TCGA_DEPMAP– Mapping Translational Dependencies and Synthetic Lethalities within The Cancer Genome Atlas

Abstract

The Cancer Genome Atlas (TCGA) has yielded unprecedented genetic and molecular characterization of the cancer genome, yet the functional consequences and patient-relevance of many putative cancer drivers remain undefined. TCGADEPMAP is the first hybrid map of TCGA patient dependencies that was built from 6,581 expression-based predictive models of gene essentiality across 791 cancer cell models within the Cancer Dependency Map (DEPMAP). TCGADEPMAP captured well-known cancer lineage and sub-lineage dependencies (e.g., ESR1, ERBB2, etc.), genetic drivers (e.g., KRAS, BRAF, etc.), and synthetic lethalities (e.g., STAG1/2, SMARCA2/4, etc.), demonstrating the robustness of TCGADEPMAP as a translational dependency map. TCGADEPMAP also unveiled novel synthetic lethalities that were experimentally confirmed using multiplexed CRISPR screening across multiple cancer cell lines. Other map "extensions" were built to capture unique aspects of patient-relevant tumor dependencies, including translating gene essentiality to drug response in patient-derived xenograft (PDX) models (i.e., PDXEDEPMAP) and predicting gene tolerability within normal tissues (GTEXDEPMAP). Collectively, this study demonstrates how translational dependency maps can be used to leverage the rapidly expanding catalog of patient genomic datasets to identify and prioritize novel therapeutic targets with the best therapeutic indices.

Mapping Genetic Interactions of All Paralogs in the Human Genome

“…Genetic mapping of all known human paralog pairs revealed relatively few synthetic lethalities that were common between cancer models, a phenomenon that has also been observed in more focused screening efforts [3,8,17]. One possible explanation is that most synthetic lethalities are complex polygenic interactions that are modified in different cellular contexts and by other endogenous factors [4].…”

“…Thus, the machine learning classifier used in the present study is distinct in that the training sets were not reliant on a limited representation of synthetic lethalities in the DepMap [7]. Secondly, our machine learning classifier was unique in that it used a weighted positive-only training set, which accounts for the widely variable penetrance of paralog synthetic lethalities that was reported here and elsewhere [8,14,17,19,27,28]. To our knowledge, this new classifier revealed for the first time that the overlap and essentiality of the PPI network that is shared by a paralog pair are the best predictors of the most penetrant synthetic lethalities.…”

“…Multiple studies have demonstrated a wide variability in the penetrance of digenic synthetic lethal interactions using focused CRISPR screens [8,14,17,19,27,28], yet interpreting these data have been difficult due to the limited overlap in the paralog sets and technologies used in each study [29]. Here, we screened all 36,648 paralog pairs in the human genome using a single gene-editing technology (enCas12a), which confirmed unambiguously that a minority of synthetic lethalities (~10%) are shared by two cancer models.…”

“…Log2 fold-changes (L2FC) of guide array abundance were calculated by comparing day 14 libraries with the plasmid library using limma-voom [34]. Genetic interactions (GI) were calculated by comparing the expected and observed L2FC of double and single knockout constructs following z-score transformation (zL2FC) [17]. Paralog pairs were filtered for expression (>2.5 TPM) and an observed activity of the median double knockout effect of <-1 zL2FC and >1 zL2FC for synthetic lethal and synthetic buffering pairs, respectively.…”