Photoreceptor cell fate specification in vertebrates

The retina is the part of the eye that senses light and communicates this information to the brain via the optic nerve. The vertebrate retina comprises several major cell types – retinal ganglion output neurons, bipolar, horizontal and amacrine interneurons, Müller glia and photoreceptors – each of which carries out specialized functions. The detection of light stimuli is mediated by photoreceptors, which are highly specialized neurons that occupy the outermost layer of the retina (Fig. 1A). Vertebrate photoreceptors have a unique polarized morphology (Fig. 1B) that includes an outer segment filled with photoreceptive pigments (opsins), a connecting cilium, the inner segment, the nucleus, the axon and the synaptic terminal (Rodieck, 1998; Swaroop et al., 2010). Normal photoreceptor function depends upon maintaining these compartments. There are two basic subtypes of photoreceptors: rods and cones. Rod photoreceptors are highly sensitive and function in dim light to mediate night vision; they use rhodopsin as the light-sensitive photopigment. Cone photoreceptors are present as several subtypes, each of which contains one or more opsins that are maximally sensitive to a different wavelength of light; cones thus mediate color perception.

Rods and cones, like other neurons in the central nervous system, express genes that are required for synapse formation and neurotransmitter release. The process of phototransduction requires the expression of many genes that uniquely mark photoreceptors. Historically, this distinction has facilitated the discovery of gene regulatory networks that control photoreceptor specification (see Glossary, Box 1) and physiology. Understanding the full cascade of gene regulatory events that control the specification of photoreceptor fate (see Glossary, Box 1) remains a fascinating and unresolved problem in developmental biology. Diseases that affect the genesis, survival or function of photoreceptors are major causes of vision loss (Swaroop et al., 2010) (Box 2). With recent increases in the prospect of regenerative therapies to reverse vision loss, interest in uncovering the mechanisms that govern photoreceptor development has grown.

In this Review, we first provide an overview of retinal development. This topic has been discussed in many recent reviews (Agathocleous and Harris, 2009; Brzezinski and Reh, 2010; Bassett and Wallace, 2012; Xiang, 2013; Boije et al., 2014; Cepko, 2014) and is covered only briefly here. We then discuss the current model of photoreceptor development that has emerged from over 25 years of studies on retinogenesis. This model of photoreceptor fate specification has had practical applications in the field of regenerative medicine; some of the key signaling molecules identified from developmental studies are now being used to enhance photoreceptor genesis from pluripotent stem cells. Throughout the Review, we highlight gaps in our understanding of photoreceptor fate specification and we finish by discussing several important questions that remain unresolved.

Photoreceptors are generated from a population of proliferative progenitors (see Glossary, Box 1) in the retinal neuroepithelium, a derivative of the neural tube. Rod and cone photoreceptors, along with all of the other retinal cell types, are generated from what are known as multipotent progenitors. Cell lineage-tracing studies in mice, frogs and fish have shown that these progenitor cells can divide both symmetrically and asymmetrically, and that their terminal division can generate different retinal cell types (Turner and Cepko, 1987; Wetts and Fraser, 1988; Turner et al., 1990). These lineage studies were widely interpreted to demonstrate that progenitors are not limited to producing a single type of retinal neuron, i.e. there are no progenitors that are solely restricted to producing photoreceptors. However, this is not always the case; it has been known for many years that teleost fish (e.g. zebrafish and goldfish) possess a progenitor that is restricted to producing only rod photoreceptors, although this has been considered an adaptation to maintain light sensitivity and rod density as the retina grows in postembryonic stages (Raymond and Rivlin, 1987), and more recent studies have provided evidence for horizontal- and cone-restricted progenitors in fish (Godinho et al., 2007; He et al., 2012; Suzuki et al., 2013). Together, these studies suggest that many of the specification events leading to the generation of photoreceptors, and even their specific subtype, can occur in either proliferative progenitor or postmitotic precursor (see Glossary, Box 1) cells.

