Abstract 3299: High-throughput combinatorial CRISPR-Cas9 gene knockout reveals most genetic interactions are context dependent

• Published in: Cancer research (Chicago, Ill.) Ill.), 2018-07, Vol.78 (13_Supplement), p.3299-3299
• Main Authors: Shen, John Paul, Zhao, Dongxin, Munson, Brenton, Birmingham, Amanda, Sasik, Roman, Bojorquez-Gomez, Ana, Licon, Katherine, Klepper, Kristin, Beckett, Alex, Sanchez, Kyle Salinas, Mali, Prashant, Ideker, Trey
• Format: Article
• Language: English
• ISSN: 0008-5472; 1538-7445
• DOI: 10.1158/1538-7445.AM2018-3299

Genetic interactions, in particular negative or "synthetic-lethal" interactions for which simultaneous disruption of two genes causes cell killing, have implications for therapeutic development as has been demonstrated by the clinical success of PARP inhibitors specifically for tumors with loss-of-function mutations in BRCA1/2. However, further applications of synthetic-lethal cancer therapy have been limited by poor understanding of the important genetic interactions in a cancer cell, and how these vary from one cancer type to another or from patient to patient. To enable systematic mapping of these genetic interaction networks, we recently developed a CRISPR-Cas9 screening methodology for knocking out single and pairs of genes in high throughput. Critical to this method is the precise determination of single-gene knockout effects, which is accomplished by serial measurement of the relative changes in gRNAs at days 3, 14, 21 and 28 post-transduction. To robustly quantify gene fitness and genetic interactions we developed a novel computational analysis framework that integrates all samples across the multiple days of the experiment; with said method we achieve Pearson correlation of 0.95 or greater between biologic replicates in the same cell line (p < 1x10-30). Additionally we demonstrate that our analysis method is robust to the compositional effects inherent in a pooled knockout experiment. To facilitate reproducibility of analyses and distribution to the scientific community, the code has been packaged into a modular series of python notebooks freely available on github. Evaluating all pairwise gene knockout combinations among a panel of 73 genes divided between tumor-suppressor genes (TSG) and cancer-relevant drug targets (DT) in a total of 5 cancer cell lines from diverse lineages (HeLa, A549, 293T, U2OS, LN229), we identified 226 synthetic lethal and 14 epistatic interactions at a Z-score cut-off of -3 (FDR ~0.3). Of the synthetic lethal interactions 203 (89.8%) were private to a single cell line, and no interaction was seen in more than 3 of 5 five cell lines. Thus far 10 (out of 16 tested) therapeutically relevant interactions have been replicated in low-throughput assays using either combinatorial drugs or CRISPR knockout of a TSG paired with a drug (71% precision or positive predictive value). The cell line specificity of interactions was also confirmed in low-throughput assays (75% negative predictive value). In summary, we have discovered many