Adeno-Associated Virus Engineering and Load Strategy for Tropism Modification, Immune Evasion and Enhanced Transgene Expression

Introduction

Gene therapy refers to the introduction of exogenous normal genes into target cells to correct or compensate for diseases caused by defective and abnormal genes in order to achieve therapeutic purposes and has gained widespread attention because of its ability to treat allopathic causes.^1,2 Various genetic materials, such as DNA, mRNA, small interfering RNA (siRNA), anti-sense oligonucleotide (ASO), and clustered regularly interspaced short palindromic repeats (CRISPR)/CRISPR-associated protein 9 (CRISPR-Cas9) systems, have been used to treat diseases. However, the clinical application of gene therapy is hindered by its immunogenicity, off-target effects, and carcinogenicity. Moreover, because genetic materials are hydrophilic, negatively charged, and easily degraded by enzymes, there is a need for delivery vehicles that effectively load gene drugs and deliver them to the site of interest. Existing gene delivery vectors can be divided into viral and non-viral delivery vectors. For example, non-viral lipid nanoparticles (LNPs), which have gained interest, exhibit good safety, biocompatibility, and high delivery efficiency. In addition, LNPs have several advantages, including the transient expression of LNP-loaded gene agents, chemical synthetic components and feasible large-scale production.³ However, LNPs have the drawbacks of limited delivery to extrahepatic tissues, temporary expression of gene agents, and passive transgene expression, which limits the application of LNPs in situations where non-liver delivery and long-term expression are required. Viral vectors have broader tissue tropism, including the brain, eyes and muscles, than nonviral vectors,^4,5 and they can transduce specific cell populations by optimizing cell-specific promoters,⁵ as well as inducing long-term gene expression in vivo. Therefore, viral vectors are promising gene delivery vehicles. Viral vectors currently in use include lentiviruses, adenoviruses, and adeno-associated viruses (AAVs), among others.⁶ AAV is particularly suitable for in vivo gene delivery because of its safety, low immunogenicity, long-lasting expression, broad tropism, and ease of production.^7,8 Since AAVs are non-integrating, they can only be expressed permanently in non-dividing cells over long periods of time.

AAV belongs to the genus Dependoparvovirus within the family Parvoviridae,⁹ and its life cycle depends on the participation of replication viruses, such as adenovirus and other auxiliary viruses. AAV can be divided into wild-type AAV and recombinant AAV according to their sources. Wild-type AAV can be integrated into AAVS1 site of human chromosome 19 at a fixed point, while recombinant AAV cannot be integrated due to the lack of two genes necessary for integration and replication. Therefore, recombinant AAV mainly exists in the form of appendages free from chromosomes and can exist in non-splinter cell for a long time. Table 1 lists the characteristics of natural AAV serotypes at present. AAV is a single-stranded DNA virus with a ~4.7 kb genome that comprises four known open reading frames, rep, cap, MAAP, and AAP, flanked by two 145 bp inverted terminal repeats (ITRs).¹⁰ The rep gene encodes for Rep78, Rep68, Rep52, and Rep40, which support viral genome replication, transcriptional regulation, and virion assembly. The cap gene encodes three viral proteins, VP1, VP2, and VP3.¹¹ The 60 subunits of VP at a 1:1:10 ratio of VP1:VP2:VP3 form an AAV capsid with a size of 23–28 nm.^5,12 The common VP3 protein forms the actual capsid and displays a core eight-stranded antiparallel β-barrel motif.^13,14 A highly conserved α-helix is present between the βC and βD strands.¹⁵ The surface-exposed loops between β strands feature nine variable regions (VRI-Ⅸ), which vary among AAV serotypes. The VRs are mainly found at capsid protrusions, which are crucial for the interaction between AAVs and host cells, determining immunogenicity and receptor binding.¹⁵ Some VRs, such as VR-IV, V and VII;, are exposed on the surface of the protrusions, making them favored site for capsid engineering.¹⁶ Capsid proteins function in endosomal escape and nuclear localization¹³ and are responsible for tissue tropism.^17,18

Table 1 Properties of AAV Serotypes

The various AAV serotypes show different tissue tropisms that involve primary receptors and co-receptors on the cell surface.^5,11 AAV is endocytosed into endosomes by binding these receptors.⁴² The additional sequences at the N-termini of VP1 and VP2 are buried within the capsid.^13,14 Once in the endosome, AAV undergoes a structural change that results in the exposure of phospholipase A₂ (PLA₂) domain located in the N-terminus of VP1. The activity of PLA₂ is required for endosomal release and nuclear location.^43,44 After AAV enters the nucleus, it releases its single-stranded genome, which is then duplicated into a double-stranded genome from which transgenes can be expressed.⁴⁵ Table 1 lists the properties of current natural AAV serotypes.

