Intravital 2-photon imaging reveals distinct

morphology and infiltrative properties of glioblastoma-

associated macrophages
Zhihong Chena,b,1, James L. Rossa,c,1, and Dolores Hambardzumyana,b,2
a
Department of Pediatrics, Aflac Cancer and Blood Disorders Center, Children’s Healthcare of Atlanta, Emory University School of Medicine, Atlanta, GA
30322; bWinship Cancer Institute, Emory University School of Medicine, Atlanta, GA 30322; and cCancer Biology Graduate Program, Graduate Division of
Biological and Biomedical Sciences, Emory University, Atlanta, GA 30322

Edited by Lawrence Steinman, Stanford University School of Medicine, Stanford, CA, and approved June 4, 2019 (received for review February 14, 2019)

Characterized by a dismal survival rate and limited response to gliomas to a degree that varies by glioma subtype and tumor com-
therapy, glioblastoma (GBM) remains one of the most aggressive partment (6). BMDMs do not universally conform to the pheno-
human malignancies. Recent studies of the role of tumor- type of microglia, but preferentially express immunosuppressive
associated macrophages (TAMs) in the progression of GBMs have cytokines and show an altered metabolism. These findings, to-
demonstrated that TAMs are significant contributors to tumor gether with ours, argue against traditional therapeutic strategies
growth, invasion, and therapeutic resistance. TAMs, which include that target TAMs indiscriminately and instead favor strategies that
brain-resident microglia and circulating bone marrow derived- specifically target immunosuppressive BMDMs.
monocytes (BMDMs), are typically grouped together in histopath- Monocytes develop from hematopoietic stem cell progenitors
ological and molecular analyses due to the lack of reliable markers and remain in the intact vasculature of the brain, with their in-
of distinction. To develop more effective therapies aimed at filtration into surrounding tissue hindered by the blood brain
specific TAM populations, we must first understand how these barrier (2). However, during GBM tumor progression, increased
cells differ both morphologically and behaviorally. Furthermore, expression of monocyte chemoattractant proteins (MCPs) (7) in
we must develop a deeper understanding of the mechanisms combination with the disruption in blood brain barrier integrity re-
encouraging their infiltration and how these mechanisms can be sults in infiltration of inflammatory monocytes into the perivascular
therapeutically exploited. In this study, we combined immuno- niche (8). Once inside the tissue, infiltrating monocytes can
competent lineage tracing mouse models of GBM with high-
then differentiate into macrophages and exert protumor effects
resolution open-skull 2-photon microscopy to investigate the
(9). We have previously demonstrated that a genetic decrease
phenotypical and functional characteristics of TAMs. We demon-
in the CCL2/CCR2 axis in GBMs reduces monocyte infiltration
strate that TAMs are composed of 2 morphologically distinct cell
but is insufficient to completely ablate it, suggesting that other
types that have differential migratory propensities. We show that
mechanisms are responsible for the recruitment of these cells
BMDMs are smaller, minimally branched cells that are highly
as well (5).
migratory compared with microglia, which are larger, highly
branched stationary cells. In addition, 2 populations of monocytic
macrophages were observed that differed in terms of CX3CR1 Significance
expression and migratory capacity. Finally, we demonstrate the
efficacy of anti-vascular endothelial growth factor A blockade for Therapeutic strategies targeting glioblastoma (GBM) tumor-
prohibiting TAM infiltration, especially against BMDMs. Taken associated macrophages (TAMs) have shown limited efficacy.
together, our data clearly define characteristics of individual TAM One reason for this limited effectiveness is our poor un-
populations and suggest that combination therapy with antivas- derstanding of the fundamental characteristics driving the
cular and antichemotaxis therapy may be an attractive option for behavior of these cells. Here we report a combination of im-
treating these tumors. munocompetent lineage tracing mouse models of GBM with an
open skull window for in vivo observation of TAM constituents
glioblastoma | macrophage | two-photon | monocyte | microglia via 2-photon microscopy. We show that TAMs are composed of
2 morphologically and behaviorally distinct cell types: brain-
resident microglia and bone marrow-derived macrophages
D espite aggressive therapeutic intervention, glioblastoma (GBM;
World Health Organization grade IV) remains one of the
deadliest human cancers, with a median survival of approxi-
(BMDMs). After treatment with anti-vascular endothelial
growth factor A antibody, we observed significant reductions
in BMDM infiltration and morphological responses. These
mately 15 mo (1). Current therapeutic regimens target properties
studies provide a platform for monitoring the multidimen-
of neoplastic cells but fail to target prevalent nonneoplastic cell
sional responses of specific cell populations to therapeutic in-
populations of the tumor, including tumor-associated macrophages
tervention in real time.
(TAMs). TAMs symbiotically interact with tumor cells through the
secretion of growth factors, matrix metalloproteinases, and other Author contributions: Z.C., J.L.R., and D.H. designed research; Z.C., J.L.R., and D.H. per-
factors that promote tumor cell invasion and progression, thus formed research; Z.C., J.L.R., and D.H. contributed new reagents/analytic tools; Z.C., J.L.R.,
rendering them attractive targets for therapeutic intervention (2, 3). and D.H. analyzed data; D.H. supervised the study; and Z.C., J.L.R., and D.H. wrote the
paper.
We have previously demonstrated that TAMs, which consist of
resident-brain microglia and bone-marrow derived monocytes The authors declare no conflict of interest.

