In the adult vertebrate retina, there are six major classes of neurons and one class of Müller glial cells. The retinal neurons include cone and rod photoreceptors and horizontal, amacrine, bipolar, and ganglion cells. Photons from the light are first caught by cone or rod photoreceptors, which convert them into signals and relay the signals to the interneurons, the bipolar cells. The horizontal cells modulate the signals from photoreceptors before they are relayed to the bipolar cells. Again, the amacrine cells integrate and modulate the signals from bipolar cells before they are relayed to retinal ganglion cells (RGCs). Finally, the RGCs transferred the signals to the specific brain regions and the brain generated the three-dimensional vision. As the only major non-neuronal cell type, Müller cells provide scaffolding supports, nutrients, metabolite recycling, etc. for the neurons. A special cell type, microglia, is not from retinal origin but arises from circulating monocytes/macrophages. Microglia is involved in immune surveillance and cleaning of the retina.

Retinal cells are highly diversified. Most of the major class of retinal cells can be further categorized into subgroups according to their morphology and function. For example, RGCs have more than 30 subgroups [1]; amacrine cells have over 29 subgroups [2, 3]. All these retinal cells are originally developed from a small group of multipotent progenitor cells in the optic vesicle. This complicated process is delicately controlled. A popular intrinsic model has been proposed to describe this process [4–8]. The main point of the model is that progenitors go through intrinsically determined competence states, during which they are capable of giving rise to a limited subset of retinal cell types under the influence of extrinsic signals. The model could gracefully explain how the intrinsic and extrinsic factors together control retinal cell genesis. However, details of how multipotent progenitors develop into each cell type are still murky. Here, I summarized the recent advances on retinal and neural development, focusing on topics such as cell division and differentiation, retinal intrinsic programs, intrinsic program versus stochastic mechanism, transitional progenitor model, early-born progenitors (EBPs) versus late-born progenitors (LBPs), and so on.

Cell division and differentiation

Retinal progenitor cell (RPC) pool contains a limited number of cells. To guarantee the normal size of the retina, the multipotent progenitors undergo fast cycles of self-replication before initiating cell differentiation. The self-replicating process is under the tight control of Rax, Meis1&2, Pax6, Notch1, Shh, and other factors [9–14]. Mutations in this group of factors generally cause microphthalmia or more serious phenotypes in human and animal models, due to far fewer progenitor cells available in the progenitor pool.

There are two scenarios when a progenitor cell undergoes mitosis (Fig. 1). One scenario is that the progenitor is evenly divided into two identical daughter cells, either two identical progenitors (Fig. 1a) or two neurons/glias of the same type (Fig. 1c). Two progenitors are usually generated at the early developmental stage when the progenitor pool needs to be expanded rapidly. On the other hand, two differentiated daughter cells will be produced at the late stage of organogenesis. The second scenario is that the progenitor divides asymmetrically and gives rise to one progenitor cell and one neuron/glia (Fig. 1b), or two neuron/glial cells of different types (Fig. 1d). Examples of these scenarios were provided in the study of Ath5 in the wild-type and lakritz mutant mice [15] and in a work by Cayouette and colleague [16, 17].

Fig. 1
figure 1

Four cell division modes for symmetric and asymmetric divisions. (A) A progenitor cell divided symmetrically and gave rise to two undifferentiated progenitor cells. (B) A progenitor cell divided asymmetrically and resulted in one undifferentiated daughter cell and one differentiated daughter cell. (C) A progenitor cell divided symmetrically and two differentiated daughter cells of same type were born. (D) A progenitor cell divided asymmetrically, and two differentiated daughter cells of different types were generated. In E&F, cleavage plane aligns in parallel or vertical to the apical-basal axis, namely horizontal or vertical division, respectively. (E) Horizontal division usually leads to symmetric division of two identical daughter cells, each inheriting equal amount of cell fate determinants. (F) Vertical division tends to asymmetrically divide a progenitor into two daughter cells of different size and morphology, each inheriting unequal amount of cell fate determinants