Recombinant AAV (rAAV), an AAV with a genome engineered for gene therapy (Figure 1), which has been widely used in clinical trials. In 2012, the first AAV product, Glybera, which expresses lipoprotein lipase in muscle to treat lipoprotein lipase deficiency, was approved by the European Medicines Agency.⁴⁶ Six AAV drugs, namely Luxturna, Zolgensma, Hemgenix, Roctavain, Elevidys and Beqvez, have been approved by the United States Food and Drug Administration (US FDA). In 2017, Luxturna, which uses AAV2 encoding RPE65 as gene therapy for Leber congenital amaurosis, was the first US FDA-approved AAV bioproduct.⁴⁷ Before the approval of Luxturna, no drugs were available for the treatment of retinal dystrophy.⁸ In 2019 Zolgensma, an AAV9 vector from Novartis Gene Therapies, was approved by the US FDA for the treatment of type I spinal muscular atrophy in children, which is the most common single-gene disease that causes infant death.⁴⁸ Hemgenix is an AAV5 vector that expresses highly active factor IX (FIX) protein variant Padua for the treatment of haemophilia B. This CSL Behring gene therapy product, originally developed by UniQure, was approved by the FDA in 2022 based on safety and efficacy studies in adult male patients with severe or moderately severe disease.^49,50 Roctavian is an AAV5 vector-based gene therapy for the treatment of patients with severe hemophilia A who do not have a history of coagulation factor VIII inhibitors or detectable antibodies to the AAV5 type. On August 24, 2022, this therapy was conditionally approved by the European Medicines Agency, and on June 29, 2023, it was approved by the US FDA.⁵¹ On June 22, 2023, Sarepta announced that Elevidys, an AAVrh74 vector isolated from non-human primates, became the world’s first single-use gene therapy for Duchenne muscular dystrophy (DMD) and received FDA approval to be marketed in the United States for the treatment of independently ambulatory children aged 4–5 years old. DMD is a commonly observed- X-linked recessive hereditary muscle degenerative disease in which most affected children are unable to walk by the age of 12 years and die at approximately the age of 30 years.⁵² Furthermore, Beqvez also used AAVrh74 as vectors for the treatment of adults with moderate-to-severe hemophilia B (congenital factor IX deficiency). Many rAAV-based clinical trials involve neurological diseases, eye diseases and muscular disorders.^5,53

Figure 1 General structure of AAVs and rAAV. (A) Wild-type AAV genome. AAV has two open reading frames, including genes necessary for replication (rep) and capsid structure (cap). Rep portion encodes genes for viral genome replication, transcriptional regulation, and virion assembly (Rep78, Rep68, Rep52, Rep40). Cap portion encodes genes for viral capsid proteins (Vp1, Vp2, Vp3). (B) General recombinant AAV (rAAV) genome structure. The rAAV is composed of a promoter, target transgene and a PolyA tail. Promoter portion can drive expression of the target transgene. PolyA tail is a polyadenylation signal, which supports RNA nuclear output and translation. All these open reading frames are flanked by ITR, which form T-shaped hairpin ends. Abbreviations: AAVs, adeno-associated viruses; rAAV, recombinant adeno-associated virus; ITR, inverted terminal repeat.

The commonly used AAV-mediated gene therapies involve gene replacement, addition, and gene silencing.¹ Gene replacement can treat single-gene diseases by introducing functional gene copies. This approach, such as the US FDA-approved Luxturna gene therapies aforementioned, is promising for the treatment of rare, otherwise untreatable diseases.^46,48 Gene addition has even greater clinical potential than gene replacement because it can be used to treat non-single-gene diseases such as common chronic or infectious diseases. For example, arthrogen, currently in clinical trials (NCT02727764), encodes an anti-inflammatory IFN-β transgene to treat rheumatoid arthritis.⁵⁴ Gene silencing typically involves the use of RNA interference (RNAi) to silence specific functional genes. AAV takes advantage of the RNAi process in target cells to generate siRNA by encoding short-hairpin RNA (shRNA). siRNA is a double stranded RNA; of the two strands, one can bind to mRNA by complementary base pairing to suppress gene expression. However, AAV-mediated gene silencing remains predominantly used in preclinical research because of off-target effects and competition with endogenous RNAi constructs.^55,56 The AAV-CRISPR/Cas system has also been used for gene silencing to address the limitations of RNAi.⁵⁷

Although different AAV serotypes have different tissue tropisms, there are still limited cell types that native AAV can target. Currently, commercially available AAV drugs are based on the natural tropism of their serotypes, which limits widening their clinical application. In addition to limited tropism, AAV has been shown to induce immune responses, including neutralizing antibodies (NAbs) and cytotoxic T lymphocyte (CTL) responses. Notably, approximately 70% of the population have NAbs for AAV1 and AAV2, 45% for AAV9 and AAV6, 38% for AAV8.⁵⁸ These NAbs limit AAV uptake by target cells and intracellular transport and uncoating in the nucleus, preventing transduction and limiting systemic delivery. Concerns regarding the transduction efficiency of AAV have also been raised. AAV engineering is one method used to address the limitations. Strategies to modify AAV can be divided into three categories, capsid modification, surface tethering, and virus load (Figure 2). Table 2 lists the advantages and limitations of the three AAV engineering strategies. The following sections address these strategies in detail.

Figure 2 Strategies for engineering AAVs. (A) Capsid modification directly changes the capsid protein of AAV by gene engineering. (B) AAV surface is synthetically tethered with desired substance by binding or compounding with capsid protein. (C) Virus loading is to completely package AAV in a larger natural or synthetic platform material.

Table 2 Advantages and Limitations of AAV Engineering Strategies

Capsid Modification Strategy

AAV capsids can be engineered to evade immune responses and enhance cellular targeting. The three forms of capsid modification are rational design, directed evolution, and in silico design.

Rational Design

The amino acid sequence and structure of the capsid protein on the surface of AAV determine its affinity for different primary receptors and co-receptors located on the cell surface, providing key information that can be used to enhance or improve cell targeting via rational design. The rational design strategy, which includes point mutations and peptide insertion, attempts to alter the amino acid residues of the AAV capsid or insert targeting peptides into the capsid structure via genetic modifications to improve the tropism or evasion of NAbs and CTLs.

