Adeno-Associated Virus (AAV) - Based Gene Therapies for Retinal Diseases: Where are We?

Introduction

AAVs are non-pathogenic viruses belonging to the Parvoviridae family under the genus Dependoparvovirus. They can infect humans but are non-pathogenic and are known to have low immunogenicity. They are naturally replication-deficient and hence termed as ‘helper-dependent virus’ as they depend on the presence of adenovirus, herpesvirus or papilloma virus as helpers to replicate.^1–3 In the absence of a ‘helper virus’ AAV remains latent by integrating into the host genome or remaining as extrachromosomal episomes. AAVs are small, non-enveloped viruses with a capsid size of 25nm and package single-stranded DNA of approximately 4700 base pairs.⁴ There are 13 naturally occurring AAV serotypes (AAV1 – AAV13) and 100s of AAV variants have been identified or engineered. These AAV serotypes possess distinct tropisms that are tissue specific and species specific. The tissue tropism/specificity of AAVs is determined by the proteins on the surface of capsids. The capsid proteins bind specific cell-surface receptors which in turn recruit co-receptors leading to internalization by endocytosis into the cell.⁵ Following internalization AAVs escape the endosomes and gain entry into the nucleus.⁶ AAV relies on the host cell mechanisms for the second strand synthesis.⁷

For experimental and clinical applications recombinant AAVs (rAAV) are created by replacing the viral genome by the transgene of interest while retaining the ITRs. The ITRs flank the transgene and ensure its encapsidation within the AAV. The other structural (capsid proteins) and non-structural (replication and helper) genes are supplied separately.^4,8 rAAVs rarely integrate into the host cell genome, thereby reducing the risk of insertional mutagenesis following gene therapy.

Gene therapy refers to the introduction of genetic material into tissues to rectify the effects of disease-causing mutations. Inherited diseases caused by gene mutations resulting in deficiency or malfunction of proteins involved in cellular functions are ideal targets for gene therapy. Such diseases can be treated by replacement of the entire defective gene, silencing of the mutation or correction/editing of the mutation. The therapeutic gene is packaged into AAVs and delivered to target tissue by local injections.⁹ AAV delivery to the retina is normally done by two routes: subretinal and intravitreal. Subretinal injection (SRI) involves injection between the Retinal Pigment Epithelium (RPE) and the photoreceptors resulting in efficient AAV delivery to the outer retina. Intravitreal injection (IVT) involves AAV delivery into the vitreous space thereby causing transduction of the Ganglion Cell Layer (GCL) and some other cell types of the inner retina.¹⁰ The retina is an ideal target for gene therapy due to two main reasons: First, many gene mutations responsible for Inherited Retinal Diseases (IRDs) have been identified,¹¹ and second, the retina is enclosed within the eye in a relatively immune-privileged space. The retina is shielded by a blood-retinal barrier and also benefits from the unique intraocular microenvironment.¹² This prevents the occurrence of a strong systemic effect to the AAV-mediated ocular gene therapy.

Clinical Trials Using AAVs for Retinal Diseases

Over the last five decades, substantial efforts into understanding the basic biology and engineering of AAV have resulted in clinical success of AAVs. The first FDA approved retinal gene therapy called Luxturna developed by Spark therapeutics is for a retinal disease called Leber Congenital Amaurosis type 2 (LCA2). LCA2 is an autosomal recessive disease caused by mutations in the RPE65 gene present in the RPE layer of the retina.¹³ The therapy aims at replacing the RPE65 gene by delivering a healthy cDNA copy using AAV2 injected directly adjacent to the RPE layer by SR injections. The proof-of-concept for vision restoration was provided in a canine model.¹⁴ This led to clinical trials involving LCA2 patients who received SR injections of RPE65 packaged in AAV2 via SR delivery. The trials reported improvement in visual acuity, pupillary reflex, light sensitivity as well as the patients’ ability to navigate in low light.^15–17 Furthermore, it was also demonstrated that it was safe and effective to inject the contralateral eye.¹⁸ However, long-term follow-up in 2 out of 3 clinical trials reported a decline in the visual gains in adult patients, indicating that the treatment was unable to halt the degeneration.^15,19 This highlights two important aspects: the level of AAV transduction required for therapeutic benefits in humans may be higher and the timing of therapeutic intervention is crucial (earlier interventions before the onset of degeneration will be more beneficial).

The earliest clinical trials used AAV2 and this serotype accounts for 53% of all trial for retinal diseases. Most of the current clinical trials (about 81%) continue to use naturally occurring serotypes. Only 19% of the current clinical trials use engineered AAVs including, AAV2tYF (generated by rational design), AAV2-7m8 and 4D-R100 (generated by directed evolution) (Figure 1A and B). Interestingly, the newly developed variant AAV-4D-R100 is being used to deliver genes to the photoreceptors and the RPE by IVT injections. It is worth noting that all the completed trials used naturally occurring serotypes, whereas among the ongoing trials 34% (10 out of 29) use engineered serotypes (Table 1).

