Toward improved AAV gene therapies for retinal disorders: challenges and advances

ABSTRACT

Adeno-associated virus (AAV) vectors have transformed the landscape of in vivo gene therapy, with retinal diseases emerging as a major area of progress. The eye offers unique advantages as a therapeutic target: it is accessible, compartmentalized, and relatively immune-privileged, allowing localized delivery with reduced systemic effects. The landmark 2017 approval of the first AAV-based gene therapy for an inherited retinal disorder sparked a surge of clinical trials using AAV vectors – underscoring their potential for treating genetic eye diseases. However, challenges remain, including AAV’s limited capacity for large genes, suboptimal precision in cell-type-specific targeting, and inefficient transduction of certain retinal cells via minimally invasive routes. In response, researchers are engineering next-generation AAV capsids, optimizing gene expression cassettes, developing novel delivery strategies, and advancing tissue and organoid-based screening platforms. This article highlights these efforts as essential to overcoming current barriers in retinal AAV gene therapy.

1. Advances in AAV design and delivery

The 2017 approval of Luxturna – the first AAV-based gene therapy for RPE65-associated Leber congenital amaurosis, a rare inherited retinal disorder – marked a milestone, highlighting the transformative potential of AAV vectors in treating genetic eye diseases [1]. AAVs are small (25 nm), with a limited genetic packaging capacity (~4.7 kb), and can transduce various retinal cell types. Their capsid determines not only packaging limits but also diffusivity and cell-type-specific binding affinity, or tropism, while regulatory elements within their genetic cargo control where the loaded gene is expressed. As retinal gene therapies increasingly require precise cell-type-specific expression and delivery of large genetic payloads – such as long transgenes or complex gene-editing tools like base and prime editors – there is strong momentum to engineer both capsids and regulatory elements to overcome AAVs’ natural constraints.

1.1. Overcoming AAV size limitations

To circumvent size limitations, dual or even triple AAV vector strategies have been developed. These systems split the therapeutic gene into smaller fragments, each delivered by a separate vector, and rely on cellular mechanisms such as concatemerization and trans-splicing, or homologous recombination to reconstitute the full sequence in vivo. Multiple studies have shown successful dual-AAV delivery of large genes like MYO7A, ABCA4, or PCDH15, and even base editors, in relevant models such as human retinal organoids (RO) or non-human primate retina [2]. Nevertheless, these approaches tend to show limited gene reconstitution and lower expression compared to single-vector systems. For genes exceeding the collective capacity of two AAVs (e.g., CDH23), triple AAV systems have been tested, though expression remained low and strongly differed among species (complete fusion protein expression was observed in fewer than 5% of photoreceptors in murine eyes, compared to approximately 40% in porcine eyes).

The interspecies variability of multi-vector systems exhibit reinforces our view that species-specific screening is not just useful but essential to translational success (see Advances in AAV Screening Platforms) [3]. Importantly, multi-vector strategies require the successful transduction of a single cell by multiple vectors, which inherently reduces expression efficiency. Thereby, even increasing vector dose to improve chances for co-transduction offers limited benefit, as successful protein expression does not scale with dose [4]. Additionally, incomplete co-transduction of a cell can result in failed gene reconstitution or truncated protein expression, raising safety concerns. For instance, a truncated protein could misfold uncontrollably, interfere with native protein function, or trigger an unexpected immune response. To address this, degron signals or miRNAs targeting such aberrant proteins are being integrated into multi-vector systems. Nevertheless, despite their disadvantages, multi-vector systems remain a promising viral strategy for treating diseases involving large genes.

1.2. Enhancing cell-type targeting and delivery

Among the strategies to improve on-target transduction, the incorporation of regulatory elements such as cell-specific promoters into genetic cassettes to promote cell-specific expression has been a key focus. Selection of cell-type-specific promoters like GRK1, RPE65, or BEST1 restrict expression to target retinal cells, while ubiquitous promoters like CAG provide broader transgene expression. Another strategy involves incorporating miRNA target sites in the 3′ UTR (untranslated region that precedes the stop codon) to silence transgene expression in off-target cell types [5]. However, regulatory elements can consume significant vector space, contributing to the challenge of limited cargo capacity.

