Integrative single-cell and spatial analyses implicate ECM2 in pathological CAF remodeling in breast cancer

Background Cancer-associated fibroblasts (CAFs) are major contributors to extracellular matrix (ECM) remodeling in breast cancer (BRCA), but the transcriptional programs underlying pathological CAF remodeling remain incompletely understood. Given shared features of chronic fibrotic and tumor stromal remodeling, we used heart failure (HF)-associated remodeling as a cross-disease discovery framework. Methods HF-derived remodeling programs were established from bulk transcriptomic data, validated in independent HF cohorts, and applied to TCGA-BRCA. Bulk, single-cell, and spatial transcriptomic analyses were integrated with patient-aware mixed-effects, pseudobulk, and cell–cell communication analyses. Leakage-controlled machine learning, ECM2-exclusion sensitivity analysis, virtual knockout, drug prediction, and molecular docking were also performed. ECM2 function was evaluated by siRNA-mediated knockdown in TGF-β1-stimulated human mammary fibroblasts. Results HF-derived pathological remodeling activity was associated with CAF- and ECM-related features in BRCA. Single-cell analyses localized higher remodeling activity predominantly to CAFs, particularly ECM-myCAFs and iCAFs. Patient-level pseudobulk analysis linked pathological remodeling CAFs to ECM- and TGF-β-related programs, while patient-aware communication analysis showed increased COLLAGEN and FN1 signaling. Leakage-controlled machine learning showed that pathological and protective states remained transcriptionally distinguishable after excluding state-defining genes. ECM2-exclusion sensitivity analysis showed minimal changes in remodeling scores and state assignments, while ECM2 remained elevated in the reconstructed pathological state. Associations between ECM2 and several ECM-remodeling markers persisted after adjustment for stromal composition or tumor purity together with PAM50 subtype. Spatial and single-cell analyses further linked ECM2 to CAF-containing stromal compartments. Virtual knockout predicted remodeling-related network perturbations, while DrugReflector nominated candidate compounds from the ECM2 virtual-knockout-derived perturbation signature, with exploratory docking identifying a predicted ECM2–dapivirine docking pose. ECM2 knockdown reduced α-SMA, COL1A1, and POSTN expression in TGF-β1-stimulated fibroblasts. Conclusion ECM2 is associated with pathological CAF/ECM remodeling in breast cancer, and its knockdown attenuates fibroblast activation in vitro. This study provides a cross-disease framework integrating patient-aware multi-omics analyses to characterize tumor stromal remodeling and prioritize candidates for further mechanistic and therapeutic investigation.