Losing, preserving, and restoring vision from neurodegeneration in the eye

SUMMARY

The retina is a part of the brain that sits at the back of the eye, looking out onto the world. The first neurons of the retina are the rod and cone photoreceptors, which convert changes in photon flux into electrical signals that are the basis of vision. Rods and cones are frequent targets of heritable neurodegenerative diseases that cause visual impairment, including blindness, in millions of people worldwide. Here, I summarize the diverse genetic causes of inherited retinal degenerations (IRDs) and their convergence onto common pathogenic mechanisms of vision loss. Currently, there are few effective treatments for IRDs. I will discuss how recent advances in disparate areas of biology and technology (e.g., genome editing, viral engineering, 3D organoids, optogenetics, semiconductor arrays) enable promising efforts to preserve and restore vision in IRD patients, with implications for neurodegeneration in less approachable brain areas.

INTRODUCTION

Before discussing how IRDs impair vision and how vision might be preserved and restored, it is worth considering how the healthy retina supports vision. This helps link the patterns of photoreceptor dysfunction and death to clinical IRD presentations and defines limits on the visual functions that different approaches can preserve and restore.

The human retina contains one type of rod and three types of cones, preferring short- (S-cones), medium (M-cones), and long-wavelength (L-cones) light, respectively¹. Rods function in dim light, whereas cones function in bright light (Figure 1A). Two types of horizontal cells provide separate feedback to rods and cones (Figure 1B)^2,3. Photoreceptor signals are relayed from the outer to the inner retina by second-order neurons called bipolar cells⁴. There are twelve types of bipolar cells in the human retina; one selectively contacts rods, and 11 receive cone input^1,5. One cone bipolar cell type selectively contacts S-cones; the others prefer L- and M-cones^6–8. Bipolar cells can be grouped into ON (six cone and one rod bipolar cell types) and OFF (five cone bipolar cell types) subclasses, which depolarize to light increments and decrements, respectively. In the inner retina, ON and OFF bipolar cell axons stratify in separate layers and provide input to specific sets of amacrine (~25 types) and ganglion cells (~17 types), the output neurons of the eye^1,5,9. The increasing neural diversity and circuit complexity from the outer to the inner retina transforms the pixel representations of photoreceptors, whose activity is inversely proportional to the light they absorb, into feature representations of ganglion cells, varying in complexity from local luminance and chromatic contrast signals to direction-selective maps¹⁰.

Figure 1. Organization of photoreceptors and retinal circuits.

(A) Rod and cone photoreceptors translate changes in photon flux into electrical signals. These signals originate in the outer segments and propagate through the connecting cilium and the inner segment past the nucleus to the synaptic terminals to control glutamate release from ribbon synapses.

(B) Retinal circuit schematic. Light enters the retina from the ganglion cell side and passes through the circuit before the photoreceptor outer segments absorb it. The retinal pigment epithelium (RPE) phagocytoses aged outer segment discs (~10% per day) and participates in chromophore recycling. In the outer plexiform layer (OPL), horizontal cells (HCs) provide feedback to rods and cones. Bipolar cells (BCs) relay photoreceptor signals from the outer to the inner retina. The axons of ON and OFF bipolar cells, which depolarize to light increments and decrements, respectively, stratify at different depths in the inner plexiform layer (IPL). Bipolar cell axons provide input to amacrine (AC) and ganglion cells (GC), the output neurons of the eye. Müller glia (MG) provide metabolic support, take up and recycle neurotransmitters, and function as light guides.

High-acuity vision in humans is mediated by the macula, a specialized region of the retina, where the density of ganglion cells and cones rises sharply, peaking in the fovea^1,11. The circuits and computations in the fovea are simpler than in the peripheral retina. Foveal midget ganglion cells connect to a single bipolar cell contacting a single cone¹². Foveal midget ganglion cells receive little inhibition, and their responses resemble a spiking cone¹³. Thus, central vision, important for pattern and object recognition, may be easier to preserve and restore than peripheral vision, which is critical for mobility.

John Dowling pronounced the retina, with its organized structure, clear structure-function relationships, exposition, and isolation in the eye, an approachable part of the brain for understanding the nervous system function¹⁴. The same qualities, combined with monogenic inheritance and pathogenic convergence, make IRDs an approachable part of neurodegeneration, with clear paths for translating insights into pathogenic mechanisms and technological advances into patient benefits.

LOSING VISION

Neurodegeneration in the eye mostly affects the photoreceptors, disrupting light detection, or the retinal ganglion cells, disrupting the communication of visual information to the brain. The most common causes of photoreceptor death and dysfunction are IRDs and age-related macular degeneration. This review focuses on IRDs, because we understand their pathogenesis comparatively well, and they serve as a testing ground for efforts to preserve and restore vision. Excellent reviews of ganglion cell^15–18 and age-related macular degeneration^19–21 are available. Some therapeutic strategies developed for IRDs will likely translate to other forms of neurodegeneration in the eye and beyond.

Genetic and phenotypic diversity

Inherited retinal degenerations (IRDs) are characterized by dysfunction and progressive loss of photoreceptors^22–24, causing visual impairment, including blindness, in approximately 1 in 3,000 (i.e., >2.5 million) people worldwide^25,26. Unlike most neurodegenerative diseases, IRDs often manifest during adolescence and early adulthood^23,24, increasing the individual burden and socioeconomic impact^27–29.

IRDs are predominantly inherited as monogenic traits and constitute the most common and heterogeneous group of monogenic diseases^23,24. So far, mutations in >270 genes have been shown to cause IRDs³⁰. Multiple mutations can disrupt each IRD gene. Thus, >150 disease-causing mutations have been identified in rhodopsin, the photopigment of rods³¹, and >600 have been mapped to the ATP-binding cassette transporter (ABCA4), which clears all-trans-retinal from rods after rhodopsin excitation^32,33. Despite the vast genetic landscape already mapped, the mutations causing 30%−50% of IRD cases remain unknown^26,34.

Monogenic IRDs can be transmitted in autosomal dominant (15%−20%), autosomal recessive (5%−25%), and X-linked patterns (5%−15%); the remaining IRDs (35%−50%) appear sporadic^23,24,35,36.

Autosomal recessive IRDs typically result from loss-of-function mutations, as in the case of RPE65, a retinoid isomerase involved in chromophore recycling³⁷. All X-linked IRD alleles identified so far are recessive, the most common being loss-of-function mutations in the RPGR (Retinitis Pigmentosa GTPase Regulator) gene, a component of the photoreceptor connecting cilium^36,38. In X-linked IRD families, males are affected, females exhibit mild or no symptoms, and no male-to-male transmission is observed³⁶.

