Abstract LB258: Branching cancer evolution biases the detection of epistatic interactions between driver mutations

Abstract

Abstract Cancer genomics datasets can be leveraged to decipher evolutionary dependencies, an important frontier for precision oncology. Phylogenetic analyses time mutations as occurring early (clonal) or late (subclonal) in the life history of a tumour. A separate class of computational tools aims to systematically detect mutually exclusive (ME) relationships, revealing evolutionary dependencies including synthetic lethal or functionally redundant interactions between mutations. Existing tools for ME detection take binary patient-level mutation matrices as input, ignoring the branching clonal architecture that characterizes most cancer genomes. This may result in a bias to detect clonal mutations. Moreover, ME interactions between subclonal mutations in separate lineages of a tumor's phylogenetic tree may be overlooked. Here, we examine the influence of mutation clonality on ME detection in a pan-cancer analysis. Whole exome-sequencing data was accessed from the TCGA data portal for 3,545 patients from nine major cancer-types. Nonsense and missense mutations in driver genes were filtered. PyClone was used to estimate the cancer cell fraction (CCF) for each variant. To detect significant (P<0.05) ME interactions, the DISCOVER and TiMEx tools were used.First, focusing on patients with lung adenocarcinoma (LUAD), the median CCF value for each gene across the cohort was used as a summary clonality metric, indicating that 58% (106/182 genes) of the driver genes assessed in LUAD tend to be clonally mutated (median CCF = 1). Across nine cancer-types, on average 58% (range=25-68%) of mutations were classified as clonal. In parallel, the DISCOVER and TiMEx tools for ME detection were applied to the LUAD dataset. Clonally mutated genes comprised the majority of the significant interactions detected in LUAD tumours for both tools: DISCOVER=89% (107 of 120), TiMEx=77% (127 of 166). This bias towards detecting clonal mutations was the general pattern across cancer-types: DISCOVER median=81% (range = 38-93%); TiMEx median=77% (range=48-81%). Using the DISCOVER tool in the LUAD cohort, the tendency of a gene to be clonally mutated exhibited a highly significant positive correlation with the proportion of significant findings in ME analysis (P=0.00685). The TiMEx tool, which uses a different underlying model and set of assumptions, also identified significant positive associations in LUAD (P=0.0251), demonstrating that the influence of mutation clonality on ME detection is not specific to a particular tool. Overall, significant positive correlations between mutation clonality and the number of statistically significant ME interactions were found for 78% (7/9) of cancer-types using DISCOVER, and 44% (4/9) using TiMEx. Our results suggest mutation clonality might represent an unaddressed confounding factor for the detection of significant ME interactions across cancer-types. Addressing this bias, in future, could help define a core set of evolutionary dependencies. Citation Format: Dhruva Biswas, Nicolai J. Birkbak, Javier Herrero, Charles Swanton. Branching cancer evolution biases the detection of epistatic interactions between driver mutations [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr LB258.