana Press, pp. 199–218.
Lester, D. S., Hanig, J. P., and Pine, P. S., 2002, Quantitative neurotoxicity, In: Ascoli, G. A. (ed.),
400.
Mainen, Z. F., and Sejnowski, T. J., 1996, Influence of dendritic structure on firing pattern in
model neocortical neurons, Nature 382(6589):363–366.
Migliore, M., Morse, T. M., Davison, A. P., Marenco, L., Shepherd, G. M., and Hines, M. L., 2003,
ModelDB: making models publicly accessible to support computational neuroscience,
Neuroinformatics 1(1):135–139.
Mitchison, G., 1992, Axonal trees and cortical architecture, Trends Neurosci. 15(4):122–126.
Mori, S., 2002, Principle and applications of diffusion tensor imaging: a new MRI technique
for neuroanatomical studies, In: Ascoli, G. A. (ed.), Computational Neuroanatomy: Principles
and Methods, Totowa, NJ: Humana Press, pp. 271–292.
Rodriguez, A., Ehlenberger, D., Kelliher, K., Einstein, M., Henderson, S. C., Morrison, J. H.,
Hof, P. R., and Wearne, S. L., 2003, Automated reconstruction of three-dimensional neu-
ronal morphology from laser scanning microscopy images, Methods 30(1):94–105.
Samsonovich, A. V., and Ascoli, G. A., 2002, Towards virtual brains, In: Ascoli, G. A. (ed.),
436.
Samsonovich, A. V., and Ascoli, G. A., 2003, Statistical morphological analysis of hippocampal
principal neurons indicates cell-specific repulsion of dendrites from their own cell, J.
Neurosci. Res. 71(2):173–187.
Samsonovich, A. V., and Ascoli, G. A., 2005, Statistical determinants of dendritic morphol-
ogy in hippocampal pyramidal neurons: a hidden Markov model, Hippocampus 15:
166–183.
Samsonovich, A. V., and Ascoli, G. A., 2005, Algorithmic description of hippocampal granule
cell dendritic morphology, Neurocomputing 65–66:253–260.
Schaefer, A. T., Larkum, M. E., Sakmann, B., and Roth, A., 2003, Coincidence detection in pyra-
midal neurons is tuned by their dendritic branching pattern, J. Neurophysiol. 89(6):3143–
3154.
Scorcioni, R., and Ascoli, G. A., 2001, Algorithmic extraction of morphological statistics from
electronic archives of neuroanatomy, Lect. Notes Comp. Sci. 2084:30–37.
Scorcioni, R., and Ascoli, G. A., 2005, Algorithmic reconstruction of complete axonal arboriza-
tions in rat hippocampal neurons, Neurocomputing 65–66:15–22.
Scorcioni, R., Boutiller, J. M., and Ascoli, G. A., 2002, A real scale model of the dentate gyrus
based on single-cell reconstructions and 3D rendering of a brain atlas, Neurocomputing
44–46:629–634.
oratories, J. Comp. Neurol. 473(2):177–193.
Segev, R., and Ben-Jacob, E., 2000, Generic modeling of chemotactic based self-wiring of neural
networks, Neural Netw. 13(2):185–199.
Senft, S. L., 2002, Axonal navigation through voxel substrates: a strategy for reconstructing
brain circuitry, In: Ascoli, G. A. (ed.), Computational Neuroanatomy: Principles and Methods,
Senft, S. L., and Ascoli, G. A., 1999, Reconstruction of brain networks by algorithmic amplifi-
cation of morphometry data, Lect. Notes Comp. Sci. 1606:25–33.
Shetty, A. K., and Turner, D. A., 1999, Aging impairs axonal sprouting response of dentate
granule cells following target loss and partial deafferentation, J. Comp. Neurol. 414(2):238–
254.
wiring, Neuron. 43(2):251–259.
Tamamaki, N., Abe, K., and Nojyo, Y., 1988, Three-dimensional analysis of the whole axonal
arbors originating from single CA2 pyramidal neurons in the rat hippocampus with the
aid of a computer graphic technique, Brain Res. 452(1–2):255–272.
Turner, D. A., Cannon, R. C., and Ascoli, G. A., 2002, Web-based neuronal archives: neuronal
morphometric and electrotonic analysis, In: Kotter, R. (ed.), Neuroscience Databases—A
Practical Guide, Amsterdam: Elsevier, pp. 81–98.
Van Ooyen, A., Duijnhouwer, J., Remme, M. W. H., and Van Pelt, J., 2002, The effect of dendritic
topology on firing patterns in model neurons, Network 13:311–325.
Van Ooyen, A., and Van Pelt, J., 2002, Competition in neuronal morphogenesis and the devel-
opment of nerve connections, In: Ascoli, G. A. (ed.), Computational Neuroanatomy: Principles
and Methods, Totowa, NJ: Humana Press, pp. 219–244.
Vetter, P., Roth, A., and Hausser, M., 2001, Propagation of action potentials in dendrites depends
on dendritic morphology, J. Neurophysiol. 85(2):926–937.
Winslow, J. L., Jou, S. F., Wang, S., and Wojtowicz, J. M., 1999, Signals in stochastically generated
neurons, J. Comput. Neurosci. 6(1):5–26.
Wolf, E., Birinyi, A., and Pomahazi, S., 1995, A fast three-dimensional neuronal tree recon-
struction system that uses cubic polynomials to estimate dendritic curvature, J. Neurosci.
Methods 63:137–145.
Yates, P. A., Holub, A. D., McLaughlin, T., Sejnowski, T. J., and O’Leary, D. D., 2004, Com-
putational modeling of retinotopic map development to define contributions of EphA–
ephrinA gradients, axon–axon interactions, and patterned activity, J. Neurobiol. 59(1):95–
113.
Young, M. P., and Scannell, J. W., 1996, Component-placement optimization in the brain, Trends
Neurosci. 19(10):413–415.
tional anatomical analysis of the basal forebrain corticopetal system, In: Ascoli, G. A. (ed.),
198.
20
Functional Connectivity of
the Brain: Reconstruction
from Static and Dynamic Data
ZOLTAN NADASDY, GYORGY BUZSAKI, and
LASZLO ZABORSZKY
INTRODUCTION
THE BUILDING BLOCKS: NEURONS, CIRCUITRIES, AND ASSEMBLIES
Classes of Neurons
Basic Circuitry
Vertical Organization of Circuitries
Horizontal Organization of Circuitries
Columnar Organization of the Neocortex
Synaptic Reconstruction
Cell Assemblies
STATIC DATA
Connectivity Matrix
The Importance of 3D Reconstruction
Statistical Modeling
Databases
DYNAMIC DATA
Dynamic System Approach
Large-Scale Recording of Neuronal Populations
Reconstruction of Functional Connectivity
Effective Connectivity
CONCLUDING REMARKS
APPENDIX
Parametric Modeling of Neuroanatomical Data
Databases
Cross-Correlation Function
REFERENCES
ZOLTAN NADASDY • California Institute of Technology, Pasadena, CA 91125

GYORGY BUZSAKI AND LASZLO ZABORSZKY • Rutgers, The State University of
New Jersey, Newark, NJ 07102
631
632 ZOLTAN NADASDY, GYORGY BUZSAKI, and LASZLO ZABORSZKY
Abstract: The central nervous system is a single complex network connecting each
neuron through a number of synaptic connections. However, only a small fraction
of the total connections functionally link neurons together. If the smallest multineu-
ronal architecture within which functional links are established constitutes circuitry,
then what are the basic operating principles of these circuitries from which we can
understand both the composition and the dynamics of the larger networks? We argue
that a finite class of circuitries, the “basic circuitries,” can be identified as repeat-
ing structural motifs tightly associated with specific dynamics. Functional circuitries,
however, cannot be derived from the static architecture simply because they do not
obey structural borders. Fortunately, since the constituent neurons do act in synergy,
we can infer from the dynamics the minimal structural conditions that constitute
a circuitry. In this chapter, instead of giving a precise definition of the “basic cir-
cuitry,” we outline a set of methods that may elucidate such a definition. We argue
that since the concept of circuitry incorporates both dynamic and static features, un-
derstanding can be achieved through combining the structural and dynamic aspects
of the available data. We review methods of extracting functional information from
static data first. Next, we review methods of extracting structural information from
dynamic data. Ideally, these two approaches should converge and define circuitry
based on the fragile concept of functional connectivity.
Keywords: cell types, circuitry, databasing, functional connectivity, large-scale record-

ing, population statistics
I. INTRODUCTION
The ultimate objective of neuronal tract-tracing is to reveal the functional

architecture of the nervous system. Progress toward this objective must rely
on a precise definition of the architecture in order to successfully explain
and predict the activity flow within its circuitries. This inferential process,
however, is rather limited since the reconstruction or prediction of putative
dynamics based on an abstract network architecture requires simulations
that are extremely sensitive to small variations of a large number of param-
eters. Therefore, we propose that the coapplication of anatomical and elec-
trophysiological methods is essential for describing a functional architecture
of the nervous system. Although the inferential process of the electrophys-
iology and the neuroanatomy are quite opposite in nature, they support
and complement each other. To find a link between them, we introduce the
“structural and dynamic compactness” criterion, that is, to determine the
smallest multiple-neuronal cluster, which generates the shortest invariant
activity pattern. This practical definition of “circuitry” is sufficient to intro-
duce the problem; however, further qualification and classification must go
beyond the structural and temporal “compactness criteria,” one of the key
challenges of research for the next decade. This chapter will review con-
stituent elements and main organization principles of cortical circuitries in
the “The Building Blocks: Neurons, Circuitries, and Assemblies” section,
specifically circuitries of the isocortex and the hippocampus will be dis-
cussed (both addressed as cortex). Next we discuss a number of innovative
FUNCTIONAL CONNECTIVITY OF THE BRAIN 633
applications of neuroanatomy and electrophysiology related to circuitries
of the brain. We separately discuss methods related to morphological data
in the “Static Data” section and physiological data in the “Dynamic Data”
section. However, emphasis is put on the convergence and dialog between
the two approaches in the “Concluding Remarks” Furthermore, we restrict
our review to the rodent brain; however, the principles we outline can be
generalized to the mammalian brain.
II. THE BUILDING BLOCKS: NEURONS,

CIRCUITRIES, AND ASSEMBLIES
Despite the complexity and size differences between the invertebrate and
the vertebrate nervous systems, both consist of large-scale repetition of com-
pact architectural modules, which we denote as “basic circuitries.” In order
to define the basic circuitries within and across brains of different species,
we first describe the constituent elements, the basic neuron types, and their
specific connectivity pattern. The basic circuitry, which is uniform within
a given brain structure, varies across different structures depending upon
the computational needs. These basic circuitries, once we understand their
dynamics, will enable us to infer the architecture from their activity pattern.
First, we need to make a distinction between structural and func-
tional connectivity1 and dynamics. The relationship between structural and
functional connectivity is best understood if we decompose the large net-
work of the nervous system into the smallest multineuronal information-
processing subunit, or “motif.” Motifs are conceived as small directed graphs
of M-nodes within a large network. It has been shown through simulations
that the number of structural motifs derived from anatomical connections of
nervous systems in various species is smaller than that from random graphs
(Sporns and Kötter, 2004). In direct contrast, when considering effective
connections, real nervous systems show more functional motifs than do ran-
dom graphs. This may suggest that the nervous system tends to maximize
the number of functional motifs but minimize the number of structural mo-
tifs. However, when considering the diversity of activity patterns generated
by networks of different functional motifs, it turns out that the number of
dynamics is smaller than the number of functional motifs. Apparently, there
is redundancy by which different functional motifs generate the same ac-
tivity pattern (Prinz et al., 2004). Therefore, dynamics may be more closely
associated with architectures than to the functional connectivity. Defining
circuitries by the dynamics they implement may allow us to reduce the nec-
essary number of basic circuitries and simplify their classification. The price
for this reduction is that the morphological composition of these circuitries
may be rather complex.
1
We also make a distinction between functional connectivity pertaining to connections
between neurons as opposed to interareal connections inferred from functional imaging.
The prototypes of cortical circuitries are those that have been described
in the hippocampus (Somogyi et al., 1998). From architectural and devel-
opmental points of view, isocortical circuitries derive from the prototypical
circuitries with a certain complexity added. Specifically, while the circuitries
are relatively homogeneously distributed and coaligned in the hippocam-
pus, the neocortical organization is complex and is conceived as an expan-
sion of the hippocampal cytoarchitecture. This expansion involves three
types of topological transformations. The first is a layer multiplication that
adds a new set of interlaminar circuitries to the isocortex, nonexistent in
the hippocampus. According to one view, the laminar structure of isocortex
can be conceived as the unfolding of the hippocampus and superimpos-
ing of three subregions, the dentate, CA3, and CA1, as different layers but
at the same time preserving the connections (Watts and Thomson, 2005).
The second type of expansion of cortical development is a superposition
of the same circuitry within the same cortical layer and often within the
same volume, which makes reconstruction of the synaptic circuitry particu-
larly difficult (Somogyi et al., 1998). The third type of expansion is a radial
specialization that forms columns and selective reciprocal tangential con-
nections with functionally similar circuitries. This extension is responsible
for creating the patchy functional architecture of the neocortex.
A. Classes of Neurons
Neuronal circuitries, at the lowest level, consist of two mutually exclu-

sive classes of neurons, excitatory glutamatergic neurons (mainly pyramidal
cells), and GABAergic inhibitory interneurons.2 On the basis of an assess-
ment from the CA1 area of the hippocampus, the pyramidal-to-interneuron
ratio is 33:1 (Aika et al., 1994). Principal neurons are the main excitatory
projection neurons as they establish long-range connections and transfer
information between different structures. In the cortex, pyramidal cells are
reciprocally connected to the thalamus and to each other via axon collater-
als. In spite of their diverse axonal projection patterns, pyramidal cells show
a characteristic bipolar dendritic arborization which consists of apical and
basal dendritic tufts. In contrast, most cortical interneurons incorporate
a more diverse morphology. Interneurons establish differential reciprocal
connections with other interneurons. The taxonomy of interneurons is still
unresolved since the categories constructed based on histochemical stain-
ing, morphological features, termination sites, and firing patterns do not
mesh. According to two extreme viewpoints on interneuronal diversity, cell
types may either represent a finite set of discrete classes or blends of continu-
ous feature distributions (Gupta et al., 2000). Several inhibitory interneuron
types have been classified solely based on morphological features, such as
2
Among the few exceptions, the spiny stellate neurons in layer 4 are excitatory and considered
interneurons.
basket, small basket, nest basket, axo-axonic, spiny stellate, aspiny stellate,
Martinotti, double-bouquet, and a number of smaller classes (Gupta et al.,
2000). The inhibitory interneuron diversity seems to scale with the evolution
of the mammalian brain. While brain structures with long evolution history,
such as cerebellum, basal ganglia, and thalamus, consist of a few cell types,
structures that specialized later, such as hippocampus and neocortex, show
larger diversity. Moreover, interneurons show layer specificity in terminals
(Freund and Buzsaki, 1996; Somogyi and Klausberger, 2005) and circuitry
specificity. One speculation about interneuronal diversity is that it is the re-
sult of circuitry specialization. According to this view, GABAergic interneu-
rons are added to the glutamatergic neurons to serve specific functional
roles (Földy et al., 2005). This is consistent with their different origin from
pyramidal cells during the early development of the nervous system and a
migration path orthogonal to that of the pyramidal cells (Rakic, 1995). The
most accepted definition of interneuronal species takes several features into
account, such as the postsynaptic target, the layer specificity, and the expres-
sion of species-characteristic markers (Freund and Buzsáki, 1996; Maccaferri
and Lacaille, 2003; Somogyi and Klausberger, 2005).
B. Basic Circuitry
The next level of organization is the “basic circuitry.”3 All extrinsic and
intrinsic glutamatergic pathways terminate on both pyramidal cells and
GABAergic interneurons. Therefore, the basic circuitry is composed of pyra-
midal → pyramidal, pyramidal → interneuron, interneuron → pyramidal,
and interneuron → interneuron functional units. The pyramidal–pyramidal
excitatory connection is feed-forward if it connects two pyramidal cells be-
tween two areas or between subregions of the cortex (Fig. 20.1A). It is re-
current if axons project back to the same pyramidal cell population (Fig.
20.1B).4
Connections between neurons, in general, can be synaptic or electri-
cally coupled (gap junction). Both pyramidal and interneurons can mu-
tually connect through gap junctions that are instrumental for gamma and
higher frequency band network synchronization. Interneurons seem to elec-
trically couple only with the same subtypes (Gibson et al., 1999; Tamás et al.,
2000). Conversely, connections between the same type of interneurons are
often mutual, and the outcome of a steady-state input is an oscillation with
zero-phase lag synchrony (empirical result and modeling: Destexhe and
3
We would like to make a clear distinction between the concepts of basic circuitry and a
“canonical microcircuit” of the neocortex (Douglas et al., 1989). While the former denotes
the basic information processing circuit involving only a few pyramidal and interneurons, the
latter describes the basic architecture of an isocortical volume incorporating all six layers and
all prototypical connections between the known neuron types.
4
Very few autaptic axons, i.e., axons projecting back to the very same neuron, have been
observed (Tamas et al., 1997).
Figure 20.1. Most common examples of basic circuitry types. (A) Feed-forward exci-
tatory connection involves excitatory projection from a different group of neurons
(b, c ) terminating on the target neuron a. The excitatory input from (b, c ) must
precede the action potentials in a. The relative timing of events is indicated by pulses
and numbers. The flow of action potentials is indicated by arrows. (Examples: Schaf-
fer collateral system in the hippocampus, layer 2–3 pyramidal → layer 5 pyramidal
synapses in the isocortex.) (B) The recurrent or feedback excitation involves excita-
tory collaterals from a to other neurons of the same group (b, c ). The excitatory input
from a must precede the action potential in b and c , which in turn may cause a second
Babloyantz, 1993; modeling: Wang and Rinzel, 1993; Vreeswijk, 1996). The
other main class of basic circuitry is the connection between projection neu-
rons (pyramidal) and interneurons. Within an interneuron–principal cell
couple, the interneuron controls the probability of the principal cell gener-
ating an action potential (AP) for a given excitatory input. The interneuron–
principal cell connection can be feed-forward (Fig. 20.1E) or recurrent
(feedback) (Fig. 20.1D). It is feed-forward if the interneuron, activated by
an excitatory input, has an inhibitory effect on the target pyramidal neuron
←
Figure 20.1. (Cont.) wave of excitatory postsynaptic potential on neuron a. (Exam-
ples: Hippocampal CA3 recurrent collateral system, recurrent connections between
layer 2–3 pyramidal neurons in the isocortex.) (C) Inhibitory–inhibitory connection
between two GABAergic neurons. The sequence of action is important in order to
understand how mutual inhibition causes oscillation with zero-phase-lag synchrony.
When both neurons inhibit each other (phase 1), they both will be released from in-
hibition at the same time (Phase 2). Then they both elicit action potentials (Phase 3),
which in turn cause them to be inactive for the next period, and the cycle starts
over. Assuming sufficient depolarizing driving force, the phases of mutual inhibi-
tion and disinhibition generate a self-sustaining oscillation within the interneuronal
network. (Examples: basket cells in hilus, layer 2–3 interneurons in the isocortex.)
(D) Recurrent or feedback inhibition. In this case, the excitatory collateral projects
to interneurons, which in turn project back to the same neuron. The recurrent in-
hibition is evoked by an excitatory input on the glutamatergic neuron (Phase 1),
which generates an action potential. The action acts on the interneuron (Phase 2)
through a collateral axonal projection, which in turn evokes an inhibition and when
backprojected to the glutamatergic neuron (Phase 3) causes suppression of action
potentials in the next phase. This type of control is effective to decrease the probabil-
ity of action potentials of pyramidal neurons. (Examples: interneuron—pyramidal
connections in the CA3 area of the hippocampus and pyramidal cell—basket cell
feedback inhibition in the CA area of the hippocampus.) (E) Feed-forward inhibi-
tion. Distant excitatory inputs often projects to interneurons which terminate on
local pyramidal cells. Functionally, the distant excitatory input (Phase 1) precedes
the interneuron’s response (Phase 2), which causes a suppression in the target pyra-
midal cells (Phase 3). A classic example of this circuitry is the lateral inhibition, com-
mon contrast-enhancement mechanism in sensory structures. Small black spheres
are GABAergic terminals. Cone-shaped terminals are glutamatergic. Principal cells
are shown in yellow, and interneurons are shown black. (F) A highly reduced model
of isocortical circuitry can be conceived as the combination of above described
feed-forward (orange) and feedback (viola) circuitries within a shared volume. This
circuit repeats in each layer of the isocortex, often in juxtaposition and in superimpo-
sition with the same circuit within the same volume. Excitatory and inhibitory inputs
are arriving on the top from left as input 1 and right as input 2, respectively. Only a
single layer is featured. Connections between pyramidal neurons (p) and four differ-
ent GABAergic interneuron types are illustrated. The laminar segregation of their
terminals relative to the pyramidal cell is emphasized. Specifically, while the feed-
forward inhibitory interneurons (a and b ; such as double-bouquet, neurogliaform,
and bitufted cells) preferentially target apical dendrites, the basket cells terminate
perisomatically on the pyramidal neurons. Axo-axonic cell terminals, in contrast,
occupy the axon initial segment. In addition, interneurons establish extensive and
laminar interconnections amongst each other. The reciprocal excitatory–excitatory
connections are also extensive. (After Somogyi et al., 1998.)
prior to the common excitatory input activating the pyramidal neuron di-
rectly. The connection is recurrent if the interneuron, activated through
recurrent axon collaterals, feeds back to the same pyramidal cell with a
delay. Note that this morphological classification includes a dynamic crite-
rion, the relative timing of the inhibitory and excitatory inputs. Accordingly,
since pyramidal cells and interneurons are mutually interconnected within
a region, whether inhibition is feed-forward or feedback can only be deter-
mined based on the relative timing of inhibition and excitation. Methods
to resolve timing relationships will be discussed in Section 4C–D.
Different interneuron classes localize in specific hippocampal and cortical
layers, and they establish connections with specific layers. Their axon termi-
nation is coaligned with excitatory input. In addition to their layer-specific
distribution, different interneurons terminate selectively on specific parts
of the pyramidal neuron. A highly simplified model of isocortical circuitry
is shown in Fig. 20.1F. Only one excitatory and one inhibitory input is il-
lustrated. Three basic types of interneurons are featured: axo-axonic, feed-
forward inhibitory to excitatory interneuron (e.g., bistratified cell), and
feed-forward inhibitory–inhibitory interneurons (e.g., basket cells). The
feed-forward interneurons preferentially terminate on the apical dendrites
and middle range dendrites. Basket cells are part of the feedback, feed-
forward, and reciprocal inhibitions. Their terminals preferentially target
the pyramidal cell soma. Axo-axonic cell terminals occupy the axon initial
segment of pyramidal cells. The combination of convergent feed-forward
and feedback connections effectively imposes a complex temporal pattern
of inhibition on pyramidal cells through spatially segregated gating of the
excitatory input from basal or apical dendrites (Fig. 20.1C). It is assumed
that a concerted action of different types of interneurons is able to impose
a complex temporal pattern of hyperpolarizations on the pyramidal cell
(Somogyi and Klausberger, 2005).
C. Vertical Organization of Circuitries
Both the laminar arborization of the dendritic tree as well as the laminar
organization of axons are cell-type specific. Furthermore, the efficacy of
synapses on the postsynaptic cell is highly dependent on the spatial local-
ization of the terminals relative to the postsynaptic cell’s morphology. The
closer the terminal is to the axon initial segment, the more effective is the in-
hibitory conductance. Therefore, inhibitory interneurons, such as the axo-
axonic cells in the isocortex and the basket cells in the hippocampus, can
effectively suppress the pyramidal cell response regardless of the excita-
tory postsynaptic potential, while interneurons terminating on the apical
dendrite can suppress the integration of excitatory postsynaptic potential
selectively for a specific dendritic cluster or branch.
Two examples of how basic circuitry types combine within the same vol-
ume to form a functional unit are shown in Fig. 20.2A. The first is a complex
A B
feed-forward feedback
rate
EC/T
Schaffer c.
PP
associated
cell
feed-forward
CA3
feedback
onset
GABAA
basket cell
EC/Commissur.
feedback
feedback
Figure 20.2. Hippocampal circuitries conceived as the composition of basic cir-

