Single-cell profiling of gastric cancer ascites reveals immunosuppressive remodeling and an epithelial-immune dual phenotype signature predictive for PD-1 immunotherapy responses

Background Gastric cancer (GC) with peritoneal metastasis frequently leads to malignant ascites (MA), a highly immunosuppressive “liquid tumor microenvironment” associated with poor prognosis. Although immune checkpoint blockade (ICB) demonstrate efficacy in some GC patients, treatment response to gastric cancer-associated peritoneal metastasis (GCPM) remains heterogeneous. The immune mechanisms driving this variability and predictive biomarkers remain unclear. Methods We performed single-cell transcriptomics and TCR/BCR repertoire analyses on paired MA and peripheral blood mononuclear cell (PBMC) samples from 10 advanced GC patients. Cellular clustering, trajectory inference, and intercellular communication analyses characterized immune remodeling. Clinical cohorts were used to validate prognostic and therapeutic predictive significance. Results Eight major cell types were identified: epithelial and fibroblast cells enriched in MA, while immune cell populations dominated PBMCs. MA exhibited significant remodeling of T cell, B cell, and myeloid cell populations. CX3CR1⁺ CD8⁺ Temra cells demonstrated clonal expansion, migratory characteristics, and an activated state associated with poor prognosis. B cells exhibited enriched memory subsets and plasma cell expansion, accompanied by functional reprogramming associated with poor prognosis. Myeloid analysis revealed cDC2 reprogramming, emergence of LAMP3⁺ regulatory dendritic cells, and immunosuppressive macrophages driven by CD47-SIRPA-TREM signaling for evasion. Notably, we identified a novel epithelial-immune dual-phenotype cell (EIDPC) population, validated by single-cell RNA sequencing and flow cytometry, exhibiting moderate malignant characteristics and potent immunoregulatory capacity. Transcriptomic and trajectory analyses suggest an epithelial origin with reprogramming toward immune evasion. Based on 10 EIDPC core genes, we developed the immune response signature of EIDPC (IRS-EIDPC), which accurately predicts PD-1 therapy response and prognosis in independent gastric cancer cohort (AUC 0.929). Conclusions This study reveals the immune landscape of malignant ascites in gastric cancer, elucidates lineage-specific reprogramming mechanisms, and confirms EIDPC as a transitional malignant subpopulation with potent immunomodulatory functions. The IRS-EIDPC signature provides reliable evidence for predicting immunotherapy response and survival, unveils novel mechanisms of immune plasticity in malignant gastric ascites, and paves new pathways for precision therapy.