Full size image

At the onset of neurogenesis, the progenitors gradually switch from proliferative, symmetric to neurogenic, asymmetric divisions. The shift is shown to be associated with a change of the orientation of the mitotic spindles in the dividing progenitors. It is proposed in CNS development that if the cleavage plane aligned in parallel with apical-basal axis, it is termed horizontal division; whereas the cleavage plane is perpendicular to apical-basal axis, it is named vertical division (Fig. 1e, f). It is found that horizontal division of a mitotic cell usually divides symmetrically and gives two identical daughter cells, while the vertical division tends to be asymmetric and gives two daughter cells of different size and morphology. This may partially attribute to the unequal inheritance of cell surface molecule Notch and intracellular cell fate determinants Numb and etc. [16, 18]. Observations by Kosodo et al. suggest that the vertical division is mostly asymmetric, but it can be either proliferative, symmetric or neurogenic, asymmetric, depending on the equal or unequal inheritance of cell constitutes [19]. Interestingly, the definition of horizontal versus vertical division of neural progenitors is similar to Dr. Hans Spemann’s definition of sagittal versus frontal division of fertilized eggs, and the apical side is counterpart of the gray crescent region.

The key question is, what is the molecular machinery determining symmetric versus asymmetric inheritance? The findings from drosophila to mammals show that the polarity is controlled by evolutionarily conserved protein complexes: the Par proteins (Par3-Par6-aPKC), the heterotrimeric G protein complex (Gαi-Pins (LGN/Gpsm2 in mammals)-Mud (NuMA in mammals)), and Insc that can bind Par3 and Pins [20, 21]. The Notch signaling pathway is also critical in regulating the division polarity. Notch is found to promote the asymmetric localization of the protein Numb and the positioning of the cleavage furrow [22]. Eya1 and Gαs also controls spindle orientation and asymmetrical division by regulating Notch signaling [23, 24]. However, it is still unclear how each fate-determining transcription factor specifically controls the asymmetrical cell division by regulating these polarity proteins during cell cycle. Mostly, the regulation was initiated at G1 phase, as G1 phase cells have permissive epigenetic environment that allows developmental programs to be activated, and G1 phase cells favorably respond to fate-determining extrinsic cues [25].

Retinal intrinsic programs specifying cell fates

Emphasis has been put on intrinsic developmental programs in specifying cell fates. However, what composes an intrinsic program is not clearly defined. An intrinsic program is established in evolution and is mainly encoded by a cascade of intrinsic factors, especially transcription factors. An intrinsic program can be initiated by a signal from intrinsic or extrinsic factors, or another program. Once initiated, a program is hardly reversible in vivo; however, it can be forcibly aborted by another program, which sometimes leads to cell apoptosis. A cell in homeostasis is in a balanced state of a series of programs. Once the balance is destroyed in the progenitor, the cell either changes its status (i.e., dividing or differentiation) or commits to apoptosis. A program may have two or more alternative subprograms at a branching point. Selective execution of one of the subprograms depends on the inputs of extrinsic signals, which underlies the basis that cell fate decision relies on both intrinsic and extrinsic factors.

In general, there are three major types of intrinsic programs in retinal or other tissue development (Fig. 2). The first type is the linear program, which usually occurs at the late stages of development (Fig. 2a). The linear program reflects the simplest biological causal relationship, and one program specifies only one cell-type fate. Disruption of the program leaves the cell nowhere to go, and it usually commits to cell apoptosis. For instance, Bhlhb4 is specifically expressed in the rod bipolar cells and guides the terminal differentiation of the cells. Deletion of Bhlhb4 causes the cells take the path to apoptosis in the mouse retina (Fig. 2a, [26]).