(BMDMs), constitute up to 40% of the total tumor mass in both This article is a PNAS Direct Submission.

human and mouse GBMs (2, 4). Furthermore, we have demon- Published under the PNAS license.
strated in a PDGF-B–driven genetically engineered mouse model 1
Z.C. and J.L.R. contributed equally to this work.
(GEMM) of GBM that >80% of these cells are BMDMs that in- 2
To whom correspondence may be addressed. Email: [email protected].
filtrate from the vasculature (5). A recent elegant study using This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
single-cell profiling of human GBMs confirmed our data and 1073/pnas.1902366116/-/DCSupplemental.
showed that monocytes significantly infiltrate pretreatment human Published online June 24, 2019.

14254–14259 | PNAS | July 9, 2019 | vol. 116 | no. 28 www.pnas.org/cgi/doi/10.1073/pnas.1902366116

 Unlike BMDMs, microglia are always present in the central
nervous system and are homogeneously dispersed throughout
the brain. These long-lived cells, which develop from embryonic
yolk sac progenitor cells during embryogenesis, are responsible
for the clearing of pathogens and maintenance of synaptic
processes (10, 11). In their activated state, microglia produce
proinflammatory cytokines and reactive oxygen species, assume
an amoeboid morphology, and become histologically indistin-
guishable from infiltrating BMDMs (12, 13). In contrast to
BMDMs, which are found in tumor bulk and infiltrating tumor
margins, microglia are sparsely distributed throughout the tumor
and accumulate at the periphery. Analysis of freshly dissociated
patient GBM samples for BMDMs and resident microglia
revealed significant infiltration of BMDMs across the GBM
subtypes. Only the mesenchymal subtype of GBM showed in-
creased microglia infiltration compared with BMDMs (14).
Two-photon microscopy has previously been used for the
Fig. 1. The 2-photon imaging setup and acquisition procedures. (A) A cus-
longitudinal analysis of GBM growth kinetics, vascular dynamics, tom acrylic adapter attaches the mouse to a steel stage to stabilize the im-
and immune cell infiltration (15–18). However, the contributions aging plane and reduce movement artifacts. (B) Superficial tumors and
of individual TAM populations have largely been ignored, be- vascular landmarks can be observed by eye and at low magnification. (C) All
cause microglia and circulating monocytes are typically grouped image analysis was performed at a depth of >50 μm below the pial surface
together and treated as a single collective entity. Although we and down to ∼500 μm deep. Blue indicates DAPI stain. S, surface; T, tumor;
have a firm understanding of TAMs in the context of tumor NB, normal brain. (D) High-resolution Z stacks were imported into Imaris for
growth promotion and immune suppression, the majority of image processing and analysis. TRITC, vessels; GFP, TAMs.
published histopathological studies are limited to static snap-

