Comparative Biology of Vertebrate Retinal Regeneration: Restoration of Vision through Cellular Reprogramming

Abstract

The regenerative capacity of the vertebrate retina varies substantially across species. Whereas fish and amphibians can regenerate functional retina, mammals do not. In this perspective piece, we outline the various strategies nonmammalian vertebrates use to achieve functional regeneration of vision. We review key differences underlying the regenerative potential across species including the cellular source of postnatal progenitors, the diversity of cell fates regenerated, and the level of functional vision that can be achieved. Finally, we provide an outlook on the field of engineering the mammalian retina to replace neurons lost to injury or disease.

The retina is the sensory tissue of the eye, the primary sense organ for vision. Figure 1 shows the basic anatomy of the eye for orientation. The anterior of the eye consists of a chamber (the anterior chamber) bounded by the cornea—a tough, transparent tissue—the lens, and the ciliary body. Behind the anterior chamber is a large cavity filled with a gel-like material called the vitreous and lining the inner surface of this vitreous chamber is the thin layer of neural cells, the retina. The retina is in turn surrounded by additional layers of cells; immediately adjacent to the retina is the pigmented epithelial layer (known as retinal pigment epithelium [RPE]), followed by the vascular choroid layer and then the fibrous scleral layer. All vertebrate retinas share a common organization and basic cell types, although the retina has adapted to suit each species’ ecological niche at the level of cell-subtype diversity and neural circuitry (Baden et al. 2020). All vertebrates have both rods and cones, the two types of photoreceptors (the cells that transduce light energy to changes in synaptic transmitter release). Photoreceptor output is modulated by interneuron subtypes: horizontal cells, bipolar cells, and amacrine cells. These interneurons relay the signals from the photoreceptors to the retinal ganglion cells, the projection output neuron of the retina (Fig. 1).

Figure 1.

Basic anatomy of the vertebrate eye and retina. (A) The key parts of the eye are identified. The ciliary marginal zone [CMZ] is at the interface between the ciliary body and the retina (yellow). (B) The neural retina and retinal pigmented epithelium (RPE) are shown at higher magnification. The cell bodies of both rods and cones are in the outer nuclear layer [ONL] (blue and red). The inner nuclear layer (INL) contains the cell bodies of horizontal cells (dark blue), bipolar cells (purple), and amacrine cells (green). The retinal ganglion cells (RGCs) have their cell bodies in the innermost layer of the retina (yellow). Created using BioRender.com.

During development, the retina begins as an outgrowth of the diencephalon and like the rest of the central nervous system (CNS) has a limited regenerative capacity. Quite a lot is known about the development of the retina, and many of the genes that control fate specification and neural differentiation in this region of the CNS are similar to those in the brain and spinal cord. The various types of neurons in the retina are generated during development by a multipotent progenitor cell that can undergo multiple rounds of cell division over the course of neurogenesis (Turner and Cepko 1987; Wetts and Fraser 1988). The different types of retinal cells are generated in a rough sequence, with ganglion cells made first, followed by cone photoreceptors and horizontal cells, then amacrine cells, rod photoreceptors, and bipolar cells. The progenitors that do not differentiate into neurons acquire a glial phenotype at the end of the neurogenesis phase; these are called Müller glia (MG) and serve many of the same functions astrocytes perform in the rest of the CNS.

In mammals and birds, most of the development of the retina occurs in the embryonic phase of development; however, in amphibians and fish with an extended larval stage, a zone of retinal stem cells exists at the border between the retina and the ciliary body, the ciliary marginal zone (CMZ) (Fig. 1). The CMZ allows fish and amphibian retina to grow throughout life to match overall body growth (Hollyfield 1968; Straznicky and Gaze 1971). In frogs, for example, during the tadpole stages, nearly 90% of the retina is generated by the CMZ zone, with only a fraction of the retina generated in the embryonic phase (Reh and Constantine-Paton 1983). The more central regions of retina are functional during the larval phase and the new retinal cells added at the CMZ are seamlessly integrated with the existing retinal circuitry. The CMZ in fish and amphibians is spatially organized with uncommitted stem cells lying in the furthest periphery and more committed retinal progenitor cells (RPCs) residing closest to neural retina (Hitchcock and Raymond 1992; Perron et al. 1998). Although nearly all of the retina develops before hatching in birds, chickens, and quails (and likely other avian and reptile species), these species have a small number of proliferating cells at their CMZ, but only a very small amount of the bird retina arises from these cells (Fischer and Reh 2000; Kubota et al. 2002; Todd et al. 2016b). In adult mammals, there is little or no proliferation at the junction between the retina and the ciliary epithelium, and with little evidence for CMZ stem cells in adult mammals (Kubota et al. 2002; Moshiri and Reh 2004; Bélanger et al. 2017).

The eye has long been of interest to biologists studying regeneration. More than 200 years ago, it was discovered that certain amphibians could regenerate parts of their eyes, including the retina, and several very interesting discoveries concerning regeneration were made in ocular tissues (Moshiri et al. 2004). More recently, a resurgence of regenerative medicine and embryonic stem cell biology has led to a renewed focus on the possibility of restoring vision to blind individuals through regeneration of damaged or diseased cells. Major blinding disorders, like glaucoma, macular degeneration, and diabetic retinopathy, all cause visual impairment because of the loss of one or more types of retinal neurons. Patients with macular degeneration, for example, lose their central vision from degeneration of the cone photoreceptors. Therefore, an impetus exists to understand how retinal repair happens successfully in other vertebrates and whether these lessons can be applied to the ultimate aim of vision restoration in humans.

