Multiscale characterization of an inflammatory macrophage state associated with an air pollutant-prioritized prognostic signature in gastric cancer.

Background: Air pollutants are implicated in carcinogenesis, but their relevance to gastric cancer (GC) across prognosis, tumor organization and cell behavior remains unclear. We integrated evidence to evaluate pollutant-prioritized GC candidates. Methods: Seven pollutants were integrated with GeneCards/OMIM GC genes using toxicology and candidate-prioritization resources. Robustness analyses used alternative disease definitions, Fisher's tests, 5,000 degree- and annotation-matched null sets, enrichment and protein-protein interaction networks. TCGA-STAD supported nested evaluation of 117 survival pipelines and a five-gene neural network, with locked external testing in GSE84437 and GSE54129. Single-cell, donor-aware macrophage, CellChat, Monocle2, multiplex immunofluorescence and cell2location-guided Visium analyses resolved tissue context. Docking, 100-ns molecular dynamics and bibliometrics examined GSK-3β-centered hypotheses. AGS cells underwent benzene/toluene dose screening followed by CCK-8, colony formation, Transwell and wound-healing assays. Results: Among 176 genes, 63 overlapped with GC beyond matched-null expectations (empirical P = 0.012), with enrichment in inflammation, hypoxia, apoptosis, metabolism and carcinogenesis. APRGSig (DNMT1, GSK3B, HNF4A, MMP9 and TERT) remained associated with overall survival after age/T/N/M adjustment (HR per 1-SD increase, 2.31; 95% CI, 1.98-2.71) and achieved a locked external C-index of 0.701; exploratory GC-ANN achieved an external AUC of 0.962. Score-defined APRGSig-high macrophages showed broader predicted crosstalk, whereas donor-aware analysis identified an inflammatory/stress-associated MALAT1-high state. Multiplex immunofluorescence provided representative compartment-level localization of APRGSig-encoded proteins; four Visium sections showed directionally consistent positive spatial concordance between APRGSig scores and inferred MALAT1-high abundance. Formal GSK-3β-benzene/toluene models yielded Vina scores of -7.1/-7.4 kcal/mol and remained numerically persistent over 100 ns; bibliometrics of 311 publications highlighted proliferation, invasion and metastasis. At 2.5 mM benzene and 20 μM toluene, both exposures increased AGS clonogenic capacity, Matrigel traversal and wound closure (all P < 0.05). Conclusions: This framework integrates predictive, tissue-resolved, structural and phenotype-level evidence to prioritize exposure-linked molecular and immune hypotheses in GC.