NEUROSCIENCE
shots of the disease over time. If we seek to identify therapeutic
points of intervention, we must understand how these cell types Although not performed in this study, this methodology also al-
interact with their microenvironment in real time. Furthermore, lows for the visualization of other cell types and cellular components
the development of cell type-specific targeted therapies will be tagged with fluorescent molecules, such as tumor cells, lympho-
possible not only by studying the TAM population as a whole, cytes, microtubules, and others (18, 20, 21).
but also by identifying biological differences between the two
aforementioned subtypes. Visualization of TAM Populations in Chimeric Mice. Using a
In the present study, we performed live in vivo 2-photon mi- Cx3cr1GFP/WT;Ccr2RFP/WT mouse model, we previously demon-
croscopy in an immunocompetent GEMM of GBM together strated that infiltrating peripheral monocytes show decreased
with myeloid cell-specific reporter mice to lineage-trace micro- CCR2-RFP expression as they mature in the GBM microenvi-
glia and BMDMs. By taking advantage of reciprocal bone mar- ronment. Since we are interested in quantifying both newly in-
row chimeras between Cx3cr1GFP/WT and wild-type (WT) mice, filtrating monocytes and differentiated macrophages, this model
we demonstrated the feasibility of using an open-skull window is not optimal for our studies, as we would inevitably miss a large
for the imaging of TAM subpopulations in tumor tissue. We population of cells as they mature. To remedy this, we used re-
observed that tumor microglia and infiltrating BMDMs are mor- ciprocal bone marrow chimera mouse models with 1 allele of the
phologically distinct cell populations with differing migratory ca- Cx3cr1 gene replaced with GFP (Fig. 2A). These models allow us
pacities. We further showed that anti-vascular endothelial growth to discern differences between infiltrating BMDMs and brain-
factor A (VEGFA antibody) treatment significantly decreases resident microglia. CX3CR1 is the sole receptor for the che-
infiltration and induces a morphological switch in BMDMs to mokine CX3CL1 (fractalkine), which is widely expressed
resemble differentiated macrophages. This study also demon- throughout the central nervous system. Replacement of 1 allele
strates that blood-brain barrier (BBB) integrity is not the sole with GFP (Cx3cr1GFP/WT) has been demonstrated to have no
driver of monocyte infiltration and provides a rationale for com- effect on the migration capacity or infiltration of TAMs (22, 23).
bining antiangiogenic and antichemotaxis (targeting members of To generate mice with GFP-expressing BMDMs and WT
the MCP family) therapies to block monocyte infiltration. microglia (i.e., no GFP expression in microglia), bone marrow
from Cx3cr1GFP/WT mice was mixed in a 1:1 mixture with bone
Results marrow from Cx3cr1WT/WT mice. This mixture was then injected
Two-Photon Imaging Permits Direct, Longitudinal Observation of into whole-body–irradiated mice to reconstitute the bone mar-
TAMs In Vivo. Advantages of 2-photon microscopy over tradi- row with 50% GFP-expressing BMDMs and 50% WT BMDMs
tional confocal microscopy include reduced autofluorescence (SI Appendix, Fig. S2). This scheme allows for single-cell reso-
and photobleaching effects, increased imaging depth, and mini- lution of infiltrating BMDMs, whereas a nondiluted GFP signal
mal photodamage to surrounding brain tissue (19). Conse- poses difficulties for resolving individual cells due to the high
quently, we used 2-photon microscopy for the in vivo analysis of density of these cells in the tumor. Non–tumor-bearing mice had
individual TAM populations. To minimize breathing artifacts few to no GFP-expressing BMDMs in the brain parenchyma, as
during imaging, a custom acrylic adapter that attaches to the these cells do not readily cross the BBB (Fig. 2B and SI Ap-
cranium after skull window placement was manufactured in- pendix, Fig. S3 A–D); however, imaging of tumor-bearing mice
house. This adapter attaches to a stainless-steel stage that holds revealed a significant infiltration of GFP-expressing BMDMs
the mouse in place throughout image acquisition (Fig. 1A). Su- (Fig. 2 B–D). These cells were seen to widely populate tumor
perficial tumors are easily visualized through the skull window and bulk, perinecrotic areas, and infiltrative tumor margins, while
can be distinguished from adjacent normal brain tissue by eye or at their presence was absent in adjacent nontumor tissue (SI Ap-
low magnification with a stereoscope (Fig. 1B and SI Appendix, pendix, Fig. S4B).
Fig. S1). To avoid visual artifacts induced from skull window To generate mice with WT BMDMs and GFP-expressing microglia,
placement, such as glial scarring and leukocyte accumulation, all donor bone marrow isolated from Cx3cr1WT/WT mice was injected into
image stacks were analyzed at a depth of 50 μm or deeper from whole-body–irradiated Cx3cr1GFP/WT recipients. Microglia popu-
the pial surface, typically between 100 and 500 μm (Fig. 1 C and late the brain during embryogenesis and are found consistently
D). High-resolution image stacks with approximate dimensions of dispersed throughout the tissue (24). In tumor-bearing brains,
400 μm × 400 μm × 400 μm were typically acquired and analyzed. microglia are sparse in tumor bulk but often accumulate in

Chen et al. PNAS | July 9, 2019 | vol. 116 | no. 28 | 14255

Fig. 2. Reciprocal bone marrow chimera allows for the detection of individual TAM populations. (A) Bone marrow chimera schematic depicting the gen-
eration of mice with GFP-labeled BMDMs or GFP-labeled microglia. (B) Raw Z stack images acquired from 2-photon imaging of naïve and tumor-bearing mice
with GFP-BMDMs (Top) and GFP-microglia (Bottom). (Scale bar: 100 μm; magnified inset scale: 30 μm.) (C) GFP-labeled TAM populations were quantified in
Imaris, with the number of cells per mm3 calculated. Each dot represents an individual animal. MO, monocyte; MG, microglia. **P < 0.01; ****P < 0.0001, 1-
way ANOVA. (D) 10× Z-stitch immunofluorescence image of a tumor-bearing Cx3cr1GFP/WT→Cx3cr1WT/WT mouse. Note the presence of GFP signal only in
tumor tissue. GFP, monocytes; DAPI, nuclei. (E) Cx3cr1WT/WT→Cx3cr1GFP/WT chimera in a tumor-bearing mouse. Microglia are unevenly distributed in the tumor
bulk and accumulate at the tumor margins in distinct clusters (white arrows). GFP, microglia; DAPI, nuclei.