Table 1 Naturally Occurring and Engineered AAVs Used in Clinical Trials

Figure 1 Distribution of (A) Naturally occurring AAVs (blue) and engineered AAVs (Orange) used in 53 clinical trials for retinal diseases; (B) distribution of AAV serotypes being used in clinical trials with the naturally occurring ones (AAV2, AAV4, AAV5 and AAV8) in shades of blue and the engineered variants (AAV2tYF, 4D-R100 and AAV2-7m8) in shades of Orange; (C) distribution of the therapeutic strategy used in clinical trials; (D) distribution of clinical trials using the subretinal delivery route for outer retina or other regions of the retina; (E) distribution of clinical trials by injection route including subretinal (Orange) and intravitreal (blue) injections; (F) distribution of trials using intravitreal delivery routes and targeting the inner, outer or other regions of the retina.

The early clinical trial success has paved the way for many other clinical trials (Table 1 and Table 2). The majority of current gene therapies focus on gene replacement for diseases caused by recessive mutations (Figure 1C). These are straightforward targets since the replacement provides the protein and restores the missing function in target tissue. However, for dominant mutations first the malfunctioning gene has to be suppressed and then the correct gene needs to be supplemented. Another widely tested therapy is the anti-VEGF therapy for wet-AMD (Age-related Macular Degeneration) (Figure 1C). AMD has a global prevalence of close to 200 million and this number is bound to increase with an increase in elderly population overtime, making the development of therapy urgent.²⁰ AMD has two forms: dry and wet AMD. It is estimated that about 15% of the patients develop what is known as the wet form of the disease, wherein neovascularization occurs and the poorly formed blood vessels result in blood leakage. Vascular Endothelial Growth Factor (VEGF) has been recognized as an important factor involved in this process of neovascularization and the present clinical practice involves the repeated injection of anti-VEGF antibodies or soluble receptor inhibitors.²¹ However, patients need frequent (often monthly) anti-VEGF injections, which are expensive.²² Hence, the anti-VEGF gene therapies aim to have stable expression of VEGF inhibitors, offering an alternative long-term solution.

Table 2 Exclusion Based on Pre-Existing Immunity and Immune Responses in Clinical Trials

All deliveries to the retina in clinical trials are by SRI or IVT routes (Figure 1D–F). Therapies that target the cell types of the inner retina can be supplied by IVT injections. Further, anti-VEGF therapy can also be provided intravitreally as no cell-type is targeted, rather it acts to curb a symptom (angiogenesis). Contrary to these, most of the replacement-based therapies are provided by SRI as the disease-causing mutations mostly occur in photoreceptors or RPE.

As mentioned earlier, one important caveat for the success of gene therapy is early intervention, since the cells where the gene correction or supplementation is required have to be present. In late stages of retinal degeneration, when the photoreceptors are mostly lost, correction of gene mutations will not provide any therapeutic effect. To circumvent this issue mutation-independent optogenetic therapy is now being developed and tested. Following the loss of photoreceptors, optogenetic therapy aims at converting the surviving retinal neurons into photosensitive cells. This is achieved by the expression of light-sensitive microbial opsins in the remaining cell types of the retina.²³ A recent study delivered the light-sensing Channelrhodopsin protein - ChrimsonR packaged in an engineered vector (AAV2-7m8) by IVT injection to mainly target the foveal retinal ganglion cells of a blind patient. In the study, the patient was reported as being able to perceive, locate, count and touch different objects using the treated eye alone. Furthermore, during visual perception, object-related activity was observed above the visual cortex.²⁴ This success has encouraged further trials involving microbial opsins (Table 1). Optogenetic therapy can benefit patients with end-stage degeneration who presently have no therapeutic options. In the near future, it might be extended to target more upstream retinal neurons with more sensitive G protein coupled opsins for better visual outcomes.²⁵