cuitries. (A) Combination of inhibitory feed-forward and feedback circuitries. The
feed-forward circuitry is established through excitatory input connections from en-
torhinal cortex or thalamus terminating on perforant path (PP) associated cells that
inhibit pyramidal cells near the soma. This is convergent with the recurrent inhibi-
tion through basket cells that are coaligned with the Schaffer collateral/comissural
input and also terminate perisomatically. The design of the two convergent but spa-
tially segregated inhibitory sources suggest coordinated inhibition on the pyramidal
cells. (B) Another example of the hippocampal CA1 circuitries where the in vitro
physiological response has been clarified. Spatially segregated recurrent (feedback)
inhibition circuitries originate and act on the same pyramidal neuron. One is act-
ing on the apical dendrites, the other is perisomatic. During stimulation through
alveus, the inhibition rapidly shifts from the soma to the apical dendrites. While the
somatic inhibition acts in a time-dependent fashion, the distant dendritic inhibition
is prolonged and frequency dependent (Pouille and Scanziani, 2004).
feed-forward/feedback circuitry of hippocampal CA1 pyramidal cell, basket

cell, and perforant-path-associated cell. Basket cell terminals are coaligned
with the Shaffer collateral or comissural input, while the perforant-path-
associated interneuron is coaligned with the entorhinal/thalamic input.
The combination of the two types of inhibition, feed-forward and feedback,
exerts an effective control over the integration in the pyramidal cell as they
both act on the soma (Somogyi et al., 1998). The other example, repre-
senting a combined feedback inhibition, is also from the CA1 area of the
hippocampus (Fig. 20.2B). By using simultaneous somatic and dendritic
recordings in vitro, a rapid shift from somatic feedback to dendritic feed-
back inhibition can be observed (Pouille and Scanziani, 2004). The somatic
feedback most likely acts via basket cells. These observations suggest that two
spatially disjoint circuitries process input onset time and prolonged rate in
the same pyramidal cells separately. This simple circuitry, by utilizing the
somatodendritic dynamics of inhibition, may enable pyramidal cells to allo-
cate separate channels for processing time-encoded and firing-rate-encoded
information.
From the layer specificity of dendritic and axonal distributions, one can
derive quantitative models of circuitries. Even if synapses are not available
for direct observation, one can estimate the number of synaptic contacts by
applying Peters’ rule (Peters and Feldman, 1976). Consider, for example, a
cortical volume traced for two types of neurons, type i and type j . According
to Peters’ rule, given S uj the number of synapses in cortical layer u established
by presynaptic neurons of type j , the number of neurons in layer u, and D u
the summed length of all dendritic branches in layer u, one can calculate
the number of synapses S that all neurons of cell type j establish with the
apical dendrite of neuron i:
Siuj = S uj diu /D u .
This formula is based on the assumption that synapses distribute evenly.
Applying this formula, Binzegger et al. (2004) were able to calculate the num-
ber of synapses between inhibitory and excitatory neurons in the primary
visual cortex of a cat. The calculation led to a few surprising revisions of the
traditional circuitry diagram of area 17 (Gilbert, 1983; Gilbert and Wiesel,
1983; Szentagothai, 1978). For example, the most important circuitry in
area 17 was believed to consist of a high-bandwidth sensory feed-forward
pathway of X/Y afferents, originating from the dorsal lateral geniculate
nuclei (LGN) and terminating in layer 4 spiny stellate cells. What changed
this view was taking into consideration that layer 4 cells establish massive
excitatory connections with layer 2/3, 5, 6 pyramidal cells and feedback to
layer 4 through a recurrent loop. Based on a quantitative assessment of the
synapses, it became evident that the feed-forward pathway comprises only
21% of all excitatory connections (Binzegger et al., 2004). In contrast, at least
34% of connections are intrinsic, i.e., establishing long-range horizontal
connections within the given layer and interconnecting different columns
(Fig. 20.3). Similar relationships were found among inhibitory–inhibitory
and inhibitory–excitatory neurons.
The selectivity of excitatory and inhibitory connections appears to be
circuitry and layer specific within the same cortical area. For example,
feed-forward projections from pyramidal neurons preferentially target other
pyramidal neurons; however, feedback connections mainly target interneu-
rons (Watts and Thomson, 2005). More specifically, feed-forward projec-
tions from layer 4 to 3 and from layer 3 to 5 target pyramidal cells and to
lesser degree interneurons. “Feedback” projections from layer 5 to 3 and
from layer 3 to 4, on the other hand, mainly target interneurons.
Using one of the most powerful techniques, optical release of caged gluta-
mate in combination with intracellular or multiple extracellular recordings,
Callaway and colleagues demonstrated an intracolumnar fine-scale organi-
zation of layer 2/3 cortical neurons (Yoshimura et al., 2005). Specifically,
adjacent pairs of layer 2/3 pyramidal neurons that are connected to each
other share common input from layer 4. Conversely, those that are not
connected share negligible input. In contrast with this fine-scale specificity,
inhibitory or layer 5 excitatory input are all shared across layer 2/3 cells,
L1
L2/3
L4
L5
L6
wm
Figure 20.3. The cortical circuitry of excitatory connections based on synapse density
in cat visual cortex area 17 (Binzegger et al., 2004). The cylindrical volumes represent
cortical columns. Arrows are the local excitatory connections. The arrow thickness is
proportional to the number of synapses (total number of synapses between excitatory
neurons = 13.6 × 1010). Gray arrows represent connections between layers. Pink
arrows are within layer connections. Some of these connections project from other
columns. Note that the within layer excitatory connections outnumber the feed-
forward sensory connections (cortical layers numbered; wm, white matter).
regardless of whether they are connected or not. Whether this example of

a layer-specific fine-scale organization of neurons represents a functional
subnetwork independent of columnar compartmentalization remains to be
investigated.
D. Horizontal Organization of Circuitries
We must make a clear distinction between the vertical specialization of cir-

cuitry and the horizontal (short and long range) associational connections.
These associational connections interconnect different vertical circuitries,
regardless of whether these circuitries cluster according to columns or not.
The range of these horizontal connections vary from adjacent columns
(50–200 µm) to different cortical areas or different hemispheres (few cen-
timeters). Excitatory associational long-range connections localize broadly
between layer 3 and layer 6. Horizontal excitatory axon collaterals and in-
terneurons from layer 3 arborize in “patches” at distant excitatory targets
within layer 2/3 and 5. Furthermore, excitatory long-range connections

originate from layer 5 where local ascending connections also derive. Layer
6 pyramidal cells also send long-range horizontal and oblique collaterals.
Starting with the short-range horizontal connections, pyramidal cells in
cats’ and monkeys’ primary visual cortex (V1) send excitatory axons to
neighbor columns. Whether target specificity is selective with respect to
orientation columns is not clear. According to a study by Das and Gilbert
(1995), spikes from the neuron’s cell body spreading in a radius of 400–1100
µm radius and subthreshold activation extending as far as 3.2 mm in cat V1
preferentially target columns with similar orientation. In contrast, the same
authors found by measuring correlation between neuron pairs located in
optically imaged maps of V1 orientation columns that the strength of local
connections between cells is a graded function of lateral separation across
cortex, largely radially symmetrical and relatively independent of orienta-
tion preferences (Das and Gilbert, 1999). Collinear facilitation observed
between pyramidal cells with nonoverlapping receptive fields supports a
key role in contour integration (Li and Gilbert, 2002).
Complementary to excitatory connections, local inhibitory interneurons
may substantially contribute to the functional segregation and dynamic as-
sembly of orientation columns. Selective suppression of different orienta-
tion domains of adjacent columns may enhance the contour integration.
The first important difference between lateral inhibitory and excitatory con-
nections is that inhibition has a shorter (250–500 µm) radius. Morpholog-
ical reconstruction of large GABAergic basket cell axonal arborization and
superposition on local orientation maps obtained by optical imaging have
revealed selective targeting. Specifically, visual area 17 layer 4 clutch cells
(a subtype of basket cells) arborize isotropically near their cell body within
50 µm, restricted to the nearest adjacent columns (Budd and Kisvarday,
2001). However, axons beyond this core show highly domain-specific topog-
raphy (Kisvarday et al., 2002).
The long-distance horizontal connections, which extend beyond specific
cortical subregions, form a massive cortico-cortical network. Probably the
best studied such interareal network is the visual pathway where the most
complete functional connectivity map is available. This map revealed a net-
work of distributed hierarchical processing (Felleman and Van Essen, 1991)
based on the systematic mapping of long-range cortical associational con-
nections that have been published during the last few decades. It illustrates
the enormous effort to extract connectivity information from published
data. Mapping associative connections between other cortical areas, such as
the prefrontal cortex (Kötter et al., 2001; Rempel-Clower and Barbas, 2000),
is in progress. This type of mapping involves a combination of retrograde
tracing with electrophysiology because the range of connections is at the
super millimeter level and cannot be tracked from the same section. High-
resolution functional and optical imaging applying voltage-sensitive fluores-
cent dyes with confocal or two-photon microscopy may substantially facili-
tate the functional verification of the connections, especially the horizontal
associational and callosal connections (Chen-Bee et al., 2000; Grinvald and
Hildesheim, 2004; Petersen, in this volume).
E. Columnar Organization of the Neocortex
The organization module that integrates both vertical and horizontal con-
nections is the cortical column. The concept of columnar architectonics of
the cerebral cortex arose originally from the early physiological observa-
tions of Mountcastle (1957) of the vertical columnar organization of the
somatosensory cortex. This was soon followed by an analogous architec-
tural principle in the visual cortex found by Hubel and Wiesel (1959). The
columnar architectonic principle of the cortex has received crucial sup-
port from studying the callosal and associational connections in primates
by Goldman and Nauta (1977). The arborization spaces of callosal columns
are one order of magnitude larger (300 µm–3 mm) as compared to the
orientation columns of Hubel and Wiesel (1972). Even after transections
of large parts of the corpus callosum, the distribution of degenerated fibers
shows a discontinuous pattern: in coronal sections hourglass-shaped ter-
ritories containing massive degeneration are alternating with areas con-
taining little or no degenerated terminals (Zaborszky and Wolff, 1982).
Cortico-cortical associational connections show inhomogeneous distribu-
tion pattern similarly to callosal columns (Zaborszky, 2002). The systematic
studies by Burkhalter, Killackey, Malach, and more recently by Sakman and
their colleagues (Coogan and Burkhalter, 1993; Johnson and Burkhalter,
1997; Koralek et al., 1990; Lubke et al., 2000; Malach, 1994; Paperna and
Malach, 1991) in the rodent cortex and that in the prefrontal cortex in pri-
mates by Patricia Goldman-Rakic, Helen Barbas, and David Lewis (Barbas
and Rempel-Clover, 1997; Goldman-Rakic, 1984; Pucak et al., 1996) am-
ply confirmed the columnar nature of associational connections that can
be utilized to predict the hierarchical organization of cortico-cortical con-
nections as shown in the often cited diagram of Van Essen (Felleman
and Van Essen, 1991). The size of associational columns in primates com-
pared with the size of the associational columns in rats show a remarkable
congruence.
The idea of columnar organization of the neocortex5 is part of a more
general hypothesis of modular organization of the nervous system, a widely
documented principle of design for both vertebrate and invertebrate brains
(Szentagothai, 1983). Some of the main characteristics of the modular prin-
ciple are summarized in a review by Liese (1990) and in a superb book
on the anatomy and functions of cerebral cortex by Mountcastle (1998).
The following features can be enlisted: (1) Modules are local networks of
cells in any region of the CNS, containing one or more electrically compact
5
A more detailed discussion of the columnar–modular organization of the cortex is beyond
the scope of this chapter and the reader is referred to a recent review by Rockland (1998).
circuits active in a particular behavioral function. (2) Modules are dynamic

entities—modules, repeated iteratively within each larger structure, func-
tion independently, or they may act together when combined in groups
whose composition may vary from time to time. (3) Modules may differ
in cell type and number in internal and external connectivity and in the
mode of neuronal processing between different large entities; but within
any single entity, like the neocortex, they have a basic similarity of internal
design and operation, ranging in diameter from about 150 to 1000 µm.
(4) The neighborhood relations between connected subsets of modules in
different entities result in nested systems that serve distributed functions. A
cortical area defined in classical cytoarchitectural terms may belong to more
than one and sometimes to several such systems. (5) Modules may develop
through ontogenesis and phylogenesis by duplication of homeobox genes
(Allman, 1998).
Modules can often be anatomically differentiable from surrounding tis-
sue, for example, associational or callosal columns in the cortex using trac-
ing methods, the striosome–matrix compartments in the striatum using
AChE histochemical reaction (Graybiel and Ragsdale, 1978), or the ap-
plication of immunocytochemical and autoradiographic methods for the
presence of various transmitters and receptors (Gerfen, 1985). AChE stain-
ing also delineates patches in the superior colliculus that represent special
sites where information from various sensory modalities can be integrated
(Chevalier and Mana, 2000). In other brain regions, such as the auditory
nuclei (Malmierca et al., 1998), the pontine gray (Leergaard et al., 2000),
or the basal forebrain (Zaborszky et al., 2005b), computational anatomical
methods helped to reveal a clustered, putatively modular organization, de-
fined by patterns of connectivity. For the historical record, we reprint here
in Fig. 20.4 Szentagothai’s imaginative models about cortical modules in
which he envisioned to place the elementary circuitries of the neocortex in
repetitive spaces of callosal columns.
F. Synaptic Reconstruction
Precise assessment of the number of synapses has been done entirely

by using electron microscopy (EM) in 2D sections (see Avendano, this
volume). Full 3D reconstruction of neurons based on EM has only re-
cently became available (Fiala and Harris, 2001; see also the chapter by
Duque and Zaborszky, this volume). As a shortcut, using the correlation be-
tween dendritic spines and synapses, it is possible to estimate the number
of synapses based on two-photon microscopic reconstruction of neurons
(Yuste and Denk, 1995) and investigate the specificity of synaptic connec-
tions relative to random dendritic contacts (Kalisman et al., 2005). These
measurements provide the most reliable quantitative assessment that could
guide further calculations on the bandwidth at different components of the
circuitry, thus allowing reconstruction of the functional connectivity.
IV.
VI.
specific afferent
corticocorticalis afferens
inhibitory
filtering
Figure 20.4. The callosal and associational columns. (A) Arrangement of neurons and
local circuits in one cortical columnar module. Pyramidal cells: red; specific afferents:
blue; corticocortical afferents: green; inhibitory neurons: solid black; TKN: spiny
stellate excitatory neuron; GGS: inhibitory neuron (double-bouquet cell of Cajal)
connected to other inhibitory neurons. The effect of the Martinotti cell in L VI spread
upto L I. (B) Interconnections of associational and callosal columns. The output of
each column originates from pyramidal cells, their terminal axonal arborizations are
labeled green. Ipsilateral connections maximally up to 10 columns. The lower right
scheme shows some of the dynamic features: in L I and VI the excitation expands
the diameter of the column, in L IV, the inhibition shrinks the cylinder (Reprinted
with permission from Szentagothai and Rethelyi, 2002.)
Combining voltage-sensitive Ca2+ imaging with two-photon microscopy it