Experiments that correlate the timing of progenitor cell cycle exit (‘birth') with cell fate have shown that the different retinal cell types are formed in a partly overlapping sequence that is largely conserved among vertebrates (Sidman, 1961; Young, 1985; Altschuler et al., 1991; Rapaport et al., 2004; Wong and Rapaport, 2009). When integrated with lineage-tracing data, these birthdating studies indicate that the frequency of each cell type generated by progenitors changes over time. In line with this, the isolation of retinal progenitors from different stages of development further demonstrated that the probability of generating specific fates in vitro changes over the period of neurogenesis (Reh and Kljavin, 1989; Watanabe and Raff, 1990). These data led several investigators to propose a model in which progenitors change their potential (see Glossary, Box 1) to control fate specification (Reh and Cagan, 1994; Cepko et al., 1996; Frantz and McConnell, 1996). In these models, both the acquisition and the removal (restriction) of potential are critical specification events. The change in potential is likely to be due to a combination of intrinsic and extrinsic factors, such as transcription factors and signaling molecules, respectively. For example, progenitors from early stages of retinal development do not express transcription factors such as Sox9 and Ascl1 (Jasoni and Reh, 1996; Georgi and Reh, 2010; Brzezinski et al., 2011), and are unresponsive to epidermal growth factor (EGF), but those from later times express these transcription factors and respond to EGF (Anchan et al., 1991; Lillien, 1995). In addition to transcription factors and signaling molecules, cells may restrict their potential by changing their epigenetic landscape, such that fate-determining transcription factors no longer have access to chromatin at key target sequences. In the absence of either event, a progenitor could still be multipotent, but such a cell would only be able to produce a restricted range of cell fates at that particular time. Such dynamic fate potential models require a clock-like mechanism, which again might be intrinsically controlled or rely on signals from surrounding cells. A discussion of the mechanisms that underlie this ‘clock' of neurogenesis, and the epigenetic control of photoreceptor development (Yang et al., 2015), are beyond the scope of this Review, although it should be noted that some of the key transcription factors and miRNA networks (e.g. Let7, Lin28) that control analogous pathways in other systems have been shown to regulate developmental timing in the retina as well (La Torre et al., 2013).

Photoreceptors are generated from retinal progenitors over most of the course of retinal development. In general, the first cone photoreceptors are formed before the initiation of rod photoreceptor production (Fig. 1C) (Carter-Dawson and LaVail, 1979). Within mammals, this is particularly evident in those species that exhibit an extended gestation, such as the monkey, where the first cone photoreceptors are born more than a week before the onset of rod photoreceptor production (La Vail et al., 1991). To better understand how retinal progenitors produce photoreceptors, a number of studies have focused on the transcription factors expressed by progenitors and early photoreceptors (see Table 1).

One of the key transcription factors necessary for photoreceptor development is Otx2. In the mouse retina, most proliferative progenitors do not express detectable levels of Otx2, but this transcription factor is rapidly upregulated as cells exit the mitotic cycle (Muranishi et al., 2011). Otx2 appears to act early in the process by which photoreceptors and bipolar cells acquire their unique identities. The deletion of Otx2 from the retina prevents the development of photoreceptors and bipolar cells, whereas its overexpression can promote the formation of both cell types (Nishida et al., 2003; Koike et al., 2007; Sato et al., 2007; Wang et al., 2014). Accordingly, chromatin immunoprecipitation (ChIP) and enhancer characterization experiments show that Otx2 associates with the promoters and enhancers of genes expressed in both photoreceptors and bipolar cells (Kim et al., 2008; Brzezinski et al., 2013; Samuel et al., 2014).

The initiation of Otx2 expression in progenitors leads to the activation of additional transcription factors required for correct fate specification. Two of these factors, Vsx2 (Chx10) and Prdm1 (Blimp1), act downstream of Otx2 and control whether Otx2-expressing cells develop as photoreceptors or bipolar cells (Fig. 2A). Vsx2 is expressed by progenitors and, after cell cycle exit, is upregulated in bipolar cells directly downstream of Otx2 (Kim et al., 2008). Overexpression and ChIP experiments have shown that Vsx2 represses photoreceptor-specific genes (Dorval et al., 2005,, 2006; Livne-Bar et al., 2006). It was also shown that Vsx2 regulates progenitor cell proliferation, and that progenitors in Vsx2 mutant mice fail to generate bipolar cells even when their proliferation defects are rescued (Burmeister et al., 1996; Green et al., 2003). Otx2 also directly activates Prdm1 (Brzezinski et al., 2010; Katoh et al., 2010; Wang et al., 2014). In Prdm1 mutants, cells begin to adopt photoreceptor identity but instead switch to a bipolar cell fate (Brzezinski et al., 2010,, 2013; Katoh et al., 2010). Thus, early-born Otx2-expressing cells, which are normally limited to photoreceptor fates, can generate bipolar cells in these mutants, suggesting that Prdm1 prevents Otx2-expressing cells from adopting a bipolar fate. These data support a model in which Otx2 expression provides precursors with the potential to become photoreceptors and bipolar cells, and then the Otx2 targets Prdm1 and Vsx2 restrict precursors to either a photoreceptor or bipolar fate (Fig. 2A), respectively. How the balance between these fate choices is dynamically regulated over developmental time is unknown.

