Predicting the response to immunotherapy from gene expression data in HER2-negative breast cancer.

Background: The addition of immunotherapy in the neoadjuvant setting is showing promising results for HER2- and triple-negative breast cancer patients, but pathological complete response is observed only in a fraction of patients. The aim of the present work was to investigate if ARIADNE, an algorithmic strategy to analyze gene expression data from bioptic samples based on epithelial-mesenchymal phenotypes, can predict the response to immunotherapy in HER2- patients.

Methods: We considered gene expression data for HER2-breast cancer patients treated with pembrolizumab in addition to chemotherapy (n = 69) and with chemotherapy alone (n = 179) from the I-SPY 2 trial. We stratified patients in two risk groups (low/high risk) according to the score of the ARIADNE algorithm and studied an additional cytokine signature. To better understand the significance of our results, we studied the interactions among genes in the PD-L1 pathway and analyzed single-cell data from TNBC patients treated with pembrolizumab.

Results: Our results show that ARIADNE predicts differential response to immunotherapy: the high-risk group has a pathological complete response (pCR) rate of 26% as compared with 62% for the low-risk group (OR 4.7, with 1.68-11.32 95% CI and p < 0.01). We also find significant correlations between a cytokine score and the rate of pCR. The ability of ARIADNE to predict pCR is associated with regulatory activity within the PD-L1 pathway. Comparison between ARIADNE and other immunological genomic signatures shows no correlations. The study of single-cell data showed that patients responding to immunotherapy display a larger number of exhausted T-cells than non-responders.

Conclusions: Our analysis shows that ARIADNE is predictive of the response to immunotherapy, but not to chemotherapy, in HER2- patients.