Genetic control of retinal ganglion cell genesis

Abstract

Retinal ganglion cells (RGCs) are the only projection neurons in the neural retina. They receive and integrate visual signals from upstream retinal neurons in the visual circuitry and transmit them to the brain. The function of RGCs is performed by the approximately 40 RGC types projecting to various central brain targets. RGCs are the first cell type to form during retinogenesis. The specification and differentiation of the RGC lineage is a stepwise process; a hierarchical gene regulatory network controlling the RGC lineage has been identified and continues to be elaborated. Recent studies with single-cell transcriptomics have led to unprecedented new insights into their types and developmental trajectory. In this review, we summarize our current understanding of the functions and relationships of the many regulators of the specification and differentiation of the RGC lineage. We emphasize the roles of these key transcription factors and pathways in different developmental steps, including the transition from retinal progenitor cells (RPCs) to RGCs, RGC differentiation, generation of diverse RGC types, and central projection of the RGC axons. We discuss critical issues that remain to be addressed for a comprehensive understanding of these different aspects of RGC genesis and emerging technologies, including single-cell techniques, novel genetic tools and resources, and high-throughput genome editing and screening assays, which can be leveraged in future studies.

Introduction

Retinal ganglion cells (RGCs) are the only type of output neurons in the retina with long axonal projections to the brain [1]. In the visual circuitry, RGCs receive visual signals from photoreceptors via interneurons (bipolar and amacrine cells) and send them to the rest of the brain through the optic nerve [2, 3]. The various aspects of vision, including color, intensity, contrast, and movement detection, are all coded through the many types of RGCs [4, 5]. According to the most recent estimate, more than 40 types of RGCs exist [6–8]. The different RGC types respond to different features of the visual cues and relay them to the brain to achieve cohesive visual perception. RGC types also include intrinsically photosensitive RGCs (ipRGCs), which respond to light directly to carry out both non-image forming functions, such as pupillary light reflex and circadian photoentrainment, and imaging forming functions [9, 10]. Individual types have distinct morphologies, as manifested by the size and shape of their receptive field, dendrite density, projection depth, electrophysiological characteristics, and gene expression profiles [3, 5–8]. An understanding of how RGCs and their types form during development is critical for understanding how visual circuits form during development and for designing regenerative therapeutic measures to treat many eye diseases in which RGCs are implicated, such as glaucoma.

All retinal cell types originate from naïve multipotent retinal progenitor cells (RPCs) in the embryonic retina [11]. RPCs are characterized by active proliferation [12, 13] and the expression of a group of pan-retinal transcription factors, including Pax6, Rax, Sox2, Six3/Six6, Vsx2 (also known as Chx10), Lhx2, Nr2e1, and components of the Notch pathway [14–22]. Generally, these transcription factors function to maintain the properties of RPCs and regulate the balance of proliferation and differentiation, although each factor plays specific roles in development. They form a complex network with significant cross-regulation, feedback, and feedforward processes through which RPC competence for differentiation into multiple cell lineages is established and the downstream regulatory pathways underlying all retinal cell types, including RGCs, are activated [23, 24].

As the earliest born retinal neurons, RGCs emerge from multipotent RPCs that are located in the neuroblast layer (NBL) of the embryonic retina [11, 12, 14, 25, 26] (Fig. 1a). The somas of proliferating naïve RPCs oscillate through the thickness of the NBL throughout the cell cycle via a process called interkinetic nuclear migration (INM) [12]. For differentiation to occur, a subset of RPCs first becomes competent (transitional) RPCs that are poised to exit the cell cycle. Despite the many types, RGCs are all generated via a common genetic mechanism, since mutation of a single transcription factor gene, Atoh7, almost completely abolishes all RGCs [27, 28]. All newly born RGCs emerge from competent RPCs expressing Atoh7 [29–31], and then migrate to the inner side of the retina to form the future ganglion cell layer (GCL) [26] (Fig. 1a, b). Newly born RGCs further differentiate into functional neurons by extending dendritic and axonal projections and forming synapses as development progresses. In the mouse, RGC genesis spans a time window of approximately ten days, with the first RGCs appearing in the central retina at E11, propagating peripherally, reaching a peak at E14.5, and tapering off until just after birth [13, 32] (Fig. 1c). All these aspects of RGC development are tightly controlled genetically. Since the first major regulator, Pou4f2 (also known as Brn3b), of RGC development was identified more than two decades ago [33–35], substantial progress has been achieved, and a general framework for the genetic network controlling RGC genesis has been established [27, 28, 36–41]. The stage-specific functions of many regulators are often reflected by the timing of their expression along the developmental trajectory of the RGC lineage, which is traditionally revealed by in situ hybridization and immunohistochemistry. Recent single-cell transcriptomics studies have further unequivocally demarcated the distinct stages of the RGC developmental trajectory with unique expression signatures [42, 43]; these stages include naïve RPCs, transitional RPCs, RGC precursors (early), and differentiating (late) RGCs [43]. We thus organize our discussion of the key regulators according to their functions in these different stages.

Fig. 1.

The spatial and temporal genesis of retinal ganglion cells in the mouse retina. a. Diagram illustrating the emergence of RGCs from naïve retinal progenitor cells (nRPCs). nRPCs divide and oscillate their nuclei across the retina throughout the cell cycle. Some become transitional RPCs (tRPCs), exit the cell cycle, and adopt one of the retinal cell fates (e.g., RGCs). Fate-committed RGCs migrate to the inner side of the retina to form the future ganglion cell layer. b. Two stages of RGC differentiation are marked by two transcription factors, Atoh7 and Pou4f2, as evidenced by immunofluorescence staining. Atoh7 marks the tRPCs located in the outer part of the retina, and Pou4f2 marks fate-committed RGCs, which are born in the outer part and then migrate to the inner side. Newly born RGCs are still located in the outer part and express both markers. Once they reach the inner side, they lose the expression of Atoh7. c. Temporal genesis of RGCs, as marked by Pou4f2. At E12.5, RGC genesis is only beginning in the central retina. At E14.5, RGC genesis spreads to the periphery and reaches a peak. At E17.5, RGC genesis has largely stopped, as very few newly born RGCs located on the outer side of the retina are observed. The images in b and c represent previously published findings [31]

