Viruses are the most used vectors and the process by which they infect and release their genetic content into target cells is termed transduction. Several different recombinant viruses, that are replication deficient, can be used to deliver therapeutic nucleic acids, with differences in terms of cargo limits, integration capabilities, transduction efficiency, cellular tropism, and risk of immune responses.

Adenoviruses (Ads) are a family of DNA viruses that can infect quiescent and dividing cells and replicate in the host nucleus without integrating their genome. Adenoviruses have been largely tested as gene therapy vectors, mainly due to their cargo capacity (approximately 8–36 kb) and ability to transduce many cell types. As far as IRDs are concerned, however, conventional Adenoviral vectors (AVs), constructed by substituting the E1 region with the transgene cassette of interest (9), had limited success, owing to the expression of some viral genes in the infected target cells, which enhanced immunogenicity and undermined treatment longevity, even in the immune-privileged environment of the human eye. These issues have been partially addressed with second- and third-generation vectors, characterized by progressive stripping of all viral coding sequences and implementation of helper-dependent AVs. However, problems with contaminating helper viruses, vector instability, and replication-competent AVs have been reported (10, 11).

Adeno-Associated Viruses (AAVs) are defective single-strand (ss) DNA parvoviruses with more than 20 integration sites in the human genome. Recombinant AAVs (rAAVs) vectors are by far the most frequently used ones in gene therapy approaches for IRDs, because of their lack of pathogenicity, favorable immunologic profile (since, unlike AVs, they do not carry any virus open reading frame), non-integrating nature in the absence of rep protein (which minimizes the risk of insertional mutagenesis, unlike LVs), ability to provide a stable transgene expression and extended retinal tropism. To date, 13 naturally occurring serotypes of AAV have been isolated from primates (AAV1–AAV13): different serotypes have a different capsid conformation and different properties, especially as far as tropism is concerned. Moreover, AAVs can be modified in several ways, for example by packaging the viral genome bordering the transgene into the capsid of a different AAV serotype, process known as pseudotyping or cross-packaging (e.g., an AAV2/8 vector is a pseudotype in which the genome of AAV2 serotype is packaged into an AAV8 capsid) (12). Both serotype and pseudotype choice are important to optimize vector design for the target disease. To date, the serotypes and pseudotypes that have been used in clinical trials for IRDs include AAV2/5, AAV2/8, and AAV8. The major disadvantage of rAAV vectors is their limited cargo capacity, which cannot exceed 4.7 kb. Although with an apparently reduced photoreceptor transduction efficiency, dual AAV vectors—each of which contains half of a large transgene expression cassette—have been shown to improve retinal phenotype in murine models of IRDs (13–15).

Lentiviruses (LVs) are retroviruses with a larger packing capacity (8 kb), which makes them a compelling alternative to AAV vectors for those IRDs whose causative gene coding sequence exceeds the 4.7 kb limit, such as ABCA4-related Stargardt's disease and MYO7A-related Usher's syndrome type 1B. So far, the retroviral variant of human immunodeficiency virus type 1 (HIV-1) and the equine infectious anemia virus (EIAV) have been studied for IRDs. Lentivirus vectors have two main drawbacks. First, LVs are integrating in nature and genomic integration, if on the one hand leads to a sustained expression of the foreign DNA, on the other hand carries a risk of insertional mutagenesis (16, 17). Such risk may not be justified in the case of IRDs, since in post-mitotic tissues, like the retina, a stable transduction can be achieved even by lentiviral episomes. This limitation can be overcome by employing integration-deficient lentiviral vectors (IDLVs), which have been successfully used in a rodent model of retinal degeneration (18). Second, LVs are capable of effectively transducing RPE cells and only to a lesser extent, which is generally insufficient for therapeutic purposes, differentiated photoreceptors (19, 20).

Non-viral delivery systems have some advantages over viral delivery systems, including potentially unlimited cargo capacity, simultaneous conveyance of multiple therapeutics, low immunogenicity, and inexpensive manufacturing procedures.

Non-viral delivery systems use physicochemical agents to compact the DNA and/or transport it across the membranes.

Physical methods include, for instance, sonoporation (21), and electroporation (22) (i.e., the use of ultrasound or electricity to temporarily increase cell permeability) and direct injection of DNA into target cells (23), respectively, offering no protection from enzymatic degradation of therapeutic nucleic acids.