Challenges with AAVs

The small size of AAV is an advantage because it allows efficient diffusion through many tissue barriers. However, smaller size also limits the genome size. Hence, transgenes including their promoters and regulatory sequences have to be below 4.7kb in size. This is a limiting factor in gene therapies that aim at replacement of defective genes, as some of the disease-causing genes are larger than 4.7kb in size. Moreover, gene replacement is not a viable option for IRDs caused by autosomal dominant mutations displaying dominant negative or toxic gain-of-function mechanisms.²⁶ For these reasons, the field of gene therapy is moving towards gene editing using the power of editing tools such as CRISPR-Cas9, Base editors (BEs) and Prime editors for correcting gene mutations.^27–29 Some Cas9 proteins such as saCas9 fit into AAV along with the target guide RNA, whereas the most commonly used one - spCas9 does not. But, base and prime editors are too large to fit into a single AAV.³⁰ Although gene-editing techniques are precise and efficient, they are still known to have off-target effects.^31,32 Newer technologies are being developed to circumvent this issue, but none of them guarantee complete avoidance of off-targets.³³ Hence, it is important to limit unwanted editing in cell-types that do not need correction. This can be done by using cell-specific promoters. The challenge of using these editing tools gets more complex when they need to be expressed in a cell-specific manner, which is achieved by using cell-specific promoters.³⁴ But, in most cases, these promoters occupy significant space out of the 4.7kb available in AAVs alongside polyadenylation sequences and other potential cis regulatory elements that are necessary for optimal transgene expression.

Although the early success and clinical usage of AAV-based gene therapy was achieved in the retina, there are still certain challenges that remain. As mentioned earlier, with SRIs transduction of the photoreceptors can be achieved. However, SRIs are limited to a small region of the retina and with this route of injection a pan-retinal transduction is not achieved. Additionally, this is an invasive technique and is likely to damage an already fragile and degenerating retina.³⁵ The IVT route of injection results in a pan-retinal transduction which is limited to the ganglion cells or some inner retinal cells. Certain engineered variants have been developed which, when delivered by the IVT route, can cross all retinal barriers and transduce the photoreceptors of the mouse retina.^36–38 A variant (AAV2-7m8) developed in mice did not translate well for non-human primate applications.³⁷ Species-specific directed evolution was performed to identify variants for NHPs but these had limited success due to the more prominent barriers to AAV transduction in NHPs compared to mice.³⁸ One of the major barriers, the Inner Limiting Membrane (ILM) is thicker in the primate and human retina compared to rodents.³⁹ Although IVT injections can be less invasive, there is a higher probability of immune response due to the presence of immune cells in the vitreous fluid.⁴⁰

AAVs are non-pathogenic and do not cause any disease in humans. But they are ubiquitously present and about 50–90% of the human population has already been exposed to AAVs.^41,42 Such exposure initiates an immune response and results in the development of anti-AAV antibodies. The anti-AAV antibodies are of two types – Total or Binding antibodies (BAbs) and a subset of BAbs called Neutralizing antibodies (NAbs). BAbs are all the antibodies that are produced against a specific AAV serotype and upon repeat exposure to the same serotype BAbs can trigger a stronger immune response that can potentially result in inflammation. A subset of BAbs called NAbs are capable of neutralizing AAV thereby reducing the efficacy of gene therapy.⁴³ Hence, prior exposure to AAVs results in the presence of pre-existing antibodies and when a patient with high pre-existing antibodies receives AAV-based gene therapy, it compromises both the efficacy and the safety of the therapy. These pre-existing BAbs and NAbs are serotype, species, age and population specific.⁴⁴ The highest seroprevalence in the human population was for anti-AAV2 antibodies, while anti-AAV8 was the lowest.^41,42 Studies in non-human primates reported the highest seroprevalence was against AAV8 and AAV9, while AAV5 was the lowest.^45,46 Irrespective of the pre-existing immunity, AAVs can trigger an immune response in the eye. Even though the eye is considered immune privileged, reports from clinical trials and experiments have provided plenty of evidence to the contrary (Table 2).⁴⁷

Strategies to Overcome the AAV Size Restriction

To overcome the size limitation of AAVs, new strategies including the use of simultaneous dual or triple AAVs are being developed, extending total AAV cargo capacity to 9kb and 14kb, respectively.^48,49 This approach is based on concatemerization and/or homologous recombination of AAV vectors that each contains a distinct part of the transgene, flanked by inverted terminal repeat sequences (ITRs) and/or transgene homologies. With these strategies, the vectors can reconstitute the full transgene coding sequence.^49,50 Thus, it provides an interesting approach to treat autosomal recessive diseases with genes longer than 5kb. Unfortunately, some of the more than 280 genes involved in IRDs would require gene replacement that exceed the packaging capacity of AAVs (https://web.sph.uth.edu/RetNet/). Recently, in vivo proof-of-concept of dual AAVs therapies have been reported by successfully delivering large therapeutically relevant genes involved in IRDs pathologies such as MYO7A (6.7kb) or ABCA4 (6.8kb).^51–53 Moreover, an evaluation of dual-AAV strategies studied the in vivo efficiency of this approach to deliver transgenes to retinal cells. The authors compared the efficiency of the three main strategies of dual-AAV reconstitution: trans-splicing (ITRs concatemerization), overlapping (homologous recombination) and the hybrid method (concatemerization or homologous recombination). They showed that the hybrid method was the most relevant when compared to the other two in the mouse retina, confirming results from previous studies showing the hybrid approach to be efficient in vivo. The authors also demonstrated about 10% reconstitution efficiency based on protein expression compared to unique AAV reporter control.⁵⁴ Even if the reconstitution efficiency is low, phenotypic improvement in disease models have been reported in previous studies.^52,54,55

