Development and evaluation of molecular signatures for predicting the efficacy of therapeutics in PI3KCA mutated breast cancer tumors.

B34

Introduction: Activating mutations in the catalytic subunit of the phosphatidylinositol kinase (PI3KCA) gene are found in 20-40% of human breast cancers and thus represent a potential driver for the neoplastic process. Loss of PTEN expression is also a common feature of the disease and these two events occur in mutually exclusive tumors. Ergo, dis-regulated activation of the PI3K/AKT survival pathway is predominant in breast cancers. Further, molecular signatures characteristic of this pathway deregulation would be useful for identifying cancers most likely to respond to targeted therapeutics. The primary goals of the current work were (1) to determine clinical correlates of PI3K mutational status, (2) to derive a gene signature algorithm that could accurately predict activation of the PI3 kinase pathway, and (3) to assess a panel of molecular pathway and chemosensitivity signatures on tumors annotated for PI3 kinase mutations.
 Methods: The entire coding region of the PI3KCA gene was sequenced from 74 human breast cancer tumor samples for which microarray data were also available. Models of molecular pathway activation (PI3K, ER+, and Her2/Neu amplification) and chemosensitivity signatures (PI3K selective inhibitors and standard chemotherapeutics) were applied to the tumor array data using binary regression analysis. Each signature was created by the selection of control and experiment training sets using either NCI-60 cell lines or independent tumor sets and the resulting top gene coefficients. The Wilcoxon rank sum test was use to determine correlations with mutational status and the prediction value for all pathway activation and chemosensitivity signatures. Finally, clinical parameters including ER/PR status, disease free survival times, age at diagnosis, recurrence, tumor size, and node status were compared to mutational status using Fisher’s Exact test.
 Results: Mutations in the PI3KCA coding sequence were detected in 27% (20/74) of our tumor population and were more frequent in non-Caucasians (P=0.020). In the molecular signature studies, using the NCI-60 cell line panel annotated for PI3KCA and PTEN status as a training set revealed a positive trend with mutated tumors, but was not significant for two of the signatures tested. (250 genes P=0.245 and 100 genes P=0.104) Further, there was no correlation with predicted sensitivity to PI3KCA selective inhibitors and mutation status in the tumors. However, tumors with PI3KCA mutations correlated with predicted resistance to paclitaxel (P=0.029), predicted ER positivity (P=0.019) and predicted lack of Her2/Neu amplification (P=0.033).
 Conclusions: The correlation between non-Caucasian race and PI3K mutations is a new finding. A consistent gene expression signature that reflects activation of the PI3K pathway in breast cancer could not be discerned using annotated NCI-60 cell lines; however, we did determine a significant correlation with predicted resistance to paclitaxel and PI3K mutations. Although clinical ER status did not correlate with PI3K mutational status, the predicted ER positivity and lack of Her2/Neu amplification correlated significantly. Future studies will include delineation of these identified chemotherapeutic windows and further determination if a PI3K activation model can be validated in breast cancer.