Background: The significant genetic heterogeneity amongst breast cancer patients is a primary obstacle to effective clinical diagnosis and management. Advances in microarray technology have enhanced understanding of the molecular mechanisms that drive tumorigenesis in individual patients, and emerging technologies based on gene expression profiling (GEP) may provide clinically useful information to improve the management of breast cancer. GEP has been utilized to refine classification of previously undistinguishable tumor subgroups, and predict prognosis and response to anti-cancer agents. Here we report the use of GEP to refine phenotypic (ER, PR, EGFR, HER2/neu) classification, and assess proliferative rate (KI-67). Methods: 437 patients were selected from the Institut Paoli-Calmettes (IPC, n=213; separated into independent identification (n=159) and validation (n=54) datasets), the Institut Bergonié (IB, n=114), and Centre Léon Bérard (CLB, n=110). Tissues were analyzed on 10K nylon cDNA microarrays (DiscoveryChip) in a GLP environment. Identification analyses were performed with ProfileSoftware™ Corporate (Ipsogen). Gene signatures were elucidated based on adjusted t-test analysis on an identification set. A robust and automated prediction model was designed to validate each identified signature in clinical laboratories. Sensitivity and specificity were calculated in comparison to standard histopathological, IHC and/or fluorescence in-situ hybridization (FISH) techniques. Results: ER, PR, HER2/neu signatures were identified on the IPC identification dataset (n=159), and then validated on the three independent datasets. Two thresholds were used for ER and PR (1% and 10% of stained cells/total nuclei) with very similar signatures (90% of concordance) on the IPC dataset. An EGFR signature was calculated on the IPC identification dataset, then confirmed on the IPC validation dataset. Tumor proliferation rate (KI-67 GES) was determined on the IB dataset, then validated on the IPC dataset. Gene signatures were subsequently cross-validated at both the RNA and protein level by RQ-PCR and IHC. All phenotypic gene signatures are robust with sensitivity and specificity higher than 80%. Conclusions: These results collectively demonstrate GEP can be utilized to augment existing pathological analysis to classify patient sub-groups with distinct behavior and prognosis, and may facilitate decisions on treatment options for individual patients. Additional studies are in progress to evaluate the clinical impact of molecular classification and risk assessment in the management of breast cancer.

[Proc Amer Assoc Cancer Res, Volume 46, 2005]