Another transcription factor that is important for eye development is Pax6, which is expressed in all progenitors. Retinas lacking Pax6 fail to form photoreceptors and most other cell types (Marquardt et al., 2001). Interestingly, some photoreceptor genes (e.g. Crx) are inappropriately activated in these mutants (Oron-Karni et al., 2008). These data suggest that Pax6 is required in progenitors for photoreceptor (and other cell type) potential, but that it also temporarily restricts the ability of progenitors to activate a photoreceptor program.

Since most progenitors are not restricted to producing only a single cell type, such as a photoreceptor, what determines which precursor cells express Otx2 and acquire a photoreceptor fate? One possibility is that only a subset of the progenitor population has the ability to activate Otx2. This could be a stochastic process in progenitors or the result of a preprogrammed process of progenitor potential regulation. Another possibility is that the activation of Otx2 in progenitors is a default outcome, which must then be repressed to generate other cell types. In support of this latter hypothesis, experiments with dissociated chick retinal cell cultures suggest that all progenitors are able to express Otx2 (Adler and Hatlee, 1989). Similarly, blocking Notch signaling in cultured mouse retinas can drive Otx2 expression and photoreceptor formation at a high rate (Nelson et al., 2007). Recently, lineage-tracing studies have revealed distinct subsets of proliferative progenitors that exhibit restricted fate potential and express varying combinations of transcription factors such as Ascl1 and Olig2. The fate mapping of Ascl1-expressing cells demonstrated that they give rise to all types except retinal ganglion cells, suggesting that Ascl1 marks a single fate-restricted lineage in the retina (Fig. 2B) (Brzezinski et al., 2011). Similarly, Olig2-expressing cells can give rise to all but ganglion cells and Müller glia, representing ∼96% of the cells in the adult mouse retina (Fig. 2B) (Jeon et al., 1998; Hafler et al., 2012). Furthermore, some of these Olig2-expressing cells were poised between two outcomes; for example, between cone and horizontal cell fates, or between rod and amacrine fates. Similar to the Olig2 lineage-tracing results, Prdm1 fate-mapping experiments revealed that the Prdm1-expressing population can give rise to all cell types except ganglion cells and Müller glia (Brzezinski et al., 2013). Given that Prdm1 is made only by Otx2-expressing cells, these results imply that the Otx2-expressing cohort can generate photoreceptors, bipolar cells, amacrine cells and horizontal cells. This result is consistent with the hypothesis that the activation of Otx2 in progenitors is a default outcome, which must then be repressed to generate other retinal cell types, although definitive Otx2 lineage-tracing studies are required to further test this hypothesis.

