| Summary |
Breast cancer is a difficult disease to manage because it is comprised of a spectrum of tumor subtypes with different biological characteristics. Previous gene expression studies using breast tumor “intrinsic” gene lists identified five distinct subtypes of breast tumors: Luminal A, Luminal B, Normal Breast-like, HER2+/ER- and Basal-like. Using a training data set of 102 unique breast tumors, we derive a new “intrinsic” gene list based upon 26 paired samples. This Intrinsic/UNC gene set recapitulates the old classifications and extends the analysis to include a large proliferation signature, which was lacking before. Using Distance Weighted Discrimination, we next created a true “test set” of 311 tumors by combining together three different publicly available breast tumor microarray data sets (1-3), which was then analyzed using the newly derived Intrinsic/UNC gene set. The test set analysis identified the same five intrinsic subtypes seen before and a new one (Luminal I), and we show that the intrinsic subtype classifications are independent predictors of relapse-free survival when adjusting for age, node status, tumor size, grade and ER status. We also developed a “Single Sample Predictor” that is able to assign an intrinsic subtype to one sample at a time, which was also shown to be predictive of outcomes on another independent test set of tamoxifen-treated ER+ patients. These analyses demonstrate that 1) common patterns of expression can be identified across different microarray platforms, 2) the breast tumor “intrinsic” subtypes are present in all breast tumor data sets studied, and 3) this classification is adding value to the existing repertoire of clinical markers for breast cancer patients.
Keywords: parallel sample
|
Less...
More...