Review

Cell fate determination in the

vertebrate retina
Erin A. Bassett1 and Valerie A. Wallace1,2,3
1

Vision Program, Ottawa Hospital Research Institute, 501 Smyth Road, Ottawa, Ont, K1H 8L6, Canada
Department of Ophthalmology, University of Ottawa, 451 Smyth Road, Ottawa, Ont, K1H 8M5, Canada
3
Department of Biochemistry, Microbiology and Immunology, University of Ottawa, 451 Smyth Road, Ottawa,
Ont, K1H 8M5, Canada
2

The vertebrate retina is a well-characterized and tractable model for studying neurogenesis. Retinal neurons
and glia are generated in a conserved sequence from a
pool of multipotent progenitor cells, and numerous cell
fate determinants for the different classes of retinal cell
types have been identified. Here, we summarize several
recent developments in the field that have advanced
understanding of the regulation of multipotentiality
and temporal competence of progenitors. We also
discuss recent insights into the relative influence
of lineage-based versus stochastic modes of cell fate
determination. Enhancing and integrating knowledge
of the molecular and genetic machinery underlying
retinal development is critically important for understanding not only normal developmental mechanisms,
but also therapeutic interventions aimed at restoring
vision loss.
Retinal development: the essentials
The retina is a thin sheet of neural tissue located on the
inner surface of the posterior eye. It functions to convert
light to electrical impulses and transmits this information to visual processing centers in the brain so that this
sensory information can be interpreted as vision. The
adult retina comprises six types of neurons [rod and cone
photoreceptors, bipolar, amacrine, horizontal, and ganglion cells (GCs)] and one type of glial cell (the Muller
glia), which are organized into three distinct cellular
layers (Figure 1a). These different cell types are generated from a pool of multipotent (see Glossary) retinal
progenitor cells (RPCs) in a sequence that is remarkably
conserved across all vertebrates (Figure 1b) [1]. During
embryogenesis in the mouse, GCs are generated first,
followed by the production of cone photoreceptors, horizontal cells, and most of the amacrine neurons. Bipolar
neurons, Muller glia, the remaining amacrine neurons,
and most rod photoreceptors are generated postnatally.
However, it is important to emphasize that there is
considerable overlap in the production of retinal cell
types at any given time. In addition, the retina also
contains astrocytes, endothelial cells, and microglia, although these are derived from separate lineages and are
not discussed here.
Corresponding author: Wallace, V.A. ([email protected]).
Keywords: retina; cell fate; lineage; competence; multipotency; micro-RNA.

Retinal cell fate specification; an ongoing, integrated

process
Landmark studies examining the descendants of individual RPCs showed that they are multipotent [24], arguing
against a simple model in which each cell type is produced
by restricted progenitors. Subsequent findings supported
the idea that RPCs generate different cell types by proceeding irreversibly through intrinsically defined competence states that can, to a certain degree, be influenced
by environmental cues (reviewed in [5]). Although the
mechanisms responsible for shifts in competence remain
largely elusive, growing lists of transcription factors have
emerged as key intrinsic regulators of retinal cell fate,
including members of the basic helix-loop-helix (bHLH),
homeodomain, and forkhead families (summarized in
Figure 1b) [68]. There are examples of genes that function instructively to direct cell fate, notably neural retina
leucine zipper protein (Nrl), which is necessary and sufficient to direct photoreceptors to the rod fate [9,10],
whereas other cell fate regulators serve more of a

Glossary
Clone: the progeny arising from an individual retinal progenitor. Determining
the composition of a clone does not provide information about the history of
the progenitor unless the lineage tree is followed.
Commitment: the process by which cell fate is fully determined and can no
longer be affected by environmental influences.
Competence: the ability of an RPC to give rise to a particular set of cell types.
Competence is linked to temporal identity and is distinct from potency, which
refers to the entire complement of cells that a progenitor can ultimately
produce.
Instructive versus permissive: an instructive factor irreversibly drives cells
towards a specific fate, being both necessary and sufficient to generate a
certain cell type. A permissive factor is required for cells to make a particular
cell type, but is not sufficient to promote a specific fate unless influenced by
additional factors.
Lineage: the order of descendents from a dividing cell. Similar to a family tree,
lineage provides information about history, and asks what the outcome was at
each cell division. Following the lineage shows how a cell progresses over time
in relation to its siblings.
Multipotent: the ability to give rise to more than one cell type. Multipotent
stem or progenitor cells are distinguished from pluripotent cells, which have a
broader developmental potential and should, by definition, be able to generate
any cell type of the body.
Precursor: a cell that has completed its terminal mitosis and is committed or
biased towards a certain cell fate, but not yet differentiated.
Progenitor: a dividing cell that, in contrast to a stem cell, cannot proliferate
indefinitely. Progenitor cells give rise to a restricted range of differentiated cell
types, often within a particular tissue or organ.
Specification: the process by which a cell becomes capable of, and biased
towards, a particular fate; however, specification precedes full commitment
and, therefore, cells can be respecified.

0166-2236/$ see front matter . Crown Copyright 2012 Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tins.2012.05.004 Trends in Neurosciences, September 2012, Vol. 35, No. 9

565

Review

Trends in Neurosciences September 2012, Vol. 35, No. 9

Rod
(a)

RPE

Cone

OS

Mller

ONL
OPL

Horizontal

INL
Bipolar

(b)

IPL

Amacrine

GCL
NF

Ganglion

Rax, Hes1, Hes5, Hesr2

Mller glia
Bipolar

Vsx2, Neurod4, Ascl1, Bhlhb5, Vsx1, Irx5, Bhlhb4

Neurod1, Ascl1, Otx2, Nrl,
Nr2e3, Rorb, Prdm1, Pias3

Rod

Six3, Foxn4, Neurod1, Neurod4, Ptf1a,

Bhlhb5, Barhl2, Isl1, Nr4a2, Neurod6,
Satb2, Neurod2

Amacrine

Neurod1, Ascl1, Otx2, Rorb, Prdm1,

Sall3, Pias3, Thrb, Rxrg, Rora, Nr2f1/2
Foxn4, Six3, Neurod4, Ptf1a, Prox1

Cone
Horizontal
Multipotency
network:
Sufu

Ganglion

Atoh7, Pou4f1/2/3, Isl1, Neurod1

Pax6, Rax, Vsx2, Six3,
Six6, Sox2, Nr2e1

RPC

E10

12

14

16

Gli2
18 0 P2

Developmental age (mouse)