Autosomal dominant inheritance can result from haploinsufficiency, gain-of-function, and dominant-negative mutations. For example, loss-of-function mutations in GCAP1 (Guanylate Cyclase Activating Protein 1) attenuate negative feedback to cyclic nucleotide-gated (CNG) channels in photoreceptors and cause autosomal dominant IRDs via haploinsufficiency^39,40. Gain-of-function mutations that substitute proline at position 23 for histidine (P23H) lead to rhodopsin misfolding and ER accumulation, which triggers rod degeneration and autosomal dominant IRDs^31,41. Dominant-negative mutations are a specific subset of gain-of-function mutations⁴², in which mutant proteins compete with wild-type proteins for binding to a target (e.g., mutations in the cone-rod homeobox transcription factor CRX that preserve DNA binding but disrupt transactivation⁴³) or sequester wild-type proteins into non-functional complexes (e.g., mutations in peripherin-2, which normally forms homodimers and higher-order oligomeric complexes to support membrane disc formation in photoreceptor outer segments⁴⁴).

Apparently sporadic IRDs can result from de novo mutations in the affected individuals (i.e., truly sporadic) or incomplete penetrance of an inherited allele. However, the most common cause of sporadic IRDs is compound heterozygosity (i.e., inheritance of different mutant alleles of the same gene from each parent)⁴⁵. The prevalence of compound heterozygosity reflects the extraordinarily high carrier frequency of IRD alleles (1 in 2–4 people)^46,47. Mutations in ABCA4 alone are found in 5%−10% of the population^48,49. Finally, high carrier frequency and pathogenic convergence can result in digenic inheritance of IRDs^50,51.

Different mutations in the same gene can cause IRDs with different inheritance patterns and clinical presentations^23,24. Clinical presentations also vary between patients with the same mutation, even in the same family, reflecting differences in environmental factors and genetic background, a phenomenon known as variable expressivity or incomplete penetrance^52,53.

The clinical presentations of IRDs, and their classification, reflect differences in the patterns of photoreceptor dysfunction and degeneration^23,24. IRDs can primarily target rods (retinitis pigmentosa; RP) or cones (cone and cone-rod dystrophies; COD and CORD) or affect both simultaneously (Leber’s congenital amaurosis; LCA)²³. Instead of progressive degeneration, some IRDs manifest as stationary dysfunctions of rods (congenital stationary night blindness; CSNB) or cones (achromatopsia)²³. Some IRDs are restricted to the cone-rich macula. The most common inherited macular dystrophy is Stargardt disease, which frequently results from loss-of-function mutations of ABCA4 and affects 1 in 6,500 people⁴⁷. Finally, IRDs can be part of multi-system disorders (i.e., syndromic IRDs), typically named after their discoverers (e.g., Bardet-Biedl syndrome and Usher syndrome)^54,55.

Pathogenic convergence

Mutations in the >270 known IRD genes converge to disrupt a few key signaling pathways, cell structures, and processes to cause photoreceptor degeneration (i.e., pathogenic convergence). Indeed, reverse IRD genetics uncovered many of the molecular mechanisms supporting these pathways, structures, and processes. Pathogenic convergence organizes the genetic diversity of IRDs and provides hope for mutation-agnostic therapies to preserve vision. This section reviews disruptions of gene regulatory networks, primary cilia, phototransduction, the visual cycle, proteostasis, and synapse function in the convergent pathogenesis of IRDs (Figure 2).

Figure 2. Losing vision - pathogenic convergence.

(A) Gene regulatory networks control the differentiation of retinal precursor cells into rod and cone photoreceptors. These networks also help maintain cell fate and survival.

(B) The connecting cilium links outer segments (specialized sensory cilia) to the inner segment by the connecting cilium, where it is anchored via the basal bodies. The axoneme forms the microtubule backbone of the connecting cilium.

(C) Phototransduction cascade converts photon absorption by the photopigment (i.e., opsin plus chromophore) into an electrical signal by controlling cyclic nucleotide-gated channels in the outer segment plasma membrane.

(D) After photoisomerization from 11-cis-retinal (11cRAL) to all-trans-retinal (atRAL), the chromophore is recycled in a multi-step process involving enzymes in different color-coded cell types. Whereas rods rely entirely on the visual cycle through the RPE, cones can utilize 11-cis-retinol (11cROL) produced by an alternative visual cycle through the Müller glia.

(E) Proteins are synthesized and folded in the endoplasmic reticulum (ER). Misfolded proteins in the ER trigger stress responses (unfolded protein response and heat shock response). Misfolded proteins and toxic aggregates can be cleared by the ubiquitin-proteasome system (UPS) and degraded by the lysosomes (including via autophagy).

(F) Complex pre- and postsynaptic machinery and transsynaptic complexes regulate signal transmission from photoreceptors to bipolar cells. The molecular mechanisms of this process are best understood at the synapse between photoreceptors and ON bipolar cells (including rod bipolar cells) illustrated here.

Gene regulatory networks

In the gene regulatory networks that drive photoreceptor differentiation and maintain their identity, OTX2 commits retinal precursor cells to the photoreceptor lineage by regulating the expression of CRX, another homeobox factor (Figure 2A)^56–58. CRX is required for rod and cone differentiation and maintenance^59–61. In rods, CRX interacts with NRL, a basic leucine zipper transcription factor, to promote rod-specific gene expression while suppressing cone-specific genes^62–64. NRL directly activates the expression of the orphan nuclear receptor NR2E3, which in turn, further represses cone-specific genes and fine-tunes the expression of rod-specific genes^65–69. In cones, CRX cooperates with TRβ2, the thyroid hormone receptor β2, to activate cone-specific genes and promote L- and M-cone fate^70,71. Mutations in these transcription regulators (except for TRβ2) can cause IRDs. Because most of them (directly or indirectly) regulate the expression of rod- and cone-specific genes, mutations can manifest as RP, COD, CORD, or LCA⁴³. While some mutations result in congenital visual deficits (LCA) resulting from developmental failures, others cause degeneration later in life (RP, COD, CORD), highlighting the importance of the gene regulatory networks in maintaining photoreceptor fate and function⁵⁸.