In this review, we discuss the various cellular strategies that different vertebrates have developed to restore retinal cells after injury. We discuss evidence for transdifferentiation and dedifferentiation, two phenomena thought to underlie the regeneration process in nonmammalian vertebrates. We also review recent studies, primarily in fish, that have examined the anatomical and functional fidelity of neural circuit reconstruction. Although molecular details of these naturally occurring reprogramming processes are only beginning to be understood, several attempts are being made to use the knowledge gained from nonmammalian species to promote similar cellular reprogramming in mice, and we will end with a section on progress in this area.

RESTORATION OF NEURONS AFTER INJURY: WHERE DO NEW CELLS COME FROM?

Neurons are terminally postmitotic and do not serve as a source for their own replacement. In all known cases of retinal regeneration, a nonneuronal cell in the retina undergoes one or more rounds of mitotic cell division to generate the additional cells needed for repair. Depending on the species, either the RPE and/or the MG serve as the source of new neurons. In both cases, these cell types retain the ability to proliferate, and there is evidence that mitotic growth factors (epidermal growth factor [EGF], fibroblast growth factor [FGF]) are sufficient to drive this process (Moshiri et al. 2004; Wilken and Reh 2016). In addition to their ability to proliferate in adult organisms, the RPE cells and the MG retain the ability to “dedifferentiate” or “transdifferentiate” into progenitor cells resembling those present in embryonic stages of development, essentially “winding the clock back.” The process of neurogenesis from these neoprogenitors is apparently similar to that seen in development, although there are also likely some differences, because the environment of the tissue is clearly different.

RPE

The RPE is the primary source for regeneration in amphibians such as salamanders and larval frogs (Fig. 2). The RPE in these species responds to retinal removal by dedifferentiating, losing pigmentation, detaching from the basement membrane, and proliferating as neurogenic progenitors (Reh and Nagy 1987; Mitashov 1996; Del Rio-Tsonis and Tsonis 2003). These progenitors go on to replace the neural retina as well as reestablish a functional RPE (Stone 1950; Reh and Nagy 1987). However, the loss of the RPE phenotype is not total, as new neurons transiently retain RPE markers (Klein et al. 1990). Thymidine labeling and transgenic lineage tracing has been used to establish that RPE reenter the cell cycle and are the source of newly derived retina in amphibians (Mitashov 1996; Casco-Robles et al. 2016). The molecular events underpinning amphibian RPE dedifferentiation are not well understood. RPE dedifferentiation is accompanied by loss of mature RPE morphology and up-regulation of progenitor makers such as N-cadherin, Klf4, Pax6, and Sox2 (Reh and Nagy 1987; Islam et al. 2014).

Figure 2.

Retinal regeneration in amphibian. (A) In response to complete removal of the retina, salamanders and tadpoles are able to mount a complete regenerative response. In a cartoon of the retina, ganglion cells (yellow), amacrine cells (pink), bipolar cells (brown), horizontal cells (purple), rods (blue), and cones (magenta) are depicted. (B) Retinal pigmented epithelium (RPE) are the primary cellular source for regeneration in amphibians and respond to injury by dedifferentiating and proliferating as neurogenic progenitors. (C) An entire functional retina can be regenerated by amphibian RPE. (D–F) Representative image of newt RPE (RPE65, red) reentering the cell cycle as progenitors (PCNA, green) in response to retinectomy. (F) Example of RPE-derived retinal regeneration in the newt using RPE65-CreERT2/floxed mCherry-based lineage tracing. (ONL) Outer nuclear layer, (OPL) outer plexiform layer, (INL) inner nuclear layer, (IPL) inner plexiform layer, (GCL) ganglion cell layer. (Panel D is reproduced from the Supplemental movie in Islam et al. 2014, courtesy of a Creative Commons Attribution 4.0 International License. Panels E and F are reprinted from Casco-Robles et al. 2016, courtesy of a Creative Commons Attribution 4.0 International License.) Additional figures created using BioRender.com.

Outside of amphibians, the potential of RPE to generate retinal neurons is limited to early embryonic stages in fish, chicken, and mammals (Coulombre and Coulombre 1965; Sologub 1975; Zhao et al. 1995). Even in embryonic stages, the RPE must be substantially manipulated to induce neural retina in these species. In fish, examples of RPE transdifferentiation occur when embryonic RPE is dissociated from its environment and transplanted into the pericardium (Sologub 1975). RPE does not contribute to the regenerated retina in adult fish, suggesting the reprogramming capacity of RPE in fish is limited (Maier and Wolburg 1979; Knight and Raymond 1995). The RPE in chick and quail does not spontaneously regenerate retina after an injury stimulus. However, at particular embryonic stages, growth factor treatments (particularly FGF) can induce avian RPE to dedifferentiate and reprogram into RPCs (Park and Hollenberg 1989; Pittack et al. 1991). Interestingly, in chick, the RPE-regenerated retina develops in inverted polarity, with photoreceptors facing the lens (Park and Hollenberg 1989). This is likely owing to the fact that the cells do not detach from Bruch's membrane in chicks, as they do in amphibians, and so they retain the inverted polarity of the RPE (with respect to the retina). Unlike amphibians, RPE-derived progenitors do not replace their own RPE, and there is no evidence that regenerated retina in chick embryos ever reaches a functional stage (Coulombre and Coulombre 1965). FGF2 is also able to induce cultured rodent RPE to express neuronal markers in a limited embryonic window, otherwise mammalian RPE cells do not retain a capacity to reprogram into neural progenitors (Zhao et al. 1995).