Establishing competence in RPCs for the RGC fate

The first step toward the RGC fate is to establish a competent state for lineage-specific differentiation. The bHLH transcription factor Atoh7 plays a critical role in this step, but other factors likely also participate in the process. Atoh7 is a vertebrate orthologue of the Drosophila gene Atonal, which is a member of the so-called proneural gene family [44, 45]. Atoh7 is expressed in a subset of RPCs from E11 to P0, with a peak at around E14.5 (Fig. 1b). The temporal expression of Atoh7 coincides with RGC genesis [13, 31, 46]. Mutations of Atoh7 or its enhancers lead to an almost complete loss of RGCs in zebrafish, mice, and humans [27, 28, 47–49]. Thus, Atoh7 occupies a critical position in the gene regulatory network controlling RGC genesis; the expression of key downstream genes for RGC differentiation, such as Pou4f2 and Isl1, depends on Atoh7 [39, 43, 50, 51]. However, Atoh7 is not sufficient to drive RPCs to the RGC fate, since Atoh7-expressing RPCs adopt essentially all retinal cell fates [29, 30, 52]. Thus, the function of Atoh7 is to confer RPCs competence for RGC fate rather than directly specify it. The majority of Atoh7-expressing RPCs, if not all, are at the last cell cycle [53]; likely, Atoh7 also functions to promote cell cycle exit [53–55]. Although many downstream genes have been identified [39, 43, 50, 51], the mechanism by which Atoh7 exerts its functions and activates these genes remains unknown.

The SoxC class of transcription factors, including Sox4, Sox11, and Sox12, likely plays key roles in establishing competence for the RGC lineage as well [37]. All three transcription factors are expressed in RGCs during development [41, 56–58]. Although discrepancies have been reported in the literature [37, 56, 58], we have unequivocally shown that Sox4 and Sox11 are also expressed in a subset of RPCs [43]. Supporting evidence for these factors playing key roles in establishing the RGC lineage at the same stage where Atoh7 functions is the severely defective RGC genesis in their knockouts. Although single knockout of individual SoxC genes results in either mild or no phenotypes, Sox4/Sox11 double knockout and Sox4/11/12 triple knockout retinas display severely diminished RGC genesis both in vivo and in cultured explants [37, 56, 59]. Thus, these three transcription factors have redundant functions in retinal development, with Sox4 and Sox11 being the major players. Consistently, the expression of RGC genes controlling RGC fate, such as Pou4f2 and Isl1, depends on SoxC transcription factors. Therefore, SoxC transcription factors play key roles in the early stage of RGC fate specification, possibly in parallel with Atoh7.

Since Atoh7 and SoxC transcription factors do not depend on each other [37, 43], they likely provide independent regulatory inputs to collaboratively activate genes such as Pou4f2 and Isl1, which are essential for RGC differentiation. Other transcription factors may function at this transitionary stage as well. These factors include NeuroD1, Neurog2 (Neurogenin2), and Ascl1. NeuroD1 may possess some, albeit weak, intrinsic property to promote the differentiation into the RGC lineage, but its role in normal RGC development is likely minimal [60]. Neurogenin2 and Ascl1, both of which are proneural transcription factors, function to facilitate the propagation of the initial wave of RGC genesis, but their functions are also minimal in overall RGC genesis [61, 62].

The Notch pathway plays key roles in the transition from naïve RPCs to competent RPCs. The canonical Notch pathway is activated when the Notch extracellular domain (NECD) binds to Delta-Serrate-LAG2 (DSL) ligands that are expressed in the membrane of neighboring cells. After two proteolytic cleavage steps, the Notch intracellular domain (NICD) translocates into the nucleus and, together with cofactors RBPJ (CSL) and MAML1, activates the transcription of target genes, such as Hes1 and Hes5. In contrast, the expression of ligands and Notch receptors on the same cell results in the cis inhibition of Notch signals and degradation of the receptor [63–65]. The Notch pathway plays pleiotropic roles in the retina; multiple Notch ligands, receptors, and downstream genes are expressed in the developing retina, and individual genes are involved in distinct lineages [66–75]. Nevertheless, the Notch pathway is critical for balancing proliferation and differentiation [14, 68, 74–77]. The pathway functions by interacting with both upstream pan-retinal transcription factors and downstream factors functioning in specific retinal lineages [20, 56, 69, 78]. For example, Notch1 expression requires Sox2, and it is likely a direct target gene of Sox2 [20]. On the other hand, downregulation of the Notch pathway is the first step toward differentiation, which is mediated by lineage-specific factors such as Atoh7, Ascl1, Ptf1a, and Foxn4 [71, 76]. The mechanism by which a subpopulation of RPCs is selectively released from Notch signaling is unclear, but classical lateral inhibition likely plays a major role [65, 69, 76, 79].

As mentioned above, recent single-cell RNA-seq studies have provided an unprecedented new perspective on the nature of RPC competence. Leveraging the power of this technology to profile the transcriptomes of thousands of retinal cells, the cellular states along individual retinal lineage trajectories have been identified and characterized [42, 42, 43, 80]. These studies on both developing human and mouse retinas identified a state, namely, transitional or neurogenic RPCs, shared by all retinal cell fates [42, 43, 80]. A significant finding is that at least during early retinal development, transitional RPCs co-express genes involved in multiple lineages, e.g., Atoh7 and the SoxC genes for RGCs and Neurod1 and Otx2 for photoreceptors, with Atoh7 expressed in almost all transitional RPCs. Furthermore, these cells have downregulated Notch signaling and upregulated the expression of Notch ligands and, are exiting the cell cycle, and are ready to assume one of the retinal cell fates. These findings indicate that the competence of retinal progenitors for different fates is determined by the expression of lineage-specific transcription factors such as Atoh7 at a particular developmental stage and that the eventual fate is determined stochastically by these available transcription factors and their relative activities. This hypothesis is supported by results from previous clonal analysis [81–83] and by findings that the levels of Atoh7 influence the numbers of RGCs being produced [55, 84]. Moreover, and somewhat surprisingly, two studies revealed that the RGC lineage still forms in the Atoh7-null retina, but the RGC precursors largely fail to reach the more differentiated stages and are eventually lost [43, 85]. When apoptosis is inhibited, these cells persist through postnatal stages, form circuits with upstream neurons, and respond to light stimuli. However, they are not normal since they fail to connect with the central targets in the brain. The underlying reason for the stalled progression of the RGC lineage is likely the insufficient or even failed expression of many RGC-specific genes and ectopic expression of many non-RGC genes. These results further support the hypothesis that additional factor(s) such as the SoxC family in transitional RPCs promote them to differentiate toward the RGC lineage. Thus, activation of early RGC genes such as Pou4f2 and Isl1 likely requires both upstream inputs, but in the absence of Atoh7, the SoxC factors still activate some of the RGC genes, although often at reduced levels.

