AAV Capsid Engineering: From Random Diversification to Precision Receptor-Guided Targeting

Abstract

Adeno-associated virus (AAV) capsid engineering has evolved from random diversification and directed evolution toward rational and computational approaches designed to improve transduction efficiency, tissue tropism, and cellular specificity. Although these strategies have generated numerous engineered capsids, their biological performance often depends on incompletely understood receptor interactions and may not translate across species. Receptor-guided engineering offers a complementary strategy by directing capsid design toward defined cellular targets. Here, we review major advances in AAV capsid engineering and introduce Precision AAV Capsid Engineering (PACE™), a receptor-first framework integrating target-cell and receptor identification, selection of characterized binding ligands, capsid incorporation, vector production optimization, and functional validation. As an illustrative application, we describe CD3scFv-modified AAV6 capsids engineered to selectively transduce human T cells. Incorporation of a CD3-binding single-chain variable fragment into surface-exposed capsid loops markedly enhanced primary human T-cell transduction compared with unmodified AAV6. In humanized mouse models of CD19-positive lymphoma, systemic administration of CD3-targeted AAV encoding a CD19-directed chimeric antigen receptor generated CAR-expressing T cells and demonstrated significant antitumor activity. These findings illustrate the potential of receptor-first capsid engineering for in vivo CAR-T generation. We further discuss the challenges of receptor specificity, immunogenicity, biodistribution, manufacturing, and clinical translation. Integrating defined receptor biology with capsid engineering and functional validation may facilitate the development of more selective AAV vectors for precision gene delivery.

1. Introduction

Adeno-associated virus (AAV) vectors have emerged as one of the leading platforms for in vivo gene therapy, owing to their ability to transduce dividing and non-dividing cells, support sustained transgene expression in multiple tissues, and exhibit a generally favorable safety profile. The clinical development and regulatory approval of several AAV-based gene therapies have established the therapeutic potential of this vector system for inherited and acquired diseases. ¹^,² Nevertheless, the broader application of AAV-mediated gene delivery remains constrained by several fundamental limitations, including inefficient transduction of certain therapeutically relevant cell populations, inadequate tissue and cell-type specificity, pre-existing anti-capsid immunity, and dose-dependent toxicities associated with systemic vector administration. ³^,⁴

The biological properties of AAV vectors are strongly influenced by their capsids, which determine interactions with cellular attachment factors and receptors, tissue distribution, cellular internalization, intracellular trafficking, and susceptibility to neutralizing antibodies. Naturally occurring AAV serotypes exhibit distinct but frequently overlapping tissue tropisms, providing useful starting materials for vector development. However, their natural tropisms do not necessarily correspond to the cellular targets required for specific therapeutic applications. Following systemic administration, substantial vector accumulation in non-target organs, particularly the liver for many AAV serotypes, may limit delivery efficiency and contribute to unwanted vector exposure. Consequently, the development of engineered capsids capable of achieving efficient and selective gene delivery has become a major focus of AAV vector research. ⁵^,⁶^,⁷

Over the past two decades, numerous capsid engineering strategies have been developed to overcome the limitations of naturally occurring AAV serotypes. Early approaches relied extensively on random sequence diversification, including peptide insertion libraries, error-prone PCR mutagenesis, and DNA shuffling among different AAV serotypes. These methods generate large populations of capsid variants that can subsequently be screened or selected for desirable properties, such as enhanced cellular transduction, altered tissue tropism, or resistance to neutralizing antibodies. When coupled with iterative functional selection, these diversification strategies form the basis of directed evolution, which has produced numerous engineered AAV variants with improved biological performance. ⁸^,⁹^,¹⁰ However, because candidate capsids are commonly identified through phenotype-driven selection, the molecular mechanisms responsible for their enhanced transduction or altered tropism may remain incompletely understood.

Advances in AAV structural biology, receptor identification, high-throughput sequencing, and computational protein design have enabled increasingly rational approaches to capsid engineering. Structural characterization of capsid surfaces and their interactions with cellular receptors provides opportunities to modify specific amino acid residues or surface-exposed regions to alter vector properties. The identification of AAV receptor (AAVR) as an important cellular entry factor for multiple AAV serotypes further demonstrated the central role of receptor biology in productive AAV transduction. ¹¹ More recently, machine learning and artificial intelligence have been integrated with experimental screening to predict sequence–function relationships and prioritize capsid variants with desirable properties. ¹² These developments are progressively transforming capsid engineering from predominantly empirical discovery toward more hypothesis-driven and predictive design.

