Abstract P3-08-02: Pharmacogenomic Predictors of Patient Response to Chemotherapy Derived from Breast Cancer Cell Lines Combining Knowledge-Based and Data-Driven Methods

Abstract

Background: Responses of breast cancer patients to chemotherapy treatments vary considerably, and population treatment response rates remain low. To improve patient outcomes, genomic profiles have been used to identify patients who would benefit from specific treatments. Several studies have used cancer cell lines to develop pharmacogenomic predictors by identifying genes associated with drug response. However, pharmacogenomic predictors derived by this data-driven approach may not readily elucidate the underlying mechanisms associated with drug responses, because the identified predictors by computational methods may not directly associate with drug responses considering the complex genetic regulatory network. To overcome this issue, we proposed a strategy to integrate data-driven methods with biological knowledge-based approaches to identify genomic predictors. We then applied this approach to breast cancer cell lines to identify genomic predictors of paclitaxel-5-fluorouracil-doxorubicin-cyclophosphamide (TFAC), the identified predictors are then evaluated by their ability to predict the clinical outcome of the breast cancer patients who are treated by TFAC. Material: Thirty immortalized breast cancer cell lines were exposed to various concentrations of TFAC using a chemosensitivity assay. Area under the dose-response curves was calculated to measure chemoresponses. Gene expression profiles of the 30 cell lines, the expression profiles as well as the pathologic complete response (pCR) information of 133 breast cancer patients treated by TFAC were publicly available. Methods: We performed pathway enrichment analysis in breast cancer cell lines to assess the association between drug response and curated gene sets predefined by molecular signature database. Pathways with p-value less than 0.01 were considered enriched. The genes from the enriched pathways whose expression values were highly correlated with drug sensitivity were selected as the pharmacogenomic predictors. To validate these predictors, the performances of their prediction for patients’ pCR were evaluated using principle component regression method. Results: Using pathway enrichment analysis, 17 pathways were identified to be related to TFAC drug response. These pathways are related to different biological functions, including breast cancer ER status and BRCA type, immune response, differentiation, and drug response. Using supervised principal component regression, 59 genes involved in at least one of these 17 pathways were selected as genomic predictors. The prediction accuracy of patient pCR was 0.70, sensitivity was 0.71, and specificity was 0.70. Conclusion: By combining knowledge-based and data-driven methods, we have identified 59 genes from breast cancer cell lines as pharmacogenomic predictors of drug response to TFAC. These results support the viability of using breast cancer cell lines to predict breast cancer patient response to chemotherapy. Further functional study of these pharmacogenomic predictors will extend our understanding of individual drug response and facilitate personalized treatment. Citation Information: Cancer Res 2010;70(24 Suppl):Abstract nr P3-08-02.