Fig. 2
figure 2

Typical intrinsic programs in retinal and other tissue development. a The linear program specifies only one cell type fate, and disruption of the program usually results in cell apoptosis. b The branching program has two or more subprograms to choose at the branching point. Which subprogram(s) to execute depends on the extrinsic signal input and will lead to different cell fate(s). For example, the extrinsic factor x will force the progenitor to take the RGC path while inhibiting Amacrine and Horizontal path. Disruption of one subprogram will result in more execution of the other programs. In this case, more amacrine and horizontal cells will be generated when RGC subprogram is destroyed during development. c The converged program is the contrary to the branching program. Two or more subprograms merged together to define one cell type fate. Block of one of subprograms will generate less target cells to some extent

Full size image

The second type is the branching program (Fig. 2b), which mostly takes place in the early and peak stage development. The progenitor cell taking the branching program has two to several subprograms to choose, and it depends on the extrinsic signals to pick the right subprogram to continue. The lack of complete reliance on intrinsic signals ensures that there are branching choices and not a linear program. Again, this is how extrinsic and intrinsic factors together determine cell fate. Taken as an example, the early-born progenitor is capable of differentiating into a RGC, horizontal or amacrine cell (Fig. 2b). The fate choices depend on whether the extrinsic signals to activate the Foxn4 or Brn3b subprograms. If Brn3b is activated, the progenitor will terminally differentiate into a RGC. If Foxn4 is triggered, the progenitor will become a horizontal or amacrine cell depending on further choice. If Brn3b gene is knocked out, there would be a temporal increase of horizontal and amacrine cells [27]. Alternatively, if Foxn4 expression is abolished, RGC population would temporally get boosted [28, 29].

The last major type is the converged program (Fig. 2c). Contrary to the branching program, the converged program has two or more subprograms merged together to specify one cell type. Evolutionally considered, this is important since it provides a redundant mechanism to guarantee the generation of a crucial cell type. Interruption of one of the subprograms usually results in a decrease but not a complete loss of target cells. For instance, Brn3b, Isl1, Sox4, and Sox11 are all important for RGC development. Deletion of any one of them would cause partial loss of RGCs, but compound knockout of two factors would lead to a near-complete loss of RGCs [30–32].

In conclusion, three types of programs are all important in fate decision during development. It should be emphasized that none of the programs is isolated in the cell, and they work synergistically in sequential or in parallel to activate/inhibit downstream genes, since conflicting programs usually lead to cell death, which were eliminated during evolution. For instance, Foxn4, Ptf1a [33], and Tfap2α/Tfap2β [34, 35] operate in a sequential cascade to determine horizontal and amacrine cell fates, while RORβ1 seems to work in parallel with Foxn4 to turn on Ptf1a expression [36]. These programs work coordinately to form functional horizontal and amacrine neurons in the retina.

Intrinsic programs versus stochastic mechanism

It is believed that intrinsic developmental programs are the most decisive forces in specifying cell fates [37, 38]. Slater [39] and Gomes [40] et al. found that stochastic mechanism also plays an important role in the process. However, Gomes’s conclusion that stochasticity is the major role during retinal development is in question. If it is largely a stochastic force, one would expect that an approximately equal number of rods, amacrine, bipolar, and Müller cells would be generated during the experiment, obviously this is not the case. Then, it raises the questions: (i) what is underlying the stochastic mechanism, and (ii) when does the stochastic mechanism take over the control? One plausible answer to (i) is random fate choice due to the balance of forces from two or more opposite programs, per se, the balance due to the dose-dependent effect of antagonistic transcription factors. As illustrated in Fig. 3, in postmitotic Otx2+ progenitors, Blimp1 and Vsx2 (Chx10) were the two intrinsic determinants to choose a bipolar or a rod cell fate [41, 42]. There exists a point which I called balance point when Blimp1 and Vsx2 reach equivalence. A small region flanking the balance point, named the stochastic zone, is the key to question (ii). Beyond the left side of stochastic zone, the Blimp1 dominates the Vsx2 and the cell will definitely become a rod, and vice versa. Only in the stochastic zone, the stochastic force has some effects on cell fate choices; however, the ratio should be close to 50:50 for rod and bipolar fates in the zone. As a result, stochasticity causes variations in cell fate determination, but will not affect the overall cell ratio statistically. We may conclude that the intrinsic program(s) is the major driving force, while the stochastic mechanism also plays a minor role and causes variations in the development.