10

12

Sox2high
Pax6low

TRENDS in Neurosciences

Figure 1. Development and organization of the vertebrate neural retina. (a) Histological image overlaid with schematic diagram of the mature vertebrate neural retina
structure, using the mouse retina as an example. The photoreceptor outer segments associate with the retinal pigmented epithelium (RPE), whereas their cell bodies reside
in the outer nuclear layer (ONL). The inner nuclear layer (INL) contains the cell bodies of horizontal, bipolar, and amacrine cells, as well as Muller glia. The ganglion cell layer
(GCL) contains the cell bodies of both ganglion and displaced amacrine cells. Connections between the photoreceptor, bipolar, and horizontal cells are found in the outer
plexiform layer (OPL), whereas synapses between bipolar, ganglion, and amacrine cells occur in the inner plexiform layer (IPL). The ganglion cell (GC) axons make up the
nerve fiber (NF) layer. (b) Chronological order and transcriptional regulation of retinal cell birth (embryonic and postnatal times based on mouse development [1]). GCs are
generated first and Muller glia are generated last. A recently described regulatory network involving Hedgehog (Hh) pathway components and a precise ratio of paired box
gene 6 (Pax6) and sex-determining region Y-box containing gene 2 (Sox2) expression maintains progenitor identity and neurogenic potential in mouse retinal progenitor
cells (RPCs) [36,37,40] (green box, right). Various transcription factors (left) direct the specification of each cell type and subtype ([16,17,7377] and reviewed in [68,20,78
80]). Abbreviation: OS, photoreceptor outer segments.

permissive role. Protein atonal homolog 7 (Atoh7/Math5)

provides the best example of an unequivocally identified
intrinsic competence factor. This factor also illustrates a
permissive function because, although it is required for
GC development [11,12], not all ATOH7+ cells differentiate as GCs [13,14], suggesting that it operates with other
factors to control cell fate. Studies on GC specification
have also revealed the complexities that exist within
neurogenic gene networks, exemplified by the fact
that replacement of Atoh7 with another bHLH gene,
neurogenic differentiation 1 (Neurod1), partially rescues
the severe loss of GCs normally observed upon deletion of
Atoh7 [15]. Recently, a possible Atoh7-independent GC
specification program directed by Neurod1 was shown
to account for the small GC population remaining in
Atoh7-null retinas [16,17].
566

It is now becoming clear that final commitment to a

particular cell type is an ongoing process that includes
negative regulatory programs required to stabilize fate
specification, many of which act postmitotically. Examples
include POU-domain class 4 homeobox 2 (Pou4f2), which
acts in early GC precursors to suppress the formation of
late-born GC, amacrine, and horizontal cells [18], and PR
domain zinc finger protein 1 (Prdm1), which represses
bipolar cell fate in photoreceptor precursors [19]. We refer
the reader to a review detailing photoreceptor development, where the rod versus cone decision is determined in
postmitotic photoreceptor precursors [20]. Downstream of
general cell type determination is the issue of neuronal
subtype specification, a subject beyond the scope of this
review. Thus far, many known factors involved in subtype
specification also have roles at earlier time points or in

Review
alternate cell types, as is the case for genes such as LIMhomeodomain 1 (Isl1; cholinergic amacrine, bipolar and
GCs) [21], BarH-like homeobox 2 (Barhl2; multiple amacrine subtypes and GCs) [22] and bHLH family member
e22 (Bhlhe22/Bhlhb5; GABAergic amacrine and Type 2
OFF-cone bipolar cells) [23]. Figure 1b lists recent factors
shown to promote amacrine cell or cone subtypes, as well as
new additions to the photoreceptor specification network.
Superimposed upon the transcriptional programs that
specify retinal cell fates are many other factors, including
the dynamic behavior of RPCs as they migrate within the
neuroblast layer and position themselves to divide. We
refer the reader to recent reviews on these topics, which
include interkinetic nuclear migration and its link to Notch
signaling, as well as the orientation of dividing RPCs and
consequent inheritance of fate determinants [2429].
Despite the remarkable number of transcription factors
known to influence retinal cell fate decisions, little is
known about their upstream regulators, and how earlier
events in RPCs are integrated with downstream transcriptional programs to explain how cells diversify. This raises
two important issues: (i) what are the molecular mechanisms regulating how RPCs generate cell types at the
appropriate time; and (ii) how can different cell types be
produced concurrently from a pool of seemingly indistinguishable progenitors? Here, we review recent developments in the field concerning early aspects of cell fate
determination. We discuss the factors that influence multipotentiality, the molecular control of RPC temporal identity and neurogenic timing, and the relative influence of
lineage-based versus stochastic mechanisms on cell fate
determination.
Factors that influence RPC multipotentiality
RPCs are specified during early eye morphogenesis
through the operation of a transcriptional network of
eye determination factors that includes paired box gene
6 (Pax6), retina and anterior neural fold homeobox (Rax),
and orthodenticle homolog 2 (Otx2) (reviewed in [30]).
Several of these factors function as key regulators of
RPC development at later stages. A case in point is
Pax6, which is essential for the maintenance of RPC multipotency [31]. Conditional inactivation of Pax6 in the
embryonic eyecup generates two types of RPCs: one that
initiates a photoreceptor lineage program before dying,
and a second that differentiates exclusively into cells of
the amacrine lineage [31,32]. This phenotype can be
explained, in part, through the effects of PAX6 on neurogenic gene expression. In the absence of PAX6, the expression of two of its direct targets, neurogenin 2 (Neurog2/
Ngn2) and Atoh7 [31,3335] is not induced, the latter being
essential for the production of GCs [11,12]. More recently,
sex-determining region Y-box containing gene 2 (Sox2) and
suppressor of fused homolog (Sufu) have been added to the
network of genes that function with Pax6 to regulate RPC
multipotency [36,37] (Figure 1b). Sox2, a high mobility
group (HMG)-containing transcription factor, is a key regulator of neurogenesis and stem cell renewal in the central
nervous system (CNS) [38] and is required for neurogenesis in the vertebrate retina [37,39]. In the absence
of Sox2, RPCs fail to differentiate as neurons and, in the