Primary cilia

Primary cilia are hair-like organelles found on the surface of most eukaryotic cells, composed of a microtubule-based structure called the axoneme surrounded by a ciliary membrane^38,72. The ciliary membrane differs from the plasma membrane, containing specific proteins and lipids^38,72. Cilia function as sensory antennas, detecting signals such as growth factors, hormones, and mechanical cues^38,72. The photoreceptor outer segment is a specialized primary cilium (Figure 2B) that captures light and translates it into electrical signals (i.e., phototransduction). Phototransduction occurs in stacks of outer segment membrane discs rich in photopigments (rhodopsin in rods, cone opsins in cones) and their downstream signaling cascade^73,74. The outer segment axoneme is connected to the cell body through a connecting cilium anchored in the inner segment by the basal body^38,72. Photoreceptors shed aged discs from the tips of their outer segments (up to 10% per day), requiring continuous disc renewal⁷³. As a result, ~1–2 μm² of membrane and 2,000 rhodopsin molecules pass through the rod connecting cilium every minute⁷⁵. Proteins and lipids are exchanged between the inner and outer segments by intraflagellar transport³⁸. Mutations that disrupt ciliary biogenesis, maintenance, and intraflagellar transport cause ciliopathies. Mutations in >240 genes are known to cause ciliopathies, and mutations in >50 are associated with IRDs^38,72. Ciliopathies can manifest as isolated IRDs but, given the wide-ranging functions of primary cilia, often feature retinal degeneration in multi-organ syndromes (e.g., Bardet-Biedl syndrome encompasses retinal degeneration, obesity, polydactyly, intellectual disability, and kidney abnormalities, while Usher syndrome causes vision and hearing loss)^55,76.

Phototransduction

The photopigments in the outer segment membrane discs consist of the chromophore 11-cis-retinal and a G protein-coupled receptor (GPCR; rhodopsin, L-, M-, and S-opsin). Upon light absorption, the chromophore isomerizes to all-trans-retinal, inducing an activating conformational change in the opsin moiety (Figure 2C)^74,77. The activated photopigment triggers a G-protein called transducin (Tαr in rods and Tαc in cones)⁷⁸, which in turn activates phosphodiesterase (PDE6αβ heterodimers in rods and Pde6α’α’ homodimers in cones), an enzyme that lowers the levels of cyclic guanosine monophosphate (cGMP)^77,79. The decline in cGMP concentration closes CNG channels in the photoreceptor cell membrane. CNG channels are heterotetramers composed of three CNGA1 and one CNGB1 subunit in rods and three CNGA3 and one CNGB3 subunit in cones^80–82. Closure of these channels blocks sodium and calcium influx and hyperpolarizes the photoreceptor. Activated photopigments recruit GPCR kinases (typically, GRK1 in rods and GRK7 in cones)⁸³. The GRKs phosphorylate the photopigment, which is subsequently bound by arrestin (arrestin-1 in rods and arrestin-4 in cones), preventing further interactions with transducin and terminating the phototransduction cascade^84,85. Some mutations that block phototransduction cause stationary dysfunction. Because most molecular components differ between rods and cones, these dysfunctions manifest either as congenital stationary night blindness (e.g., mutations in the genes encoding rhodopsin, Tαr, PDE6β, GRK1, and arrestin-1)⁸⁶ or achromatopsia (e.g., mutations in the genes encoding Tαc, PDE6α’, CNGA3, and CNGB3)⁸⁷. Mutations of cone opsins result in dichromacy (loss of L-opsin: protanopia, loss of M-opsin: deuteranopia, loss of S-opsin: tritanopia). Red-green colorblindness (protanopia or deuteranopia) is common among males (8%−10% in the USA) because of the proximity and frequent recombination of the L- and M-opsin genes on the X chromosome⁸⁸. Dichromacy, unlike achromatopsia, is not considered an IRD. Gain-of-function mutations (e.g., in genes encoding rhodopsin, Tαr, PDE6α, and PDE6β) and loss-of-function mutations (e.g., in the gene encoding arrestin-1) can also constitutively activate the phototransduction cascade, fueling progressive degenerations. How persistent phototransduction kills photoreceptors is debated but may overlap with the mechanisms of phototoxicity (including calcium dysregulation), as proposed in the equivalent light hypothesis^89,90.

Visual cycle

The photoisomerization of 11-cis to all-trans-retinal exposes the covalent bond between chromophore and opsin to hydrolysis. Unbound all-trans-retinal is converted back to 11-cis-retinal in a multi-step process known as the visual cycle (Figure 2D)^91,92. In the first steps, all-trans-retinal is transported to the cytoplasmic side of the disc membrane by ABCA4 and reduced to all-trans-retinol (vitamin A) by retinol dehydrogenase (RDH) within the photoreceptor^93,94. All-trans-retinol interacts with a transporter protein called interphotoreceptor retinoid-binding protein (IRBP) to get to the RPE^95,96, where it is esterified by lecithin-retinol acyltransferase (LRAT) to form all-trans-retinyl esters. These esters are substrates for RPE65, which isomerizes and hydrolyzes them into 11-cis-retinol^37,97. Subsequently, 11-cis-retinol is oxidized back to 11-cis-retinal by 11-cis-retinol dehydrogenase (11-cis-RDH)^94,98,99. Finally, 11-cis-retinal is returned to photoreceptor cells via IRBP. Here, it recombines with opsins to form functional photopigments, completing the visual cycle. Rods rely entirely on the visual cycle through the RPE, while cones themselves express 11-cis-RDH and can utilize 11-cis-retinol recycled by Müller glia via an RDH and RGR opsin^92,100,101. In addition to IRBP, cellular retinaldehyde-binding protein (CRALBP) and cellular retinol-binding protein (CRBP) solubilize and transport retinoids between different enzymatic reactions, ensuring efficient conversion and recycling of retinoids in the visual cycle^95,96. Mutations in all components of the visual cycle mentioned here (and several others) can cause IRDs with diverse clinical presentations^30,91. Disruption of the visual cycle can kill photoreceptors through the build-up of toxic intermediates (e.g., accumulation of the lipid-conjugated all-trans-retinal in Stargardt’s patients with ABCA4 mutations and retinyl esters in RP patients with RPE65 mutations)^97,102. In addition, opsins without chromophores activate spontaneously at higher rates than opsins bound to chromophores^103–106. Thus, a lack of 11-cis-retinal from visual cycle disruption continuously activates phototransduction causing degeneration similar to gain-of-function mutations in the phototransduction cascade^89,90.

