Improving Cross-Species Translation of Engineered AAV Capsids

Introduction

Adeno-associated virus, or AAV, has become one of the most important delivery vehicles in modern in vivo gene therapy. Its broad tissue tropism, relative safety profile, physical stability, and ability to support long-term transgene expression have made it a preferred vector platform for many therapeutic areas, including ophthalmology, neurology, neuromuscular disease, metabolic disease, and rare genetic disorders.

A recent review article published in Gene Therapy, titled “Recent advancements in improving cross-species applicability of bioengineered AAV capsids,” highlights both the promise and the translational complexity of AAV capsid engineering. The authors note that AAV is widely accepted as an in vivo gene delivery vector because of its relatively low immunogenicity, minimal toxicity, sustained efficacy, and broad tropism. At the same time, they emphasize that unpredictable cross-species applicability remains a major obstacle for broader clinical translation. The review focuses on rational design, directed evolution, and AI-based capsid design as strategies to create AAV variants with stronger human translatability.

This issue is highly relevant for gene therapy development. Many AAV capsids are first discovered or optimized in mice, but the biological environment in mice can differ significantly from that in non-human primates, large animals, and humans. A capsid that appears highly potent in a mouse model may lose activity in humans, show altered biodistribution, or trigger unexpected safety concerns. Therefore, the question is no longer simply: Can we engineer a stronger capsid? The more important question is: Can we engineer a capsid whose performance can translate across species and ultimately support safe, effective human therapy?

Why Cross-Species Translation Is a Core Challenge in AAV Development

AAV-based gene therapy has already achieved important clinical milestones. The review references the landmark approval of Luxturna for inherited retinal disease and Zolgensma for spinal muscular atrophy as examples that helped establish AAV as a clinically validated gene delivery platform.

However, AAV’s success in approved therapies does not eliminate the translational challenge. Each new capsid, target tissue, dose, route of administration, and therapeutic payload introduces new development risk. In particular, capsid performance may vary dramatically between species because AAV transduction depends on multiple biological steps:

1. Attachment to cell-surface receptors or co-receptors
2. Endocytosis and intracellular trafficking
3. Endosomal escape
4. Nuclear entry
5. Capsid uncoating and genome release
6. Transgene expression
7. Immune recognition and clearance
8. Tissue-specific toxicity

Each of these steps can be affected by species-specific biology. The review notes that interspecies genetic differences can influence protein structure, post-translational modifications, splicing mechanisms, epigenetic regulation, and host factors involved in AAV transduction. These differences affect the AAV life cycle from cellular attachment to genome release.

This is why mouse efficacy alone has limited predictive value. A knockout mouse may reproduce a human disease phenotype, but that does not guarantee that an AAV vector will enter the same human cell type, use the same receptor, traffic through the same intracellular pathway, or produce the same safety profile.

The AAV-PHP.B Lesson: Potency in Mice Is Not Enough

One of the clearest examples of cross-species uncertainty is AAV-PHP.B. This engineered capsid showed strong CNS transduction in certain mouse strains, which generated major interest in systemic AAV delivery to the brain. However, subsequent studies showed that its neurotropic properties were highly dependent on mouse-specific biology and did not translate predictably to non-human primates.

The review specifically highlights AAV-PHP.B as an example of a capsid developed through directed evolution in a single mouse model that later showed restricted tropism and poor performance in other species.

The key lesson is that a capsid can be excellent in one biological context but fail when the selection pressure is not aligned with human-relevant biology. For translational AAV development, capsid discovery must be designed around conserved mechanisms, clinically relevant tissue barriers, and appropriate validation models.

Five Barriers That Determine Cross-Species AAV Performance

The review organizes cross-species translation around several species-dependent barriers: receptor usage, intracellular trafficking, immune recognition, transcriptional regulation, and toxicity. These factors determine whether an engineered capsid can retain its function beyond the original discovery species.

1. Receptor Usage

AAV capsids interact with cell-surface receptors, glycans, and co-receptors to enter cells. If the target receptor is conserved across species, a capsid has a stronger chance of translating from animal models to humans. If the receptor is species-specific, the vector may perform well in mice but poorly in NHPs or humans.