Chemical agents, which protect the payload from the action of nucleases, include, among others, cationic liposomes, lipopolyplex, and nanoparticles (NPs) (24–27).

Though some preclinical gene therapy studies have successfully used non-viral DNA systems (28–30), these agents are hard to export to an in vivo clinical setting, mainly because of transiency of gene expression (31, 32), ultimately resulting in a relatively inefficient delivery (9).

Since its initial conceptualization back in the 1960s (39), the idea of gene therapy was based on the straightforward assumption that monogenic recessive disorders could be cured by replacing a faulty gene with a normal copy of it delivered through therapeutic vectors. This approach is called gene augmentation (or gene replacement) because the synthesis of the protein is augmented and its function, at least partially, restored. The gene of interest can be delivered as DNA or mRNA. Though having the advantage of not requiring delivery into the nucleus, thereby reducing the risk of integration into the host genome, the mRNA platform is far more immunogenic and less stable than the DNA platform, which is therefore the preferred one for ocular gene therapy (40). Most clinical gene therapy trials, as well as the first, and currently only, FDA- and EMA-approved treatment for an IRD, voretigene neparvovec-rzyl (Luxturna), rely on gene replacement, that does not require modification of native DNA and therefore is particularly compelling, owing to its simple design and relative ease of investigation.

Gene augmentation is an established approach for recessive monogenic disorders, but it is not suited to AD diseases resulting from pathological gain-of function mutations. In these cases, the therapeutic goal is to prevent the altered gene from expressing and encoding an aberrant protein that would interfere with normal cell function. To do so, it is possible to adopt several different approaches, which, schematically, can act at three levels: DNA, mRNA, and the intermediate process in between them (i.e., transcription).

In all of these cases, the host nucleic acids can be targeted in two ways, that is in a mutation-dependent fashion, whereby specific allele inhibition is sought in order to allow the expression of the wild-type copy of the gene, or in a mutation-independent fashion, in which a combined approach with gene augmentation is mandatory, since both copies of the gene (the mutated and the functional one) are silenced and replaced by a non-silenced wild-type form of it. Although allele-specific strategies do not disrupt the endogenous wild-type genome, allele-independent approaches are more far more practical since they don't have to be customized for specific disease-causing mutations. Allele-independent strategies, however, require the expression of both nucleic acid molecules in the same vector and are therefore limited by packaging issues.

With over 270 genes associated with IRDs, developing mutation-specific or even gene-specific approaches becomes challenging. In this context, the role of non-targeted gene therapy is to provide alternative strategies, aimed at improving vision independently of the causative gene (86). Attempts have been made by delivering via AAV vectors molecules capable of prolonging photoreceptor cell survival, including neurotrophic factors, such as ciliary neurotrophic factor (CNTF), glial cell line-derived neurotrophic factor, basic fibroblast growth factor and rod-derived con viability factor, and anti-apoptotic agents, such as X-linked inhibitor of apoptosis (87–92). Other than CNF, whose efficacy was rather modest (93), none of these approaches have been tested in humans so far.

As an alternative mutation-independent strategy, optogenetics delivered as non-targeted gene therapy for advanced RP is also being tested. In this case, the assumption is that inducing expression of light sensitive opsins in bipolar or RGCs could activate the visual pathway even in the absence of viable photoreceptors (94–97). There are three ongoing clinical trials using optogenetics in RP patients, RST-001 (NCT02556736), GS030 (NCT03326336), and BS01 (NCT04278131), while one phase 1/2 clinical trial, vMCO-I, has been recently completed (NCT04919473).

Leber congenital amaurosis is mostly inherited in an AR fashion, though for some genes, like CRX, AD patterns have been reported (107). So far, more than 25 genes, overall accounting for at least 80% of all LCA cases, have been described, the most common of which are listed in Table 1 (106–126). Accordingly with the current literature, the most common LCA-causing genes are, in descending order, GUCY2D, CEP290, CRB1, RDH12, and RPE65 (1, 3, 4). LCA-associated genes encode proteins, whose functions can be divided into four main categories: phototransduction (e.g., GUCY2D), photoreceptor morphogenesis (e.g., CRB1 and CRX), retinoid cycle (e.g., RDH12 and RPE65), and ciliary transport processes (e.g., CEP290) (98).