Proteostasis

The proteostasis network balances the synthesis, folding, trafficking, and degradation of proteins (Figure 2E)¹⁰⁷. It involves chaperones, degradation systems (proteasomes, autophagy), and stress response pathways (unfolded protein response and heat shock response) to prevent misfolded or damaged proteins from accumulating^35,107,108. The continuous disc renewal increases the turnover of outer segment proteins and makes photoreceptors particularly susceptible to mutations that cause protein misfolding^35,109. Rhodopsin makes up more than 80% of the outer segment proteome^110,111, and rhodopsin misfolding is a frequent cause IRDs, including the P23H exchange that is the most common cause of autosomal dominant RP in the USA (15%−18% of cases)^31,41. The unfolded protein and heat shock responses initially counteract ER stress in photoreceptors by reducing protein synthesis, upregulating chaperones to assist in protein folding, and increasing the degradation of misfolded proteins. However, if protein misfolding and ER accumulation overwhelm these responses, their signals switch to pro-apoptotic^{31,35,108,109}. Misfolding mutations of other photoreceptor proteins (e.g., PDE6β and ABCA4) or proteins in the RPE (e.g., fibulin and bestrophin) can similarly cause IRD through persistent ER stress^112–115. Indeed, misfolded protein aggregates and ER stress are common drivers of neurodegeneration throughout the nervous system^116–118.

Synapse function

The electrical signals generated by phototransduction in the outer segment propagate to the synaptic terminals of rods and cones, where specialized protein machinery converts them into chemical signals transmitted to bipolar and horizontal cells (Figure 2F). Mutations that disrupt presynaptic machinery components can cause IRDs with progressive degeneration. For example, mutations in the calcium channel Cav1.4 or RIMS1, a protein that regulates vesicle docking, have been identified in CORD patients^119–121. Mutations in pre- and postsynaptic proteins in rods (Cav1.4, α2δ4, CABP4, and LRIT3) and rod bipolar cells (mGluR6, TRPM1, GPR179, and NYX) can also cause stationary dysfunction (i.e., CSNB). In many neurodegenerative diseases, including IRDs, synapse loss long precedes cell death^122–126. Therefore, synapse maintenance is an important therapeutic goal for preserving visual function. The temporal sequence (i.e., synapse loss before cell death) also suggests that synapse loss and dysfunction may accelerate cell death in IRDs that are not caused by mutations in the synaptic machinery. This hypothesis and whether synapse rescue can be neuroprotective remains to be tested.

PRESERVING VISION

This section covers approaches to prevent, arrest, or slow down photoreceptor degeneration and optimize remaining visual functions. Given space limitations, I focus on recent advances in gene therapy, neuroprotection, and homeostatic plasticity (Figure 3). Excellent reviews covering pharmacotherapy and supportive clinical care are available^127,128.

Figure 3. Strategies for preserving vision.

(A) Gene augmentation delivers genetic materials encoding functional versions of mutant or lost proteins (loss-of-function mutations) through viral carriers. AAVs are the most commonly used carrier for gene augmentation. They deliver single-stranded DNA and are limited in packaging capacity to ~4.7 kb. LVs have greater capacity (~8kb) and carry single-stranded RNA genomes that, after reverse transcription, integrate into the host genome.

(B) Recent advances in gene editing (CRISPR/Cas nucleases, base editors, and prime editors) allow for increasingly precise and versatile corrections of mutations in the host genome, including in postmitotic photoreceptors. The illustrations in this panel are inspired by Raguram et al. (2022)¹⁴⁸ .

(C) Neuroprotective strategies use broadly trophic factors (e.g., CNTF) or leverage insights into the convergent pathogenesis to supply agents that bypass or boost affected pathways. A particular focus in neuroprotection is on preventing secondary cone degeneration in RP, including by delivery of the rod-derived cone viability factor (RdCVF) and its longer splice variant (RdCVFL).

(D) Homeostatic plasticity involving diverse mechanisms across retinal circuits can preserve visual function in the face of partial photoreceptor loss.

Gene therapy

Monogenic inheritance, the accessibility of the retina in the eye, and its immune privilege make IRDs promising targets for gene therapies^129,130. Gene therapies fall into two categories: gene augmentation and gene editing.

Gene augmentation

In gene augmentation (Figure 3A), genetic materials encoding functional versions of lost or mutant proteins are delivered to prevent or slow degeneration. Gene augmentation is a natural fit for IRDs caused by loss-of-function mutations. Genetic materials are most commonly delivered by adeno-associated viruses (AAVs) or lentiviruses (LVs). AAVs carry single-stranded DNA encapsulated by a shell comprising 60 copies of three viral proteins (i.e., the capsid)^131,132. The viral capsid mediates infection by binding to cellular receptors. Naturally occurring capsid variants (i.e., serotypes) have been surveyed, and mutants created to optimize cell targeting for IRDs^132–137. As a result, photoreceptors and RPE cells can be reliably infected by subretinal AAV injections^132–137. However, subretinal injections risk retinal detachment and bleeding, accounting for most adverse events in IRD gene therapy trials^138,139. Intravitreal injections are clinically routine and low-risk, but intravitreally injected AAVs infect few cells in the primate retina^140,141. In addition to more viral engineering, suprachoroidal AAV injections may further improve the risk-reward balance of IRD gene therapies^140,141.

Voretigene neparvovec (Luxturna^™) delivers RPE65 to RPE cells by subretinal injection of an AAV2 carrier. It can improve vision in patients with loss-of-function RPE65 mutations (autosomal recessive RP or LCA) and has been in clinical use since 2017 and 2018 in the USA and Europe, respectively^142–145. Benefits from a single injection can last four years, indicating that genes delivered by AAVs, which after nuclear conversion into double-stranded DNA remain episomal, can be stable^143,146. Voretigene neparvovec is the only approved IRD gene therapy to date, but clinical trials of subretinal AAV injections to augment PDE6, PDE6B, CNGA3, CNG3B, RPGR, and other genes are ongoing^138,139.

An important limitation of AAVs is their cargo capacity (~4.7 kb)^147,148. The frequency of loss-of-function mutations correlates with gene size, evidenced by the relationship between transcript length and the number of families affected by X-linked recessive IRDs³⁶. By contrast, the frequency of gain-of-function mutations does not correlate with transcript length, indicating that there are many ways to disrupt a protein but only a few places to enhance or alter its function^36,42. Because of this relationship, some genes frequently targeted by loss-of-function mutations exceed the size limit for AAV-mediated augmentation. The most prominent of these is ABCA4 (coding sequence: ~7 kb), whose >600 identified mutations, collectively, are the most common cause of IRDs¹⁴⁹.

