Single-cell and spatial transcriptomic profiling reveals impaired tertiary lymphoid structure maturation associated with T cell exhaustion in gastric cancer

Background. Gastric cancer (GC) remains a leading cause of cancer-related mortality, and only a minority of patients derive durable benefit from immune checkpoint blockade (ICB). Tertiary lymphoid structures (TLS) are predictive of favorable ICB response in several solid tumors, yet their organization and functional context in GC remain incompletely defined. Methods. We integrated public single-cell RNA sequencing (scRNA-seq; GSE206785) and 10x Visium spatial transcriptomics (GSE251950) of human GC. Intercellular communication was inferred by CellChat v2 and NicheNet v2. Spatial CXCL13–CXCR5 co-localization was examined by Pearson correlation and Ripley’s K-cross analysis across three molecularly distinct specimens (GC1 MSI-H, GC2 EBV-positive, GC3 epithelial), each supported by matched H&E histology. A three-tier spatial TLS maturity score (None/Early/Mature) was developed and tested by a 144-combination sensitivity analysis. T-cell trajectory was reconstructed by Slingshot. Prognostic relevance was tested in TCGA-STAD; selected genes were checked by exploratory qPCR. Results. Eight major cell populations were resolved from ~ 50,000 cells. TLS signature genes were strongly enriched in B and T lymphocytes (χ² P = 2.86×10⁻⁹⁵; Cramér’s V = 0.42). Despite robust single-cell TLS expression, CXCL13–CXCR5 was spatially uncorrelated across all three specimens (representative: Pearson R = − 0.003, P = 0.85; Ripley’s K-cross below theoretical envelope). A three-tier maturity score showed that although TLS transcripts were widespread (739–1,083 Early spots per specimen), spots meeting mature-TLS criteria were exceedingly rare (GC1 MSI-H 0%, GC2 EBV-positive 0.33%, GC3 0.72%) and remained ≤ 1.05% across 144 parameter combinations; the immunogenic MSI-H specimen contained no mature spots at all. Matched H&E showed only sparse, non-organized lymphoid aggregates. Immunosuppressive CXCL12–CXCR4 was significantly co-localized (R = 0.136, P = 2.5×10⁻¹⁷). TLS and exhaustion scores correlated positively (single-cell R = 0.27; spatial R = 0.13). TLS signature did not stratify OS in TCGA-STAD (log-rank P = 0.47). Conclusions. GC harbors a widespread transcriptional TLS signal enriched in lymphocytes that only rarely progresses to spatial maturity (mature-TLS spots ≤ 1.05% across all specimens and parameters), with the immunogenic MSI-H specimen showing none. This maturation deficit co-occurs with progressive T-cell exhaustion and immunosuppressive signaling, motivating therapies that target the structural maturation of TLS rather than their transcriptional induction alone.