Müller Glia

It has been known for more than half a century that teleost fish can regenerate neural retina (Lenkowski and Raymond 2014). The cellular source responsible for this regeneration was originally postulated to be constitutive proliferating rod precursors that add rod photoreceptors throughout adulthood (Raymond et al. 1988). In 2006, transgenic lineage tracing established that MG were the cellular source of newly regenerated neurons in the zebrafish retina (Fig. 3; Fausett and Goldman 2006). When fish MG sense tissue damage, they undergo a large shift in their transcriptome and chromatin accessibility to allow for dedifferentiation (Powell et al. 2013; Goldman 2014; Hoang et al. 2020). This leads to reentry into the cell cycle and acquisition of neural progenitor genes (Nagashima et al. 2013). When MG reprogram, they reinitiate developmental gene programs found in early progenitors such as pax6 and atoh7 (Fimbel et al. 2007; Lahne et al. 2021). A large body of studies have begun delineating a complex web of gene regulatory networks and cell signaling mechanisms that control the reprogramming capacity in fish (these studies are thoroughly reviewed by Goldman 2014; Lenkowski and Raymond 2014; García-García et al. 2020; Lahne et al. 2020).

Figure 3.

Retinal regeneration in fish. (A) Retinal injury stimulated by ganglion cell death (NMDA), photoreceptor death (light damage), or a broad neurotoxin (oubain) all lead to Müller glia (MG)-mediated regeneration. Cartoon depicts ganglion cells (yellow), bipolar cells (dark purple), amacrine cells (green), horizontal cells (blue), rods (light purple), and cones (brown). (B) In response to various modes of injury, MG reenter the cell cycle to produce multipotent progenitors. (C) MG-derived progenitors are capable of generating all classes of retinal neurons. (D) Example of lineage-traced MG reentering the cell cycle (BrdU⁺) and (E) up-regulating the proneural transcription factor Ascl1. (Panel E from Wan et al. 2012; reprinted, with permission, from Elsevier © 2012.) (F) Circuitry analysis of regenerated bipolars (xfz43⁺) shows reestablishment of connections with red, green, and ultraviolet (UV) cones. (Panel F from D'Orazi et al. 2016; reprinted, with permission, from Elsevier © 2012.) Additional figures created using BioRender.com.

The first evidence that MG can serve as a source of regenerated neurons came from the chick retina (Fig. 4; Fischer and Reh 2001). In response to excitotoxic injury, chick MG robustly proliferate, down-regulate glial genes, and up-regulate proneural progenitor genes (Fischer and Reh 2001; Hayes et al. 2007; Ghai et al. 2010; Todd and Fischer 2015; Hoang et al. 2020). These reprogrammed MG can persist up to weeks after injury and still express markers of undifferentiated progenitors (Fischer and Reh 2001; Todd et al. 2016a). This suggests that although chick MG can reenter the cell cycle and acquire progenitor characteristics, the majority of reprogrammed MG do not produce neurons, possibly because they do not fully acquire an RPC identity. Without transgenic or growth factor manipulation, mammalian MG are unable to spontaneously reprogram in response to retinal damage (Fig. 4; Dyer and Cepko 2000; Karl et al. 2008; Todd et al. 2015). MG in mouse retina can respond to injury by migrating to the injury site and up-regulating the RPC marker Pax6, but these cells do not fully reenter the cell cycle (Joly et al. 2011).

Figure 4.

Retinal regeneration in chick. (A) NMDA injury that ablates retinal ganglion cells (yellow) and amacrine cells (green) stimulates a regenerative response in posthatch chick retina. (B) In response to damage chick Müller glia (MG) (pink) robustly proliferate and (C) a small portion of these proliferating MG differentiate into amacrine cells (brown). (D) Example of MG (Sox2⁺) proliferating (BrdU⁺) after retinal damage. (Panel D is reprinted from Todd et al. 2016a, courtesy of a Creative Commons Attribution 4.0 International License.) (E) Proliferating chick MG up-regulate progenitor genes like Ascl1 (Cash1). (ONL) Outer nuclear layer, (INL) inner nuclear layer, (IPL) inner plexiform layer, (GCL) ganglion cell layer. Scale bar, 50 μm. (Panel E from Fischer and Reh 2001; reprinted, with permission, from Elsevier © 2001.) (F) Representative image of a regenerated Brdu⁺ MG-derived HuC/D⁺ amacrine cell in damaged chick retina. (Panel F is reprinted from Todd et al. 2016a under the terms of the Creative Commons CC BY license.) Additional figures created using BioRender.com.