The RPE65 gene product plays a critical role in the retinoid cycle, so that RPE65 mutations affect visual function before photoreceptor structure. Therefore, in contrast with many other IRDs in which visual dysfunction results from rods and cones death, RPE65-LCA patients retain viable cells for years before significant degeneration becomes evident. This structure-function dissociation makes RPE65-related retinal dystrophies a particularly compelling target for gene replacement strategies.

Proof of principle for retinal gene therapy came from the pioneering studies conducted in the early 2000's on a peculiar canine model (127, 128). These preclinical studies employed a subset of Briard dogs with a homozygous 4-bp deletion in the RPE65 gene resulting in a premature stop codon, thereby appearing to be an excellent spontaneous model for human RPE65-related LCA (129). These studies reported, after a single SRI of AAV-mediated RE65, an improvement in blue light stimulated dark-adapted ERGs and cone flicker, pupillometry, and VEP in the injected eyes and in qualitative behavioral assessments in the treated dogs, which were stable 3 years after the procedure (127, 128). Further evidence of the efficacy of this approach came from the naturally Rpe65-mutated rd12 murine model and from the genetically built Rpe65^−/− knockout mouse (130, 131). Following the success of animal studies, clinical trials were initiated in 2007 by groups from the UK and the US (132–134), culminating in the first FDA- and EMA-approved AAV-based retinal gene therapy drug, voretigene neparvovec-rzyl (Luxturna) (135). Follow-up studies revealed stable improvements in most patients, peaking at 6–12 months after injection (136–138), but observational trials aimed at evaluating the long-term effects of Luxturna are still ongoing (NCT03602820, NCT01208389).

The protein encoded by the CEP290 gene localizes to the photoreceptor connecting cilium and, besides microtubule-associated transport across the cilium, is required for outer segment (OS) regeneration and phototrandusction. The most common CEP290 mutation is the so called IVS26 c.2991+1655A>G mutation (p.Cys998X), an adenine to guanine point mutation located within intron 26 creating a novel splice donor site, which results in the inclusion of a pseudoexon in the mRNA and in the consequent creation of a premature codon stop. This mutation has been addressed by means of two innovative approaches.

The first strategy relies on a CRISPR/Cas9 system, called EDIT-101, consisting of an AAV5 vector used to deliver the Sthaphylococcus aureus Cas9 and CEP290-specific gRNAs with no identified off-targets. EDIT-101, or a non-human primate (NHP) surrogate vector, were shown to restore normal splicing in vitro (in photoreceptor-containing retinal explants) and in vivo (in mice and NHPs) with no serious adverse events (139).

The second strategy exploits the AON technology to remove the 128-bp pseudoexon included in the IVS26-mutated CEP290 mRNA transcript. Preclinical evidence of the efficacy of the AON designed to restore IVS26 splicing defects, called QR-110, came from in vitro studies on LCA10 fibroblasts (140).

In the wake of these results, IVI of this oligonucleotide was successfully attempted in NHP s (141), finally reaching the clinical setting with a phase I/II trial showing vision improvement at 3 months with no complications in LCA type10 patients treated with multiple doses of intravitreal QR-110 (142). A phase II/III multiple-dose clinical trial is still ongoing and is aimed at evaluating efficacy, safety, tolerability, and systemic exposure of QR-110 administered via IVI in patients with LCA type 10 due to CEP290 c.2991+1655A>G mutation after 24 months of treatment (NCT03913143).

Although so far only two LCA-associated genes have made it to the human trial stage, for many other disease-causing genes, including GUCY2D, CRB1, and RDH12, preclinical studies are underway, showing promising results (143–148).

To date, 16 genes associated to Usher syndrome have been identified, two of which are good candidates for gene therapy and deserve a more detailed description.

The first of such genes is MYO7A. MYO7A encodes myosin VIIA, involved in in transport of melanosomes and phagosomes along actin filaments in the RPE and of opsin and other phototransduction proteins in photoreceptors (195). Mutations in MYO7A are associated with Usher Syndrome type 1B (USH1B).