Early RGC-specific transcription factors: determining the RGC fate

The commitment of RPCs to the RGC fate marks the no-return point of the RGC developmental trajectory. RGC fate specification coordinates with cell cycle exit, migration to the inner retinal region, and activation of the gene expression program for RGC maturation and function [13, 26, 41]. Pou4f2, a POU homeodomain factor, and Isl1, a Lim homeodomain factor, are critical transcription factors functioning at this critical point of RGC fate specification [32–34, 36, 38–41, 86–88], although other transcription factors likely play key roles as well.

Pou4f2 and Isl1 are the earliest genes activated in RGC precursors. They overlap transiently with and are dependent on Atoh7 but continue to be expressed in RGCs throughout life when Atoh7 is no longer expressed [31, 39, 40, 51]. Knockout of Pou4f2 and Isl1 leads to similar defects in the RGC lineage; although RGCs form, they assume a hybrid cell identity, are defective in differentiation, and most of them die by P0 [26, 32–34, 39, 40, 89]. The mechanism underlying these defects is that Pou4f2 and Isl1 interact and regulate both overlapping and specific sets of downstream genes [36, 39, 40, 87]. They achieve this regulation by activating genes required for RGC differentiation and repressing genes required for other cell lineages [39, 88]. More importantly, when Pou4f2 and Isl1 are expressed ectopically in place of Atoh7 in the Atoh7-null retina, RGC formation is rescued in almost all aspects of RGC differentiation, including cell cycle exit, migration, RGC-specific gene expression, survival, and physiological function [41]. This last experiment shows that Pou4f2 and Isl1 are two critical transcription factors not only for RGC maturation but also for the initial fate specification at the transition stage from competent RPCs to fate-committed RGC precursors.

Based on these findings, a model of the gene regulatory cascade from Atoh7 to Pou4f2 and Isl1 in RGC fate determination and differentiation has been postulated [41]. In this model, the function of Atoh7 and other factors (e.g., SoxC factors) functioning in transitional RPCs is to activate a core group of RGC fate-determining transcription factor genes, including but not limited to, Pou4f2 and Isl1, in a subset of competent RPCs. Once this core group of genes is activated, the encoded transcription factors sustain their own expression by cross-regulation and/or autoregulation and no longer rely on upstream activators. This core group of transcription factors then activates the general genetic program required for all aspects of RGC differentiation and likely maintenance. The finding that endogenous Isl1 and Pou4f2 are activated by ectopic Pou4f2 and Isl1 and that their expression continues even after the ectopic expression is turned off supports this model [41]. The core group likely includes other factors, since some RGC-specific genes are not significantly affected, even in the Pou4f2-null, Isl1-null, or the double-null retina [38, 41, 51, 87]. Nevertheless, even these genes are rescued by ectopic Pou4f2 and Isl1 expression in the Atoh7-null retina, further supporting the hypothesis of cross-regulation among this core group of genes.

Currently, the identities of the other core RGC-determining factors are unknown. Dlx1 and Dlx2 also exert modulatory functions in the early stages of RGC development. They depend on Atoh7 and are expressed at the transition stage when RGC fate is determined [90] but are repressed by Pou4f2 and Isl1 once the RGCs have migrated to the inner side [38, 41, 43, 91]. Although double knockout of Dlx1 and Dlx2 leads to only a moderate loss of late-born RGCs [90], Dlx1/2/Pou4f2 triple knockout results in an almost complete failure of RGC genesis at early stages [92]. In addition, although Pou4f2 and Isl1 repress Dlx1 and Dlx2 expression, Dlx1 and Dlx2 promote the expression of Pou4f2 and Isl1 [92]. Noticeably, Dlx1 and Dlx2 also promote amacrine cell and photoreceptor production, but this ability is inhibited by Pou4f2 through a physical interaction [91]. These results strongly suggest that Dlx1 and 2 are part of the regulatory core factors specifying RGC fate. Dlx1/2 are likely involved in specifying multiple lineages through cross-regulation with Pou4f2 and probably Isl1; they collaborate with Pou4f2 to promote the RGC fate but promote other fates in the absence of Pou4f2.

SoxC factors may also be part of the core group of RGC-determining transcription factors. Although we propose that the SoxC factors function at the level of Atoh7 in conferring RGC competence to RPCs, they likely also participate in determining RGC fate, since they continue to be expressed at high levels in fate-committed RGC precursors and RGCs [37, 41, 43, 56]. Thus, SoxC transcription factors may function at multiple stages along the developmental trajectory of the RGC lineage and may be subject to different modes of regulation at these stages.

The recent finding that the RGC lineage still forms in Atoh7-null retinas does not refute the key roles of Pou4f1 and Isl1 in specifying RGC fate but rather indicates that some key regulators independent of Atoh7 exist. RGC fate specification is equivalent to activation of the gene expression program essential for RGC structure and function, and each gene receives a unique combination of upstream inputs. Some of these genes depend on the Atoh7 → Pou4f2/Isl1 regulatory cascade, whereas others do not. However, most genes likely require both Atoh7-dependent and Atoh7-independent upstream inputs to reach the full levels of expression critical for RGC differentiation and survival.

