Gene expression signature predicts chemoresponse of microdissected papillary serous ovarian tumors
5064 Background: The purpose of this study was to identify a predictive gene signature for chemoresponse in patients with advanced stage papillary serous ovarian cancer. Methods: Expression profiling was performed on 50 chemonaive, microdissected advanced stage papillary serous ovarian cancers using Affymetrix Human Genome U133 Plus 2.0 microarrays. Chemoresistance was defined as disease progression while the patients remained on primary chemotherapy. Nine normal human ovarian surface epithelial (HOSE) brushings were also assessed to quantify normal gene expression levels. Validation was performed by quantitative real time PCR using the HOSE isolates and microdissected ovarian tumor samples. Results: A supervised learning algorithm applied to genes differentially expressed between chemosensitive/resistance tumors (p < 0.001) using leave-one-out cross-validation (LOOCV), identified over 2000 genes associated with tumor chemosensitivity. The chemoresponsive gene list was further refined to 576 genes by including only genes used for all LOOCV iterations. An independent gene list was generated comparing expression profiles of chemoresistant tumors to HOSE. The two lists were compared to identify common genes, generating final classifier list of 75 genes that included genes involved in apoptosis, RNA processing, protein ubiquitination, transcription regulation, and other novel genes. We hypothesized genes identified in both data sets would be predictive and biologically relevant. Of these 75 genes, 20 were validated by real-time PCR. Validated genes were ranked by a univariate t-stat value to further resolve the predictor. 4 multivariate predictor algorithms demonstrated the 10 top ranked validated genes maximixed prediction accuracy (compound covariate, 91%; diagonal linear discriminant analysis, 91%; 3-nearest neighbor, 86%; nearest centroid, 95%). The predictive value of these genes will be evaluated on an independent sample set. Conclusions: Gene expression profiling can distinguish between chemosensitive and chemoresistant ovarian cancers. This signature can predict response to therapy and has identified novel biologically and clinically relevant targets. No significant financial relationships to disclose.