MOLECULAR STRATIFICATION OF INTESTINAL AND DIFFUSE SUBTYPES OF GASTRIC ADENOCARCINOMA AND ITS PROGNOSTIC IMPACT

Abstract

Gastric adenocarcinoma is routinely classified by Lauren histology into Intestinal and Diffuse types; however, its prognostic use is limited by molecular heterogeneity within each subtype. We tested whether transcriptomic features refine risk beyond histology and whether biomarker signals generalize across populations. The premise is that when underlying programs diverge across tumors and cohorts, a single biomarker is unlikely to be universally informative; accordingly, integrating molecular signatures with histology may improve risk stratification. We analyzed TCGA-STAD and GTEx gene expression data for discovery and validated the findings in ACRG. Cell-program markers guided unsupervised clustering, and a predictive signature was learned using recursive feature elimination with a random forest and cross-validation. Pathway context was assessed with GSEA, and clinical relevance was evaluated within each Lauren subtype. Clustering of canonical gastric and intestinal programs yielded two molecular clusters that did not explain survival. However, Diffuse tumors with stronger intestinal-like programs tended to experience better outcomes. Consistently, the derived signature was associated closely with Lauren categories and defined two molecular groups. Nevertheless, neither histology nor the signature alone stratified overall survival at conventional thresholds. Moreover, gene-level survival analyses indicated subtype specificity rather than universality. In TCGA, 28 signature genes were associated with overall survival (P < 0.05; HR, 0.79-1.63). ROC analyses for Lauren typing were heterogeneous, with CRLF1 achieving an AUC of 0.92, followed by CYP1B1 and MMRN2. ACRG gene expression revealed an inversion of the TCGA pattern, with prognostic risk shifting toward the Diffuse subtype. Specifically, CRLF1, ERG, and FRZB shifted from Intestinal risk to Diffuse risk, while CDX1 shifted from Intestinal to Diffuse protection. Taken together, these findings suggest that the utility of biomarkers in gastric cancer depends on histology and the population context. Consequently, a contextual strategy that integrates molecular signatures with the Lauren classification has the potential to improve risk stratification beyond either source alone.