Fantastic AAV Gene Therapy Vectors and How to Find Them-Random Diversification, Rational Design and Machine Learning

Abstract

Parvoviruses are a diverse family of small, non-enveloped DNA viruses that infect a wide variety of species, tissues and cell types. For over half a century, their intriguing biology and pathophysiology has fueled intensive research aimed at dissecting the underlying viral and cellular mechanisms. Concurrently, their broad host specificity (tropism) has motivated efforts to develop parvoviruses as gene delivery vectors for human cancer or gene therapy applications. While the sum of preclinical and clinical data consistently demonstrates the great potential of these vectors, these findings also illustrate the importance of enhancing and restricting in vivo transgene expression in desired cell types. To this end, major progress has been made especially with vectors based on Adeno-associated virus (AAV), whose capsid is highly amenable to bioengineering, repurposing and expansion of its natural tropism. Here, we provide an overview of the state-of-the-art approaches to create new AAV variants with higher specificity and efficiency of gene transfer in on-target cells. We first review traditional and novel directed evolution approaches, including high-throughput screening of AAV capsid libraries. Next, we discuss programmable receptor-mediated targeting with a focus on two recent technologies that utilize high-affinity binders. Finally, we highlight one of the latest stratagems for rational AAV vector characterization and optimization, namely, machine learning, which promises to facilitate and accelerate the identification of next-generation, safe and precise gene delivery vehicles.

Keywords: AAV; adeno-associated virus; capsid engineering; gene therapy; molecular evolution.

Figure 1

Structure of the AAV cap gene and technologies for its diversification. (A) Schematic of the AAV cap gene including variable regions (VRs I-IX according to Govindasamy et al. [36]) and transcriptional start sites for VP1, VP2 and VP3, as well as MAAP [37] and AAP [38]. p5, p19 and p40 are the endogenous AAV promoters. poly-A, polyadenylation signal. (B) Tropisms of AAV vectors can be defined by choosing one of 13 primate AAV serotypes (AAV1-13) or a plethora of other naturally occurring isolates from various species. (C) Wild-type tropisms can be modified by mutagenesis of one or several capsid residues (e.g., Kern et al. [39]). (D) Insertion of pre-defined or randomized peptide sequences (e.g., a randomized 7 mer peptide “P7”; red indicates the peptide sequence and black the flanking residues, such as glycine or alanine that can be used as linkers) can be performed within WT cap backbones (e.g., Müller et al. [40]), in synthetic capsids such as shuffled variants (e.g., Tan et al. [41]), or in backbones already carrying an independent peptide insertion in another position (e.g., Goertsen et al. [42]). The colors of the individual capsid fragments denote the serotype origin according to the legends in the upper right corner of this figure. (E) Recombination of larger cap stretches from several parental capsids can be performed via domain swapping (e.g., Shen et al. [43]), SCHEMA-based shuffling through pre-defined optimal crossover points (marked with “x”) (e.g., Ojala et al. [44]), DNA family shuffling based on partial sequence homology (e.g., Grimm et al. [45]), or virtual VR shuffling (e.g., Marsic et al. [46]).

Figure 2

Synthetic biology-inspired approaches to modify AAV vector tropism. Antibodies can be coupled to AAV capsids via a covalent interaction between a HUH tag in the AAV capsid protein and the antibody, which is enabled through an oligonucleotide bridge (HUH-AAV) [32]. Non-covalent interactions can also be harnessed, for instance, by using an Fc-binding Z34C domain integrated into the AAV capsid (AAV-Z34C) [135] or a bispecific antibody that recognizes a conformational epitope [136] or a tag inserted into the AAV capsid (F(ab)₂-AAV) [137]. Other molecules such as nanobodies (Nb) inserted into the GH2/GH3 loop of VP1 [29] or DARPins integrated into the same loop [30], fused to the VP2 N-terminus [138,139] or covalently linked [140] can also be used to efficiently retarget AAV vectors. This figure contains free clipart from https://smart.servier.com/ (accessed on 1 April 2022).