Beyond Subretinal Injections: How AI is Reshaping AAV Capsid Engineering for Retinal Gene Therapy
Overcoming a Central Barrier in Retinal Gene Therapy
For retinal gene therapy, an effective therapeutic sequence is only part of the solution. The vector must also reach the right retinal cells, achieve sufficient expression, and perform consistently across the experimental systems used to support clinical development. Photoreceptors are particularly challenging targets because they lie deep within the retina and are difficult to access following intravitreal administration.
A recent study published in Molecular Therapy Advances, “AI-engineered AAV Capsid Enables Intravitreal Delivery for the Treatment of Diverse Retinal Degenerations,” addresses this long-standing delivery challenge through artificial intelligence-guided AAV capsid engineering.
Co-authored by scientists from Cyagen’s preclinical CRO platform and collaborating research teams, the study reports the development and validation of AAV2.PN168, an engineered AAV2 capsid designed to improve retinal penetration and transduction after intravitreal injection. The work integrates AI-based capsid design, vector production, species-specific screening, Cyagen-developed retinal disease models, ocular procedures, and multimodal efficacy evaluation.
Rather than evaluating capsid performance alone, the researchers connected vector engineering directly with therapeutic proof-of-concept studies in GUCY2D-associated Leber congenital amaurosis type 1 and VEGFA-driven wet age-related macular degeneration.
Why Intravitreal Retinal Gene Delivery Remains Challenging
AAV vectors are widely investigated for ocular gene therapy because they can support sustained transgene expression in retinal cells. However, therapeutic performance depends heavily on both the capsid and the route of administration.
For diseases involving photoreceptors or the retinal pigment epithelium, many investigational therapies rely on subretinal injection. This approach places the vector near the target cells but requires the creation of a localized subretinal bleb. It can introduce additional mechanical stress into an already compromised retina, while transduction is generally concentrated within the treated region.
Intravitreal injection is less invasive and already routinely used for ophthalmic drug delivery. Its principal limitation is biological. Conventional AAV capsids often struggle to cross the inner retinal barriers and reach photoreceptors in the outer nuclear layer.
A capsid capable of achieving broad retinal transduction from the vitreous could therefore support less invasive delivery while expanding the retinal area exposed to a therapeutic transgene.
An Experimental Data-Driven AI Engineering Workflow
The researchers developed AAV2.PN168 from a diverse library of AAV2 capsid variants containing seven- to twelve-amino-acid peptide insertions near residues R587 and R588, a region commonly used to modify AAV tropism.
They then used large-scale experimental data to train AI models to predict two key properties: capsid viability and retinal targeting. Viability was assessed by comparing variant abundance before and after vector production, while retinal targeting was evaluated by sequencing RNA recovered from mouse and cynomolgus monkey retinas after intravitreal administration. Because the readout was RNA-based, it reflected not only retinal delivery but also cellular entry and transgene expression.
Separate mouse and non-human primate targeting models were integrated into iterative workflows combining sequence generation, AI scoring, and multi-objective optimization. This enabled simultaneous selection for production fitness and species-relevant retinal activity, ultimately leading to AAV2.PN168. The capsid also showed favorable manufacturing performance, including a higher purified yield than wild-type AAV2 under the reported conditions.
Broad Retinal Transduction in Non-Human Primates
The non-human primate findings represent an important translational component of the study. Differences in ocular anatomy, receptor expression, and retinal barriers can prevent capsids identified in rodents from reproducing the same performance in primates.
To identify candidates with activity in primate retinal tissue, the researchers evaluated a focused library containing 43 AI-designed capsids alongside wild-type AAV2 and AAV2.7m8, an engineered capsid previously investigated for intravitreal retinal delivery.
Each vector carried a barcoded eGFP expression cassette, enabling capsid activity to be resolved at the single-cell level. Following intravitreal administration and single-cell RNA sequencing, AAV2.PN168 emerged as a leading candidate.
The capsid transduced multiple retinal cell populations, with notable activity in rod and cone photoreceptors. In a subsequent comparison at an equivalent dose, AAV2.PN168 produced broader retinal GFP expression than AAV2.7m8 and a higher proportion of GFP-positive cells in the outer nuclear layer across macular, central, and peripheral regions.
These findings are particularly relevant because the outer nuclear layer contains the photoreceptors required for light detection. Reaching these cells through intravitreal administration remains a major objective in retinal gene therapy.
The study nevertheless represents an early proof of concept. The authors noted that larger non-human primate cohorts, additional biological replicates, longer follow-up periods, and formal safety studies will be required to evaluate clinical translation. Crucially, at an equivalent dose, AAV2.PN168 produced broader retinal GFP expression than AAV2.7m8, achieving a higher proportion of targeted cells across macular, central, and peripheral regions.