2. Structural and Biological Foundations

The adeno-associated virus (AAV) capsid is a non-enveloped, approximately 25-nm protein shell composed of 60 viral protein (VP) subunits arranged with T = 1 icosahedral symmetry. Three overlapping capsid proteins, VP1, VP2, and VP3, are encoded by the cap gene and incorporated into mature virions at an approximate stoichiometric ratio of 1:1:10, although their relative abundance varies among serotypes and production conditions. ¹³^,¹⁴ VP3 constitutes the principal structural framework, while VP1 and VP2 contain additional N-terminal sequences associated with intracellular trafficking and infectivity.

The three capsid proteins share a common C-terminal sequence corresponding to VP3. VP1 contains a unique N-terminal region (VP1u) harboring a phospholipase A2 (PLA2) domain involved in endosomal escape. ¹⁵^,¹⁶ VP2 contains an N-terminal extension relative to VP3 but lacks the VP1-specific PLA2 domain. The common VP3 region adopts a conserved eight-stranded antiparallel β-barrel, or jelly-roll fold, characteristic of parvoviral capsids. Surface-exposed loops extending from this structural core form prominent features surrounding the twofold, threefold, and fivefold symmetry axes. ¹⁷^,¹⁸

Comparative structural analyses have identified nine major variable regions (VR-I through VR-IX), located predominantly within surface-exposed loops. These regions contribute to differences in receptor recognition, tissue tropism, and antigenicity among naturally occurring AAV serotypes. ¹⁷^,¹⁸

Natural AAV serotypes recognize distinct cellular attachment factors and entry-associated receptors. AAV2 binds heparan sulfate proteoglycans (HSPGs), whereas other serotypes interact with sialylated glycans or terminal galactose residues. ¹⁹^,²⁰ These interactions influence cellular attachment and susceptibility but do not necessarily determine productive transduction.

The identification of AAV receptor (AAVR), encoded by KIAA0319L, established an important role for protein-mediated interactions in AAV entry. ¹¹^,²¹ AAVR contains extracellular polycystic kidney disease (PKD) domains that interact with multiple AAV serotypes. Structural studies have demonstrated differences in capsid–AAVR binding interfaces and domain usage, contributing to the diversity of AAV entry mechanisms.

AAV transduction involves sequential processes of cell-surface attachment, internalization, intracellular trafficking, endosomal escape, nuclear entry, genome uncoating, and transgene expression. Following receptor engagement, AAV particles undergo endocytic uptake and intracellular trafficking, during which endosomal acidification promotes conformational changes that facilitate exposure of the VP1u PLA2 domain. ²² This activity contributes to endosomal escape, allowing subsequent steps required for nuclear delivery and transgene expression.

Importantly, cellular attachment, vector internalization, and productive transduction represent distinct biological events. Efficient receptor binding or cellular uptake does not necessarily result in successful intracellular delivery or transgene expression. This distinction is fundamental to understanding the biological tropism of AAV vectors.

3. Empirical Diversification and Directed Evolution

Empirical AAV capsid engineering relies on generating sequence diversity and selecting variants with desirable biological properties. Major diversification strategies include peptide insertion and display, DNA family shuffling, and random or targeted mutagenesis. These approaches are frequently combined with directed evolution, in which libraries of capsid variants undergo functional selection to enrich candidates exhibiting improved transduction efficiency, altered tissue tropism, or resistance to neutralizing antibodies. The principal methods and representative applications are discussed below.

3.1. Peptide insertion and display

Random peptide insertion and display represent one of the earliest and most widely investigated strategies for altering the cellular tropism of adeno-associated virus (AAV) vectors. This approach involves genetically incorporating short, randomized peptide sequences into surface-exposed regions of the AAV capsid, thereby generating libraries of capsid variants with potentially novel receptor-binding properties. Unlike rational engineering approaches that begin with a predefined cellular receptor or ligand, random peptide display relies on functional screening or selection to identify capsids capable of efficiently interacting with and transducing the desired cell population.

The success of random peptide display depends strongly on the selection of capsid regions capable of accommodating foreign sequences while maintaining particle assembly, genome packaging, and infectivity. Structural studies have identified multiple surface-exposed variable regions within the AAV capsid, particularly loops surrounding the threefold symmetry axes, that may tolerate sequence modifications. ²³^,²⁴^,²⁵

Random peptide-display libraries are commonly generated by introducing degenerate nucleotide sequences into defined positions of the AAV cap gene. These randomized sequences encode short peptides, frequently containing approximately seven amino acids, although other lengths and configurations have been investigated. The resulting plasmid library provides a genetically diverse population of capsid variants that can be packaged into AAV particles for subsequent screening.

