Determining therapeutically actionable genetic interactions in human cancer at scale using multiplexed CRISPR screening

Abstract

Synthetic lethality (SL) is a compelling yet underutilised strategy in precision oncology, enabling the selective targeting of gene loss in cancer. This study employed a high-throughput multiplexed combinatorial CRISPR/Cas9 pooled screening approach to systematically identify SL interactions. A novel tRNA-based dual guide expression system was used to construct the Synergy library, enabling comprehensive and unbiased monogenic and digenic perturbation of 931 prioritised gene pairs. These pairs were identified through two large-scale bioinformatic analyses. The first leveraged mutual exclusivity and co-occurrence analysis of genetic alterations across thousands of sequenced cancer genomes (Chapter 2), while the second utilised a machine-learning classifier to predict SL among paralog pairs, prioritising those with clinical relevance and therapeutic tractability (Chapter 3). Pooled screens were conducted using the Synergy library across 12 cancer cell lines, including those derived from the colon, kidney, stomach, and biliary tract, ensuring the identification of interactions robust to genetic heterogeneity. We identified 52 strong, robust SL interactions, all involving paralog pairs, suggesting that mutual exclusivity of genetic alterations is not primarily driven by SL. Screen findings were further validated through low-throughput competitive growth assays, confirming the reproducibility and robustness of the Synergy screens. A key finding was the strong correlation between the machine-learning classifier predictions and experimental results, confirming its utility in computationally prioritising SL paralog pairs for functional validation. Among the identified interactions, three high-value paralog pairs (ESCO1/ESCO2, CDS1/CDS2, and DOCK1/DOCK5) were prioritised for further characterisation. Rescue experiments in cell models with inherent loss of one paralog confirmed that this loss creates a dependency on the remaining paralog, reinforcing the SL relationship and demonstrating the feasibility of therapeutic targeting. However, drug experiments revealed the challenge of developing truly selective inhibitors for highly similar paralogs, as the tested DOCK1 inhibitor failed to recapitulate the genetic perturbation phenotype, highlighting the need for more precise targeting strategies. Future work will employ base editing mutagenesis screens to delineate functionally critical regions within SL paralogs, providing increased resolution for the development of selective inhibitors. Collectively, this study maps genetic interactions at scale, providing a solid foundation for translational efforts to exploit paralog-targeted SL strategies in cancer treatment.