CMZ

As noted above, in some amphibians and fish, the retina continues to grow throughout the life span, along with the rest of the body, and this additional growth is mediated by the CMZ. The continued addition of new neurons at the retinal margin and their integration into the retinal circuit implies that this tissue has maintained a developmental state into adulthood. This may require that more central regions of the retina also retain some features of the developing retina because the axons of newly generated ganglion cells likely use developmental cues to navigate to their targets in the brain. Although the regeneration of the retina in amphibians occurs primarily from the reprogramming of the RPE cells to RPCs, the CMZ responds to retinal injury by increasing its production of new neurons to contribute to the repair process (Maier and Wolburg 1979; Reh and Nagy 1987; Raymond et al. 1988; Miyake and Araki 2014). Ultimately, the contributions of new neurons from the RPE (in amphibians) or the MG (in fish) fuse with those derived from the CMZ to make a complete retina. In goldfish retina, for example, the peripheral retina, presumably regenerated from the CMZ, has a higher density of cones than the central retina that was regenerated from MG (Stenkamp et al. 2001). However, both in fish and amphibians CMZ plays a secondary role as the source for regenerated neurons.

Birds do contain CMZ progenitors; however, their proliferative and neurogenic capacity is substantially reduced compared to amphibians and fish (Fischer and Reh 2000). Growth factor treatments can increase neurogenesis in the chick CMZ (Fischer et al. 2002; Todd et al. 2018). CMZ cells in the bird do not seem to be responsive to injury, suggesting they do not contribute to damage-induced regeneration (Fischer 2005). In mouse, live imaging has revealed a subset of neurons derived from a CMZ-like region in development (Marcucci et al. 2016). Lineage tracing experiments found at the retinal margin a distinct progenitor subtype expressing Msx1 gives rise to both neural retina and ciliary epithelium (Bélanger et al. 2017). However, this population of progenitors does not persist into the postnatal stages in mouse retina (Kubota et al. 2002; Moshiri and Reh 2004; Bélanger et al. 2017). A set of studies reported that in vitro, pigmented cells from the ciliary body could give rise to neural progenitors with the competency to generate retinal neurons (Ahmad et al. 2000; Tropepe et al. 2000). Follow-up studies cast doubt on whether these ciliary epithelial-derived cells are competent to generate neurons (Cicero et al. 2009). New high-resolution genomic methods, such as single-cell RNA-seq, should be able to resolve this controversy as these techniques have proven capable of capturing and characterizing rare cell populations in retina (Tran et al. 2019; Kölsch et al. 2021) Therefore, if a population of progenitors persists at the retinal margin of mammalian species, these unbiased approaches should reveal their existence.

The above survey highlights the fact that neurogenic progenitors, similar to those that are present during normal development, can be generated after retinal injury from two different differentiated cell types: the MG and the RPE cells. The cells that serve as the source for regeneration in nonmammalian vertebrates are still present in mammals. However, in mature mammals, both the MG and the RPE lack a high degree of proliferation ability, and after a transient phase during early development, they lack the ability to spontaneously reprogram to progenitor cells. What is holding them back? Several mitotic inhibitors, including p27kip and p53, normally repress MG and RPE proliferation after injury (Dyer and Cepko 2000; Levine et al. 2000; Ueki et al. 2012; Ul Quraish et al. 2016). Circumventing these barriers may be critical for replacing large numbers of retinal cells after injury. In addition to their reluctance to proliferate after injury, the mammalian RPE and MG lack the pliability to lose terminally mature features and revert back to an earlier developmental state, allowing acquisition of neural progenitors and the recapitulation of retinal development. The fact that this endogenous reprogramming ability is present at very early developmental stages in mammals and birds has led to some attempts to define factors responsible for the loss in reprogramming competence. A few factors have been shown to restrict RPE transdifferentiation, including the FGF signaling pathway and TGF-β related factors, such as activin (Sakami et al. 2008; Fuhrmann et al. 2014; Casey et al. 2021). Recent reports also suggest some epigenetic changes to the chromatin may restrict reprogramming to a progenitor-like state in mature bird and mammalian RPE (Dvoriantchikova et al. 2019; Luz-Madrigal et al. 2020). Importantly, both the ability to proliferate and the potential to dedifferentiate/reprogram in the nonneuronal cells are key properties that allow them to serve as a source for new neurons, and both these features are vestigial in birds and mammals.

HOW COMPLETE IS REGENERATION IN FISH AND AMPHIBIANS?

Generation of All Retinal Cell Types

Various types of injury to the retina require the regeneration of different types of retinal neurons. This is true experimentally, with selective types of cell ablation, but this is also true clinically, in which different retinal diseases lead to the loss of different types of retinal neurons (e.g., glaucoma leads to RGC loss), but spares other retinal cell types. As noted in the beginning section, during development, there is a defined sequence for generation of all the types of retinal cells. This temporal production of diverse cell types is controlled by a “clock” in development that appears to control the transition of the progenitor from making early cell types to making later-born cell types (Reh and Kljavin 1989; Wallace 2011; La Torre et al. 2013). These transitions through time are accompanied by the regulated expression of a cascade of transcription factors (TFs) that drive neural fate acquisition and ultimately subtype diversification. After retinal injury, this sequential cell production needs to be restarted if early cell types are to be restored. Is the clock turned back all the way so that both early and late cell types can be generated and in sequence? Are the same TFs used by RPCs reinitiated?