This strategy has been used to develop gene therapies for DMD, a life-threatening disease caused by mutations in the dystrophin gene. Achieving these therapeutic goals, a chimeric AAV capsid variant (AAV2.5) was designed by engrafting five mutations from AAV1 onto the AAV2 capsid.⁸¹ These five amino acid residues are responsible for the high muscle tropism. As shown in a clinical trial, compared with AAV2, AAV2.5 reduced cross reactivity against NAbs and enhanced skeletal muscle transduction.⁸¹ This technique utilizes the knowledge of AAV’s receptor-binding domains.

Site-directed mutagenesis technique is one way to improve brain tropism and reduce the liver targeting of AAV. Wang et al hypothesized that even a few changes in the capsid residues would alter the tissue tropism of AAV. Among the AAV9 variants, the single mutant vector rAAV9.R533S (which indicates that position 533 of the R residue becomes the S residue) and the double mutant vector AAV9.HR containing H527Y and R533S changes through rational design demonstrated significantly reduced liver targeting following intravascular (IV) and intramuscular (IM) administration. Quantitative polymerase chain reaction (qPCR) analysis showed that genome copy number of that of AAV9.HR in the liver was only 0.5% or 8% of wild-type (WT) AAV9 after IV or IM injection, respectively. Moreover, AAV9.HR retained its ability to cross the blood–brain barrier (BBB), resulting in widespread neuronal transduction.⁸² Seo et al constructed two engineered rAAVs, namely PHP.eB and CAP-B10. The PHP.eB vectors contained a modification at position 588 relative to AAV9, and the CAP-B10 vectors had further residue changes in the position 455 relative to PHP.eB. Although AAV9, PHP.eB, and CAP-B10 showed similar liver accumulation, the two engineered AAVs were mainly internalized by Kupffer cells and thus showed reduced hepatocytes transduction compared with AAV9. In addition, PHP.eB and CAP-B10 displayed enhanced brain accumulation in comparison to AAV9.⁸³ In addition, PHP.eB variants show high levels of transduction into the retina in mice.⁸⁴ Gilkes et al constructed two AAV8 variants: a double mutant (Y447 + 733F) and a triple-mutant (Y447 + 733F + T494V) and showed the triple-mutant AAV8 variant enhanced brain transduction compared with AAV8 and double-mutant AAV8 variant.⁸⁵ With this approach, it is possible to increase bioavailability, which will maximize the therapeutic potential.

The peptide insertion strategy ligates degenerate oligonucleotides into the cap ORF in order to insert peptide sequences into defined sites of capsid.⁸⁶ Several studies have attempted to insert specific nucleic acid sequences encoding targeting peptides into appropriate sites of the capsid genome. In 1998, Yang et al fused the single-chain fragment variable region (sFv) of an antibody against CD34 to the N-terminus of VP by linking it to the capsid gene. Of the three sFv-VP containing AAV2s, only sFv-VP2 containing AAV2 showed increased transduction in CD34-expressing KG-1 cells, a human acute myeloid leukemia cell line with a positive surface CD34 molecule. However, further improvements are required for in vivo application.⁵⁹ Wu et al performed mutational analysis of the AAV2 capsid gene and constructed 93 mutants by inserting peptides at the N-termini of VP1 and VP2. Insertion of the serpin receptor ligand at the N-termini of VP2 resulted in increased targeting of IB3 cells,⁸⁷ a lung epithelial cell line believed to express the serpin receptor. They also showed that fusion of the serpin receptor ligand into the N-termini of VP1 resulted in increased infectivity compared with that of wild-type AAV2.⁸⁷ Residue 587 (I587) is also an appropriate insertion site for targeting peptides.^60,88–92 For example, Girod et al constructed six AAV2 mutants by inserting an L14 targeting peptide into I-261, I-381, I-447, I-534, I-573, and I-587. Of the mutants, only I-587-L14 containing rAAV2 re-targeted viral tropism and increased transduction in AAV2-resistant Co-115 and B16F10 cells.⁶⁰ Nicklin et al used the same position to insert the SIGYPLP peptide, which efficiently targets endothelial cells (ECs) that are otherwise resistant to AAV vectors. Their results showed that SIGYPLP mediated AAV2 enhanced reporter gene expression in various ECs lines, including human umbilical vein endothelial cells and primary ECs.⁸⁸

Recently, this peptide insertion strategy has been used for other capsid proteins, serotypes, and targeting cell types. For example, in a recent study, the P1 peptide (RGDLGLS) was introduced into the VP3 of AAV9 to form a high astrocyte-targeting AAV variant (rAAV9P1).^93,94 The rAAV9P1 showed >90% astrocyte transduction in intact cerebral cortex and efficiently transduced reactive astrocytes in a mouse model of brain injury after intravenous injection.⁹⁴ The rAAV9P1 vectors also transduced human astrocytic cell lines more efficiently (>75%) than the wild-type rAAV9 (<20%) did.⁹⁴ Cell-penetrating peptides (CPPs) can also cross biological membranes.⁹⁵ Several studies had shown that the complexes of AAV and CPP can enhance the transduction of AAV2, AAV8 and AAV9.^96–98 Furthermore, some studies attempted to insert CPP into AAV capsids generate an AAV variant library. Yao et al identified an optimized CPP sequence (TVSALK) with increased brain penetration in mice through iterative screening and optimization of AAV variants. The obtained AAV.CPP.16 displays enhanced transcytosis at the BBB as well as increased efficiency of cellular transduction relative to AAV9, and that it can be used to deliver antitumour payloads in a mouse model of glioblastoma.⁹⁹ For transducing cardiomyocytes, AAV2-THGTPAD and AAV2-NLPGSGD variants were identified using an in vivo peptide display library and showed significantly enhanced efficacy in cardiomyocyte transduction compared with the AAV2 serotype.¹⁰⁰ For achieving high-level retina transduction, an AAV2-based peptide display library was used to select AAV2 variants.¹⁰¹ For lung delivery, the peptide THALWHT was inserted into the AAV2 capsid to obtain an AAV2 variant with enhanced affinity and high transduction efficiency to human airway epithelium.^61,102,103

