Ultra-deep sequencing validates safety of CRISPR/Cas9 genome editing in human hematopoietic stem and progenitor cells

Abstract

As CRISPR-based therapies enter the clinic, evaluation of safety remains a critical and active area of study. Here, we employ a clinical next generation sequencing (NGS) workflow to achieve high sequencing depth and detect ultra-low frequency variants across exons of genes associated with cancer, all exons, and genome wide. In three separate primary human hematopoietic stem and progenitor cell (HSPC) donors assessed in technical triplicates, we electroporated high-fidelity Cas9 protein targeted to three loci (AAVS1, HBB, and ZFPM2) and harvested genomic DNA at days 4 and 10. Our results demonstrate that clinically relevant delivery of high-fidelity Cas9 to primary HSPCs and ex vivo culture up to 10 days does not introduce or enrich for tumorigenic variants and that even a single SNP in a gRNA spacer sequence is sufficient to eliminate Cas9 off-target activity in primary, repair-competent human HSPCs.

Fig. 1. Experimental design and confirmation of on-target activity.

a Experimental design: CD34⁺ HSPCs from 3 donors underwent gDNA harvesting at d0 (to establish germline variants) and were then subject to mock electroporation or Cas9 treatments with gRNAs corresponding to sites at AAVS1, HBB, and ZFPM2. Cells were cultured and gDNA was harvested again at d4 and 10 post-editing. b Predicted off-target cut site (OT1) of ZFPM2 guide in exon 5 of the EZH2 gene, based on sequence homology. Mismatch in gRNA is shown in red. c On-target activity of AAVS1, HBB, and ZFPM2 gRNAs determined by PCR amplification of the genomic region surrounding the predicted cut sites followed by Sanger sequencing and analysis of indels by TIDE 4 days post-editing. Bars represent median. N = 3 separate HSPC donors.

Fig. 2. Summary of TSO500 sequencing data.

a MEC for each treatment for each donor. Treatments are Mock electroporated (-), AAVS1- (A), HBB- (H), and ZFPM2-targeted (Z). Individual points represent technical replicates. Columns and error bars represent mean and standard deviation. Dotted line indicates recommended lower specification limit, set at 1300 MEC. b Number of reproducible variants across technical replicates from total called by treatment group. Columns represent mean variants called for the three donors within each treatment. c VAF x MEC for all variants found among technical replicates for Cas9 treatments for each donor at d10. Large white points are those that remained after removing germline and synonymous variants. Large black points are those that remain after removing variants present in Mock within each donor.

Fig. 3. Variants identified in Cas9 treatments.

a VAF x MEC for variants remaining from all Cas9 treatments after removal of synonymous, germline, and Mock calls. Donor 1 had no variants remaining after filtering. b Percent indels in EZH2 identified by PCR amplification, Sanger sequencing, and TIDE analysis using d4 gDNA. Donor 1 had no detectable activity. c Mosaic plot of genes harboring mutations within each donor and Cas9 treatment at d4 and 10. Area is proportional to the number of times variants were called in a particular gene within a particular treatment group. Filtering removed germline and synonymous variants. For each donor and timepoint, conditions are ordered as Mock, AAVS1, HBB, and ZFPM2 (-, A, H, and Z, respectively). d Sanger chromatograms at predicted EZH2 off-target site. The PAM site and spacer are depicted as blue and red lines, respectively. Homozygous SNP in Donor 1 abrogated detectable editing activity.

Fig. 4. Variants identified by whole-exome sequencing.

a VAF x Coverage for all variants called by exome sequencing pipeline in Mock and AAVS1 d10 conditions. b VAF for variants called by tumor-normal pipeline when AAVS1 is used as tumor and Mock as normal inputs (left panel), and when Mock is used as tumor and AAVS1 as normal (right panel). c VAF x Coverage for 137 variants shown in b. 30 large white points are those that remained after removing variants with Mock VAF > 0.01. d Annotation for all 30 AAVS1 variants. Homology to gRNA is defined as 10 or more matches to spacer+PAM within 20 bp upstream or downstream of variant.

Fig. 5. Variants identified by whole-genome sequencing.

a VAF x Coverage for all variants called by WGS pipeline at d3 in AAVS1 treatment with Mock as background input. Large white points depict on-target AAVS1 variants. b AAVS1 VAF x Mock VAF for all called variants in AAVS1 treatment Large white points indicate on-target AAVS1 variants. c VAF x Coverage for all variants depicted in a after subtraction of Mock VAF from AAVS1 VAF. Large white points depict on-target variants. d AAVS1 read depth x Mock read depth for all called variants in AAVS1 treatment. e VAF x Coverage for all variants depicted in a after filtering to remove calls below 0.1% VAF in the AAVS1 treatment and VAFs > 1% in the Mock sample. Large white points depict on-target variants. f Mock VAF x AAVS1 VAF for all called variants in Mock treatment (i.e., in Mock treatment with AAVS1 as background input).