The presentation can be viewed by creating an account here: http://theleadingstrand.cshl.edu/activate/160970/2017/GENOME
Expanding discovery from cancer genomes by integrating network analyses with in vivo tumorigenesis experiments
Heiko Horn 1,2 , Michael Lawrence 2,3 , Candace Chouinard², Yashaswi Shresta², Jessica Hu 1,2 , Elizabeth Worstell 1,2 , Emily Shea², Nina Ilic 2,4 , Ejung Kim 2,4 , Atanas Kamburov 2,3 , Alireza Kashani 1,2 , William Hahn 2,4 , Joshua Campbell 2,5 , Jesse Boehm², Gad Getz 2,3 , Kasper Lage 1,2
1 Massachusetts General Hospital, Department of Surgery, Boston, MA, ²Broad Institute of MIT and Harvard, Cancer Program, Cambridge, MA, 3 Massachusetts General Hospital, Department of Pathology, Boston, MA, 4 Dana Farber Cancer Institute, Department of Medical Oncology, Boston, MA, 5 Boston University, Department of Medicine, Boston, MA
Gene-based statistical tests to find cancer genes look for increased rates of somatic mutations or genomic copy number changes in cancer genomes. However, considerable sample sizes are required to find driver genes with intermediate or low mutation frequencies, and additional cancer genes remain to be discovered. Previous analyses have shown that cancer mutations in some cases converge on specific functional genomics sub networks. This suggests that mutations in a genes’ functional network can be predictive of whether it is a cancer gene itself. However, this hypothesis has never been systematically explored across hundreds of known cancer genes, tens of tumor types, and thousands of cancer genomes. More importantly, analyses of cancer gene networks have not previously been coupled to systematic experimental validation assays and their predictive power to provide new insight into tumor biology remains unclear. We develop a statistic (NetSig) that combines molecular protein network information and existing cancer sequencing data to identify genes with a significantly mutated gene network (excluding data on the gene itself). We apply NetSig to data from 4,742 tumors spanning 21 tumor types and identify known and recently proposed driver genes in most (~ 60%) tumor types. NetSig also identifies 62 other genes with a significantly mutated gene network many suggesting new cancer biology. We test 25 known driver genes (positive controls), 33 NetSig candidates, and 79 random genes (random controls) in a massively parallel in vivo tumorigenesis cell assay. We demonstrate that the NetSig candidates induce tumors at rates that are comparable to the known driver genes and eightfold higher than random genes when injected into mouse models. Guided by the NetSig results and functional validation experiments, we looked for mutations and copy number changes in these genes that could explain 242 (out of a total of 660) lung adenocarcinomas without any known driver event; the analysis identified significant amplifications of several NetSig candidates in this patient subgroup. Overall, we present an integrated workflow that complements gene-based statistical tests by combining molecular network information, cancer sequencing data, and in vivo tumorigenesis assays to find and validate new driver genes in existing cancer genome data. The framework we describe is scalable to the rapid production of data and should become increasingly powerful as more tumors are sequenced in the future.