Beyond payload considerations, the route of delivery critically influences therapeutic outcomes. Subretinal injection (SRI) is currently the most used method, enabling localized delivery to the outer retina and effective transduction of photoreceptors and retinal pigment epithelium. However, SRI is invasive, involving localized retinal injury and the creation of a subretinal bleb, which can lead to retinal thinning or detachment. Intravitreal injection (IVI), by contrast, is less invasive and can potentially provide broader retinal coverage. Yet, its effectiveness is restricted by the inner limiting membrane (ILM), a significant barrier to AAV diffusion in larger animal, like pigs and primates, as well as in human eyes.

Because AAV capsid design determines both diffusivity and cell tropism, multiple strategies have been pursued to engineer capsids with improved properties. Rational design uses structural insights – such as crystallography and residue function – to guide specific amino acid modifications. However, rational design alone has not yet produced variants with robust and consistent pan-retinal expression in higher eye models, limiting its translational promise [6].

A powerful alternative is directed evolution, an approach that creates large capsid libraries through random mutagenesis or peptide insertions to show improved ILM penetration and subjects them to selection pressures. For instance, the AAV2.7m8 vector system engineered by directed evolution demonstrated enhanced retinal delivery in both animal models and early human studies [7].

To further accelerate capsid development, machine learning (ML) and artificial intelligence (AI) are increasingly integrated into AAV engineering workflows [8]. These tools can analyze large datasets from directed evolution to predict functional capsid designs in silico. However, it is noteworthy that the utility of ML models remains constrained by the quality and diversity of the underlying datasets – often limited by narrow mutation libraries – and by the challenge of significant vector performance differences across species and tissue types. Nevertheless, combined, rational engineering, directed evolution, and computational tools could redefine AAV capsid development for retinal gene therapies.

1.3. AAV-coupled technologies

Despite such significant advancements in capsid engineering and vector delivery, many clinical trials still rely on natural AAV serotypes, which we believe reflects both regulatory conservatism and the real-world inertia of manufacturing pipelines. This has spurred a growing trend to develop AAV-accompanying technologies that can universally enhance the performance of existing, clinically well-explored serotypes.

Promising approaches include enhancing delivery of AAVs with camelid antibody fragments, exosomes, or magnetic nanoparticles. Chemically attaching small, high-specificity camelid antibody fragments to AAV capsids has enhanced cell-specific transduction in vitro across multiple serotypes, although in vivo validation is pending [9]. Likewise, associating AAVs with exosomes (exo-AAVs) has improved pan-retinal transduction after IVI in rodent models – though their efficacy in larger eye models remains unproven [10]. Magnetic nanoparticles (MNPs) offer yet another promising avenue: these biophysical carriers can be guided by external magnetic fields to enhance AAV delivery across the ILM and have demonstrated efficacy in large eye models, like porcine eyes [11]. MNPs have been reported to be in vivo trackable in rodent eyes via standard MRI, and the extent of their penetration into retinal layers depends on the duration of magnetic field application. A recent study has shown that this approach can selectively target specific retinal layers after IVI in ex vivo porcine eyes using various AAV types [11]. In this and other studies on non-ocular magnetic virus delivery, MNPs have been shown to be compatible with various AAVs and other virus types [12]. Their ability to complex with viral particles via nonspecific electrostatic interactions – unlike camelid antibody fragments or exo-AAVs, which require capsid-specific engineering – suggests they could serve as versatile carriers for all AAV serotypes. How MNP-mediated enhancement of local AAV load translates into altered tropism behavior and improved layer-specific transduction remains to be fully understood. Moreover, although IVI of MNPs in rodents has been shown to be nontoxic across multiple ocular parameters – including retinal structure, photoreceptor function, aqueous drainage, and intraocular pressure – evidence in larger animal models remains limited to structural assessments in ex vivo porcine eyes using imaging alone, with no in vivo validation or long-term toxicity data [12]. Whether this method can also be universally and beneficially combined with capsid engineering or multi-vector strategies is still an open question – but represents a highly promising avenue to complement existing capsid and cargo innovations. Lastly, mild enzymatic digestion of the ILM has been explored to enhance AAV penetration, with some studies showing improved transduction even in primates [13]. However, intravitreal enzyme use has also been linked to rare ocular adverse events, including inflammation and vision loss, warranting caution for clinical application [14].

