Abstract 3732: Drug signaling network-based prediction of therapeutic response to anti-cancer agents

Abstract

Abstract Background: A large number of biomarkers that are potentially predictive of a patient's clinical response to various anti-cancer agents have been reported in recent cancer therapeutics. These biomarker models, however, have not been able to provide a comprehensive understanding on why a specific drug works or does not work for different patients even with the same type of tumor. In our current study, we demonstrated that drug signaling network-based biomarker models could not only predict such therapeutic responses but also provide more direct insight on drug signaling gene networks of individual patients. Methods: In vitro drug activity and gene expression data of 60 NCI and 106 GSK cancer cell lines were first used to identify candidate drug sensitivity biomarkers for Paclitaxol, a standard chemotherapeutic agent, and erlotinib, a targeted therapeutic EGFR inhibitor. Next, known functional biomarkers related to each drug's molecular mechanisms of action were obtained from literature and then examined and filtered for their expression correlations with each drug activity using the above cell line data. Two biomarker sets of each drug were combined to discover gene-network modules by a hierarchical clustering analysis. We then developed candidate multivariate gene-network models and identified the final models that were highly predictive of drug activities on the cell lines. The final signaling network signature models were independently applied to three human patient sets treated with relevant anti-cancer agent therapies. Results: For the application of Paclitaxol in breast cancer, we derived a highly predictive gene-network signature model of 8 taxane signaling pathways including ATP-binding cassette transporters and tumor progression mechanisms. Its prospective applications to two independent human datasets of 133 and 100 breast cancer patients treated with a neoadjuvant combination chemotherapy with Paclitaxol were highly significant in stratifying patients with a complete response against the ones with poor or no clinical response (two-sample t-test p-value=0.048, 5.3E-06, respectively). Our gene network model of erlotinib signaling pathways was also made with 16 sub-networks including drug metabolism and tumor proliferation mechanisms for predicting this drug's sensitivity in metastatic colorectal cancer patients. Predictive scores of this model for 43 colon cancer patients with K-RAS wild type treated with cetuximab, another EGFR inhibitor, were again highly associated with heterogeneous clinical responses of patients (Jonckheere-Terpstra ordered difference test p-value=0.03). Conclusions: Drug signaling network-based biomarker models were highly predictive of taxane and cetruximab therapeutic responses of patients with breast and colorectal cancer providing a better understating on individual patients’ molecular mechanisms of drug response. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 101st Annual Meeting of the American Association for Cancer Research; 2010 Apr 17-21; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2010;70(8 Suppl):Abstract nr 3732.