Rational design strategies have also been employed for immune evasion. For example, owing to the compatibility of different AAV serotypes, Chai et al generated haploid AAV vectors by co-infecting AAV2 and AAV8 helper plasmids at ratios of 3:1, 1:1 and 1:3. These haploid viruses, such as AAV2/8 3:1, obtained higher transduction than their parental AAV2 and AAV8 after muscular administration. Moreover, they also produced the triploid virus AAV2/8/9 to further improve NAb evasion compared with haploid serotypes.¹⁰⁴ Inspired by the self-peptide (SP), a bioactive region of CD47, that acts as a “don’t-eat-me” signal for macrophages, Robinson et al inserted SP onto the capsid of AAV2 to obtain AAV2 mutants with decreased phagocytic susceptibility in vitro.¹⁰⁵ In another study, after injecting the AAV capsid fused to Tregitopes, the human MHC class II epitopes, CD8+ T in splenocytes of mice displayed reduced responses to AAV capsid.^104,106

In addition, several studies have leveraged point mutations and peptide insertions to achieve a more delicate function than that reported in the literature. For example, the rAAV-retro vectors, which include point mutation of N382D and V708I, as well as LADQDYTKTA peptide insertion at position 587 of AAV2 capsid, were constructed for achieving specific neural circuits targeting and retrograde tracing.¹⁰⁷ Further study of rAAV-retro displayed that the 10-mer peptide insertion is mainly responsible for the retrograde transport of neurons projecting to striatum and the restoration of V708I enhances the retrograde functionality in the amygdala-striatum pathway; nevertheless, the change in N382D has little effect on the retrograde ability.¹⁰⁸

In summary, these results suggest that a rational design strategy is able to efficiently generate mutants and alter tissue tropism, reduce the capture of immune system, and improve the transduction efficiency and specificity of AAV. However, because this strategy is based on an adequate understanding of AAV biology and the diversity of mutations is relatively low, its clinical use may be limited. With the in-depth understanding of AAV capsid protein and biological structure, through the comprehensive application of site-directed mutagenesis technique and peptide insertion strategy, rational design can greatly enhance the application potential of AAV in gene therapy.

Directed Evolution

Directed evolution is another capsid modification strategy that targets viral vectors via mutagenesis, including error-prone PCR, gene shuffling, Cre recombination-based AAV targeted evolution (CREATE), and a combination of above strategy and rational design. This strategy is useful when information on the structure or properties is insufficient.

Error-prone PCR is low fidelity PCR which incorporates random point mutations into the AAV cap ORF at a pre-set and adjustable rate, which leads to the novel variants library. Subsequently, selection can be performed by binding affinity columns, in vitro cell culture models or in vivo model.⁸⁶ Maheshri et al first used error-prone PCR and a staggered extension process to generate AAV2 mutants,⁶² generating random point mutations in the cap gene throughout VP1-3. Subsequently, a large AAV2 mutant library was obtained, and high-throughput selection processes were used to isolate mutants with high expression of primary receptor HSPG and low NAbs-binding.⁶² This process is time-consuming because of many variants. Structural analyses combined with error-prone PCR may narrow the size of the AAV variant library. For instance, Pulicherla et al obtained 43 viable variants using only mutations in the GH loop at amino acids 390–627 (VP1 numbering) on the outer surface of VP3. AAV9.45 and AAV9.61 showed liver-detargeted functions, and the number of vectors was reduced by approximately 10- to 25-fold. Furthermore, AAV9.45 displayed higher than AAV9.61 transduction in the cardiac and musculoskeletal tissues and decreased liver transduction.¹⁰⁹ A diagram of the error-prone PCR is shown in Figure 3.

Figure 3 Error-prone PCR. The main principle of error-prone PCR is to introduce random mutation into the gene of interest (red, yellow, green lines represent variants of cap gene), and form a plasmid with a mutated cap gene. Subsequently, the transfection of AAV-producing cells led to the production of mutant AAV capsids, and various AAV capsid libraries were formed. Following that, three methods, namely binding affinity columns, in vitro cell culture models and in vivo model were used to select AAV variants.