Transcription factors required for RGC differentiation

Once the RGC fate is determined, the RGC precursors embark on the process of differentiation and maturation. This process entails the activation of genes encoding proteins essential for the structure and function of RGCs, which is likely mediated by the many transcription factors specifically expressed in RGCs. The fate-determining core factors participate in regulating these downstream genes, either directly or by regulating other downstream transcription factor genes. Many of these downstream factors likely regulate specific aspects of RGC differentiation, as knockout of their genes tends to result in moderate or mild defects. Additionally, many of these factors are members of transcription factor families functioning redundantly, further confounding the tasks of discerning their roles. Here, we try to provide as extensive a list as possible of these factors. Although the current understanding of their functions in RGCs is often scant, we hope the compiled information will inspire further investigations. The formation of individual RGC types is part of the differentiation process, but we will separately discuss this process in the next section.

Pou4f2, Pou4f1, and Pou4f3: In addition to Pou4f2, the two other class IV POU transcription factors, Pou4f1 and Pou4f3, are also expressed in RGCs during development [35], but they are dependent on Pou4f2, and their expression occurs at later stages after RGCs have migrated to the future GCL [33, 34, 38, 93]. Initial analyses of the Pou4f1 and Pou4f3 mutant retinas suggest that they are not critical for RGC development [33, 94–96], but further studies show that knockout of Pou4f1 leads to the loss of a distinct population (20%) of RGCs with small dense dendritic arbors [86, 97] and that redundancy exists among the three factors, as Pou4f1/2 and Pou4f2/3 double knockout retinas manifest a more severe RGC loss than Pou4f2 knockout retinas [97, 98]. Knockin and ectopic experiments indicate that all three members are capable of promoting RGC genesis and have similar, if not identical, intrinsic biochemical properties in DNA binding and in promoting RGC genesis [93, 99–101]. On the other hand, the three transcription factors are expressed in different RGC subpopulations, particularly in the postnatal retina, although with significant overlaps [7, 86, 102–105]. Furthermore, the deletions of Pou4f1 and Pou4f3 differentially affect RGC dendritic morphology, suggesting that these factors function in shaping the identities of RGC types. Interestingly, these factors also mark distinct ipRGC subtypes [97, 106–108], although how they are involved in their formation is not known. Nevertheless, expression profiling found very few genes with altered expression in the Pou4f1-null retina [109], likely due to redundancy among these factors. One major outstanding issue is whether each of the three factors has unique properties in promoting RGC differentiation, or whether they have equivalent intrinsic properties and the phenotypes are largely due to the total cumulative activities of these proteins in the cell.

Klf7 and Klf4: Both proteins belong to the Krüppel-like zinc finger factor family. This family of transcription factors normally functions as transcriptional repressors [110]. Klf7 is expressed immediately after RGCs are born. Knockout of Klf7 results in aberrant RGC axon projections but no overt change in RGC numbers, indicating that Klf7 functions in the pathfinding of RGC axons [111]. Klf7 is regulated by both Pou4f2 and Isl1 [38, 39, 43], further supporting the hypothesis that it is part of the gene regulatory network governing RGC differentiation. In contrast, Klf4 is expressed in developing RGCs at much lower levels, but its expression increases significantly in postnatal RGCs [112]. Knockout of Klf4 results in no overt RGC defects [113, 114]. Nevertheless, Klf4 represses RGC axon regeneration, although Klf7 instead promotes RGC axon growth [111, 112, 115]. Thus, the two related transcription factors function differently in RGC development and maintenance.

Barhl2: Barhl2 is a member of the Barh family and is expressed in RGCs, amacrine cells, and horizontal cells [116]. Studies in both mice and zebrafish indicate that Barhl2 regulates type specification of amacrine cells [116–118]. Barhl2 is also critical for the development of a normal complement of RGCs [117]. In the RGC lineage, Barhl2 is downstream of Atoh7 and Pou4f2 and is expressed in a subset of RGCs [43, 117]. In Barhl2-null mice, no change was observed in the initial generation of RGCs, but 35% of RGCs were subsequently lost due to apoptosis, suggesting that Barhl2 is dispensable for RGC genesis but necessary for the survival of a subset of RGCs. The function of Barhl2 is cell autonomous; only cells normally expressing Barhl2 die when Barhl2 is deleted. How Barhl2 exerts its functions, its target genes, and the properties of Bahl2-expressing RGCs are all important questions to be addressed.

Onecut transcription factors: The Onecut family of transcription factors is characterized by a ‘cut’ domain and an atypical homeobox domain, and both domains are involved in DNA binding [119]. Three Onecut factors have been identified in the mouse, all of which are expressed in the retina, but Onecut1 and Onecut2 play major roles since Onecut3 is expressed at much lower levels. Onecut1 (also known as Hnf6) and Onecut2 are expressed in retinal progenitor cells, developing RGCs, and horizontal cells [120]. Onecut factors function redundantly in the retina, since the single knockout of Onecut1 or Onecut2 results in only a partial loss of horizontal cells [121–124], but in Onecut1/Onecut2 double knockouts, the mutant retina exhibits more profound defects in all early retinal cell types, including reduced production (by 30%) of RGCs [123]. Since only a subset of RGCs is affected, the Onecut factors are likely involved in the differentiation of specific RGC types, although their type identities have not been determined. Consistent with this hypothesis, Onecut1 and Onecut2 levels vary significantly among RGCs, and they activate some type-specific markers but repress others [120, 123]. Since their expression does not require Atoh7, Isl1, or Pou4f2, Onecut1 and Onecut2 are likely regulated by other upstream regulators [43, 120].

Early B-cell factors (Ebfs): Ebfs are characterized by a non-basic HLH dimerization domain and an atypical DNA-binding domain [125]. Four Ebf genes (Ebf1-4) are present in the mouse genome, all of which are expressed in multiple retinal cell types, including RGCs, a subset of amacrine cells, a subset of bipolar cells, and horizontal cells [126]. The expression of all four genes in RGCs, but not other cell types, is regulated by Pou4f2 and Isl1 [36, 38–40, 43, 87]. Knockout of Ebf1 leads to only mild defects and axonal projection errors in RGCs [127]. However, ectopic expression of a dominant negative form of Ebf or repression of all Ebfs by RNAi results in defects in all the cell types expressing them, including RGCs, suggesting redundancy among these members [126]. Ebf1 and Ebf3 may play major roles, since they are expressed at much higher levels than Ebf2 and Ebf4 [43]. A better understanding of how the Ebf factors regulate RGC development will require the inactivation of at least both Ebf1 and Ebf3 specifically in RGCs during development.