Fig. 3
figure 3

Illustration of stochastic mechanism in the rod versus bipolar cell fate decision. The rod versus bipolar cell fate decision was determined by the dose-dependent effect of Blimp1 and Vsx2 protein. When Blimp1 dominates Vsx2, the cell will become a rod cell; otherwise, it will become a bipolar cell if Vsx2 is dominant. There is a balance point when Blimp1 and Vsx2 reach equivalence. Flanking the balance point is the stochastic zone where the cell fate was decided by stochastic mechanism. In the stochastic zone, the progenitor has equal chances to differentiate into a rod or a bipolar cell

Full size image

Transitional progenitor model for retinogenesis

A pluripotent progenitor cell usually does not give rise to terminally differentiated cells directly; instead, it generates the multipotent progenitors. The multipotent progenitors lose the pluripotency, are partially determined cells, and can only produce a limited set of cell types. The multipotent progenitors also give birth to more restricted intermediate progenitors, the oligopotent progenitors, which are capable of generating only three or more cell types. The oligopotent cells then differentiate into bipotent and finally into unipotent progenitors that can only produce terminally differentiated progeny. This multi-step commitment seems to be the case in the retinogenesis as well as in other tissue development, such as in hematopoietic genesis. Each step is controlled by one or a serial of intrinsic programs. The intermediate oligopotent, bipotent, and unipotent retinal progenitors are defined as transitional progenitors in this model. The transitional progenitor model emphasizes the heterogeneity of RPCs, reflecting the fact that the retina, at any given developing point, is a bag of mixed multipotent and transitional RPCs with terminal cells.

Here, a model is proposed that the intrinsic programs define the competence state of transitional progenitors and specify retinal cell types and subtypes. Previous popular models describing retinal development were reviewed in details recently [43]. These models suggest that the RPCs gradually lose the competence during the developmental process [4, 5, 43]; however, the transitional progenitor model here proposes that the multipotent RPCs keep their competence to generate all potential retinal cell types. The ‘restricted’ competence is due to a gradual decrease of multipotent RPCs and a steady increase of the transitional progenitors and terminal cells over the developing course. The model is supported by discoveries that the late RPC pool is comprised of mixed population with stem cell-like multipotent progenitors and heterogeneous lineage-restricted progenitors [44, 45]. Even in the mature retina, there are adult stem cells (equivalent to multipotent RPCs by definition) that have the full potential to generate all retinal cell types and maintain the integrity of the retina. Another difference is that previous models suggest that differentiating RPCs follow a linear route during the retinal development (Fig. 4a), while the transitional progenitor model proposes a tree-structure route which is more reasonable (Fig. 4b). For example, readers could be misled by the linear route to believe that amacrine cells are generated from the RGC-fate-incompetent progenitors and are born later than RGCs. As a matter of fact, the early progenitors are competent for RGC, amacrine, horizontal, and cone cell fates, and birthdays of these cells are interweaved.

Fig. 4
figure 4

Models of RPC differentiation and fate determination. a One representative intrinsic model is shown. According to the model, multipotent RPCs pass through temporal order of states of competence over the course of development, and get more and more restricted in cell fate choices from one state to another. During each competence state, the RPC is capable of terminally differentiating into one or several specific subtypes of cells (for simplicity, only one type of cell is depicted in each state in the cartoon). b The model proposed here emphasizes transitional changes and complex heterogeneity among progenitors. The cell fate determination follows an order from multipotent progenitor to oligopotent then bipotent and unipotent progenitor, and the RPCs finally differentiated into particular cells. At any given developmental stages, the retina is composed of mixed population of multipotent, oligopotent, bipotent, and unipotent progenitors and terminally differentiated cells. The multipotent RPCs are getting fewer and fewer in number, but they keep their competence and are not restricted in fate choices. The fates of multipotent progenitors were determined by both the environmental cues and intrinsic programs

Full size image

The crucial transitional progenitors, including the early-born and late-born oligopotent progenitors, bipotent progenitors, unipotent progenitors, and the associated transcription factors defining the intrinsic programs in these progenitors, are listed in Fig. 5 and discussed below.

