Adeno-Associated Virus (AAV) Vectors: Rational Design Strategies for Capsid Engineering

Abstract

Adeno-associated virus (AAV) consists of a simple genome, infects mammalian cells, displays nonpathogenicity in humans, and spans an array of serotypes and variants bearing distinct tissue tropisms. These attributes lend AAV tremendous promise as a gene delivery vector, further substantiated by its extensive testing in human clinical trials. Rational design approaches to capsid engineering leverage current scientific knowledge of AAV to further modulate, enhance and optimize the performance of the vectors. Capsid modification strategies include amino acid point mutations, peptide domain insertions, and chemical biology approaches. Through such efforts, insights regarding AAV capsid sequence-structure-function relationships can be learned. Developments over the last 5 years in rational design-based capsid engineering approaches will be presented and discussed.

Graphical Abstract

The Basics of AAV

Adeno-associated virus (AAV) is a member of the Parvoviridae family that primarily infects mammalian cells and is purportedly nonpathogenic in humans. First reported in 1965 as a contaminant of adenovirus, it has since been characterized as naturally replication-deficient, requiring helper viruses such as adenovirus for propagation [1].

AAV’s linear single-stranded DNA genome (~4.7kb) encodes two genes, rep and cap, flanked by inverted terminal repeats (ITRs) necessary for packaging the viral genome inside the capsid. The ITRs act as primers for second-strand synthesis and are the only elements in the genome required in cis for viral production [2]. For recombinant AAV (rAAV) vector production, the remainder of the viral genome can be removed, provided in trans on separate plasmids, and replaced with a desired transgene.

The rep gene encodes four overlapping non-structural proteins for replication, integration, and packaging. Rep78 and Rep68 bind to the ITRs and demonstrate helicase and endonuclease activity necessary for AAV genome replication [3]. Rep52 and Rep40 demonstrate 3’ to 5’ helicase activity and package viral genomes into capsids during virus production [4]. The cap gene encodes three structural proteins, VP1, VP2, and VP3, that self-assemble into a 60-mer icosahedral capsid at a ratio of approximately 1:1:10. These three proteins are transcribed from the same open reading frame and share a C-terminal domain but have different N-termini due to alternative start codons and alternative splicing [5]. cap also encodes for a non-structural protein, assembly-activating protein (AAP), in an alternate open reading frame from the VPs initially shown to be required for AAV2 capsid assembly [6]. Capsid assembly dependence on AAP is serotype-specific, as AAV4, −5, and −11 do not require AAP to assemble [7]. Wild-type AAV2 can undergo Rep-mediated site-specific integration into human chromosome 19 without helper virus, and rAAV vectors lacking rep integrate non-specifically at low frequencies [8]. rAAV achieves high levels of long-term gene expression without chromosomal integration and persists episomally in the nucleus in the form of head-to-tail concatemers [9]. rAAV episomes may be able to replicate in proliferating cells, albeit at low frequency [10].

VP1 is the largest (87 kDa), followed by VP2 (72 kDa) and VP3 (62 kDa). They share a common C-terminal domain, while VP1 and VP2 also contain longer N-terminal domains that are packaged inside the capsid but externalize in the endosome during intracellular trafficking [11]. The VP1 N-terminal domain contains a phospholipase A₂ domain for endosomal escape and nuclear localization sequences for nuclear trafficking. VP2 is nonessential for capsid formation and viral infection [12]. Techniques including X-ray crystallography and cryo-electron microscopy have been employed to visualize the capsid structures of many serotypes, revealing key domains that can be exploited in rational design strategies to modify functionality [13,14]. All AAV capsids share a core β-barrel motif. The β-strands connected by variable surface loops produce capsid surface topological variations – much of which can be found around the capsid’s three-fold spikes (a region frequently implicated in receptor binding and antibody recognition) [15,16].