Iroquois-like homeobox (Irx) factors: Iroquois-like homeobox (Irx) factors are encoded by genes homologous to the Drosophila gene Iroquois; they are characterized by a 63 a.a. TALE family homeodomain and a 9 a.a. conserved motif outside of the homeodomain known as the Irx box [128]. Six Irx genes in the mouse are clustered in two loci (Irx1, 2, and 4 on chromosome 13 and Irx3, 5, and 6 on chromosome 8). All six members are expressed in developing RGCs and are downstream of Atoh7, Pou4f2, and Isl1, indicating that they play important roles, but likely in a redundant manner [36, 129–132]. Although several Irx genes have been studied in the retina, very little is known about their function in RGCs. Moreover, Irx genes are also expressed and function in other retinal cell types, further complicating the situation. Although knockout of either Irx6 or Irx5 leads to defects in the specification of bipolar cells, no effects on RGCs have been reported [133–135]. Knockout of Irx2 also does not cause overt RGC defects [136]. However, a role for Irx transcription factors in RGCs was supported by a study of Irx4 in the chicken retina, which showed that Irx4 inhibits RGC axon outgrowth by repressing Slit1 [137]. Inactivation of multiple Irx genes will be needed to delineate their functions in RGCs.

Myt1 and Myt1l: Myt1 and Myt1l are related C2H2 zinc finger transcription factors expressed extensively in the neural system, including RGCs, although Myt1 is also expressed in RGC precursors [138, 139]. Myt1l, together with other transcription factors, including Pou3f2 and Ascl1, is capable of reprogramming fibroblasts into neuronal cells [140], indicating that it plays a major role in neuronal differentiation. Early studies in Xenopus indicated that Myt1 can overcome the inhibition of bHLH genes such as Atoh7 by the Notch pathway and promote neurogenesis, including RGC formation [141, 142]. However, the exact roles of Myt1 and Myt1l in the RGC lineage have not been definitively established. Given their roles in other parts of the central nervous system these two related transcription factors are plausibly required for some aspects of RGC differentiation.

Pou6f2: Pou6f2 (formerly known as RPF-1) is one of the earliest identified RGC-specific transcription factors [143]. Its expression is highly dependent on Pou4f2, Isl1, and Onecut1/2 [38, 39, 43, 87, 123]. Pou6f2 is expressed in all RGCs during development but in only a subset in the mature retina [143, 144]. However, no overt defects were observed when Pou6f2 was knocked out (http://www.informatics.jax.org/marker/phenotypes/MGI:2443631). Interestingly, single nucleotide polymorphisms (SNPs) associated with Pou6f2 are linked to glaucoma, a disease characterized by RGC death, in humans [144]. In the postnatal mouse, the subset of RGCs expressing Pou6f2 included the ON–OFF directionally selective RGCs, which are more sensitive to death in the DBA/2 J glaucoma model as well as induced by axon injury [144, 145]. All these findings indicate that Pou6f2 functions in RGCs.

In addition to the transcription factors mentioned above, many others are also specifically expressed in RGCs. For example, Lhx9, a member of the LIM homeodomain (LIM-HD) family of transcription factors, is expressed in RGCs during the early stages of development [146]. Rxrg is expressed in both cones and RGCs [43, 123]. In addition to its roles in RPCs, Pax6 is extensively expressed in RGCs beginning early in development [46]. Large-scale in situ hybridization databases such as Eurexpress Atlas [147] and the Allen Developing Mouse Brain Atlas [148], as well as global gene expression profiling [7, 8, 42, 43, 109, 149], particularly using single-cell transcriptomics, have identified many additional transcription factors expressed in RGCs, including Groucho/Tle factors (1–6), Zfhx2-4, Nhlh2, Nkx6-2, Thra, and Pbx3. The highly RGC-specific expression of these factors indicates that they all likely regulate some aspects of RGC differentiation.

Generation of diverse RGC types

As output neurons, RGCs are tasked to relay all aspects of visual signals, such as color, movement, and contrast, as well as the non-image forming light inputs, to the brain. These tasks are carried out by a remarkably diverse set of RGC types. RGCs are traditionally categorized based on morphology and electrophysiology, which rely on their dendritic arbor shapes and responses to different light stimuli [4, 5]. Several tour de force studies using lipophilic dye labeling, genetic labeling, serial electron microscopy, and large-scale recording have provided a general picture of RGC diversity [150–154], and efforts are being made to integrate the different aspects associated with individual RGC types (http://rgctypes.org/). The classification based on these methods proves reliable and realistic, as distinct physiological properties are often associated with specific morphologies. However, for many years, molecular and genetic markers, which are essential for studying their development, have been absent. Transgenic mouse lines in which RGC types are labeled with fluorescent proteins, either directly or indirectly by Cre-mediated recombination, have proven to be very useful tools to study their properties, but they cover only a limited set of RGC types [5, 6]. Nevertheless, an increasing number of marker genes for specific RGC types are being identified [103, 105, 155]. Single-cell transcriptomics has enabled a comprehensive account of RGC types at the molecular level; more than 40 types of RGCs, each with a unique gene signature, were recently recognized, although the physiological properties of many of them remain to be elucidated [7, 8].

Despite this progress, we are only beginning to unravel the mechanisms controlling the development of diverse RGC types. A critical issue that remains is whether all types arise from a generic RGC population or are destined already when they are born. One of the RGC types, ON–OFF direction-selective RGCs that are responsive to vertical motion and express Cdh6, arises from RPCs already expressing Cdh6 [156], but it is not clear whether such a scenario is the rule or an exception. The fact that the vast majority of RGCs are subject to control by Atoh7, Pou4f2, and Isl1 indicates that the former scenario is the case. However, many type-specific transcription factors, e.g., Tbr2 and Isl2, are already expressed in only subsets of RGCs at very early stages of retinal development [157, 158]. These facts argue for both shared and specific gene regulatory programs underlying the formation of different RGC types. Additionally, the timing of the birth of most RGC types is completely unknown. Nevertheless, several known transcription factors regulating specific RGC types have been identified recently. These studies are beginning to provide important insights into the genetic mechanisms controlling RGC type specification.