This is especially important for CNS, muscle, lung, and tumor-targeted AAV programs, where tissue barriers and receptor expression patterns can differ substantially across species.

2. Intracellular Trafficking

After cell entry, AAV must avoid degradation, escape the endosome, reach the nucleus, and release its genome. Some intracellular mechanisms are broadly conserved, such as ubiquitination-related degradation pathways, while others may differ depending on cell type or species. Capsid mutations that improve intracellular trafficking may therefore offer a route to broader cross-species performance when they target conserved cellular pathways.

3. Immune Recognition

AAV vectors are not invisible to the immune system. Capsids can be neutralized by pre-existing antibodies, activate complement, or trigger T-cell responses against transduced cells. Immune recognition patterns can differ between rodents, NHPs, and humans, making it difficult to rely on one model to predict clinical immunogenicity.

4. Transcriptional Regulation

Even when a capsid reaches the right cell, transgene expression depends on promoter and enhancer activity. A promoter that performs well in mouse tissue may not have the same activity profile in human tissue. This means that cross-species translation cannot be solved by capsid engineering alone; expression cassette design also matters.

5. Toxicity

Toxicity may be the most clinically consequential barrier. The review discusses hepatotoxicity, dorsal root ganglion pathology, complement activation, and thrombotic microangiopathy as safety concerns that can differ across species.

Strategy 1: Rational Design

Rational capsid design uses existing knowledge of AAV structure and function to make targeted modifications. Scientists may modify receptor-binding regions, introduce peptide insertions, alter surface-exposed amino acids, reduce degradation signals, or modify immune epitopes.

The review describes rational design as a knowledge-driven strategy that relies on structural and biochemical understanding of the capsid. It can target regions involved in receptor binding, cellular trafficking, intracellular degradation, and immune recognition.

Peptide Insertion for Retargeting

One rational design strategy is inserting receptor-recognizable peptides into the capsid surface. These peptides may help the vector interact with a desired receptor or cross a biological barrier. For example, the review discusses AAV-ie, where a CPP-like peptide was inserted into an AAV-DJ backbone to help the vector reach inner ear cell populations.

This type of strategy is attractive because it can directly modify capsid-cell interactions. However, for clinical translation, the receptor or cellular pathway engaged by the inserted peptide should ideally be conserved across species.

Reducing Capsid Degradation

Another rational strategy is to reduce capsid ubiquitination. Surface-exposed tyrosine residues on AAV capsids can be phosphorylated, leading to ubiquitination and proteasomal degradation. Mutating these residues can improve intracellular trafficking and increase the number of particles that reach the nucleus.

The review notes that tyrosine-mutant AAV vectors have shown efficient transduction in ocular cell lines derived from both mice and humans, suggesting that ubiquitination-mediated degradation may represent a conserved intracellular barrier.

Reducing Immune Recognition

Capsid engineering can also be used to reduce immune recognition. For example, antibody-binding regions can be modified to avoid neutralization. However, immune evasion is difficult because antibody epitopes and complement recognition patterns may differ between humans and animal models. A capsid designed to evade human serum may not show the same behavior in rodents or NHPs, complicating preclinical testing.

Strengths and Limitations of Rational Design

Rational design is powerful when the biology is known. It is relatively predictable, mechanism-driven, and well suited for solving defined problems. However, its discovery space is limited by what is already understood. If the key receptor, trafficking pathway, or immune determinant is unknown, rational design may not identify the best solution.

For this reason, rational design is often most effective when paired with broader discovery approaches such as directed evolution and AI-assisted prediction.

Strategy 2: Directed Evolution

Directed evolution takes a more empirical approach. Instead of engineering one or a few specific mutations, researchers create large capsid libraries and apply selection pressure to identify variants with desired properties.

The review describes directed evolution as a method that generates large libraries of AAV variants, screens them against target tissues or cell lines over multiple rounds, and then sequences and amplifies enriched candidates. The final candidates are typically evaluated further in models such as NHPs to assess translational potential.

