Predicting treatment outcome using kinome activity profiling in HER2+ breast cancer biopsies

Abstract

In this study, we measured the kinase activity profiles of 32 pre-treatment tumor biopsies of HER2-positive breast cancer patients. The aim of this study was to assess the prognostic potential of kinase activity levels, to identify potential mechanisms of resistance and to predict treatment success of HER2-targeted therapy combined with chemotherapy. Indeed, our system-wide kinase activity analysis allowed us to link kinase activity to treatment response. Overall, high kinase activity in the HER2-pathway was associated with good treatment outcome. We found eleven kinases differentially regulated between treatment outcome groups, including well-known players in therapy resistance, such as p38a, ERK, and FAK, and an unreported one, namely MARK1. Lastly, we defined an optimal signature of four kinases in a multiple logistic regression diagnostic test for prediction of treatment outcome (AUC = 0.926). This kinase signature showed high sensitivity and specificity, indicating its potential as predictive biomarker for treatment success of HER2-targeted therapy.

Keywords: Oncology; cancer; molecular biology; proteomics.

Graphical abstract

Figure 1

Study overview (A) Experimental workflow. (B) Number of kinase T-loops quantified per patient biopsy in each treatment outcome group. (C) T-loop abundance reproducibility of two technical workflow replicates that were created by splitting a single patient sample in two.

Figure 2

Low kinase activity levels within HER2 pathway are associated with treatment resistance (A) Graphical representation of HER2-pathway. Kinases quantified in this study are highlighted. (B) Unsupervised clustering of HER2-pathway kinase activities (Euclidean distance, Z-scored data). HER2 protein expression levels are displayed above the heatmap. ∗Significantly regulated kinase; #kinase T-loop peptide was shared with at least one paralog.

Figure 3

Kinase activities (by T-loop phosphorylation) differentiate treatment outcome groups (A) Heatmap of unsupervised clustering of all significantly regulated kinase T-loops (ANOVA p value <0.05) (Euclidean distance, data were Z-scored). # kinase T-loop was shared with at least one close paralog. (B–D) Boxplots of regulated kinase T-loops. ∗ p value <0.05; ∗∗ p value <0.001. E) Receiver operator characteristics (ROC) curve analysis to predict treatment outcome based on kinase T-loop abundance for single kinases. (F) ROC curve to predict treatment outcome based on a panel of kinases. Six-kinase panel consisted of kinase T-loops detected in at least 80% of the patients, with a minimum of 1.5-fold difference and max p value of 0.05. Four-kinase panel consisted of kinase T-loops detected in all patients with p value <0.05.