clusters at the periphery of tumor margins (Fig. 2 B and E and SI BMDMs were observed to be stationary (0.003478 μm/s),
Appendix, Figs. S3B and S4A). As previously demonstrated by our reflecting the phenotype of differentiated macrophages (Fig. 4 B
laboratory (5), we did not observe an increase in the microglial and C and Movies S1 and S2).
population in tumor-bearing brains but did find a significant in- Microglia were found to be stationary; however, their processes
crease in the BMDM population. We also observed a slight in- were continuously extending and retracting in both tumor and
crease in GFP intensity in tumor-associated microglia compared nontumor brains (Fig. 4D). These results reinforce previous find-
with nontumor microglia (SI Appendix, Fig. S3F). In addition, we ings that even in the absence of a stimulus, these cells sense their
found a decrease in microglial branching with increases in surface microenvironment through the continuous extension and retraction
area and volume in tumor-bearing mice, depicting the accepted of ramified projections (27, 28). Although no instances of micro-
notion of microglia adopting an amoeboid morphology in patho- glial cell migration were observed, phagocytic behavior was seen in
logical conditions as they phagocytose dead or damaged cells (SI 1 imaging volume (Movies S3 and S4). The complex branching
Appendix, Fig. S5) (12, 13, 25, 26). phenotype coupled with the sensing behavior is likely how these
cells detect pathogenic insults in the local microenvironment (27).
TAMs Dynamically Alter Their Phenotype in the Tumor Microenvironment.
After image acquisition, XYZ stacks were imported into Imaris
for creation of 3D representations (Fig. 3A). Images acquired via
2-photon microscopy were highly amenable for the detection and
quantification of morphology, whereas traditional confocal mi-
croscopy was unable to resolve the fine microglial processes due to
GFP signal loss and reduced Z stack depth (SI Appendix, Fig. S4 C
and D). After image processing and quality control measures,
morphology analysis was performed on both BMDMs and
microglia (SI Appendix, Fig. S6). Microglia were found to be larger
in mean surface area (5,516 ± 267.3 μm2) and volume (4,729 ±
258.5 μm3) compared with BMDMs (970.5 ± 24.32 μm2 and
2,154 ± 94.24 μm3 respectively) (Fig. 3 B and C). Microglia also
had significantly more primary branches per cell (6 vs. 1), causing
them to assume a less spherical morphology with numerous ter-
minal points (28 vs. 1) (Fig. 3 D and E).
Time-lapse images were analyzed to determine migratory
differences between the two cell types. BMDMs were found to
Fig. 3. 3D morphology analysis of TAM populations. (A) 3D renderings of
be migratory, consisting of two phenotypically distinct populations GFP-microglia (Top) and GFP-BMDMs (Bottom) image stacks using Imaris.
(Fig. 4A). There were low–GFP-expressing BMDMs that were (Scale bar: 50 μm; magnified inset scale: 20 μm.) (B) Single-cell morphology
highly mobile (0.07826 μm/s), reflecting the phenotype of newly statistics for surface area. ****P < 0.0001, 2-tailed t test. (C) Cell volume. P <
infiltrating inflammatory monocytes (Fig. 4B) (5). These cells had 0.0001, 2-tailed t test. (D) Number of primary branches for each cell. P <
an average directional track “straightness” of 0.45 (0, nonstraight; 0.0001, Mann–Whitney U test. (E) Terminal points per cell. P < 0.0001,
1, perfectly straight), and were visually observed to migrate Mann–Whitney U test. Tumor MO, 543 cells from 7 mice; tumor MG, 123 cells
through the tissue in a nonspecific fashion. High-GFP-expressing from 4 mice.

14256 | www.pnas.org/cgi/doi/10.1073/pnas.1902366116 Chen et al.

Anti-VEGFA Treatment Reduces Monocyte Infiltration. Given their of circulating BMDMs and induces the differentiation of existing
localization and protumor behavior, TAMs have been suggested monocytes into tissue macrophages.
to promote resistance to antiangiogenic agents, such as bevacizumab,
but the underlying mechanisms remain unknown (29–31). Evidence Discussion
suggests that resistance may be due to changes in the tumor Previous morphological studies of microglia in a pathogenic state
microenvironment and the presence of myeloid-derived TAMs have been informative, yet future studies may benefit from larger
(32, 33). Mechanisms of TAM infiltration may include passive imaging dimensions to reduce the loss of cellular information
migration due to a leaky BBB or active migration via chemokine (37–40). In addition, studies of BMDM morphology in the
gradients; however, which of these mechanisms is most re- context of GBM are lacking. The 2D immunofluorescence im-
sponsible remains unclear. To address this, we treated mice with ages acquired from fixed slices 40 μm and thinner inevitably miss
a neutralizing anti-VEGFA antibody (B20-4.1.1) and performed the numerous 3D structures of microglia. Our study included 3D
2-photon imaging (Fig. 5A). We previously found that B20-4.1.1 image stacks up to 500 μm in vertical depth to ensure that cell
reduced tumor vessel area and vascular leakiness in a PDGF-B– bodies and all of their many branches were included in the
driven mouse model (34); therefore, it can be inferred that any analyses. Our methodological approach closely resembles re-
reduction observed in the infiltration of TAMs is due in part to cently published findings demonstrating that tumor microglia
normalized vasculature (20, 35, 36). We found increased post- respond to therapeutic treatment via morphological changes
injection survival in tumor-bearing mice treated with B20-4.1.1 (40); however, contrary to our findings that BMDMs are the
compared with vehicle-treated mice (42.5 d vs 30 d) (Fig. 5B). In predominate TAM subtype in immunocompetent mice, the au-
addition, the number of infiltrating monocytes/macrophages was thors of that study reported greater infiltration of microglia in
significantly lower in the B20-4.1.1–treated mice, as observed on patient-derived xenograft (PDX) models using immunocom-
2-photon imaging (8,619 cells/mm3 vs. 21,317 cells/mm3) (Fig. 5 promised nod-scid mice. Further studies are needed to evaluate
C and D). A morphological switch also occurred in the B20- whether the differences observed in monocyte infiltration in
4.1.1–treated mice. Compared with vehicle-treated mice, the PDX models compared with GEMMs of GBM are a result of the
cells assumed a more highly branched phenotype (2.46 vs. 0.99 immunocompromised status of nod-scid mice or due to species
branches per cell) and significantly increased total cell volume incompatibility between human chemokines and mouse chemo-

