Abstract / Summary
Abstract Background : A type of non-apoptotic, regulated, inflammatory cell death called necroptosis has emerged as a growing player in the regulation of breast cancer progression and anti-tumor immunity, but its transcriptomic signature has not been systematically defined. Methods : Three discovery cohorts (GSE42568, GSE15852 and GSE29431, totaling 273) were merged with an independent validation cohort (GSE70947, n=296). Differentially expressed genes (DEGs) between tumor and normal tissues were determined after quantile normalization and ComBat batch correction using limma and intersected with 159 curated necroptosis regulators. Penalized logistic regression, support vector machines, and random forests were used to develop predictive models and rank the most important features, using SHAP values. The infiltration of immune cells was quantified using xCell, and the immune-crosstalk between necroptosis and immune cells were analysed by Spearman correlation.
Results: We found 840 DEGs (304 up and 536 down), including seven genes associated with necroptosis (PLA2G4A, STAT5A, PYGL, TLR4, MAPK10, and IL33 all down, while STAT1 was up). A 31-gene consensus signature, which consisted of the most discriminatory genes identified by machine learning, discriminated well in held-out test samples (SVM AUC=0.968, AUPRC=0.987, Brier score=0.049) and showed strong predictive power in the independent validation cohort (AUC=0.842). The most important genes for model prediction were chosen: PCK1, RBP4, TNMD, ITIH5, and PLIN1, through the SHAP analysis. Large-scale immune profiling showed significant enrichment of epithelial cells (0.381 vs 0.231, p<0.001), a depletion of adipocytes (0.216 vs 0.442, p=8.0e-27) and stromal contraction (0.201 vs 0.440, p=8.2e-25) and 148 significant necroptosis-immune associations. There were comparable gene-immune correlation patterns between the discovery cohort and the validation cohort (r=0.757). Summary and Perspectives : Coordinated inhibition of multiple components in the necroptosis pathway is the general mechanism by which necroptosis is suppressed in breast cancer, and is strongly linked to remodeling of the immune microenvironment. The resulting 31-gene signature is able to discriminate between tumor and normal tissue with high accuracy (SVM AUC 0.968) and is generalizable to an independent cohort (AUC 0.842).