Furthermore, some genes involved in genetic disease are still larger than dual-AAVs capacity. In a recent study, Maddalena et al performed triple-AAV-based gene delivery of a reporter transgene (Fusion of EGFP and DsRed that was split into 3 vectors) and therapeutic transgenes (CDH23; ALMS1). Interestingly, when considering the reporter construct, a low percentage of full fusion protein expression was observed in mouse retina with ~2% expression of the single control AAV encoding the entire reporter, whereas a higher efficiency was achieved in pig retina at ~39%, suggesting structural and molecular interspecies specificities impacting viral transduction and intracellular reconstitution.⁴⁸

Gene editing has been proposed as an alternative to gene replacement as it allows to correct the genetic defect in situ in the chromosome. This strategy has rapidly moved towards clinical application after pre-clinical proof of concept studies. In 2019, Maeder et al delivered a single AAV-SaCas9 construct with two different guide RNAs to excise an aberrant splice donor in the 7kb gene CEP290 in mouse and non-human primate, removing the ISV26 disease-causing mutation for LCA type 10, thus avoiding technical genetic supplementation with multiple vectors. With this strategy, they could avoid the technical challenge of genetic supplementation with multiple vectors.⁵⁶ However, this elegant therapeutic strategy depends on mutational context. To overcome such obstacles, some groups are developing multiple AAV strategies to deliver new gene editing tools that usually exceed AAV cargo capacity, such as prime editors or base editors.⁵⁷ BEs are DNA editing tools that use a dead Cas9 (dCas9; no/residual catalytic activity) fused with cytidine (CBEs) or adenine (ABEs) deaminases allowing single base conversion of C:G>T:A or A:T> G:C.^27,58 This strategy is particularly relevant in both monogenic autosomal recessive and autosomal dominant diseases contexts for repairing pathogenic variants. BEs were used recently in several cell types/organs, including the retina, and were delivered by dual-AAVs approaches.^27,59 The delivery methods are usually based on trans-splicing inteins, where each part of the transgene containing split protein fused to N-terminal and C-terminal intein sequences. Thus, reconstitution occurs at the protein level.^60,61 Two recent studies showed interesting results in rd12 mice, an LCA model involving the RPE65 gene with nonsense mutation c.130C>T, p.(R44X).^28,62 Both teams performed dual-AAV delivery of split-ABEs to correct point mutation and reached 3% and 6% A>G editing at the DNA level in RPE cells, leading to retinal function improvements.

Dual or triple AAV approaches offer new perspectives that would not be possible with other available viral vectors, especially in the retina. Despite their greater packaging capacity, other vectors such as adenovirus and lentivirus have variable immunogenicity and transduction efficiencies/specificities.^63,64 Nevertheless, these multiple-AAV approaches imply co-transduction of the same cell by 2 or 3 vectors, decreasing the probability of full reconstitution. One of the problems with dual/triple AAV strategies is the risk of producing truncated proteins, which have been reported previously.^48,51,55 As it raises safety concerns, these issues must be addressed for future clinical applications. Finally, to circumvent the use of multiple vectors, a new trend is the development of size-minimized tools, such as ABE involving a smaller Cas9 protein that thus can fit in a single AAV.^65,66

