Abstract 385: Network-driven discovery of cancer drivers and pathways using 2,500 whole cancer genomes

Abstract

Abstract Cancer driver genes exhibit unexpectedly high mutation rates in large cancer genomic datasets. We hypothesize that driver mutations specifically alter molecular interaction networks by disrupting “active sites” - interaction interfaces in proteins and DNA. We present ActiveDriverWGS, a novel computational method to discover cancer drivers in whole-genome sequencing (WGS) data. ActiveDriverWGS finds genome regions that are significantly enriched in somatic single nucleotide variants (SNVs) and indels and ascertains whether these associate to known active sites. Analysis of active sites allows us to predict the mechanisms of mutation on three layers of the central dogma: regulatory DNA with transcription factor (TF) binding sites (TFBS), mRNA with microRNA binding sites in untranslated regions (UTRs), and post-translational modification (PTM) sites in proteins. To discover cancer driver genes and pathways, we analysed the WGS dataset of >2,500 samples from the International Cancer Genome Consortium (ICGC) Pan-Cancer Analysis Working Group (PCAWG). We found 61 protein-coding candidates with 34 known drivers (P=10-40), validating the high accuracy of our method. 40 genes have significant mutations of PTM sites, suggesting that rewiring of PTM signalling networks is a common oncogenic mechanism. For example, the BRAF V600E SNV flanks two phosphorylation sites and one ubiquitination site (FDR P=10-44), a novel interpretation and potential avenue for precision therapies targeting the kinase and ubiquitin network of BRAF. In the non-coding genome, we detected known lncRNAs (NEAT1, MALAT1), promoters (TERT, WDR74) and novel candidates with mutation enrichment. For example, an enhancer on chr6 has a mutation hotspot in 33 patients (FDR P=10-19), with 20 SNVs affecting binding motifs of cancer-associated TFs FOXO3, SOX2, HMGA2 (FDR P=10-10). Thus our method discovers non-coding drivers and their candidate mechanisms in a single analysis. Our ActiveDriverPW method extends coding and non-coding mutations to biological pathways. We found >600 mutation-enriched pathways in the PCAWG pan-cancer dataset. Of these ~200 are also significant when only non-coding mutations are analysed, showing that the non-coding genome includes previously unstudied mutations in pathways. The DNA double-strand break response pathway (FDR p=10-10) includes non-coding SNVs in ~20 histones and chromatin modifiers, such as the demethylase KDM4B with 46 SNVs in its promoter and enhancers. ActiveDriverPW maps mutations of the long tail that affect genes in hallmark cancer processes yet remain undiscovered in gene-focused analyses. Our methods accurately capture known drivers in the ICGC-PCAWG dataset and suggest specific mechanistic details. Our benchmarks also emphasize the robust performance of our methods. ActiveDriverWGS and ActiveDriverPW are valuable additions to the toolbox for cancer genome analysis. Citation Format: Jüri Reimand. Network-driven discovery of cancer drivers and pathways using 2,500 whole cancer genomes [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 385. doi:10.1158/1538-7445.AM2017-385