Gene shuffling strategy shuffles short gene fragments obtained by a modified PCR with other strands to produce new genes library, which leads to the production of novel AAV capsid variants (Figure 4).^{14,62,110–113} The idea behind gene shuffling libraries is to combine multiple advantageous features of different AAV serotypes, for example, to randomly reassemble the cap genome fragments from AAV1-9. The obtained chimeric-1829, with its genome composed of fragments from AAV1, 2, 8, and 9, showed improved tropism in AAV-resistant integrin minus hamster melanoma cells by binding to heparan sulfate receptors on the cell surface.¹¹² Compared with AAV1 and 2, chimeric-1829 displayed a three-fold and five-fold increase, respectively, in transduction efficiency in the tumor model; a lower level of transgene expression was observed in the liver and skeletal muscle.¹¹² Liu et al combined DNA shuffling and peptide display to generate AAV libraries, followed by cardiovascular EC-enriched variant isolation in a mouse model using flow cytometry and PCR. EC71 and EC73 variants were enriched in cardiovascular cells and transgene expression increased 8.8-fold and 6.3-fold, respectively, compared with AAV1 in vivo. By contrast, the vector genome (vg) copy numbers in the liver were reduced by 23.3-fold for EC71 and 28.5-fold for EC73, compared with AAV1. EC71 and EC73 also mediated long-term transgene expression in ECs for up to 4 months and show an approximately four-fold higher expression level in the heart than AAV1 dose.¹¹⁴ Moreover, progress has also been made for developing the AAV variants to effectively transduce brain,^115–118 spinal cord,¹¹⁹ muscle,^115,120,121 myocardium,^110,114 pancreas,¹¹⁵ liver¹²² and lung.¹¹⁵

Figure 4 Gene shuffling. Orange, purple and green lines represent variants of the Cap gene, which then undergoes fragmentation and shuffling of fragments. Subsequently, the transfection of AAV-producing cells led to the generation of new chimeric capsids, forming diverse AAV capsid libraries.

CREATE can be used to develop brain-targeting AAVs because the BBB hinders the entry of delivery vehicles into the CNS.³⁹ In this approach, AAV9 is used as a template to generate an AAV library. Subsequently, seven randomized amino acids are inserted at position 588–589 (VP1 numbering). AAV libraries are then administered intravenously to transgenic animals with specific Cre-expressing cells. The most abundant variant in the brain and spinal cord, AAV-PHP.B (containing the TLAVPFK peptide), was selected in vivo and showed over 40-fold greater transduction efficiency in the CNS than AAV9 did.³⁹ By using this approach, multiple variants with high permeability across the BBB were also obtained.^123–127

A directed evolution strategy has also been used to create novel immune evading AAV. In 2008, Kay et al used a gene shuffling strategy to construct a library of hybrid capsids from 8 WT viruses. By iteratively amplifying this library in human liver cells, the obtained single chimeric AAV variant, AAV-DJ, resisted NAb neutralization better than AAV2 did.¹¹³ Tse et al developed a structure-guided evolutionary approach based on the hypothesis that antibody recognition sites are evolutionarily conserved. An evolved capsid antigenic motif (CAM), CAM130, was able to effectively evade NAb while maintaining tissue tropism in non-human primate and human serum compared with the parent AAV.¹²⁸ Likewise, Havlik et al evolved a humanized AAV8 capsid, AAVhum.8, which displayed human hepatocytes tropism and NAb evasion.¹²⁹

Directed evolution can also generate highly diverse mutants and expand the pool of rAAV. Unlike rational design strategies, directed evolution strategies do not require understanding AAV biology. However, this approach is time-consuming because of the many variants and the in vivo selection process. Among diverse mutants, AAV with targeting specificity, safety and durability may be included. In combination with a rational design, it could have wide applications and reduce the workload of future clinical trials and achieve the goal of safe and lasting gene therapy.

In silico Design

Some studies have also adopted in silico designs to discover novel AAV variants, such as ancestral reconstruction and machine learning.

Ancestral reconstruction is a computational design approach that uses phylogenetic data and gene sequences to generate new capsids.^63,130 Unlike modern serotypes, ancestral variants exhibit unique cellular entry mechanisms. For example, Ortiz et al used 32 variable sites to obtain an ancestral AAV library. These variants had similar infection efficiencies as the most efficient modern serotypes in various cell lines, such as C2C12 mouse myoblast and glioblastoma tumor, but showed 19–31-fold higher muscle tropism than AAV1 did. The ancestral variants also displayed a NAb-binding ability similar to that of AAV1. Furthermore, these ancestral capsids show higher thermostability and do not bind to the usual receptors of modern AAV serotypes, such as HSPG, sialic acid, and galactose.⁶³ In another study, the Anc80 ancestral AAV vector was selected from 75 AAV capsids via in silico ancestral sequence reconstruction and tested in mice and non-human primates. This Anc80 vector is considered the ancestor of AAV1, 2, 8, and 9 and was able to transduce multiple tissues more efficiently, including the muscle, retina, and liver, than AAV2 and AAV8 can.¹³⁰ Anc80 was also used in a study leveraging the Anc80L65 clone as a vehicle to deliver the USH1C gene encoding harmonin to the cochlea and vestibular sensory organs in a murine model of Usher syndrome to improve hearing loss. The results showed that Anc80L65-mediated harmonin expression rescued sensory transduction in Ush1c hair cells, auditory brainstem responses, and auditory and vestibular behaviors in an Ush1c mouse model.¹³¹