NEUROSCIENCE
(2,661.96 ± 72.82 vs. 2,154 ± 94.24 μm3) and surface area kine receptors (41).
(1,503 ± 41.78 vs. 970.5 ± 24.41 μm2), causing their sphericity to Our findings demonstrate that microglia are large, branched
decrease (0.668 ± 0.005 vs. 0.734 ± 0.004) (Fig. 5 E–J). Inter- cells with highly active processes that extend and retract in a
estingly, we did not observe any significant difference in cell sensing manner in both naïve and tumor contexts. These results
motility or cell directionality with B20-4.1.1 treatment (Fig. 5 K confirm our previous understanding that even in the absence of a
and L). This treatment also appeared to normalize the tumor stimulus, microglia are constantly surveying their microenviron-
vasculature; however, due to the leaky nature of our vehicle-treated ment through the continuous extension and retraction of rami-
tumors, the TRITC signal was either too weak or too diffuse to fied projections (27, 28). We did not observe any instances of
enable a reliable quantitative vascular comparison. These changes microglial cell body migration, in line with a previous finding of
suggest that B20-4.1.1 treatment partially prevents the infiltration limited movement of microglial soma (42). Conversely, we found
that infiltrating BMDMs are minimally branched and highly
migratory. As cells up-regulate their expression of CX3CR1 as
indicated by GFP intensity, their migration capacity decreases.
These results imply that BMDM cell migration is a function of
the state of differentiation or morphological phenotype.
Recent studies have demonstrated that anti-VEGFA therapy
causes increased infiltration of monocytic macrophages into the
tumor in recurrent GBM using a xenograft mouse model (43).
Evidence also suggests that in response to anti-VEGFA block-
ade, infiltrating myeloid cell populations initially decrease, but
then rapidly increase as the tumor progresses (33). Although we
found a decrease in this population via 2-photon imaging, our
model captured the biology of primary tumors, not recurrent
tumors, and was performed in immunocompetent mice, which
may explain discrepancies between our findings and those
reported by others. Vascular normalization has been attributed
to anti-VEGFA therapy, which may explain the early decrease in
myeloid cell populations (29, 30, 34); however, as the tumor
grows over time, hypoxia increases, providing infiltrating cells
with the chemokines and stimulating factors necessary for their
recruitment (44–46). In our study, we observed a decrease in the
CX3CR1+ myeloid population after B20-4.1.1 treatment. In
addition, in response to treatment, monocytes assumed a more
branched phenotype, suggesting that VEGFA sequestration
promotes differentiation of these cells. As TAMs have been
shown to promote tumor vascularization, a possible scenario
Fig. 4. TAM migration analysis in time-lapse images. (A) Cell tracking using
arises: do BMDMs present during anti-VEGFA therapy become
the Imaris spots function. Spheres indicate individual cell position at each
time point and are colored according to GFP intensity. Movement tracks are
trapped in the tumor and promote vascularization over time (47,
colored by distance traveled by each cell. (B) Speed of individual cells with
48)? Although not performed in this study, longitudinal imaging
low (22 cells from 7 mice) and high (22 cells from 3 mice) GFP intensity. over the course of anti-VEGFA treatment could answer this
****P < 0.0001, 2-tailed t test. (C) Individual cell tracking of a low–GFP- question and may identify mechanisms of resistance that other-
expressing monocyte (white circle) compared with high–GFP-expressing wise would go unobserved using other methods.
cells (purple circles). The dashed line indicates the path of the cell over 75 In conjunction with our previous study demonstrating that ge-
min. (Scale bar: 30 μm.) (D) Time-lapse imaging of a stationary tumor netic loss of CCL2/CCR2 partially decreases monocyte in-
microglia over 180 min. Arrows indicate extension and retraction of indi- filtration, our present finding that B20-4.1.1 also partially reduces
vidual cell processes. (Scale bar: 30 μm.) this population demonstrates that monocyte infiltration is not due

Chen et al. PNAS | July 9, 2019 | vol. 116 | no. 28 | 14257

Fig. 5. Anti-VEGFA treatment reduces TAM infiltration and induces a morphological response. (A) Timeline of treatment schedule and imaging of anti-
VEGFA–treated mice. DPI, days postinjection. (B) Kaplan–Meier survival curves for vehicle-treated mice (n = 9; median survival, 30 d) and anti-VEGFA–treated
mice (n = 10; median survival, 42.5 d). P = 0.0436. ms, median survival. (C) Quantification of cell populations acquired from individual imaging volumes. *P <
0.05. (D) Representative image stacks acquired during 2-photon imaging of mice with GFP-labeled BMDMs for vehicle-treated and anti-VEGFA–treated
animals. (Scale bar: 100 μm; magnified inset scale: 30 μm.) (E) Single-cell morphology statistics for surface area. ****P < 0.0001, 2-tailed t test. (F) Cell volume.
P < 0.0001, 2-tailed t test. (G) Sphericity (1.0, perfect sphere; 0, not spherical). P < 0.0001, 2-tailed t test. (H) Number of primary branches for each cell. P <
0.0001, Mann–Whitney U test. (I) Number of secondary branches for each cell. P < 0.0001, Mann–Whitney U test. (J) Total number of branches for each cell. P <
0.0001, Mann–Whitney U test. Vehicle MO, 543 cells from 7 mice; anti-VEGFA MO, 717 cells from 7 mice. (K) Speed of individual cells, 2-tailed t test. (L) Cell
migration track straightness (1.0, straight line), 2-tailed t test. Vehicle, 26 cells from 5 mice; anti-VEGFA, 19 cells from 5 mice.