The second relevant gene is USH2A, which encodes usherin. Mutations in USH2A, besides being the commonest association with type 2 Usher syndrome (80%), are the most frequent cause of AR non-syndromic RP (10–15%) (196–199). Clear genotype-phenotype correlations for USH2A mutations are not easy to establish. Generally, however, nonsense mutations, frameshifts mutations, or canonical splice site mutations in USH2A, either biallelic or combined with one missense allele, are associated with Usher syndrome type II, whereas the association of two missense mutations tends to result in non-syndromic RP (200). Of note, a peculiar feature of both non-syndromic and syndromic USH2A-related retinopathy is the fact that fundoscopy generally shows mild or no pigment deposits.

The MYO7A gene has 49 exons and spans approximately 87 kb of genomic sequence on chromosome 11q13.5, therefore significantly exceeding the cargo capacity of AAV vectors. For this reason, attempts have been made to deliver MYO7A in the shaker1 mouse model of Usher syndrome type 1B by means of UshStat, an EIAV lentiviral vector carrying the wild-type gene (EIAV-CMV-MYO7A) (201). Later on, Sanofi sponsored two phase 1/2 clinical trials, one of which is currently ongoing (NCT02065011), while the other has been stopped not for safety reasons (NCT01505062).

As far as USH2A is concerned, though being associated with recessive IRDs, a number of factors, including poor understanding of the physiological function of usherin and the USH2A gene size (15 kb), stand in the way of gene replacement strategies. Therefore, in order to develop an alternative approach, great interest was focused on a subset of mutations on exon 13 resulting in aberrant pre-mRNA splicing, which leads to the inclusion of a pseudoexon in the mature USH2A transcript. Since exon 13 consists of a multiplier of three nucleotides, skipping this exon does not disturb the open reading frame and likely results in the synthesis of a shorter protein with predicted residual function. This is the rationale behind the employment of AONs in an AR IRD. Encouraging results came from preclinical studies (73, 202) and ProQR Therapeutics sponsored the STELLAR phase 1/2 clinical trial (NCT03780257), whose purpose is to evaluate the safety and tolerability of a single intravitreal administration of AONs (QR-421a) in subjects with RP due to mutations in exon 13 of the USH2A gene. Enrolled patients receive one single IVT injection of QR-421a or sham-procedure in one eye (subject's worse eye) and are then followed up for 24 months. In March 2021, ProQR announced positive results from clinical trial of QR-421a and planned to start two phase 2/3 trials (SIRIUS and CELESTE).

Choroideremia is caused by mutations in the CHM gene, which encodes component A of Rab geranylgeranyl-transferase, referred to as Rab escort protein 1 (REP1), a key mediator of post-translational lipidation (prenylation) and subcellular localization of a family of intracellular protein trafficking regulators, known as the Rab GTPases (211). The CHM gene is ubiquitously expressed, but in most tissues, including adrenal gland, brain, and thyroid, the homolog REP2 protein partially counterbalances REP1 deficiency. The reasons why REP2 does not prevent disease manifestation in the eye are yet to be elucidated. In rare instances, CHM may be part of a contiguous gene syndrome involving Xq21. Indeed, males with large interstitial deletions of an additional X-chromosome portion, other than Xq21, may develop CHM together with birth defects (cleft lip and palate and agenesis of the corpus callosum) and severe cognitive deficits (212). Moreover, previous reports described cases of males with a small deletion of Xq21 presenting with CHM, mixed sensorineural and conductive hearing deficits (in case of deletion of POU3F4), and varying degrees of cognitive deficits (in case of deletion of RSK4) (213). In addition, a previous report described the case of a female patient affected by CHM, sensorineural deafness, and primary ovarian failure secondary to a balanced X-4 translocation (214). A large study, involving more than 70 patients affected by this disease, reported a 94% rate of identification of a CHM mutation (204). Mutations of uncertain significance included non-contiguous duplications, insertion, deletion, point mutations, and aberrant splicing (204). It is worth of notice the absence of disease-causing missense mutations, in contrast to the majority of human genetic diseases, which are mainly determined by such mutations (204). To date, no defined genotype–phenotype correlation has been identified for CHM.