If the expression of Otx2 is indeed the default program, how then does the retina make cells other than bipolar cells and photoreceptors? Recent insights have come from characterizing other transcription factors expressed in retinal progenitors. For example, the transcription factors Foxn4 and Rorβ are expressed by progenitors (Chow et al., 1998; Li et al., 2004; Luo et al., 2012; Liu et al., 2013; Fu et al., 2014) and combine to induce the expression of Ptf1a, which encodes a transcription factor that is necessary for amacrine and horizontal cell genesis (Fujitani et al., 2006; Liu et al., 2013). Lineage tracing of Ptf1a-expressing cells labels only horizontal and amacrine cells (Fujitani et al., 2006), suggesting that any Otx2-expressing cells that also express Ptf1a have already (or will) become restricted to these two fates. The observation of two-cell clones containing a photoreceptor and an amacrine/horizontal outcome (Hafler et al., 2012) also suggests that this potential restriction occurs during the last division or in postmitotic precursors in mice. By contrast, Otx2-expressing precursors do not appear to generate ganglion cells, the first cell type generated in the retina. The ability to form these cells is mediated by Atoh7 (Math5), but lineage-tracing studies show that only a small subset of cells acquires the potential to generate ganglion cells (Brown et al., 2001; Wang et al., 2001; Yang et al., 2003; Feng et al., 2010; Brzezinski et al., 2012). This suggests a model in which the initial progenitor pool produces precursors that express Atoh7, and these undergo subsequent specification events to produce ganglion cells. Those progenitors that remain in the cell cycle will express Ascl1 and generate Otx2-expressing precursors when they exit the cell cycle (Fig. 3A). Inhibition of Notch signaling seems to be the event that triggers cell cycle exit and the expression of Atoh7 at early stages of retinal development, and then Otx2 at later stages (Nelson et al., 2006,, 2007; Maurer et al., 2014).

Because of their unique morphology and gene expression profiles, photoreceptors were among the few neuronal subtypes that could be identified reliably in vitro. This allowed for the identification of soluble regulators of photoreceptor cell fate specification. These early experiments provided evidence for both an intrinsic potential ‘clock' and for cell-to-cell interactions that control photoreceptor fate. In addition, co-culturing cells from early and late stages of retinal development provided evidence that factors in the progenitor cell environment might influence their ultimate fate (Watanabe and Raff, 1990,, 1992; Altshuler and Cepko, 1992; Reh, 1992; Belliveau and Cepko, 1999; Belliveau et al., 2000). In particular, it was shown that factors present in the postnatal rat or mouse retina promoted the development of rhodopsin-expressing cells in the cultures. These initial findings were followed by studies that aimed to define the ‘rod-inducing' molecule in postnatal cultures, and eventually both taurine and retinoic acid were found to promote rhodopsin expression in cultures of retinal cells (Altshuler et al., 1993; Kelley et al., 1994). Other soluble factors, such as activin and Shh, also promote rhodopsin expression in some in vitro assays (Levine et al., 1997,, 2000; Davis et al., 2000). It is not currently known where in the process of photoreceptor specification these factors act, but understanding the regulation of photoreceptor development by soluble factors is important for the derivation of rods and cones from stem cell sources (Box 3).

In addition to these soluble factors, the Notch signaling pathway has also been shown to play a key role in regulating the transition from multipotent progenitor to photoreceptor precursor. Notch signaling among progenitors and postmitotic cells is known to maintain the progenitor pool in multiple neuroepithelia, including the developing retina (Perron and Harris, 2000; Nelson et al., 2007; Xiang, 2013; Cepko, 2014). An unknown process causes a subset of the progenitors to inactivate the Notch signal and initiate the precursor gene expression program (Fig. 3A,B). For photoreceptors, as noted above, one of the earliest signs of this transition is Otx2 expression; inhibition of Notch signaling causes the induction of Otx2 expression in most progenitors of the mouse retina (Jadhav et al., 2006; Nelson et al., 2006,, 2007; Yaron et al., 2006), while conditional deletion of Notch pathway components reduces the number of progenitors and increases the number of both rod and cone photoreceptor precursors (Jadhav et al., 2006; Yaron et al., 2006; Riesenberg et al., 2009; Mizeracka et al., 2013). There are at least three non-exclusive ways that Notch signaling might inhibit photoreceptor genesis: (1) by controlling the fraction of progenitors that exit the cell cycle at any given point during development; (2) by regulating photoreceptor potential (i.e. Otx2 expression); and (3) by influencing fate choice within postmitotic Otx2-expressing precursors. Further studies are required to dissect the mechanisms by which Notch acts in this context, although it should be noted that the ability to control retinal progenitor proliferation and photoreceptor specification using small molecule inhibitors of the Notch pathway has proven useful in several stem cell protocols (Box 3).

