Matrix-Associated Gene Expression and Annotated Tumor–Stroma Composition of Transcriptomically Defined Cold Regions in Triple-Negative Breast Cancer

Background/Objectives: Regional transcriptomic labels can reflect cell mixtures and expression states. We examined how transcriptomically defined cold regions in triple-negative breast cancer relate to matrix-associated expression, annotated composition, spatial neighborhoods, and cell-type attribution. Methods: We reanalyzed public data from 43 discovery sections (22 patients) and 11 cancer captures of a secondary dataset with discovery-derived labels. Computational spot labels included desert (D), excluded (E), inflamed (I), and ambiguous-cold. Hierarchical models estimated E−D contrasts and E−I gene contrasts with and without sequencing-depth and estimated tumor-abundance adjustment; tests underwent Holm correction. Further analyses addressed annotated composition, neighborhoods, and pooled exploratory cell-type attribution. Results: Estimated cancer-associated fibroblast (CAF) abundance was higher in E- than D-labeled regions (standardized within-section effect 0.78; 95% confidence interval 0.55–1.00), as was expression of PDGFRB, COL3A1, COL5A1, and COL6A1; the three collagen genes remained higher than in D- and I-labeled regions after adjustment. Across 11 captures, the median within-capture difference in tumor-annotated spot fraction (E minus D) was −14.1 percentage points (95% bootstrap interval −33.5 to −10.7; nine negative, two ties; interval reached zero without in situ annotations). In discovery, 330 of 822 ambiguous cold spots neighbored both D- and E-labeled spots, exceeding two null models. Pooled models estimated positive COL3A1 and COL5A1 differences within CAFs; immune readouts depended on scoring and adjustment. Conclusions: We characterized a stromal–transcriptomic contrast: E-labeled regions showed higher matrix-associated expression and estimated CAF abundance than D-labeled regions and, in the secondary dataset, lower annotated tumor fraction. These measurements refine interpretation without implying functional immune exclusion.