Document resource
Background The tumor microenvironment (TME) is a dynamic ecosystem comprising of diverse cell types that play essential roles in tumor growth and clinical outcome through their interactions. While the multifarious cell-cell interactions (CCIs) within the TME have been studied extensively, especially in the context of immunotherapy, understanding their role in chemotherapy outcome remains less explored. To this end, we present a novel, generic computational framework named DECODEM ( DEcoupling Cell-type-specific Outcomes using DEconvolution and Machine learning) that facilitates exploring the association of cell-type-specific gene expression with the effectiveness of neoadjuvant chemotherapy (NAC) in breast cancer (BC).Methods DECODEM applied cellular deconvolution of bulk transcriptomics to extract cell-type-specific gene expression within the TME, and subsequently leveraged machine learning to build cell-type-specific predictors of clinical outcomes. We used DECODEM to analyze the bulk gene expression from three cohorts of HER2-negative BC patients treated with NAC, validating model generalizability. We systematically assessed the predictive capabilities of individual cell types and their collective influence within the TME, benchmarking against a state-of-the-art predictor using both clinical and transcriptomic features. We further validated our findings in two single-cell (SC) cohorts of triple negative breast cancer (TNBC) patients undergoing NAC and immune checkpoint blockade (ICB) therapy. Moreover, we extended DECODEM to DECODEMi to investigate the CCIs involving top immune cell types, offering potential insights into the mechanism.Results Our findings highlight the importance of active pre-treatment immune infiltration in the TME in achieving positive outcome to chemotherapy in HER2-negative BC, going beyond the impact of malignant cells. The key insights from our study are as follows:Gene expression of specific immune cells (myeloid, plasmablasts, B-cells) and stromal cells (endothelial, normal epithelial, cancer-associated fibroblasts) are highly predictive of chemotherapy response.The ensemble of the expressions of top immune and stromal cells (endothelial, myeloid and plasmablasts) achieve the highest predictive accuracy across cohorts, outperforming the predictor built on the original tumor gene expression.The top cell-type-specific predictors generalize to SC expression, predicting patient response to both chemotherapy and ICB therapy in TNBC.DECODEMi identifies key immune CCIs mediating chemotherapy response in the TME of HER2-negative BC, validated in SC data.Conclusions We provide an unprecedented computational methodology to quantitatively assess the cell-type-specific influences of the TME on clinical outcome. This approach holds promise for elucidating the nuanced ways by which the TME can modulate therapeutic efficacy, potentially paving the way for effective personalized cancer treatments.