Once progenitors leave the cell cycle and stably express Otx2, other factors must influence and stabilize the ultimate pattern of gene expression that is characteristic of photoreceptors. As noted above, photoreceptors are neurons that express a unique set of genes required for phototransduction. The mechanism by which a photoreceptor precursor acquires this unique gene expression profile has been the subject of considerable investigation. This process could occur as the result of: (1) positive acting factors that instruct a fate choice; (2) negative regulators that block competing fate choices; or (3) more likely a combination of both mechanisms.

Downstream of Otx2 is a related transcription factor, Crx, which is necessary for the expression of most photoreceptor genes (Fig. 3B,C) (Chen et al., 1997; Freund et al., 1997; Furukawa et al., 1997). Deletion of Crx in mice does not prevent photoreceptor specification like Otx2 deletion, but instead leads to a dramatic reduction in the expression of photoreceptor genes. A comparison of Crx and Otx2 ChIP-seq datasets shows that their transcriptional targets largely overlap in the retina, although there are also some unique targets (Samuel et al., 2014). Crx is expressed in both rods and cones and, as a result, mutations in this gene can lead to diseases that affect one or both photoreceptor types, such as Leber's congenital amaurosis and cone-rod dystrophy (Box 2) (Freund et al., 1997; Swaroop et al., 1999; Tran and Chen, 2014).

The distinction between rod and cone gene expression profiles occurs downstream (or independently) of Otx2 and Crx, and two key factors that control this specification event are Rorβ and Nrl (Swaroop et al., 2010). Otx2 and Rorβ are needed to initiate the expression of the transcription factor Nrl in a subpopulation of newly postmitotic precursors: these Nrl-expressing cells will develop into rod photoreceptors (Fig. 3C) (Akimoto et al., 2006; Kautzmann et al., 2011; Fu et al., 2014; Roger et al., 2014). Without either Rorb or Nrl, rod photoreceptor cells do not develop; instead, there is an approximately one-to-one fate shift to S-opsin-expressing cone cells (Mears et al., 2001; Fu et al., 2014). Nrl is needed to activate several essential rod-specific genes, including Nr2e3, a transcription factor that can both co-activate rod genes and silence cone genes, such as Opn1sw (S-opsin) (Fig. 3C) (Cheng et al., 2004,, 2006,, 2011; Chen et al., 2005; Peng et al., 2005; Hao et al., 2012). Accordingly, Nr2e3 mutants exhibit excess S-opsin expression, reduced rod gene expression and significant rod defects (Akhmedov et al., 2000; Corbo and Cepko, 2005). When Nrl is overexpressed using a Crx enhancer (which drives expression in photoreceptors and bipolars), only cones, and not bipolar cells, become reprogrammed to a rod fate (Oh et al., 2007). Together, these data argue that achieving a normal rod photoreceptor gene expression profile requires the combination of rod activators (e.g. Nrl, Nr2e3) and a cone repressor (e.g. Nr2e3).

The factors that control the expression of these rod-specification factors are not clear, but are likely to exert their effects after the cells exit from the mitotic cycle in mice. This is because in lineage-tracing studies two-cell clones derived from Olig2⁺, Neurog2⁺ and Ascl1⁺ progenitors sometimes contain two photoreceptors and sometimes a photoreceptor and a non-photoreceptor type (Brzezinski et al., 2011; Hafler et al., 2012). Live imaging experiments in zebrafish show that terminal divisions can generate similar diversity, although there are also lineage-restricted cone and rod progenitors (Godinho et al., 2007; He et al., 2012; Suzuki et al., 2013). These results imply that the specification events that define rod and cone fates are not strictly linked to the status of the cell cycle. The data collected in mice suggest that a bipotential photoreceptor precursor is established shortly after cell cycle exit (Fig. 4A) (Swaroop et al., 2010). In this model, only these precursors can respond to Nrl and adopt rod fate. Several lines of evidence support this bipotent precursor model: (1) some rod and cone marker [i.e. Nrl and Thrβ2 (TRβ2)] overlap has been observed during development (Ng et al., 2011); (2) Nrl expression in developing cones can partially convert cells to a rod identity (Oh et al., 2007); and (3) removing Nrl from adult rods can partially convert them into cones (Montana et al., 2013). An alternative explanation (Fig. 4B) is that another factor causes progenitor cell commitment (see Glossary, Box 1) to rod identity upstream of Nrl, but after cell cycle exit (Emerson et al., 2013). In this model, Nrl would repress cone development and execute rod physiological maturation in photoreceptor precursors. The existence of a specification event prior to Nrl expression would account for the observation that photoreceptors are still specified in Nrl mutants. This would also explain why Nrl cannot reprogram all Crx-expressing cells into rods and why the cones seen in Nrl mutants are not identical (Daniele et al., 2005) to wild-type S-cones. Future work is needed to identify the earliest events in photoreceptor specification and discriminate between these two models.