Trends in Neurosciences September 2012, Vol. 35, No. 9

mouse, Sox2-deficient RPCs adopt a non-neural ciliary

epithelium (CE) fate that is characteristic of the cells
located at the peripheral margin of the eyecup [40]. Interestingly, the expression levels of SOX2 and PAX6 are
inversely correlated with neurogenesis, such that SOX2
is higher and PAX6 lower in the central eyecup where
neurons differentiate, and PAX6 is higher and SOX2 lower
in the peripheral eyecup, which develops into non-neural
CE [40]. The conversion of RPC to CE associated with Sox2
loss is accompanied by increased Pax6 expression and,
notably, could be partially compensated by reducing the
dosage of Pax6 [40]. These findings suggest that SOX2
normally antagonizes Pax6 expression and that a precise
ratio of SOX2 and PAX6 is required for the maintenance of
RPC identity and neurogenic potential.
In the frog retina, Sox2 expression is regulated by Wnt/
b-catenin signaling [39,41]. However, this regulatory cascade is not conserved in the mouse, as b-catenin is not
required for neurogenic competence in the mouse retina
[42]. More recent evidence points to a role for the Hedgehog
(Hh) signaling pathway in Sox2 regulation in mouse RPCs.
Secreted Hh ligands activate a signal transduction cascade
that converges on the activity of Gli transcription factors,
the primary transcriptional regulators of this pathway
[43]. Work in several vertebrate species has highlighted
the importance of Hh signaling in controlling cell fate and
progenitor proliferation in the retina (reviewed in [44]). In
addition to positive pathway regulators (such as the Hh
ligands), the Hh pathway is controlled by several negative
regulators, most notably SUFU, which is required to inhibit the activity of Gli transcription factors [45]. RPCs
express Sufu and Gli2, and conditional Sufu inactivation in
embryonic RPCs results in Gli2-dependent ectopic Hh
pathway activation and RPCs adopt amacrine and horizontal cell fates exclusively [36]. Thus, active SUFU inhibition of GLI2 activity is required in embryonic RPCs to
inhibit Hh activation and to maintain RPC multipotency.
Sufu inactivation was associated with increased SOX2 and
reduced Pax6 expression, and a failure to induce the
expression of key cell fate determination genes, including
Atoh7 [36]. These findings raise the possibility that SUFU
acts upstream or at the level of SOX2/PAX6 to regulate
multipotency, a key first step in retinal neural cell differentiation (Figure 1b). Although not tested directly in that
study, these findings also implicate SUFU/GLI2 in the
regulation of Sox2, which is consistent with the evidence
that Hh regulates Sox2 expression in neural stem cells
[46]. It will be important to investigate how Hh regulation
of Sox2 in RPCs controls growth and fate at later stages of
retinal development.
RPC temporal identity and neurogenic timing
Multipotent RPCs cannot give rise to all retinal cell types
at all stages of histogenesis, so what limits the repertoire of
cell types that can be generated at a particular time? The
answer may lie in RPC heterogeneity. Gene expression
profiling of individual mouse RPCs has confirmed that they
exhibit significant heterogeneity over time and even
among progenitors at the same developmental stage, particularly with respect to transcription factors [47]. Thus,
groups of genes whose expression is confined largely to
567

Review
early or late mouse RPCs have been identified. However,
this information did not indicate whether there was a
defined temporal sequence of gene expression that could
explain the chronological order of cell birth, similar to that
observed in Drosophila melanogaster [48].
Homologs of key transcription factors that define the
temporal identity of Drosophila neuroblasts have now been
detected in the developing mammalian retina [47,49,50].
One such factor, IKAROS family zinc finger 1 (Ikzf1/
Ikaros), is a mouse ortholog of hunchback (hb), which is
necessary and sufficient to specify early-born neurons in
Drosophila (reviewed in [48]). Expression analyses combined with lineage tracing showed that IKAROS is
expressed in early but not in late mouse RPCs [49].
Ikaros-deficient retinas exhibit a transient decrease in
proliferation at embryonic day (E) 13 and a permanent
reduction of most early-born cell types (GCs, amacrine, and
horizontal cells, but interestingly, not cones), which is not
accompanied by an increase in late-born types. Accordingly, ectopic expression of Ikaros in late-stage progenitors is
sufficient to promote development of most early-born retinal neurons (GCs, amacrine, and horizontal cells, but
again, not cones) alongside late-born neurons. The fact
that cones are not affected by the gain or loss of Ikaros
suggests that different regulatory mechanisms control the
timing of cone production. Notably, Ikaros overexpression
also causes a slight reduction of bipolar cells and prevents
Muller glia formation, suggesting that loss of Ikaros is
required for the progression to a late temporal (and gliapermissive) state [49]. Thus, Ikaros appears to define
temporal identity in the mouse retina, whereby its timing
of expression corresponds to a stage when RPCs are competent to generate particular cell types, and ectopic expression at the wrong time promotes the production of
temporally inappropriate cell types (Figure 2a). Future
studies may establish whether other factors act in sequence with IKAROS to determine temporal windows
within mammalian RPC lineages.
Interestingly, Ikaros mRNA is expressed throughout
retinal development, whereas the protein is present only
in early RPCs, suggesting that the timing of translation is
crucial to Ikaros regulation [49] (Fig. 2a). Key mediators of
this process could be miRNAs. Hundreds of miRNAs are
expressed in the developing and/or mature mammalian
retina [51,52] and, given their role in translational control,
they may prove to be of central importance during shifts in
temporal identity or fine control of neurogenic timing.
Similar to Ikaros, the expression of Xenopus laevis homeobox genes orthodenticle homolog 5-B (otx5b), visual system
homeobox 1 (vsx1) and otx2 is temporally regulated at the
translational level, and knockdown of the miRNA processing enzyme dicer (dicer1, ribonuclease type III) interferes
with their sequential protein expression patterns [53,54].
Additional work in Xenopus showed that a set of four
miRNAs control the timing of bipolar cell genesis by inhibiting vsx1 and otx2 translation in early progenitors, an
effect that is dependent on cell cycle length rather than
developmental stage [55] (Figure 2b). Cell cycle length,
which is known to increase as retinogenesis progresses
[56], was hypothesized to provide an intrinsic timer that
regulates cell birth through miRNA activity [55,57]. Given
568

