Abstract 360: Thousands of published cancer signatures and pathways can be collapsed into a handful of non-redundant gene programs: a TCGA pan-cancer analysis

Abstract

Abstract Background: The OMIC revolution has produced tens of thousands of published gene expression signatures, pathways, and modules associated with cancer biology or patient outcome. Thanks to the TCGA effort to systematically collect molecular data from multiple cancer types, it is now possible to compare these signatures within and across cancers, and to collapse them into a smaller set of non-redundant features. Methods: We scored 3602 TCGA samples representing 12 cancer types for expression of a compendium of 6,898 published gene expression prognostic signatures, co-expression modules, amplicons and pathways, using RNAseq expression profiles. 6477 of these features were from the database MsigDB, and the rest from the literature. To define non-redundant gene-programs, we applied a bimodality filter and used Weighted Gene Correlation Network Analysis (WGCNA) to aggregate the bimodal signatures and pathways into coherent clusters. Associations were assessed using standard statistical methods. Results: From the thousands of cancer signatures in our analysis we identified 22 non-redundant gene-programs, defined as groups of gene signatures with high correlation across the Pan-Cancer dataset. These gene programs represent many of the hallmarks of cancer, including sustained proliferative signaling and DNA repair; immune system signaling; altered glycolytic and fatty acid metabolism yielding resistance to stress; apoptosis evasion; epithelial to mesenchymal transition; co-opted squamous and ‘stemness’ developmental programs; adhesion and cell-cell communication programs, some using plasma membrane and lipid vesicles; and self-sufficiency in growth signals employing estrogen, EGF, and MYC. More than a third of the original signatures clustered together in the proliferation gene-program. In addition to cell cycle and proliferation pathways, this program includes prognostic signatures for lymphoma, bladder, lung, breast, ovarian and other cancers, suggesting that most prognostic signatures are sensing a common proliferative signal associated with cancer aggression. The presence of two dominant immune signals showed a clear divergence in immune signaling between T cell/B cell immune activation and interferon responsive signaling, with the former correlated with immuno-modulatory treatment targets like CTLA4 and PD1. Interestingly, basal, hypoxic, and squamous differentiation programs were associated with poor outcome in many cancer types; in addition, expression of a more novel program dominated by tumor suppressing miRNA associated with good prognosis. Conclusion: The compendium of gene expression signatures is correlated, and can be reduced to a small set of non-redundant gene-programs that mirror the cancer hallmarks and may provide opportunities for investigating cancer heterogeneity and treatment approaches beyond tissue of origin. Citation Format: Denise M. Wolf, Cheng Fan, Katherine A. Hoadley, Christina Yau, Artem Sokolov, TCGA Network, Josh Stuart, Charles Perou, Laura van ’t Veer. Thousands of published cancer signatures and pathways can be collapsed into a handful of non-redundant gene programs: a TCGA pan-cancer analysis. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 360. doi:10.1158/1538-7445.AM2014-360