In amphibians, regeneration after retinal removal is complete in the sense that once the RPE undergoes reprogramming to an RPC-like state, an entirely new retina is formed, with appropriate lamination and apparently normal ratios of the various cell types. The ability of amphibians to regenerate all types of neurons in the retina suggests that the RPE cells reprogram to an early progenitor state, although this has not been verified by the characterization of the transcriptome of these cells. The RPE-derived progenitor cells likely regenerate retina in the same order as development (i.e., ganglion cells first) (Cheon et al. 1998).

In fish, the MG-derived progenitors can regenerate all classes of retinal neurons (Fausett and Goldman 2006; Bernardos et al. 2007). The order of cell genesis in the zebrafish has been characterized during development and follows the same pattern as other vertebrates. Studies of Atoh7, which marks the early developing RGCs, have shown these are the first cells generated in zebrafish retina (Kay et al. 2001). During regeneration, MG in fish are able to induce progenitor-associated TFs like Atoh7, Pax6, Islet1, and Otx2 (Sherpa et al. 2008; Goldman 2014; Lahne et al. 2021). The temporal order of proneural TFs that occurs in development is recapitulated in MG-derived progenitors, suggesting fish MG are able to “wind back” their molecular clock to a developmental progenitor state (Lahne et al. 2021). Lin28 is a developmental “clock” regulator that is conserved from Caenorhabditis elegans to mammals (Ambros 2011; La Torre et al. 2013). Lin28 regulates cell fate choice across developmental time by modulating miRNAs. During zebrafish retinal regeneration, MG up-regulate Lin28, which represses the miRNA let7, allowing for induction of pluripotency genes in MG (Ramachandran et al. 2010). Taken together, these studies suggest the ability of fish MG to generate all retinal cell types is likely because of their ability to fully reprogram to an early progenitor state, and recent molecular characterization of the MG-derived progenitors is consistent with this possibility (Hoang et al. 2020).

Chick MG are able to robustly proliferate after retinal injury; however, only a small percentage of the MG-derived cells ultimately differentiate into retinal neurons and they are exclusively interneurons (Fischer and Reh 2001; Hayes et al. 2007; Todd et al. 2016a). Why are avian MG fate-restricted to regenerate predominantly amacrine-like cells? Proliferating chick MG up-regulate some RPC genes such as Ascl1, Pax6, and Klf4 (Fischer and Reh 2001; Todd and Fischer 2015). However, chick MG do not turn on Otx2, a TF critical for photoreceptor fates (Todd et al. 2018). This lack of induction of the full repertoire of progenitor genes likely explains the fate restriction of avian MG.

Comparing the spontaneous reprogramming of nonneuronal cells to an RPC in fish, amphibian, and birds indicates that cells with a very different starting point can converge in their gene expression to an RPC fate. Although we do not know at a detailed level that the progenitors derived from MG in fish and those derived from RPE cells in amphibians are exactly the same; operationally, they both regenerate all the various cell types of the retina, apparently in the same order and ratios as they do in development. It will be interesting to see whether this is true at the molecular level, but results from a recent comparative study of fish, birds, and mice indicate that MG-derived progenitors in fish are transcriptionally similar to those in embryogenesis (Hoang et al. 2020). The relatively recent discovery that some amphibians can also regenerate retinal neurons from MG-derived progenitors will allow direct comparative transcriptional profiling between these different sources of progenitors (Langhe et al. 2017).

Generating a Functional Circuit

For the regenerated retina to function properly and restore vision, the cell types must be generated in proper ratios to one another and synapse with the appropriate partners. This is even more of a challenge in pathology in which inflammation, loss of cells, and remodeling provide a hostile landscape for new cells to wire into established networks or form de novo circuitry.

In fish, the cell morphology and ultrastructural features of regenerated ganglion cells, bipolar cells, and photoreceptors are similar to their original counterpart (Hitchcock and Cirenza 1994; Cameron and Carney 2000; D'Orazi et al. 2016; McGinn et al. 2018). Only a small handful of studies have investigated the integration of regenerated neurons into existing circuitry. Early studies using gap channel tracers showed regenerated patches of retina integrated with the old uninjured circuitry (Hitchcock 1997). More recent investigations have focused on particular regenerated neuron subtypes and whether their stereotypic connection pattern is restored in regeneration. Horizontal cell types will preferentially rewire with their selective cone partner subtype during regeneration; however, if the absence of that partner persists, they will rewire into abnormal circuitry (Yoshimatsu et al. 2016). This indicates there may be a time limit to establish proper connectivity before abnormal circuitry develops. Despite regenerated bipolar cells regaining their proper morphology and synapsing with their normal photoreceptor partners, their connectivity bias between photoreceptor subtypes is not reestablished (D'Orazi et al. 2016). This suggests that although regenerated circuity resembles the native zebrafish retina, cues for proper circuit formation are not fully reengaged in regeneration.

