Cancer biomarker discovery: speed-bumps and tire shredders.

In this remarkable era of modern molecular biology we have been fortunate to be able to perform high throughput investigations of cancer-specific DNA sequence variations, RNA expression profiling, proteomics, and metabolomics with tremendous expectations for their application to the early detection and treatment of human cancer. The ease of obtaining such high dimensional data and the throughput of these approaches to discover cancer-specific changes in various molecular moieties belie the difficulties in the full development and implementation of these molecules for clinic-ready biomarker tests. The problems can be rooted in the high-dimension data sets that may be over-fitted in the discovery phase of this type of study (too many analytes and too few specimens) or they may be related to the nature of the samples used in discovery and validation that may not properly represent the true clinical implementation. Few recently developed molecular diagnostic tests have passed the regulatory hurdles to FDA-approval for the pre-symptomatic diagnosis of cancer [2] and many tests in use today might not pass in the current regulatory environment especially given the recognition of the over-diagnosis and over-treatment of otherwise indolent cancers [4]. The field of cancer biomarkers is grounded in the principle that if we can identify individuals with cancer at its earliest known stage we can reduce the morbidity and mortality from the disease. Through this credo of early detection as a means to reduce cancer mortality we can look back with confidence at cervical cancer and the reduction of mortality from this disease resulting from the PAP test [1] as evidence that the advances in molecular diagnostics have enormous potential [3]. The reduction in mortality from cervical cancer was not without its cost as there is certainly over-treatment of what are likely indolent lesions, cervical dysplasia, in order to assure the removal of malignant cells. Therefore sensitive diagnostic tests with low specificity result in costs we currently bear at the expense of morbidity from overtreatment. Because particular cancers with similar presentations can have very different courses of malignant development due to subtle differences in molecular etiology, it is unlikely that a single biomarker could provide an accurate diagnostic test for that disease. Therefore panels of biomarkers will be required to achieve sufficient diagnostic sensitivity to comprehensively identify even a single type of human cancer. Assuming the appropriate technology is in place, three prominent elements are required for outstanding biomarker discovery, technical validation and prospective clinical validation: specimens, specimens, specimens! One only needs to look at the standard approaches to biomarker discovery to identify specimens as the major barrier to arriving at sensitive and specific biomarker panels for each clinical application. Investigators who discover and validate biomarkers do not generally have access to well-curated prospective casecontrol samples representative of the future screening population and so biomarkers often fail at the validation stage or in pre-symptomatic clinical trials. Commonly the study design begins with samples obtained at the time of disease diagnosis for the discovery phase of the work. Similar but independent samples are