Random peptide-display libraries can be screened or selected using cultured cells, organotypic systems, or animal models, depending on the desired biological phenotype. In vitro selection typically involves exposing target cells to a capsid library, recovering vector-associated genetic sequences from the selected population, and identifying enriched variants. Where the library system supports biological amplification, selected variants may undergo additional rounds of enrichment.

A prominent subsequent application was the development of AAV2-7m8 for retinal gene delivery. Dalkara et al. used in vivo-directed evolution to identify an AAV2-derived capsid variant containing a peptide insertion that improved gene delivery to the outer retina following intravitreal administration. ²⁶ This approach addressed a major limitation of conventional retinal AAV delivery, in which efficient photoreceptor transduction frequently requires subretinal administration. The AAV2-7m8 variant demonstrated that peptide insertion combined with selection in an intact tissue environment could generate capsids capable of overcoming anatomical barriers to gene delivery. Later studies investigated the mechanisms underlying its enhanced retinal transduction and demonstrated that the functional consequences of peptide insertion extend beyond simple changes in receptor attachment. ²⁷

While AAV2-7m8 demonstrated the potential of peptide insertion to overcome anatomical barriers in retinal gene delivery, similar capsid engineering strategies have also been applied to central nervous system delivery. However, studies of AAV-PHP.B and related variants revealed that their enhanced central nervous system tropism is strongly dependent on the mouse genetic background. Mechanistic investigations identified the glycosylphosphatidylinositol-anchored protein LY6A as an important receptor mediating enhanced central nervous system delivery in permissive mouse strains. ²⁸^,²⁹ This discovery explained important aspects of the strain-dependent phenotype and highlighted the limitations of extrapolating results from selected animal models to human applications.

Random peptide insertion offers several important advantages for AAV capsid engineering. First, it enables the discovery of novel receptor-binding interactions without requiring prior identification of the target-cell receptor. This feature is particularly valuable when the molecular determinants of cellular susceptibility are incompletely understood. Second, large combinatorial libraries permit exploration of peptide sequence space beyond the range that could be efficiently investigated through individual rational designs. Third, screening peptides directly in the context of assembled AAV particles allows selection to account for aspects of capsid compatibility that may not be captured by isolated ligand-binding assays.

3.2. DNA family shuffling and recombination

DNA family shuffling is a combinatorial engineering strategy that generates novel adeno-associated virus (AAV) capsids through recombination of homologous cap gene sequences derived from different parental serotypes or naturally occurring isolates. ¹⁰^,³⁰ Unlike random peptide insertion, which introduces foreign sequences into defined capsid regions, DNA shuffling exploits the naturally evolved sequence diversity of existing AAV capsids to generate chimeric proteins containing combinations of parental sequence elements. The underlying principle is that favorable biological properties associated with different parental capsids may be combined through recombination, potentially producing variants with improved transduction efficiency, altered tissue tropism, or reduced susceptibility to neutralizing antibodies.

Naturally occurring AAV serotypes exhibit substantial differences in receptor recognition, tissue tropism, intracellular trafficking, and susceptibility to neutralizing antibodies. These functional differences arise from sequence variations distributed throughout the capsid, particularly within surface-exposed variable regions. Recombination among parental serotypes therefore provides an opportunity to generate novel capsids that combine selected structural and functional characteristics.

Domain swapping and variable-region exchange represent more targeted forms of capsid recombination. Rather than generating numerous random recombination junctions throughout the cap gene, these approaches replace defined capsid segments with corresponding sequences from another serotype. Such modifications may involve surface-exposed loops, receptor-binding regions, or larger structural domains associated with particular biological properties.

The generation of a diverse shuffled capsid library does not ensure that all resulting variants are biologically functional. Recombination may disrupt protein folding, capsid assembly, genome packaging, or essential intracellular functions. Accordingly, functional selection is required to identify recombinant capsids capable of maintaining the properties necessary for gene delivery.

One of the best-characterized products of AAV DNA family shuffling is AAV-DJ, developed by Grimm et al. through recombination of multiple parental AAV serotypes followed by functional selection. ¹⁰ The selected capsid contained sequence contributions from AAV2, AAV8, and AAV9 and differed from AAV2 at approximately 60 amino acid positions. AAV-DJ demonstrated efficient transduction of cultured cells and improved liver transduction relative to AAV2 in the experimental models evaluated. The study also reported altered susceptibility to neutralization by pooled human immunoglobulins, demonstrating that DNA shuffling could simultaneously influence transduction and antigenic properties.

Importantly, AAV-DJ retained heparin-binding activity, which influenced its biodistribution and biological performance. This observation illustrates that capsid recombination does not necessarily eliminate parental receptor interactions. Instead, the final phenotype reflects the combined effects of retained and newly configured structural features.