Not merely an inert protein shell, the capsid dictates virus-cell receptor interactions and intracellular trafficking. Recent research suggests that VPs may contribute to second-strand synthesis and genomic transcription [17]. Twelve AAV serotypes and numerous variants from human and nonhuman primates have been identified with different serological profiles, cell surface receptor usage, and tissue tropisms [18]. Serotypes can be loosely categorized based on their primary cell surface receptor usage: AAV2, −3, and −6 bind heparan sulfate proteoglycan (HSPG), AAV1, −4, −5 and −6 bind sialic acid, and AAV9 binds galactose [16]. Co-receptors for internalization also vary and include laminin receptor, epidermal growth factor receptor, hepatocyte growth factor receptor, platelet-derived growth factor receptor, and several integrins (reviewed in [19]). A novel receptor AAVR recently identified appears to be required for infection for some variants [20]. Most serotypes depend on AAVR for successful cell internalization, but bind to and interact with it differently [21]. Exceptions include AAV4 and the chimeric variant AAVrh32.33, which use an AAVR-independent pathway [22].

AAV as a Gene Therapy Vector

rAAV was first produced in the early 1980’s; rAAV containing an antibiotic resistance gene in place of cap successfully transduced mammalian cells, establishing AAV’s potential as a gene delivery vector [23,24]. The first FDA approval for gene therapy treatment of a hereditary disease was granted in December 2017 for an AAV2-based product for RPE65-mediated inherited retinal dystrophy.

rAAV possesses several key features that make it highly promising for gene therapy. Its genome and capsid structure are relatively simple, and the ITRs are the only cis-acting elements essential for packaging transgenes into the capsid [2]. Additionally, AAV is nonpathogenic and demonstrates relatively low levels of immunogenicity and genotoxicity [25,26]. Serotypes exhibit a diverse range of tropisms and immune response profiles desirable for different applications. AAV8 is preferential for targeting the liver, whereas cardiac and skeletal muscle gene transfer appears mediated best by AAV1, −6, and −9 [19]. AAV9 and AAVrh.10 have demonstrated the ability to cross the blood-brain barrier (BBB) when injected intravenously [27]. Remarkably, despite AAV predominantly persisting episomally, transgene expression can be detected as long as 10 years post-AAV injection [28].

Rational Design Strategies for AAV Capsid Engineering

Despite AAV’s successes as a gene delivery vector, achieving greater control and predictability of function remains a non-trivial task. Fortunately, illuminating studies on AAV structure and biology continue to uncover new insights. Rational design strategies draw from this ever-expanding body of AAV knowledge as a framework for harnessing virus behavior. Three prominent rational capsid engineering strategies employed over the last 5 years are presented below.

Genetic mutation of AAV parts

Several studies have investigated the role of specific capsid amino acid residues in AAV’s functionality. Specifically, the efficiency and specificity of AAV gene delivery can be improved using point mutations on the viral capsid. For example, it has been postulated that undesirable post-translational modification leads to capsid degradation [29]. To address this problem, various serine, threonine and lysine residues in the AAV2 capsid were mutated to alanine or arginine [30]. The majority of these substitutions lead to enhanced transduction efficiency in HeLa cells, as well as greater gene expression in the livers of mice. The triple mutant, Y444F/Y500F/Y730F, is a promising AAV2 vector in the field [31], although similar mutations in other AAV serotype capsids do not enhance gene delivery efficiency. In a different study that addresses gene delivery specificity, an array of naturally occurring AAV variants from non-human primate tissues was isolated and capsid alignment of new isolates to currently available variants identified several residues of interest [32]. Based on this information, a new variant, AAV9.HR, was generated from the parental AAV9 by changing only two residues, H527Y and R533S. This vector transduces cells in the central nervous system (CNS), although not as robustly as AAV9. However, AAV9.HR has increased specificity since its transgene expression in peripheral tissues is reduced. In a murine model of Canavan disease, AAV9.HR-mediated delivery of the human ASPA gene successfully improves motor function. Thus, results demonstrate key point mutations on the AAV capsid can alter gene delivery efficiency and specificity.

Point mutations to the AAV capsid can also be used to mitigate recognition by host antibodies. Using cryo-electron microscopy reconstruction, site-directed mutagenesis of candidate residues, and cellular assays, Bennett et al. identified residue K531 as the contributor to AAV6 recognition by ADK6, a monoclonal antibody [33]. Mutation of K531, therefore, has the potential to impart immune evasion properties to AAV6. In a separate study, an AAV6 mutant, AAV6.2FF, was generated by introducing three point mutations (F129L, Y445F, and Y731F) to the capsid [34]. This mutant exhibits enhanced transduction efficiency in vitro relative to AAV6. Moreover, it is also more resistant to neutralization by intravenous antibodies. Although AAV6.2FF accumulates more rapidly in the lungs and muscles of mice, long-term expression levels do not reveal significant differences with AAV6. Point mutations to the AAV capsid, therefore, can be a useful strategy for preventing vector neutralization by preexisting antibodies in the host.