Depending on the targeted cell type, dominant or recessive form of the disease, therapeutic aims, and desired cargo, using AAVs is not always the most straightforward choice. Further, other drawbacks of AAVs have to be considered despite the lack of data such as the consequences of delivery in post-mitotic cells in which the viral genome can persist for years leading to long-term endonuclease expression;^17,67–69 and immune response against the transgene;⁴⁴ and potential genotoxicity.⁷⁰ Due to such considerations, over the last few years, non-viral vectors were increasingly studied and their therapeutic potential has been demonstrated.^71–73 These delivery approaches offer some advantages such as relatively low immunogenicity, no integration risks and delivery of a large range of components like DNA, RNAs or proteins.⁷⁴ Non-viral vectors are a heterogenous family including various components such as lipid nanoparticles (LNPs), polymer-based vectors, or inorganic materials.^75,76 Nonetheless, when it comes to the retina, those vectors, which are bigger (~100–200nm) than AAVs (~25nm), may face numerous physical and physiological barriers such as limiting membranes, hyper structure of photoreceptors outer segments, RPE phagocytosis activity, tight cell junctions and low extracellular space volume.^77–81 Moreover, other variables have to be considered such as the stability of the cargos that could be degraded in the extracellular space.⁸¹ Despite these difficulties, some studies reported retinal delivery of DNA plasmid or minicircle in mouse models of disease-associated genes like RPE65.^82,83 These vectors also offer an opportunity to deliver gene-modifying tools in mRNA or proteins form, limiting their intracellular presence and potentially reducing the risk of off-targets effects.^70,84,85 So far, CRISPR/Cas9 transfer into the retina in its ribonucleoprotein (RNP) conformation has been used to target the VEGFA gene, whose overexpression is involved in AMD. Recent studies delivered RNP complexes using LNPs and achieved significant editing in RPE cells.^86,87 Finally, despite interesting characteristics, their efficacy is lower compared to AAV viral vectors. For example, in two different studies, the same group was able to introduce ABEs to the retina as RNPs by LNPs, as well as via AAV, showing 1.8% and 6% efficiency, respectively, leading to phenotype improvement only with AAV-ABEs suggesting a better clinical relevance.^62,88 These advances are encouraging and their development in parallel with that of AAV approaches is crucial to develop safer therapeutic methods.

Strategies to Engineer AAVs with Enhanced Properties for Retinal Gene Therapy

The search for AAV variants with improved tropism and infection profiles has walked hand in hand with the development of AAV as vectors for gene therapy. In the retina, the natural serotypes of AAV have broad tropism for all cell types, and infection is dependent on route of delivery. SRIs permit infection of mainly RPE and photoreceptors, while IVTs lead to transduction in retinal ganglion cells and amacrine cells.^89–91 SRI, while currently the method of choice in clinical trials aiming to reach the outer retina, results in retinal detachment due to the formation of the surgical bleb, an undesirable prospect for patients with retinas already undergoing degeneration. Formation of the bleb under the fovea has also resulted in retinal thinning.¹⁸ Moreover, AAV has limited reach outside of the bleb due to restricted lateral spread.^92,93 While IVTs permit AAV to reach a larger area of the tissue, they nevertheless require vitrectomies and result in poor transduction of the inner and outer retina due to physical barriers such as the ILM.⁸⁹ The ILM is coated in a thick layer of glycans, of which heparan-sulfate (HS) is essential for AAV2 binding. While binding to HS reduces AAV2’s dilution in the vitreous, it also leads to reduced penetration of the retinal tissue. For almost two decades now there has been a search for AAV variants capable of overcoming these obstacles and show increased transduction efficiencies to assist the delivery of a lower AAV dose. Different methods have been developed for identifying such variants, and many are now the vectors of choice in clinical trials (Table 1).

With crystallography of natural occurring serotypes and the understanding of the role of different amino acids on the capsid surface, researchers have explored a rational design approach. By swapping amino acids important for HS and receptor binding (R484, R487, K532, R585, and R588) or ubiquitination pathways (substitution of threonine, tyrosine or serine residues), it is possible to modify tropism or to increase transduction efficiencies in different tissues, including the retina.^36,94–96 As such, the variant AAV2tYF (three tyrosine to phenylalanine substitutions, Y444F, Y500F, Y730F) has shown improved transduction profiles in canine animal models by evading degradation through the proteasome complex.^97,98 AAV2tYF is currently in use in four clinical trials (Table 1), both through SR and IVT injections. However, rational design has its limitations and extensive pan retinal expression still has not been shown with this method.

Along similar lines to rational design, targeting AAVs to known cell receptors has recently been explored with the use of nanobodies.⁹⁹ By harnessing the desirable characteristics of camelid antibodies, a fragment of the heavy-chain variable domain can be attached to the AAV capsid, conferring the desired cell tropism by targeting known receptors or other membrane proteins. Incredibly small and with high specificity, these fragments, also known as nanobodies, have been successfully incorporated onto the capsids of different AAV serotypes (AAV1, 2, 8 and 9) and they have also been targeted to different cell proteins (a single-pass transmembrane protein, a multi-pass ion channel, and a glycosylphosphatidylinositol-anchored ectoenzyme) expressed in HEK cell membranes. While not yet tested in vivo, these AAVs targeted via nanobodies greatly enhanced their transduction efficiencies and present a potential future for solving issues of transduction in specific cell types from different tissues, including the retina.