Together, AAV-DJ developed by Grimm et al., the chimeric AAV variants generated by Koerber et al., and AAV-LK03 developed by Lisowski et al. illustrate the versatility of capsid DNA shuffling. ¹⁰^,³⁰^,³¹ These studies demonstrated that recombination of parental capsid sequences, followed by functional selection, can generate vectors with altered antibody recognition, enhanced transduction efficiency, and species-dependent cellular tropism. Importantly, the distinct properties of these engineered capsids highlight the influence of selection conditions and host-specific molecular interactions on vector performance, emphasizing the need to evaluate engineered AAV variants in biologically relevant experimental models.

DNA family shuffling provides a powerful strategy for AAV capsid engineering by exploiting naturally evolved sequence diversity to generate novel combinations of capsid sequences. Unlike unrestricted random mutagenesis, this approach recombines sequence elements derived from functional parental capsids, potentially preserving structural features important for capsid assembly and infectivity. However, recombination does not guarantee functional compatibility between sequence segments, and many resulting variants may exhibit impaired assembly, infectivity, or transduction efficiency. Consequently, the success of DNA family shuffling depends on the generation of sufficiently diverse libraries and the application of appropriate selection strategies to identify variants with desirable biological properties.

3.3. Mutagenesis and sequence–function landscapes

Mutagenesis is a fundamental strategy for exploring sequence–function relationships in adeno-associated virus (AAV) capsids. Unlike DNA family shuffling, which recombines sequences from parental serotypes, random mutagenesis introduces new substitutions into a selected capsid gene. Error-prone PCR is commonly used to generate diverse mutant libraries. Early studies by Maheshri et al. and Perabo et al. demonstrated that random mutagenesis could produce AAV variants with altered receptor interactions and susceptibility to antibody neutralization. ⁸^,⁹

Saturation mutagenesis provides a more systematic approach by introducing multiple amino acid substitutions at selected capsid positions. Single-site saturation mutagenesis can theoretically evaluate all 19 alternative amino acids at a given residue, revealing its functional tolerance. Targeted combinatorial mutagenesis extends this approach by introducing substitutions at multiple positions simultaneously, enabling investigation of interactions between mutations, known as epistasis.

Deep mutational scanning (DMS) combines large-scale sequence diversification with high-throughput functional measurements and next-generation sequencing to map relationships between capsid sequence and biological performance. This approach enables systematic identification of functionally constrained and mutation-tolerant residues, providing valuable insights into capsid assembly, genome packaging, and transduction. The resulting sequence–function datasets can also serve as a foundation for subsequent computational engineering, as discussed in Section 4.2.

The principal strength of mutagenesis-based approaches is their ability to identify functionally important residues without requiring complete prior knowledge of capsid structure or molecular mechanisms. Systematic mutagenesis and DMS further provide empirical sequence–function datasets that can guide subsequent capsid engineering. However, the effects of individual mutations are often context-dependent, and combinations of substitutions may produce unpredictable outcomes. These limitations highlight the importance of experimentally validating engineered capsids.

Despite differences in library construction, empirical capsid engineering strategies commonly involve library generation, AAV vector production, functional screening or selection in target cells or animal models, and characterization of enriched variants. Directed evolution may require multiple rounds of vector production, selection, and optimization to identify capsids with desirable biological properties. The complete development process can be labor-intensive and time-consuming, potentially extending from months to years when extensive in vivo selection and validation are required. Moreover, improved transduction efficiency does not necessarily indicate enhanced cellular specificity, and capsid performance may vary across experimental systems and species. These limitations have encouraged the development of rational and computational approaches that incorporate structural information, receptor biology, and sequence–function relationships into capsid design.

4. Rational and Computational Capsid Engineering

4.1. Structure-guided design

Rational and structure-guided engineering provides a complementary approach to empirical diversification and directed evolution for modifying adeno-associated virus (AAV) capsids. By exploiting knowledge of capsid architecture, receptor-binding sites, and antigenic epitopes, this strategy enables targeted modifications designed to alter vector tropism, transduction efficiency, or antibody recognition. Early studies demonstrated that structure-informed modifications of capsid surface residues could alter receptor interactions and neutralization profiles. ³²^,³³ Subsequent investigations further established that receptor-binding footprints could be modified or transferred between AAV serotypes to generate vectors with distinct biological properties. These approaches can narrow the experimental search space and facilitate mechanistic interpretation, although the effects of individual modifications remain dependent on capsid context and require experimental validation.

Rational engineering strategies include substitution of defined amino acid residues, modification of native receptor-binding interfaces, alteration of surface charge, introduction of targeting ligands, and engineering of regions involved in intracellular trafficking. These approaches can be guided by structural comparisons among AAV serotypes, mutational analysis, receptor-binding studies, or experimentally determined structures of capsid complexes.