Trends in Neurosciences September 2012, Vol. 35, No. 9

its roles in regulating cell cycle length and exit (reviewed in

[24,44]), Sonic Hedgehog (Shh) is a possible mediator of
this process. In support of this, lengthening the cell cycle by
inhibiting Shh signaling downregulates the set of miRNAs
that control vsx1 and otx2 translation in Xenopus [55].
Thus, the function of Shh during temporally regulated
aspects of retinogenesis in other vertebrate species warrants further study.
Loss of miRNA function in the embryonic mouse retina
also interferes with neurogenic timing. In mutants with
Cre-mediated deletion of Dicer1, GCs continue to be produced beyond the normal period of GC genesis compared
with controls [58,59]. Dicer1-deficient RPCs do not express
the late-stage RPC markers Sox9 and achaete-scute complex homolog 1 (Ascl1/Mash1), suggesting that DICER1
processes miRNAs that regulate particular progenitor
characteristics [59]. However, the retinal phenotypes of
mouse Dicer1 mutants are complex and eventually result
in extensive cell death and degeneration; therefore, future
studies examining the roles of specific miRNAs are required. The mechanisms by which miRNAs could regulate
retinal cell fate are not clear, although reduced expression
of Notch and Shh pathway components in Dicer-deficient
mouse retinas are beginning to provide some clues [58,60].
The influence of lineage versus stochastic mechanisms
on cell fate determination
In Drosophila and Caenorhabditis elegans, neurons and
glia are produced from neuroblasts in a reproducible sequence that instructs cell fate [48,61]. Cell fate choices of
individual RPCs may be regulated in a similar fashion
where there is a reproducible order or pattern of cell type
development (e.g., a lineage), or the process could be more
stochastic. It is difficult to resolve this issue by analyzing
the clonal progeny of single marked RPCs because the
clone represents the endpoint of the process and does
not provide information about the sequence of events that
generated it.
Several studies that have used reporter genes and inducible Cre-marking approaches to trace the development of
distinct subsets of RPCs suggest that lineage has a role in
RPC cell fate choice. In the zebrafish (Danio rerio), atoh7
expression defines a reproducible lineage where one daughter of an atoh7+ progenitor adopts a GC fate and the other
daughter adopts a different neural fate, typically a photoreceptor, amacrine, or horizontal cell [14] (Figure 3a). More
recently, it was shown that vsx2 and vsx1 expression also
defines distinct cell subsets in zebrafish [62] (Figure 3a).
Similar to atoh7+ cells, vsx1+ cells have limited proliferation
and give rise primarily to bipolar neurons, whereas vsx2+
cells have greater proliferative potential and give rise to
Muller glia and a subset of bipolars that is distinct from the
vsx1+ cells [62]. Within the Atoh7 and Vsx2 lineages, the
induction of ptf1a controls the development of inhibitory
neurons [63]. Interestingly, the expression of vsx2, vsx1, and
atoh7 appears to define a hierarchy, where vsx2 is expressed
first in most RPCs, followed by the appearance of vsx1+ and
atoh7+ progenitors [62]. Time-lapse imaging in transgenic
fish expressing different reporter genes to discriminate
between vsx2+ and vsx1+ cells was used to demonstrate
that vsx2+ cells give rise to vsx1+ cells and that this

Review

Trends in Neurosciences September 2012, Vol. 35, No. 9

(a)

Temporally inappropriate
Ikaros expression
Ikaros
misexpression
by retroviral
infection

Mouse
?miRNA

Ikaros

Ikaros

P1

X
X

Ikaros

Ikaros KO

Late-born types

Early-born types
E11

E13

E15

E17

P0

P2

P4

P6

Developmental age

(b)

Production of temporally
inappropriate early-born types
(GC, horizontal, amacrine)
Block of Mller glia production
Temporally inappropriate
miRNA downregulation

Xenopus

miRNA
knockdown by
lipofection

miRNA
otx2

otx2

X
vsx1
X

vsx1

Stage 18
otx2
vsx1

Non-bipolar types

Bipolars

miRNA
Bipolar production

Cell cycle length

25

30
35
Developmental stage

40
TRENDS in Neurosciences

Figure 2. Regulation of neurogenic timing in the developing retina. (a) In mouse, the temporal identity of retinal progenitor cells (RPCs) is defined by IKAROS protein
expression, which is a feature of early progenitors [49]. Accordingly, Ikaros-knockout (KO) retinas exhibit a reduction of most early-born cell types, whereas temporally
inappropriate expression of Ikaros in late RPCs promotes the production of early-born types (on right). Interestingly, Ikaros is transcribed throughout retinogenesis but
translated only in early RPCs [49]. Although not currently proven, miRNA-mediated translational control of Ikaros may provide a possible link between miRNA activity and
temporal aspects of retina cell fate in mouse. (b) In Xenopus, bipolar fate is driven by visual system homeobox 1 (vsx1) and orthodenticle homolog 2 (otx2), which are
transcribed in RPCs from stages 15 and 25, respectively, but only translated from stages 37 and 3839, respectively [54]. A set of four cell cycle-regulated miRNAs (miR-129,
miR-155, miR-214, and miR-222) were shown to bind the 30 untranscribed region (UTR) of vsx1 and otx2, inhibiting their translation in early RPCs [55]. Temporally
inappropriate knockdown of these miRNAs in the stage 18 optic vesicle increases both the proportion of vsx1/otx2-translating cells and the proportion of bipolar cells in the
inner nuclear layer (INL; on right) [55]. As in other species and neural tissues, the cell cycle length in Xenopus RPCs increases over time; therefore, bipolar cells are produced
by slower dividing late RPCs [54,56,81]. Lengthening the cell cycle upon treatment with the Sonic hedgehog (Shh) signaling inhibitor cyclopamine downregulates this set of
miRNAs and leads to earlier translation of otx2 and vsx1, whereas blocking cell cycle progression maintains the miRNA expression and inhibits otx2 and vsx1 translation
[54,55]. Thus, translational control may link the temporal identity of RPCs (measured by cell cycle length) to cell fate [57]. Abbreviations: E, embryonic day; GC, ganglion cell;
P, postnatal day.

