Cancer chemosensitivity prediction of drug combinations using single drug responses and microarray profiling data

B46

Currently, the choice of therapy for metastatic cancer remains empiric due to the lack of chemosensitivity prediction for the available combination chemotherapeutic regimens. Here we present a novel approach that identifies chemosensitivity biomarker genes as predictors of tumor growth inhibition for three widely-used chemotherapeutic agents: Cisplatin, Paclitaxel, and Gemcitabine. Measuring cell proliferation responses 72hr after-treatment with each of these three agents on 40 human urothelial cancer cell lines, we obtained dose response curves and GI50 (50% of growth Inhibition) estimates. Based on these single drug response information and genome-wide gene expression profiling data of the 40 cell lines, we applied the MiPP (misclassification penalized posterior) approach, to discover molecular prediction models for these single drug responses. These MiPP-identified prediction models, each with a small number of gene biomarkers, predicted the responses of human bladder cancer cell lines to each of the three agents with a sensitivity of 0.93 to 0.96. Based only on these single-drug prediction models, we developed an in silico approach to predict responses to 3 clinically used two drug combinations of the aforementioned agents. We evaluated these predictions in vitro and found that 80% of 15 randomly-chosen cell lines were correctly predicted for their responsiveness to combination treatment (p=0.0002). Together, our results suggest for the first time that chemosensitivity to drug combinations can be accurately predicted based on gene expression profiling. Furthermore, this study provides the framework for the evaluation of predicted drug sensitivity in patients undergoing combination chemotherapy for cancer, based on gene expression profiles. If validated, this approach will lead to a breakthrough in personalized cancer therapy.