Integrated Single-cell and Bulk Transcriptomic Analysis Links NOTCH Signaling to Stromal Features and Prognosis in Gastric Cancer
Mingxin Jin, Zhonghui Cui, Jing Zhang, Jiayi Mi, Shoutang Lu, Xin ZhangIntroduction:
NOTCH signaling is frequently dysregulated in Gastric Cancer (GC) and has been implicated in stromal and immune regulation. This study aimed to characterize NOTCH-associated intercellular communication and develop a transcriptome-based risk model for survival stratification and exploratory prediction of treatment response.
Methods:
The Seurat package was used to process scRNA-seq data. CellChat was applied to infer intercellular communication networks, with a focus on NOTCH-associated crosstalk. In the TCGA-STAD cohort, we calculated a NOTCH pathway activity score and used WGCNA to identify the module most strongly correlated with that score. Univariate Cox, LASSO Cox, and multivariate Cox regression were then used to derive a five-gene Riskscore. A separate multivariable Cox model evaluated its association with Overall Survival (OS) after adjustment for age, sex, AJCC stage, and histological grade. Prognostic performance was assessed using Kaplan-Meier analysis and time-dependent ROC curves and was externally evaluated in GSE66229. Immune and stromal features were quantified using ssGSEA, MCPcounter, and ESTIMATE. TIDE and pRRophetic were used for in silico estimates of immunotherapy response and drug sensitivity, respectively.
Results:
Single-cell analysis showed shifts in the relative cellular composition of GC samples, including higher proportions of T cells and macrophages/monocytes and a lower proportion of B cells. CellChat inferred broader NOTCH-associated communication in tumor samples, particularly between fibroblasts and endothelial cells. WGCNA identified a green-yellow module that was most strongly correlated with the NOTCH activity score and was enriched in extracellular matrix organization. A five-gene module-derived Riskscore stratified OS in TCGA-STAD and GSE66229, although time-dependent discrimination varied across cohorts and follow-up intervals. In the complete-case TCGA-STAD analysis, the Riskscore remained associated with OS after adjustment for age, sex, AJCC stage, and histological grade (HR per SD = 1.28, 95% CI: 1.07-1.53, p = 0.007). High-risk tumors had higher immune and stromal scores and higher expression of CD80, CTLA4, and HAVCR2. TIDE and drug-sensitivity estimates suggested subgroup differences; these results remain hypothesis-generating because they were not validated experimentally.
Discussion:
This integrative computational study identified associations among NOTCH pathway activity, inferred fibroblast-endothelial communication, extracellular-matrix programs, and a five-gene prognostic score. The five genes are not canonical NOTCH components; rather, they were selected from a co-expression module correlated with the NOTCH activity score. The findings therefore support an indirect, datadriven association and do not establish that NOTCH signaling causes stromal remodeling, immune evasion, or drug response.
Conclusion:
NOTCH pathway activity was associated with stromal and immune features in GC, and a five-gene module-derived Riskscore provided prognostic stratification with variable discrimination across cohorts. Independent single-cell and experimental validation is required before the inferred crosstalk or treatment-response estimates can inform clinical decisions.