LVs are replication-incompetent retroviruses that can carry a cargo of ~8 kb, which they deliver via an RNA genome that integrates into the host genome after reverse transcription^150,151. Integration carries a risk of insertional mutagenesis^152,153. To minimize this risk, integration-deficient LVs (IDLVs) have been developed^154,155. In RPE65 knockout mice, LV- and IDLV-mediated gene augmentation can preserve vision similar to AAV-mediated gene augmentation^154,156. However, while LVs infect RPE cells broadly after subretinal injection, they infect fewer photoreceptors (development > adults and primates > mice)^157–159. The tropism of LVs depends on the glycoproteins in their host-derived membrane envelope^160,161. LVs can be produced with different envelope glycoproteins (i.e., pseudotyping). LVs currently tested for IRD gene therapy are pseudotyped with the vesicular stomatitis virus glycoprotein^154,162. Exploring different pseudotyping strategies may increase the reach of LV-mediated treatments, including ABCA4-augmentation for Stargardt’s disease^149,158,163.

Gene editing

Gene editing promises to expand the reach of gene therapy, allowing more mutations to be targeted and enabling permanent corrections and potential cures^164,165. I will briefly summarize three gene editing approaches (CRISPR-Cas nuclease editing, base editing, and prime editing), their potential, and their recent applications to IRDs (Figure 3B).

CRISPR-Cas nuclease editing harnesses a naturally occurring bacterial defense mechanism^166,167. The system involves two components: a single-guide RNA (sgRNA) and a Cas nuclease (commonly Cas9). The sgRNA guides the Cas nuclease to a specific DNA sequence, complementary to the sgRNA, in the target genome^166,167. The Cas nuclease then introduces a double-stranded break at the target site, which the cell’s endogenous machinery repairs, either via non-homologous end joining or homology-directed repair^166,167. Homology-directed repair is inefficient in non-dividing cells like photoreceptors. Non-homologous end joining frequently introduces small insertions and deletions (indels)¹⁶⁸. CRISPR-Cas can, thus, be used to disrupt IRD mutations with indels. Targeting gain-of-function rhodopsin mutations, CRISPR-Cas editing enhanced photoreceptor survival and visual function in mouse and rat models^169,170. This strategy, combined with gene augmentation in a ‘knockout and replace’ approach, is slated for clinical trial¹⁷¹. CRISPR-Cas nucleases have also been used to target an intronic mutation in CEP290 that causes the inclusion of cryptic exon, producing a non-functional ciliary protein in LCA 10¹⁷². Following success in animal models¹⁷², this approach was tested in a phase1/2 trial, with favorable safety profiles and encouraging signs of treatment efficacy^173,174.

A significant limitation of CRISPR-Cas nucleases in IRD therapy is that the editing outcomes are uncontrollable. Different indels could have different effects on disease progression¹⁴⁸, and the double-stranded breaks introduced by the CRISPR-Cas nucleases occasionally result in large deletions, translocations, and other chromosomal abnormalities raising safety concerns^175–177.

Base editors overcome many limitations of the CRISPR-Cas nuclease system by allowing accurate gene correction via single-nucleotide conversions in genomic DNA without the need for double-stranded breaks¹⁷⁸. Base editors combine a catalytically impaired Cas nuclease, which via a sgRNA, binds a specific DNA sequence but does not cleave it with a base-modifying enzyme. This enzyme converts one nucleotide to another directly at the target site. Since the initial development of cytosine base editors, which convert C·G base pairs into T·A, and adenine base editors, which convert A·T pairs into G·C, additional base editors have been created^165,178. Adenine base editors have been used to correct a premature stop in RPE65 in an LCA mouse model, slowing photoreceptor degeneration and providing lasting improvements in visual function^179,180.

Base editors can correct many pathogenic single-nucleotide polymorphisms¹⁷⁸. However, they cannot perform all possible single-nucleotide conversions nor mediate targeted insertions or deletions. Prime editors were created to address these limitations¹⁸¹. They comprise a catalytically impaired Cas nuclease fused to a reverse transcriptase enzyme and a prime editing guide RNA (pegRNA). The pegRNA directs the prime editor to the target DNA site and encodes the desired edit. The Cas nuclease nicks one DNA strand, and the reverse transcriptase uses the pegRNA as a template to synthesize the edited DNA sequence. The cell’s repair machinery then incorporates the edited sequence into the genome, achieving the desired genetic modification. Prime editors have been used successfully to correct the same premature RPE65 stop, previously targeted with adenine base editors^179,180,182 and a PDE6B missense mutation using a Cas variant with an expanded DNA binding scope¹⁸³. Both prime edits preserved vision in the respective mouse models^182,183.

Gene editing systems generally exceed the packaging limit of AAVs. Thus, they were either split into two AAVs or delivered by LVs in the studies mentioned in the previous paragraphs. However, the stable viral expression needed for gene augmentation is a risk for gene editing, as the likelihood of off-target edits accumulates over time. Therefore, non-viral delivery methods are being explored. Virus-like particles that use viral capsids to package and deliver editing systems as ribonucleoproteins hold particular promise^184,185. A single subretinal injection of virus-like particles carrying an adenine base editor in LCA-model mice achieved similar on-target editing efficiencies compared to lentiviral delivery while reducing off-target editing¹⁸⁴.

Neuroprotection

Alongside efforts to correct disease-causing mutations, strategies to safeguard photoreceptors against death, regardless of the specific mutation (i.e., neuroprotection), are being developed (Figure 3C).

Neuroprotective approaches apply broadly trophic factors or use knowledge of IRD pathogenesis to boost or bypass the function of affected signaling pathways, cell structures, and processes. Neurotrophic factors (including CNTF, bFGF, and BDNF) can slow photoreceptor degeneration after light damage and in IRD mouse models^186,187. CNTF has made it farthest toward the clinic. Delivered by AAVs or encapsulated cells, CNTF protected photoreceptors in mouse and dog IRD models^188,189. Encapsulated CNTF was found to be safe in humans¹⁹⁰ but failed to slow the decline of visual function in RP patients¹⁹¹.

Insights into pathogenic convergence gave rise to neuroprotective retinoid therapies for IRD mutations in the visual cycle. Exogenous chromophores can protect photoreceptors threatened by disruptions in chromophore recycling⁹¹. This approach uses 9-cis-retinal rather than 11-cis-retinal, as it is easier to synthesize, more stable, and combines with opsins to form functional photopigments^192,193. After success in mouse and dog models^194–197, 9-cis-retinal entered clinical trials and was shown to improve visual function in pediatric LCA patients with loss-of-function mutations in LRAT and RPE65¹⁹⁸. Similarly, transgenic increases in proteasome activity can dramatically improve photoreceptor survival in an IRD model with rhodopsin misfolding (P23H mice)^199,200. Reduced ER stress and improved protein folding also contribute to the protective effects of tauroursodeoxycholic acid, an FDA-approved compound found naturally in the bile of hibernating bears¹²⁸. Because of the pathogenic convergence, these pathway-specific mutation-agnostic approaches could apply to larger groups of IRD patients than mutation-specific gene therapies.