Tbr2 (Eomes): Tbr2 is a T-box transcription factor. In the retina, it is expressed in a small subset of RGCs immediately after they have migrated to the inner side [157]. Tbr2 is downstream of Pou4f2 and Isl1; thus, Tbr2-expressing RGCs are subject to the general regulation of the RGC gene regulatory network [39, 157]. In the mature retina, Tbr2 marks several non-overlapping RGC types, including all ipRGCs, all Cdh3-GFP-labeled RGCs, and the majority of Unc5d + (Unc-5 homolog D-expressing) RGCs, which send axons to both image forming and non-image forming areas, and subsets of displaced amacrine cells [105, 159]. Initially knockout of Tbr2 was shown to lead to the loss of only a subset of RGCs [157]. Further analyses indicate that Tbr2 is required for the genesis, differentiation, and survival of all ipRGC subtypes [105, 159]. One of the downstream genes of Tbr2 is Opn4. Unlike Tbr2, ipRGCs still form when Opn4 is knocked out, suggesting that Tbr2 regulates additional downstream genes [105]. Also, definitive roles of Tbr2 in non ipRGCs and displaced amacrined cells remain to be established.

Tbr1: Tbr1 is another T-box-related transcription factor expressed in the retina. It is exclusively expressed in subsets of RGCs [109]. Independent studies on its roles in RGC types were reported by two groups [103, 104]. Although the findings from these two studies differ in several details, the consensus of the two studies is that Tbr1 is expressed in several distinct OFF RGC types, including J- and α-OFF-s-RGCs, with all their dendrites projecting to the OFF substrata of the inner plexiform layer (IPL), and one major function of Tbr1 is to specify the dendritic projections. The study by Liu et al. suggests that Tbr1 functions in individual types by regulating different downstream target genes. The study by Kiyama et al. revealed that although Tbr1 + RGCs are born early (before E14.5), Tbr1 is activated much later in these RGCs but is essential for their differentiation and survival. Further studies are necessary to identify the complete set of target genes expressed in the different Tbr1 + RGC types and understand how they are differentially regulated in different RGC types. How Tbr1 itself is activated also portends great interest and significance in understanding the general rules governing RGC type differentiation.

Satb1: Satb1 (special AT-rich sequence-binding protein 1) is a homeodomain transcription factor that is selectively expressed in ON–OFF direction-selective ganglion cells (ooDSGCs) [160]. Mutation of Satb1 leads to the specific loss of the ON arbors and thereby the ON response, but not the OFF arbors and OFF response, of ooDSGCs. Furthermore, Satb1 is essential for the stabilization of ooDSGC ON arbors instead of their initial formation. In the Satb1 mutant retina, the number of ooDSGCs, their mosaic distribution, their central projection to the brain, and the overall retinal organization were not affected, indicating that Satb1 plays highly specific roles. Satb1 carries out this function in part by regulating Contactin 5 (Cntn5), which in turn mediates the ooDSGC interaction with ON starburst amacrine cells through hemophilic contact.

Isl2: Isl2 is expressed in a subset (30–40%) of RGCs throughout development [161]. It is closely related to Isl1 but is expressed at a later stage; thus, unlike Isl1, it does not have a role in the initial RGC fate specification. It is downstream of Pou4f2 and Isl1 [39, 40, 43, 93]. Knockout of Isl2 leads to an expansion of ipsilateral RGCs, suggesting that it has a role in repressing the genetic program underlying ipsilateral projections (see the additional discussion below). However, this repression may not be the only function of Isl2, since most RGCs in the Isl2 knockout retina remain contralateral. In fact, similar to Isl1, Isl2 also forms a complex with Pou4f2 [87], indicating that Isl2 has some redundant functions with Isl1. Genetic labeling indicates that Isl2-positive RGCs have distinct dendritic stratification patterns in the IPL and central projection targets, and most αRGCs belong to this group. Based on these results, Isl2 + RGCs belong to a distinct functional group, although additional investigations are required to understand their physiological properties and the roles of Isl2 in their formation and function.

Clearly, with more than 40 RGC types present and only a very limited number of them studied to date, we have just scratched the surface regarding the genetic control of RGC diversity. Nevertheless, several general principles have emerged from the few types already studied. Individual types may be specified at distinct developmental time points. Each type is likely regulated by a combinatorial code of transcription factors. Conversely, single transcription factors may participate in regulating the development of multiple types, and they may play a broad role, such as specification and survival, or a very specific role, such as instructing dendritic projections. Many additional transcription factors marking only subsets of RGCs, such as Gfi1, FoxP2, Prdm16, and Tfap2d, have been identified [162–166], and even more have been reported from transcriptomic studies [7, 8, 109, 109], but virtually nothing is known about their roles in RGCs. As mentioned above, some of the transcription factors involved in overall RGC genesis in early retinal development may also participate in the differentiation of specific RGC types at later stages. For example, Isl1, but not the Pou4f factors, is required for the differentiation of ipRGCs [97]. An understanding of how different RGC types are specified is likely a major focus in the field of retinal development in the foreseeable future, although the combinatorial nature of transcription factors regulating individual types may pose a major challenge in this endeavor.

Gene regulation of RGC central projections

RGC axons project to as many as 40 target sites in the brain [6]. Two aspects contribute to the central projection of RGC axons: laterality and target specificity. In mice, the vast majority (95%) of RGC axons project to the opposite side of the brain (contralateral), and the remaining 5% project to the same side (ipsilateral) [167]. On the other hand, although many RGC types have preferred brain targets [6, 103, 105, 106, 155, 168, 169], the central projection patterns for most of them are unknown. Much has been learned about the genetic pathways controlling the laterality of RGC axon projections, but essentially no knowledge is available regarding how individual RGC types select their central target(s).

