Single-cell and bulk transcriptome-based identification of myeloid-derived suppressor cell-associated immune evasion signatures and a prognostic model for gastric cancer

Abstract

Gastric cancer (GC) immunotherapy has limited efficacy. We developed a myeloid immunosuppressive niche score (GC-MIS) to quantify immune evasion and assess its association with the tumor microenvironment and patient prognosis. Two single-cell and three bulk transcriptomic cohorts were analyzed. A 12-gene MDSC-like signature (MIS) and an 8-gene tumor-intrinsic (TU) module were identified, and GC-MIS was defined as the standardized composite score of the two. Bulk cohort analyses included multivariable regression, Cox modeling, and prognostic evaluation (AUC, C-index, calibration, and decision curves). CellChat characterized epithelial–myeloid interactions, and in vitro GC cell lines exposed to inflammatory and hypoxic stimuli with pathway-specific inhibitors were assessed for PD-L1, STAT3 phosphorylation, and chemokine responses. MIS and TU modules were stable and moderately correlated at the single-cell level. In bulk cohorts, GC-MIS was positively associated with T-cell exhaustion and myeloid abundance, and inversely associated with antigen presentation (adjusted R² = 0.33–0.41). High GC-MIS independently predicted worse survival (pooled HR = 1.32, P  < 0.001). A GC-MIS–based model combined with clinical variables showed consistent but moderate prognostic performance across cohorts (5-year AUC ≈ 0.66). Pathway analyses implicated JAK–STAT, HIF-1, and PD-1/PD-L1 signaling, and GC-MIS correlated with CXCL8, IL6, and CCL2 communication axes. In vitro, inflammatory and hypoxic stimuli induced PD-L1 and STAT3 activation, which were significantly attenuated by targeted inhibitors ( P  < 0.05). GC-MIS is a reproducible transcriptional score associated with features of the tumor immune microenvironment and prognosis in gastric cancer. It reflects tumor–myeloid interaction programs and may serve as a prognostic biomarker. The associated inflammatory and hypoxia-linked signaling pathways warrant further investigation in functional and immunotherapy-treated cohorts.