Fig. 5
figure 5

A hypothetical model of intrinsic programs for hierarchy lineage specification in the mouse retina. The multipotent progenitors give rise to transitional progenitors from oligopotent to bipotent and unipotent. RGC, horizontal, amacrine, and early-born cone photoreceptor cells originate from a common pool of intermediate progenitors (EBPs). Bipolar, Müller, and late-born photoreceptor cells share a pool of common intermediate progenitors (LBPs). The differentiated cells can give negative feedback signals to its progenitor pool. Critical factors are listed on each pathway

Full size image

EBP groups

There is evidence to suggest that there is a pool of common oligopotent progenitors for early-born cell types, such as horizontal, amacrine, RGC, and cone photoreceptor cells (Fig. 5). All EBPs are Pax6-positive. Pax6 is a transcription factor containing paired box and homeodomain and is highly conserved in vertebrates. In the chick, the onset of Pax6 starts at Hamburger and Hamilton stage 8.5 in primordial eye field and persists at high levels throughout optic cup morphogenesis [46]. The continuous expression of Pax6 activates Math5 expression in some of the EBPs [47]. Math5 endorses the progenitors to differentiate into RGCs, and Math5-negative progenitors become amacrine, horizontal, and photoreceptor cells. In the Math5 null retina, there is an increase of amacrine and cone cells but not bipolar cells [48], supporting the idea that RGCs, cones, and horizontal and amacrine cells share common EBPs, since promotion of one cell type will inhibit the other cell types as they share the same pool of progenitors. Continual expression of Pax6 can also activate the expression of Otx2 and Trβ2 in some of the progenitors and leads to generation of cone but not rod photoreceptors [49], which explains why cone photoreceptors are generated earlier than rod photoreceptors. It also lends evidence to support the idea that rods were developed from cones evolutionarily, since the embryonic development is a reflection of evolution history to some extent.

Onecut1 (OC1) and Onecut2 (OC2) are initiated at E11.5 which is later than Pax6. They seem to be specifically expressed in EBPs at the stage. Compound deletion of both will result in complete absence of horizontal cells and early-born cholinergic amacrine cells, and partial losses of RGCs and cone photoreceptors [50, 51]. Other than expressed in EBPs, OC1/2 also expresses in RGCs and horizontal cells at late stages [50]. During RGC development, OC1/2 seems to function independently of Math5.

The intrinsic programs defining unipotent RGC progenitors seem to be complicated. The RGCs have more than 30 subtypes with morphological and functional differences. The specification of the subtypes is decided by differential expression of Brn3b, Dlx1, Dlx2, Islet1, and others [30, 32, 52, 53]. For example, Brn3b (Pou4f2) is a Pou domain and homeobox domain transcription factor expressed in the postmitotic precursors. Expression of Brn3b promotes RGC fate at the expense of amacrine, horizontal, and cone cells [27], which is another evidence supporting the existence of common EBPs.

Horizontal and amacrine cells share a pool of bipotent progenitors that are progenies of EBPs. This is strongly supported by evidence from the Foxn4 and Ptf1a mutant mice. Foxn4, a forkhead box domain transcription factor, is critical in the retinal and other tissue development [28, 54]. In the Foxn4 null retina, all horizontal cells and the majority of amacrine cells are lost; instead, more RGCs and photoreceptors are temporarily generated [28], showing a fate switch from amacrine and horizontal cells to RGCs and photoreceptors. Similar phenotypes are found in the Ptf1a knockout mice [33, 55]. These evidences support the ideas that these four cell types share the common pool of EBPs and that horizontal and amacrine cells rise from the same bipotent progenitors. The activating protein-2 (AP-2) transcription factors Tfap2α and Tfap2β act downstream of Ptf1a and regulate the genesis of horizontal and amacrine cells as well [34, 35].

Horizontal cells are interneurons situated between photoreceptors and bipolar cells and form synaptic connections with both cell types. There are two kinds of horizontal cells in the vertebrates, the axon-bearing and the axon-less ones, though there are only axon-bearing ones in some vertebrates, i.e., rodents. The terminal differentiation of horizontal cell depends on the homeodomain protein Lim1, which is exclusively expressed in the unipotent horizontal progenitors [56]. Loss of Lim1 causes the horizontal cell precursors stuck in the wrong laminar position. As a result, the ectopic horizontal cells adopt a morphology more reminiscent of amacrine cells.