Choroideremia is a promising candidate for gene therapy since the 1.9 kb CHM cDNA is small enough to fit the size capacity of AAV vectors. Preclinical proof-of-concept studies on the feasibility of CHM gene replacement have been conducted both in vitro, by inducing the CHM gene in pluripotent stem cells (iPSCs) from patients with CHM, and in vivo, by delivering the AAV2-CHM virus in normal sighted mice and zebrafish, with no evidence of toxicity (215, 216), paving the way to clinical trials on CHM patients. The results of the first-in-human clinical trial date back to 2014, when MacLaren et al. reported phase I safety and efficacy data on six patients treated with low-dose subretinal AAV2-REP administered subfoveally, demonstrating an improvement in BCVA and retinal sensitivity for up to 3.5 years after treatment (217). These findings were confirmed by the 24-month data coming from the phase II trial (218). Less encouraging results came from another phase I clinical trial using the same vector at higher doses in six patients, since one of the six untreated eyes exhibited an improvement of >15 ETDRS letters, prompting the authors to conclude that VA should not be used as a primary outcome measure for future CHM gene therapy trials (219). So far, several clinical trials employing SRI of AAV2.REP1 have been conducted (NCT02671539, NCT02077361, NCT02553135, NCT02341807, NCT03496012). Combining together their results, overall 40 patients have been treated with a median gain of 1.5 ±7.2 SD in ETDRS letters, highly variable between the different trials (220). Although some issues still need to be addresses, such as the identification of proper endpoints and the development of safety enhancements to facilitate subretinal gene delivery, gene therapy for CHM has reached phase III clinical trials, providing real promise for patients. It is worth of notice the launch of the first phase I CHM clinical trial aimed at evaluating safety, tolerability, and preliminary efficacy of a single IVI of a rAAV gene therapy, 4D-110, in male patients with genetically confirmed CHM (NCT04483440).

RS1 encodes retinoschisin, a secretory protein exclusively expressed in retinal photoreceptors and bipolar cells, that can however be detected in all neuroretinal layers (230–232). Retinoschisin is found in a homo-oligomeric forms and, more specifically, it is an octamer made up of eight identical discoidin domains joined by intramolecular disulphide bonds (233). Mutations in the RS1 sequence disrupt subunit assembly, thus interfering with retinoschin's role in retinal cell adhesion and organization of retinal architecture (233, 234).

Preclinical studies have shown that IVI administration of AAV8-scRS/IBPhRS vector, as well as SRI of the AAV5-mOPs-RS1 resulted in significant morpho-functional improvement in the retinoschisin knockout (Rs1-KO) mouse, with evidence of good tolerability in rabbits (235–238). Building upon these encouraging results, two phase 1/2 trials were initiated with different constructs, administered via IVI, since XLRS is an inner retinal pathology. For the first of such trials (NCT02317887), sponsored by the National Eye Institute (NEI), employing AAV8-scRS/IRBPhRS in adults (≥18 years old), promising initial findings have already been reported (239). The second trial (NCT02416622) uses rAAV2tYF-CHhRS1, intravitreally injected in adults (≥18 years old) in the first dose-escalation phase, with subsequent enrolment of individuals ≥6 years of age, after the maximum tolerated dose is identified.

Stargardt disease can be distinguished in three different forms: (I) STGD1, the most common form, displays an AR homozygous or compound heterozygous transmission and is caused by mutations in the ABCA4 gene; (II) STGD3, which is determined by AD mutations in the ELOVL4 gene; (III) STGD4, a rare AD form associated with mutations in the PROM1 gene.

ABCA4, whose mutations are responsible for STGD1, is a 150 kb gene encoding an ATP-binding cassette transporter localized along the rims of photoreceptor OSs (246). This ATP-dependent flippase importer transports phosphatidylethanolamine (PE) and the all-trans-retinal (atRAL)/PE Schiff base (N-Ret-PE) into the cytosol, where atRAL is converted to all-trans-retinol (atROL) by retinol dehydrogenase RDH8 and RDH12. The absence of ABCA4 results in the accumulation of photo-toxic bisretinoids (A2E) with lipofuscin buildup in the RPE.

STGD3-causing ELOVL4 gene contains six exons and plays a fundamental role in the synthesis of very long-chain polyunsaturated fatty acids (247).

