Integrated single-cell and bulk RNA sequencing analyses identify a myeloid state-related gene signature for molecular subtyping in stomach adenocarcinoma

Abstract

Purpose: Stomach adenocarcinoma (STAD) is characterized by significant heterogeneity, within which myeloid cells play crucial yet incompletely understood roles. The relationship between the functional states of myeloid cells, patient prognosis, and therapeutic response requires further elucidation. Methods: We integrated single-cell RNA-seq profiles and 443 bulk RNA-seq profiles from the TCGA-STAD cohort. By integrating myeloid cell differentiation trajectories inferred from Monocle2 pseudotime analysis with survival analysis, we identified myeloid state-related prognostic genes (MSRPGs) and constructed a molecular classification (STAD-MSC). We also explored its prognostic significance and multi-omics features. Additionally, we utilized correlation analysis to establish regulatory networks and predict candidate inhibitors. The 5-gene risk model was evaluated in a public 355-patient validation cohort, and the STAD-MSC framework was further assessed at the protein level in a 70-patient retrospective cohort using immunohistochemistry for NNMT, AXL, and COL1A1. Results: We identified 32 MSRPGs across five distinct myeloid states. Consensus clustering stratified the patients into three subtypes, including low immune infiltration STAD (LI-STAD), moderate immune infiltration STAD (MI-STAD), and high immune infiltration STAD (HI-STAD). The HI-STAD subtype, characterized by high immune infiltration accompanied by an immunosuppressive and dysfunctional microenvironment, exhibited the poorest overall survival (global log-rank p = 0.018). The multi-omics analysis revealed subtype-specific genomic and immune landscapes. A 5-gene prognostic signature was constructed and evaluated as a risk-associated prognostic model. In silico analysis identified subtype-associated differences in predicted drug response. Exploratory pharmacogenomic analysis revealed nominal associations for dabrafenib (p = 0.0051) and ruxolitinib (p = 0.041), suggesting potential subtype-specific therapeutic vulnerabilities. Importantly, the three-protein classifier (NNMT/AXL/COL1A1) stratified a retrospective 70-patient cohort into three subgroups with significantly different OS and PFS. Conclusion: Using public-cohort and protein-level clinical validation, we established STAD-MSC, a myeloid state-centric molecular taxonomy that stratifies STAD patients into subgroups with distinct prognoses and immunosuppressive microenvironmental features, providing a framework for immune-informed patient stratification.