Clinical validation of in vitro drug sensitivity microarray data: Regimen-specific signatures predict pathological complete response to neo-adjuvant chemotherapy for breast cancer in a randomized trial (EORTC 10994/BIG 00–01)

544 Background: We previously described gene expression signatures that predict sensitivity to common chemotherapeutic agents and published promising results of their applicability in patients (Nature Med 2006). The goal of this study was to confirm their validity in a larger series of breast cancer patients with hormone-receptor negative (HR negative) since these tumours are more sensitive to chemotherapy. We used pathological complete response as a surrogate for chemosensitivity. We analyzed samples from a subset of patients included in a recently completed large neoadjuvant phase III trial. The trial compares a non-taxane regimen (fluorouracil + epirubicin + cyclophosphamide × 6; FEC arm) with a taxane regimen (docetaxel × 3 then epirubicin + docetaxel × 3; T->ET arm). Methods: RNA prepared from frozen samples obtained at diagnosis were hybridized to Affymetrix arrays. In vitro single agent signatures generated using a metagene approach were combined to obtain a FEC and a T->ET regimen-specific signatures. Predictions were blinded to patient outcome. With both signatures we calculated the receiver operating curve, its AUC, and the cut-point with maximal Youden index- accuracy, positive predictive value (PPV), sensitivity (Sens), negative predictive value (NPV) and specificity (Spec). Results: Samples from 124 patients (55 pCR) with HR negative tumours underwent a successful gene-expression array: 65 patients were treated in FEC arm and 59 patients in T->ET arm. The results are summarized below. Conclusions: We have validated the approach of using regimen-specific genomic signatures developed in vitro, in the context of a multicenter randomized trial. These results support the activation of a prospective trial comparing the conventional random choice of chemotherapy versus a specific array based approach. [Table: see text] [Table: see text]