With the development of technologies capable of reconstructing ancestral proteins, libraries of hypothetical ancestral AAV capsids can also be manufactured and screened. By looking at the node Anc80, the theoretical ancestor of AAV1, 2, 3, 6, 7, 8 and 9, a library of potential ancestors was developed in silico and tested in vitro and in vivo.¹⁰⁰ The variant Anc80L65 outperformed all others in its photoreceptor infectivity profile, and, interestingly, showed low development of an immune response. So far, this vector has been tested in mouse and NHP retinas, and confirms the promising possibilities of capsid development in silico.¹⁰¹

Directed evolution has become an important, high-throughput method for the development of new AAV variants. Mutations to the cap gene of naturally occurring serotypes are introduced, and new variants are identified after several rounds of selection. There are several methods to generate highly diverse libraries, including the insertion of random peptide sequences into a defined location on the capsid, error-prone PCR and site directed mutagenesis, gene shuffling, and insertion of peptide sequences in random locations.^102–105 Selective pressure is applied, such as reaching the outer retina through IVT route, and variants that fulfill the desired tasks are identified. The first AAV variant capable of overcoming the ILM barrier was found through directed evolution and is called AAV2-7m8, a variant that has a seven-amino acid insert at R587.³⁷ This mutation confers the ability to partially evade HS binding and the “sink” effect of these proteoglycans at the ILM. Therefore, AAV2-7m8 is capable of passing through the physical barrier of the ILM and subsequently infecting all retinal layers through an IVT route achieving pan retinal expression. This is not only true in animal models such as mice, which have thinner ILMs, but also in sheep and NHPs,^37,106 as well as in patients (Table 1). However, in animal models with more complex and thicker ILMs, AAV2-7m8 does not yield a uniform pattern of expression. In the search for variants with more efficient profiles in human patients, other AAV variants have been identified in the last few years with improved infectivity or transduction characteristics after IVT injections in NHPs.^38,107,108 Of these, 4D-R100 is currently being used in two clinical trials (Table 1). Directed evolution has also been employed to develop AAV variants that are capable of evading immune responses (described in the next section).

With the identification of several new AAV variants, accurate and quantitative methods to compare their efficiencies and profiles are required. Traditionally, such comparisons required large numbers of animals and variations between injections and individual animals can lead to unreliable results. This is undesirable, especially when working with large animal models. With this in mind, methods such as scAAVengr were developed, whereby barcoded libraries of known AAV variants allow for the comparison of different naturally occurring or engineered capsids side by side.¹⁰⁷ By using single-cell RNA-seq, it is possible to compare expression patterns of different variants in parallel within not only the same retina, but also within individual cells. This method is especially valuable for comparing different variants in NHP models in vivo or in human retinal explants ex vivo, indicating AAV reliability for future clinical trials.¹⁰⁹

As a high-throughput method that uses next-generation sequencing, directed evolution of AAV capsids now generates large data sets that can be further explored through artificial intelligence (AI) and machine learning. AI-based algorithms can aid in the development of new diverse libraries and in the selection of variants that perform the required tasks and increase the success rate in finding relevant AAVs. Recently, computational models have predicted AAV viability in silico, which was further tested with AAV libraries in vitro.^110,111 Different methods to identify variants after initial screening rounds include calculations of enrichment factors and credibility scores.¹¹² Moreover, computational methods have recently helped guide capsid engineering and the understanding of AAV biology.¹¹³ Currently, it is possible to develop AAV variants with high viability using in silico design, and in future, with the study of complex directed evolution datasets, machine learning and AI can aid research groups in predicting and designing AAV variants with improved infectivity and transduction characteristics for uses in the retina.

Strategies to Overcome Immune Responses

Adverse immune outcomes due to AAV delivery to the eye include inflammation, reduced efficacy and clearance of transduced cells. The most common strategy to prevent or manage the immune response in clinical trials or clinics is by immunosuppression (Table 2). Non-specific immunosuppressive agents are provided to patients before, during and after the AAV injections. These corticosteroids treatments vary from one study to another and systematic comparisons evaluating their effects are lacking. Immunosuppression is also routinely used in large animal models such as canines and non-human primates. But these are provided as per the veterinarian’s recommendation and have not been systematically evaluated for retinal studies and there are not any standard recommendations available.