Structural characterization of AAV interactions with cellular attachment factors and entry receptors has provided important mechanistic insights for capsid engineering. The identification of receptor-binding interfaces enables investigators to distinguish residues directly involved in ligand recognition from neighboring structural elements that influence receptor accessibility or capsid conformation.

One of the most direct applications of rational capsid engineering is the modification of known receptor-binding regions to alternative tropism or reduce undesirable cellular interactions. Such modifications may involve substitution of residues that mediate attachment to cellular glycans, disruption of existing receptor-binding motifs, or introduction of alternative binding determinants.

Preexisting immunity to naturally occurring AAV serotypes represents a major challenge for systemic gene delivery. Neutralizing antibodies can interfere with vector attachment, cellular entry, or other steps required for productive transduction, thereby reducing therapeutic efficacy. Antibody recognition can also complicate vector readministration, particularly when the initial administration induces a strong humoral immune response.

Structural characterization of antibody-binding epitopes has enabled rational approaches to modifying AAV capsid antigenicity. Cryo-EM and related structural techniques can identify antibody footprints on the capsid surface, while mutational analysis can determine which residues contribute to neutralization. These data provide a foundation for designing substitutions that reduce antibody recognition while preserving essential capsid functions.

Although receptor recognition is an important determinant of AAV transduction, intracellular trafficking and post-entry processing frequently limit the efficiency of gene delivery. Following internalization, AAV vectors must avoid excessive degradation, escape from endosomal compartments, reach the nucleus, and release their genomes in a form capable of supporting transgene expression.

Rational engineering can also target capsid features involved in intracellular trafficking. Zhong et al. demonstrated that substitution of selected surface-exposed tyrosine residues with phenylalanine reduced AAV2 capsid ubiquitination and enhanced transduction efficiency, consistent with improved intracellular trafficking. ³⁴ This study illustrates how mechanistic insights can guide specific capsid modifications to improve vector performance.

An important strength of rational design is its mechanistic interpretability. Modifications targeting defined receptor-binding interfaces, antigenic epitopes, or intracellular trafficking determinants can be evaluated within a specific molecular framework, facilitating subsequent optimization. Structure-guided engineering may also integrate multiple design objectives, such as reducing native receptor interactions, introducing targeting ligands, and preserving capsid assembly and intracellular trafficking functions. These capabilities are particularly relevant to applications requiring precise cellular targeting.

Nevertheless, rational engineering remains constrained by incomplete biological knowledge. Structural information alone may not fully explain the dynamic processes of cellular entry and intracellular trafficking, while interactions among capsid residues can produce unpredictable, context-dependent effects. Consequently, rational design often requires experimental validation and may benefit from integration with empirical diversification and functional selection.

4.2. Computational and AI-assisted engineering

Computational and artificial intelligence (AI)-assisted approaches have emerged as complementary strategies for AAV capsid engineering, enabling the analysis of large sequence–function datasets and the identification of promising variants beyond those readily accessible through conventional experimental screening. Machine learning (ML) models can identify relationships between capsid sequences and experimentally measured properties, including assembly, genome packaging, and transduction efficiency, without requiring complete knowledge of the underlying molecular mechanisms.

Early applications demonstrated the feasibility of data-driven AAV capsid engineering. Ogden et al. combined comprehensive AAV2 mutational fitness measurements with computational modeling to guide the identification of viable capsid sequences. ³⁵ Building on this concept, Bryant et al. employed machine learning to design highly diverse AAV capsid libraries enriched for functional variants. ¹² Together, these studies established that experimentally derived sequence–function relationships can support computational prediction and guide the design of novel capsid variants.

More recently, generative models, including protein language models and neural network-based approaches, have expanded the potential for designing novel capsid sequences. Rather than evaluating only existing variants, these methods can propose new amino acid combinations while incorporating learned sequence constraints. However, computationally predicted viability does not necessarily establish efficient genome packaging, productive transduction, or desirable tissue tropism.

A major challenge is that capsid performance depends on multiple biological processes, including receptor recognition, intracellular trafficking, and tissue-specific barriers. Models trained on capsid production or cell-culture transduction data may therefore have limited predictive value for systemic delivery in humans. Furthermore, optimizing multiple properties simultaneously, such as production efficiency, cellular specificity, and antibody resistance, remains difficult.

Integrating computational predictions with experimental validation through iterative design–build–test–learn cycles offers a promising strategy for improving AAV capsid engineering. The effectiveness of these approaches ultimately depends on the quality, diversity, and biological relevance of the training data.

