Quadruple adenine base edited allogeneic CAR T cells outperform CRISPR/Cas9 nuclease engineered T cells

Genome editing technologies have enabled the clinical development of allogeneic cellular therapies, yet the optimal gene editing modality for multiplex editing of therapeutic T cell product manufacturing remains elusive. In this study, we conducted a comprehensive comparison of CRISPR/Cas9 nuclease and adenine base editor (ABE) technologies in generating allogeneic chimeric antigen receptor (CAR) T cells, utilizing extensive in vitro and in vivo analyses. For the SureSelect panel, baits were designed against all predicted on- and off-target regions in the human genome (hg38) for the gRNAs utilized in this study, based on sequence homology between the gRNA protospacer and the genome. Off-targets were defined as having up to five mismatches to any on-target or one indel and up to four mismatches to any on-target. This resulted in 15,920 captured regions for the Cas9 quadruple editing scenario and 25,988 captured regions for the ABE quadruple editing scenario. For targets with <=2 mismatches (including on-target), baits extended +/-150 bp around the expected cut site. For off-target sites with >2 mismatches, baits extended +/-100 bp around the cut site. Baits were tiled to predicted genomic targets to minimize binding within cut sites while maintaining optimal capture efficiency, irrespective of INDEL formation due to editing. For Cas9 and ABE, the cut site was assumed to be 3 bp upstream of the Protospacer Adjacent Motif (PAM). On-target sites and predicted off-targets with <=2 mismatches were tiled with 16 baits per site for increased sensitivity in translocation analysis, while other off-targets were tiled with eight baits per site. Bait tiling balanced coverage for plus and minus strands upstream and downstream of predicted cut sites. Final bait balancing was calculated by SureDesign software. Two sets of SureSelect capture baits were synthesized for on- and off-target sites in the Cas9 or ABE quadruple editing scenarios. Multiplex gene-edited cells from conditions (No EP, ABE8e mRNA Mock, ABE8e mRNA 4KO, Cas9SpyFI Protein Mock, and Cas9 SpyFi Protein 4KO) from three independent donors (ND365, ND567, ND584) were used for this experiment. DNA was isolated at the end of expansion (7 days post-electroporation) using the DNeasy Blood & Tissue Kit (Qiagen) and quantified with the Qubit 2.0 Fluorometer (Life Technologies). Library preparation was performed using the SureSelect XT HS2 DNA System (Agilent Technologies) with 200 ng of DNA per sample following the manufacturer's instructions. Primary DNA libraries were qualified and quantified using the Agilent 2100 Bioanalyzer system. Up to 1,000 ng of prepared DNA libraries were hybridized to custom SureSelect capture probes for target enrichment, followed by qualification, quantification, pooling, and sequencing in a 2x150 bp paired-end (PE) configuration on a NovaSeq system (Illumina). Sequencing reactions and adapter trimming were conducted at GENEWIZ, Inc. (South Plainfield, NJ, USA). For INDEL and base edit identification, paired-end reads were aligned using BWA-mem. Consensus read pairs were used when groups of read pairs shared the same mapped start and end coordinates. For Cas9, % indel was calculated as % read pairs with indels within +/-5 bp of the cut site. A site was considered edited if edits exceeded 2% in a Cas9-edited sample while remaining <1% in controls. For ABE, % base edit was calculated as % read pairs with A->G edits within the ABE editing window (5-15 bp upstream of PAM). A site was considered edited if base edits exceeded 2% in ABE-edited samples while remaining <1% in controls. Translocation events were identified when paired reads mapped to different loci.