A Stromal-Immune Gene Signature for Precision Prognosis and Therapeutic Guidance in Gastric Cancer

The tumor microenvironment (TME), especially the extracellular matrix (ECM) and immune milieu, drives gastric cancer (GC) progression and therapy resistance. However, prognostic biomarkers integrating these two dimensions remain lacking, limiting precise risk stratification and personalized treatment. We analyzed bulk transcriptomic data (TCGA-STAD, GSE15459, GSE62254), single-cell RNA-seq data (GSE206785), and drug sensitivity data. Unsupervised clustering identified ECM-defined GC subtypes, and machine learning (LASSO, XGBoost) constructed a prognostic signature. Single-cell analyses dissected ECM-active cell origins and mechanisms. Two ECM subtypes (ECM-high/low) were identified; ECM-high had shorter OS (18.2 vs. 36.5 months, HR = 2.13, P < 0.0001). A 7-gene ECM-immune signature (GPC3, TGFB2, ADAMTS1, EFEMP1, LOX, SVEP1, VWF) was validated across 3 cohorts (5-year OS AUC: 0.735/0.636/0.676), outperforming AJCC staging (C-index = 0.72 vs. 0.65, P < 0.01) and serving as an independent prognostic factor (HR = 1.62, P < 0.001). High-risk patients showed immune exclusion (M2 macrophages:21.3% vs.12.5%, P < 0.001) and enhanced ferroptosis. Single-cell data revealed ECM-active cells derived from endothelial cells (62.3%) and CAFs (28.5%), interacting via TGFB2-TGFBR1. The signature predicted drug sensitivity (high-risk: cisplatin IC50 = 3.8 ± 0.7 µmol/L, P < 0.001); low-risk patients gained no benefit from combination chemotherapy (5-year OS:45.3% vs.42.1%, P = 0.38). A web tool (https://gc-ecm-signature.shinyapps.io/) was developed. This ECM-immune signature captures TME heterogeneity, offers robust prognostic value, and guides personalized GC therapy.