Machine learning (ML) is a promising strategy for variants AAVs designs. Ogden et al replaced random mutagenesis with an algorithmic approach. In this study, the AAV2 mutation landscape was restricted to capsid positions 561–588.¹³² Bryant et al leveraged deep learning to construct an AAV2 variants library. The authors tested three training data libraries designed using complete, random, or additive sampling strategies. The three datasets contained (i) a complete single variant set plus randomly chosen variants with double mutants, (ii) complete single mutation variants plus, and (iii) randomly chosen variants with multiple-mutants between 2 and 10 mutations plus the additive model-designed variants. Three machine algorithms were trained using the three datasets and used to generate the AAV2 variants.¹³³ Nevertheless, these starting libraries often consist of a substantial number of variants that are incapable of assembling properly or packaging their genomes, which leads to a significant waste of library. Therefore, Zhu et al developed ML-guided AAV peptide insertion libraries to balance the diversity and packaging fitness. In a selection of primary human brain tissue, the ML-guided library produced a 10-fold increase in the number of infectious variants in comparison to standard NNK library, with minimal impact on diversity.¹³⁴ The data-driven approach displays significant viability compared with random mutagenesis and allows deeper research on unexplored AAV sequences, transforming future capsid engineering strategies. However, the predictive power of machine learning depends heavily on model design, data representation and training data generation.⁶⁴

Surface Tethering Strategy

In contrast with gene engineering, surface tethering strategies use chemical modifications to evade NAbs or achieve tissue targeting. Thus, it provides a simple, multi-functional, and universal approach to extending the reach of AAV.

Several studies have used the polyethylene glycol (PEG) strain to attach polypeptides, nanoparticles, and biomedical molecules.^65–67 PEGylation is the process by which PEG tethering or coating endows molecules or nanoparticles with the ability to evade the immune response, prolonging their circulation time. PEGylation has also been used to modify AAV surfaces. Lee et al first investigated whether PEG length and the PEG:lysine ratio affect serum neutralization and viral infectivity.⁶⁸ PEG with succinimidyl propionic acid groups reacts with the abundant lysine residues on the viral capsids. However, an excessive PEG coating ratio may compromise the interactions between AAV and its receptors, impairing viral infectivity. A small PEG coating range showed improved infectivity and NAbs escape, and the width varied with PEG size.⁶⁸ Another study examined the covalent conjugations of PEG to AAV.¹³⁵ Succinimidyl succinate and tresyl chloride linkers improved lung transduction efficiency despite similar liver and muscle targeting. The use of tresyl chloride for PEG attachment provided the best protection from NAbs, whereas the cyanuric chloride linker induced AAV aggregation, compromising transduction.¹³⁵ Other conjugation methods have been used. For example, an azide moiety has been incorporated into the AAV capsid to achieve precise site-specific PEGylation using click chemistry.¹³⁶

In addition to PEGylation to protect against neutralization, the use of a targeting peptide or protein incorporation strategy can improve or alter the tropism of AAV capsids. Owing to the high affinity of streptavidin for biotin, proteins genetically fused with core streptavidin have been used to conjugate biotinylated AAV2 capsid.⁷⁰ In a study using the epidermal growth factor (EGF) or fibroblast growth factor 1α (FGF1α) as a targeting ligand, EGF-AAV2 and FGF1α-AAV2 showed increased infectivity in EGFR-positive SKOV3.ip1 cells or FGFR-1α-positive M07e cells. Nevertheless, the capsid-modified AAV2 vectors retained their natural tropism.⁷⁰ Liu et al developed a site-specific modification strategy that involved the genetic inserting of a 13-amino acid peptide (LCTPSRAALLTGR) at amino acid position 587, in which the cysteine in the peptides was converted to an aldehyde-containing formylglycine residue in the presence of a formylglycine generating enzyme. The aldehyde group showed good reactivity toward hydrazide- or aminooxy-functionalized molecules such as targeting ligands or fluorophores.⁶⁹ In this case, incorporation of the anti-HLA antibody into AAV2 significantly increased 293T cell transduction, whereas AAV2 with a chemically modified arginine-glycine-aspartate (RGD) motif increased HeLa cell transduction compared with wild-type AAV2.⁶⁹ Zdechlik et al genetically introduced an HUH tag into the AAV capsid. Because the HUH tag can covalently link to sequence-specific single-stranded DNA (ssDNA), an ssDNA-conjugated antibody was incorporated into the viral capsid. The AAV-HUH-antibody composites displayed increased transduction in HEK293 cells compared with wild-type AAV. Furthermore, anti-L1CAM-AAV displayed significantly increased transduction efficiency toward neurons but not glial cells, which do not express L1CAM.¹³⁷

In addition to the biochemical materials aforementioned, nanoparticles can be linked to AAV to enhance transduction. Magnetic nanoparticles have been widely used in nanomedicine because of their unique magnetic tropism.¹³⁸ Several attempts have been made to combine magnetic nanoparticles with AAV. Viral vectors were linked with heparin-coated superparamagnetic iron oxide nanoparticles (HpNPs), which magnetize in a magnetic field.⁷¹ Magnetically guided AAV was shown to enhance delivery efficiency in PC12 and HEK293T cell lines. Moreover, HpNP-mediated AAV delivery has a shorter transduction time (<180 min) than but a similar transduction efficiency to the conventional protocol which typically requires 24 h of incubation.⁷¹ In another study, streptavidin-coated superparamagnetic iron oxide nanoparticles (SPIONs) were linked to biotin-nitrilotriacetic acid (NTA), which can chelate Ni ions. For further conjugation with AAV, hexa-histidine was inserted into AAV capsids. The acquired complexes (AAV/NiStNPs) displayed enhanced infection and rapid internalization by non-permissive human neural stem cells.¹³⁹ Although magnetic nanoparticle-mediated AAV delivery shows rapid, efficient transduction in various cell types, in vivo magnetic delivery of AAV remains a challenge. Recently, Mukherjee et al explored local magnetic targeting of AAV in vivo. The treatment of noise-induced hearing loss is limited by the absence of an effective drug delivery strategy. Mukherjee et al used SPION-AAV complexes for local magnetic delivery of brain-derived neurotrophic factor. AAV2 (quad Y-F), in which the four capsid tyrosine residues were replaced with four phenylalanine residues, was conjugated with commercial Mag4C-Ad SPIONs. Through magnetic targeting of delivery to the inner ear in a noise-induced hearing loss rat model, SPIONs-AAV2 (quad Y-F) carriers increased brain-derived neurotrophic factor gene expression, improving hearing ability.¹⁴⁰