Mouse ipsilateral RGCs reside in the ventrotemporal crescent (VTC) of the retina [167]. Intrinsic differences exist between ipsi- and contralateral RGCs. They are generated on distinct time courses and possess unique expression profiles [56, 170–172]. These differences are likely dictated by key regulators that control the laterality of RGC axon projections, including Zic2, Isl2, and SoxC factors. Zic2, a zinc finger transcription factor, is expressed exclusively in RGCs in the VTC zone [173]. Loss- and gain-of-function analysis revealed that Zic2 is both necessary and sufficient for specification of ipsilateral RGCs. Zic2 functions in specifying ipsilateral RGCs by regulating molecules involved in axon pathfinding, such as EphB1 [174]. Isl2, on the other hand, is expressed only in a subset of contralateral RGCs across the retina, including in the VTC, but not in ipsilateral RGCs [158, 161]. Thus, Isl2 and Zic2 mark two distinct RGC populations. Although Isl2 may have roles in RGC type development, the function of Isl2 in regulating projection laterality of RGC axons is to restrict ipsilateral RGCs, likely by directly repressing Zic2 and/or EphB1. Nevertheless, Isl2 is unlikely to have a dominantly instructive role in directing contralateral RGC projections, since it is only expressed in one-third of contralateral RGCs and most RGCs outside of the VTC remain contralateral even when Isl2 is knocked out. SoxC family members also play a role in the differentiation of contralateral RGCs but not ipsilateral RGCs or their axon growth and midline crossing at the chiasm [56]. SoxC factors carry out these two functions by antagonizing the Notch pathway and by regulating Plexna1 (encoding Plexin-A1) and Nrcam (encoding a neural cell adhesion molecule, NrCAM), respectively. However, their roles do not appear to be instructive either, as only a small percentage of RGCs from the non-VTC region project ipsilaterally when the SoxC genes are ablated [56].

Notably, ipsilateral RGCs have been largely overlooked in most studies of RGC development in the mouse, likely due to their small population in the mouse and the narrow time window of their genesis. Most studies have not distinguished these two populations. In that sense, the several published studies on the SoxC genes were largely consistent and point to a role for the SoxC transcription factors in RGC genesis, except that some of these studies did not emphasize contralateral RGCs [37, 56, 59]. However, it remains unclear whether the general gene regulatory pathway plays equivalent roles in ipsi- and contralateral RGCs.

Extrinsic signaling pathways influencing RGC formation

Multiple signaling pathways are involved in retinal development. The Notch pathway, as discussed above, functions at the RPC level to negatively affect RGC production. Three other pathways, the hedgehog pathway, the BMP pathway, and the Vegf pathway, also play definitive roles in RGC formation. Shh from RGCs acts on RPCs through its receptors Patched1 and 2 [175]. Knockout of Shh leads to reduced RPC proliferation and increased RGC genesis [176]. The importance of Shh was further illustrated in four mouse models, Pou4f2-null, Isl1-null, Atoh7-null and Pou4f2-Dta mice; all these models exhibit reduced RPC proliferation because of detective or failed RGC differentiation and thus compromised Shh signaling [38, 39, 43, 43, 51, 177]. Therefore, the genetic pathway promoting RGC genesis also negatively feeds back onto itself by activating Shh to balance differentiation and proliferation in the developing retina. This mechanism is at least partially achieved by promoting the expression of cyclin D1 in RPCs [176, 177]. Since all retinal cell types arise from naïve RPCs and RGCs are the first cell types to form, this feedback mechanism ensures sufficient cell numbers are produced for late-born cell types.

Gdf11 and Gdf8 (also known as myostatin, Mstn) are members of the Tgfβ family. Binding of BMP ligands to one of the receptors leads to a cascade of phosphorylation and translocation of the Smad proteins to the nucleus, which eventually activates downstream target genes [178, 179]. The deletion of Smad4 leads to defects in RGC axon projection [180]. However, since signaling from all Tgfβ ligands converges at Smad4, this result cannot distinguish the functions from the individual ligands, as Tgfβ many ligands are expressed in the developing mouse retina, including Bmp4, Gdf8, and Gdf11 [181–183]. The main function of Bmp4 appears to be in the retina-lens interaction and in defining the ventral-dorsal domains in the retina, although it may also play some roles in RGC development [184]. Gdf8 and Gdf11 are the most closely related proteins within the Tgfβ family. They are both expressed in developing RGCs, although Gdf11 is also expressed in RPCs. Similar to Shh, Gdf11 also negatively regulates RGC production, since knockout of Gdf11 leads to increased and prolonged RGC genesis [185]. Although Gdf8 is well known for its role in regulating skeletal muscle formation, the roles of Gdf8 in retinal genesis have not been clearly elucidated. Notably, Gdf8 is also under the control of the Atoh7-Pou4f2/Isl1 cascade. Given the close homology between Gdf8 and Gdf11 and the fact that both proteins are expressed in RGCs, Gdf8 may conceivably have redundant functions with Gdf11, although this hypothesis remains to be confirmed experimentally.

Vegf and its receptors, which are traditionally known to regulate vasculogenesis, are expressed in the developing retina before the formation of the retinal vasculature, suggesting that the pathway has non-vasculogenesis-related functions in the retina [186, 187]. Similar to the Shh pathway, the Vegf pathway also functions in a feedback fashion to modulate the balance between proliferation and differentiation; Vegfa promotes RPC proliferation and represses RGC formation in the avascular chicken retina [188]. Additionally, mice lacking Nrp1, which is also as a Vegf receptor, and Nrp1-binding Vegfa isoforms have defects in both RGC projections and vascular structures, indicating additional functions of the Vegf pathway [189].

Notably, significant cross-talk exists between these pathways. For example, both the Shh pathway and the Vegf pathway interact with the Notch pathway by co-regulating the Notch target gene Hes1, highlighting the complexity of the regulatory mechanisms involved [175, 188, 190, 191].