Cross-Species Performance in the Mouse Retina
Although AAV2.PN168 was prioritized through the non-human primate design workflow, it also demonstrated substantial activity in mice.
The study assessed retinal transduction using several complementary endpoints, including single-cell RNA sequencing, in vivo bioluminescence imaging, fundus fluorescence imaging, retinal histology, and molecular quantification of transgene expression.
Across these analyses, AAV2.PN168 achieved stronger or broader retinal activity than wild-type AAV2 and outperformed AAV2.7m8 in several experiments.
Its significance therefore lies not only in high activity within a single species, but also in its ability to retain retinal performance across mice and non-human primates. Cross-species activity may be especially valuable during preclinical development, where efficacy, pharmacology, biodistribution, and safety programs often progress through multiple animal systems.
Therapeutic Validation in Cyagen-Developed Retinal Disease Models
To determine whether improved retinal delivery could produce therapeutic benefit, the researchers evaluated AAV2.PN168 in two retinal disease models developed by Cyagen.
GUCY2D-Associated LCA1 Model
The researchers used a Gucy2e/Gucy2f double-knockout mouse model to represent GUCY2D-associated Leber congenital amaurosis type 1 (LCA1). The model exhibited substantial retinal dysfunction and outer nuclear layer thinning, providing a defined system for evaluating photoreceptor-directed gene replacement.
AAV2.PN168 delivered human GUCY2D under the control of the photoreceptor-specific hGRK1 promoter. Following a single intravitreal injection, treated mice demonstrated improved scotopic and photopic electroretinography responses and higher retinal hGUCY2D expression than mice treated with the comparator capsid.
Retinal function was not restored to wild-type levels, and structural recovery of the outer nuclear layer was not demonstrated. The data therefore support functional rescue of remaining photoreceptors rather than reversal of established degeneration.
hVEGFA Transgenic Wet AMD Model
The researchers also evaluated AAV2.PN168 in a Cyagen-developed hVEGFA transgenic mouse model representing wet age-related macular degeneration (wet AMD). In this model, retinal expression of human VEGFA produced abnormal vascular activity and fluorescein-detectable leakage.
AAV2.PN168 was used to deliver an anti-VEGFA antibody fragment. A single intravitreal administration produced higher retinal anti-VEGFA expression than wild-type AAV2 and AAV2.7m8. Fluorescein angiography showed progressive improvement in leakage, with AAV2.PN168 producing a faster and stronger response than the comparator capsids during the 21-day study.
These findings illustrate how AAV-mediated expression could potentially extend intraocular anti-VEGFA activity. Longer studies remain necessary to assess durability, expression control, immunogenicity, and long-term safety.
A Demonstration of Cyagen’s Integrated Preclinical CRO Capabilities
This publication reflects more than the identification of an engineered capsid. It demonstrates how AI-guided vector design can be integrated with disease-relevant models and comprehensive in vivo validation.
Scientists from Cyagen’s CRO platform contributed directly to the study, while Cyagen-developed experimental mice supported therapeutic evaluation in two distinct retinal disease settings.
The broader workflow incorporated AAV production and purification, intravitreal administration, fundus imaging, fluorescein angiography, electroretinography, retinal histopathology, ddPCR, single-cell RNA sequencing, biomarker analysis, and quantitative efficacy assessment.
Cyagen’s preclinical platform can support retinal gene therapy programs through:
- Vector Engineering: AI-guided AAV capsid design, screening, and customized vector construction/packaging.
- In Vivo Modeling & Delivery: Custom retinal disease models and precise intraocular administration.
- Multimodal Efficacy Evaluation: OCT, fundus imaging, ERG, and behavioral testing.
- Molecular & Histological Analysis: Single-cell RNA sequencing, ddPCR, and immunostaining.
By combining vector engineering, Cyagen-developed animal models, and integrated ophthalmic CRO services, researchers can evaluate capsid tropism, transgene expression, functional outcomes, and therapeutic efficacy within a coordinated preclinical workflow.
👉 Advance Your Retinal Gene Therapy Program with Cyagen
👉 Connect with our scientific team to identify the right preclinical strategy for your program.
Reference
Cui M, Liu H, Cai L, Zhang Q, Yuan L, Gao C, et al. AI-engineered AAV capsid enables intravitreal delivery for the treatment of diverse retinal degenerations. Mol Ther Adv. 2026. doi:10.1016/j.omta.2026.201806..