We light the world around us to engage our high-acuity colorful cone vision at all times. As a result, photoreceptor deficits are most disruptive when they affect the cones. Patients with stationary rod dysfunction (i.e., CSNB) live relatively unperturbed, whereas progressive rod degeneration (i.e., RP), which accounts for a majority IRDs (1 in 4,000 people), is a debilitating disorder²⁶. This is because rod dysfunction has a limited impact on cones, but when rods die, the cones eventually follow suit.

The following observations begin to explain the secondary cone death and spawned efforts to prevent it. First, it was found that as rods die, cones begin to starve²⁰¹. Cone starvation and degeneration can initially be overcome by insulin injections and mTOR activation, which increase glucose uptake into cones^201,202. Second, rods secrete a trophic factor called rod-derived cone viability factor (RdCVF), whose depletion causes cones to die^203,204. Third, RdCVF released by rods binds Basigin-1 on cones to increase glucose import via GLUT1²⁰⁴. Thus, the loss of RdCVF accounts, at least partly, for the starvation of cones as RP progresses. RdCVF is generated by alternative splicing of nucleoredoxin-like 1 (Nxnl1) transcripts. Nxnl1 also gives rise to a longer protein, RdCVFL, containing a thioredoxin domain that protects cones against the hyperoxia they experience with fewer rods around^205–207. AAV-mediated delivery of RdCVF and RdCVFL rescues cones in RP mouse models and has entered phase 1/2 clinical trials²⁰⁷.

Another clever approach for saving cones leverages insights into the gene regulatory networks that govern photoreceptor differentiation. Without NRL, precursors destined to be rods become cone-rod hybrids (i.e., cods)^62,208. Cods downregulate many rod-specific genes and, therefore, are protected from degeneration caused by mutations in these genes. This transdifferentiation works, to a large extent, even when NRL is removed transgenically or by AAV-mediated CRISPR-Cas editing in adults^209,210. Transdifferentiation preserves vision in RP mouse models because cods produce RdCVF and, thus, support the cones^209,210.

Homeostatic plasticity

Adjustments in retinal circuits that maximize information transmission from the remaining photoreceptors can help preserve visual function. Homeostatic plasticity refers to a circuit’s drive to return to a functional set point following perturbation and the mechanisms that accomplish this^211,212. The retina has an extraordinary capacity for functional homeostasis, involving plasticity throughout the circuit (Figure 3D)^212,213. Cones undergoing secondary degeneration can maintain light responses even as they lose their outer segments²¹⁴. Bipolar cell dendrites can recruit new partners and scale up existing synapses to compensate for the partial loss of photoreceptors^215,216. Finally, changes in the inner retina can maximize the visual information encoded in the retinal output^217–220. However, there are obstacles to functional homeostasis in IRDs. First, homeostatic plasticity declines with age. Thus, photoreceptor loss that is asymptomatic in young mice causes visual deficits in older mice²¹⁵. Second, as bipolar cells lose photoreceptor input, circuits in the inner retina develop spontaneous high-frequency oscillations that obscure the remaining light responses^221–223. Third, as degeneration progresses, Müller glia proliferate, hypertrophy, and form scars (i.e., reactive gliosis) that prevent neural rewiring^224,225. These obstacles must be overcome to unlock the therapeutic potential of homeostatic plasticity in IRDs.

RESTORING VISION

In the late stages of IRDs, when most photoreceptors are gone, innovative approaches to restoring vision are being explored (Figure 4). These either try to replace photoreceptors from external or internal sources (i.e., photoreceptor replacement), make downstream neurons light-sensitive (i.e., optogenetics), or substitute the function of photoreceptors or the retina with prostheses and cognitive assistants (i.e., visual prostheses and cognitive assistants).

Figure 4. Strategies for restoring vision.

(A) In the late stages of IRDs, when most/all photoreceptors are gone, efforts to replace them from external and internal sources are underway. In external cell replacement, photoreceptor progenitors are derived from retinal organoids. In internal cell replacement, Müller glial reprogramming could allow for photoreceptor regeneration.

(B) Optogenetic approaches to vision restoration convert second or third-order neurons into light sensors. A variety of optogenetic actuators, including microbial opsin (e.g., ChrimsonR), animal opsins, chimeras (e.g., Opto-mGluR6), and photosensitized glutamate receptors (Li-GluR) are being tested.

(C) Retinal prosthesis can be implanted subretinally or epiretinally and either receive data and power from external sources or be directly driven by light (i.e., photovoltaic arrays). In addition to retinal prostheses, cortical prostheses, and cognitive assistants may restore vision and replace visual functions, respectively.

Photoreceptor replacement

Like all retinal neurons, photoreceptors are born from multipotent retinal precursor cells (RPCs), which undergo temporal changes in competence to generate the different neuron classes in overlapping waves of development^226,227. The progenitors of rods and cones (i.e., photoreceptor progenitors or PRPs) are born at different times (rod PRPs ~P4, cone PRPs ~E15.5 in mice) and mature over the following weeks (in mice) and months (in humans)²²⁸. RPCs, PRPs, and mature PRs have all been transplanted in xenogenic (between species) and allogenic (within species) experiments (Figure 4A). Consistently, PRPs survived best, maintained their fate, and showed some ability to integrate into host circuits^228,229.

In humans, PRPs are born during the second trimester, making primary PRPs for therapeutic photoreceptor replacement impractical. The isolation of embryonic stem cells (ESCs)^230,231 and the ability to de-differentiate somatic cells into induced pluripotent stem cells (iPSCs)^232,233 promised to eliminate the need for primary PRPs. However, it proved difficult to differentiate ESCs and iPSCs into PRPs in 2D cultures²³⁴. This changed with the introduction of 3D retinal organoids^235,236, which replicate retinal development and generate outer nuclear layers that, in principle, provide an unlimited source of PRPs for transplantation^237–239.

The transplantation of PRPs as tissue sheets (i.e., outer nuclear layer sections isolated from organoids) improves graft survival over the injection of single-cell suspensions into the subretinal space^237,240,241. In addition to tissue sheets, increasingly complex biopolymer scaffolds seeded with RPCs or PRPs are being developed to optimize transplant delivery and survival^242–244.