The scope of surface tethering strategies continues to expand, because this strategy can be used to attach various materials (eg, polymers, antibodies, peptides, fluorescent probes, nanoparticles) via diverse conjugation methods especially chemical conjugation. Surface tethering strategies can affect circulation and clearance, accumulation and retention in tissues, and interaction and internalization of cells, so it has a wider application range than capsid modification does. However, complex protocols may limit their clinical applications and compromise their clinical value.

Virus Load Strategy

The load strategy is another efficient method for protecting AAVs from neutralization and changing the natural targets. Most importantly, this strategy provides a protocol for the co-delivery of AAVs and other drugs and endows rAAVs with the additional benefit of responsive or controllable release.

Liposomes have a bilayer membrane structure and can load various substances such as plasmids, proteins, and small-molecule drugs. Since 1984, liposomes consisting of cholesterol, phosphatidylcholine, and phosphatidylglycerol prepared using the reverse-phase evaporation method have been used to encapsulate retroviruses, and have shown improved transduction efficiency in resistant cell lines.¹⁴¹ The encapsulation of rAAV into liposomes also yielded exciting results. For example, rAAV-associated liposomes, which comprise N-(α-trimethylammonio-acetyl)didodecyl-D-glutamate chloride, 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine (DOPE), and dilauroyl phosphatidylcholine, exhibited more than six-fold targeting efficiency compared with rAAV alone in three glioma cell lines.⁷³ However, in vivo viral delivery involves a more complex physiological environment; thus, the requirements for rAAV-associated liposomes are higher. Fiandaca et al used cationic lipids containing gadolinium and rAAV to achieve real-time magnetic resonance imaging (MRI) visualization of rAAV in non-human primate brain. Liposomes containing gadolinium were prepared via a thin-film hydration method using 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC), cholesterol, and 1,2-distearoyl-sn-glycero-3-methoxy (polyethylene glycol)-2000 (PEG-DSG). Gadolinium-loaded liposomes were monitored for rAAV distribution. The distribution of rAAV shown using immunohistochemistry (IHC) was consistent with that of the gadolinium-loaded liposomes on MRI scans.¹⁴² In another study, two cationic lipids were synthesized to encapsulate AAV2/9 and AAV2/6.2, which showed unique tropism for the alveolar epithelium and conducting airway/alveolar epithelium, respectively. After encapsulation, the complexes exhibited positive zeta potentials and increased infection efficiency in A549 cells. In the analysis of β-galactosidase (β-gal) expression by AAV, AAV-encapsulating liposomes showed a four-fold increase in β-gal expression in vivo compared with control AAV2/9 or AAV2/6.2 after intranasal administration. Moreover, AAV2/9-encapsulating liposomes change the viral tropism and increased the conducting airway transduction efficiency of AAV2/9.¹⁴³ AAV vectors were entrapped in the liposome, hindering the AAV capsid-cell membrane interaction. The altered tropism might result from the uptake of liposomes into the cells. These cationic lipids are also responsible for retargeting AAV vectors. Moreover, cell targeting can be conferred on rAAV-associated liposomes via conjugation with specific antibodies or targeting peptides.

Exosomes, lipid membrane vesicles derived from cells (40–160 nm in the diameter), contain multiple components including DNA, RNA, lipids, and cell-surface proteins and function as transit systems between cells.¹⁴⁴ With the advantages of non-immunogenicity, natural tropism and non-toxicity, exosomes have attracted increasing attention as natural drug delivery systems, especially for gene delivery. Notably, exosomes can protect against NAbs and target various tissues because of their natural tropism. Some AAVs are associated with exosomes in viral production and are termed vexosomes or exosome-AAV (Figure 5). Vexosomes were shown to display enhanced transduction in 293T and U87 cell lines and improve NAbs evasion compared with standard AAV.⁷⁴ Moreover, their tropism of vexosomes can be modified by expressing transmembrane receptors on the surface. In one study, the biotin acceptor peptide-transmembrane domain (BAP-TM) and biotin ligase BirA were introduced into 293T cells using lentivirus transduction to produce BAP-TM vexosomes. After incubating BAP-TM vexosomes with magnetic beads containing streptavidin, BAP-TM vexosomes showed two-fold enhanced transduction compared with that of vexosomes within magnetic fields.⁷⁴ Vexosomes are also suitable gene delivery tools for treating CNS diseases. Maguire et al isolated vexosomes from conditioned media of 293T producer cells via ultracentrifugation. The exosome associated AAV vector effectively targeted the CNS after systemic administration and crossed the BBB in vitro, which may be explained by the ability of exosomes to cross the BBB. Furthermore, intravenously injected exosomes-AAV9 showed higher transduction of neurons and astrocytes than with AAV9 alone.¹⁴⁵ Other studies have reported that exosome-AAVs were able to improve gene therapy in various tissues, including the liver,¹⁴⁶ lung,¹⁴⁷ muscle,¹⁴⁸ ear,¹⁴⁹ and retina.¹⁵⁰ These findings highlight the obvious benefits of exosome-AAV, which can be summarized as (i) natural or modified tropism, (ii) NAbs evasion, and (iii) the ability to cross biological barriers. Unlike synthetic liposomes, which are mainly composed of natural or synthetic lipids, multiple transmembrane receptors are naturally expressed in exosomes derived from parent cells.¹⁵¹ The cell source influences exosome membrane protein types, which further influences the biodistribution and cell tropism of exosomes. The heterogeneity and endogeneity of exosomes provide unique advantages over liposomes regarding targeted delivery and immune evasion. However, despite exosome-AAV, there are multiple benefits of exosome-AAV compared to AAV or liposome-AAV, a manufacturing process of exosome-AAV is challenging. For example, the extraction and purification of therapeutic exosome-AAV in a homogeneous manner to meet both scale and clinical requirements present great difficulties.⁷⁵