Restoration of visual function after inducing blindness has been shown in the fish retina. Injections of oubain, an ATPase inhibitor, destroys RGCs and quickly leads to visual deficits (Sherpa et al. 2008). After allowing the fish retina to regenerate for 100 days, visual behaviors such as place preference, predator escape response, and dorsal light reflex can be reestablished by regenerated retina (Mensinger and Powers 1999; Sherpa et al. 2008, 2014). Although these gross behaviors recover, regenerated fish retinas do not reach the same amplitude as uninjured controls when tested with retinal ERGs (Mensinger and Powers 1999; Lindsey and Powers 2007; McGinn et al. 2018). The lack of full recovery in ERGs is not surprising because regenerated neurons do not recover the same density they had before injury. Neither the density of regenerated bipolar cells nor their plexiform layer thickness reaches that of normal levels (D'Orazi et al. 2016; McGinn et al. 2018), nor do regenerated cones reach normal density or mosaic organization (Vihtelic and Hyde 2000; Stenkamp et al. 2001; Stenkamp and Cameron 2002; D'Orazi et al. 2020).

The differences in the numbers and ratios of retinal cells in regenerated zebrafish retina may be caused by the lack of appropriate regulation of their generation by the progenitors. Zebrafish MG can regenerate all types of retinal neurons, but do so indiscriminately. For example, specific ablation of RGCs or photoreceptors will lead to MG production of all types of retinal neurons, including interneuron types that were not ablated (D'Orazi et al. 2016, 2020; Powell et al. 2016). This invariably will lead to nonphysiological ratios of cell types. In addition, although all photoreceptors types can be regenerated by fish MG, the proper ratio between types is not reestablished (D'Orazi et al. 2020). These supernumerary neurons last at least 30 days, arguing against dynamic feedback to cull new cells (Powell et al. 2016). Although MG do not selectively replace the neuron type lost, evidence in amphibians suggest that biased cell replacement can occur from the CMZ (Negishi et al. 1982; Reh and Nagy 1987; Tyler et al. 2005). The difference may lie in the fact that the CMZ is a part of the normal growth of the retina, rather than a response to injury, but better characterization of CMZ progenitors with those derived from RPE or MG might shed some light on differences in their response to the changing cellular microenvironment.

Taken together, the data from fish suggest that a rebuilt retina does not have to perfectly correspond with its normal counterpart to be useful in restoring vision to an animal. The photoreceptors and inner retinal circuit must reach a level in which enough of a circuit regenerates to provide detection, initial processing, and appropriate relay of visual information to the RGCs. The RGCs must then relay their information to the brain via the axons through the optic nerve. Newly regenerated ganglion cells must be able to navigate the distance to the brain, a distance much longer than what they had to traverse during embryonic development. A thorough review of RGC axon regeneration is beyond the scope of this review; however, in general, those animals that show effective cellular regeneration in the retina also have maintained their ability to sustain reconnection of the new cells with visual brain nuclei.

OUTLOOK ON ENGINEERING REGENERATION IN MAMMALS: SYNTHETIC BIOLOGY FOR RETINAL REGENERATION

In humans, death of retinal neurons is permanent and leads to irreversible blindness. The fact that some vertebrates are able to regenerate retina to restore vision has led to hope that investigators can recapitulate this phenomenon in mammals and ultimately human patients. Over the past decade, interest in endogenous reprogramming of MG in mammals has intensified.

Rather than regenerating neurons, mammalian MG respond to injury by up-regulating inflammatory cytokines, fibrotic proteins, and intermediate filaments (Bringmann et al. 2009). Early strategies used exogenous growth factors to coax rodent MG into a regenerative phenotype and these studies showed evidence that a small amount of MG-mediated neurogenesis could occur in vivo in the rodent retina (Ooto et al. 2004; Osakada et al. 2007; Karl et al. 2008). Interestingly, the background of mouse strain used partially determined the efficacy of MG cell cycle reentry (Suga et al. 2014). These early studies established the potential of mammalian MG to reenter the cell cycle as progenitors and possibly regenerate neurons; however, these strategies were only minimally successful, with only a small number of possible neurons reported, little characterization of the cells, and no evidence for function or synaptic connections.

Using Transcription Factors for MG Reprogramming

Engineering mammalian MG to express developmental TFs has proven the most promising strategy to date. Comparisons between the transcriptomes of MG and RPCs revealed that many of the TFs normally expressed in developing progenitors (Rax, Lhx2, Pax6, Vsx2) are retained into adulthood in mouse MG (Roesch et al. 2008; Nelson et al. 2011). However, key proneural and neurogenic genes needed for neurogenesis during normal development are not expressed in mature MG, either in quiescent conditions or after injury (Karl et al. 2008). These results suggested that expressing proneural TFs in MG might return them to an RPC state, leading to neural regeneration. As discussed above, something very much like this occurs spontaneously in fish, and to a more limited extent in birds. In zebrafish, drastic shifts in TF networks occur in MG during regeneration (Ramachandran et al. 2010, 2012). A recent comparative study found in zebrafish, chick, and mouse retina that MG respond to retinal damage by rapidly altering their transcriptomes and many of these changes occurred in TFs (Hoang et al. 2020). A key difference between species likely is that TFs get up-regulated in response to injury (i.e., proneural vs. inflammatory). An essential regeneration-associated TF network in fish is controlled by the proneural basic helix-loop-helix TF Ascl1 (Ramachandran et al. 2010, 2011).

