Validation of an immunomodulatory gene signature algorithm to predict response to neoadjuvant immunochemotherapy in patients with primary triple-negative breast cancer.

Abstract

3117 Background: A 101-gene algorithm established as a molecular subtyping method for triple-negative breast cancer (TNBC) includes assignment of an immunomodulatory (IM) subtype based on genes active in immune cell processes. Recently, we isolated the IM concept to an independent 27-gene algorithm and its predictive ability for immunotherapy response was demonstrated in lung cancer. The objective of this study was to validate the predictive accuracy of the IM subtype as determined by the 27-gene algorithm for pathological complete response (pCR) compared with PD-L1 immunohistochemistry (IHC) staining in TNBC. Methods: We obtained RNA sequencing data from pretreatment core needle biopsies in 55 patients with stage I-III primary TNBC who received neoadjuvant immunotherapy (durvalumab with weekly nab-paclitaxel followed by ddAC) in phase I/II trial (NCT02489448). The 27-gene algorithm was used to determine IM positivity using a cutoff point previously validated from 71 lung cancer biopsy patients treated with immunotherapy. Results from the algorithm and PD-L1 IHC (antibody, SP263) were compared with pCR. Predictive accuracy of both methods was determined by diagnostic indicators. In cases positive for the IM subtype and pCR, we analyzed the immune microenvironment by deconvoluting the immune infiltration using a computational algorithm. Results: Of the 55 patients, 25 (45%) had pCR. Compared with previous subtyping methods, the 27-gene algorithm showed stronger predictive value (odds ratio, 4.125; 95% CI, 1.36-13.47; P < 0.015). For PD-L1 IHC, the odds ratio was 2.63 (95% CI, 0.82-9.21; P = 0.11). The positive predictive value, negative predictive value, positive likelihood ratio, and negative likelihood ratio for PD-L1 IHC were 0.55, 0.68, 1.43, and 0.54, respectively. For the 27-gene algorithm, these metrics were 0.65, 0.69, 2.09, and 0.51, indicating its superior accuracy for predicting pCR. The computational algorithm showed that the IM subtype and pCR were negatively associated with a macrophage-enriched microenvironment. CD4+ T cells and dendritic cells were significantly increased among the baseline immune cell population in IM-positive patients. Conclusions: We conclude that 27-gene algorithm is a clinically applicable and a possible predictive marker for response to neoadjuvant immunochemotherapy for patients with primary TNBC. Our immune microenvironment results suggest that antigen-presenting immune cells have a crucial role in immunochemotherapy response.