Gene expression profile-based predictors of response to chemotherapy

A molecular test that could help in selecting the most effective chemotherapy for a particular individual could save patients from unnecessary toxicity, and the right choice of drugs may save lives, particularly in the adjuvant treatment of breast cancer. Administration of chemotherapy before surgery provides an attractive opportunity to discover predictors of response [1]. Pathologic complete eradication of cancer from the breast and lymph nodes (pCR) represents an extreme form of chemotherapy sensitivity and invariably heralds excellent long-term survival. We adopted pCR as an early surrogate of clinically meaningful benefit from therapy and as an outcome that is worth predicting. There are simple clinical and histological parameters, including grade, estrogen receptor status and tumor size, that can be combined into powerful prediction scores. However, these clinical variables do not yield treatment regimen specific predictions, and they cannot be used to select one therapy over another. Assessment of traditional single gene markers of chemotherapy sensitivity has not yet resulted in clinically useful tests. Gene expression profiling, which enables simultaneous measurement of thousands of genes, represents a promising new tool that may be applied to this clinical problem. It is currently unknown what the best strategy is to discover response predictors from high dimensional gene expression data. The simplest approach may be to search for the single most informative gene that is differentially expressed between responders and nonresponders. This may lead to new mechanistic insights into the biology of chemotherapy response and could yield easy-to-use but moderately powerful single gene predictive markers [2]. Another approach is to identify gene expression signatures that are predictive of response, assuming that the combined information provided by multiple genes would result in more accurate predictions than any single gene can do. Several small studies have suggested that this is feasible [3]. Large-scale validation of these results is needed and is currently underway. Yet another approach is to recognize the different molecular subtypes of breast cancer and attempt to develop distinct predictors for each subtype [4]. This approach assumes that, by focusing on the molecularly more homogenous subgroups, more accurate predictors could be developed than by analyzing all breast cancers together. We shall present results from our own research program, illustrating the successes and limitations of each of these approaches.