Both the developmental studies and the fish regeneration studies suggested that a key difference between MG and progenitors was the expression of Ascl1; this gene is expressed in RPCs in both fish and mice, and up-regulated in MG after injury in fish and birds, but not mice. Could overexpression of this TF reprogram mouse MG to an RPC identity? This was first shown in vitro, in which cultured mouse MG infected with Ascl1 lentivirus reprogrammed into a progenitor state that could produce retinal-like neurons (Pollak et al. 2013). This finding led to creation of a transgenic mouse that specifically overexpressed Ascl1 in MG in vivo. Similar to Ascl1's effects on MG in vitro, MG overexpression of Ascl1 proved capable of leading to regeneration of retinal neurons in vivo (Ueki et al. 2015). However, this depended on retinal injury and the age of the animal. At early postnatal ages, Ascl1 was able to effectively reprogram MG in damaged retinas into neurogenic progenitors capable of generating bipolar and amacrine neurons. In adult retinas, Ascl1 overexpression combined with retinal injury was insufficient to stimulate neural regeneration (Ueki et al. 2015; Jorstad et al. 2017). Epigenetic changes in promoter/enhancer accessibility were hypothesized to explain the loss in competency of MG for Ascl1-induced neurogenesis. Therefore, in a follow-up study, Ascl1 overexpression was combined with an epigenetic modifier (histone deacetylase inhibition), and this proved to be effective in regenerating retinal neurons in vivo in adult mice (Jorstad et al. 2017). Detailed analysis using scRNA-seq, ATAC-seq, electron microscopy (EM), and electrophysiology showed that these regenerated neurons were molecularly similar to endogenous neurons, physiologically functional, and integrated into retinal circuits (Jorstad et al. 2017, 2020). These findings were proof-of-principle that MG can be targeted to serve as a source for endogenous regeneration in the mammalian retina (Fig. 5).

Figure 5.

Retinal regeneration in mouse. (A) In the normal mouse retina, Müller glia (MG) respond to retinal damage by undergoing an inflammatory processed called “gliosis.” Cartoon depicts ganglion cells (yellow), bipolar cells (dark purple), amacrine cells (green), horizontal cells (blue), rods (light purple), cones (brown), and MG (orange). (B) The combination of MG-specific Ascl1 overexpression, retinal injury, and trichostatin-A (TSA) treatment leads to MG-mediated regeneration of bipolar and amacrine neurons in the adult mouse retina. (C) Example of MG-derived (GFP⁺) bipolar neurons (CABP5⁺) that were regenerated through proliferating MG (EdU⁺). (D) Patch-clamp electrophysiology recordings have shown that MG-derived neurons can respond to light, suggesting circuit integration. (E) Single-cell RNA-seq analysis of MG-mediated regeneration show transitional states between MG, MG progenitors, and MG-derived neurons. (Panels C–E from Jorstad et al. 2020; reprinted, with permission, from Elsevier © 2020.) Additional figures created using BioRender.com.

The finding that experimentally driving the expression of Ascl1 in MG enables these cells to regenerate new retinal neurons in adult mice has stimulated new research in this area, using alternative TF combinations to direct the MG to other cell fates. Ultimately, this approach might provide a way to repair the human retina after injury or disease. Determining how close these in vivo “reprogramming” strategies are to the regeneration that occurs naturally in animals like fish and urodele, will require additional investigation.

Key Questions to Address for In Vivo Reprogramming Strategies

First, are the neurons in the recent mouse regeneration reports really newly born in the adult retina? For many years, it was thought that neurogenesis was not present in adult mammals. However, in studies of the hippocampus and olfactory bulb, colabeling of neuronal markers with [3H]-thymidine (or the analogs, BrdU or EdU) that is incorporated into dividing progenitors, was established as a “gold standard” to identify newly born neurons (Altman 1962; Kuhn et al. 2018). Early studies using exogenous growth factors to stimulate MG neurogenesis confirmed that the small number of neurons generated were indeed from MG that reentered the cell cycle (Ooto et al. 2004; Karl et al. 2008). Recent reports have used EdU labeling to show that overexpression of factors such as β-catenin, Lin28, or Ascl1 or the loss of factors such as NFIa/b/x or Hippo-signaling can lead to MG proliferation in the in vivo damaged adult mouse retina (Yao et al. 2016, 2018; Elsaeidi et al. 2018; Hamon et al. 2019; Rueda et al. 2019; Jorstad et al. 2020; Todd et al. 2020). In the case of the neurons generated from Ascl1-reprogrammed MG, many arise from mitotic divisions, but not all can be labeled with EdU, suggesting that some proportion of the MG-derived neurons arise from direct transdifferentiation of the glial cells. If MG undergo transdifferentiation to generate neurons, this precludes the use of Edu/BrdU to show the newly derived nature of regenerated neurons. Nevertheless, although it is not possible to conclude that all presumptive regenerated neurons are in fact newly generated from mitotic divisions, it is possible to conclude that many are.