Once delivered PRPs (or RPCs) need to mature and integrate into host circuits. Initial results were encouraging. After subretinal transplantation of fluorescent PRPs into mice, many fluorescent photoreceptors with mature morphologies and synaptic connections to bipolar cells were observed, and visual function was restored to blind hosts^{241,245–247}. However, it was subsequently realized that rather than PRPs integrating into host circuits, host photoreceptors were labeled and function recovered by material transfer from the immature grafts^248–252. What (RNA and/or proteins) and how (tunneling nanotubes²⁵³ and/or extracellular vesicles²⁵⁴) materials are exchanged between cells are interesting questions with potential therapeutic implications, but for photoreceptor replacement from external sources, the focus has shifted to promoting (and correctly assessing) the integration of PRPs into host circuits²²⁸.

Some non-mammalian vertebrates, including zebrafish and chicks, can replace degenerating photoreceptors from within. The source of this regeneration are Müller glia, the last cells born from RPCs during development^226,255,256. Müller glia in regenerating species can de-differentiate, re-enter the cell cycle, and undergo asymmetric divisions in which one daughter replaces the original Müller glia (Figure 4A); the other gives rise to a neuron^255,256. Regeneration can be triggered by indiscriminate injuries or cell-type-specific degenerations^257,258. For photoreceptor degenerations, regeneration outcomes are biased toward cell-type replacement^258,259. Thus, when all zebrafish cones are ablated, they regenerate in approximately their normal proportions, whereas regeneration is biased toward the lost cell type after selective red-cone and UV-cone removal²⁵⁸.

In mammals, Müller glia can also proliferate in response to photoreceptor degeneration. However, this proliferation fails to produce neurons, resulting instead in gliosis and scarring^224,260,261. Why and how mammalian Müller glia lost their neuroregenerative capacity is unclear^262,263. Therapeutic efforts to restore this potential try to translate insights into the degeneration-induced reprogramming of Müller glia from regenerating species to non-regenerating species or turn back time in the maturation of mammalian RPCs to Müller glia^263–266. As a result, some late-born neuron types (i.e., bipolar cells and glycinergic amacrine cells) can now be regenerated from Müller glia in mice^263–267 (more spectacular successes have been reported, but need to be reexamined for technical reasons^268,269). The regeneration of cones, among the first retinal neurons to be born, will likely pose significant challenges. Nonetheless, internal and external photoreceptor replacements hold considerable promise for future vision restoration.

Optogenetics

The discovery of microbial light-gated ion channels and pumps (i.e., microbial opsins) that can be expressed in mammalian neurons to control their activity (i.e., optogenetics) raised the possibility that when photoreceptors die, downstream neurons could be turned into light detectors (Figure 4B)²⁷⁰. Indeed, in one of the first demonstrations of optogenetics, Zhuo-Hua Pan’s group showed that viral delivery of channelrhodopsin-2 could restore light responses to retinal ganglion cells and downstream brain areas of blind mice²⁷¹. Since then, remarkable progress has been made in exploring diverse opsins, developing alternative light-sensitizing strategies, and delivering optogenetic actuators to different retinal cell types to optimize optogenetic vision restoration, which has entered clinical trials²⁷².

In microbial opsins, the absorption of a photon directly gates a single ion channel or pump. By contrast, in vertebrate GPCR opsins, photon absorption is amplified by a second messenger cascade allowing a single photon to close hundreds of CNG channels²⁷³. Photoreceptors also express vertebrate opsins at extraordinarily high levels. Thus, each rod contains 10⁸-10⁹ rhodopsin molecules²⁷³, whereas optogenetic responses suggest that bipolar and ganglion cells can express 10⁴-10⁵ microbial opsins, similar to the ~10⁴ melanopsin molecules in intrinsically photosensitive ganglion cells²⁷⁴. Because of these discrepancies, optogenetic light responses are orders of magnitude less sensitive than photoreceptor-driven light responses^271,275,276 and require external stimulus amplification through goggles for clinical translation²⁷⁷. For channelrhodopsin-2 (i.e., the first microbial opsin widely studied), whose absorption peaks at 470 nm, the light levels needed for vision restoration exceed biological safety limits and risk photochemical damage^278,279. The risk of photochemical damage decreases exponentially with the light’s wavelength^278,279. Therefore, current microbial opsin-based efforts to restore vision focus on naturally occurring and engineered variants with red-shifted absorption^277,278.

To benefit from signal amplification, several studies have ectopically expressed rhodopsin and cone opsins in bipolar and ganglion cells^280–282. This enhanced the light sensitivity of retinal responses and visual behaviors compared to microbial opsins, eliminating the need for signal-amplifying goggles at the cost of slower kinetics^280–282. Another potential drawback of rhodopsin and cone opsins is their dependence on the visual cycle; microbial opsins recycle their chromophores internally²⁸³. This is particularly relevant to IRDs caused by disruptions of the visual cycle and late stages of other IRDs when RPE cells undergo secondary degeneration^91,224,225.

To further optimize opsins for vision restoration, several chimeras have been created. Opto-mGluR6 combines the light-sensing parts of melanopsin, whose therapeutic potential is also being explored separately^284,285, with the signal transduction parts of mGluR6²⁸⁶. After delivery into ON bipolar cells, which are used to amplifying mGluR6 signals, Opto-mGluR6 can mediate retinal light responses with fast kinetics and high sensitivity and guide visual behaviors^286,287. Similarly, the extracellular domains of rhodopsin and cone opsins have been fused to the transmembrane and intracellular domains of melanopsin to combine the preferable kinetics of the former with the improved coupling to the ubiquitous G_q proteins by the latter²⁸⁸.

As an alternative to microbial opsins, other ion channels can be sensitized to light via synthetic compounds. Like microbial opsins, these allow for faster kinetics than vertebrate opsins. However, a disadvantage is that the photosensitizing compounds must be continuously supplied to the ion channel. Maleimide-azobenzene-glutamate tethered mutant GluK2 ionotropic glutamate receptors (Li-GluRs) are the most developed of these actuators and have been shown to restore retinal and behavioral light responses after expression in bipolar and ganglion cells in mice^289–291.

