Abstract
Triple-negative breast cancer (TNBC) is an aggressive subtype of breast cancer with limited treatment options, which warrants the identification of novel therapeutic targets. Deciphering nuances in the tumor microenvironment (TME) may unveil insightful links between antitumor immunity and clinical outcomes; however, such connections remain underexplored. Here, we employed a data set derived from imaging mass cytometry of 71 TNBC patient specimens at single-cell resolution and performed in-depth quantifications with a suite of multiscale computational algorithms. The TNBC TME reflected a heterogeneous ecosystem with high spatial and compositional heterogeneity. Spatial analysis identified 10 recurrent cellular neighborhoods-a collection of local TME characteristics with unique cell components. The prevalence of cellular neighborhoods enriched with B cells, fibroblasts, and tumor cells, in conjunction with vascular density and perivasculature immune profiles, could significantly enrich long-term survivors. Furthermore, relative spatial colocalization of SMA(hi) fibroblasts and tumor cells compared with B cells correlated significantly with favorable clinical outcomes. Using a deep learning model trained on engineered spatial data, we can predict with high accuracy (mean area under the receiver operating characteristic curve of 5-fold cross-validation = 0.71) how a separate cohort of patients in the NeoTRIP clinical trial will respond to treatment based on baseline TME features. These data reinforce that the TME architecture is structured in cellular compositions, spatial organizations, vasculature biology, and molecular profiles and suggest novel imaging-based biomarkers for the treatment development in the context of TNBC.