Concluding remarks

Substantial progress has been achieved in understanding the genetic mechanisms controlling the formation of RGCs during development. Based on what we have learned, the developmental trajectory of the RGC lineage can be presented in the traditional Waddington diagram in the context of retinal development to depict the different stages (Fig. 2). The direction of the trajectory is shaped by changes in the epigenetic landscape at different stages. Details of the shifts in the epigenetic landscape should emerge soon with the use of powerful new techniques such as single cell ATAC-seq, which uses the Tn5 transpose to probe chromatin regions that are accessible to the enzyme at single cell resolution [192, 193]. This technique will facilitate the identification of enhancers whose activities are associated with stage-specific gene expression. As discussed throughout this review, many regulatory factors functioning at the different stages along the developmental trajectory of the RGC lineage have been identified and investigated (Table 1), and more are being added to the list from global transcriptomics studies. These regulatory factors shape the epigenetic landscape along this trajectory. However, we are far from having a comprehensive and cohesive understanding of many of the key steps. The functions of most regulators are still anecdotal and fragmented, the relationships among many of them remain obscure, and major gaps persist. Looking forward, the major tasks are to fill these gaps and integrate the fragmented pieces of information to obtain a systematic view of how this major retinal cell class and the various types of RGCs are generated. Below, we list three major directions we consider critical to understanding the process of RGC development.

Fig. 2.

A Waddington diagram showing the epigenetic landscape directing the RGC lineage. Pools represent different stages, and ridges and valleys illustrate the epigenetic landscape determining the developmental trajectories. Directions of the trajectories are indicated by arrows. Other than the RGC trajectory, the other two trajectories, amacrine cells (ACs) and horizontal cells (HCs), and photoreceptors (PHs), are also shown. Conceptually, the epigenetic landscape is shaped by the transcription factors expressed at each stage

Table 1.

Transcription factors functioning at different stages of the RGC developmental trajectory

Stages	Transcription factors
Naïve RPCs	Pax6, Rax, Six3, Six6, Vsx2, Sox2, Lhx2, Notch1/Hes1, 5, Nr2e1, Gli1
Transitional RPCs	Atoh7, Neurog2, Neurod1, Sox4, 11, 12, Dlx1, 2
Nascent RGCs	Isl1, Pou4f2, Sox4,11,12, Eya2, Dlx1,2
RGC differentiation	Pou4f1, 2, 3, Pou6f2, Myt1, Myt1l, Sox4,11,12, Ebf1-4, Irx1-6, Klf7, Klf4, Pax6, Lhx9, Rxrg, Tle1-6, Zfhx2-4, Nhlh2, Nkx6-2, Thra, Pbx3
RGC types	Tbr1, Tbr2, Isl2, Pour4f1-3, Satb1, Onecut1-3, Barhl2, Gfi1, Foxp2, Prdm16, Tfap2d
RGC projection	Zic2, Isl2, Sox4,11,12

First is the establishment of the transitional RPC state. As revealed by recent single-cell transcriptomics studies, these cells are in a plastic state, display downregulated Notch signaling and upregulated expression of Notch ligands, are multipotent, co-express genes for multiple cell fates, are on the verge of exiting or have just exited the cell cycle, and are ready to assume one of the retinal cell fates. All these events likely occur in a coordinated manner controlled by complex cross-regulation and signaling, particularly through the Notch pathway, but the details remain to be worked out. Since Atoh7 marks this state, at least in the early stages of retinal development, an understanding of how Atoh7 is activated may be the first step. Key enhancers mediating Atoh7 activation have been identified [47, 194, 195], which will greatly facilitate the investigation of its regulation. The identification of the upstream regulators acting on these enhancers will be key to understanding how this transitional state is established, since these factors likely also regulate the other aspects of the transitional RPC state.

The mechanism by which the RGC fate is fixed from competent/transitional RPCs is still not well understood. Recent findings indicating that the RGC lineage still forms, although it is stalled prematurely, in the absence of Atoh7 present us with a new perspective. Isl1 and Pou4f2 are the major transcription factors involved in this process, but other factors are likely also involved. Since activation of Isl1 and Pou4f2 coincides with the emergence of RGC fate, identifying the upstream inputs and deciphering the mechanisms regulating these two genes would be a critical step toward understanding the process. Identifying the additional factors and their functions will be another critical aspect of future research. In this regard, the functions of the SoxC transcription factors in the RGC lineage must be further analyzed in a stage-specific manner to discern their roles in the different stages.

The genetic mechanism underlying the formation and differentiation of individual RGC types has generally been a black box until recently but will continue to be an active area of research in the future. The identification of type-specific markers will greatly facilitate this area of research, since these markers can be used to assess RGC type-specific defects in mutant mice, and studies examining their regulation may be the entry point to understanding the formation of the relevant RGC types. Mutant mice affecting specific RGC types may exhibit relatively mild defects, which often dampened the motivation for continued study in the past, but this situation is changing. Some of the mouse lines mutant for genes expressed in RGCs but without overt defects observed previously should be revisited. For genes that are not exclusively expressed in RGCs, cell type-specific deletion will be needed. Individual RGC types can be and have been genetically marked by fluorescent proteins directly or indirectly, and their physiological and molecular properties can thus be investigated. Mouse lines in which RGC types are labeled with fluorescent proteins will also facilitate the delineation of the developmental history of individual types. Using fluorescence as a direct readout, these marked cells can be further used to study the regulatory mechanisms controlling the corresponding marker genes, which are integral aspects of the differentiation of specific RGC types.

Addressing these questions will require a combined approach integrating conventional experimental investigations, such as genetic perturbations, genetic labeling, cell sorting, and imaging, with new technologies such as single cell transcriptomics, single cell epigenomics, and CRISPR-based genome editing. For example, a comparison of the wild-type and mutant epigenetic landscapes using single cell ATAC-seq and other similar technologies will reveal the specific roles of individual transcription factors in shaping the epigenetic landscape along the RGC developmental trajectory. Major challenges will include experimental assessments of the function of many cis elements identified by epigenomic studies, identification of the transcription factors acting on them, and determination of their combinatorial interactions. High-throughput assays combining CRISPR and next-generation sequencing will be needed to enable functional testing of enhancers at global levels. The goal is to obtain a systemic view of the transcriptomic state and epigenetic landscape at each stage of RGC formation and the genetic regulators and pathways defining each stage and driving it to the next.

Authors’ contributions

XM conceived the project, reviewed the literature, and wrote the paper. JL reviewed literature and participated in writing the paper.

Funding

National Institutes of Health (R01EY020545 and R01EY029705).

Availability of data and materials

Not applicable.

Code availability

Not applicable.

Declarations

Conflict of interest

The authors declare that they have no conflict of interest.

Ethics approval

Not applicable.

Consent to participate

Not applicable.

Consent for publication

Not applicable.