Cre-Dependent Directed Evolution

Cre-dependent systems use cell type-specific Cre expression to select vectors that transduce specific cell populations in vivo. For example, M-CREATE uses Cre-expressing target cells to selectively amplify AAV variants that enter the desired cell type. This approach can improve cell-type specificity during screening.

However, many Cre-dependent screens are performed in mice, which creates translational risk. If the selected capsid depends on a mouse-specific receptor or tissue environment, it may not perform similarly in humans. AAV-PHP.B is an important cautionary example because its strong CNS activity in mice was linked to biology that did not translate well to primates.

Cre-Independent Directed Evolution

Cre-independent approaches can reduce some of these limitations by screening in more translationally relevant tissues or species. The review describes AAV.Ep, which was identified through NHP brain screening and showed elevated transduction in mouse brain, NHP brain, and human iPSC-derived cortical neurons. It also describes AAV2.GL and AAV2.NN, which were selected for photoreceptor transduction and showed activity in mice, dogs, and NHPs.

These examples show how screening directly in large-animal or human-relevant systems can improve confidence that a capsid will maintain activity beyond the mouse model.

DELIVER and Muscle-Tropic AAV Engineering

The DELIVER platform is another important example. It uses tissue-specific promoter-driven selection to identify AAV variants with functional activity in target tissues across animal models. In this system, capsid library expression is controlled by muscle-specific promoters, allowing researchers to enrich variants that successfully transduce muscle cells.

The review highlights MyoAAV4a, a variant identified after selection in cynomolgus macaque muscle tissue. MyoAAV4a showed enhanced muscle transduction in both cynomolgus macaques and humans. Mechanistically, RGD peptide insertion was linked to integrin-related muscle targeting, and integrin expression was demonstrated across mouse, NHP, and human muscle.

This is an important translational principle: a capsid is more likely to cross species when it targets a receptor or pathway conserved across species.

CNS and Blood-Brain Barrier Targeting

The review also discusses TRACER, a platform conceptually similar to DELIVER. AAV variant VCAP-102 was identified through TRACER and shown to interact with the conserved brain vascular receptor ALP, enhancing blood-brain barrier crossing.

For CNS gene therapy, this type of conserved receptor targeting is especially important. The blood-brain barrier is a major delivery obstacle, and species differences in vascular biology can make mouse-to-human translation particularly difficult.

Strengths and Limitations of Directed Evolution

Directed evolution can uncover capsids that would be difficult to design rationally. It does not require complete prior knowledge of receptor biology, and it can screen large sequence spaces under functional selection pressure.

However, it has limitations. The output can be stochastic, and the mechanism behind improved performance may remain unclear. The review notes that variants such as AAV.CAP-B10 or AAV2.GL may show superior performance, but mechanistic understanding is often lacking. It recommends integrating receptor discovery, membrane proteomics, receptor knockout or overexpression validation, and structural biology to create a “selection—target discovery—mechanism validation” loop.

Strategy 3: AI-Assisted Capsid Design

AI-assisted design is becoming an increasingly important tool in AAV engineering. Machine learning models can analyze sequence-function relationships and predict capsid variants with desired properties. This can help reduce the burden of animal screening and accelerate candidate prioritization.

The review notes that AI design uses training data to identify patterns and make predictions on new, unseen sequences.

AAV.Anc80L65 and Ancestral Reconstruction

One of the best-known AI-related capsid examples is AAV.Anc80L65, developed through maximum likelihood ancestral sequence reconstruction. This approach computationally reconstructed an ancestral AAV sequence from wild-type AAV data.

According to the review, AAV.Anc80L65 demonstrated superior liver transduction in both mice and primates compared with wild-type AAV, and later studies expanded its application to NHP cochleae.

AI for Multi-Parameter Optimization

AI approaches may eventually help optimize multiple capsid properties at once, including tissue tropism, manufacturability, immune evasion, cross-species applicability, and safety. The review highlights Fit4function as an example of a model that links capsid amino acid sequence features with functional traits such as liver targeting and cross-species applicability. However, the authors also note that incorporating NHP transduction data could improve prediction accuracy.

