Single-cell atlas of gastric cancer reveals malignant epithelial evolution and regulatory reprogramming of the tumor microenvironment

The inflammation–intestinal metaplasia (IM)–carcinoma cascade has been proposed as a framework for gastric cancer (GC) development, yet the cell-level heterogeneity and microenvironmental remodeling underlying this progression remain poorly characterized. Here, we constructed a single-cell transcriptomic atlas by integrating scRNA-seq data from chronic gastritis (superficial, CGS), IM, cancer-adjacent, and tumor tissues through a unified analytical pipeline. Seven major cell lineages were resolved. Relative to CGS, IM and GC tissues exhibited a progressive contraction of epithelial compartments accompanied by expansion of immune and stromal populations. Copy number variation (CNV) inference identified two tumor-restricted malignant epithelial subgroups—one biased toward differentiation and the other enriched for inflammatory and epithelial–mesenchymal transition (EMT) signatures—as well as putative proto-malignant intermediates that coexisted with phenotypically normal epithelium. Cell–cell communication analysis indicated broadly augmented crosstalk between epithelial cells and T cells, myeloid cells, and fibroblasts, with prominent involvement of a CD44–extracellular matrix (ECM) axis. Pseudotime trajectory analysis placed malignant epithelium at late positions along gastric and pyloric mucosal cell differentiation backbones, coinciding with increasing CNV burden and enrichment of stem-like transcriptional programs. Gene regulatory network analysis revealed coordinated activity of lineage-specification modules (HNF4/CDX, NR1H4/ESRRA), proliferative regulons (MYC/TFDP1), and inflammatory/EMT-associated programs (FOSL1/REL/NF-κB). In independent cohorts, elevated expression of several malignant-epithelium-associated transcription factors—including HNF4A, KLF3, FOSL1, TCF7L2, BCL3, RELB, ONECUT2, and MAF—correlated with unfavorable overall survival. Collectively, these findings provide single-cell-resolution evidence consistent with the proposed three-stage model of gastric carcinogenesis and highlight candidate transcriptional regulators warranting further investigation as potential early-detection biomarkers.