was possible to monitor activity of 100–1000 neurons from a 150 × 150 or
300 × 300 µm cortical volume of area 18 in cat and rat in vivo at a < 1-µm spa-
tial resolution, allowing unambiguous cell body identification (Ohki et al.,
2005). By improving the stability and signal-to-noise ratio of the voltage-
sensitive dyes, the temporal resolution of this technique can be reduced
to sub-second rates to achieve a dynamic imaging of whole populations of
neurons simultaneously.
For submicroscopic structures, such as gap junction, EM verification re-
mains to be necessary (Fukuda and Kosaka, 2000). However, the combina-
tion of precise morphology-based modeling with physiological level simula-
tions allows predictions about network dynamics to be made. For example,
multicompartmental modeling of the cellular morphology of interneurons
and pyramidal cells, including the number of synapses and various input cur-
rents, led Traub et al. (1999) to conjecture that gap junctions are necessary
for high-frequency synchronization, commonly observed in the hippocam-
pus and neocortex.
G. Cell Assemblies
The next level of organization is the cell assembly level (Hebb, 1949;
Wickelgren, 1999). Although the original definition of cell assembly was
purely conceptual, the empirical definition relies on both structural and
dynamic criteria. Structural criteria are the “overlapping set,” “sparse cod-
ing,” and “high density of excitatory connections,” and the dynamic criteria
are “persistence,” “completion,” and “Hebbian learning.” The morpholog-
ical substrate of cell assemblies and the mechanism by which neurons dy-
namically form functional groups are still undetermined. In general, a cell
assembly represents a coalition of neurons within which neurons act in a
synergistic fashion. These coalitions can be established in a topological or
a topographical basis in the spatial domain, as well as dynamic or static ba-
sis in the temporal domain. Neurons with correlated activity are likely to
group together during development and form a topographically compact
network. This type of cell assembly is static and supports stable functions over
a period of time. The other type of cell assembly is dynamic and may not
segregate into topographical cell clusters. This is typical in structures sup-
porting flexible associative function between cells such as the hippocampus
(Harris et al., 2003) or interface structures with a large input/output diver-
gence (higher sensory cortical areas). While topographical cell clustering is
not a defining feature of cell assemblies, temporal synergy is. On the other
hand, since temporal synergy often derives from the common input to the
constituent neurons, topographical and temporal compactness are usually
codetermined. Temporal compactness can be detected as coherence, while
spatial compactness is a morphological feature. Since cells with overlapping
dendritic arborization are likely to share input, they must respond to the
common input with a synergistic fashion (see section “Statistical Modeling”).
However, temporal compactness can be derived from the high interconnect-
edness of a group of neurons that generate the same spike pattern regardless
of the input. Therefore, cell assemblies created by the common input must
be distinguished from cell assemblies formed by the synergy of neurons. To
consider all these possibilities of neuronal ensemble formation, one needs
to combine anatomy with physiology. Fortunately, the combination of mor-
phological and dynamical features results in a finite set and, as we argue,
the lexicon of connection patterns and dynamics fall into several categories.
For example, neurons with overlapping dendritic trees sharing a common
input can either form a feed-forward or a recurrent network. If the output
of these cells reliably reflects the temporal structure of the input, then the
network is feed-forward, otherwise it is likely to be recurrent.
The ability to record simultaneous activity from a population of identi-
fiable neurons brings up important questions concerning the relationship
of anatomical connections between neurons (“structural connectivity”), the
observed correlations between the activity of different neurons (“functional
connectivity”), and the causal sequence (“effective connectivity”). It has
been considered for many years that several neuronal configurations are
compatible with the same observed firing correlations leading to the concept
of a simplest “effective” mechanism that can account for the data (e.g., Aert-
sen et al., 1989). Furthermore, the same observed activity pattern recorded
from a group of neurons may be underlain by different network connectiv-
ity, suggesting that the robustness and self-organization of activity patterns
are more important than the precise architecture (Prinz et al., 2004). Much
further work will need to be done before we can unequivocally specify the
relationships between structural and functional connectivity, the number of
their distinct configurations, and the potential benefit of redundancy at any
of these levels.
III. STATIC DATA
A. Connectivity Matrix
The static architecture of neuronal information processing relies on the

map of connectivity, i.e., the connectivity matrix. We will refer to this as
“static connectivity” in contrast with the “functional connectivity,” which
represents active connections, equivalent with the map of information flow,
defined by measuring spike-train to spike-train correlations (see section
“Dynamic Data”). Theoretically, one can construct an immense matrix in-
cluding all neurons as i indices and list the same neurons as j indices to
represent all the monosynaptically connected neuron pairs with the number
of synapses. Within this matrix, we should find numerous isolated hot spots
representing major hubs and symmetric blocks indicating long-range con-
nections as well as local reciprocal connections. More detailed analysis of the
matrix would resolve the primary associations between basic neuron types
(e.g., association of Purkinje and mossy cells in the cerebellum, or basket
cells and granule cells in the dentate area of the hippocampus), those that
we have described as circuitries. The next level of associations would indi-
cate static assemblies, those that represent functionally related groups, such
as ocular dominance columns of the visual cortex, barrels in the somatosen-
sory cortex of the rat, or glomeruli of the olfactory bulb. The next level of
associations would reveal brain regions, such as the thalamus, hippocam-
pus, colliculus-superior, etc., including functionally distinct cortical areas.
To construct such a matrix is beyond the current technological capabilities
and may not be feasible at all. However, what is technically feasible is the
regional mapping of connections. We refer to this as the connectivity map,
a small part of the theoretical connectivity matrix limited to a specific struc-
ture. This is within a reach of the current technology, since a database like
this could be developed incrementally (see section “Databases”). Detailed
regional mapping and tracing of connections of various brain areas over
a century have revealed the critical organizational features of these struc-
tures allowing to make generalizations and construct accurate models (see
chapter by Ascoli and Scorcioni, in this volume).
B. The Importance of 3D Reconstruction
In order to create connectivity maps, first one has to identify the ele-
ments of connections, such as dendrites, axons, spines, and synapses. There
are two basic approaches of extracting these structures from histochemi-
cally prepared slices. Both approaches are based on human operators to
recognize these elements; however, data registration has been substantially
accelerated by computer technology. One is based on image analysis and
the other is based on vectorial tracing. Both methods start with application
of specific markers (Amunts and Zilles, Ascoli and Scorcioni, and Bjaalie
and Leergard, this volume). The goal of image analysis is to recognize ele-
ments of connections from images. To date, there are a number of image-
enhancement methods, such as edge detection, contrast enhancement, and
texture analysis, that can aid or make the recognition of different structures
unsupervised (He et al., 2003; Rodriguez et al., 2003).
The other approach derives from the technique of camera lucida, which
is based on manual tracing of outlines under the microscope at different
levels of details, including cellular, population, and structure level. Origi-
nally, this method was introduced to trace the contours of cells and con-
nections with maximal precision. Today, under computerized microscope
control, the tracing of individual sections is still done by an operator; how-
ever, traces are registered with their X, Y , and focal depth coordinates to a
database by computer. The computer encodes each contour segment by a
vector in the 3D Cartesian coordinate system in addition to the coordinates
indexed by the actual section under the microscope. By combining these
sections, we can reconstruct the virtual 3D structure of an entire traced
neuronal system. The reconstruction may involve interpolation between ad-
jacent sections unless the sectioning was gapless. The result is a 3D vectorial
representation of the cells that may include, besides cell bodies, the corre-
sponding dendritic processes and eventually axons. One major advantage
of 3D representation in a Cartesian coordinate system is that it enables one
to view the data from different angles and virtually zoom in to any level of
detail. Furthermore, using a standard stereotactic coordinate system, the 3D
representation allows incremental development of a database by adding new
elements. The most widely adopted commercial system for vectorial data ac-
quisition is MicroBrightField’s Neurolucida r
(MicroBrightField Inc., VT;
see Ascoli and Scorcioni, this volume).
The two types of data representation, image and vector, are fundamen-
tally different and combining them is a major challenge in developing neu-
roanatomical databases for the future. Although the 3D vectorial represen-
tation is better suited for tracking neural processes across sections than the
image format, it does not automatically identify the connections. Synaptic
connections can only be verified with EM or physiology. Therefore, connec-
tions revealed by light microscopy are only putative and inferentially based
on a set of critical attributes. These attributes are synaptic boutons, den-
dritic spines, or the proximity of axons and dendrites. A less reliable but
still useful attribute is the overlapping dendritic arborization that indicates
shared input source since en passant axons are likely to establish contact
with all neurons with overlapping dendritic arbors.
Ideally, for a population database, one would like to map neurons along
with their cell body, dendrites, and axons in relation to other neurons and
structure outlines or other available morphological markers, such as cortical
layers. Axons, however, are difficult to trace due to their small diameter and
that they may traverse across multiple sections and depth planes. However,
an intermediary solution for reconstruction of the axonal tree is to scan cam-
era lucida paper-and-pencil drawings and apply an algorithm that follows a
nearest-neighbor strategy (Ascoli and Scorcioni, this volume). Because the
connectivity is not readily available from tracings, reconstruction of the con-
nectivity map of large populations of neurons requires using a few inferential
heuristics. Such heuristics are the following: (a) neurons that group together
are likely to be functionally related; (b) neurons with overlapping dendritic
arborization are likely sharing input; (c) spatial association of chemically
identified cell types reflects functional synergy. Guided by these heuristics,
methods have been developed to extract and visualize hidden association
of neurons (Stepanyants et al., 2004). The first group of methods is based
on cell body distribution; the second is based on dendritic morphology.
C. Statistical Modeling
When a population of cell bodies have been selected with an unbiased

sampling, traced, and registered in 3D, a valid question is whether the dis-
tribution of neurons suggests a pattern that further implicates functional
relatedness. Obviously, the anatomical relationship is only suggestive and a
functional relationship must be tested by physiological methods. The null
hypothesis (H0 ) is that neurons distribute evenly within the volume of inter-
est. In contrast, if neurons cluster, we need to reject H0 , which again does
not necessarily imply any functional relatedness among neurons of the same
cluster. One algorithm of testing for homogeneity is the following:
1. For digitization, design a set of grids with linearly increasing grid sizes.
This grid serves to partition the total volume of interest into voxels.
2. Apply the largest grid size that is smaller than the smallest possible
cluster size.
3. Count the number of cell bodies within each voxel. Cell counts repre-
sent local densities.
4. Impose a density threshold. Voxels with local density larger than thresh-
old can be visualized by surface rendering.
5. Employ the next larger grid size and recursively apply steps 3–5 until
the grid is larger than the largest possible cluster size.
If H0 is correct, then the average cell density within the filled voxels
should increase linearly (that is, the total number of cells within the filled
voxels should increase exponentially) with the grid size, while the number of
filled voxels should also increase linearly. Deviation from the linear density
change, for example, a stepwise increase in average cell density, indicates
inhomogeneity. The precise cell density and grid size at which the largest
stepwise increase of density occurs is the one that corresponds to the critical
density and cluster size, respectively. This algorithm was applied to reveal
hidden clustering of cell bodies within the basal forebrain cholinergic sys-
tem (Nadasdy et al., submitted; Nadasdy and Zaborszky, 2001; Zaborszky
et al., 2002, 2005b). Having the critical density and size determined, one
can apply a grid size that fits to the cluster size and compute the local cell
density within the voxels constructed by this grid (Fig. 20.5B). Next, select
only those voxels where the density is equal to or larger than the critical
density. These voxels delineate putative neuronal clusters (Fig. 20.5E). To
visualize these clusters, apply a surface rendering on the selected voxels.
To eliminate sharp edges between adjacent polygons, it is recommended to
interpolate and smoothen the polygons before surface rendering is applied
(Fig. 20.5F). For construction of volumetric database and isodensity surface
rendering, see the Appendix..
Visualization of such clusters can place these clusters into the context
of the macro-architecture (Fig. 20.5D–F). We can take the density analy-
sis one step further and apply it to the association of different cell types.
For example, if the brain sections were histochemically stained for choline
acetyltransferase, parvalbumin, calretinin, and calbindin, then each of these
markers will label a subpopulation of neurons specifically associated with a
specific cell type. These populations can be considered either as four in-
dependent spatial distributions (H0 ) or a coordinated distribution of four
markers (HA ). One could ask the following question: Knowing the spatial
distribution of each marker population, what is the probability of finding
a parvalbumin and a cholinergic neuron within the same voxel by chance?
The combined by-chance probability is the product of the local probabilities
of occurrences of the two markers. If the observed coregistration probabil-
ity is larger than the expected by-chance coregistration of markers, then
we must reject H0 . Rejection of H0 does not necessarily imply functional
relatedness. Nonetheless, assuming that neurons within the same density
cluster share common input, HA is suggestive of a functional association of
different cell types (Zaborszky et al., 2005b).
Dendritic morphology, when combined with cellular density data, further
supports the inference from the spatial distribution to the connectivity map.
Utilizing the overlapping dendritic volume as an indicator for shared input
provides an additional attribute of functional cell clusters. To explore this
option, we used a large database of basal forebrain cholinergic neurons
(n = 15,700 neurons) and extracted an unbiased random sample of 750
neurons for dendritic morphology tracing. Next, we determined the 3D
orientation of the main axis of the dendritic tree and, for each individual
cell, we replaced the dendrite with a vector pointing from the cell body.
The vector’s orientation was identical to the orientation of the dendritic
Figure 20.5. Visualization of neuron clusters by using isodensity surface rendering.

(A) Microscopic image of immunohistochemically labelled neurons. (B) Camera lu-
cida reconstruction of neuron outlines with processes. (C) A grid is defined that
incorporates the size of clusters to be determined. Cell bodies are highlighted.
(D) Cell bodies enter with their voxel coordinates. (E) The number of cell bod-
ies within each 3D grid is determined and the local density is assigned to each
voxel. (F) Voxels at which the density exceeds a certain threshold create a volume.
This volume, encompassing the spatial structure of neuron clusters, can be visual-
ized by surface rendering and superimposed on clusters obtained from other cell
markers.
volume, estimated as the center of dendritic mass relative to the cell body,
and vector length was proportional to the dimension of the dendrites. Here,
one can use a different abstraction of parameters. The vector can represent
dendritic length, dendritic density, dendritic volume, etc. Alternatively, in-
stead of a single vector one can apply two vectors that better describe bipolar
dendritic arborizations. The importance of dendritic vector abstraction is
twofold. First, it allows one to estimate the overlapping of the volume that
a dendritic tree can sample, thus providing a quantitative assessment of
the shared input. Second, proportional enhancement of vectors allows a
visual representation of the dendritic orientation pattern in the context
of the cellular distribution pattern, which otherwise would not be seen at
a true microscopic scale. When neurons represented with their dendritic
vectors were projected to a 3D coordinate system and visualized relative to
the outlines of the cholinergic column, the dendritic orientation revealed
a systematic pattern (see Fig. 5 in Zaborszky et al., 2002). Along the septal
cholinergic volume, a spiral staircase pattern of dendritic orientation was
observed that was orthogonal to the orientation of the septal column. This
architecture suggested an optimal sampling of the en passant axons parallel
to the septal column (Somogyi et al., manuscript in preparation.
Although synapses with their respective postsynaptic target can reliably
be identified only by EM, the presence of synaptic marker proteins such
as synaptophysin or ProSAP2/Shank3 can give an estimation of synapses
at the population level using light microscopy. Also, using a tour de force
confocal regimen (see Wouterlood, this volume) putative synapses from 3D
reconstructions can be determined. Using different proximity scales, Stepa-
nyants and colleagues determined the number of “graphical” contacts, i.e.,
the potential synapses, between overlapping dendritic and axonal segments
(Stepanyants et al., 2004). Next, they shifted the coordinates of the dendritic
arbor in silico by ∼30 µm to establish a different set of potential synapses.
Their rationale was as follows. If the axons and dendrites establish contacts
on a by-chance basis, the shift should not affect the number of potential
synapses. By varying the shifts, it was possible to determine the necessary
shift beyond which the misalignment destroys the potential synapses to the
by-chance level. The precision at which axons establish potential synaptic
contacts was found to be consistent within the shaft dimension (0.4 µm)
and dendritic spine dimension (2.0 µm). Considering this metric, it was
also evident that the number of actual synapses represent only 10–30% of
the potential synapses, indicating significant room for plasticity to convert
potential synapses to actual (Stepanyants et al., 2002, 2004). Exploration of
the regional and laminar differentiation of potential synaptic contacts will
hopefully elucidate further details of cortical circuitries. The use of com-
putational and statistical methods could play a major role in exploiting the
richness of data available from the 3D morphology and population scale
reconstruction. Therefore, systematic construction and incremental com-
bination of morphological data collected by different laboratories using a
common data registration and database system is vital for these quantitative
methods to gain momentum.
D. Databases
To be able to derive synaptic connectivity from static neuronal popula-

tion data, we first need a representative sample of neurons which have been
identified and mapped in 3D. More than 100 years of systematic mapping of
a variety of structures and the morphological characterization of different
neurons have amounted to a vast data source, which have been partially ana-
lyzed and the results published. Unfortunately, most of the original data are
no longer available. It is now imperative to integrate independently gener-
ated data into one consistent database. The first problem is the data structure
of choice to incorporate the full scale of complexity of static neuronal data.
The language of the database must be vectorial to represent connections and
flexible enough to accommodate new aspects. Born from this motivation,
MorphML, a unified vectorial data structure has been designed for flexible
representation of various neuromorphological objects from the subcellular
level to the macroscopic scale morphology (http://www.morphml.org). The
purpose of this data format is multifold. First, it enables incremental data
integration across different laboratories and platforms. Second, it supports
visualization of anatomical data as a rendering tool, designed specifically
for this data format. Third, it provides a “core” data format that allows con-
versions between different morphological databases. Fourth, it supports the
import of morphology data into dynamic modeling environments and sim-
ulation software such as Neuron (Hines and Carnevale, 1997) and Genesis
(Bower and Beeman, 1998). Fifth, it could be maintained by industrial stan-
dard database management software (e.g., XML, IXL, Oracle). Currently,
data conversion modules allow conversion of the Neurolucida data format
to the 3D annotation system.
Arguably, it is now feasible to integrate different levels of morphological
data and construct a detailed 3D representation of large population of neu-
rons. With multiple-level data integration, it will soon be possible to reveal
parts of the connectivity matrix by incremental development of the connec-
tivity map based on morphological tracing. Furthermore, automatic data
acquisition, either image analysis based or vectorial tracing, is expected to
speed up data generation. Within the next 10 years, we anticipate that the
complete 3D vectorial database of the rat brain will be available to address
specific questions about hidden organization principles. A preliminary ver-
sion of a database that allows integration and analysis of different 3D data sets
collected at different levels and different laboratories that uses MorphML
format can be viewed at http://www.ratbrain.org/ (Zaborszky et al., 2005a).
More details on this database and a list of other databases or electronic brain
atlases can be found in the Appendix.
IV. DYNAMIC DATA
A. Dynamic System Approach
The inductive approach to understand network dynamics as a multiplica-

tion of single neuron operations fails for multiple reasons. First, inferences
about the connectivity matrix, a prerequisite of functional connectivity,
could not be made even with the complete neuromorphometric database.
Recovering the connectivity matrix would require mapping all synapses onto
each neuron, which is impractical. Secondly, given that all synaptic connec-
tions are mapped, calculating the range of dynamics exhibited by a connec-
tivity matrix is intractable. Specifically, when trying to solve all the differential
equations describing channel dynamics in multicompartmental models of
a population of neurons, the effect of small uncertainties and the sensitivity
for small perturbations scales with iterations so quickly that it could render
the state of the neurons unpredictable.
Fortunately, it is not necessary to know all the connections and solve
the differential equations for each compartment in order to predict the
dynamics of a given network of neurons. There is a shortcut. Simulations on
20 million versions of a three-cell model of the pyloric network suggest that
the number of possible different dynamics is much less than the possible
synaptic strength configurations (Prinz et al., 2004). Given the chance of
reduction, theoretically we can classify circuitry architectures based on the
invariant dynamics they generate. These dynamics at the mesoscopic6 level
may be robust enough to tolerate small differences in the subnetwork level
organization (Freeman, 2000). Using this classification we can infer from the
dynamics the architecture and vice versa. Recently, these dynamics became
empirically permeable by using large-scale high-resolution recording of a
population of neurons.
B. Large-Scale Recording of Neuronal Populations
1. Limitations and Constraints
To attain an empirical ground on large-scale network activity and capture