Several studies have shown that the dose of the vector administered influences the immune outcome. A study analyzing safety thresholds of AAV2 and AAV8 in NHPs reported anti-AAV NAB production and anti-transgene T-cell response at the highest doses tested.¹¹⁴ A recent study on 41 NHPs concluded that systemic levels of both anti-AAV BABs and NABs vary in a dose-dependent manner post ocular AAV injections.⁴⁶ A study comparing two engineered variants (AAV2-7m8 and AAV8BP2) reported that AAV2-7m8 was more efficient in transducing most retinal cell types, whereas AAV8BP2 had a better safety profile at higher doses.¹¹⁵ There are multiple factors such as the capsid, transgene or promoter that influence the immune response and with increased dose the effect is cumulative. Promoter-associated ocular toxicity and reduced function have been reported in mice. A comparison CAG (ubiquitous) and Rho (cell-specific) promoters reported reduction in retinal function (measured by ERG) with ubiquitous promoter.¹¹⁶ Another study compared the effect of many ubiquitous (CAG, CMV, UbiC) and cell-specific (RedO, Rho, CAR, GRK, Best1) promoters. They reported higher toxicity to both retina and RPE by ubiquitous promoters. They further reported higher microglial activation and migration caused by ubiquitous (CMV) promoter compared to cell-specific (RedO) promoter. They reported higher innate immune responses with ubiquitous promoters, as measured by levels of TNF-α, IL-1β, IL-6 and IL-γ.¹¹⁷ More recently, we compared serum antibody levels after ocular delivery of AAV-mediated gene delivery in nonhuman primates (NHPs). We showed that NHPs receiving ubiquitous (CAG) promoter had higher serum antibody levels compared to the ones that received cell-specific (SNCG) promoter.⁴⁶ Hence, using lower doses can ensure lower immune responses. This is particularly important when there is a requirement for readministration of AAV injections to maintain or augment therapeutic benefits.^18,19 Readministration may be required because of the bilateral nature of the diseases affecting the eyes, wherein the contralateral eye may need therapy months or years after the first eye. Such readministration has been tested in non-human primates and also reported in clinical trials.^18,118 Furthermore, presently efficient delivery to the photoreceptors is achieved by subretinal injections in clinical settings. This mode of delivery transduces only a small part of the retina, thus readministration to target other areas at later time points may be necessary to increase the therapeutic benefits.¹¹⁹

It would be highly desirable to have AAV variants that can evade immune responses and there have been attempts to engineer such viruses by rational design or directed evolution.¹²⁰ A study employed the directed evolution strategy to develop variants that can escape immune response. A library of AAV2 variants was generated by error-prone PCR and then these variants were screened for their ability to transduce cells in the presence of serum antibodies. The best variants identified in this screen had 10 to 100 fold higher resistance to neutralizing antibodies.¹⁰³ Another study used the capsid shuffling method (wherein the capsid sequences of known AAVs are fragmented and reassembled to generate variant sequences) to generate AAV variants library and screen these to find a less immunogenic variant. A novel variant identified from this screen named chimeric 1829 showed better transduction and lower reactivity to anti-AAV antibodies.¹²¹ A study identified the antibody binding regions on the AAV capsid using structural information acquired by cryo-electron microscopy. They further mutated these regions and reported that the variants thus obtained were more resistant to antibody neutralization.¹²² Apart from these attempts to engineer variants with desirable properties, another approach is to biomine variants to which humans would have no/low exposure. A study identified an ancestral AAV called Anc80 by in silico sequence reconstruction and reported this variant to be less immunogenic.¹⁰⁰ A recent study isolated 4 new AAVs originating from bats and reported these variants as being more efficient in evading neutralizing antibodies. However, the overall transduction efficiencies of these variants were lower than the ones currently used.¹²³

Another strategy to prevent the recognition of AAVs by NABs is to mask the antibody binding regions on the capsid surface by chemical modification. A polymer called Polyethylene Glycol (PEG) was conjugated with AAV to form PEGylated AAVs. These were reported to have higher resistance to NABs and result in enhanced transduction in the presence of NABs.^124–126 However, when the ratio of PEG/AAV was increased the transduction efficiency reduced most likely because excessive coverage with PEGs reduced the exposure of the AAV capsid regions essential for infectivity.¹²⁴ Some researchers have also attempted encapsulating the AAV capsid within gel beads or polymeric particles with the intention of providing initial protection against NABs and then progressively degrading the polymers and releasing AAVs.¹²⁷ Although this strategy can be useful for lentivirus to some extent, it proved inefficient for AAVs as the release and spread of AAVs is dependent on diffusion mechanisms.¹²⁸

Another strategy utilizes empty capsid (capsid decoys) that bind the anti-AAV antibodies, leaving the AAVs carrying the therapeutic gene intact. This was achieved by mutating the receptor binding site of AAVs and generating empty capsids. The empty capsids were injected along with the AAV capsid containing the transgene of interest. A higher transduction and low immunogenicity was reported when delivered systemically.¹²⁹ Another study used a similar strategy with additional inhibition of macrophages reporting improved transduction.¹³⁰ In these studies, the AAV delivery was systemic, wherein, on one hand, the threat and severity of immune response is higher but, on the other hand, higher doses can be tolerated. The eye is relatively immune privileged, but the dose and volume of AAV that can be tolerated is limited. The feasibility and usefulness of this technique for applications in the eye has not yet been tested.