Second, are the new neurons really generated from MG? Classically, in developmental biology, the use of lineage tracing strategies coupled to the reprogramming factor confers the ability to track the progeny of a cell, or even its transdifferentiation. When the lineage tracing is performed using a transgenic approach, the reagents can be rigorously defined in control studies. Groups have found both EdU⁺ and EdU^– neurons with MG-specific lineage tracing and interpreted this as MG undergoing both direct transdifferentiation and transitioning through an intermediate cycling progenitor (Jorstad et al. 2017, 2020; Hoang et al. 2020). However, in cases in which viral vectors are used for the lineage tracing and experimental overexpression of reprogramming factors, some disconnect between the processes of reprogramming and lineage may occur. Studies solely relying on AAV-mediated lineage tracing have come into question as technical artifacts of capsid and promoter specificity can lead to erroneous cell fate conclusions (Wang et al. 2020; Blackshaw and Sanes 2021). In these cases, we suggest the combination of well-established transgenic lineage-tracing strategies, combined with scRNA-seq analysis, can reveal the proper cellular source of regenerated neurons (Blackshaw and Sanes 2021). The advent of scRNA-seq technologies has given investigators an unprecedented window to observe molecular transition states of reprogramming cells. scRNA-seq has been leveraged to observe in vivo MG reprogramming, where, in response to Ascl1, NFI knockout, or Hippo inactivation, transition states can be seen where MG lose their typical glial gene expression, up-regulate neural progenitor genes, and then ultimately begin to express markers of immature neurons (Jorstad et al. 2017, 2020; Rueda et al. 2019; Hoang et al. 2020; Todd et al. 2020). This technology can be used to ensure that newly derived neurons in the mouse retina arise from transitioning MG rather than unintended labeling of endogenous neurons.

Third, can all retinal cell types be generated from mouse MG? Unlike fish, in which MG reprogram into multipotent progenitors, strategies stimulating MG reprogramming in mice typically find MG are restricted to one or two cell fates. Studies that have used transgenic lineage tracing, EdU labeling, and scRNA-seq have found MG are able to regenerate bipolar-like and amacrine-like cells in the in vivo adult mouse retina (Jorstad et al. 2017, 2020; Elsaeidi et al. 2018; Hamon et al. 2019; Hoang et al. 2020; Todd et al. 2020). These reports used either Ascl1 overexpression, NFIa/b/x knockout, Lin28/Ascl1 overexpression, or Hippo inactivation. It is unclear why these diverse strategies all seem to lead to an interneuron fate, although it is possible that the common driver is Ascl1, because the deletion of NFI family TFs in MG leads to an increase in Ascl1 in the cells (Hoang et al. 2020). Other TFs may therefore be needed to stimulate the MG to adopt other neuron fates. For example, regeneration of retinal ganglion cells and rod photoreceptors from mouse MG in vivo has been reported using other TFs and AAV-mediated strategies. In the case of rods, a cocktail of AAV viruses was used to first transduce MG with β-catenin to reprogram the cells to an RPC-like state, and then to fate specify MG to rods by overexpressing Otx2, Nrl, and Crx (Yao et al. 2018). For RGCs, AAV viruses carrying a CRISPR-CasRx cassette to knock out the RNA-binding protein Ptbp1 from MG was reported to transdifferentiate them into the RGC fate (Zhou et al. 2020). Although these studies suggest that combinations of TFs may be needed to generate diverse cell types from MG, an important caveat is that no examples of EdU-labeled neurons or intermediate molecular states were shown in these reports. In addition, there is evidence that the AAV vectors used to target MG lack the necessary specificity for this cell type, and direct infection of presumed progeny neurons by the AAV is possible (Wang et al. 2020). Follow-up experiments using these TF combinations with better controls will be needed to validate these studies.

Fourth, can newly generated neurons integrate into circuity and become functional? The initial report that Ascl1 overexpression combined with epigenetic modulation stimulated regeneration in the adult mouse retina attempted to assay circuit integration of these newly regenerated bipolar cells (Jorstad et al. 2017). Synaptic structures in regenerated neurons was shown at the ultrastructural level using serial block face EM combined with GFP-lineage tracing (Jorstad et al. 2017). Furthermore, in two reports, in vivo patch clamp electrophysiology has been used to show light responses of newly regenerated bipolar neurons suggesting they integrate into host circuitry (Jorstad et al. 2017, 2020). As noted in the previous section, functional recovery does occur in zebrafish after ouabain injury, but even after 200 days of regeneration, the b-wave of the ERG remains only at 50% of the pre-lesion value or global retinal function; evidence from patch clamp recordings showed that the light responses of the MG-derived bipolar neurons were <50% of normal bipolar responses in the same retinas. Future studies will hopefully provide better characterization of functional recovery and potentially visual behavioral assessment.

CONCLUDING REMARKS

In the past few decades, we have learned a great deal about normal vertebrate retinal development, as well as the process of retinal regeneration in zebrafish. The application of the lessons from developing and regenerating systems has allowed a proof-of-principle that functional regeneration of neurons is possible in the mammalian retina. However, the devil is in the details, and we have only scratched the surface of what needs to be performed for translation of this approach to one day treat human retinal degenerations. The fact that amphibians and fish are so effective at replacing their retina, like we would heal a skin lesion or a broken bone, shows what is possible in biology, but it will take a much deeper understanding of the underlying gene regulatory networks to rigorously take control of the process and create a synthetic biology for regeneration.