A signature-agnostic test for differences between tumor mutation spectra reveals carcinogen and ancestry effects

TL;DR

A new statistical test detects differences in tumor mutation patterns, revealing effects of carcinogens and ancestry that are often missed by traditional methods.

Contribution

The paper introduces a signature-agnostic metric called AMSD to detect significant differences in mutation spectra between groups.

Findings

AMSD identifies mutation spectrum shifts in 11 out of 20 carcinogen-exposed mouse studies.

Ancestry is associated with distinct mutational patterns in human tumors across ten cancer types.

AMSD complements signature analysis by detecting environmental and genetic influences on mutagenesis.

Abstract

Despite dozens of tools to identify mutational signatures in cancer samples, there is not an established metric for quantifying whether signature exposures differ significantly between two heterogeneous groups of samples. We demonstrate that a signature-agnostic metric - the aggregate mutation spectrum distance permutation method (AMSD) - can rigorously determine whether mutational exposures differ between groups, a hypothesis that is not directly addressed by signature analysis. First, we reanalyze a study of carcinogen exposure in mice, determining that eleven of twenty tested carcinogens produce significant mutation spectrum shifts. Only three of these carcinogens were previously reported to induce distinct mutational signatures, suggesting that many carcinogens perturb mutagenesis by altering the composition of endogenous signatures. Next, we interrogate whether patient ancestry has a measurable impact on human tumor mutation spectra, finding significant ancestry-associated differences across ten cancer types. Some have been previously reported, such as elevated SBS4 in African lung adenocarcinomas, while some have not to our knowledge been reported, such as elevated SBS17a/b in European esophageal carcinomas. These examples suggest that AMSD is a robust tool for detecting differences among groups of tumors or other mutated samples, complementing descriptive signature deconvolution and enabling the discovery of environmental and genetic influences on mutagenesis.

A statistical test identifies significant differences in mutational biases between groups of tumors, revealing carcinogen- and ancestry-associated effects that are often missed by standard mutational signature analysis approaches.

Figures