Although a great deal of progress has been made in understanding the transcriptional network involved in rod photoreceptor specification, much less is known about how cones are committed to their fate. Only a few cone-specific markers have been characterized during embryonic retinal development. One such factor is the thyroid hormone receptor Thrβ2. Cones in mice are divided into blue (short wavelength) or green (medium wavelength) subtypes based on whether S-opsin (Opn1sw) or M-opsin (Opn1mw) is predominantly expressed, respectively. In contrast to human cones that predominantly express one opsin (blue, green or red) per cone, most mouse cones co-express both opsins in opposing dorsal-ventral gradients, such that M-opsin is highest dorsally and S-opsin ventrally (Applebury et al., 2000; Swaroop et al., 2010). Mice deficient for Thrb2 lack M-opsin expression and instead robustly express S-opsin in all cones, indicating a loss of the green cone subtype (Ng et al., 2001; Roberts et al., 2006). However, Thrb2 null mice still generate cones, demonstrating that Thrb2 is not required for commitment to cone identity. Thrb2 is not sufficient for cone genesis either, as its expression in place of the Nrl gene in mice does not promote cone formation or prevent rod development (Ng et al., 2011). The transcription factor Rxrγ (Rxrg) is also expressed in developing cones and ganglion cells (Mori et al., 2001). As with Thrb2 mutants, loss of Rxrg has no effect on the initial formation of cones (Roberts et al., 2005). M-opsin is expressed relatively normally, but S-opsin is robustly expressed by all cones in Rxrg mutants, disrupting subtype identity. Therefore, it appears that green cone subtype development is very similar to rod formation, requiring the action of a positive factor (i.e. Thrb2) to express M-opsin and the negative activities of both Thrb2 and Rxrg to suppress S-opsin and blue cone identity (Fig. 3C).

How then are cones committed to their identity? There are three possibilities. First, and as discussed above, cone commitment might be the default outcome in the retina, requiring only negative regulators, such as Nr2e3 or Notch signaling, to generate non-cone fates (Fig. 4A,C). Second, it is possible that a cone-specific factor triggers commitment to a cone fate and/or suppresses rod fate, analogous to Nrl (Fig. 4D). The identity of such a factor(s) remains unknown. Third, it is possible that the combinatorial action of more widely expressed regulators (i.e. not lineage restricted to photoreceptor fate) promotes cone commitment (Fig. 4E). A potential regulator is the transcription factor Onecut1, which is expressed in subsets of progenitors and Otx2-expressing cells, and later in horizontal and ganglion cells (Wu et al., 2012,, 2013; Emerson et al., 2013). The forced expression of Onecut1 results in the formation of excess cones at the expense of rods (Emerson et al., 2013). It was also shown that Onecut1 and Otx2 synergize to activate Thrb2 expression, suggesting that the combinatorial action of these two factors is necessary for the acquisition of cone potential or cone fate commitment. In mice that lack Onecut1 and its similarly expressed paralog Onecut2, there is an initial lack of Rxrγ-expressing cones, but the number of cones is only modestly reduced by the end of development (Sapkota et al., 2014). In addition, nearly all cones in these mutants express high levels of S-opsin and few cones express M-opsin, similar to Thrb2 mutant mice. Thus, the combinatorial action of Onecut1/2 and Otx2 is not necessary for cone potential or commitment, but is needed for normal M-cone subtype development. Several other transcription factors that intersect early with Otx2, such as Neurod1, Ascl1, Neurog2 and Olig2, have been identified, although these do not prevent cone development when mutated (Tomita et al., 1996; Morrow et al., 1999; Hufnagel et al., 2010; Brzezinski et al., 2011; Hafler et al., 2012). However, it should be noted that mice lacking Neurod1 fail to express Thrb2 and have cone subtype defects (Liu et al., 2008). A model of cone commitment via the intersection of transcription factors and signaling cascades may therefore contain redundant components or compensatory mechanisms. The identification of additional early cone-specific markers and complex multi-factor gain- and loss-of-function studies are needed to distinguish between these three mechanisms of cone commitment.