conversion was associated with the rapid downregulation of

vsx2 [62]. Using similar approaches, vsx2 and atoh7 were
found to be expressed in non-overlapping populations [62].
The basis for this non-overlap could be due, in part, to the
repressive effects of Vsx2, which is known to inhibit Vsx1
expression in murine RPCs [64]. A similar relation between
Vsx2 and atoh7 may also exist, as atoh7 expression is derepressed in vsx2 morphants and Vsx2 binds to an upstream
element in the atoh7 gene, suggesting that the basis for this
repression is direct [62]. Thus, Vsx2 expression defines the
multipotent progenitor pool and may directly repress the

expression of key determinants like vsx1 and atoh7. Over

time, vsx2+ progenitors generate these distinct progenitor
subsets with restricted developmental potential. What is
not known is how vsx2 expression is regulated and how the
choice between progenitor types is controlled.
Can the expression of cell fate determinants define
restricted lineages in the mammalian retina? In the murine retina, ASCL1, NEUROG2 and oligodendrocyte transcription factor 2 (OLIG2) are expressed in all
combinations in dividing RPCs [65,66]. To investigate
whether expression of these bHLH transcription factors
569

Review

Trends in Neurosciences September 2012, Vol. 35, No. 9

Multipotent RPC

Restricted RPC

Major cell types

(a) Fish

Mller Inhibitory:
amacrine
Bipolar
cell subtypes

Vsx2
+Ptf1a

Vsx2

Vsx1
Atoh7

Inhibitory:
Bipolar amacrine
cell subtypes

Vsx2

Vsx1

+Ptf1a
GC

Atoh7

Inhibitory:
or + Ptf1a
Photoreceptor

All horizontal and some

amacrine subtypes

Amacrine Horizontal
(b) Mouse

Extremely
low GC
representation

Ascl1
Photoreceptor Bipolar

Mller

GC amacrine horizontal
All types

Neurog2
Mller
Photoreceptor

Bipolar
TRENDS in Neurosciences

Figure 3. Model of the origin and development of distinct retinal lineages. (a) In zebrafish, visual system homeobox 2 (vsx2) expression marks multipotent progenitors, a
state that is proposed to be maintained by active repression of determinants such as protein atonal homolog 7 (atoh7) and vsx1 [62]. vsx2+ progenitors give rise to nonoverlapping populations of vsx1+ and atoh7+ cells, and well as other non-vsx2+ progenitors (not shown) with restricted developmental potential [62]. Vsx2+ and Vsx1+
lineages have different proliferative potential and they are expressed in distinct populations in the adult retina. Vsx2+ cells become bipolar cells and Muller glia, whereas
Vsx1 marks a different bipolar subset and some amacrine cells [62]. Atoh7 marks cells that are in their last division, which always generate a ganglion cell (GC) and one
other cell type, typically a photoreceptor [14]. ptf1a expression in the Atoh7, Vsx2, and possibly the Vsx1, lineages promotes the development of inhibitory neurons [63]. (b)
In mouse, retinal progenitor cells (RPCs) give rise to progenitors that express all combinations of achaete-scute complex homolog 1 (Ascl1) and neurogenin 2 (Neurog2)
[65]. Fate mapping using drug-inducible Cre-ER Ascl1 and Neurog2 knock-in alleles reveals that Ascl1+ progenitors can generate all cell types, with the exception of GCs,
which are under-represented in this lineage [65]. Neurog2 fate-mapped cells can develop into all cell types but, compared with the Ascl1 lineage, generate fewer cells at any
given stage, suggesting that these cells are in their last division [65]. Thus, Neurog2 marks cells that are in their last division and Ascl1 defines a restricted progenitor pool
that does not make GCs. Blue arrows indicate the final division of a restricted progenitor, whereas the number of brown arrows indicates the approximate proliferative
potential of a restricted progenitor.

could define developmental potential, a drug-inducible

Cre-recombinase based approach to fate map Neurog2
and Ascl1-expressing progenitor cells over time was used.
Ascl1+ cells could generate all cell types, with the exception
of GCs, which were under-represented in these progeny
[65]. Ascl1+ cells still initiated Atoh7 expression and Ascl1
knockout retinas did not overproduce GCs; therefore, this
restriction in developmental potential in the Ascl1+ lineage does not appear to involve effects on Atoh7 expression
or Ascl1-mediated antagonism of GC development [65]. By
contrast, Neurog2+ cells differentiated as GCs, as well as
570

other retinal cell types [66]. However, compared with

Ascl1+ cells, Neurog2+ RPCs tended to adopt fates that
were typical of the cell types being produced at that stage
in the retina and gave rise to fewer cells, suggesting that
Neurog2+ cells were typically in their last division [65].
The notion that Neurog2 is expressed before cell cycle exit
is consistent with the data showing that induction of
Neurog2 precedes Atoh7 expression and GC differentiation
in the murine retina [67]. Thus, Ascl1 and Neurog2 expression define distinct RPC developmental outcomes, with
Ascl1+ cells being restricted in their competence to adopt

Review

Trends in Neurosciences September 2012, Vol. 35, No. 9

the GC fate and having greater proliferative potential than

Neurog2+ RPCs (Figure 3b).
Although it is well established that RPC competence to
generate early cell types is an intrinsic property of the cells,
Ascl1 is one of the first markers that can define RPCs with
limited or no competence to generate GCs. One question
that arises from these studies is how the production of
lineage-restricted Ascl1+ progenitors is linked to signaling
pathways, such as Shh, and growth and differentiation
factor 11 (GDF11), that negatively regulate GC development [68]. Ascl1 has been shown to be a direct Hh/Gli
target gene in P19 embryonal carcinoma cells [69] and loss
of Hh signaling in the retina is associated with reduced
Ascl1 expression [70]. Thus, a future avenue for investigation is how Shh and other extrinsic signaling pathways
contribute to the development of this Ascl1-restricted
lineage.
Although these lineage-tracing studies in fish and
mouse point to the existence of restricted progenitors,
other studies using time-lapse video microscopy to reconstruct entire lineage trees of rat RPCs in clonal density
cultures [71] have led to a different conclusion. Because the
cells were grown under clonal conditions and, therefore,
free from the influence of extrinsic cues in the retina, this
analysis would reveal the intrinsic biases of progenitors as
well as potential effects of locally acting cues within the
clone. Here, the outcomes of each division and cell type
composition in the lineage trees fit with a stochastic pattern of behavior. For example, the frequency of divisions
where one daughter remained a progenitor and the other
differentiated (Pr/D), or where both daughters differentiated (D/D) or remained as progenitors (Pr/Pr), was fixed
during the development of the clone and did not depend on
the outcome of the previous division (Figure 4a). Similarly,
(a)
Div 1: 5% 20% 75%

