"Right drug for the right patient": hurdles and the path forward in colorectal cancer.

Predictive biomarkers have been heralded as the way to develop the "right drug for the right patient." However, despite many studies incorporating novel biomarkers with targeted therapies, there has been little progress over the 5 years since the identification of KRAS mutations' ability to predict resistance to epidermal growth factor receptor (EGFR) monoclonal antibodies. Recently approved therapeutics (regorafenib, aflibercept) or label extensions for existing therapies (bevacizumab) lack companion biomarkers. The current model of biomarker development, "target-based biomarker" design, attempts to identify individual biomarkers that are closely tied to the activity of a particular treatment. There are several limitations to prospective utilization of predictive biomarkers in novel therapy development, including technical validation of the assay and the logistics of timely biomarker determination with available material that limit the options. Tumor heterogeneity, both between different regions in the tumor and as a result of changes induced over time and under the selective pressure of chemotherapy, can reduce the precision of biomarker determination. Biomarkers present in low frequencies are increasingly common in drug development and will require efficient screening infrastructure to be feasible. Although development efforts will continue in the current target-based biomarker model for the near future, it is increasingly apparent that a new model is needed. A "taxonomy-based biomarker" model has been proposed, which is less tied to novel drug development and instead attempts to classify individual tumors based on their intrinsic biology. This requires integrating multiple characteristics of the tumors, including gene mutations, amplifications, methylation, as well as RNA and protein expression. Identification of the taxonomy of colorectal cancer will then allow more efficient development of targeted agents that can leverage the distinct molecular vulnerabilities of the resulting subsets. A transition to a taxonomy-based biomarker model would provide the classification structure and biologic insights needed to advance the ultimate goal of the right drug for the right patient.