5. Receptor-Guided Targeting and Precision AAV Capsid Engineering

5.1. Established receptor-mediated targeting strategies

Receptor-mediated targeting provides a strategy for redirecting adeno-associated virus (AAV) vectors toward selected cell populations. Natural AAV serotypes recognize distinct attachment factors, including heparan sulfate, sialic acid, and galactose, while AAVR contributes to productive entry by multiple serotypes. Understanding these interactions has enabled the development of engineered capsids with altered receptor specificity.

Peptide display was among the earliest AAV retargeting strategies. Girod et al. demonstrated that insertion of a receptor-binding peptide into the AAV2 capsid could redirect transduction toward cells expressing the corresponding integrin receptor. ²⁵ Subsequent peptide-display libraries expanded the range of targeting motifs and enabled selection of capsids with altered cellular tropism. ²³^,²⁴

Antibody-derived targeting domains, including single-chain variable fragments (scFvs) and nanobodies, offer greater molecular specificity through recognition of selected cell-surface antigens. Although their incorporation can be constrained by protein folding and capsid assembly, successful display of structured binding domains has been demonstrated. ³⁶

Designed ankyrin repeat proteins (DARPins) provide an alternative class of targeting ligands. Münch et al. demonstrated selective transduction of HER2/neu-, EpCAM-, and CD4-expressing cells using DARPin-displaying AAV vectors. ³⁷ Theuerkauf et al. subsequently developed bispecific DARPin-displaying capsids that preferentially transduced cells coexpressing CD4 and CD32a, demonstrating the feasibility of dual-receptor targeting. ³⁸

Adaptor-mediated and modular conjugation strategies provide alternatives to direct genetic incorporation of large targeting ligands. Bartlett et al. demonstrated that a bispecific antibody fragment redirected AAV2 transduction toward megakaryocytic cells. ³⁹ More recently, SpyTag/SpyCatcher-based approaches have enabled covalent attachment of receptor-binding proteins to assembled AAV capsids. Armbruster et al. demonstrated selective targeting of EGFR-, EpCAM-, and HER2-expressing cells using SpyCatcher-fused DARPins, while related approaches have supported nanobody- and scFv-mediated AAV retargeting. ⁴⁰^,⁴¹ These modular systems permit ligand exchange without requiring genetic incorporation of the complete targeting domain into the capsid.

Across these strategies, successful targeting depends on ligand accessibility, orientation, density, and compatibility with capsid assembly. Importantly, receptor binding does not guarantee productive transduction, which additionally requires cellular uptake, intracellular trafficking, and nuclear delivery. Furthermore, receptor expression in nontarget tissues and species-dependent differences remain important challenges for translating receptor-targeted AAV vectors into clinical applications.

5.2. PACE: a receptor-first design framework

Conventional AAV capsid engineering frequently begins with sequence diversification followed by functional selection of variants exhibiting desirable biological properties. Although effective, such phenotype-first approaches may identify improved vectors without revealing the molecular mechanisms responsible for their activity. Precision AAV Capsid Engineering (PACE™) is proposed as a complementary receptor-first framework that begins with a defined therapeutic cell population and a selected cell-surface receptor, followed by ligand selection, capsid engineering, and functional validation.

The PACE workflow comprises five interconnected stages: (i) identification of the target cell population and a suitable cell-surface receptor; (ii) selection of a previously characterized receptor-binding ligand; (iii) incorporation of the ligand into an appropriate AAV capsid configuration; (iv) optimization of AAV production; and (v) experimental validation of functional transduction, receptor dependence, cellular specificity, and off-target activity. This receptor-first approach integrates established receptor biology with capsid engineering and functional validation, reducing reliance on unrestricted empirical screening.

Receptor and ligand selection. A suitable receptor should be enriched on the intended cellular population, accessible to administered vectors, and capable of supporting productive intracellular delivery. Its distribution in nontarget tissues, expression variability, and internalization characteristics must also be considered. PACE prioritizes previously characterized binding ligands, including natural receptor-binding peptides, antibody-derived fragments, nanobodies, and engineered protein scaffolds. Published biochemical and structural information can facilitate ligand selection and subsequent verification of receptor specificity.

Ligand affinity alone does not determine targeting performance. High-affinity binding may enhance cellular attachment but can also increase recognition of nontarget cells expressing lower receptor levels. Moreover, receptor engagement may trigger signaling, internalization, or downregulation, potentially affecting both vector delivery and cellular physiology. These considerations are particularly important when targeting immune-cell receptors.