the mesoscopic dynamics, we need to record as many neurons simultane-
ously as possible. Coincidentally, the number of neurons required to decode
the location of a rat in an open space or the motion of the hand/arm on
monkeys (both with better than 10-cm precision) is at least a hundred (Ke-
mere et al., 2004; Musallam et al., 2004; Zhang et al., 1998). To record this
quantity of neurons at a single spike temporal precision is now feasible by
extracellular recording. Imaging techniques are approaching this scale and
6
An intermediate level between microscopic and macroscopic (Freeman, 2000).
will soon complement the extracellular recording methods. Albeit the spa-
tiotemporal resolution of the blood oxygenation level-dependent signal and
the indirect link between neural activity and hemodynamic changes make
fMRI less suitable to capture the functional architecture on the cellular
scale, recent advances in voltage-sensitive dye (Petterson, this volume) and
two-photon calcium imaging (Goldberg et al., this volume) are promising
techniques to study the spatiotemporal dynamics of hundreds of neurons in
the living brain. We focus on the extracellular recording technique in this
chapter; however, the statistical approach for recovering functional connec-
tivity from the spike pattern of multiple neurons is indifferent to the data
acquisition technique.
Concerning the traditional extracellular recording technique, it is impor-
tant to consider the limitation of the electrode. To date, the most commonly
used electrodes are sharp electrodes, microwire electrodes, tetrodes, multi-
ple wire electrodes, and electrode arrays. Positioning electrodes individually
imposes a serious constraint on the number of electrodes. Practically, remote
control operated individually movable electrode microdrives can control up
to 16 electrodes. This may be sufficient to record activity of up to 30–40 neu-
rons. To attain a simultaneous recording of > 100 neurons requires chroni-
cally implanted electrodes. Several configurations have been devised. One
is an array of parallel microwires, mounted together allowing about 400 µm
spacing between the 72 µm-diameter wires (MicroProbe Inc., MD). Another
option is using silicon-substrate, micromachined probes that come in var-
ious shapes and configurations and cause much less tissue displacement
relative to microwires. Among other configurations, multiple shank silicon
probes allow simultaneous multisite data acquisition from the same cell as
well as from the different cell groups (Vetter et al., 2004). The third technol-
ogy is the Utah electrode (the “Utah array”), which is a penetrating array of
electrodes mounted on a 4 mm × 4 mm base containing 100 silicon spikes
that are up to 1.5 mm long and designed to be implanted in primate and
human cortical structures (Nordhausen et al., 1996).
Another constraint in extracellular recording is sampling rate. To resolve
spike shapes, the sampling rate must exceed 20 kHz per electrode. Sampling
rates higher than 25 kHz do not improve spike discrimination significantly
(Nadasdy et al., 1998). A simultaneous recording of 100 channels at this
sampling rate would require 2 MHz multiplexing with the digitized data
streamed to a hard disk. If the signal is digitized at 16 bits per channel (12 bits
is sufficient, 16 bits is recommended), then the necessary data transfer rate
is 32 MB/s, which can be easily transferred in standard network cards and
streamed to a hard disk. However, we still have to be cautious with disk
space since 2 h of continuous recording with this bandwidth takes a total
(nonredundant) disk space of about 28 GB per session.
Another important constraint is the type and geometry of the electrode.
Multiplying electrodes can be done by bundling multiple microwires to-
gether. The standard technique is to bundle four electrodes to make a
tetrode. This technique multiplies the recording sites around a single
neuron, improving spatial resolution and consequently spike discrimina-

tion (Buzsaki, 2004; Henze et al., 2000). There is no substantial gain in
neuron yield of bundling more than four microwires. Another way to in-
crease the number of recording sites is by multiplying electrode arrays that
are mounted separately. Several such electrode configurations have been
deployed through the last 15 years. There are two basic strategies, concern-
ing the type of scientific questions addressed. One strategy is to chronically
implant as many electrodes as possible, where it is certain that a number of
electrodes will not detect any signal from active neurons, simply because they
cannot be individually positioned near the neurons. The alternative strategy
is to make the individual electrodes independently movable. This technique
is better suited for acute preparations where the research objective requires
inserting electrodes in different locations every day for systematic mapping.
Chronically implanted tetrodes are between these two strategies since the
whole tetrode array can manually be lowered every day. Electrode configu-
rations for chronic implants range from 16 (MicroProbe Inc., MD) to 100
(Utah probe, University of Utah, Salt Lake City, UT) recording sites, on a
single connector. Many laboratories use custom-made electrode arrays. Elec-
trode implants can be skull mounted or floating. To minimize movement
artifacts, it is critical to keep the electrode and signal stable. The stabil-
ity of floating electrodes is higher than that of skull mounted because the
electrode moves with the brain.
Extracellular recording strategies have to consider several trade offs. In
order to increase the neuronal yield of recordings, the obvious strategy is
to multiply electrodes. However, the cost of multiplying electrodes is multi-
fold. First, the tissue displacement increases with the number of electrodes.
According to simulations, a vertically inserted 50-µm-diameter microwire
electrode collides with 80% of the dendrites of the nearest recorded neuron
(Claverol and Nadasdy, 2004). Second, the increased number of channels
requires upgrading the data acquisition hardware to keep the sampling rate
at least 20 kHz per channel and stream larger data blocks to a hard disk.
The third factor is that the isolation of neurons from an increased number
of channels substantially increases the volume and time of data processing.
For example, a thorough manual spike sorting of a 2-h recording from eight
channels may take 2 weeks in our practice. Considering that for publication
it is necessary to process at least 30 such recordings and to replicate it at least
on one other preparation, it would take more than 2 years of graduate stu-
dent life with no weekends off to publish. Consequently, 100 channels would
take at least 28 years to analyze which may eventually delay graduation.
2. Signal-Based Source Identification
A typical microwire electrode or tetrode can detect signals from neurons

as far as 150 µm away (Fig. 20.6). Such a volume contains about 1100 neurons
based on hippocampal cell density estimates (Henze et al., 2000). The prob-
ability of detecting neurons decreases with the distance from the electrode.
In the outer part of this volume no individual spikes are discernible primarily
Figure 20.6. The simplified scheme of single-unit isolation from extracellular record-
ing. The montage is an artistic rendering of a ∼300-µm-wide cortical volume with
the tip of a ∼50-µm-diameter tetrode inserted in the vicinity of pyramidal cells. The
broad blue cylinder represents the largest distance from the electrode within which
spikes can be detected. This volume contains ∼1100 neurons. The 50-µm-radius
inner cylinder is the volume within which spikes are discernible. A pyramidal cell
(highlighted) at the vicinity of the electrode projects differential signals to the four
electrode tips as a function of distance. Since the voltage attenuates proportionally
with the distance, the location of the source can be recovered from the differential
amplitudes reaching the electrodes. Although this volume contains ∼140 neurons,
one can usually record up to 8 neurons. In the volume around the 50 µm radius,
the signal amplitude drops below 60 mV. From this point we can consider the sig-
nal “multiunit,” where individual spikes are still detectable but cannot be identified
with discrete sources. From this range to the 140-µm range, the signal asymptotically
converges to noise as spikes from an increasing number of neurons start to overlap.
Numbers are based on hippocampal estimates (Henze et al., 2000). (Background
image is enhanced GFP-expressing lentivirus injected into the parenchyma of rat
layer 2/3 somatosensory cortex from Brecht et al., 2004.)
because spikes at average firing rate overlap in time and space, making the
signal similar to 1/f type noise. Within the inner volume of 100 µm, spikes
become discernible; however, due to the additive background noise, the
amplitude variation is too large to classify neurons. It is only within a 50-µm
radius volume around the electrode that spikes can reliably be associated
with neurons. Intriguingly, based on a hippocampal cell density estimate,

∼140 neurons should be detectable from this volume, we typically discrim-
inate only a small fraction, on average less than 10 neurons. Whether this
discrepancy is due to the silence of neurons or other attenuating factors,
remains to be clarified.
In order to associate spikes with neurons, most spike-sorting methods
use waveform-discrimination-based algorithms. The underlying assumption
behind this group of methods is that the spike waveform differences corre-
spond to different neurons. The biological justification for this assumption
is the point source model of action potential generation. According to this
model, action potentials going down the axon of the same neuron are al-
most identical if you measure it near the axon hillock or intracellularly from
the soma. Consequently, the main source of variation of mixed waveforms
is the distinct spatial location of the neurons relative to the electrode. How-
ever, the point source is affected by a number of other factors that need to
be considered. For example, the intrinsic dynamics of spike generation (the
interaction of back-propagating action potentials and dendritic spikes), the
cell type (interneuron versus pyramidal cell), and ongoing local field oscil-
lations substantially contribute to the spike waveforms (Buzsaki et al., 1996)
either by moving the point source or by adding a nonlocalized variance to
the waveform.
An alternative approach in spike sorting is based on the spatial localiza-
tion of the source. This principle was first utilized by the stereotrodes and
further exploited by the tetrodes. Projecting simultaneous traces of spike
waveforms (all data points or only peak-to-peak values), measured from two
adjacent electrodes as X and Y coordinates against each other, results in clus-
ters of points. These clusters reflect the differential amplitude ratios caused
by the unequal distance of the neurons relative to the two electrodes. Using
multiple recording sites within the critical 50–60 µm volume and knowing
the spacing between the sites, one can triangulate the sources (Nadasdy
et al., 1998). Sources determined from the amplitude ratios of collinearly
arranged recording sites discriminate between neurons more reliably than
spike waveforms (Harris et al., 2000). Silicon multiprobes further expand
the “scope” of electrodes. For example, multiple shanks and collinearly
arranged recording site geometry with 20 µm spacing provide precise lo-
calization of neurons as well as reliable segregation of different groups of
neurons from each shank (Csicsvari et al., 2003). The distance and geometry
of recording sites can be optimized for a given structure of interest.
3. Spike Sorting
While electrodes and amplifiers allow recording from more than 100
channels simultaneously, isolation of single units from all of these channels is
one of the main bottlenecks of data processing. Spike sorting is a critical step
not only for isolating the sources of spike generation but also for classifying
neurons. More research is needed to elucidate the relationship between
action potential generation and current propagation in the extracellular
space in order to maximize information available from the waveforms. For
example, it has been reported that spikes generated by interneurons have a
shorter polarity reversal (spike half width) than do pyramidal cells (Bartho
et al., 2004). The correlative inference from the average spike shape to the
type of neuron can serve only as a putative classification considering that
the waveform can be affected by many other factors. Combining evidences
including firing rate statistics and cross-correlation analysis is necessary to
classify neurons based on dynamic properties.
Spike-sorting methods consist of two main steps: feature extraction and
clustering. The goal of feature extraction is to determine a set of discrimi-
native features which, when projected to an n-dimensional space, best sepa-
rate individual spikes. When spikes are classified based on waveforms, either
the whole spike shape or several descriptive features are used (Abeles and
Goldstein, 1977; Fee et al., 1996, respectively, and review by Lewicki, 1998).
In either case, the database usually contains redundant dimensions that are
suboptimal for clustering. The most common method to reduce dimension-
ality of spike features is based on the principal component analysis (PCA).
Using PCA, the main axes of “ideal” projection are determined by the largest
two eigenvalues (Abeles and Goldstein, 1977). Although the largest eigen-
value projections maximize the global variance of the whole spike sample,
they may not be optimal to separate the clusters. A more recent approach
is based on clustering wavelet coefficients computed from the individual
waveforms (Quian-Quiroga et al., 2004).
It has been argued that the main causes of unit identification errors re-
gardless of the method being used are (1) the additive noise derived from
overlapping background activity, (2) the intrinsic amplitude modulation of
the spikes due to subthreshold membrane oscillations, (3) somatodendritic
back propagation of action potentials (Buzsaki, 2004), and (4) the human
operator’s ability to supervise the clustering (Harris et al., 2000). The human
operator’s suboptimal performance and the time constraint are the main
motivations to develop quasi or fully automatic and unsupervised spike-
sorting tools. As the number of electrodes and channels increases, isola-
tion of single units (spikes generated by a distinct neuron) “online” during
the experiment simply cannot be performed reliably by a human opera-
tor. Such unsupervised spike-sorting methods have already been deployed
(KlustaKwik, http://klustakwik.sourceforge.net/; WaveClus, Quian-Quiroga
et al., 2004). For a while, off-line spike sorting will remain necessary simply
because the online classification is CPU intensive and requires more time
to classify a channel than to record from it.
C. Reconstruction of Functional Connectivity
Spike sorting identifies spikes with a distinct neuron, thus allowing the
reconstruction of multiple spike trains. The next major step is to recover
the hidden functional architecture of these neurons, based on the tem-

poral correlation between the spikes in different spike trains. Before this
step is taken, we need to distinguish between sources of temporal correla-
tions. Correlations in neuronal activity can be caused by the spatiotemporal
structure of stimuli or can be derived from the functional connectivity of
neurons.7 The former one is referred as signal correlation and the latter
one is referred as noise correlation. Usually both sources contribute to the
actual activity pattern. To elucidate the functional connectivity, we must
consider the intrinsic variance in neuronal activity, i.e., the variance that is
independent of the input signal characteristics. Therefore, the input to the
studied circuitry should either be kept constant or assumed to be evenly
distributed. If each neuron is an ideal encoder, which encodes the stimulus
independent of other neurons, then all correlations can be explained by
the stimulus and there is no circuitry to uncover. In contrast, the typical
scenario is that the input can only be partially recovered from the activity
of neurons, suggesting that a nontrivial circuitry is involved where neu-
rons are not independent. The answer for the general question “whether
neurons encode the input independently from one another or not” likely
depends on the studied structure. The complexity of dependence greatly af-
fects the level of statistics efficient to recover the circuitry from the activity of
neurons.
If neurons are highly interdependent, then recovering the circuitry is a
complex and computationally NP-complete problem∗ where the computa-
tion time is an exponential function of the number of cells (spike trains).
Therefore, the first practical step is to reduce the complexity to pairwise
cross-correlations of spike processes and build the functional connectivity
from the pairwise relationships. Several cross-correlational techniques have
been developed, primarily between spike trains of two neurons. Nonethe-
less, higher order correlations are also practical to compute.
The first-order relationship between two spike trains is captured by the
cross-correlation and cross-coherence of the two spike trains. For the cal-
culation of the cross-correlation function see the Appendix. The graphical
representation of cross-correlation is the cross-correlogram. The qualita-
tive evaluation of cross-correlogram can tell (i) the independence of the
two spike generation processes, (ii) the time precedence between the two
processes, and (iii) the polarity of the effect (inhibitory/excitatory). What
a cross-correlogram does not determine is whether the dependency be-
tween the two neurons is direct or indirect. The interpretation of cross-
correlograms is illustrated in Fig. 20.7. Cross-correlograms are constructed
by treating one spike train as the reference and the other train as the
7
The method of reconstructing the stimulus from the activity is called the reverse correlation
technique.
∗ NP-complete problems are decision problems verifiable in non-deterministic polynomial
time.
Figure 20.7. Inference from cross-correlations to circuitries. (A) A central peak with
a few millisecond delay indicates a monosynaptic excitatory drive from neuron a
to b . (B) A central through is suggestive of a monosynaptic inhibitory drive from
c to b. (C) A central bipolar wave with positive peak followed by the through indi-
cate a feed-forward excitatory drive (a–b ), followed by recurrent inhibition (a–c –b).
The opposite polarity order would suggest feed-forward inhibition combined with
a non-monosynaptic recurrent excitation (not shown). (D) Periodicity in the cross-
correlogram is consistent with an excitatory drive from a to b, modulated by an oscil-
latory input from a locally phase-coupled interneuron circuitry c . Alternatively, the
same cross-correlogram can reflect two inhibitory interneurons firing in synchrony
(not shown). (E) A fictitious example for a three-neuronal excitatory circuitry illus-
trates the limitation of cross-correlation analysis. Neurons a and c both terminate on
b. It is evident from the cross-correlogram that a drives b with certain delay (repre-
sented by outlined area in the cross-correlational matrix and corresponding empty
ticks in the spike train inset). The c –b cross-correlogram also indicates an excitatory
drive between neurons c and b (blue area and blue ticks in the spike train inset).
However, the flat a–c cross-correlogram suggests that spikes of neurons a and c co-
incide only by chance (gray area in the cross-correlogram). The dilemma, whether
a–b and c –b drives are conjunctive or disjunctive, i.e., a and c drive must coincide
to excite b or any of them is sufficient to drive b , cannot be resolved based solely on
the pairwise cross-correlograms. One needs to look at the triple cross-correlogram
and evaluate whether the frequency of a − b − c triplets (indicated by black ticks
in the spike train inset) is higher than chance or not. (This panel is modified from
Nadasdy, 2000).
subject. The computational algorithm of the cross-correlogram goes as

follows:
1. Define a time window W of interest. This is usually longer than the re-
fractory period (3 ms) and smaller than the longest causal interactions
within a network (200 ms). Moreover, define a t precision of spike

time measurement. Reasonably, this should be larger than the spike
width and smaller than the expected smallest spike time precision, but
not larger than W /20.
2. Then iteratively, take the next/first spike at the reference train and look
for spike coincidences on the subject train within the time window cen-
tered around the reference spike. The time lag τ (see the Appendix)
of the subject spike train can either be negative when preceding or be
positive when following the reference.
3. Next, the bin of the cross-correlation histogram at τ is incremented.
4. After all subject spikes within W have been registered, move the time
window to the next reference spike and repeat steps 2–3 until the last
spike has been used as the reference spike.
The interpretation of cross-correlations is summarized in Fig. 20.7. The