Pr Pr

Pr D

D D

5%

20%

75%

Div 2: 5% 20% 75%

(b)
P

70%

10%

10%

10%

TRENDS in Neurosciences

Figure 4. Lineage reconstruction of rat retinal progenitor cells (RPCs). (a)

Frequency of divisions resulting in two progenitors (Pr/Pr), a progenitor and
differentiated cell (Pr/D), or two differentiated cells (D/D) observed by time-lapse
imaging of clonal density cultures [71]. For every division in the lineage tree shown
on the right, the frequency of the different outcomes does not change. (b) The cell
fate choice as a function of position within the lineage. The cell fate choice of
daughter cells at each division fits with a stochastic model where the probability of
a given cell type is the same as the frequency of that cell type in the retina [71].
Thus, at each division where a daughter differentiates, the choice of photoreceptor
(P) is higher (as indicated by the unequal weighting of the dice towards the P
choice) compared with other cell fates (A, amacrine; B, bipolar; and M, Muller).

the types of cell generated, with the exception of a few

outliers, fit with a stochastic model of cell behavior where
the probability of generating a particular cell type is the
same as the frequency of that cell type in the mature retina
(Figure 4b). Thus, with every cell division, there is a higher
probability of generating a rod than a bipolar cell because
the frequency of rods is greater than the frequency of
bipolar cells. Although there were reproducible lineages,
such as one containing photoreceptors exclusively, this
outcome would also have been predicted by comparing
the data with a stochastic model. Finally, in contrast to
the chronological birth order of retinal cells at the population level in the mammalian retina, this order was not
reproduced within a lineage [71]. For example, there were
instances where bipolar cells were generated before amacrine cells in a lineage and Muller cells before photoreceptors [71]. These results contrast with those in the frog,
where bromodeoxyuridine (BrdU)-labeling studies
showed, with a few exceptions, a regular sequence of cell
type development [72]. This difference could reflect constraints imposed on developmental timing related to the
length of histogenesis in frogs, which is considerably
shorter than in rodents.
Is it possible to reconcile a stochastic mode of differentiation with the evidence for lineage restriction in Ascl1+
and Ath5+ cells or the immediate cell cycle exit of Neurog2+
progenitors? It is important to emphasize that a stochastic
model does not imply an absence of regulation. The different probabilities for the generation of each cell type probably reflect the operation of a complex regulatory machinery
that influences the expression and activities of cell cycle
and cell fate determinants. Thus, the activities of bHLH
proteins could be involved in either establishing or mediating the operations of this intrinsic machinery. Indeed, it
would be interesting to couple clonal observation studies in
vitro with the analysis of transcription factor expression.
One might predict that divisions that result in a D/D
outcome are associated with Neurog2 induction before that
final division. It is also important to note that, in the clonal
RPC lineage reconstruction, the investigators were looking
at a stage when RPCs are normally only producing four cell
types and, therefore, were unable to develop lineage trees
to describe the behavior of progenitors that could generate
early cell types such as GC and horizontal cells. Finally, it
is possible that cells within a particular competence state
make stochastic choices. Whereas Ascl1+ cells tended to
generate cell types that were typical of the population at
that stage, it would be interesting to compare the order of
cell generation within the cohort.
Concluding remarks
The field is moving towards combining the identification of
cell fate determinants with investigations into how these
factors are regulated and the mechanisms that control the
timing of cell fate production. Key themes are emerging that
involve the control of cell fate determinant translation and
the timing of cell determination, as well as the conservation
of fly neurogenic genes in the temporal competence to
generate early cell types. Sophisticated time-lapse imaging
and Cre-dependent lineage tracing have pointed to a role for
lineage in the behavior of RPCs. Although stochastic
571

Review
mechanisms of diversification are not completely incompatible with a lineage model, it will be important to define the
extent to which lineage versus stochastic processes operate
in cell fate decisions, as this information will define the next
line of enquiry. Aside from deepening the understanding of
how retinal development proceeds, these differences will
have an impact on the approaches towards directed differentiation of retinal cell types from stem cells for therapeutic
purposes. Finally, we have highlighted recent advances in
understanding of the function of transcription factors in the
cell diversification process; however, it will be important for
future studies to link the function of these genes with the
extrinsic cues in the retina that are known to regulate
growth and differentiation, and that modulate cell fate
decisions.
Acknowledgments
We are grateful to Michel Cayouette for helpful discussion and critical
comments on the manuscript. We thank Glen Oomen for his assistance
with the illustrations in Figure 4. This work was supported by operating
grants from the Canadian Institutes of Health Research, the Foundation
Fighting Blindness Canada, and the Cancer Research Society (to V.A.W.).

