Comprehensive Integrated Genomic Perturbations Reveal Molecular Mechanisms of Red Blood Cell Trait Associations

Discovery of molecular mechanisms responsible for trait associations as discovered by genome-wide association studies (GWAS) is hampered by difficulty in identifying causal genetic variants due to linkage disequilibrium. Typical assays of genetic function are low throughput or evaluate sequences in heterologous ectopic settings. Genome editing enables perturbation of trait-associated genetic sequences within relevant genomic, chromatin and cellular context. Here we perform comprehensive analysis of genetic variants associated with red blood cell traits by pooled CRISPR screening. We performed a genome-wide Cas9 gene knockout screen in immortalized erythroid precursors (HUDEP-2 cells) during erythroid maturation to define functional erythroid genes required for cell growth or differentiation. We evaluated 952 loci associated with nine red blood cell traits (Astle et al, Cell 2016) comprising 24,843 SNPs. We linked 7,187 (28.9%) of these SNPs to genes by at least one of four routes: sharing topological associated domain, physical proximity (<20 kb), long-range chromatin interaction (promoter HiC), or eQTL with a functional erythroid gene. We designed ~5 guide RNAs per SNP requiring cleavage position within at least 50 bp and exceeding an off-target score threshold, resulting in 32,710 sgRNAs testing 5,592 SNPs at 481 loci. We utilized four editors: Cas9 nuclease to produce indels, dCas9-VP64 for gene activation, dCas9-KRAB for gene repression, and dCas9 as a DNA targeting control. By pooled lentiviral transduction, erythroid differentiation culture, and guide RNA library deep sequencing, we found reproducible results across biological replicates, with guide count clustered by Cas9 protein type. We performed fine-mapping of association results by Bayesian inference to calculate posterior probability of inclusion (PPI). We found a strong correlation between PPI and CRISPR significance score indicating agreement between CRISPR screening and genetic fine-mapping, despite examples of validated CRISPR signals with low PPI scores. We identified numerous CRISPR-implicated functional SNPs at regulatory elements including promoters and enhancers but also at noncoding sequences lacking chromatin marks. The editing of CRISPR -implicated functional SNPs correlated well with expression of linked genes. Editing of CRISPR-implicated functional SNPs caused altered proliferation and/or differentiation of both HUDEP-2 cells and CD34+ HSPC-derived primary erythroid precursors. We validated a functional SNP at the BCL2L1 enhancer, where both Cas9 disruption and dCas9-KRAB inhibition resulted in decreased expression of BCL2L1,and reduced cell survival during erythropoiesis. We validated several CRISPR-implicated functional SNPs at GFI1B that controlled erythroid cell differentiation, including one at a distal enhancer element and another that impacted exon splicing. These results demonstrate the potential and challenges of comprehensive integrated genomic perturbation to complement genetic, biochemical, and statistical approaches to uncover molecular underpinnings of human blood cell traits.