Capsid incorporation and optimization. Following ligand selection, the targeting domain is incorporated into a capsid configuration that supports receptor accessibility while preserving particle assembly, genome packaging, and intracellular delivery. The optimal display position depends on ligand size, folding, orientation, and structural compatibility with the parental capsid. Multivalent presentation may enhance receptor engagement but can also introduce steric constraints. Accordingly, PACE optimization emphasizes overall vector performance rather than maximal receptor affinity or ligand density alone.

Functional validation. Demonstrating receptor-dependent transduction is essential for distinguishing genuine targeting from nonspecific enhancement of vector activity. Supporting evidence may include preferential transduction of receptor-positive cells, reduced activity following receptor blockade or genetic depletion, and restoration of transduction after receptor re-expression. Importantly, receptor binding and cellular uptake must be distinguished from productive gene delivery, which requires successful intracellular trafficking and transgene expression. Evaluation of relevant nontarget cell populations is equally important for establishing cellular selectivity.

CD3-targeted AAV as an illustrative application. The PACE concept is exemplified by engineering AAV6 capsids to display a CD3-binding single-chain variable fragment (CD3scFv) for targeting human T lymphocytes ⁴². CD3 was selected because of its expression on mature T cells and the availability of characterized CD3-binding antibodies. Alternative capsid display configurations were evaluated to determine their effects on functional transduction.

Incorporation of CD3scFv into surface-exposed capsid loops markedly enhanced primary human T-cell transduction compared with unmodified AAV6. These findings suggest that interactions between capsid-displayed CD3scFv and cell-surface CD3 facilitate vector attachment and may promote subsequent internalization and productive transduction. Importantly, the effectiveness of CD3scFv display depended on its position within the capsid, highlighting the critical role of ligand presentation in determining functional performance. Differences in ligand accessibility, orientation, receptor engagement, and intracellular trafficking may contribute to these observations, although the precise mechanisms remain to be elucidated.

Because CD3 is a component of the T-cell receptor signaling complex, CD3-targeted vectors also require evaluation of potential effects on T-cell activation, viability, proliferation, and function. Collectively, this example demonstrates how defined receptor recognition, capsid configuration, and functional validation can be integrated within the receptor-first PACE framework to develop cell-targeted AAV vectors.

Broader applications and future directions. PACE may be applicable to other immune-cell subsets and therapeutically relevant tissues, provided suitable receptors and compatible targeting ligands can be identified. However, receptor accessibility, tissue barriers, species-dependent expression, and intracellular trafficking may limit generalizability. Future integration with structural modeling, directed evolution, and machine-learning-assisted optimization could improve ligand presentation and downstream transduction. Multispecific receptor recognition and complementary transcriptional or post-transcriptional detargeting strategies may further enhance cellular selectivity.

PACE should therefore be viewed as an integrated, hypothesis-driven framework rather than a deterministic method for predicting vector tropism. Its potential value lies in combining defined receptor biology with capsid engineering and rigorous experimental validation to develop more selective AAV vectors.

6. Translational Challenges and Therapeutic Applications

The clinical translation of engineered AAV capsids requires consideration of immunogenicity, biodistribution, manufacturing feasibility, and target-cell specificity. Pre-existing neutralizing antibodies can substantially reduce vector transduction, while capsid modifications may alter antibody recognition and potentially introduce additional immunogenic epitopes. ⁴³ Although antigenic engineering can reduce recognition by selected antibodies, ⁴⁴ broad immune evasion cannot be assumed, particularly when heterologous targeting ligands are incorporated into the capsid.

Biodistribution and vector dose are equally important determinants of therapeutic efficacy and safety. Systemically administered AAV vectors frequently accumulate in the liver, potentially increasing off-target exposure. High systemic doses have been associated with serious adverse events, including hepatotoxicity, complement activation, and thrombotic microangiopathy. ⁴⁵^,⁴⁶ Receptor-guided targeting may improve cellular selectivity and potentially reduce the vector dose required for therapeutic activity, but these benefits must be demonstrated experimentally. Nonclinical studies should therefore assess tissue biodistribution, dose-dependent transduction, and relevant safety endpoints.

Manufacturing represents another critical consideration. Incorporation of targeting ligands, particularly folded protein domains such as scFvs and nanobodies, may affect capsid assembly, genome packaging, stability, and vector yield. Consequently, receptor-targeted capsids must be evaluated for both functional performance and manufacturability, including product identity, purity, potency, and consistency.

In vivo CAR-T therapy represents a promising application of receptor-guided AAV engineering. Conventional CAR-T manufacturing requires ex vivo T-cell collection, genetic modification, expansion, and reinfusion. In contrast, in vivo approaches aim to generate CAR-expressing T cells directly within the patient, potentially simplifying treatment and improving accessibility. Several delivery platforms are under investigation, including lipid nanoparticles, engineered lentiviral vectors, and AAV-based systems.