Increased understanding of mechanisms by which AAV capsid and DNA interact with immune cells has enabled designing some recent immune evasion strategies. Immune cells have surface receptors called Toll-like Receptors (TLR) that sense DNA from pathogenic viruses containing unmethylated CpG motifs. CpG binding to TLR promotes its dimerization and activates TLR9 signaling. This leads to innate immune responses eventually recruiting other immune cells to the site of infection and further triggers adaptive immune responses.¹³¹ In 2019, two groups independently reported that the antibody response to AAV is mediated by toll-like receptor 9 (TLR9). They showed that TLR9 was responsible for activating T-cells against an AAV epitope and TLR9 also induces a cytotoxic T-cell response against the transgene resulting in loss of expression.^132,133 Another study identified TLR9 inhibitory sequences (TLRi) that were 12–24 nucleotides in size. They designed the transgene sequences to include the TLRi sequences and reported lower immune responses and higher transduction efficiency in mouse and pig models.¹³⁴ A study evaluated immune responses in Tlr null mutant (Tlr9-/-) and wild-type mice after injections of an immunogenic AAV variant (AAVrh32.33). They reported inhibition of IFN-γ T-cell responses toward capsid and transgene resulting in minimal cellular infiltrate and stable transgene expression.¹³⁵ As mentioned earlier, immune responses occur upon interaction of CpG motifs on viral DNA with TLR receptors on immune cells. CpG islands are clustered regions rich in CG motifs that are common in AAV vector sequences and promoter sequences. Hence, in addition to modifying TLR9 receptors, attempts have been made to optimize the CpG content of the AAV genome that was reported to reduce T-cell expansion, lowering the immune response.^135,136 The ITRs of AAVs are rich in CpG motifs. A recent study modified the ITRs and generated AAVs that contained CpG-free ITRs. They reported that the elimination of CpG motifs from the ITRs did not affect the biological activity of AAVs.¹³⁷ The use of these strategies, either independently or in combination, can result in improved gene delivery in the presence of antibodies and facilitate repeat AAV injections.

Conclusion and Future Perspectives

In the context of immune responses, it is important to make a distinction between tolerable immune responses and detrimental immune responses. When any substance is introduced in the ocular space, an immune response can occur, which indicates a normal and expected reaction. However, when this immune response escalates, spreads outside the area of the intervention and cannot be managed, it can be referred to as a detrimental immune response. It becomes a cause of concern for ocular gene therapy applications when the inflammation does not subside over time, resulting in worse visual outcomes and clearance of transduced cells. Hence, while designing and optimizing different parameters of the gene therapy, one must not only consider the type of immune response that may occur, but more importantly, if these responses will adversely affect the therapeutic outcome. So far, most studies observe or measure distinct aspects of the innate and adaptive immune system, and report the presence or absence of the immune response. Future studies can benefit from a combination of immunomonitoring tools that can provide an accurate account of the situation. Another aspect that is lacking is studies investigating cellular and molecular mechanisms of the immune responses in the ocular space. These can provide valuable insights for effective evasion and management of immune responses.

While discussing “where are we now?” with AAVs it is also important to consider “where we are going?”. The present clinical trials mostly attempt to use a “replacement strategy” to deliver a small transgene (below 4.7kb) for monogenic diseases with recessive mutations. But, as the strategies get more complicated – involving large editing tools (CRISPR, Prime and BE) and large genes (above 4.5kb) – alternatives to AAVs will be required. Although the feasibility of double and triple AAVs has been reported, presently the efficiency is not sufficient for therapeutic benefits.⁴⁹ Hence, there is a worldwide quest to develop non-viral vectors with the intention of delivering large cargo and to avoid viral components. Non-viral delivery can be as simple as delivering naked DNA or more sophisticated like using Virus-Like Particles (VLPs). There have been attempts of direct delivery of plasmid DNA,¹³⁸ but this results in transient expression and is prone to being easily degraded by host cellular mechanisms. Several studies have also explored deliveries of liposomes, polymers and nanoparticles conjugated with DNA or protein.⁹¹ Some of these result in stable particles but the lack of cell-specificity in most cases and large size in some cases restricts their use. Recently, created synthetic vectors called VLPs are nanoscale structures made up of one or more different molecules with the ability to self-assemble, mimicking the form and size of a virus particle, but lacking the viral genetic material and being non-infectious.¹³⁹ These new developments are interesting but none of these have efficiencies comparable to AAVs, thereby preventing their usage in vivo for retinal gene therapy applications. Moving forward, the field will benefit from development of both novel AAV vectors with enhanced properties as well as non-viral vectors that can carry larger or different cargo (DNA, mRNA or protein) while having AAVs’ desirable diffusion and cell entry properties.