Identification and validation of a genomic predictor to distinguish classes of patients with distinct outcomes among poor prognosis breast tumors after anthracycline-based adjuvant therapy Free

3586

Background: The significant genetic heterogeneity among breast cancer patients is a primary obstacle to effective clinical diagnosis and management. Emerging technologies based on gene expression profiling (GEP) may provide clinically useful information to improve the management of breast cancer. GEP has been used to refine classification of previously undistinguishable tumor subgroups, and predict prognosis and response to anticancer agents. Here we report a multicentric GEP analysis to identify and validate a predictor in order to improve tumor classification and predict clinical outcome of patients after standard anthracycline-based adjuvant chemotherapy. Material & methods: 504 patients with early breast cancer treated with adjuvant anthracycline-based chemotherapy were selected from Institut Paoli-Calmettes (IPC, Marseille), Centre Léon Bérard (CLB, Lyon), Institut Bergonie (IB, Bordeaux), and from two prospective randomized therapeutical trials of adjuvant chemotherapy (FNCLCC: arm A of PACS01, and arms A and B of PEGASE01). Tumor RNAs were analyzed on 10K nylon cDNA microarrays. Metagenes for tumor classification were identified based on adjusted t-test analysis and hierarchical clustering on an identification set (IPC). A Cox-based method was applied to metagenes on an identification set of 323 patients to find predictors able to discriminate patients with favorable outcome (no metastasis) after chemotherapy. The stability and robustness of the model were assessed on an independent validation set (n=181). Results: A predictor was identified on 323 patients treated with chemotherapy (anthracyclines). This predictor was based on a linear combination involving different metagenes, and allowed the computing of a metastatic score = ∑(a_i x metagene_i). This score separated two groups of patients with different outcome with respective 5-year MFS of 79% and 52% (p<0.0001, log-rank test). The robustness of this predictor was then confirmed on a validation set of 181 patients, with respective 5-year MFS of 80% and 60% in the so-defined good-prognosis and poor-prognosis groups (p=0.01, log-rank test). In multivariate analysis, our multigenic predictor compared favorably with other classical prognostic parameters. Conclusion: Our metagene-based predictor is highly efficient to discriminate patients with unfavorable outcome after adjuvant anthracycline-based chemotherapy. It uses a validated combination of genes known for their biological relevance, and is valid irrespective of the clinical centre. Additional clinical studies and technical developments are ongoing to translate this new tool into a test designed for routine clinical practice.