A multi-modal transcriptomic integration suggests spatial co-localization of a stromal neural-adhesion program with tertiary lymphoid structure niches in gastric cancer

Tertiary lymphoid structures (TLS) are immune niches linked to antitumor responses. Whether neural-like programs spatially associate with TLS in gastric cancer (GC) is unclear. We integrated scRNA-seq (GSE183904), Visium spatial transcriptomics (GSE251950), and TCGA-STAD RNA-seq. A 15-gene neural-like (Neuro) program and TLS program were evaluated by Spearman correlation, Lee’s L, and a ReLU-corrected coupling score. Ligand–receptor co-expression and supportive CellChat domain inference were assessed; GSE245704 provided limited corroboration. Neuro activity localized mainly to fibroblasts and was driven by neural adhesion/glial markers. Unadjusted Neuro–TLS associations were positive (Spearman rho 0.143–0.559; Lee’s L 0.038–0.317; permutation p ≤ 0.006), but adjustment for microenvironmental components attenuated both measures and reversed their direction in a minority of sections, indicating partial composition dependence. NeuroHigh/TLSHigh domains had higher exhaustion-signature scores ( p = 0.002). GSE245704 showed a directionally concordant association (rho = 0.193; Lee’s L = 0.109). In TCGA-STAD, Coupling_ReLU was associated with inferior overall survival after adjustment (HR = 1.291, 95% CI 1.062–1.569; p = 0.010); however, Neuro alone was significant (HR = 1.322; p = 0.001), suggesting that the prognostic signal may be partly driven by the stromal neural-adhesion component. A stromal neural-adhesion program co-localizes with TLS niches, but the relationship is partly composition-dependent. Ligand–receptor co-expression and CellChat provide supportive domain-level inference rather than direct cell–cell interaction. These findings are hypothesis-generating and require larger cohorts, higher-resolution spatial profiling, and functional validation.