objective of cross-correlational analysis is to determine the interaction be-
tween two neurons (reference and subject neurons). Since correlation can
occur due to a causal effect by one neuron on the other neuron, as well as
due to a causal effect by an outside source on either of the two neurons, the
reduced model of interactions must include at least three neurons. When
considering the dynamics of a three neuronal (a, b, c ) subnetwork based on
the information available from two neurons, the possible cross-correlations
are consistent with one of the following schemes: (i) excitatory drive from a
to b (Fig. 20.7A); (ii) inhibitory drive from c to b (Fig. 20.7B); (iii) combined
excitatory and inhibitory drives (Fig. 20.7C); and (iv) common oscillatory
drive to a and b is also evident from cross-correlograms (Fig. 20.7D).
D. Effective Connectivity
Although action potentials are uniform in shape and generation site, the
inputs that elicit them derive from many neurons firing at different times.
In principle, a strong input connection has a larger contribution to a spike
series than a weak input. In order to recover the functional connectivity be-
tween neurons from an extracellular recording, it is important to determine
the most effective source of input for a given neuron. The tight temporal
correlation between the input and the output spikes is the key to resolving
which spike was caused by which neuron in the network, even when higher
order (hidden) dependencies are involved.
Higher order dependencies, involving three or more neurons, are beyond
the scope of the cross-correlation method. Consider the example illustrated
by Fig. 20.7E. Let us assume that we know the connectivity between the
three neurons (a, b , c ), where neuron a and c both have an excitatory
drive on neuron b . The excitatory dependencies between a–b and c –b are
evident from the combined cross-correlograms (on the right). Nonetheless,
we would like to know whether a single input from neuron a or c is sufficient
to excite b or the two inputs must coincide to excite neuron b.8 From pairwise
cross-correlations this information is not available. To infer the “effective
connectivity” from multiple spike trains involving more than two neurons
requires higher order statistics.
Motivated by mapping the “effective connectivity,” Gerstein and col-
leagues introduced the joint peristimulus time histogram (JPSTH) method
(Aertsen et al., 1989). The JPSTH was a generalization of cross-correlation
method to spike triplets of different neurons that coincide within the time
window W . The appropriate statistics were developed much later (Abeles
and Gat, 2001; Baker et al., 2001; Frostig et al., 1990; Palm et al., 1988). As
an illustration of such statistical analysis on >2D dependency between spike
trains, consider Fig. 20.7E again. The joint occurrence of an abc triplet is sim-
plified by reducing it to the coincidences of two intervals, τ ab and τ ac inter-
spike intervals. In a random spike process, we assume that the acb triplets are
merely by-chance co-occurrences of the ab and ac intervals. Therefore, their
joint probability is the product of the component probabilities. Otherwise,
the triplets are not random coincidences:
H0 : Pac b = P τ ab · P τ ac
HA : Pac b = P τ ab · P τ ac
If it turns out that Pac b > (Pτ ab · Pτ ac ), that is, the observed frequency
of the triplet is higher than the product of the individual probabilities,
then the acb triplets must be coordinated above chance coincidence of ab
and ac intervals. The Pτ ab and Pτ ac probabilities are estimated from the
pairwise cross-correlation functions; however, the two interspike-interval
pools are not mutually exclusive because they both contain a fraction of
the other interval pool. For example, the ab cross-correlations include the
abc events and likewise the ac cross-correlations, consequently these events
will be counted twice. As a result, the product of cross-correlation functions
overestimates the expected probability of triplets. Therefore, the product
must be corrected and renormalized to fit to the individual firing rates and
cross-correlograms. This renormalization must conform not only with the
Pτ ab Pτ ac cross-correlations but also with the Pτ b c . Because the true prob-
ability density functions are not directly available from the data, various
computational methods have been developed for estimating it. The correct
estimation of random coincidence is critical in order to determine the con-
fidence interval for significance testing. Among those methods, we point
the reader’s attention to the method of calculating a “surprise function”
(Palm et al., 1988), spike jittering or “histogram blurring” technique (Abeles
and Gat, 2001) and time resolved cross-correlation method (Baker et al.,
2001).
8
Although there may be thousands of excitatory inputs terminating on a pyramidal cell, the
above question, whether the two cells (a, c ) belong to populations that have a conjunctive or
disjunctive effect on neuron b , remains relevant.
An alternative approach to detecting hidden dependency time struc-

tures from spike trains is based on searching for joint events using pattern-
searching algorithms. The rationale behind this approach is the following.
When several neurons are recorded simultaneously, these neurons are part
of a larger hidden network. We further assume that neurons are finite state
machines and spikes are generated from neuron to neuron as a causal se-
quence of state transitions. If a sequence occurs more often than chance,
we have a statistical basis to believe that the spike sequence follows a deter-
ministic process, which is constrained by the network architecture (unless it
was imposed by the temporal structure of the input to the structure which
we do not consider here). If the circuitry favors one spike sequence over
other sequences, then this sequence must recur more often than others.
Consequently, the most invariant sequence must be related to the activity
flow, which in turn, is dependent upon the circuitry. Thus, the objective
of pattern searching spike trains is to determine repeating sequences and
reconstruct their underlying functional connectivity. Various algorithms of
searching the high dimensionality space of multiple spike trains for repeat-
ing motifs of spikes were introduced during the last 10 years (Abeles and
Gerstein 1988; Nadasdy et al., 1999; Lee and Wilson, 2002; Ikegaya et al.,
2004). The common feature of these algorithms is to define a template mo-
j1 j2 j3
tif Sn = {ti , ti+1 , ti+2 , tlnm } from the data, which is a vector of spike latency
t of the ith spike generated by the j th neuron where j = [1 . . . n]. Let us
name this as an n sequence. Then the computer program performs itera-
tive searches by comparing the template n sequence with each instance of
n sequences on a data segment where the length of n sequence is equal
to the dimension of searching space n. After every partial match between
the template and the data, the dimension is incremented until n. When
the search is complete for a given template, the template is replaced by a
new n sequence from the data and the searching session starts over. After
an exhaustive search of the whole database for all possible Sn sequences,
the detected recurring n sequences can be rank ordered and those that
occur more than chance are considered to be representing the predomi-
nant flow of activity within the network of the given set of neurons. As we
have mentioned, the searching process is CPU intensive and computation
time increases exponentially with the number of spike trains. A spike train
database of 100 neurons would require a supercomputer to run the search.
Since dependency and causality are interchangeable terms, reconstruc-
tion of effective connectivity from dynamic data can be approached as re-
covering the hidden causal sequence of activity flow within the network. The
formal definition of causality on time series was first introduced by Granger
(1969) in the context of linear regression models of stochastic processes and
since then it is referred as “Granger causality.” Accordingly, if the variance
of the prediction error from a time series Y (t) is reduced by including the
past measurements from time series X(t), then the time series X(t) “caused”
the time series Y (t). Independent of the Granger causality, numerous other
attempts were made to quantify causality from neuronal data. To take a
full advantage of multichannel data, such as simultaneously recorded spike
trains, Kaminsky and Blinowska (1991) suggested a multivariate spectral
measure to determine directional influence, defined as the “directed trans-
fer function,” which is fully compatible with Granger causality (Kaminsky et
al., 2001). Using the formalism of directed transfer functions, one can re-
construct the causal-dependency network within a population of neurons.
It is important to note that Granger causality does not imply direct causal
effect between the two spike trains (i.e., neurons). The effect can be medi-
ated by other (hidden) neurons or distributed over the whole population of
neurons, which would make the causal sequence untraceable. However, the
question of direct causality can also be addressed within the framework of
directed transfer functions (Kaminski et al., 2001). Granger causality spectra
have been successfully applied on local field potentials, clarifying interac-
tions between different recording sites at specific frequency bands, (Brovelly
et al., 2003) and can be applied to point process spectra such as spike trains
(Kaminski et al., 2001).
A problem common in all these methods is the definition of confidence.
To obtain confidence intervals for significance estimations, we need to know
the probability density function of the given variable, such as triplet occur-
rence, sequence repetition, or causality. Since there is no a priori knowledge
about the generative process of these variables, we must rely on Monte Carlo
methods, such as the simulation of the spike generation process and the con-
struction of surrogate spike trains (Nadasdy et al., 1999). The limitation of
the Monte Carlo approach is that we can never be sure that randomization of
a variable leads to a biologically plausible null hypothesis or not. Consensus
on these methods has yet to be achieved.
In summary, to recover the functional connectivity from a local cell as-
sembly based on the dynamics, the following methods should be applied.
Use electrodes with multiple recording sites at 20–50 µm apart allowing
multiple spike measurements from the same neuron. Combining several
such electrodes with at least 100–200 µm spacing between them allows the
recording of nonoverlapping groups of cells. Configurable recording ge-
ometry is preferred to suit the electrode to the size and cytoarchitecture of
the target tissue. For an optimal design, the potential tissue displacement
and interference with normal neuronal functions should be considered.
Using an unsupervised spike-sorting software, the classification of 3–7 neu-
rons from a single electrode (shank or tetrode) is feasible, which scales up
with the number of electrodes used. Cross-correlations can be applied to
reveal first-order dependencies. Higher order statistics, essential for recon-
structing the network dynamics, can be computed by joint probability and
sequence-searching algorithms.
An example of the complete process of the reconstruction of effective
connectivity is shown in Fig. 20.8. Cortical recording from the somatosen-
sory cortex of an awake rat was performed by using a silicon microprobe
(Buzsaki, 2004). For illustration purposes, the image of the electrodes was su-
perimposed on the histological section using electrode traces as landmarks.
Figure 20.8. Reconstruction of effective connectivity between neurons based on the

temporal coherence of spiking activity recorded extracellularly from the somatosen-
sory cortex of a rat. (A) The theoretically possible network of neurons is illustrated
by white lines (only a subset) connecting putative pyramidal (red triangles) and
interneurons (blue circles). Neurons were isolated using spike sorting and localized
by calculating the “center of mass” of the spike amplitudes. The putative locations
enabled identification of neurons with their histological traces (background) and
recovery of the electrode traces. The image of the multiple shank silicon electrode
was aligned and superimposed on the traces. Spacing between adjacent recording
sites of the electrode was 20 µm with 200 µm intershank distance (electrode shanks
The neurons isolated from each electrode were localized by calculating
the center of mass of their spike amplitude. Possible functional connec-
tions were considered by pairing all these neurons and computing pairwise
cross-correlograms of their spike times. Neurons with the largest modu-
lation were selected as active nodes of the functional network, using the
cross-correlation function. Interneurons and pyramidal cells were identified
based on their firing rate and the polarity of the cross-correlation modula-
tion. Reciprocal connections were also evident from the cross-correlograms.
Note that not all of the recorded neurons are engaged in the network. The
fact that certain neurons did not participate in the functional network sup-
ports the validity of detected connections. Using such heuristics, one can
construct a network that is most consistent with the dynamics sampled by
the cross-correlations. Although this network is rather incomplete and con-
nections are only putative, it is the most reliable method to date for parsing
the architecture of circuitries.
V. CONCLUDING REMARKS
After a century of extensive piece-by-piece brain mapping, finally all the

main pathways of the central nervous system have been described and val-
idated across different species. Although for most structures the cytoarchi-
tecture has been defined and major cell types have been identified, the key
principles that relate the microscopic architecture to activity patterns are
still unknown. We argue that the conceptual bridge between these two lev-
els is the dynamics of microcircuitries. The microcircuitry is a mesoscopic
entity between neurons and macroscopic networks that generates a dynamic
mosaic of functional cell assemblies by assigning neurons to different subnet-
works, and these subnetworks are dynamically allocated to various functions.
We reviewed classification of these microcircuitries as precursors for “basic
circuitries.” Although the catalog of basic circuitries is at the doorstep, to
be able to make inferences from the architecture to the activity pattern and
vice versa requires a coordinated anatomical and physiological approach
that embraces both static and dynamic aspects of neuronal data acquisition.
←
Figure 20.8. (Cont.) were brought closer for illustration purpose). (B) Auto- and
cross-correlation histograms featuring the coherency between the three interneu-
rons (numbers 3, 35, and 40). The autocorrelograms for interneurons are purple,
interneuron–pyramidal cell cross-correlograms are black, and pyramidal cell auto-
correlograms are red. (C) The main PCA clusters show clear isolation of single units.
Numbers at the right are the Mahalanobis distances for the given cluster. (D) Effec-
tive connections verified by the cross-correlational analysis among the theoretically
possible network. Neurons whose connectedness could not be verified are marked
by empty symbols. Note that most pyramidal cells receive inhibitory connections
from the interneurons. A few of these connections are reciprocal. In contrast, no
pyramidal–pyramidal connection was observed.
Basic circuitries are recognized as instrumental for filtering and gener-

ating specific activity patterns and oscillations. Their cellular composition
must be specific for the given structures and consistent across different
species. Plausibly, the evolving and developing brain multiplies these cir-
cuitries as basic information-processing units to construct large networks.
The unique features of basic circuitries are the constituent neuron types,
connections, and the specific activity patterns they implement. We argued
that these activity patterns cannot be derived from the physiology of isolated
neurons. Instead, the focus needs to shift from single-cell recording to the
circuitry of ensembles of neurons, an approach that involves a combination
of neuroanatomical and physiological methods.
Among these methods, this chapter discussed algorithms designed to
extract information from a large population of neurons that is relevant
for functional considerations, such as neuronal clusters. We also reviewed
methods of collecting physiological data that capture the dynamic aspects
of circuitry function, such as spike pattern generation. We highlighted on
methods of the multiple electrode recordings and the analysis of action po-
tential patterns generated by a population of neurons. We illustrated the
inferential processes from the two opposite directions: one that proceeds
from the structure to the dynamics and the other that proceeds from the
dynamics to the structure. The two types of inferences complement each
other and they should converge to a solution where the circuitry architec-
ture is consistent with the activity pattern. The circuitry and activity pattern
as a functional unit may be considered as a fundamental building block of
the nervous system.
Concerning strategies of neuronal tract-tracing methods, two dominat-
ing tendencies, the expanding computational approach in neuroanatomical
data analysis and combining existing methods, are broadly covered by differ-
ent chapters of this volume. Specifically, using a sophisticated combination
of tracing methods, the synaptology of any circuitry can be accurately de-
termined (Sesack et al., this volume). Using extracellular, juxtacellular, and
intracellular recording methods (Duque and Zaborszky and Sik, this vol-
ume), the anatomical features can be correlated with an electroencephalo-
gram, multiunit activity, local field potential, and intrinsic membrane char-
acteristics. Functional networks or entire brain regions are available for
reconstruction (Bjaalie and Leergard, this volume) and statistical methods
on testing structural hypotheses have already been developed (Bjaalie and
Leergard, this volume). Furthermore, neural functions can be correlated
with microstructural variation (Amunts and Zilles, this volume).
Last but not least, we emphasized the importance of 3D representation
and the use of volumetric data structure that is a prerequisite for exploit-
ing the information available from neuronal tracings. The third dimension
is critical to complete the paradigm shift we are witnessing in three ma-
jor fronts of research: (i) functional neuroimaging, (ii) computational data
analysis and visualization, and (iii) data sharing and integration into incre-
mental databases.
APPENDIX
A. Parametric Modeling of Neuroanatomical Data
1. Definition of Volumetric Database
A volumetric database consists of a 3D distribution of one or more vari-

ables. For the sake of simplicity, let us focus only on one variable. This
variable must have spatial gradients that are continuous in space. Usually
this variable is the density of a given feature. Therefore, the first step of con-
structing a volumetric database is to define this structural feature (such as
the cell body, the dendritic segment, or the expression of certain marker),
which has spatial density gradients. Before extracting this variable the space
needs to be discretized. Discretization is generally achieved by defining a
grid structure that is optimal for sampling the data. Reasonably, the grid size
must be large enough to contain multiple instances of the given feature, but
small enough to represent the fine-scale spatial variation of the variable. We
refer to the 3D units of the grid as voxels. The voxel size determines the
spatial resolution of the density of the variable. Voxels can also be consid-
ered as 3D bins within which the instances of the variable are counted. The
voxels usually, but not necessarily, have equal edge lengths and each voxel
is addressed by its Cartesian coordinates (x, y , and z).
2. Construction of Volumetric Data of Neuronal Density
The given camera lucida database (we recommend Neurolucida, however,

it can be another 3D vectorial data type) must be exported as an ASCII file
and parsed for cell bodies and structure outline coordinates. As a result, the
data should contain only cell bodies {b t } of a given cell type t, represented
individually by their single Cartesian point coordinates { p xy z } and section
identifiers s . If the z coordinates were not recorded, the section identifier
should be considered as the z coordinate.
b t = p xy z (20.1)
Conversely, since a given position could be occupied by only one cell body,
the Cartesian coordinates together with the cell type completely define a cell
b t as xb , y b , z b , and t. Structure outlines can be compiled to separate files. The
medial, lateral, dorsal, and ventral extremes of the cell population should
be taken as edges of a 3D framework to incorporate the region of interest.
The total volume V occupied by neurons (neuronal space) is expressed as
a vector r with minimal r density function:
V = {
r : ρr > 0} (20.2)
For quantitization of local density differences, V was subdivided into “vox-
els” and defined as follows:
v = d x , d y , dz , (20.3)
where d is the edge of the voxel. If the within section depth of the coordinates
were not registered, the section index s can be used as dz . Thus, voxel and
cell definitions can be simplified as vs = dx , d y and b i = p xy s , respectively.
The i, j , and k indices of a voxel containing a b i = p xy s cell body is calculated
as
xp yp
i = dx , j = dy , k = s. (20.4)
Vlength Vwidth
Cells are then counted in each voxel for each cell type separately, provid-
ing a local density function ρ:
ρ(vi j k ) = count (vi j k ) = nb i . (20.5)
For visualization, we define a volume based on a function of minimum
density σ :
= vi j k : ρ(vi j k ) ≥ σ . (20.6)
The anatomical distribution of high-density locations is often visualized by
a manifold rendered around the volume and denoted as δ. The manifold
can be defined by surface rendering and carried out using commercially
available visualization tools such as the Matlab software package (Math-
Works, Natick, MA). After surface rendering, one must pay attention to
convert the voxel coordinates back to stereotaxic coordinates or coordinates
used by the data acquisition system to be able to superimpose the voxels on
structure outlines.
Note that the critical step was the conversion of the 3D point-coordinate
database, where the entries were the cells, to density data (Eq. 20.5),
which became a volumetric database. In contrast to the original parametric
database where cell bodies are represented by their x, y , and z coordinates,
the entries of the volumetric database are cell density or cell count. The
key advantage of the volumetric database is that it allows one to employ
parametric statistical methods. Moreover, it enables one to slice data at any
angle and visualize it from any point of view.
3. Differential Density 3D Scatter Plot
Valuable information can be extracted from the volumetric database that

is not available from the 3D tracings of cells. To extract structural features
and visualize them, we use the example of cell density. The density of neu-
rons is indirectly related to functional clusters of the population because
neurons close to one another are likely sharing input. We can apply a density
threshold and highlight voxels with higher than threshold density. Taking
the visualization one step further, we randomly choose a neuron from the
selected voxels and visualize that neuron superimposed on the background
of local density. Furthermore, we can select not only one but n number of
neurons from a given voxel, where n is proportional to the density.
4. Isodensity Surface Rendering
The spatial organization of a large population of neurons can be very

complex. Instead of visualizing each neuron, the global pattern of den-
sity differences may elucidate important structural principles that are not
apparent from a large population of neurons. Surface rendering around
high-density cell groups can capture this global pattern. We developed an
algorithm that renders a surface around voxels where cell density is larger
than a certain threshold. The procedure of discretization of the 3D space
into voxels and calculating the per voxel cell density is described under sec-
tion “Construction of Volumetric Data of Neuronal Densities.” By applying
different density thresholds, the sharp transition of neuronal density cap-
tured by two thresholds can be indicative of the size of a neuronal aggregate.
The stepwise increase in density may be related to deviation of randomness
that requires further statistical tests.
5. Isorelational Surface Rendering
In order to simplify the complex spatial relationship of large neuron pop-

ulations and extract the associative relationship of different cell types, we can
calculate the density ratio of the two cell types in each voxel. Highlighting
the locations at which the density ratio exceeds a certain level reveals the
spatial configuration of the cell-to-cell associations between different cell
types. To illustrate this, first we constructed volumetric databases of density
for each cell type and discretized the space by dividing it into voxels. For
this, the different cell types must be carefully mapped to a common 3D coor-
dinate system. If the different cell types were traced from different sections,
it must be considered that inference of the joint density from separated
sections may be affected by section distortion. If density changes between
adjacent sections are negligible, the within-section z coordinates of differ-
ent cell types can be collapsed into the same section s to obtain an estimate
for the real joint density. The next step is to calculate local density ratios
between the cell types for each voxel. When density ratios had been assigned
to each voxel of the volumetric database, we imposed a predefined density
ratio criterion and selected voxels that met this criterion for visualization.
The volume of these voxels represented a specific numerical association
between cell types. Voxels where the predefined density ratio of cell types
had been established were surface rendered. Thus, cell bodies with density
ratios larger than a specific value were covered by the surface. Conversely,
voxels with density ratios smaller than the critical one were located outside
of the surface. The unique feature of the “isorelational surface rendering”
method is the visual representation of an abstract relationship that may be
more important for understanding the functional architecture of neurons
than the exact locations of cell bodies. For visualization purposes, a range of
critical density values must be applied for testing the integrity of clouds and
to make sure that there are no hollow spaces covered. Examples for analyses
described under sections “Differential Density 3D Scatter Plot,” “Isodensity
Surface Rendering,” and “Isorelational Surface Rendering” can be found
in Nadasdy and Zaborszky (2001) and Zaborszky et al. (2005b).
B. Databases
1. List of Electronic Databases of Rodent Brains
Brain Maps Swanson Computer Graphic Files, Elsevier,

