Introduction: Pancreatic ductal adenocarcinoma (PDAC) is the fourth leading cause of cancer deaths. PDAC is a broadly heterogeneous disease making it difficult to treat; each patient’s tumor is unique, but it can be classified into different subtypes, and identifying the right therapeutic strategy for each subtype is key to improving patient outcomes. Three sets of transcriptional signatures for PDAC subtype classification have been proposed in the literature; however, preliminary data from in vitro cell line models indicates that they are associated with sensitivity for only a handful of compounds. Computational methods can identify effective individual therapies by robustly analyzing genome-scale data from multiple experimental platforms to derive molecular classifiers predictive of drug response (drug signatures).

Methods: We developed a novel Bayesian framework approach called Gene-wise prior Bayesian Group Factor Analysis (GBGFA) for predicting drug sensitivity in a given sample from genomic data. Our approach integrates data from different omic platforms by modeling the statistical dependencies of measurements made for the same gene via a model parameter called “gene prior”. We identify gene signatures for the clinically relevant to PDAC compounds olaparib and palbociclib, leveraging data from 44 pancreatic cell lines. Using hybrid capture mutation, gene expression, and copy number data, we applied our model to score genes for their ability to predict drug response in pancreatic cell lines and compiled the top predictors into drug signatures.

Results: Our results across a pan-cancer cell line collection and >100 compounds show improved ability to recapitulate known biomarkers and drug targets compared to other state of the art methods. For the MEK1/2 inhibitor selumetinib, we successfully recapitulated BRAF mutation as the top-scored biomarker of sensitivity. In addition, we identified four genes that carry additional information that can be used to stratify samples into sensitive or resistant groups: all were wild type in the sensitive cell lines and mutated in some of the resistant lines. We hypothesize that these genes could be used in conjunction with BRAF mutation status to refine the stratification to selumetinib therapy. We also identified several highly expressed genes in the resistant cell lines, and we hypothesize that these genes may mediate drug resistance. Similar analysis are currently ongoing for olaparib and palbociclib, comparing our findings to the subtypes currently proposed in the literature.

Conclusions: Our integrated approach identifies drug-specific gene signatures in PDAC cell lines. We derived genomic signatures for two clinically relevant to PDAC compounds in the context of the established in the literature PDAC subtypes, and hypothesize that these signatures could be used to identify patients most likely to respond to these therapies.

Citation Format: Olga H. Nikolova, Laura Heiser, Adam A. Margolin. Genomic signatures for tumor-to-treatment stratification in pancreatic cancer [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 5548. doi:10.1158/1538-7445.AM2017-5548