Abstract 5322: Collections of simultaneously altered genes as highly predictive markers of cancer cell drug response

Abstract

Abstract Introduction: Many genetic markers of drug response have been reported in cancer pharmacogenomics studies. However, most of these markers fail to achieve clinical utility because they capture response in only a small fraction of the patient population; this shortcoming might result from a reliance on single-gene markers in contemporary personalized medicine. While there is growing evidence that therapeutic response can be modulated by the concerted impact of multiple genetic alterations, the combinatorics associated with many-body interactions has mostly prohibited the discovery of multi-gene biomarkers, even using computational methods. We developed a computational approach, MOCA (Multivariate Organization of Combinatorial Alterations), to partially address these challenges. Extending our method that accounts for pairwise interactions, MOCA combines many genomic alterations into biomarkers of drug response, using Boolean set operations coupled with optimization; in this framework the union, intersection, and difference Boolean set operations are proxies of molecular redundancy, synergy, and interference. The algorithm is fast, broadly applicable to genomics data, of immediate utility for prioritizing cancer pharmacogenomics experiments, and recovers known clinical findings without bias. Furthermore, results from this work connect many important, previously isolated observations. Major Findings: When applied to 416 pharmacogenomically characterized cancer cell lines from the Cancer Cell Line Encyclopedia, MOCA identified many known and potential markers of drug response. For example, correlation of ERBB inhibitor response drastically increased when considering EGFR (ERBB1), ERBB2, ERBB3, ERBB4, and KRAS alterations in a single marker. Similarly, a marker combining IGF1, IGF1R, and RAD51 alterations drastically increased correlation with IGF1R inhibitor response, relative to any of these three genetic markers considered in isolation. This approach is also powerful for determining subsets of gene-specific mutations that increase correlation with drug response. For instance, MOCA accurately captures the differential EGFR inhibitor response conferred by known activating EGFR mutations. Similarly, we find specific HDAC1 mutations cooperate with HDAC5 overexpression to potentiate cells to the HDAC inhibitor panobinostat. Additionally, considering all pairwise gene-drug interactions, MOCA recovers known and compelling correlations, including: RTK inhibitor resistance via c-MET, EGFR, ERBB2, and PDGFRB kinase switching; mutual exclusivity of TP53 mutation and response to the MDM2 inhibitor nutlin-3; greater nutlin-3 potentiation via MDM4, rather than MDM2, overexpression; MEK and RAF inhibitor response in BRAF mutated cell lines; and MEK inhibitor potentiating NRAS mutations. Masica, David L., & Rachel Karchin. Cancer Research 73 (2013): 1699-1708. Citation Format: David L. Masica, Rachel Karchin. Collections of simultaneously altered genes as highly predictive markers of cancer cell drug response. [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 5322. doi:10.1158/1538-7445.AM2014-5322