1992–1998
3D Rat Brain Timsari et al. (1999) http://www-hbp.usc.edu/Projects/
3dAtlas.htm
The Rat Brain Paxinos and Watson Academic Press, 1998–2005 CD-ROM
Brain Browser Bloom and Young Academic Press, 1993
Rat Brain Toga et al. (1995) www.loni.ucla.edu/Research/Atlases/
Rat/Atlas.html
Rat Brain Nissanov and www.neuroterrain.org
Bertrand (1998)
IMGEM Mouse Brain Sugaya et al. http://sugaya.ucf.edu
Mouse Brain Allen Project www.brain-map.org
Mouse Brain Library Reed et al. (1999) www.mbl.org/mbl main/atlas.html
Comp Mouse Brains Hof et al. http://www.neuroscion.com/
laboratory/mousebrainmaps.html
Brain for Macintosh Nissanov, Tretiak http://ece.drexel.edu/ICVC
Mouse Atlas Project Baldock et al. http://genex.hgu.mrc.ac.uk
Rat-Brain Atlas Pich, Danckaert http://www.gwer.ch/qv/ratatlas/
ratatlas.htm
Mouse MRI Images Jacobs et al. http://www.gg.caltech.edu/hbp/
atlas.html
High-Resolution Sidman et al. http://www.hms.harvard.edu/
Mouse Atlas research/brain/
Virtual RatBrain Zaborszky et al. http://www.ratbrain.org/
CCDB Martone et al. http://ccdb.ucsd.edu
2. Virtual Rat Brain Project (Zaborszky et al., 2005a;

http://www.ratbrain.org/)
The software architecture is based on the NetBeans platform, which pro-

vides a modular component framework for Java applications. Layered on
top of this are library modules for connecting to the database, dealing with
MorphML, and viewing 3D data. The top layer is a collection of applications
for warping and submitting data to the database, browsing database con-
tents, and performing analysis on multiple data sets. Figure 20.9 shows the
system architecture that consists of three major entities: (a) the contributor
who does the data acquisition and registers the data into a standardized
Contributor User
Data
Warping Tool
Web Visualization and Analytical Tools
Acquisition
System
NL
files
Warped Data XML Searching on Download XML

Meta Data XML the metadata XML Queries data
Data Submission Web Services Data Services

Interface
Database
Data Center
Figure 20.9. System architecture of the Virtual Rat Brain Project. For more details,
see at website www.ratbrain.org.
coordinate space (reference brain) by using a special software (warping

tool) and uploads it to the database; (b) the data center that stores and pro-
vides data services to the user; and (c) the user who access the data by using
the analytical and visualization tools or web browser.
The warping tool was created to register data sets into a common coor-
dinate space. The GUI was designed for easy spatial navigation and spatial
point selection (Fig. 20.10). The loaded data sets are visualized in a 3D
window. The program generates and visualizes three transparent normal
section planes that give the basis of the spatial navigation. The user can
move these planes along the x, y , and z axes and define points where the
planes intersect. The GUI contains three additional 2D windows designed to
show “virtual slices” corresponding to each plane. The registration process
begins by loading the reference brain (from the database) and the data set
to be warped (from the local disk) into the same 3D space. Then the user
can define reference points in his/her data set and corresponding points
in the reference brain. When all the reference point pairs are defined, the
program creates an affine transformation and applies it to the experimental
data set on user demand. The result of the action is a new separate data file
that contains the experimental data transformed into the standard coordi-
nate space of the reference brain. At the end of the warping procedure,
the user is allowed to add descriptive data to the new data set. The tool is
able to save the generated data on the local disk as well as upload it to the
database. The tool is able to upload original, transformed, and descriptive
data. The program currently supports MicroBrightField’s Neurolucida files
and MDPlot’s Accustage files.
The overlapping analysis tool (Fig. 20.11) is suited to compare density and
overlap between two or more cell populations. The program divides the
Figure 20.10. The warping tool is used for registering data onto the reference brain
and submitting it to the database. Corresponding points for a newly acquired data
set and the reference contours can be selected and used to construct a mapping
function.
whole brain volume occupied by the data sets into boxes (bins, voxels) of
a given size and counts the cell types within each box. If the cell number
of each population within a box are equal to or above a certain thresh-
old, the program shows this box in a different color indicating the spatial
segregation or overlap. The user selects which cell populations to analyze.
The program also makes density measurements comparing the cell num-
bers from different populations in each box. The program can open multi-
ple data files for cross-brain analyses. A similar approach has been used by
Alloway et al. (1999) and Leergaard et al. (2000). There are several outputs
of the program including visualization of bin distributions and summarized
cell and bin numbers in table and Microsoft r
Excel format (Fig. 20.12;
http://www.ratbrain.org/modules/Tools/).
C. Cross-Correlation Function
Consider spike processes as oscillations. Then, the temporal interac-

tion between two cells can be captured as coherent oscillation of the two
spike processes, which is quantified by cross-correlation and cross-coherence
Figure 20.11. The overlap analysis tool allows comparison of overlap of population
data sets. The tree display on the left shows named features in the reference brain and
reference markers used for warping. The color of the bins is determined by the cell
type that is represented by the highest number in the particular bin. Red: cholinergic;
blue: calbindin; yellow: calretinin; green: parvalbumin; white: overlapping bins.
functions. The simple cross-correlation function is the product of two spike

processes r 1 and r 2 with different time lags τ applied within a time window
of stimulus or behavioral event S at time t:
C C 12S (τ ) = r 1S (t + τ )r 2S (t)t . (20.7)
Since part of the coherent oscillations derive from stimulus induced

nonstationarities of the firing rate, in practice we normalize the cross-
correlation function by the shift predictor to get a shift-predictor-corrected
cross-correlation function:
C C 12S (τ ) = r 1S (t + τ )r 2S (t) − m 1S (t + τ )m 2S (t)t , (20.8)
where m 1S and m 2S are the mean responses of the two neurons for stimu-
lus S at time t. The stimulus-condition-independent component of cross-
correlation can be obtained by averaging CC over S:
C C 12 (τ ) = C C 12S (t)s . (20.9)

Figure 20.12. The outputs of the overlapping analysis tools in Excel format. (A) The
table shows the bin numbers and the occupying cell numbers from a section. (B)
The colored table shows the spatial distribution of the bins from the same section.
The color of the bins is determined by the cell type that is represented by the highest
number in the particular bin. This example is from the basal forebrain (horizontal
limb of the diagonal band nucleus), and shows the locations in which the majority
of cells are cholinergic (CH, red) or calretinin-containing (CR, yellow). In each
bin the upper number represents the cholinergic cells and the lower number the
calretinin-containing neurons. Beige: both cell types are represented by an equal
number.
For the interpretation of cross-correlation function, see the text (see sec-
tion “Reconstruction of Functional Connectivity”).
Acknowledgments. We thank Grant Mulliken and Rajan Bhattacharyya

for valuable comments and suggestions and thank Peter Bartho for pro-
viding figure components for Fig. 20.8. The development of the “Virtual
Rat Brain Project” was supported by NIH/NINDS grant NS023945 to Laszlo
Zaborszky and funding from UK research councils (MRC and BBSRC) under
the eScience program to Fred W. Howell. We are thankful to Peter Varsanyi
and Ms. Nicola McDonnell for the database development.
REFERENCES
Abeles, M., and Gat, I., 2001, Detecting precise firing sequences in experimental data, J. Neurosci.
Methods 107(1–2):141–154; Erratum in J. Neurosci. Methods 112(2):203.
Abeles, M., and Gerstein, G. L., 1988, Detecting spatiotemporal firing patterns among simul-
taneously recorded single neurons, J. Neurophysiol. 60(3):909–924.
Abeles, M., and Goldstein, M. 1977, Multispike train analysis, Proc. IEEE 65:762–773.
Aertsen, A. M., Gerstein, G. L., Habib, M. K., and Palm, G., 1989, Dynamics of neuronal firing
correlation: modulation of “effective connectivity,” J. Neurophysiol. 61(5):900–917.
Aika, Y., Ren, J. Q., Kosaka, K., and Kosaka, T., 1994, Quantitative analysis of GABA-like-
immunoreactive and parvalbumin-containing neurons in the CA1 regions of the rat hip-
pocampus using a stereological method, the dissector, Exp. Brain Res. 99:267–276.
Allman, J. M. 1998, Evolving Brains, New York: Scientific American Library, W. H. Freeman and
Co.
Alloway, K. D., Crist, J., Mutic, J. J., and Roy, S. A., 1999, Corticostriatal projections from rat barrel
cortex have an anisotropic organization that correlates with vibrissal whisking behavior, J.
Neurosci. 19(24):10908–10922.
Baker, S. N., Spinks, R., Jackson, A., and Lemon, R. N., 2001, Synchronization in monkey motor
cortex during a precision grip task: I. Task-dependent modulation in single-unit synchrony,
J. Neurophysiol. 85(2):869–885.
Barbas, H., and Rempel-Clower, N., 1997, Cortical structure predicts the pattern of corticocor-
tical connections, Cereb. Cortex 7(7):635–646.
Bartho, P., Hirase, H., Monconduit, L., Zugaro, M., Harris, K. D., and Buzsaki, G., 2004, Char-
acterization of neocortical principal cells and interneurons by network interactions and
extracellular features, J. Neurophysiol. 92(1):600–608.
Binzegger, T., Douglas, R. J., and Martin, K. A. C., 2004, A quantitative map of the circuit of cat
primary visual cortex, J. Neurosci. 24(39):8441–8453.
Bower, J. M., and Beeman, D., 1998, The Book of GENESIS: Exploring Realistic Neural Models with
the General Neural Simulation System, 2nd ed., New York: Springer-Verlag.
Brecht, M., Fee, M. S., Garaschuk, O., Helmchen, F., Margrie, T. W., Svoboda, K., and Osten,
P., 2004, Novel approaches to monitor and manipulate single neurons in vivo, J. Neurosci.
24:9223–9227.
Brovelli, A., Ding, M., Ledberg, A., Chen, Y., Nakamura, R., and Bressler, S. L., 2004, Beta
oscillations in a large-scale sensorimotor cortical network: directional influences revealed
by Granger causality, Proc. Natl. Acad. Sci. U. S. A. 101(26):9849–9854.
Budd, J. M., and Kisvarday, Z. F., 2001, Local lateral connectivity of inhibitory clutch cells in
layer 4 of cat visual cortex (area 17), Exp. Brain Res. 140(2):245–250.
Buzsaki, G., 2004, Large-scale recording of neuronal ensembles, Nat. Neurosci. 7(5):446–451.
Buzsaki, G., Penttonen, M., Nadasdy, Z., and Bragin, A., 1996, Pattern and inhibition-dependent
invasion of pyramidal cell dendrites by fast spikes in the hippocampus in vivo, Proc. Natl.
Acad. Sci. U. S. A. 93(18):9921–9925.
Chen-Bee, C. H., Polley, D. B., Brett-Green, B., Prakash, N., Kwon, M. C., and Frostig, R. D.,
2000, Visualizing and quantifying evoked cortical activity assessed with intrinsic signal
imaging, J. Neurosci. Methods 97(2):157–173.
Chevalier, G., and Mana, S., 2000, Honeycomb-like structure of the intermediate layers of
the rat superior colliculus, with additional observations in several other mammals: AChE
patterning, J. Comp. Neurol. 419(2):137–153.
Claverol, E., and Nadasdy, Z., 2004, Intersection of microwire electrodes with proximal CA1
stratum-pyramidal neurons at insertion for multiunit recordings predicted by a 3-D com-
puter model, IEEE Trans. Biomed. Eng. 51(12):2211–2216.
Coogan, T. A., and Burkhalter, A., 1993, Hierarchical organization of areas in rat visual cortex,
J. Neurosci. 13(9):3749–3772.
Csicsvari, J., Henze, D. A., Jamieson, B., Harris, K. D., Sirota, A., Bartho, P., Wise, K. D., and
Buzsaki, G., 2003, Massively parallel recording of unit and local field potentials with silicon-
based electrodes, J. Neurophysiol. 90(2):1314–1323.
Das, A., and Gilbert, C. D., 1995, Long-range horizontal connections and their role in cor-
tical reorganization revealed by optical recording of cat primary visual cortex, Nature
375(6534):780–784.
Das, A., and Gilbert, C. D., 1999, Topography of contextual modulations mediated by short-
range interactions in primary visual cortex, Nature 399(6737):655–661.
Destexhe, A., and Babloyantz, A., 1993, A model of the inward current Ih and its possible role
in thalamocortical oscillations, Neuroreport 4(2):223–226.
Douglas, R., Martin, K., and Witteridge, D., 1989, A canonical microcircuit for neocortex.
Neural. Comput. 1:480–488.
Fee, M. S., Mitra, P. P., and Kleinfeld, D., 1996, Automatic sorting of multiple unit neuronal
signals in the presence of anisotropic and non-Gaussian variability, J. Neurosci. Methods
69(2):175–188; Erratum in J. Neurosci. Methods 71(2):233.
Felleman, D. J., and Van Essen, D. C., 1991, Distributed hierarchical processing in the primate
cerebral cortex. Cereb. Cortex 1:1–47.
Fiala, J. C., and Harris, K. M., 2001, Extending unbiased stereology of brain ultrastructure to
three-dimensional volumes, J. Am. Med. Inform. Assoc. 8(1):1–16.
Földy, C., Dyhrfjeld-Johnsen, J., and Soltesz, I., 2005, Structure of cortical microcircuit theory,
J. Physiol. 562.1:47–54.
Freeman, W. J., 2000, Mesoscopic neurodynamics: from neuron to brain, J. Physiol. (Paris)
94(5–6):303–322.
Freund, T. F., and Buzsaki, G., 1996, Interneurons of the hippocampus, Hippocampus 6(4):347–
470.
Frostig, R. D., Frostig, Z., and Harper, R. M., 1990, Recurring discharge patterns in multiple
spike trains: I. Detection, Biol. Cybern. 62(6):487–493.
Fukuda, T., and Kosaka, T., 2000, Gap junctions linking the dendritic network of GABAergic
interneurons in the hippocampus, J. Neurosci. 20(4):1519–1528.
Gerfen, C. R., 1985, The neostriatal mosaic: I. Compartmental organization of projec-
tions from the striatum to the substantia nigra in the rat, J. Comp. Neurol. 236(4):454–
476.
Gibson, J. R., Beierlein, M., and Connors, B. W., 1999, Two networks of electrically coupled
inhibitory neurons in neocortex, Nature 402(6757):75–79.
Gilbert, C. D., 1983, Microcircuitry of the visual cortex, Annu. Rev. Neurosci. 6:217–247.
Gilbert, C. D., and Wiesel, T. N., 1983, Functional organization of the visual cortex. Prog. Brain
Res. 58:209–218.
Goldman, P. S., and Nauta, W. J., 1977, Columnar distribution of cortico-cortical fibers in the
frontal association, limbic, and motor cortex of the developing rhesus monkey, Brain Res.
122(3):393–413.
Goldman-Rakic, P. S., 1984, Modular organization of prefrontal cortex, Trends Neurosci. 7:419–
429.
Granger, C. W. J., 1969, Investigating causal relations by econometric methods and cross-
spectral methods, Econometrica, 34:424–438.
Graybiel, A. M., and Ragsdale, C. W., Jr., 1978, Histochemically distinct compartments in the
striatum of human, monkeys, and cat demonstrated by acetylthiocholinesterase staining,
Proc. Natl. Acad. Sci. U. S. A. 75(11):5723–5726.
Grinvald, A., and Hildesheim, R., 2004, VSDI: a new era in functional imaging of cortical
dynamics, Nat. Rev. Neurosci. 5(11):874–885.
interneurons and synapses in the neocortex, Science 287(5451):273–278.
Harris, K. D., Csicsvari, J., Hirase, H., Dragoi, G., and Buzsaki, G., 2003, Organization of cell
assemblies in the hippocampus, Nature 424(6948):552–556.
Harris, K. D., Henze, D. A., Csicsvari, J., Hirase, H., and Buzsaki, G., 2000, Accuracy of tetrode
spike separation as determined by simultaneous intracellular and extracellular measure-
ments, J. Neurophysiol. 84(1):401–414.
He, W., Hamilton, T. A., Cohen, A. R., Holmes, T. J., Pace, C., Szarowski, D. H., Turner, J. N.,
and Roysam, B., 2003, Automated three-dimensional tracing of neurons in confocal and
brightfield images, Microsc. Microanal. 9(4):296–310.
Hebb, D. O., 1949, Organization of Behavior, New York: Wiley.
Henze, D. A., Borhegyi, Z., Csicsvari, J., Mamiya, A., Harris, K. D., and Buzsaki, G., 2000,
Intracellular features predicted by extracellular recordings in the hippocampus in vivo, J.
Hines, M. L., and Carnevale, N. T., 1997, The NEURON simulation environment, Neural Comput.
9:1179–1209.
Hubel, D. H., and Wiesel, T. N., 1959, Receptive fields of single neurones in the cat’s striate
cortex, J. Physiol. 148:574–591.
Hubel, D. H., and Wiesel, T. N., 1972, Laminar and columnar distribution of geniculo-cortical
fibers in the macaque monkey, J. Comp. Neurol. 146(4):421–450.
Ikegaya, Y., Aaron, G., Cossart, R., Aronov, D., Lampl, I., Ferster, D., and Yuste, R., 2004, Syn-
fire chains and cortical songs: temporal modules of cortical activity, Science 304(5670):
559–564.
Isseroff, A., Schwartz, M. L., Dekker, J. J., and Goldman-Rakic, P. S., 1984, Columnar organiza-
tion of callosal and associational projections from rat frontal cortex, Brain Res. 293(2):213–
223.
Johnson, R. R., and Burkhalter, A., 1997, A polysynaptic feedback circuit in rat visual cortex, J.
Neurosci. 17(18):7129–7140.
Kalisman, N., Silberberg, G., and Markram, H., 2005, The neocortical microcircuit as a tabula
rasa, Proc. Natl. Acad. Sci. U. S. A. 102(3):880–885.
Kaminski, M., Ding, M., Truccolo, W. A., and Bressler, S. L., 2001, Evaluating causal relations
in neural systems: granger causality, directed transfer function and statistical assessment
of significance, Biol. Cybern. 85(2):145–157.
Kaminski, M. J., and Blinowska, K. J., 1991, A new method of the description of the information
flow in the brain structures, Biol. Cybern. 65:203–210.
Kemere, C., Shenoy, K. V., and Meng, T. H., 2004, Model-based neural decoding of reaching
movements: a maximum likelihood approach, IEEE Trans. Biomed. Eng. 51(6):925–932.
Kisvarday, Z. F., Ferecsko, A. S., Kovacs, K., Buzas, P., Budd, J. M., and Eysel, U. T., 2002, One
axon-multiple functions: specificity of lateral inhibitory connections by large basket cells,
J. Neurocytol. 31(3–5):255–264.
Koralek, K. A., Olavarria, J., and Killackey, H. P., 1990, Areal and laminar organization of
corticocortical projections in the rat somatosensory cortex, J. Comp. Neurol. 299(2):133–
150.
Kötter, R., Hilgetag, C. C., and Stephan, K. E., 2001, Connectional characteristics of areas in
Walker’s map of prefrontal cortex, Neurocomputing 38–40:741–746.
Lee, A. K., and Wilson, M. A., 2002, Memory of sequential experience in the hippocampus
during slow wave sleep, Neuron 36(6):1183–1194.
Leergaard, T. B., Alloway, K. D., Mutic, J. J., and Bjaalie, J. G., 2000, Three-dimensional topogra-
phy of corticopontine projections from rat barrel cortex: correlations with corticostriatal
organization, J. Neurosci. 20(22):8474–8484.
Leise, E. M. 1990, Modular construction of nervous system: a basic principle of design for
invertebrates and vertebrates, Brain Res. Rev. 15:1–23.
Lewicki, M. S., 1998, A review of methods for spike sorting: the detection and classification of
neural action potentials, Network 9(4):R53–R78 (Review).
Li, W., and Gilbert, C. D., 2002, Global contour saliency and local collinear interactions, J.
Lubke, J., Egger, V., Sakmann, B., and Feldmeyer, D., 2000, Columnar organization of dendrites
and axons of single and synaptically coupled excitatory spiny neurons in layer 4 of the rat
barrel cortex, J. Neurosci. 20(14):5300–5311.
Maccaferri, G., and Lacaille, J. C., 2003, Interneuron diversity series: hippocampal interneuron
classifications—making things as simple as possible, not simpler, Trends Neurosci. 26:564–
571.
Malach, R., 1994, Cortical columns as devices for maximizing neuronal diversity, Trends Neurosci.
17(3):101–104 (Review).
Malmierca, M. S., Leergaard, T. B., Bajo, V. M., Bjaalie, J. G., and Merchan, M. A., 1998,
Anatomic evidence of a three-dimensional mosaic pattern of tonotopic organization in
the ventral complex of the lateral lemniscus in cat, J. Neurosci. 18(24):10603–10618.
Mountcastle, V. B., 1957, Modality and topographic properties of single neurons of cat’s somatic
sensory cortex, J. Neurophysiol. 20(4):408–434.
Mountcastle, V. B., 1998, Perceptual Neuroscience: The Cerebral Cortex, Harvard University Press,
Cambridge, MA.
Musallam, S., Corneil, B. D., Greger, B., Scherberger, H., and Andersen, R. A., 2004, Cognitive
control signals for neural prosthetics, Science, 305(5681):258–262.
Nadasdy, Z., 2000, Time sequences and their consequences, J. Physiol. (Paris) 94:505–524.
Nadasdy, Z., Csicsvari, J., Penttonen, M., Hetke, J., Wise, K., and Buzsáki, G., 1998, Extracellular
recording and analysis of neuronal activity: from single cells to ensembles, In: Eichenbaum,
H. B., and Davis, J. L. (eds.), Neuronal Ensembles: Strategies for Recording and Decoding, New
York: Wiley.
Nadasdy, Z., Hirase, H., Czurko, A., Csicsvari, J., and Buzsaki, G., 1999, Replay and time com-
pression of recurring spike sequences in the hippocampus, J. Neurosci. 19(21):9497–9507.
networks, Anat. Embryol. (Berl.) 204(4):303–317.
Nordhausen, C. T., Maynard, E. M., and Normann, R. A., 1996, Single unit recording capabilities
of a 100 microelectrode array, Brain Res. 726(1–2):129–140.
Ohki, K., Chung, S., Ch’ng Y. H., Kara, P., and Reid, R. C., 2005, Functional imaging with cellular
resolution reveals precise micro-architecture in visual cortex, Nature 433(7026):597–603.
Palm, G., Aertsen, A. M., and Gerstein, G. L., 1988, On the significance of correlations among
neuronal spike trains, Biol. Cybern. 59(1):1–11.
Paperna, T., and Malach, R., 1991, Patterns of sensory intermodality relationships in the cere-
bral cortex of the rat, J. Comp. Neurol. 308(3):432–456.
Peters, A., and Feldman, M. L., 1976, The projection of the lateral geniculate nucleus to area
17 of the rat cerebral cortex: I. General description, J. Neurocytol. 5:63–84.
Peters, A., and Payne, B. R., 1993, Numerical relationships between geniculocortical afferents
and pyramidal cell modules in cat primary visual cortex. Cereb. Cortex 3:69–78.
Pouille, F., and Scanziani, M., 2004, Routing of spike series by dynamic circuits in the hip-
pocampus, Nature 429(6993):717–723.
Prinz, A. A., Bucher, D., and Marder, E., 2004, Similar network activity from disparate circuit
parameters, Nat. Neurosci. 7(12):1287–1288.
Pucak, M. L., Levitt, J. B., Lund, J. S., and Lewis, D. A., 1996, Patterns of intrinsic and associa-
tional circuitry in monkey prefrontal cortex, J. Comp. Neurol. 376(4):614–630.
Quian-Quiroga, R., Nadasdy, Z., and Ben-Shaul, Y., 2004, Unsupervised spike detection and
sorting with wavelets and superparamagnetic clustering, Neural Comput. 16(8):1661–1687.
Rakic, P., 1995, Radial versus tangential migration of neuronal clones in the developing cerebral
cortex, Proc. Natl. Acad. Sci. U. S. A. 92(25):11323–11327.
Rempel-Clower, N. L., and Barbas, H., 2000, The laminar pattern of connections between
prefrontal and anterior temporal cortices in the Rhesus monkey is related to cortical
structure and function, Cereb. Cortex 10(9):851–865.
Rockland, K. S., 1998, Complex microstructures of sensory cortical connections, Curr. Opin.
Neurobiol. 8(4):545–551 (Review).
Rodriguez, A., Ehlenberger, D., Kelliher, K., Einstein, M., Henderson, S. C., Morrison, J. H., Hof,
P. R., and Wearne, S. L., 2003, Automated reconstruction of three-dimensional neuronal
morphology from laser scanning microscopy images, Methods 30(1):94–105.
Somogyi, P., and Klausberger, T., 2005, Defined types of cortical interneurone structure space
and spike timing in the hippocampus, J. Physiol. (London) 562:9–26.
Somogyi, P., Tamas, G., Lujan, R., and Buhl, E. H., 1998, Salient features of synaptic organisation
in the cerebral cortex. Brain Res. Brain Res. Rev. 26(2–3):113–135.
Sporns, O., and Kötter, R., 2004, Motifs in brain networks, PLoS Biol. 2(11):e369.
Stepanyants, A., Hof, P. R., and Chklovskii, D. B., 2002, Geometry and structural plasticity of
synaptic connectivity, Neuron 34(2):275–288.
wiring, Neuron 43(2):251–259.
Szentagothai, J., 1978, The neuron network of the cerebral cortex: a functional interpretation.
Proc. R. Soc. Lond. B. 201:219–248.
Szentagothai, J., 1983, The modular architectonic principle of neural centers, Rev. Physiol.
Biochem. Pharmacol. 98:11–61.
Tamas, G., Buhl, E. H., Lorincz, A., and Somogyi, P., 2000, Proximally targeted GABAergic
synapses and gap junctions synchronize cortical interneurons, Nat. Neurosci. 3(4):366–371.
Tamas, G., Buhl, E. H., and Somogyi, P., 1997, Massive autaptic self-innervation of GABAergic
neurons in cat visual cortex, J. Neurosci. 17(16):6352–6364.
Traub, R. D., Jefferys, J. G. R., and Whittington, M. A., 1999, Fast Oscillations in Cortical Circuits,
Bradford Books, MIT Press, Cambridge, MA.
Vetter, R. J., Williams, J. C., Hetke, J. F., Nunamaker, E. A., and Kipke, D. R., 2004, Chronic neural
recording using silicon-substrate microelectrode arrays implanted in cerebral cortex, IEEE
Trans. Biomed. Eng. 51(6):896–904.
Vreeswijk, C. V., 1996, Partial synchronization in populations of pulse-coupled oscillators, Phys.
Rev. E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics 54(5):5522–5537.
Wang, X. J., and Rinzel, J., 1993, Spindle rhythmicity in the reticularis thalami nucleus: syn-
chronization among mutually inhibitory neurons, Neuroscience 53(4):899–904.
Watts, J., and Thomson, A. M., 2005, Excitatory and inhibitory connections show selectivity in
the neocortex, J. Physiol. (London) 562:89–97.
Wickelgren, W. A., 1999, Webs, cell assemblies, and chunking in neural nets: introduction, Can.
J. Exp. Psychol. 53(1):118–131.
Yoshimura, Y., Dantzker, J. L., and Callaway, E. M., 2005, Excitatory cortical neurons form
fine-scale functional networks, Nature 433(7028):868–873.
Yuste, R., and Denk, W., 1995, Dendritic spines as basic functional units of neuronal integration,
Nature 375(6533):682–684.
Zaborszky, L., 2002, The modular organization of brain systems. Basal forebrain: the last fron-
tier, Prog. Brain Res. 136:359–372.
Zaborszky, L., Buhl, D. L., Pobalashingham, S., Bjaalie, J. G., and Nadasdy, Z., 2005b, Three-
dimensional chemoarchitecture of the basal forebrain: spatially specific association of
cholinergic and calcium binding protein-containing neurons, Neuroscience, 136(3):697–
713.
Zaborszky, L., Csordas, A., Buhl, D. L., Duque, A., Somogyi, J., and Nadasdy, Z., 2002, Com-
putational anatomical analysis of the basal forebrain corticopetal system, In: Ascoli, G. A.
(ed.), Computational Neuroanatomy: Principles and Methods, New Jersey: Humana Press.
Zaborszky, L., Varsanyi, P., Mc Donnell, N. C., and Howell, F. W., 2005a, 3D cellular database of
the rat brain: integrated anatomical data, Soc. Neurosci. Abstr. 31. Program No. 570.3.2005
Abstract Viewer/Itinerary Planner. Washington, DC: Society for Neuroscience, 2005.Online.
Zaborszky, L., and Wolff, J. R., 1982, Distribution patterns and individual variations of callosal
connections in the albino rat, Anat. Embryol. (Berlin) 165(2):213–232.
Zhang, K., Ginzburg, I., McNaughton, B. L., and Sejnowski, T. J., 1998, Interpreting neuronal
population activity by reconstruction: unified framework with application to hippocampal
place cells, J. Neurophysiol. 79(2):1017–1044.
Index
1,2-Bis(o-aminophenoxy)ethane-N, N, axonal connectivity, digital conversion of