Unlike horizontal cell, the amacrine cell has more than 29 subtypes and its terminal differentiation is more complicated, most possibly controlled by many terminal programs. Transcription factors NeuroD and Math3 are genetically downstream of Ptf1a and redundantly control the genesis of amacrine cells. In NeuroD and Math3 double knockout retinae, all amacrine cells fail to differentiate; however, ganglion and horizontal cells are increased [57–60], which is also consistent with the EBP hypothesis. Similar to NeuroD or Math3, Prdm13 is also a downstream gene of Ptf1a and regulates the development and function of a subset of glycinergic and GABAergic amacrine neurons [61]. Only a few genes are known to control the development of one specific subtype of amacrine cells. For instance, the dopaminergic amacrine cell fate is controlled by the orphan nuclear receptor Nr4a2 (Nurr1) [62], and the development of cholinergic amacrine cells depends on LIM-homeodomain factor Isl1 [63].

LBP groups

Evidence also suggests existence of a pool of common oligopotent progenitors for late-born cell types, including Müller, bipolar, and photoreceptor cells (Fig. 5). The common transcription factors controlling LBPs were unclear; however, LBPs might oscillatorily express Notch1 [64, 65] and temporally express the paired-type homeodomain transcription factor Vsx2 [66]. The onset of Vsx2 expression starts no earlier than Hamburger and Hamilton stage 12 [67]. Vsx2 expression at the early stage of development controls proliferation of RPCs, and Vsx2 mutation causes ocular retardation due to RPC proliferation defects [66].

The progeny fate of LBPs depends on differential expression of Vsx2, Otx2, Blimp1, and Notch1. It has been shown that Vsx2 directly controls bipolar cell genesis by inhibiting rod differentiation [68]. Mutation of Vsx2 results in the total loss of bipolar cells; however, the genesis of horizontal and amacrine cells is pretty much normal in the Vsx2 mutants [66], implying that only LBPs but not EBPs are affected. By regulating Vsx2 and Otx2, Blimp1 promotes photoreceptor cell fate while inhibiting bipolar and Müller cell fates [41, 42, 69]. Ablation of Blimp1 results in far fewer photoreceptors, more bipolar and Müller cells, but unchanged amacrine, horizontal, and RGCs [41, 42], which also supports our hypothesis that Müller, bipolar, and photoreceptor cells share the same pool of LBPs, and consistent with the notion that promotion of one cell fate will inhibit the other cell fates. LBP fate determination is also controlled by transcription factor Math3 and Mash1. Compound knockout of Math3 and Mash1 also results in loss of all bipolar cells, but Müller cells are significantly increased at the cost of bipolar cells [60], suggesting bipolar and Müller cells originate from the same progenitor pool.

Knockout of Otx2 results in loss of all photoreceptor cells in the mouse retina [70], which indicates that there is a repertoire of common bipotent progenitors for cone and rod photoreceptor cells (Fig. 5). It is supported that the default fate of the cone and rod progenitors is S-cone cell, since mutations of numerous genes in the common progenitors unanimously lead to enhanced S-cone syndromes, such as mutations in Trβ2 and Nr2E3 [71, 72]. Expression of wild-type Nrl or Nr2E3 blocks this default pathway in the progenitor and initiates rod cell genesis, while expression of Trβ2 or Rxrγ blocks the default pathway and leads to generation of M-cones [71]. Conditional deletion of Otx2 also leads to a vast increase of GABAergic and glycinergic amacrine cells that span into the space where a normal ONL layer should be. The Müller and bipolar cells also increased due to cell fate switches from photoreceptor cells [70, 73]. Cone-rod homeobox gene Crx works at the downstream of Otx2, it seems not involved in photoreceptor fate specification, but rather to enhance the expression of photoreceptor-related terminal genes [70, 74, 75]. Similarly, transcription factor Mef2d works with Crx to drive the retina-specific gene expression in photoreceptor, but not involved in the fate decision [76, 77].