As an example, CD3scFv-modified AAV6 capsids have been developed to selectively target human T cells for in vivo CAR-T generation ⁴². Incorporation of CD3scFv into surface-exposed capsid loops substantially enhanced primary human T-cell transduction compared with unmodified AAV6. In mouse models engrafted with human peripheral blood mononuclear cells and CD19-positive lymphoma cells, systemic administration of CD3-targeted AAV encoding a CD19-directed CAR resulted in CAR expression in human T cells and significant antitumor activity. Dose-de-escalation studies further demonstrated antitumor efficacy across a broad range of vector doses. These findings provide preclinical evidence supporting receptor-guided AAV delivery for in vivo CAR-T generation.

Nevertheless, several challenges remain. CD3 engagement may influence T-cell activation and function, while the predominantly episomal nature of recombinant AAV genomes raises questions about CAR expression persistence during T-cell proliferation. In addition, receptor expression and accessibility may differ across cell populations and species. Long-term CAR expression, T-cell phenotype, off-target transduction, and immunological safety therefore require further evaluation.

Beyond immune-cell targeting, engineered AAV capsids are being investigated for gene delivery to the central nervous system, retina, muscle, and liver. These applications face distinct anatomical and biological barriers, emphasizing the importance of selecting appropriate experimental models and validating receptor-dependent transduction in human-relevant systems.

Ultimately, clinical translation requires integration of receptor specificity, functional transduction, biodistribution, immunogenicity, and manufacturing quality. The development of CD3-targeted AAV vectors illustrates the therapeutic potential of precision capsid engineering while highlighting the need for rigorous preclinical and clinical evaluation.

7. Future Perspectives and Conclusions

AAV capsid engineering has evolved from empirical diversification and functional selection toward increasingly knowledge-driven and computationally assisted design. Random peptide insertion, DNA shuffling, mutagenesis, and directed evolution have generated capsids with substantially altered biological properties and remain essential tools for discovering unexpected sequence–function relationships. However, these approaches can require extensive screening, and the mechanisms responsible for improved transduction are not always immediately apparent.

The integration of receptor biology, capsid structural information, and artificial intelligence represents a promising direction for next-generation AAV engineering. Receptor-expression datasets can help identify biologically relevant cellular targets, while structural analysis can guide the selection of capsid regions suitable for ligand incorporation or modification.

Machine-learning models may further assist in predicting capsid viability, structural compatibility, and selected functional properties. When combined with experimental datasets, these methods could prioritize candidate sequences and reduce the number of variants requiring extensive laboratory evaluation.

Modular receptor-guided engineering offers an opportunity to expand AAV targeting beyond the natural tropism of existing serotypes. By separating the selection of a cellular receptor from the engineering of its corresponding binding ligand, it may become possible to develop adaptable capsid platforms for different therapeutic cell populations.

PACE™ provides a conceptual framework for this approach by beginning with a biologically defined target receptor, selecting an appropriate ligand, and engineering its presentation within a functional AAV capsid.

Achieving clinically meaningful cellular specificity remains one of the most important challenges in AAV gene delivery. Enhanced transduction of a target population does not necessarily indicate exclusive targeting, particularly when the selected receptor is also expressed in non-target tissues or when residual native capsid interactions remain functional.

Moreover, receptor binding and cellular uptake do not guarantee productive intracellular delivery. Efficient transduction requires successful completion of multiple downstream processes, including intracellular trafficking, endosomal escape, nuclear delivery, and vector genome processing.

One important direction is the identification of capsid insertion configurations that accommodate diverse receptor-binding domains while preserving particle assembly and genome packaging. Structural modeling, targeted mutagenesis, and experimental screening may help define the limits of ligand size, orientation, and display density.

A second priority is the integration of receptor-guided design with directed evolution and AI-assisted optimization. Receptor biology can define the initial targeting hypothesis, while empirical and computational methods can refine capsid compatibility, intracellular delivery, and overall functional performance.

AAV capsid engineering has progressed from random diversification and phenotype-driven selection toward increasingly rational, structure-guided, and computationally assisted approaches. Each strategy has contributed important advances, but no single method fully addresses the biological complexity of cellular targeting and productive gene delivery.

Correspondence

Haifeng Chen, Ph.D., Virovek, Inc., Houston, Texas, USA.

Declaration of Interests

H.C. is founder and an equity holder of Virovek, Inc. and AAVivo, Inc., and is involved in the development of T-cell-targeted AAV and AAV-mediated in vivo CAR-T technologies discussed in this article. H.C. is also an inventor on patent applications related to T-cell-targeted AAV technology. Published evidence and proposed mechanisms are distinguished throughout the article.