UC Berkeley

Adeno-associated virus (AAV) has emerged as the most prevalently used viral vector for gene delivery in the past decade, owing to its unique attribute of low immunogenicity, lack of pathogenicity and efficient targeting of somatic cells. The AAV viral genome are flanked by inverted terminal repeats (ITRs) and encodes two major classes of proteins, namely replication (Rep) and structural (Cap) associated proteins. The Rep gene encodes four proteins that recognize ITRs to initiate genome replication and encapsidation, while the Cap gene encodes three VP structural proteins that assemble into a 60-mer AAV capsid with a T=1 icosahedral geometry. Initially, the VP1:VP2:VP3 ratio during capsid assembly was proposed to be 1:1:10. However, studies using advanced mass spectrometry instruments have revealed that the assembly ratio is stochastic, and small variations in VP components can significantly impact AAV function. As AAV gains more popularity and attention in the clinical field, thorough characterization of its heterogeneity is essential for quantifying and understanding its underlying biology, thereby improving engineering and manufacturing processes. Another obstacle, other than characterization, faced by AAV based gene therapies, is the prevalence of pre-existing neutralizing antibodies. The binding of these antibodies onto AAV can inhibit its interaction with cellular receptors, and further impairing its function by hindering the intracellular trafficking processes. Developing solutions to evade neutralization would greatly expand the applicability of AAV capsids to a broader patient population. A potential solution to address characterization of heterogeneous AAV is through the use of newly developed charge detection mass spectrometry (CDMS). Unlike conventional mass spectrometry, where the measurement only outputs mass-to-charge ratios and requires assumptions about charge states for calculation of total mass, CDMS allows simultaneous measurements of mass-charge ratio and the charge carried by each individual ion. This allows the determination of total mass and the respective charge at single ion resolution. Utilizing this advantage, we investigated the behavior of AAV9 capsid underwent acidification in different buffers and different incubation temperatures. Interestingly, AAV9 exhibited a subpopulation with a lower charge state configuration under environmental stress. These charge states were less abundant in ammonium acetate buffers and were accompanied by the appearance of lower-mass species, indicating that ammonium acetate destabilizes AAV9. In contrast, the acidification in PBS results in more abundant lower charge populations and the lack of lower mass species, highlighting the importance of performing these experiments in more biological relevant buffer to avoid misleading 1 conclusions. Furthermore, the nuclease digestion experiment at acidified conditions indicated the species with lower charge states are associated with altered capsid integrity, posing the viral genomes more accessible for TLR9 receptor recognition. To address the challenge of pre-existing neutralizing antibodies, we engineered a highly divergent, non-mammalian AAV variant—Serpentine AAV (SAAV). Due to the high divergence in the amino acid identities of SAAV and other serotypes, we hypothesized we could leverage this property for high resistance of neutralization antibodies while engineering it to obtain humoral infectivity. Characterization of SAAV’s binding and internalization profiles with HEK293 cells revealed superior binding efficiency compared to AAV2 and a comparable level of internalized virions. A novel genome-tracking approach allowed us to identify trafficking defects that limited SAAV’s infectivity. Functional rescues of SAAV was achieved by replacing the VP1/2N domains with those of AAV2 or AAV5. Using these chimeric SAAV variants as templates, we further enhanced its transduction ability through iterative directed evolution, expanding the AAV toolkit into previously uncharted evolutionary space Finally, I will provide an overview of commonly used AAV diversification techniques and iterative directed evolution strategies to identify superior variants. I will also highlight the tradeoffs of high diversity with lower viability, necessitating more iterative rounds of screening for better variants. With the advent of advanced machine learning models and the advances in computation hardware, ML has begun to influence the field of AAV engineering. I will review existing ML guided approaches for AAV library design and screening, highlighting their current applications and the potential to accelerate future vector development. I will conclude with perspective on how ML can guide the design of AAV capsids and transgene elements to create better vectors, ultimately benefiting society.