Multiple experimental paradigms and animal model systems have expanded our knowledge of photoreceptor development, but significant gaps remain. Our understanding of early events in photoreceptor genesis, such as the regulation of potential and cone fate commitment, remains conspicuously incomplete. There is also a significant gap in our understanding of how intrinsic and extrinsic factors combine to regulate photoreceptor development both in vivo and in stem cell-derived cultures. The success of future regenerative therapies, such as cell replacement or endogenously stimulated regeneration, is highly dependent upon gaining a better understanding of photoreceptor fate specification.

Several studies have shown that retinal progenitors are multipotent and that their potential changes over developmental time. These data indicate that progenitors form a heterogeneous population, and recent experiments have shown that this heterogeneity can correlate with differential proliferative and fate outcomes (Brzezinski et al., 2011; Hafler et al., 2012). Furthermore, it has been shown that cells with the potential to form photoreceptors can also have the potential to form horizontal, amacrine and/or bipolar cells (Hafler et al., 2012; He et al., 2012; Brzezinski et al., 2013; Emerson et al., 2013). Does this mean that individual progenitors have the potential for all of these options simultaneously, or are cells subdivided into a series of states that have the potential for one or two fates at a time? In the latter case, progenitors and precursors might exist in a bistable gene expression state where one competing network becomes more stabilized and ‘wins', thereby changing the potential or committing the fate of the cell. It is unclear how many discrete bistable states a cell might experience to become specified as a photoreceptor or whether the path to a stable photoreceptor gene expression network is invariant. Also of importance is how potential is suppressed in progenitors and postmitotic cells. Recent data suggest that Prdm1 restricts bipolar cell potential and that, in the absence of this inhibition, photoreceptor cell fate is mutable (Brzezinski et al., 2013). Does this type of fate plasticity exist more broadly in the retina? Further experiments are needed to test how providing potential and preventing potential restriction alters gene expression networks (i.e. fate outcomes) in both the developing and mature retina.

We have also learned much about rod development and cone subtype choice, but how cells are committed to cone identity is unclear. Although Thrβ2 and Rxrγ are useful for marking developing cones, the earliest marker of an irreversibly committed cone photoreceptor is yet to be identified. Further work is needed to discover whether cones: (1) are the default outcome of not becoming a rod photoreceptor; or (2) are actively committed from cells that have photoreceptor potential. In either case, it is unknown why some cells decide to activate Nrl and adopt rod identity whereas others commit to cone fate. Since the rod-to-cone ratio varies across vertebrate species, comparative developmental studies are likely to help unravel cone fate choice mechanisms.

In recent years, much effort in the field has been focused on generating retinal cells from embryonic and induced pluripotent stem cell sources (Box 3). Success generating retinal progenitors from these sources has been achieved by recapitulating the early events of eye development using small molecules and growth factors (Lamba et al., 2008). The relative ease of using extrinsic factors to program stem cell fate has fueled renewed interest in understanding how these molecules could be used to directly control photoreceptor development. For example, pro-rod factors identified in dissociated culture systems (e.g. taurine and retinoic acid) have been used to promote rod fate and maturation in stem cell-derived cultures (Zhong et al., 2014). Evaluating how extrinsic factors (whether native to the retina or not) control photoreceptor potential, commitment and maturation will facilitate the specific generation of rods and cones from stem cell cultures. Evidence suggests that the maturation state of donor photoreceptors strongly influences transplant success (MacLaren et al., 2006). Thus, using extrinsic factors to achieve a synchronous state of photoreceptor maturation is expected to significantly improve the efficacy of cell replacement approaches.

We thank Tatiana Eliseeva, Grace Randazzo, Jhenya Nahreini, Jason Silver and Santos Franco for critical reading of the manuscript; and Paul Nakamura and Matt Wilken for critical discussion of some of the ideas.