References
1 Young, R.W. (1985) Cell differentiation in the retina of the mouse. Anat.
Rec. 212, 199205
2 Holt, C.E. et al. (1988) Cellular determination in the Xenopus retina is
independent of lineage and birth date. Neuron 1, 1526
3 Turner, D.L. and Cepko, C.L. (1987) A common progenitor for neurons
and glia persists in rat retina late in development. Nature 328, 131
136
4 Wetts, R. and Fraser, S.E. (1988) Multipotent precursors can give rise
to all major cell types of the frog retina. Science 239, 11421145
5 Livesey, F.J. and Cepko, C.L. (2001) Vertebrate neural cell-fate
determination: lessons from the retina. Nat. Rev. Neurosci. 2, 109118
6 Hatakeyama, J. and Kageyama, R. (2004) Retinal cell fate
determination and bHLH factors. Semin. Cell Dev. Biol. 15, 8389
7 Marquardt, T. (2003) Transcriptional control of neuronal
diversification in the retina. Prog. Retin. Eye Res. 22, 567577
8 Ohsawa, R. and Kageyama, R. (2008) Regulation of retinal cell fate
specification by multiple transcription factors. Brain Res. 1192, 9098
9 Mears, A.J. et al. (2001) Nrl is required for rod photoreceptor
development. Nat. Genet. 29, 447452
10 Oh, E.C. et al. (2007) Transformation of cone precursors to functional
rod photoreceptors by bZIP transcription factor NRL. Proc. Natl. Acad.
Sci. U.S.A. 104, 16791684
11 Brown, N.L. et al. (2001) Math5 is required for retinal ganglion cell and
optic nerve formation. Development 128, 24972508
12 Wang, S.W. et al. (2001) Requirement for math5 in the development of
retinal ganglion cells. Genes Dev. 15, 2429
13 Feng, L. et al. (2010) MATH5 controls the acquisition of multiple
retinal cell fates. Mol. Brain 3, 36
14 Poggi, L. et al. (2005) Influences on neural lineage and mode of division
in the zebrafish retina in vivo. J. Cell Biol. 171, 991999
15 Mao, C.A. et al. (2008) Rewiring the retinal ganglion cell gene
regulatory network: Neurod1 promotes retinal ganglion cell fate in
the absence of Math5. Development 135, 33793388
16 Kiyama, T. et al. (2011) Overlapping spatiotemporal patterns of
regulatory gene expression are required for neuronal progenitors to
specify retinal ganglion cell fate. Vision Res. 51, 251259
17 Mao, C.A. et al. (2011) Neuronal transcriptional repressor REST
suppresses an Atoh7-independent program for initiating retinal
ganglion cell development. Dev. Biol. 349, 9099
18 Qiu, F. et al. (2008) A comprehensive negative regulatory program
controlled by Brn3b to ensure ganglion cell specification from
multipotential retinal precursors. J. Neurosci. 28, 33923403
19 Brzezinski, J.A.T. et al. (2010) Blimp1 controls photoreceptor versus
bipolar cell fate choice during retinal development. Development 137,
619629

572

Trends in Neurosciences September 2012, Vol. 35, No. 9

20 Swaroop, A. et al. (2010) Transcriptional regulation of photoreceptor

development and homeostasis in the mammalian retina. Nat. Rev.
Neurosci. 11, 563576
21 Elshatory, Y. et al. (2007) Islet-1 controls the differentiation of retinal
bipolar and cholinergic amacrine cells. J. Neurosci. 27, 1270712720
22 Ding, Q. et al. (2009) BARHL2 differentially regulates the development
of retinal amacrine and ganglion neurons. J. Neurosci. 29, 39924003
23 Feng, L. et al. (2006) Requirement for Bhlhb5 in the specification of
amacrine and cone bipolar subtypes in mouse retina. Development 133,
48154825
24 Agathocleous, M. and Harris, W.A. (2009) From progenitors to
differentiated cells in the vertebrate retina. Annu. Rev. Cell Dev.
Biol. 25, 4569
25 Latasa, M.J. et al. (2009) Cell cycle control of Notch signaling and the
functional regionalization of the neuroepithelium during vertebrate
neurogenesis. Int. J. Dev. Biol. 53, 895908
26 Pierfelice, T. et al. (2011) Notch in the vertebrate nervous system: an
old dog with new tricks. Neuron 69, 840855
27 Shimojo, H. et al. (2011) Dynamic expression of notch signaling genes in
neural stem/progenitor cells. Front. Neurosci. 5, 78
28 Taverna, E. and Huttner, W.B. (2010) Neural progenitor nuclei IN
motion. Neuron 67, 906914
29 Willardsen, M.I. and Link, B.A. (2011) Cell biological regulation of
division fate in vertebrate neuroepithelial cells. Dev. Dyn. 240, 1865
1879
30 Graw, J. (2010) Eye development. Curr. Top. Dev. Biol. 90, 343386
31 Marquardt, T. et al. (2001) Pax6 is required for the multipotent state of
retinal progenitor cells. Cell 105, 4355
32 Oron-Karni, V. et al. (2008) Dual requirement for Pax6 in retinal
progenitor cells. Development 135, 40374047
33 Riesenberg, A.N. et al. (2009) Pax6 regulation of Math5 during mouse
retinal neurogenesis. Genesis 47, 175187
34 Skowronska-Krawczyk, D. et al. (2009) Conserved regulatory
sequences in Atoh7 mediate non-conserved regulatory responses in
retina ontogenesis. Development 136, 37673777
35 Willardsen, M.I. et al. (2009) Temporal regulation of Ath5 gene
expression during eye development. Dev. Biol. 326, 471481
36 Cwinn, M.A. et al. (2011) Suppressor of fused is required to maintain
the multipotency of neural progenitor cells in the retina. J. Neurosci.
31, 51695180
37 Taranova, O.V. et al. (2006) SOX2 is a dose-dependent regulator of
retinal neural progenitor competence. Genes Dev. 20, 11871202
38 Pevny, L.H. and Nicolis, S.K. (2009) Sox2 roles in neural stem cells. Int.
J. Biochem. Cell Biol. 42, 421424
39 Van Raay, T.J. et al. (2005) Frizzled 5 signaling governs the neural
potential of progenitors in the developing Xenopus retina. Neuron 46,
2336
40 Matsushima, D. et al. (2011) Combinatorial regulation of optic cup
progenitor cell fate by SOX2 and PAX6. Development 138, 443454
41 Agathocleous, M. et al. (2009) A directional Wnt/beta-catenin-Sox2proneural pathway regulates the transition from proliferation to
differentiation in the Xenopus retina. Development 136, 32893299
42 Fu, X. et al. (2006) Beta-catenin is essential for lamination but not
neurogenesis in mouse retinal development. Dev. Biol. 299, 424
437
43 Jiang, J. and Hui, C.C. (2008) Hedgehog signaling in development and
cancer. Dev. Cell 15, 801812
44 Wallace, V.A. (2008) Proliferative and cell fate effects of Hedgehog
signaling in the vertebrate retina. Brain Res. 1192, 6175
45 Ruel, L. and Therond, P.P. (2009) Variations in Hedgehog signaling:
divergence and perpetuation in Sufu regulation of Gli. Genes Dev. 23,
18431848
46 Takanaga, H. et al. (2009) Gli2 is a novel regulator of sox2 expression in
telencephalic neuroepithelial cells. Stem Cells 27, 165174
47 Trimarchi, J.M. et al. (2008) Individual retinal progenitor cells display
extensive heterogeneity of gene expression. PLoS ONE 3, e1588
48 Brody, T. and Odenwald, W.F. (2005) Regulation of temporal identities
during Drosophila neuroblast lineage development. Curr. Opin. Cell
Biol. 17, 672675
49 Elliott, J. et al. (2008) Ikaros confers early temporal competence to
mouse retinal progenitor cells. Neuron 60, 2639
50 Blackshaw, S. et al. (2004) Genomic analysis of mouse retinal
development. PLoS Biol. 2, E247