N N -tetraacetic acid (BAPTA), 455 semiautomated vectorization procedure,
1-naphthol/azur B, 353 612
3,3 -diaminobenzidine (DAB), 11 computational derivation for, 614–615
models, 615–616
A axonal flow, 339
ABI Prism 7700 sequence detection system,
157 B
acetoxy-methyl (AM) ester calcium Bartha PRV
indicators, 458 as neuranatomical tracer
derivatives of, 460 with conventional neural tracers,
acridine orange histofluorescence, 113–114, 269–271
133 dendritic filling by, 273–274
anesthetics, 187–188 double retrograde tracing techniques
agonal state, 113 with, 271–273
airy disk, 399 electrophysiological recordings, 274
airy distribution, 399 genetically engineered, for highly
aldehydes, 112–113 specific tracing, 274–276
Alexa FluorTM 488, 413 phenotypic characterization of
Alexa FluorTM 633, 416, 420 PRV-labeled neurons, 269
Alexa FluorTM 647, 417 sources of, 293
Alexa488-dextran, 457 specific retrograde transport of,
Alexa594, 455 276–277
Allen brain atlas, 574 transneuronal transfer of, at synaptic
alpha-amino-3-hydroxy-5-methyl-4-isoxazole- terminals, 277–281
propionic acid (AMPA) receptors, 83 basal forebrain neurons, 145, 149, 161
Alzheimer’s disease (AD), 115, 143, 145 general methodologies, in labeling
AmiraTM , 425 procedure of, 200
amydgala, role of, 145 anesthesia, 202
amyotrophic lateral sclerosis (ALS), 129 control experiments, 208–210
antemortem variables, 111–113 electrodes used in, 203–205
anterograde degeneration, 18 histological and immunohistochemical
advantages of, 19–20 procedures, 210–212
anterograde neuroanatomical tracer, 340 electron microscopy reconstructions,
antibody–antigen coupling, 81 210–212
antibody–epitope interaction, 85 juxtacellular labeling, 205–208
anti-PHA-L antibodies, 347 juxtacellular labeling markers, choice
antisense RNA (aRNA), 111 of, 202–203
artificial cerebrospinal fluid (ACSF), 457 recording apparatus used in, 203–205
autoradiographic method, 338 recording solutions used in, 202–203
avidin–biotin complex method, 181 tracer techniques, 212–213
avidin–biotin peroxidase (ABC) method, basal ganglia connectivity, 343
347 BDA labeling characteristics
axonal collateralization, 339 at EM level, 329
axonal configurations, 338 at LM level, 327–328
682
INDEX 683
BDA protocols human map, of Internationational
for EM double-labeling studies, 327 consortium of human brain mapping,
for EM studies 576
BDA visualization, 327 maps and atlases based on other
fixation, 326–327 modalities, 574
injection, 326–327 Brodmann map, 568, see also classical brain
sectioning, 326–327 atlases
for LMdouble-labeling studies buffering curve, 164
BDA visualization, 325–326
fixation, 325 C
injection, 325 calcium green-1 (CG-1), 455
for LM single-labeling studies calcium homeostasis, during aging, 145
BDA visualization, 324 calcium-binding proteins (CaBP), 146
fixation, 323 camera, commercial sources of, 446
injection, 322–323 CapSure, 117
mounting and viewing, 325 HS LCM Cap, 117
sectioning, 323–324 carbocyanine dye tracing
BDA solution for injection, preparation of, in human pathological specimens, 384
358–359 with immunohistochemistry, 378
BDA solution, preparation of, 358–359 limitations of, 380–381
BDA track labeling, 349 methodology of
BDA10kDa, 307, 309–311, 313, 318 analysis of, 376–377
anterograde labeling with counterstaining, 376–377
neuroanatomical tracers, 313–315 fixation and storage of tissue, 372
BDA3kDa, 307, 310, 313, 318 incubation period, 374–375
retrograde labeling with neuroanatomical mounting, 376–377
tracers, 316 sectioning of tissue, 375
benzidine hydrochloride, 354 tracer delivery, 373
bidirectional tracer, 339 tracer targeting, 374
biocytin, 178 in postmortem human tissue, 383–384
biotin, 341 in primates, 384–385
biotin–lysine complex, 181 technical consideration, 380–381
Biotinylated Dextran Amine, 21–22, 306, caspase-3 antibody, 129–130
340–341, 349, see also BDA caveolae, 349
“bleedthrough”, in filter settings, 377 cDNA clones, use of, 240
blood-oxygen-level-dependent cDNA protocols, amplification of, 145
(BOLD)signal, 567 cDNAarray analysis, 116–117
brain anatomy, intersubject variability of cell identification, 144
brain macroscopy variability, 577–578 cell somata, 338, 340
cortical area and macroanatomical central nervous system, 1
landmarks, 579–580 cholera toxin β-subunit (CTb), 264, 339,
image registration techniques, 351
applicability of, 581–583 choline acetyltransferase (ChAT), 146
microconstructural variability, 578–579 cholinergic basal forebrain neurons, 116
brain atlases Ciphergen ProteinChip reader, 130
advantages of, 591 clathrin-mediated endocytotic vesicles,
atlas tools and links of, 593 349
classical cloning, 169–170
Brodmann map, 569–570 cognitive processes, 145
other historical maps, 571 colloidal gold particles, 74
limitations, 592 application of, 74
three-dimensional anatomical features of, 74
atlas of Talairach and Tournoux, confocal laser scanning microscope (CLSM),
571–573 398
fiber tract maps, 574–576 parameters in, 406–407
684 INDEX
confocal laser scanning microscopy (CLSM), digital morphometry, of single neurons

307 computer acquisition, 606–608
confocal laser scanning microscopy, digital files, 608
339 dendritic modeling, 610–612
confocal microscopy, 377 DiI label, see also carbocyanine dye tracing
cortical circuitries photoconversion of, 379–380
basic circuitry in, 635–638 use of, 380
cell assemblies in, 646–647 DiI placement and histological processing,
columnar organization of neocortex in, protocol for, 387–388
643–644 dinitrophenol, 341
horizontal organization of, 641–643 discrete cell microdissection methodologies,
neurons, classes of, 634–635 111
synaptic reconstructions in, 644–645 distal arborizations, 338
vertical organization of, 638–641 DNA preparation, 169–170
cortical circuitries, reconstruction from dye separation program, 418
dynamic data
dynamic system approach, 654 E
effective connectivity, 662–667 electrophysiology rigs, 115
large scale recording of, 654–659 endogenous peroxidase activity, 359
reconstruction of functional enkephalin-positive (ENK-ir), 353
connectivity, 659–662 ENK-ir terminals, 353
static data enzyme analysis and sequencing, 170
connectivity matrix, 647 epifluorescence analysis, 377
databases, 653 Eppendorf transjector, for single-cell
importance of 3D reconstruction, microaspiration, 133
648–649 Ernst Abbe’s equation, 398
statistical modeling, 649–653 etching step, 95, 97
cresyl violet, 116
cross-correlogram, computational algorithm F
of, 661–662 Falck–Hillarp fluorescence method, 2
cryoprotection, 359 fiber tractography, 574
cytoarchitectonic map, 568 fixatives, 112
generation of, 592–593 choice of, 113
Cytoplasmic RNAs, 113 fluo-4, 455
fluorescence, photoconversion of, 340
D fluorescence-based retrograde tracing, 339
DAB–Ni reaction product, immunolabeling fluorescent dyes, names of,
of, 191–192 diamidino yellow, 339
deconvolution, 401 Evans blue, 339
“delayed-fixation” approach, 386 fast blue, 339
dendrites, 338, 340 Lucifer yellow, 339
dendritic region of interest (ROI), 454 nuclear yellow, 339
dentate gyrus granule cells, 119 propidium iodide, 339
dextran amines, 306–307 true blue, 339
anterograde labeling with, 307–310 fluoroemerald, 307
labeling with immunohistochemistry, fluorogold (FG), 119, see also retrograde
316–318 tract-tracing
limitations of, 320–322 tracing, 340, 351
merits of, 318–320 fluororuby, 307
retrograde labeling with, 310 FluVRTM (SVI), 424
retrograde collateral labeling of, formaldehyde concentration, 92
310–313 freeze fracture–immunogold labeling
diffusion tensor imaging, method of, 3–4 technique, 80, 84
diffusion tensor MRI (DT-MRI) technique, freeze substitution procedure, 94
386 advantages of, 94
INDEX 685
functional magnetic resonance imaging imaged dendritic segments
techniques, 437, 567 electron microscopic reconstruction of,
functional magnetic resonance tomography 472–473
(fMRI), 567 ultrastructural identification of, 458
imaging software, commercial sources of,
G 446
GABAergic inhibition, 444 IMGAP, see IMmuno-Gold-Analysis-Program
GAPDH expression level, 159 immunoblotting, quantitative, 83
gel-based identification versus real-time immunocytochemical staining, 359
fluorescent PCR analysis immunocytochemically labeled cells, 117
agarose gel PCR, 154–155 immunocytochemistry, 113, 116
semiquantitative real-time PCR, 155–15 of transmission electron microscopy
gel-based PCR, optimization for, 171 antibody controls, 46
gene expression profiling antibody dilutions, 45–46
using fixed tissues antibody incubations, 45
acridine orange histofluorescence and penetration enhancement, 41
bioanalysis, 113–114 preembedding immunogold-silver,
antemortem and postmortem variables, 44–45
111–113 preembedding immunoperoxidase,
Gerhardt, map of, 568 42–44
glial fibrillary acidic protein (GFAP), 129 postembedding procedures, 88–100
glutamate, 239 dehydration and embedding, 93–95
glutamic acid decarboxylase (GAD), 146 fixation, 90–93
glutaraldehyde concentration, 92 immunoincubation, 95–100
glyceraldehyde 3-phosphate dehydrogenase immunogold cytochemistry
(GAPDH), 146 applications, 88
glycine, 239 controls, 85–88
glycoprotein enzyme horseradish peroxidase antibody testing, 85, 87
(HRP), 338 quantification of, 81–85
golden rule requirements, for tracer, resolution of, 75–81
345–346 lateral, 75–76
Golgi method, 1, 198–199 length factor in, 75
Golgi silver impregnation, 338 IMmuno-Gold-Analysis-Program, 84,
Granger causality, 664 103
immunogold-silver, criteria for, 51
H immunoincubation procedure, 95, 101
hematoxylin and eosin (H&E) staining, immunoperoxidase
116 criteria for, 50–51
hippocampal CA1 neurons, 119 technique, 11
hippocampal CA3 neurons, 119 in situ hybridization, steps in
histological stains, 116 hybridization, 245, 256
horseradish peroxidase (HRP), 178, 199 prehybridization, 245, 256
HRP transport technique, 339 rinsing, 246, 257
human serum albumin (HSA), 97 in vivo extracellular recording, for neural
Huygens II professional software, 407 networks, 178–180
hybridization signal intensity, 115 advantages of, 186
hydroxystilbamidine derivative, 340 limitations of, 177, 186–187
preparation methods of, 188
I use of metal wires, 179
image registration techniques, for data in vivo intracellular recording, for neural
transformation networks, 180–181
affine transformation, 581–582 advantages of, 186
medium complexity transformation, 582 limitations of, 177, 186–187
nonlinear and nonrigid transformations, preparation methods of, 188–189
582–583 use of sharp electrodes, 180
686 INDEX
in vivo recording negative extraction, 117–118