In addition to point mutations, larger peptide domains from one AAV serotype can be transferred to another serotype to impart new functions. For example, the ‘receptor binding footprint’ of the AAV9 capsid was incorporated into AAV2, which imparted the latter with galactose (Gal) binding properties of AAV9 [35]. The two resulting chimeras, AAV2G9 and AAV2i8G9, effectively bind to both HSPG and Gal receptors for cell entry. Moreover, the latter vector also exhibits liver de-targeting akin to its parental strain AAV2i8. This work revealed that grafting the Gal receptor recognition domain onto the AAV2 capsid does not require substantial sequence alteration and invites further investigation into extending this design approach using receptor binding domains from other AAV serotypes.

More recently, directed evolution was used in combination with rational design to develop AAV variants that could traverse the BBB with greater efficiency and specificity. DNA shuffling was used to generate capsid chimeras between AAV1 and AAVrh.10, which were then selected in vivo for their ability to cross the BBB. Structural analysis of one successful candidate identified three AAVrh.10 domains that may contribute to this property. Further studies reduced the functional domains down to eight key amino acid residues, and this minimal AAVrh.10 ‘BBB traversing footprint’ was grafted onto AAV1. The resulting vector AAV1RX not only transduces cells in the CNS readily, but also demonstrates improved specificity as evidenced by diminished transduction in the liver and vasculature [36].

In order to facilitate the rational design of new AAV capsid chimeras with functional domains of one serotype transplanted into another, the SCHEMA algorithm can be used to calculate the extent of structural disruption during chimeragenesis [37]. Using results from the algorithm as a guide, a small panel of AAV chimeras between AAV2 and AAV4 were generated. Experimental validation revealed that SCHEMA could be a useful tool for AAV capsid design, specifically in assessing capsid intactness and transduction efficiency. In sum, larger peptide domains from one AAV variant can be incorporated into another variant to rationally design new AAV mutants.

Insertion of nonviral parts into AAV capsid

A second rational design approach is to introduce functional domains nonviral in nature into the AAV capsid to elicit desired functions. For example, hexahistidine (His)-tagged designed ankyrin repeat proteins (DARPin) specific for Her2, CD4, and EpCAM have been inserted into the VP2 subunit of AAV2 [38]. Enrichment of DARPin-expressing viral particles by immobilized metal ion affinity chromatography eliminates off-target vector delivery, suggesting that subpopulations of capsids deficient in DARPin moieties lead to off-target transgene expression. The modified vectors demonstrated efficacy in in vitro and in vivo experiments: DARPin (anti-Her2)-AAV carrying a transgene that disrupts DNA replication allows for a temporary halt in breast tumor growth; DARPin (anti-CD4)-AAV selectively transduces target cells both in vitro and in vivo; and DARPin (anti-EpCAM)-AAV can discriminate between tumor cells and blood cells in whole blood samples, transducing only the former.

Nonviral parts can also be inserted into the AAV capsid to render them stimulus-responsive [39,40]. For example, small ‘peptide locks’ consisting of tetra-aspartic acid residues flanked by various protease cleavage sequences have been inserted in close proximity to the HSPG binding domain of AAV2 [41]. The peptide locks prevent the vector from transducing cells until they are cleaved off the capsid by extracellular proteases, such as matrix metalloproteinases (MMPs). Peptide locks with other amino acid compositions have also been tested, and results suggest the locks function primarily via steric obstruction of capsid-receptor binding interactions [42]. The protease-activatable vectors can perform Boolean AND gate logic, requiring detection of two different MMPs to transduce cells. The transduction efficiency of the AAV protease-activatable vectors can be improved by combining different ratios of wild-type and protease-activatable subunits [43]. Transduction efficiency increases with incorporation of more wild-type subunits; however, higher levels of non-specific transduction are also observed.

While endogenous stimuli can prompt viruses to respond accordingly within their microenvironment, external regulation may facilitate more temporal and spatial control of transgene delivery and expression. For instance, AAV transduction may be controlled by an externally applied chemical stimulus [44]. An AAV2 vector was developed with its natural cell receptor binding ability ablated and replaced with human FK-binding protein (FKBP). When supplied with a fusion protein containing an FKBP-rapamycin binding (FRB) domain attached to a DARPin moiety (targeting the human epidermal growth factor receptor - EGFR) as well as the small molecule rapamycin analog, the small molecule induces binding of FKBP to FRB, resulting in the mutant AAV vector able to transduce cells overexpressing EGFR.