Figure 5 Overview of exosomes-AAV production protocol. 293T cells were transfected with AAV plasmid expression vector encoding targeting ligand, and then the cells secreted free AAV and exosome-AAV, and it was obtained by ultracentrifugation.

Hydrogels can also endow AAV with unique pharmacological characteristics such as sustained release, controlled release, and local administration. Hydrogels are widely used to deliver biomacromolecules. Thus, PEG-based hydrogels can serve as biocompatible scaffolds for AAV release. By incorporating polyhistidine into PEG hydrogels, pH-sensitive hydrogels were synthesized for AAV2 encapsulation. These hydrogels released AAV2 at different rates under different pH conditions. Owing to the more acidic physiological environment of injured tissues, PEG-polyhistidine hydrogels have faster viral release.⁷⁶ Self-assembled peptide-based hydrogels have also shown ideal biocompatibility. Cucchiarini et al constructed pure RAD peptide hydrogels or RAD hydrogels combined with hyaluronic acid to achieve sustained release of AAV into human mesenchymal stem cells, providing effective delivery vehicles for tissue lesions.⁷⁷ In another study, AAV9SLR vectors were embedded in alginate hydrogels for targeted delivery into the abdominal aorta of mice.⁷⁸ These studies demonstrate that hydrogels are safe, effective biomaterials for coating viral vectors and improving in vivo kinetics and targeting abilities.

Microneedles (MNs) are small needles that can overcome biological barriers and deliver biomacromolecules to specific regions. Ye et al used AAV-harboring MNs to target rat hearts and burst-release agents, thereby improving heart function.⁷⁹ These MNs were prepared using polyvinyl alcohol (PVA) and did not degrade. The AAV was dispensed onto the surface of the MNs. The MNs displayed good drug-loading capacity, and over 4.93 × 10¹⁰ vg of virus could be loaded in the MN array. MNs showed a burst release, releasing 90.93% of AAV within 2 s and 92.42% of AAV within 5 s. Twenty-eight days after the administration of MN-AAV loaded with the vascular endothelial growth factor (VEGF) gene, myocardial infarction model rats showed improved cardiac function, reduced infarct size, and improved adverse remodeling.⁷⁹ Liu et al used MN-AAV-VEGF to treat ischemic stroke. MNs were prepared using gelatin methacryloyl (GelMA), which is biocompatible, biodegradable, and photocrosslinkable via photoinitiators. In contrast with PVA, the GelMA MNs released AAV in a sustained and controlled manner, avoiding potential burst release-induced neurotoxicity. MN-AAV-VEGF delivery promoted angiogenesis and neurogenesis and reduced modified neurological severity scores (mNSS) in a stroke rat model.⁸⁰ These findings support the value of MN-AAV as a minimally invasive and locally distributed strategy.

The load strategy regards AAV as a biomacromolecule drug and takes advantage of pharmaceutical methods, offering more possibilities than the capsid modification or surface tethering strategies. The loading strategy, as a noncovalent surface modification strategy, can increase AAV transduction efficiency and change the tropism, in addition to endowing AAV with responsive and controlled release characteristics, which will be beneficial for clinical translation.

Conclusions

Gene therapy has substantial potential for the treatment of various diseases. Thus, the attention to developing safe, effective gene delivery vectors is increasing. Owing to its well-tolerated properties and effective mechanism of entry into the nucleus, AAV is a useful delivery vector for gene therapy and has been used in many clinical trials. However, the use of AAV is limited by its low transduction efficiency in the human body, limited tissue tropism, low payload, immune responses against viral capsids or transgenes, and insufficient viral production. AAV engineering strategies have attempted to improve tissue or cell tropism and evade immune responses. Modern technologies, including bioinformatics, structural analysis, and proteomics, provide useful information on capsid modification and surface tethering. Various pharmaceutical methods have been applied to AAV load strategies to provide additional advantages. Despite numerous preclinical studies suggesting that these novel rAAV forms have obvious advantages over standard AAV in animal models, additional evidence is required to confirm whether these benefits extend to humans. Moreover, the challenges are encountered to make these novel rAAV forms to effectively manufacture. For example, the AAV engineering strategies result in lower yield and higher cost and increase the risk of high variability of product quality. Furthermore, limited understanding on the AAV engineering also hinders the manufacture. The above has posed challenges to regulatory. Therefore, further studies are warranted in this regard. Engineering of rAAV should consider various parameters, such as serotypes, target cells, administration routes, and dosage forms, all of which influence the overall curative effect. Improving the design and AAV manufacturing will benefit clinical outcomes and facilitate US FDA approval. Besides, the novel clinical frameworks are required for the assessment and management of these novel AAV therapies.