Keywords: gene expression, drug repositioning, companion predictive biomarker, personalized medicine
Many cancer therapeutic agents have shown to be effective for treating multiple cancer types. However, it is highly challenging to introduce a novel drug approved for treating one cancer type to a different cancer type, especially when only a small portion of patients in the other cancer types can benefit from the drug. Many new agents are often introduced to different cancer types together with companion biomarkers. It is, however, still uncertain whether such biomarker models are directly predictive for therapeutic response in different cancer types. To tackle this challenging question, we have developed a gene expression biomarker prediction technique based on concordantly expressed biomarkers between different cancer types, namely CONCORD, to predict a new drug’s therapeutic response in a different cancer type. Our development of a CONCORD cross-cancer prediction model for a therapeutic agent consisted of six sequential steps: 1) discovery of initial drug sensitivity biomarkers on diverse cancer cell lines 2) identification of different cancer types to be concordantly predicted from the original cancer type 3) selection of three-way drug biomarkers concordantly expressed among cancer cell line, original, and new cancer sites 4) in vitro multivariate prediction modeling and model test on the original cancer type 5) validation on independent patient cohorts in the original cancer type and 6) once a drug’s CONCORD model is confirmed with its significant prediction performance in the original cancer type, the model will then be used to predict patient response to the drug in other cancer types in a prospective manner. In particular, we obtained and validated CONCORD gene expression signatures on more than 20 patient cohorts from diverse cancer types, simultaneously predicting a drug’s sensitivity and response in patients of the original and different cancer types. Applying CONCORD to standard chemotherapy agents for breast cancer treatment and targeted drugs recently used in multiple cancer types, we show its potential in assisting with novel drug introduction by selecting new cancer types as well as predicting therapeutic response in different cancer types.