Re-interpretation of PAM50 gene expression as quantitative tumor dimensions shows utility for clinical trials: application to prognosis and response to paclitaxel in breast cancer

Abstract

Background: We recently showed PAM50 gene expression data can be represented by five quantitative, orthogonal, multi-gene breast tumor traits. These novel tumor ‘dimensions’ were superior to categorical intrinsic subtypes for clustering in high-risk breast cancer pedigrees, indicating potential to represent underlying genetic susceptibilities and biological pathways. Here we explore the prognostic and predictive utility of these dimensions in a sub-study of GEICAM/9906, a Phase III randomized prospective clinical trial of paclitaxel in breast cancer. Methods: Tumor dimensions, PC1–PC5, were calculated using pre-defined coefficients. Univariable and multivariable Cox proportional hazards (PH) models for disease-free survival (DFS) were used to identify associations between quantitative dimensions and prognosis or response to the addition of paclitaxel. Results were illustrated using Kaplan–Meier curves. Results: Dimensions PC1 and PC5 were associated with DFS (Cox PH p = 6.7 × 10−7 and p = 0.036), remaining significant after correction for standard clinical–pathological prognostic characteristics. Both dimensions were selected in the optimal multivariable model, together with nodal status and tumor size (Cox PH p = 1.4 × 10−12). Interactions with treatment were identified for PC3 and PC4. Response to paclitaxel was restricted to tumors with low PC3 and PC4 (log-rank p = 0.0021). Women with tumors high for PC3 or PC4 showed no survival advantage. Conclusions: Our proof-of-concept application of quantitative dimensions illustrated novel findings and clinical utility beyond standard clinical–pathological characteristics and categorical intrinsic subtypes for prognosis and predicting chemotherapy response. Consideration of expression data as quantitative tumor dimensions offers new potential to identify clinically important patient subsets in clinical trials and advance precision medicine.