solely to a compromised BBB or chemokine gradients (5). Taken transplantation using the following coordinates: bregma AP (anterior/posterior),
together, our data provide a clear rationale for combining anti- 0.2 mm; ML (medial/lateral), 1.0 mm, over which a burr hole was made with
VEGF therapy with chemotaxis blockade, such as anti-CCL2/ a sterile 0.7-mm-diameter drill bit (19007-7; Fine Science Tools). A Hamilton
CCR2, to block BMDM infiltration into GBM. The findings of syringe affixed to a stereotactic apparatus was then lowered to a depth
these studies have far-reaching implications beyond their appli- of 1.0 mm, and 1 μL of cell suspension was injected through a 30-gauge
cations in GBM. Open-skull window 2-photon imaging techniques needle.
coupled with chimeric mouse models could be applied to the study
of stroke and other neurologic disorders in which microglia and Skull Window Placement and 2-Photon Imaging. At approximately 3 wk after
BMDMs have been known to play differential roles (49–52). In tumor transplantation, mice began to show neurologic signs of brain tumor.
addition, deciphering mechanisms of macrophage recruitment to A 5-mm-diameter sterile circular coverslip (Electron Microscopy Sciences) was
pathogenic sites may lead to improved targeted therapies for the carefully placed to facilitate visual access to the tumor bulk and tumor
treatment of these diseases. margin. At the time of imaging, TRITC-dextran (2.5 mg/mL; Sigma-Aldrich)
was injected i.v. to outline the blood vessels. Mice were fixed to a custom-
Materials and Methods built stage to minimize breathing artifact during image acquisition. Im-
Mice. All experimental procedures utilizing live mice were approved by ages were acquired with a 25× water immersion objective (Leica; NA 0.95) on
Emory University’s Institutional Animal Care and Use Committee (Protocol a Leica SP8 confocal microscope equipped with a tunable pulsed chameleon
2003253). Details are provided in SI Appendix. infrared multiphoton laser (Coherent) and 2 high-sensitivity hybrid-PMT
(HyD) detectors (Leica). High-resolution XYZ stack images (1,024 × 1,024
Bone Marrow Chimera. Bone marrow transplants were performed as reported pixels per Z step) were taken with a step size of 2.5 μm, and images of XYZT
previously (5) and described in detail in SI Appendix. stacks were taken every 2.5 min for approximately 3 h on average. At the
end of each imaging session, the mice were euthanized and their brains
Orthotopic Cell Injections. We used the RCAS/tv-a system to generate de novo processed for confocal imaging as described previously (5). More details are
primary murine GBM as reported previously (5, 53) and described in detail in provided in SI Appendix.
SI Appendix. Once tumors were formed in donor transgenic mice, tumor cells
were freshly dissociated, and 2.5 × 104 cells were superficially injected into B20-4.1.1 Treatment. At 15 d after tumor cell injection, mice were injected i.p.
the right frontal striatum of recipient mice at 8 to 10 wk after bone marrow with either anti-VEGFA (B20-4.1.1, 5 mg/kg; Genentech) or vehicle control

14258 | www.pnas.org/cgi/doi/10.1073/pnas.1902366116 Chen et al.

(bacteriostatic 0.9% sodium chloride) every 4 d until terminal neurologic means between 3 or more independent groups. Kaplan–Meier survival
symptoms appeared. analysis was performed using the log-rank test. A P value < 0.05 was con-
sidered to indicate statistical significance.
Image and Data Analysis. High-resolution image stacks were imported into
Imaris version 9.0.2 (Bitplane). Total cell numbers within each XYZ stack were ACKNOWLEDGMENTS. We thank Dr. John Gale for contributing to the
quantified using the spots function. Morphology analysis of BMDMs and design of the imaging stage, Tony Swaine (Cleveland Clinic) for creating the
microglia were done using the surfaces and filaments functions, respectively. prototype, and Dr. Horace Dale for recreating the stage at Emory University.
Additional details are provided in SI Appendix. We also thank Dr. Alex Huang (Case Western Reserve University) for advice
on live imaging and Dave Schumick for the scientific illustrations. The B20-
4.1.1 used in this study was supplied by Genentech. This research project was
Statistical Analysis. Graphs were created using GraphPad Prism 6. Results are supported in part by the Emory University Integrated Cellular Imaging
expressed as mean ± SD. Comparisons between 2 groups were performed Microscopy Core and the Flow Cytometry Core of the Emory + Children’s
using the unpaired 2-tailed t test or nonparametric Mann–Whitney U test, as Pediatric Research Center and by National Institutes of Health Grants 1F31
indicated in the figure legends. One-way ANOVA was used to compare the CA232531 (to J.L.R.) and R21 NS106554 and R01 NS100864 (to D.H.).

