Exploration of Low-Abundance Tundrisphaera and Tissue 3-Hydroxybutyric Acid (β-HB) in Gastric Cancer Lymphovascular Invasion: A Multi- Cohort Multi-Omics Study with Clearly Defined Evidence Boundaries

Abstract Background Lymphovascular invasion (LVI) is one of the early routes of gastric cancer dissemination. Tissue-associated microorganisms and ketone-body metabolism may be associated with aggressive disease, but parallel between-group differences cannot establish a microbial–metabolic mechanism. This study integrated a discovery cohort with paired gastric tissue 16S rRNA and targeted metabolomics data, together with public transcriptomic, single-cell, methylation, dependency, and perturbation resources. Methods Clinical LVI labels were available for 66 gastric cancer cases, of which 65 had both 16S and metabolomics data (32 LVI-negative [V0] and 33 LVI-positive [V1] cases). We evaluated microbial diversity, community composition, genus-level differences, and a random forest classifier with rigorous control of information leakage, and implemented a mutually exclusive two-stage descriptive genus screen. The strict tier required a mean relative abundance of ≤ 0.01% in V0 and ≥ 0.05% in V1, and a pseudocount-adjusted fold change of ≥ 10; genera that did not meet the strict-tier criteria entered a relaxed additional tier, which required a V1 cumulative genus-table count of ≥ 20 and a ≥ 10-fold increase in mean abundance. Genera with a V0 cumulative count of 0 were treated as zero-baseline candidates, for which fold changes were not estimated. We subsequently analyzed the tissue β-HB signal and the paired bridging relationship between continuous Tundrisphaera abundance and β-HB. We evaluated 6 prespecified non-immune programs in GSE62254, TCGA-STAD, GSE246963, GSE237876, GSE163558, GSE239676, and GSE261635: ketone-body handling, butyrate metabolism, fatty acid degradation, EMT/ECM invasion, endothelial/angiogenic activity, and epithelial integrity. DepMap and LINCS analyses were used only to prioritize experimental candidates. Results The V1/V0 cumulative genus-table counts for Tundrisphaera were 33/1, the mean relative abundances were 0.002250%/0.000070%, and the pseudocount-adjusted fold change was 31.57; cumulative counts are not equivalent to the number of positive samples. The β-HB signal was higher in V1 (median, 68.49) than in V0 (median, 39.31; Mann–Whitney P = 0.0220318), but the metabolome-wide FDR was 0.9758. In the 65 paired samples, Tundrisphaera and β-HB showed a significant inverse bridging relationship (Spearman ρ = −0.6797; P < 0.001), with Tundrisphaera abundance and β-HB varying in opposite directions. In GSE62254 and TCGA-STAD, EMT/ECM and endothelial programs were associated with poorer survival. DepMap did not show widespread gastric cancer dependency on the candidate genes; LINCS reversal scores were used only to generate hypotheses. Conclusions The β-HB signal measured by targeted LC–MS was nominally elevated in LVI-positive gastric cancer tissues; the two-stage descriptive screen identified the low-abundance candidate Tundrisphaera in the relaxed additional tier, and a significant inverse bridging relationship was observed.