Computationally guided AAV engineering for enhanced gene delivery

Abstract

Gene delivery vehicles based on adeno-associated viruses (AAVs) are enabling increasing success in human clinical trials, and they offer the promise of treating a broad spectrum of both genetic and non-genetic disorders. However, delivery efficiency and targeting must be improved to enable safe and effective therapies. In recent years, considerable effort has been invested in creating AAV variants with improved delivery, and computational approaches have been increasingly harnessed for AAV engineering. In this review, we discuss how computationally designed AAV libraries are enabling directed evolution. Specifically, we highlight approaches that harness sequences outputted by next-generation sequencing (NGS) coupled with machine learning (ML) to generate new functional AAV capsids and related regulatory elements, pushing the frontier of what vector engineering and gene therapy may achieve.

Keywords: AAV libraries; ancestral sequence reconstruction; directed evolution; machine learning; next-generation sequencing; protein engineering.

Figure 1.. AAV capsids engineering strategies.

AAV libraries can be created through various methods, including random diversification techniques, computationally aided design, and the utilization of computational models specifically trained to construct capsids with desired properties. These diversified capsid libraries are then screened within a chosen system of interest, in vivo or in vitro. The top variants, identified via sequencing, can then be further validated, or serve as templates for subsequent diversification efforts. Graphics are created with BioRender.com.

Figure 2:. Overview of AAV Intracellular Trafficking Pathways and Barriers.

(A) AAVs selectively bind to specific receptors (illustrated in light purple), followed by endocytosis. (B) Intracellularly, AAVs encounter progressively acidified endosomal environments, which trigger AAV capsids to undergo conformational changes. The resulting exposure of the VP1 unique region, which harbors the phospholipase A2 domain, enables viral escape into the cytosol and subsequent trafficking to the nucleus for genetic cargo release[26]. Variants that do not escape the endosomal network are routed towards lysosomal degradation. (C) The vesicular trafficking pathway is not fully elucidated, but current studies indicate a possible transit to the trans-Golgi network, culminating in perinuclear localization[27,28]. The entry into the nucleus is believed to occur through the nuclear pore complex[26,29,30], and mutations in VP1u basic cluster regions, putative nuclear localization sequences, impair nuclear import[31]. (D) For expression of its transgenes, AAV must navigate through the nucleolus and return to the nucleoplasm, a process necessary for the release of the AAV genome[30,32]. Variants that reach the nucleus but fail to release their genome remain non-expressive. Graphics are created with BioRender.com.

Figure 3:. Selective pressure can be harnessed through various approaches to enhance the translational potential of viral capsids for clinical applications.

In vivo methods (A) allow for the evolution of capsids that may exhibit the ability to overcome anatomical barriers (e.g., blood-brain barrier). Conversely, in vitro screening of diversified libraries (B) can yield variants capable of efficiently transducing particular cell types, with the risk that cell identity or function in vitro may differ from that in vivo. (C) Ex vivo tissue or organoids enable selection in human cells. Notably, a synergistic approach, combining in vitro and in vivo screening modalities, offers the potential to evolve capsid variants that can overcome anatomical hurdles and infect human cells. Graphics are created with BioRender.com.

Figure 4.. Machine learning guided AAV capsid engineering.

(A) Computational and randomly diversified AAVs libraries can be selected, then sequenced through Next-Generation Sequencing tools to yield data to train predictive models for optimizing packaging fitness or tissue tropism. (B) Supervised models use labeled data to train and validate models for prediction such as classification and regression. (C) Unsupervised models use unlabeled data to find hidden patterns of data structures, typically used for clustering. Graphics are created with BioRender.com.

Figure 5.. AAV regulatory elements engineering workflow.

(A) A promoter library is assembled through one of three distinct methods. (B) This library is subsequently packaged into a single AAV variant for screening in the target tissue, and reporter gene expression levels are quantified through high-throughput sequencing. (C) The data generated from this screening process is employed to train a predictive model aimed at forecasting heightened promoter activity. (D) Finally, the most promising promoter is validated through the utilization of AAV constructs. Graphics are created with BioRender.com.