1. R. Stupp et al.; European Organisation for Research and Treatment of Cancer Brain 28. M. E. Tremblay, C. Lecours, L. Samson, V. Sánchez-Zafra, A. Sierra, From the Cajal
Tumor and Radiotherapy Groups; National Cancer Institute of Canada Clinical Trials alumni Achúcarro and Río-Hortega to the rediscovery of never-resting microglia.
Group, Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. Front. Neuroanat. 9, 45 (2015).
N. Engl. J. Med. 352, 987–996 (2005). 29. J. Kloepper et al., Ang-2/VEGF bispecific antibody reprograms macrophages and
2. D. Hambardzumyan, D. H. Gutmann, H. Kettenmann, The role of microglia and resident microglia to anti-tumor phenotype and prolongs glioblastoma survival. Proc.
macrophages in glioma maintenance and progression. Nat. Neurosci. 19, 20–27 Natl. Acad. Sci. U.S.A. 113, 4476–4481 (2016).
(2016). 30. A. Scholz et al., Endothelial cell-derived angiopoietin-2 is a therapeutic target in
3. X. Z. Ye et al., Tumor-associated microglia/macrophages enhance the invasion of treatment-naive and bevacizumab-resistant glioblastoma. EMBO Mol. Med. 8, 39–57
glioma stem-like cells via TGF-β1 signaling pathway. J. Immunol. 189, 444–453 (2012). (2016).
4. D. Hambardzumyan, G. Bergers, Glioblastoma: Defining tumor niches. Trends Cancer 31. K. Turkowski et al., VEGF as a modulator of the innate immune response in glio-
1, 252–265 (2015). blastoma. Glia 66, 161–174 (2018).
5. Z. Chen et al., Cellular and molecular identity of tumor-associated macrophages in 32. T. Liu et al., PDGF-mediated mesenchymal transformation renders endothelial re-
sistance to anti-VEGF treatment in glioblastoma. Nat. Commun. 9, 3439 (2018).

