Meta-analysis of breast cancer expression data using published gene signatures to reveal key cellular processes implicated in chemosensitivity and resistance.

Abstract

509 Background: Although many pathways and cell states are implicated in therapeutic resistance, predicting response to chemotherapy remains a challenge. Methods: We assembled expression data and pathologic complete response (pCR) versus residual disease (RD) status for 1,507 breast cancer samples collected prior to neoadjuvant chemotherapy. For each sample, we evaluated 118 published gene signatures including signaling, lineage, cell state, immune, and microenvironmental processes. Results: Among all signatures, the strongest predictors of pCR vs. RD were proliferation-related. Microarray data from a breast epithelial cell morphogenesis assay demonstrated that five separate proliferation signatures correlated with in vitro proliferation. To assess how proliferation contributes to chemosensitivity, we evaluated genes differentially expressed in patients with pCR versus RD (FDR-adjusted p < 0.05, limma) before and after normalizing data for proliferation using two distinct proliferation signatures. Among ER+ breast cancer and ER+ subsets Luminal A and B, > 95% of differentially expressed genes were proliferation-associated, suggesting that proliferation differences account for most of the variation in ER+ chemosensitivity. In comparison, among triple-negative breast cancers (TNBCs) and the basal-like subset, only 72.0% and 35.4% of differentially expressed genes, respectively, were proliferation-associated. For TNBCs, signatures associated with chemotherapy response clustered into five key areas: proliferation, mesenchymal phenotype, TGF-beta signaling/stromal features, cyclin and Src activation, and ER signaling (possibly low-level not detected by IHC). Stratifying 175 TNBCs with zero (N0) or few positive lymph nodes (N1) by high versus low activity of three signatures—proliferation, epithelial-to-mesenchymal transition, and Src signaling—was highly predictive of failure to achieve pCR (negative-predictive value 0.965; 95% CI 0.912-0.990). Conclusions: Interrogating multiple signatures in a large expression data set allows insights into key processes associated with chemoresistance and sensitivity in breast cancer.