Data Quality Is the Key Limitation

AI models are only as reliable as their training datasets. If a model is trained mainly on mouse or in vitro data, it may not predict human performance accurately. High-quality datasets from NHPs, human primary cells, organoids, and multi-omics studies will be essential for improving AI-driven capsid prediction.

The review emphasizes that AI-based design can reduce the cost of broad animal screening, but it requires robust training datasets and computer-generated capsids must still undergo comprehensive in vivo validation.

Safety and Toxicity: Engineering a Wider Therapeutic Window

AAV capsid engineering is not only about improving potency. It is also about expanding the therapeutic window. A better capsid should ideally deliver more vector to the target tissue, reduce off-target exposure, lower the required dose, and reduce immune or toxic burden.

The review describes toxicity as perhaps the most clinically consequential cross-species barrier. Hepatotoxicity, DRG pathology, complement activation, and thrombotic microangiopathy can differ across species and may not be reliably predicted by mouse studies.

Dose Reduction Through Better Targeting

One of the most practical safety benefits of improved capsid design is dose reduction. If an engineered capsid can deliver the payload more efficiently to the desired tissue, the total administered vector dose may be reduced. This can lower liver exposure and reduce systemic immune burden.

The review cites MyoAAV4a as an example, noting that it achieved therapeutic muscle transduction in NHPs at a 10-fold lower dose than AAV9, potentially reducing liver accumulation and associated hepatotoxicity. It also highlights AAV.CAP-B10, which showed reduced liver targeting in marmosets.

Species-Specific Toxicity Requires Large-Animal Testing

Some toxicities cannot be solved by dose reduction alone because they may involve species-specific molecular interactions. The review notes that DRG pathology following high-dose AAV administration has been documented in pigs and NHPs but not in mice. For such barriers, empirical testing in multiple large-animal models remains essential.

ITR Engineering as an Orthogonal Safety Strategy

Although capsid engineering is central to tropism and delivery, other vector design elements can also influence safety. The review discusses ITR engineering as an orthogonal strategy, noting that reshaping the ITR region to eliminate p53 binding sites may reduce AAV-induced apoptosis in human embryonic stem cells.

This highlights an important point: safer AAV therapy will likely require optimization of the entire vector system, not just the capsid.

Method	Core advantage	Main weakness	Translational lesson
Rational design	Precise modification based on known capsid structure and mechanism.	Limited by incomplete knowledge and by mechanisms that may not be conserved across species.	Works best when the target biology is already understood and broadly conserved
Directed evolution	Broad discovery without requiring complete prior mechanistic knowledge.	Often yields strong variants without fully explaining why they work.	Selection must occur in biologically relevant systems, or the result may be optimized for the wrong species context.
AI-based design	Can model sequence-function relationships and optimize multiple traits efficiently.	Strongly dependent on training-data quality and still requires substantial in vivo validation.	Prediction improves when cross-species and human-relevant datasets are included.

Capsid Engineering Alone Is Not Enough

AAV capsid engineering can improve receptor binding, biodistribution, intracellular trafficking, and immune profile. However, it does not fully solve every translational challenge.

The review points out that transcriptional regulation is a critical gap because capsid engineering alone does not determine promoter or enhancer activity. Cell-specific promoters and regulatory elements are needed to drive expression in the intended cell type while limiting off-target expression.

For example, a muscle-targeted capsid may improve delivery to skeletal or cardiac muscle, but promoter selection still determines whether transgene expression is restricted to the desired cells. Similarly, a CNS-targeted capsid may cross the blood-brain barrier, but promoter choice influences whether expression occurs in neurons, astrocytes, oligodendrocytes, or other cell populations.

This means that translational AAV design should combine:

• Capsid selection or engineering
• Tissue- or cell-specific promoter design
• Payload optimization
• Dose selection
• Route of administration
• Manufacturing quality control
• Biodistribution analysis
• Immunogenicity assessment
• Large-animal and human-relevant validation

Building a More Predictive AAV Development Workflow

The review argues that early AAV research often relied too heavily on mice, making human efficacy unpredictable. To improve translation, current solutions rely increasingly on large-animal models, especially NHPs, as well as human-derived systems such as cultured cells and organoids.

A more predictive AAV development workflow should include several layers.

