Abstract A130: Joint cell-line (CCLE, Sanger) and patient (TCGA) modeling of drug sensitivity reveals novel molecular biomarkers for targeted therapy and conventional chemotherapy.

Abstract

Abstract Background: The unresponsiveness to anticancer drugs in patients outlines the need to identify novel and robust biomarkers of response to therapy. The recent release of large molecular cell line datasets labeled for drug sensitivity enables the development of such predictors. Cell line based models poorly reflect the extent of molecular heterogeneity and of tumor microenvironment relationships existing at the patient level. We aimed to overcome these limitations by combining cell line datasets with patient cohorts to uncover novel molecular traits associated with drug response at the patient level. Methods: For building our model, we used the following publicly available datasets: CCLE (1057 cell lines, 24 drugs), Sanger (790 cell lines, 138 drugs) and TCGA (patients: breast cancer, colorectal cancer, lung squamous carcinoma, lung adenocarcinoma, ovarian cancer, kidney cancer and melanoma). A joint semi-supervised trained Elastic Net approach was used to model drug sensitivity (both conventional chemotherapy and targeted therapy) using gene-expression, copy-number alteration, and somatic mutation data. We introduced a false discovery rate estimation for selecting significant aberrations. As validation, we used pooled RNAi screens (NKI, Netherlands) and also a dataset of 120 patients prospectively enrolled in the ongoing MOSCATO molecular screening program (Gustave Roussy, France) (presented at ASCO 2013). Results: Drug sensitivity models trained in the entire set of cell lines performed better than models trained using tissue specific cell lines, whatever the drug and the tissue type. Novel molecular traits (mutation, copy number variation) were identified by selecting highly performant models (AUC > .75) across drug and tissue type combinations. We were able to recover all of the recognized landmark biomarkers across drug - tissue combination (e.g. Erlotinib: EGFR mutation in lung adenocarcinoma; Lapatanib: ERBB2 amplification in breast cancer; Vemurafenib: BRAF mutation in melanoma, etc). Moreover, we uncovered novel molecular traits (copy, number mutations) associated (FDR<25%) with drug response (e.g. Erlotinib: CDKN2A deletion in lung adenocarcinoma; Lapatanib: CDH1 deletion in breast cancer; Vemurafenib: FH amplification in melanoma, etc.). The validation of our framework in the pooled RNAi screens of colorectal and lung adenocarcinoma cell lines and in the MOSCATO patient's cohort will be presented. Conclusions: We provide a comprehensive annotation of mutations and copy-number events in tumors predicted to be associated with sensitivity and resistance in 142 anti-cancer drugs and across 7 tumor types. These molecular traits can be used to refine the results of large scale functional genomic experiments such as RNAi screens and ultimately in the early clinical trial setting. Citation Information: Mol Cancer Ther 2013;12(11 Suppl):A130. Citation Format: Charles Ferté, Elias Chaibub Neto, Chong Sun, Cecilia Noecker, Frederic Commo, Olga Nikolova, In Sock Jang, Mehmet Gonen, Benjamin Besse, Fabrice André, Eric Angevin, Ludovic Lacroix, Clement Mazoyer, Antoine Hollebecque, Christophe Massard, Adam Margolin, Roderick Beijersbergen, Stephen Friend, Rene Bernards, Jean-Charles Soria, Justin Guinney. Joint cell-line (CCLE, Sanger) and patient (TCGA) modeling of drug sensitivity reveals novel molecular biomarkers for targeted therapy and conventional chemotherapy. [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference: Molecular Targets and Cancer Therapeutics; 2013 Oct 19-23; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2013;12(11 Suppl):Abstract nr A130.