Gene expression profiling for prediction of treatment response

Abstract

9594 Background: We designed a strategy for patient stratification aimed at identifying tumor gene expression profiles that predict a favorable response to chemotherapy and allow selection of patients who would most benefit from treatment. Methods: A statistical model was developed based on global gene expression analysis of samples from 119 individuals. Two approaches were compared for their ability to disclose predictive gene expression profiles: (1) Compare normal ovary and ovarian adenocarcinoma and test that gene profile as a marker set to distinguish responsive from similarly treated, non-responsive carcinomas or (2) Directly compare gene expression profiles of responders and non-responders and apply this profile to patient samples to predict outcome. Samples were involved tissues from patients with ovarian cystadenocarcinoma obtained prior to treatment (n = 26) and histologically normal ovaries from controls (n = 93). 85% (22/26) of the ovarian carcinoma patients were advanced (Stage III-IV) disease and all were treated with paclitaxel-carboplatin. Follow-up was 6–59 months with an overall response rate of 61.5%. Gene expression was determined using Affymetrix U133 microarrays that contain >40,000 known genes and ESTs. Marker profiles were sought from pair-wise comparisons and successive validation performed by drop-replace analysis within the training sets. Results: Gene expression profiles were identified from the normal-disease comparison and were predictive of patient response to treatment. Genes can be individually weighted for their value in predicting treatment response and potentially chosen and assembled into a prognostic marker set. Conclusion: Cross-validation of the training set of 26 ovarian cancers, gives preliminary evidence that a gene set predicts outcome. The reference database of normal samples adds dimension to the building of classification models for patient disease stratification. This approach can now be compared to the predictive value of within-disease gene sets derived in the absence of normal samples. Gene expression promises to be valuable in selecting patients for clinical trials and disease management. Author Disclosure Employment or Leadership Consultant or Advisory Stock Ownership Honoraria Research Funding Expert Testimony Other Remuneration Gene Logic, Inc. Gene Logic, Inc. Gene Logic, Inc.