2. Advances in AAV screening platforms

Central to advances in retinal AAV gene therapy are tissue platforms that closely mimic the human ocular system, enabling the evaluation of emerging technologies for their therapeutic potential in patients. Traditionally, most retinal AAV studies have relied on rodent models, despite their poor rates for clinical translation due to disparities in retinal structure. Non-human primates (NHP) have the most similar retinal structure to humans but exhibit comparably higher variability and costs as well as ethical requirements. In vitro platforms such as human retinal explants (HRE), RO, and organ on a chip (OoC) have shown promise for mimicking the structure, function, and diseases of the human eye [15].

Of currently used models, HREs most closely recapitulate the cellular organization and architecture of the human retina as found in vivo [15]. HREs have demonstrated utility in comparing the performance of wild-type and engineered AAV candidates in terms of infection efficiency and cellular tropism to aid more efficient translation to the clinic [16]. However, HREs have innate associated challenges including the limited availability of cadaveric tissue and lack of utility for predicting long term efficacy of AAV gene therapy due to cell death and deterioration [15,16]. For instance, uptake of the AAV K91 was increased in HREs and NHP explants compared to a standard NHP model, which was attributed to disruption of the ILM at the explant edges [16]. Additionally, differences in donor age and health as well as tissue handling techniques impact HRE performance, while the cadaveric source of HREs limits their use in disease models and personalized medicine.

In contrast, ROs can be derived from induced pluripotent stem cells (iPSCs) for use in patient-specific inherited retinal disease models and have been maintained in culture for up to 200 days [17]. ROs have successfully recapitulated disease phenotypes associated with specific mutations, such as rod or cone death progression in RP2 or PRPF31 mutated cells derived from retinitis pigmentosa patients [18,19]. These studies further present that AAV-mediated augmentation of the disease gene could restore photoreceptor functionality within the disease ROs, even after long-term culture. Nevertheless, ROs are not yet capable of precisely mimicking the mature human retina or the functional blood-retinal-barrier (BRB), critical elements for supporting retinal photoreceptors and thus improving in vitro models for testing AAV efficacy. Specifically, the BRB’s strict regulation of oxygen and molecular transfer from the choroid to the sub-retinal space is necessary for normal retinal function, and without a functional BRB, retinal organoids may exhibit oxidative stress characteristic of diseases such as macular degeneration [20,21]. Additionally, the absence of glial cells that support and protect neurons as well as immune cells in ROs restricts effective modeling of neuroinflammation in retinal degenerative diseases [22].

Screening drugs for use in patients with different genetic backgrounds is an end goal for the development of human-mimetic in vitro models in general, including in ROs and OoCs. However, efficacy of RO generation varies significantly between stem cell lines, with some lines consistently failing to develop into ROs or mature. This failure decreases reproducibility and hampers efforts to establish reliable RO-based models, and may be associated with inhibited BMP signaling [23]. Maturation of ROs is also significantly limited by the derivation of Matrigel from mouse sarcoma, a xenogeneic material with high compositional variability between batches that compromises researchers’ abilities to elucidate the mechanisms behind RO development [24,25]. Greater reproducibility for generation of ROs with RPE and neural retina has been achieved using a Matrigel-free protocol based on the timing of BMP-4 supplementation, but lacks environmental mechanical cues that play a role in cell fate and maturation [22,24,25].

Complexity of ROs may be improved by identifying alternative culture substrates, other than Matrigel, that more closely approximate human retina-like microenvironments or by integrating them into OoC platforms [17,20,26]. Alternative culture substrates are an understudied area with potential to improve the maturation and efficacy of ROs as disease models for AAV gene therapy [25]. For instance, a high-throughput PMDS-based platform based on a brain organoid protocol was able to generate ROs with neural retina, ciliary margin, and RPE without manual transfer and isolation steps, Matrigel, or BMP supplementation [25]. Additionally, methodologies utilizing agarose substrates with human ECM protein coatings such as laminin and vitronectin have recently generated ROs on similar timescales to the standard Matrigel-based protocols [27].

Careful control of nutrition and gas can help improve RO morphology, including laminar stratification, photoreceptor maturation, and long-term maintenance of neuron functionality [28,29]. OoC platforms have been used to produce dynamic retinal models with greater cell viability and improved mimicry of the in vivo environmental conditions of retinogenesis [17,26]. For instance, the introduction of a physiological oxygen gradient within retinal OoC between the hypoxic conditions of the inner retina (2% O2) and outer retina (18% O2) was shown to promote differentiation and longer-term culture of retinal ganglion cells, which has historically been a limiting factor in Ros [26]. However, OoCs exhibit challenges such as limited oxygen and nutrient diffusion into the organoid core due to the lack of integrated vasculature, and culture artifacts, and affect on cell fate associated with constraints on organoid shape and size [30,31].

