A Computational and Experimental Platform for Detecting Full Transcriptome Cell Type Tropism of Lowly Expressed Barcoded and Pooled AAV Variants via Single-Cell RNA Sequencing

Abstract

Despite being one of the primary gene therapy delivery vehicles, adeno-associated viruses (AAVs) are limited in their specificity towards certain cell types implicated in disease. Recombinant AAVs (rAAVs) are addressing these limitations through both capsid engineering and gene regulatory approaches that alter viral tropism or viral expression patterns. Current rAAV targeting, selection, screening, and characterization methods are typically based on single- or few-molecule read-outs, such as promoter and enhancer-driven constructs, mouse lines expressing Cre recombinase under a cell type-specific promoter, or cell type-specific antibodies for imaging. Such methods harbor challenges for parallelizing rAAV characterization, or extending characterization and engineering to complex or previously unknown cell types. The recent advent of single-cell RNA sequencing (scRNAseq) has revealed a rich diversity of cell types and states, many of which are not associated with canonical cell type markers, and can even be defined by multi-gene programs. To aid in the engineering of rAAVs aimed at such complex cell states and aid in the discovery of novel tropisms, we have developed a scRNA-seq AAV screening method, whereby we inspect full transcriptomes of cells transduced with pools of AAV vectors in a single animal. To generate pools of variants that can be differentiated in sequencing, we package variants with either unique transgenes, or the same transgene with unique barcodes incorporated in the polyA region. We then co-inject mice with these pools of variants, wait for expression, and harvest tissue slices for downstream cell dissociation and single-cell sequencing using the Chromium 10X Single Cell Kit. In order to accommodate the low expression rates of virally delivered cargo and the loss of the region of mRNA upstream of the polyA capture site that identifies the capsid variant, we amplify the viral transcripts from the full cDNA library with primers near the differentiating region of the cargo. To characterize variants, we developed a customized computational pipeline that addresses the unique challenges of these datasets: (1) to discern the variant that delivered each transgene read, we demultiplex the amplified viral transgene reads based on their differentiating sequences; (2) to reduce the effects of PCR amplification noise, we convert variant transgene reads into probabilistic estimates of the number of transcripts per cell; and (3) to calculate cell type biases, we automatically identify a cell type hierarchy and compare the distribution of viral transcripts by cell type to a null model of empty droplets. Thus far, our platform has corroborated several expected virus tropism findings from imaging (e.g. for brain vasculature or neuronal preference). To apply this barcoding strategy to even larger pools without individually cloning and producing each variant, and link arbitrary mutations in the capsid genome to the barcode, we have further developed a plasmid that contains both the expressed transgene and the capsid gene, but inverted in orientation, with their 3' ends adjacent. With these barcoding strategies and computational methods, we enable fast identification and characterization of rAAV variant pools with precise disease-relevant tropisms, with the ultimate goal of aiding the gene therapy field in developing precision delivery vehicles.

Additional details