Abstract LB_A02: Predicting breast cancer patients’ response to neoadjuvant chemotherapy from the deconvolved transcriptome of different cell types in the tumor microenvironment

Abstract

Abstract Introduction: The tumor microenvironment (TME) constitutes a multifaceted ecosystem that influences tumor growth and clinical response via dynamic interactions between diverse cell types. Considerable attention has been devoted in recent years in exploring the TME’s impact on clinical response following immunotherapy, warranting further investigation into its impact in the context of chemotherapy. To address this, we present a novel computational framework called DECODEM (DEcoupling Cell-type-specific Outcomes using DEconvolution and Machine learning). Through DECODEM, we investigate the association of cell-type-specific transcriptomic profiles within the TME with response to neoadjuvant chemotherapy (NAC) in breast cancer (BC). Methods: DECODEM leverages cellular deconvolution of bulk transcriptomics data to extract cell-type-specific transcriptomics profiles for different cell types present in the TME, and further utilizes machine learning to build cell-type-specific predictors of clinical response. Using DECODEM, we analyzed the bulk transcriptomics data across three publicly available cohorts of HER2-negative BC patients who underwent NAC. We compared the predictive powers of these cell types, both individually and in concert with each other, against a state-of-the-art predictor employing clinical and transcriptomic features. We then validated our findings in a single-cell chemotherapy cohort of triple-negative breast cancer. Furthermore, we extended DECODEM to DECODEMi that can incorporate the cell-cell interactions (CCIs) among the prominent cell types to predict clinical response, hinting at the underlying mechanism at play. Results: Our analysis revealed that the diverse cell types within the TME significantly contribute to the prediction of chemotherapy response in HER2-negative BC, thus extending beyond the impact of malignant tumor cells alone. Notably, our key findings are: 1. We demonstrated the superior predictive powers of immune cells (myeloid, plasmablasts, B-cells) and stromal cells (endothelial, normal epithelial, cancer-associated fibroblasts), as compared to the bulk mixture and the state-of-the-art predictor. 2. We illustrated the complementary impacts of the prominent cell types via building highly accurate multi-cell-ensembles, where, remarkably, a three-cell-ensemble of endothelial, myeloid, and plasmablasts displayed the best predictive capability across cohorts. 3. We displayed that DECODEMi can capture the regulatory roles of CCIs and identified new CCIs likely to mediate chemotherapy response in HER2-negative BC. Conclusion: We provide an unprecedented computational approach to objectively delineate the cell-type-specific impacts of the TME in clinical response. Our analysis provides comprehensive insights into the intricate interplay of the cells in TME and their associations with chemotherapy response in HER2-negative BC, contributing to a deeper understanding of BC treatment efficacy. Citation Format: Saugato Rahman Dhruba, Sahil Sahni, Binbin Wang, Di Wu, Yael Schmidt, Eldad Shulman, Sanju Sinha, Stephen-John Sammut, Carlos Caldas, Kun Wang, Eytan Ruppin. Predicting breast cancer patients’ response to neoadjuvant chemotherapy from the deconvolved transcriptome of different cell types in the tumor microenvironment [abstract]. In: Proceedings of the AACR-NCI-EORTC Virtual International Conference on Molecular Targets and Cancer Therapeutics; 2023 Oct 11-15; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2023;22(12 Suppl):Abstract nr LB_A02.