Compared to standard RO and OoC platforms, bioprinting has been suggested to better reproduce the BRB by providing angiogenic ECM structures, which are not feasible with RO and OoC systems. However, bioprinting has faced challenges with poor resolution, resulting in vessels with larger diameters and low stability compared to capillaries in the human choroid [20]. Recently, approaches such as printing dense patterns of endothelial cells, pericytes, and fibroblasts have produced more stable capillaries with physiological diameters (5–20 µm) [20]. Interestingly, it was presented that culturing iPSC-derived RPE patches on bioprinted scaffolds stimulated the expression of Bruch’s membrane proteins and rendered them capable of fusing into porcine and rat BRBs upon transplantation [32]. Such research underlines the vast potential that combining RO, OoC, and tissue engineering approaches has in developing models that best possibly approximate the human retina in vitro – as required for appropriate AAV evaluation.

At present, efforts to improve ROs focuses on troubleshooting individual challenges such as maturation of functional photoreceptors, achieving more complex cellular composition such as by differentiation of iPSCs into microglial cells, or improving BRB structure, which may develop through interrelated mechanisms in the human retina [33]. To produce ROs that are truly representative models of the human retina, these and forthcoming innovations will ultimately need to be integrated. Since ROs can be derived from iPSCs and embryonic stem cells (ESCs), their use is subject to data protection regulations and requires appropriate ethical approval and patient consent [34]. Compared to ESCs, iPSCs raise fewer ethical concerns, as they originate from adult somatic cells, which avoids the controversies around embryonic stem cells. Moreover, the ability to generate patient-specific iPSCs enables personalized disease modeling without the need for embryo-derived cells, while reducing the ethical burden of animal models.

3. Conclusion

Designing effective, long-lasting AAV gene therapies requires improved cell targeting and delivery strategies, as well as models with greater relevance to human disease. Current trends in AAV engineering aim to provide variants with robust cell-type specific expression, improved capacity for delivering large therapeutic cargos, and enhanced diffusivity across the ILM to enable IVI. As this approach is centered around developing novel AAV types, non-viral delivery platforms are being explored that could potentially enhance afore-mentioned properties across multiple AAV types – including those that already have obtained clinical approval. Lastly, integrating tissue engineering design strategies with RO and OoC to produce more clinically-relevant human retina models has potential to enable reproducible screening of AAV gene therapies, while reducing reliance on animal retina models.

4. Future perspective

Looking ahead, we anticipate that the future will bring a shift from selection between limited numbers of clinically-approved serotypes toward modular design of AAV vectors tailored for specific retinal cell types, delivery routes, and therapeutic cargos. Advances in high-throughput capsid screening, informed by ML and multi-species data integration, could transform capsid discovery into a predictable, data-driven process. At the same time, emerging delivery technologies like MNPs and exo-AAVs may be used not as alternatives but as enhancers of clinically validated vectors. Finally, we foresee a paradigm shift in preclinical testing, where increasingly sophisticated in vitro platforms – combining organoid systems and advanced tissue engineering – will gradually replace animal models as the gold standard for evaluating AAV performance and safety.

Funding Statement

This manuscript was in part supported by The City of Dublin Skin and Cancer Hospital Charity. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Article highlights

• This article highlights the current advances in retinal AAV design (e.g., size limitations and tropism) and delivery (e.g., exosomes and magnetic nanoparticles) for treating retinal disorders

• This article discusses the current advances in retinal AAV screening platforms including traditional in vivo animal retinal models and in vitro retinal explants or organs-on-chip platforms.

• This article foresees that this field will move toward developing more optimized design of retinal AAV vectors tested with higher throughput and more clinically relevant in vitro human retinal models

Author contributions

Oliver Siontas: Writing/editing of manuscript

Mika Brown: Writing/editing of manuscript

Seungkuk Ahn: Conceptualization, Writing/editing of manuscript

Disclosure statement

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

No writing assistance was utilized in the production of this manuscript.

Reviewer disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.