1. Start With Human-Relevant Biology

Instead of asking only which capsid works best in a mouse model, developers should ask which receptors, transport mechanisms, and intracellular pathways are conserved between the model and humans.

The review recommends identifying receptors and intracellular partners that AAV variants engage with, then prioritizing capsids targeting proteins with high conservation in sequence, structure, and tissue-specific expression.

2. Use Multi-Species Screening When Possible

Directed evolution can be more predictive when selection pressure is applied across multiple species or human-relevant systems. This approach reduces the chance of selecting variants that depend on one species-specific receptor.

3. Integrate Human Cells and Organoids

Human primary cells, iPSC-derived cells, organoids, and ex vivo tissue models can provide valuable translational insight. These models cannot fully replace animal testing, but they can help identify capsids that interact with human biology before costly in vivo studies.

4. Use AI as a Prediction Tool, Not a Replacement for Validation

AI can help prioritize candidate capsids, but predicted variants still need biological validation. The strongest AI workflows will likely combine sequence data, functional screening, multi-omics, NHP data, and human cell data.

5. Validate Safety Across Relevant Models

Toxicity is not always linear across species. The review cautions against the idea that cross-species research follows a simple path from mice to pigs to NHPs to humans. Evolutionary biology does not always follow a linear pattern, and variants selected through multiple animal models may still fail in human trials.

Practical Implications for AAV Researchers

For academic and translational researchers working with AAV, the key message is clear: capsid choice should be guided by the biological question, target tissue, model system, and translational objective.

For early discovery studies, standard serotypes may be sufficient. But for therapeutic development, especially systemic delivery, researchers should consider whether the capsid’s mechanism of action is relevant to humans.

Important questions include:

• What receptor or pathway does the capsid use?
• Is that receptor conserved across mouse, NHP, and human tissues?
• Does the target cell type express the receptor in human tissue?
• Is the capsid expected to increase off-target liver exposure?
• Does the capsid allow dose reduction?
• Has the capsid been tested in human-derived cells or organoids?
• Has it been evaluated in large animals or NHPs?
• Does the promoter behave similarly across species?
• What immune or toxicity risks are expected?
• Can the vector be manufactured consistently at the required scale?

These questions help move AAV development beyond simple potency ranking and toward clinically meaningful vector selection.

How PackGene Supports AAV Research and Development

As AAV programs become more sophisticated, researchers need integrated support across vector design, production, capsid selection, analytical testing, and translational planning. PackGene supports AAV research with customizable AAV production and vector development solutions designed to help researchers evaluate capsid performance, construct design, and application-specific delivery needs.

For early-stage academic work, this may involve selecting a suitable serotype, optimizing vector genome design, and producing RUO-grade AAV for in vitro or in vivo studies. For more advanced translational programs, additional considerations may include scalable production, impurity control, full/empty capsid analysis, genome integrity assessment, potency assay development, and comparability planning.

The future of AAV development will require stronger coordination between discovery biology and manufacturability. A capsid must not only perform well biologically; it must also be producible, purifiable, characterizable, and suitable for downstream development.

Conclusion

AAV capsid engineering is advancing rapidly, but cross-species translation remains one of the most important challenges in the field. A capsid that works well in mice may not retain the same performance in NHPs or humans. This is because AAV delivery depends on many species-sensitive variables, including receptor usage, intracellular trafficking, immune recognition, transcriptional regulation, and toxicity.

Rational design offers mechanistic precision. Directed evolution offers discovery power. AI-assisted design offers prediction and scale. However, none of these strategies is sufficient alone. The strongest future approaches will combine mechanism-guided engineering, multi-species screening, human-relevant models, multi-omics, NHP validation, and rigorous safety assessment.

For the next generation of AAV therapies, success will depend not only on finding more potent capsids, but on developing vectors that are more predictable, safer, manufacturable, and clinically relevant. Cross-species applicability is therefore not just a scientific challenge; it is a central requirement for translating AAV innovation into real therapeutic impact.

References

1. Recent advancements in improving cross-species applicability of bioengineered AAV capsids. Haolai et al., Gene Ther . 2026 Jun 27.