As one of the most important pathways affecting the development of many tissues, the Notch pathway is not only involved in early RPC maintenance but also in the rod and Müller cell fate determination. Overexpression of Notch1 promotes Müller cell fate [78]. Overexpression of its downstream gene Hesr2 or Hes1 also promotes Müller genesis at the expense of rod cells [78, 79]. Retroviral-mediated conditional ablation of Notch1 at postnatal stage induces the generation of rod photoreceptors at the expense of bipolar and Müller but not amacrine cells [80]. These findings together strongly support that bipolar, Müller, and photoreceptor cells share a pool of common LBPs.

The terminal differentiation of Müller cells involves in Sry-related HMG box gene Sox9. Specific deletion of Sox9 from developing retina resulted in loss of Müller cells [81]. Apparently, other Sry-related HMG box genes were involved in Müller cell development as well, such as Sox2 [82, 83], Sox8 [84], etc. The cone bipolar cells have around nine subsets, and specification programs for the subsets depend on Vsx1 [85], Irx5 [86], Bhlhb5 [87], etc.

Genesis of EBPs and LBPs

Multipotent RPCs were maintained in the collaborated network of Rx [9, 13], Notch1 [64, 78], Meis1/2 [10], Pten [88], Pax6 [14, 49], etc. A multipotent progenitor’s decision to be either an EBP or a LBP was mostly controlled by extracellular environmental cues, including cell-cell contacts (neighboring effect), interaction between Notchs and ligands [29, 80, 89–91], extrinsic factors such as the BMPs gradient [92–95], hormones and growth factors [96–98], ephrins and receptors [99, 100], etc., which initiated intrinsic developmental programs. Early developmental environment favorably promotes the EBP fates, and late developmental environment prefers LBP fates. Genesis of EBPs, LBPs, and major retinal cell types is in a timely overlapping manner. RGCs initiated at about E11 and were the first cell type differentiated, followed by amacrine, cone, and horizontal cells at about E12 (Fig. 6, [101]). Rod photoreceptors first appeared at around E14 and then joined by bipolar and Müller cells. We could deduct that EBPs arise from E10 to P4 and peak at E16, while LBPs appear from E13 to P10 and peak at P2 (Fig. 6).

Fig. 6
figure 6

Developmental stages for transitional EBPs and LBPs in the mouse retina. EBPs initiated at about E10 and ended at about P3. LBPs started around E13 and ended around P10. The development of EBPs and LBPs overlapped spatiotemporally. The specification timing of RGC, amacrine, horizontal, bipolar, Müller, cone, and rod photoreceptor cells was also illustrated

Full size image

Conclusion

The transitional progenitor model defines progenitor status from multipotent to oligopotent, bipotent, and unipotent stepwisely. In each transitional status, the competence is controlled by intrinsic developmental programs as well as environmental cues. The multipotent RPCs become fewer and fewer with the ongoing differentiation and are almost depleted in the adult retina. The terminally differentiated retinal cells usually are not capable of proliferating, except some stem cell-like Müller glial cells [102–105] and Lgr5+ amacrine cells [106]. The epithelial cells in the ciliary margin zone were previously identified as retinal stem cells [107, 108]; however, they are unable to differentiate into retinal neurons in vivo or in vitro [109, 110] and thus disqualified as stem cells.

It needs to be pointed out that there are many mechanisms, besides transcriptional control, regulating retinal development, i.e., RNA-level (microRNAs, ncRNAs, etc.) regulation [111, 112], protein modification and degradation [113, 114], epigenetic modification [115, 116], cell death regulation [117, 118], and etc. Neighbor tissues, such as RPE and lens, certainly have indispensable effects on retinal development as well, mostly through the mechanism of mutual induction and inhibition. We will not discuss these topics in further details.

Understanding the programs underlying the developmental process not only will provide insights into how each individual cell type is formed and what function it may have, but also will give clues on how to treat retinal diseases in the future. Recent advances in stem cell researches illustrate the importance of understanding such developmental processes. Many efforts have been made worldwide to induce retina-specific cells from adult somatic cells or induced pluripotent stem cells (iPSCs) [119–121], which shed some light on cell replacement therapies to treat patients in the near future.