Review
51 Hackler, L., Jr et al. (2010) MicroRNA profile of the developing mouse
retina. Invest. Ophthalmol. Vis. Sci. 51, 18231831
52 Karali, M. et al. (2010) miRNeye: a microRNA expression atlas of the
mouse eye. BMC Genomics 11, 715
53 Decembrini, S. et al. (2008) Dicer inactivation causes heterochronic
retinogenesis in Xenopus laevis. Int. J. Dev. Biol. 52, 10991103
54 Decembrini, S. et al. (2006) Timing the generation of distinct retinal
cells by homeobox proteins. PLoS Biol. 4, e272
55 Decembrini, S. et al. (2009) MicroRNAs couple cell fate and
developmental timing in retina. Proc. Natl. Acad. Sci. U.S.A. 106,
2117921184
56 Alexiades, M.R. and Cepko, C. (1996) Quantitative analysis of
proliferation and cell cycle length during development of the rat
retina. Dev. Dyn. 205, 293307
57 Pitto, L. and Cremisi, F. (2010) Timing neurogenesis by cell cycle? Cell
Cycle 9, 434435
58 Davis, N. et al. (2011) Roles for Dicer1 in the patterning and
differentiation of the optic cup neuroepithelium. Development 138,
127138
59 Georgi, S.A. and Reh, T.A. (2010) Dicer is required for the transition
from early to late progenitor state in the developing mouse retina. J.
Neurosci. 30, 40484061
60 Georgi, S.A. and Reh, T.A. (2011) Dicer is required for the maintenance
of Notch signaling and gliogenic competence during mouse retinal
development. Dev. Neurobiol. 71, 11531169
61 Bertrand, V. and Hobert, O. (2010) Lineage programming: navigating
through transient regulatory states via binary decisions. Curr. Opin.
Genet. Dev. 20, 362368
62 Vitorino, M. et al. (2009) Vsx2 in the zebrafish retina: restricted
lineages through derepression. Neural Dev. 4, 14
63 Jusuf, P.R. et al. (2011) Origin and determination of inhibitory cell
lineages in the vertebrate retina. J. Neurosci. 31, 25492562
64 Clark, A.M. et al. (2008) Negative regulation of Vsx1 by its paralog Chx10/
Vsx2 is conserved in the vertebrate retina. Brain Res. 1192, 99113
65 Brzezinski, J.A.T. et al. (2011) Ascl1 expression defines a
subpopulation of lineage-restricted progenitors in the mammalian
retina. Development 138, 35193531
66 Ma, W. and Wang, S.Z. (2006) The final fates of neurogenin2expressing cells include all major neuron types in the mouse retina.
Mol. Cell. Neurosci. 31, 463469

Trends in Neurosciences September 2012, Vol. 35, No. 9

67 Hufnagel, R.B. et al. (2010) Neurog2 controls the leading edge of

neurogenesis in the mammalian retina. Dev. Biol. 340, 490503
68 Esteve, P. and Bovolenta, P. (2006) Secreted inducers in vertebrate eye
development: more functions for old morphogens. Curr. Opin.
Neurobiol. 16, 1319
69 Voronova, A. et al. (2011) Ascl1/Mash1 is a novel target of Gli2 during
Gli2induced neurogenesis in P19 EC cells. PLoS ONE 6, e19174
70 Sakagami, K. et al. (2009) Distinct effects of Hedgehog signaling on
neuronal fate specification and cell cycle progression in the embryonic
mouse retina. J. Neurosci. 29, 69326944
71 Gomes, F.L. et al. (2011) Reconstruction of rat retinal progenitor cell
lineages in vitro reveals a surprising degree of stochasticity in cell fate
decisions. Development 138, 227235
72 Wong, L.L. and Rapaport, D.H. (2009) Defining retinal progenitor cell
competence in Xenopus laevis by clonal analysis. Development 136,
17071715
73 Cherry, T.J. et al. (2011) NeuroD factors regulate cell fate and neurite
stratification in the developing retina. J. Neurosci. 31, 73657379
74 Jiang, H. and Xiang, M. (2009) Subtype specification of GABAergic
amacrine cells by the orphan nuclear receptor Nr4a2/Nurr1. J.
Neurosci. 29, 1044910459
75 Kay, J.N. et al. (2011) Neurod6 expression defines new retinal
amacrine cell subtypes and regulates their fate. Nat. Neurosci. 14,
965972
76 Lelievre, E.C. et al. (2011) Ptf1a/Rbpj complex inhibits ganglion cell
fate and drives the specification of all horizontal cell subtypes in the
chick retina. Dev. Biol. 358, 296308
77 Luo, H. et al. (2012) Forkhead box N4 (Foxn4) activates Dll4-Notch
signaling to suppress photoreceptor cell fates of early retinal
progenitors. Proc. Natl. Acad. Sci. U.S.A. 109, E553E562
78 Andreazzoli, M. (2009) Molecular regulation of vertebrate retina cell
fate. Birth Defects Res. C 87, 284295
79 Poche, R.A. and Reese, B.E. (2009) Retinal horizontal cells: challenging
paradigms of neural development and cancer biology. Development
136, 21412151
80 Reese, B.E. (2011) Development of the retina and optic pathway. Vision
Res. 51, 613632
81 Salomoni, P. and Calegari, F. (2010) Cell cycle control of mammalian
neural stem cells: putting a speed limit on G1. Trends Cell Biol. 20,
233243

573