Finally, the PROM1 gene contains 23 exons distributed within a genomic sequence of more than 50 kb and encodes a protein involved in the organization of the plasma membrane and in the biogenesis of photoreceptor disks (248).

Genotype-phenotype correlations are challenging due to the heterogeneous disease manifestations. The most common genotypic classification includes three groups: genotype A (carriers of two or more deleterious variants); genotype B (one deleterious variant and >1 missense or in-frame insertion/deletion variants); genotype C (two or more missense or in-frame insertion or null variants) (249). Previous investigations highlighted how deleterious variants tend to be associated with more aggressive forms of the disease, while missense mutations yield to milder phenotypes (249–252). On the other hand, null alleles result in more severe STGD forms with an earlier onset (253). Other causes of severe phenotypes include truncating and severely misfolding mutations, deletions, stop codons, and insertions (254, 255). Furthermore, hypomorphic and deep intronic variants influencing the splicing process, have been also described (72, 256–258).

The main obstacles to the development of gene therapy approaches for STGD regard the dimension of the ABCA4 gene, whose coding sequence exceeds the cargo capacity of AAV, and the extreme complexity of deleterious variants, as previously described.

In the wake of the success of animal studies using lentiviral gene therapy to deliver the corrected ABCA4 gene (259), starting from 2011, Sanofi sponsored a phase I/II clinical trial (NCT01367444) to test the efficacy of a SRI of SAR422459, a recombinant lentiviral vector (EIAV) transporting a modified form of the ABCA4 gene. Notwithstanding the encouraging preliminary findings, in 2020, the sponsor decided to stop development of the product for non-safety reasons.

In 2019, the Applied Genetic Technologies Corporation announced the development of a hybrid AAV dual vector and published preclinical data supporting the potential of this technology in STGD (260).

More recently, promising results have been reported in animal models with a non-viral technique relying on subretinal delivery of self-assembled NPs (261, 262). Compared to viral vectors, non-viral delivery systems have unlimited payload, low immunogenicity, and minimal side effects, features that may allow to circumvent the obstacles which are currently standing in the way of STGD gene therapy (263).

Mutations in six genes are responsible for over 90% of all ACHM cases, five of which are involved in the phototransduction process.

CNGA3 and CNGB3, encoding for the a- and b-subunit of the cyclic nucleotide-gated (CNG) cation channel 3 found in cones OSs, together account for 70–80% of cases of ACHM worldwide (272).

GNAT2 encodes the catalytic a-subunit of the G-protein transducin and is responsible for an infrequent form (<2%) of ACHM (273).

Equally rare subtypes are those caused by mutations in the PDE6C and PDE6H gene, encoding for the catalytic a- and inhibitory g-subunits of the photoreceptor-specific phosphodiesterase (274, 275).

Most recently, ATF6 has been identified as a sixth ACHM-associated gene. ATF6 encodes a transmembrane TFs ubiquitously expressed and involved in endoplasmic reticulum homeostasis (276). The frequent finding of foveal hypoplasia in ATF6-ACHM lead to the suggestion that this gene may be crucial for foveal development (276).

Promising preclinical results have been reported for ACHM caused by mutations in the CNGA3, CNB3, and GNAT2 genes. Taken together, these studies, which have been conducted on knock-out mouse models and naturally occurring mouse, sheep, and canine models, suggest that gene replacement approaches are effective and durable in ACHM, especially if administered early in life (277–284).

Other than for these encouraging data, ACHM seems particularly suited for gene augmentation for a number of reasons, including the presence of viable cones in all patients, as demonstrated by means of AOSLO, and its stationary or slowly progressive nature, which provides a wide window of opportunity.

The first phase I/II clinical trial commenced in November 2015 (NCT02610582) in Germany to assess the safety and efficacy of SRI of rAAV.hCNGA3 in patients with CNGA3-ACHM, using a dose-escalation protocol. Short after, another phase I/II clinical trial for CNGA-related ACHM (NCT02935517) was initiated in the US and in Israel and is still ongoing.

As far as CNGB3-ACHM is concerned, a phase I/II dose-escalation trial has been conducted in the UK (NCT03001310) to test the efficacy and safety of subretinal delivery of AAV2/8-hCARp.hCNGB3 and another similar multicentric phase I/II trial is still ongoing in the US and in Israel (NCT02599922).