One important consideration for optogenetic vision restoration is which cells to target. In the outer retina, signals from photoreceptors diverge into parallel bipolar cell channels that interact with amacrine cells in the inner retina to generate the diverse feature representations sent to the brain via ganglion cell spike trains^4,10. The further upstream optogenetic actuators are expressed, the more likely they are to restore retinal feature representations. AAV-mediated expression of halorhodopsin, a light-activated microbial chloride pump, in cones restored light responses in an RP mouse model and human retinal explants²⁷⁶. However, vision restoration via cones will be transient unless their secondary degeneration can be stopped. Another appealing target for optogenetic vision restoration are bipolar cells, which survive in late-stage IRDs^224,225. ON bipolar cells are particularly attractive for vision restoration because they can be selectively targeted for AAV-mediated expression and drive light responses in the OFF pathway via cross-over inhibition^134,275,292. Indeed, targeting microbial opsins or Opto-mGluR6 to ON bipolar cells can restore some ganglion cell feature representations and behavioral responses that rely on them^220,275,286.

At least four phase 1/2 clinical trials of optogenetic vision restoration are currently ongoing²⁷², and promising results from one patient receiving an intravitreal AAV injection targeting foveal ganglion cells for expression of ChrimsonR, a red-shifted microbial opsin, have been reported²⁷⁷.

Visual prostheses and cognitive assistants

Before optogenetics, electrical stimulation was the primary way of activating neurons at will. In retinal prostheses, electrode arrays are implanted on the subretinal side, contacting bipolar cells (when photoreceptors are gone), or on the epiretinal side, contacting ganglion cells (Figure 4C)²⁹³. In addition to choosing a side, differences in the stimulation patterns, including anodic vs. cathodic delivery and pulse duration, allow for selective activation of bipolar or ganglion cells²⁹⁴.

The only retinal prosthesis approved by the FDA is the Argus^® II system (Second Sight Medical Products), which delivers stimuli through a 60-electrode epiretinal array²⁹⁵. The system can restore partial sight to patients with end-stage IRDs, allowing them to perceive light and motion and recognize common objects²⁹⁵. However, no IRD patient achieved a level of vision that would eliminate the need for a white cane or a guide dog for mobility, and in March 2019, Second Sight discontinued the Argus^® II system.

A major challenge for epiretinal prostheses is to stimulate ganglion cell bodies but avoid ganglion cells axons passing between the array and the cell bodies. Axonal stimulation elicits arcuate percepts that severely distort the retinotopic map of the visual field²⁹⁶. Stimulation thresholds are three times lower near the soma than along the axon²⁹⁷. Thus, in principle, newer arrays that can record and stimulate could calibrate current amplitudes at each electrode to avoid axonal stimulation²⁹⁸. Another hope for such bidirectional arrays is that they can selectively stimulate different ganglion cell types, distinguished by their action potential waveforms, to recreate and appropriately distribute retinal feature representations²⁹⁹.

Subretinal prostheses were pioneered with the Alpha IMS system (Retina Implant AG), which contains 1,500 active microphotodiodes, each with its own amplifier and stimulation electrode³⁰⁰. Thus, the system directly senses light from the environment and converts it into local electrical signals delivered to bipolar cells, counting on downstream circuits to extract salient features. This contrasts with epiretinal systems, including the bidirectional arrays currently in development²⁹⁹, which use cameras and external processing units to compute feature representations and deliver them to the retinal output neurons, bypassing upstream circuits. Alpha IMS allowed previously blind IRD patients to read letters, combine them into words, and recognize shapes³⁰⁰. However, many Alpha IMS systems and their successors failed within a year of implantation, and in March 2019, Retinal Implant AG closed³⁰¹.

Systems with conventional stimulation electrodes require an external power supply. In Argus^® II and Alpha IMS, power and data (for Argus^® II) were delivered through wires, complicating surgeries and adding failure points^293,295. Power and data can now be delivered wirelessly by radio frequency or ultrasound transmission, reducing risk and improving stability. Another exciting advance is the development of photovoltaic arrays, which use semiconductor materials found in solar panels to directly convert light into electrical signals, eliminating the need for external power and data^302–305. Photovoltaic arrays also allow for denser stimulation patterns, critical for restoring high-acuity vision³⁰², and can be delivered on flexible supports that unfold, expanding visual field restoration into the periphery, which is crucial for mobility³⁰⁴.

Another possible evolution is photovoltaic nanoparticles^306,307. In IRD mouse models, subretinal injections of photovoltaic nanoparticles restored light responses and some visual behaviors^306,307 (but see³⁰⁸). Conducting (i.e., gold) nanoparticles that emit heat after light absorption have also been combined with the expression of heat-sensitive ion channels to impart infrared light sensitivity to remaining cones in an RP model to could complement the remaining visual function³⁰⁹.

In parallel to retinal prostheses, significant progress has been made in restoring visual function through cortical prostheses^310,311. It has long been recognized that electrical stimulation in the visual cortex of blind patients elicits the perception of light flashes called phosphenes³¹². Using modern brain-computer interfaces (e.g., Utah arrays), efforts are underway to dynamically steer stimuli to shape phosphenes into useful representations, with early successes enabling blind patients to recognize letters^310,311. In addition to IRDs, cortical prostheses are particularly relevant for restoring vision to patients with neurodegenerative diseases of ganglion cells (e.g., glaucoma).

Finally, although outside the scope of vision restoration, it is worth noting the development of cognitive assistants in which machine vision systems map and parse scenes and relay information to blind patients via speech. Early applications have focussed on navigation^313–315, but consumer-driven advances in augmented reality systems and artificial intelligence will likely accelerate and expand progress in this area.

CONCLUSION

IRDs are a genetically heterogenous, collectively common group of heritable neurodegenerative diseases. Over the past decades, vast sections of the genetic IRD landscape have been mapped, and >270 disease-causing genes identified. Yet, the mutations underlying ⅓ - ½ of IRDs remain to be uncovered. Recent advances in gene editing, which allow mutations to be corrected in somatic cells, motivate comprehensive genetic explorations. In addition to increasing the efficacy of delivering editing tools to target cells in the retina, innovation and reform are needed to streamline and incentivize treatment development for rare mutations and ensure wide access to gene therapies³¹⁶. The realization that diverse IRD mutations converge on a few pathogenic pathways to vision loss fuels hopes for mutation-agnostic, pathway-specific treatments that may help large groups of IRD patients, akin to efforts to prevent secondary cone degeneration. Even when photoreceptors are gone, all is not lost, as advances in photoreceptor replacement, optogenetics, retinal and cortical prostheses, and cognitive assistants promise to restore visual functions. The insights and innovations discussed in this review are beginning to make their way into the clinic, and while continued investment in the fundamental science generating these and future advances is critical, there is genuine optimism that benefits to patients afflicted with IRDs are near.