chemicals, 193 positive extraction, 117
materials, 193 linear unmixing program, 418
solutions, 193–194 Lucifer yellow, 178, 181
in vivo tractography, 3
injected tracers, histological processing of, M
357–358 magnetoencephalography (MEG),
injection site events 567
BDA track labeling, 349–350 manganese chloride (MnCl2), 386
CTB delivery, 351 MatLab software, 461
deposition of tracer, 352–353 McIntyre HSV1, 290
fluoro-gold delivery, 351–352 McIntyre-B strain, 290
PHA-L delivery, 351 microarray analysis, 119
size and shape of, 350 of microdissected samples, 125–126
survival time, 352 microarray studies, 114
uptake mechanism, 349 microdissection studies, 114
Institutional Animal Care and Use microdissection, methods of
Committee, 200 advantages and limitations, 131–
intracardial perfusion, for transmission 132
electron microscopy LCM techniques, 114
choice of fixative, 39–40 single-cell microaspiration, 114
postfixation and sectioning, 40–41 microdissections of
pretreatments, 39 basal forebrain, 115
intracellularly filled cells, visualization hippocampal formation, 115
preparation of, 190–191 midbrain accumbens, 115
iontophoresis, 352 nucleus accumbens, 115
iontophoretic delivery, 340 microfuge tube, 117
microruby, 307
J midbrain dopaminergic neurons, 116
juxtacellular labeling, of single neurons, miniruby, 307
application of molecular biological methodology, 116
advantages of, 226 molecular fingerprint, 116
electrophysiological recordings, 215– morphometric parameters with L-measure,
216 extraction of, 625–626
limitations of, 226 multiple tracing, rules for successful,
morphological recordings, 215–216 345–346
retrograde labeling, 216–218 multiplex PCR, 145–146
juxtacellularly labeled neurons multitracing experiment, completion of,
chemical identification and morphometry 334
of, 218–221 myelin stains, see postmortem silver
synaptology of electrophysiologically and techniques, use of
chemically identified neurons, myelinated fiber bundles, 3
221–225 myelinated pathways, 3
K N
kissing fibers, 576 Nauta–Gygax method, 1
network modeling
L anatomically realistic, 619–621
labeling efficiency, 84 application perspectives, 624–625
laser capture microdissection, see LCM physiological relevance
LCM, 111, 117–119 design principles, 623–624
in combination with proteomic morphology on neuronal
applications, 126–131 electrophysiology, application of,
instrumentation, 118 621–622
principal means of network dynamics, 622–623
INDEX 687
system-level boundaries and virtual calcium indicators, bulk loading of,
stereology, 616–619 460–461
neural connections imaging ensemble activity, 461–463
artifacts in principle component analysis
deformation, 489–493 advantages and limitations, 470–471
observation, 487–489 example, 464–468
assessment of, 478–479 fundamental concept of, 463–464
by axonal length, 502 two-photon calcium imaging, in dendrites,
isotropy reefs, 503–507 454, 471–472
linear features of, 508–509 advantages and limitations, 468–470
by axon numbers calcium indicator selection, 455–456
in 2D, 498–499 simulation electrode, targeting of,
preterminal and terminal fields, 456–457
499–502 single synapses synaptic activation,
bias estimation in, 486–487 457–458
case study, 518–522 neuronal preparations, method of,
morphometry and stereology, 484–486 166–167
advantages and limitations, 516–518 neurons, reconstruction after intracellular
by neuronal numbers, 494–497 labeling, 192–193
precise measurement of, 515–516 neuropeptide detection, 339
by synapse number, 509–515 neurotransmitter detection, 339
target parameters of nickel-enhanced DAB (DAB-Ni), 353
parameters, 480–484 nigral dopaminergic neurons, 343
structures identified, 479–480 NIH, 117
neural mechanisms, microstructural Nissl and immunocytochemical stains, 116,
localization of, 588–591 376
neural networks noncontact laser extraction, see negative
combined techniques, 180, 186 extraction
in vivo recording nonradioactive riboprobes technique
advantages, 186 advantages of, 253–254
challenges in, 177, 186–187 application of
extracellular, 178–180 ISH combined with juxtacellular
intracellular, 180–181 labeling, 252
single-cell labeling, 180–181 tract-tracer combined with double ISH,
neuroanatomical data, parametric modeling 250
of, 669–672 tract-tracer combined with ISH,
neuroanatomical tracers, groups of 247–249, 250–251
anterograde tracers, 338 viral tracing combined with double ISH,
retrograde tracers, 338 252–253
neuroanatomical tract-tracing techniques, disadvantages, 254
338 methodological considerations
classical, 368 cDNA clone production, 241
in human specimens, 382 cDNA linear template production,
modern tract-tracing techniques, using 243–244
carbocyanine dye, 368–372 cDNA plasmid transformation,
specifications of, 369 241–243
tracers used, 369 controls, 246–247
trends in, 386 immunocytochemistry for revealing
neuroanatomist, 114 digoxigenin and proteins, 246,
neurocrine, 478 257
neuronal circuit analyses, at multiple levels, in situ hybridization, 245–246, 256
methods of in vitro transcription of cRNA, 244
neuronal ensembles imaging, 458–460, modifications, for double ISH, 246,
473–474 258–259
advantages and limitations, 470 riboprobe design, 241
688 INDEX
nonradioactive riboprobes technique (cont.) postmortem variables, 111–113

test for incorporation of nonradioactive preembedding immunogold approach, 11
label using a dot blot on a nytran presynaptic boutons, 510
strip, 244, 255–256 primate caudate–putamen, 343
tissue preparation, 244–245 Primer Express r , 153
PrimerQuest tool, 241
O primers and probe lists
oligonucleotide probes, use of, 240 for gel-based PCR, 168
radioactive-labeled, 240 for RT-PCR, 168
one-round single-cell RT-PCR protocol, procion yellow, 178
146 ProSAP2/Shank3, 396, 425, 427
optical microscope ProteinChip array, 127, 129
optical theory, practical implications of, ProteinChip biology system, 127
398 ProteinChip reader, 127
resolution and image improvements proteomics, 119
deconvolution, 404–405 PRV neuroanatomical tracing, pros and cons
monochromatic light, use of, 401–403 viral labeling patterns, interpretation of
pinhole application, 403–404 route determination, 286–287
pixelising the image, 404 thorough neuroanatomical
shape of PSF, 405–406 hypothesis-testing, 289
oregon green BAPTA-1 (OGB-1), 455 viral tracing, in CNS, 281
PRV injection site, 282
P Bartha PRV, in ventricular system,
pallidal outflow, 343 282–283
PALM Microlaser Technologies, 117 Bartha PRV, in vasculature, 283
PALM system, 117 uptake by fibers of passage, 283–284
parvalbumin (PV), 186 controlled viral tracing, in CNS,
PCR primers and probe design, 152–153 284–285
PCR solutions, method of, 167–168 viral tracing, in peripheral nervous system,
pCRII-TOPO vector, 243 285–286
peroxidase substrates, commonly used, 347 viral transneuronal tracing, data after,
peroxidase substrates, different types of, 289–290
DAB, 354 pseudorabies virus (PRV)see also retrograde
DAB-Ni, 354 tract-tracing
V-VIP, 354 dilution of, 295
PHA-L, 20, 265, 305, 340, 347, 349, 351, 409 disposal of PRV-infected material, 296
disadvantages of, 21 immunohistochemical staining, 296
tracing, 341, 349 injection of, 295
Phaseolus vulgaris leucoagglutinin, see PHA-L storage of, 293–295
phenotypic labeling, 38–39 viral growth, 293–295
photoconversion, protocol for, 388–389
point spread function (PSF), 399 Q
polymerase chain reaction (PCR), 111 qPCR, 119
Ponceau S stained tissue sections, 129
positron emission tomography (PET), R
567 rat preparation, for immunogold procedure,
postembedding immunogold labeling, steps 55
in, 95 Rayleigh’s criterion, 399
postembedding procedures, sequence steps real-time RT-PCR techniques
in, 88–100 additional methods for, 170–171
dehydration and embedding, 93–95 advantages of, 157, 165–166
fixation, 90, 92–93 applications of, in single-cell mRNA
immunoincubation, 95–100 expression
postmortem interval (PMI), 112 of phenotypic markers, 160–161
postmortem silver techniques, use of, 3 and patch-clamp recording, 161–162
INDEX 689
patch-clamp recording, and fluorescent RNase inhibitor, 148
calcium measurements in young and RNase-free precautions, 131
aged rodent barrel cortex, sensory analysis of,
cells, 162–164 442–446
limitations of, 166 excitatory neuronal network, activity of,
quantification and validation experiments, 444
157–160 neocortical processing, first level of, 442
comparative C T methods, 158 whiskers, behavior of, 444
negative control in, 160
special methods, 169–170 S
standard curve method, 158 scRT-PCR, 143–144
receptor PET (rPET), 567 advantages of, 165–166
regional microdissection methods, 114–115 limitations of, 166
restriction enzymes, 170 probe design for, 152
retrograde axonal transport, 338 SELDI-TOF MS approach, 127, 135
retrograde labeling, of neuron, preparations selective vulnerability, 110
in silver stain, 117
3D light microscopy reconstructions, silver staining method of axons, 1
232–234 as neurotoxic assessment tool, 2
for cutting and pretreatment of sections, postmortem silver techniques, use of, 3
228–229 silver-degeneration methods, 338
for electrophysiology, 227–228 single cell harvesting, for molecular analysis,
embedding, for electron microscopy, 232 146–152
fluorescent signal conversion, to DAB, 231 preparations of, 148–149
for labeling, 228 PCR-dedicated equipments for, 148
for perfusion, 228 single photon emission computed
neurochemical identification, of tomography (SPECT), 567
biocytin-filled neuron, 230 single-cell microaspiration methods,
prior to electrophysiological recordings, 115–117
227 single-cell reverse-transcription/polymerase
staining, 231–232 chain reaction, see scRT-PCR
visualization, of biocytin-filled neuron and single-cell RNA amplification, 116
digital photography, 229–230 spike sorting, 658–659
retrograde tract-tracing spinal motor neurons, 119
BDA, 22–23 standard buffers, recipes for, 54–55
disadvantages over FG, 348 standard reference brain, 583
FluoroGold, 23–24 average brain, 585–586
pseudorabies virus (PRV), 24–30 individual reference brain, 584–585
reverse transcription preparation, method stereological methods, 84
of, 167 sterile phosphate buffer (PB), 351
rhodamine dextran amine (RDA), 353 Stokes shift, 417
RNA amplification, 111–112, 115–116 substance P- (SP-ir), 353
aRNA, 121–123, 134 substrate diaminobenzidine (DAB;Sigma)
for single-cell PCR, 145–146 surface-based atlases and flat maps, 587–
terminal continuation, 123–125, 134– 588
135 SYBER r Green, 155, 157
RNA assessment, 113 synapse, definition of, 509–510
RNA extraction, applications of, synaptic boutons, 502
cDNA arrayanalysis, 114 synaptic cleft, 510
differential display, 114 synaptic junction, 339
quantitative real-time PCR (qPCR), 114 Szentágothai, models of, 198
serial analysis of gene expression (SAGE),
114 T
RNA quality, 113 T7 RNA polymerase, 145
RNA–RNA hybrid, 240 Taqmanr , 153, 155, 157, 160
690 INDEX
TEM, see transmission electron microscopy anterograde tract-tracing, 18–22

terminal boutons, 338 retrograde tract-tracing, see retrograde
terminal continuation (TC) RNA tract-tracing
amplification, 111, 123–125 identification of neurochemical
tetramethylbenzidine (TMB), 353 phenotype, 30–31
theory of Pathoklise, 3 triple labeling studies, 31–37
thermostable DNA polymerase, 144 future prospects of, 54
thionin, 116 general application and appraisal of,
three dimensional computerized 8–9
reconstruction limitations of, 7, 52–54
advantages and limitations of, 554–556 principles of methods
coordinate systems, 543–544 animals, 37–38
application, for rat pontine nuclei, immunocytochemistry, 41–46
556–560 intracardial perfusion, 39–41
data acquisition, 536, 538–541 phenotypic labeling, 38–39
using image-combining computerized tissue preparation, 46–47, 63, 100
microscopy, 536–537 tissue sampling, 48–51
using image analysis techniques, 537 tract-tracing, 38
data alignment ultrastructural immunocytochemistry of
fiducial markers and block-face dual labeling procedures, 16–17
imaging, 541–542 general issues of preembedding
landmarks and boundaries, 542 immunocytochemistry, 9–11
template sections, atlases and preembedding immunogold–silver
tomography data, 542 labeling, 12–16, 61
data translation, of local and global preembedding immunoperoxidase
coordinates, 561–562 labeling, 11–12, 60
example data, 532–533 transneuronal labeling, 381
preparatory steps in transport vesicles, 338
fiducial markers and block-face triple channel confocal laser scanning,
imaging, 536 methodology of
section plane orientation, 534–535 advantages of, 427–428
shrinkage extimation, 536 anterograde neuroanatomical tracing,
tissue block imaging, 534 409
tissue block size, 534 confocal instrument operation, 419–420
visualization and analyses, 544–545 controls
density gradient analysis, 552–553 immunofluorescence specificity
primary viewing, 546–548 controls, 410
slicing, 548 confocal instrument controls, 410–411
spatial orientation analysis, 549–551 crosstalk
stereo-imaging, 553 determination of emission, 416–417
surface modeling of labeling patterns, determination of excitation, 417–419
548–549 result authenticity of, 419
whole brain atlas systems, 553–554 types of, 413–416
tissue preparation, for transmission electron disadvantages of, 427–428
microscopy fluorochromes
counterstaining, 47 bleaching of fluorescence signal,
dehydration, 47 412–413
osmication, 46–47 emission spectrum, 412
Titan One Tube RT-PCR kit, 241 excitation spectrum, 411–412
tract-tracing method, 1, 4 image processing and 3D reconstruction
with discrete cell microdissection, deconvolution, 420–423
119–121, 133–134 image mismatch correction, 423–424
transmission electron microscopy method, 7 multichannel 3D reconstruction,
advantages of, 51–52 424–425
analysis of synaptic connections, 17 result interpretation, 425–426
INDEX 691
triple immunofluorescence protocol, 408 viral transneuronal tracing
triple-labeling studies, 31–37 Bartha PRV, see Bartha PRV, as
Triton X-100, 378 neuranatomical tracer
triton, 97 historical background, 267–268
tyrosine hydroxylase (TH) practical considerations and pitfalls, see
immunofluorescence, 341 PRV neuroanatomical tracing, pros
tyrosine hydroxylase, 160 and cons
visualization methods, 353–356
U voltage-sensitive dye
ultrastructural visualization, 340 commercial sources of, 446
imaging of, 447–448
V imaging technique, 438
vacuuming method, 150 advantages and limitations of, 446
Vector r very intense purple (V-VIP), application of, 2
354 staining neocortex with, 447
vesicular glutamate transporters, 160
vibratome sectioning, 375 W
Vibratome r , 186, 229 WGA-HRP tracing, 339–340, 353
viral transneuronal tracing studies, viruses wheat germ agglutinin (WGA), 339
used for –horseradish peroxidase (WGA-HRP), 265
HSV1, 290–291 whole-cell patch clamp recording, 438–442,
perspectives of, 292 448
rabies, 291–292 advantages and limitations of, 446
perspectives of, 292 Wizardr DNA preparations, 170

0 387 28942 9

Uploaded by

Copyright:

Available Formats

0 387 28942 9

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

0 387 28942 9

Uploaded by

Copyright:

Available Formats

Neuroanatomical

José Luis Lanciego

Cover illustration: Reconstruction of functional connectivity between neurons based on the

Library of Congress Control Number: 2005932857

Printed in the United States of America. (TB/KYO)

Between the ﬁrst edition of Neuroanatomical Tract-Tracing Methods in 1981

development of new ﬂuorescence probes, single- and multiphoton confo-

It has been an honor as well as a pleasure to select contributors for the

2. Preembedding Immunoelectron Microscopy: Applications for

3. Postembedding Immunogold Cytochemistry of Membrane

4. Cell and Tissue Microdissection in Combination with Genomic

5. Molecules and Membrane Activity: Single-Cell RT-PCR and

6. Merging Structure and Function: Combination of In Vivo

7. Juxtacellular Labeling of Individual Neurons In Vivo:

8. Nonradioactive In Situ Hybridization in Combination with

9. Viral Tracers for the Analysis of Neural Circuits . . . . . . . . . . . 263

10. Dextran Amines: Versatile Tools for Anterograde and Retrograde

11. Multiple Neuroanatomical Tract-Tracing: Approaches for

12. Tract-Tracing in Developing Systems and in Postmortem Human

13. Combined Fluorescence Methods to Determine Synapses in

14. Advances in Understanding Cortical Function through

15. From Dendrites to Networks: Optically Probing the Living Brain

16. Stereology of Neural Connections: An Overview . . . . . . . . . . . 477

17. Three-Dimensional Computerized Reconstruction from Serial

19. Neuron and Network Modeling . . . . . . . . . . . . . . . . . . . . . . 604

20. Functional Connectivity of the Brain: Reconstruction from Static

LENNART HEIMER • Departments of Neurosurgery and Neuroscience, University of

most easily explained by reference to his unfailing persistence and ability to

brain; neurosurgeons and neuroradiologists, in particular, are promoting

SUSAN R. SESACK, LEEANN H. MINER, AND NATALIA OMELCHENKO • De-

A. General Applications and Appraisal of the Methods

The higher magniﬁcation and resolution afforded by the TEM method of

B. Ultrastructural Immunolocalization of Neurobiological Proteins

The functional impact of any given neurotransmitter is determined by

1. General Issues of Pre- and Postembedding

The main techniques used for immunologically based electron-dense

preembedding (Yi et al., 2001) and postembedding (see chapter by Mathisen

2. Preembedding Immunoperoxidase Labeling

The preembedding immunoperoxidase technique involves localizing

a. Strengths and Weaknesses

The major beneﬁt of the immunoperoxidase technique is its superior

Another potential drawback of immunoperoxidase staining is that some

b. Applications and Analysis

The immunoperoxidase technique is well suited for qualitative analysis

3. Preembedding Immunogold–Silver Labeling

Immunogold labeling exploits the conjugation of colloidal gold particles

a. Strengths and Weaknesses

In contrast to the precipitates formed in immunoperoxidase staining,

Another possible weakness is that preembedding immunogold may be bet-

b. Applications and Analysis

The ultrastructural immunolocalization of neurobiological proteins is a

For analysis of tissue in which the subcellular localization of gold–silver

A major strength of the immunocytochemical procedures discussed here

C. Analyses of Synaptic Connections

In order to determine whether neurons of interest are synaptically con-

The induction of anterograde degeneration following brain lesion is the

Figure 2.4. Electron micrographic examples of anterograde tract-tracing in the rat

There are several disadvantages to the anterograde degeneration method:

1. The correct survival time after lesion is critically important. At short