In addition to chemical stimuli, externally applied light can be used to control AAV transduction. A light-activatable platform based on the heterodimerization of Phytochrome B (PhyB) and Phytochrome Interacting Factor (PIF) has been developed [45]. Upon delivering AAV2 displaying PIF on its capsid surface, a PhyB-NLS plasmid, and the chromophore phycocyanobilin to mammalian cells, efficiency of virus nuclear translocation can be modulated using different ratios of red to far red light. Moreover, cells exposed to increasing red light intensities through a photomask can exhibit higher gene expression in a spatially controlled fashion [46].

Lastly, nonviral motifs can be inserted into the AAV capsid to bring about new functional outputs. Recently, a panel of mosaic AAV capsids with varying lengths of VP2 truncation mutant subunits with a His tag at the N-terminus have been generated [47]. By harnessing the AAV capsid’s natural mechanism of activatable peptide display, the resulting virus particles exhibit varying degrees of His tag exposure pre- and post- temperature activation. Among the elucidated design principles, capsid mosaicism appears to be a requirement for robust activatable peptide display, with incorporation of fewer mutant subunits improving activatability. The length of the truncation subunits does not impart any significant functional effects. In sum, motifs from nonviral sources can be incorporated into the viral capsid in order to dramatically expand the functionality of AAV vectors.

Chemical biology approaches for AAV capsid modification

The third rational design approach involves using chemical biology strategies to make more precise modifications to the capsid. For example, an aldehyde tag was inserted into all three VP subunits of the AAV2 capsid, to which various types of molecules could be attached [48]. Despite overall low transduction efficiency, conjugating cyclic RGD peptides to the modified AAV improves transduction of HeLa cells compared to controls lacking the functionalized peptides.

In another study, the non-canonical amino acid AzK was genetically incorporated at five different surface-exposed regions on the AAV2 capsid [49]. A synthetic peptide targeting α_vβ₃ integrin receptors was then chemically conjugated to the AzK residues of two AAV variants, T454AzK and R588AzK. The latter mutant demonstrates effective vector retargeting and transduction of high α_vβ₃-expressing ovarian cancer cells.

More recently, a small tetracysteine motif was introduced into AAV9 for subsequent chemical attachment of a maleimide dye [50]. Fluorophore labeling does not disrupt the virus’s ability to pass through the BBB in mice, and the fluorescent vectors can be tracked in real-time using intravital microscopy. Insights on the capsid interactome were gleaned from studies employing a maleimide-biotin AAV9 variant in human embryonic kidney (HEK) cells, namely that transduction decreases in the absence of αVβ6 integrin but increases with lower levels of histone deacetylase 4. In sum, chemical biology methods to capsid modification allow for site-specific attachment of moieties, such as targeting ligands and fluorophores, to the vector.

Conclusion

Rational design strategies for AAV capsid engineering have yielded numerous vectors with enhanced functionalities. They rely on fundamental insights derived from the continuous discovery of naturally occurring virus variants, structural characterization, predictive modeling, and mechanistic studies. In consideration of clinical translation, areas that have been explored but still require further progress include improving AAV transduction efficiency, targeting specificity, and minimizing recognition by the host immune system. While mutation of various capsid residues may impart desirable characteristics, such as enhanced transduction or specific tissue de-targeting, the precise mechanisms behind these outcomes often remain poorly understood. Therefore, additional comprehensive experiments are needed to further our understanding of capsid sequence-structure-function relationships. When inserting exogenous motifs into the AAV capsid, it is sometimes difficult to minimize their effects on capsid assembly and hence vector production. If the vectors are to progress towards scale-up and clinical testing, improved designs must be investigated to lessen any adverse impacts of motif insertion on vector titers. As new AAV capsid variants continue to be developed and studied, key design principles will be discovered which will inform future capsid improvement strategies.

Highlights.

• Rational design of AAV relies on virus sequence, structure, and function knowledge.

• Strategies include point mutations, motif insertions, and chemical biology methods.

• Goals are to improve AAV transduction efficiency, specificity, and immune evasion.