NEUROSCIENCE
glioblastoma. Cancer Res. 77, 2266–2278 (2017).
6. S. Müller et al., Single-cell profiling of human gliomas reveals macrophage ontogeny 33. Y. Piao et al., Glioblastoma resistance to anti-VEGF therapy is associated with myeloid
as a basis for regional differences in macrophage activation in the tumor microen- cell infiltration, stem cell accumulation, and a mesenchymal phenotype. Neuro Oncol.
vironment. Genome Biol. 18, 234 (2017). 14, 1379–1392 (2012).
7. M. Massara et al., Neutrophils in gliomas. Front. Immunol. 8, 1349 (2017). 34. K. L. Pitter et al., Corticosteroids compromise survival in glioblastoma. Brain 139,
8. D. F. Quail, J. A. Joyce, The microenvironmental landscape of brain tumors. Cancer 1458–1471 (2016).
Cell 31, 326–341 (2017). 35. E. R. Gerstner et al., VEGF inhibitors in the treatment of cerebral edema in patients
9. D. I. Gabrilovich, S. Ostrand-Rosenberg, V. Bronte, Coordinated regulation of myeloid with brain cancer. Nat. Rev. Clin. Oncol. 6, 229–236 (2009).
cells by tumours. Nat. Rev. Immunol. 12, 253–268 (2012). 36. W. S. Kamoun et al., Edema control by cediranib, a vascular endothelial growth factor
10. F. Ginhoux et al., Fate mapping analysis reveals that adult microglia derive from receptor-targeted kinase inhibitor, prolongs survival despite persistent brain tumor
primitive macrophages. Science 330, 841–845 (2010). growth in mice. J. Clin. Oncol. 27, 2542–2552 (2009).
11. B. Ajami, J. L. Bennett, C. Krieger, W. Tetzlaff, F. M. Rossi, Local self-renewal can 37. S. Heindl et al., Automated morphological analysis of microglia after stroke. Front.
Cell. Neurosci. 12, 106 (2018).
sustain CNS microglia maintenance and function throughout adult life. Nat. Neurosci.
38. C. Kozlowski, R. M. Weimer, An automated method to quantify microglia morphol-
10, 1538–1543 (2007).
ogy and application to monitor activation state longitudinally in vivo. PLoS One 7,
12. H. Kettenmann, U. K. Hanisch, M. Noda, A. Verkhratsky, Physiology of microglia.
e31814 (2012).
Physiol. Rev. 91, 461–553 (2011).
39. B. A. Plog et al., A novel technique for morphometric quantification of subarachnoid
13. C. K. Donat, G. Scott, S. M. Gentleman, M. Sastre, Microglial activation in traumatic
hemorrhage-induced microglia activation. J. Neurosci. Methods 229, 44–52 (2014).
brain injury. Front. Aging Neurosci. 9, 208 (2017).
40. G. Hutter et al., Microglia are effector cells of CD47-SIRPα antiphagocytic axis dis-
14. K. Gabrusiewicz et al., Glioblastoma-infiltrated innate immune cells resemble M0
ruption against glioblastoma. Proc. Natl. Acad. Sci. U.S.A. 116, 997–1006 (2019).
macrophage phenotype. JCI Insight 1, 85841 (2016).
41. Y. Pan et al., Athymic mice reveal a requirement for T-cell-microglia interactions in
15. C. Ricard et al., Phenotypic dynamics of microglial and monocyte-derived cells in
establishing a microenvironment supportive of Nf1 low-grade glioma growth. Genes
glioblastoma-bearing mice. Sci. Rep. 6, 26381 (2016).
Dev. 32, 491–496 (2018).
16. C. Ricard, F. C. Debarbieux, Six-color intravital two-photon imaging of brain tumors
42. J. Meller et al., Integrin-Kindlin3 requirements for microglial motility in vivo are
and their dynamic microenvironment. Front. Cell. Neurosci. 8, 57 (2014).
distinct from those for macrophages. JCI Insight 2, 93002 (2017).
17. C. Ricard et al., Dynamic quantitative intravital imaging of glioblastoma progression
43. B. A. Castro et al., Macrophage migration inhibitory factor downregulation: A novel
reveals a lack of correlation between tumor growth and blood vessel density. PLoS
mechanism of resistance to anti-angiogenic therapy. Oncogene 36, 3749–3759 (2017).
One 8, e72655 (2013).
44. R. Du et al., HIF1alpha induces the recruitment of bone marrow-derived vascular
18. M. Osswald et al., Brain tumour cells interconnect to a functional and resistant net-
modulatory cells to regulate tumor angiogenesis and invasion. Cancer Cell 13, 206–
work. Nature 528, 93–98 (2015). 220 (2008).
19. K. S. Madden, M. L. Zettel, A. K. Majewska, E. B. Brown, Brain tumor imaging: Live 45. F. Shojaei et al., G-CSF-initiated myeloid cell mobilization and angiogenesis mediate
imaging of glioma by two-photon microscopy. Cold Spring Harb. Protoc. 2013, tumor refractoriness to anti-VEGF therapy in mouse models. Proc. Natl. Acad. Sci.
pdb.prot073668 (2013). U.S.A. 106, 6742–6747 (2009).
20. L. von Baumgarten et al., Bevacizumab has differential and dose-dependent effects 46. S. C. Wang, J. H. Hong, C. Hsueh, C. S. Chiang, Tumor-secreted SDF-1 promotes glioma
on glioma blood vessels and tumor cells. Clin. Cancer Res. 17, 6192–6205 (2011). invasiveness and TAM tropism toward hypoxia in a murine astrocytoma model. Lab.
21. G. Moncayo et al., SYK inhibition blocks proliferation and migration of glioma cells Invest. 92, 151–162 (2012).
and modifies the tumor microenvironment. Neuro-oncol. 20, 621–631 (2018). 47. S. Brandenburg et al., Resident microglia rather than peripheral macrophages pro-
22. M. J. Ruitenberg et al., CX3CL1/fractalkine regulates branching and migration of mote vascularization in brain tumors and are source of alternative pro-angiogenic
monocyte-derived cells in the mouse olfactory epithelium. J. Neuroimmunol. 205, 80– factors. Acta Neuropathol. 131, 365–378 (2016).
85 (2008). 48. X. Chen et al., RAGE expression in tumor-associated macrophages promotes angio-
23. X. Feng et al., Loss of CX3CR1 increases accumulation of inflammatory monocytes and genesis in glioma. Cancer Res. 74, 7285–7297 (2014).
promotes gliomagenesis. Oncotarget 6, 15077–15094 (2015). 49. J. Neumann et al., Beware the intruder: Real time observation of infiltrated neutro-
24. H. W. Morrison, J. A. Filosa, A quantitative spatiotemporal analysis of microglia phils and neutrophil-microglia interaction during stroke in vivo. PLoS One 13,
morphology during ischemic stroke and reperfusion. J. Neuroinflammation 10, 4 e0193970 (2018).
(2013). 50. A. E. Cardona et al., Control of microglial neurotoxicity by the fractalkine receptor.
25. C. U. Kloss, M. Bohatschek, G. W. Kreutzberg, G. Raivich, Effect of lipopolysaccharide Nat. Neurosci. 9, 917–924 (2006).
on the morphology and integrin immunoreactivity of ramified microglia in the mouse 51. K. Bhaskar et al., Regulation of tau pathology by the microglial fractalkine receptor.
brain and in cell culture. Exp. Neurol. 168, 32–46 (2001). Neuron 68, 19–31 (2010).
26. N. Stence, M. Waite, M. E. Dailey, Dynamics of microglial activation: A confocal time- 52. K. A. Jones et al., Peripheral immune cells infiltrate into sites of secondary neuro-
lapse analysis in hippocampal slices. Glia 33, 256–266 (2001). degeneration after ischemic stroke. Brain Behav. Immun. 67, 299–307 (2018).
27. A. Nimmerjahn, F. Kirchhoff, F. Helmchen, Resting microglial cells are highly dynamic 53. C. J. Herting et al., Genetic driver mutations define the expression signature and
surveillants of brain parenchyma in vivo. Science 308, 1314–1318 (2005). microenvironmental composition of high-grade gliomas. Glia 65, 1914–1926 (2017).

Chen et al. PNAS | July 9, 2019 | vol. 116 | no. 28 | 14259