A saliva-derived metabolic signature with multi-omics support for oral squamous cell carcinoma diagnosis.

Background: Oral squamous cell carcinoma (OSCC) lacks practical and non-invasive biomarkers for case detection. Saliva is an attractive source for biomarker discovery; however, many reported candidates lack support from independent biological data layers. Methods: We performed untargeted salivary metabolomics in 40 patients with OSCC and 40 healthy controls. Differential metabolites were identified using multivariate and univariate analyses. Candidate diagnostic metabolites were prioritized by support vector machine-recursive feature elimination, random forest, and receiver operating characteristic (ROC) analyses. The biological context of the salivary findings was further examined using two GEO transcriptomic cohorts, the TCGA-HNSC cohort, a single-cell RNA sequencing dataset, and public protein-level resources including the Human Protein Atlas and CPTAC. Results: OSCC and control samples showed distinct salivary metabolic profiles, and 19 annotated endogenous differential metabolites were retained for downstream analysis. These alterations were mainly enriched in amino acid- and lipid-related pathways, including arginine and proline metabolism. Within the discovery cohort, internal cross-validation and feature ranking converged on a seven-metabolite panel; however, because feature selection and evaluation were performed on the same 80 samples without nested cross-validation, performance estimates are subject to optimism bias and are not reported here as validated diagnostic metrics. In GSE6631 and GSE37991 showed concordant downregulation of linoleic acid metabolism, arachidonic acid metabolism, bile acid-related programs, fatty acid metabolism, and peroxisome pathways in tumor tissue. Single-cell analysis further suggested reduced activity of curated metabolic modules in malignant cells and tumor-associated tissue states, though these comparisons did not account for patient-level dependence. At the protein level, ALDH1A1 and VDR were decreased, whereas SLC7A11 was increased in tumor tissue. Conclusion: OSCC was associated with a saliva-derived metabolic signature supported by pathway-level concordance across multiple public data layers. The overall pattern suggests lipid remodeling, reduced bile acid-related signaling, altered amino acid metabolism, and enhanced redox defense. These findings support further targeted quantification and external validation of saliva-based biomarker candidates for OSCC, while the current panel should be regarded as a preliminary discovery-set candidate signature.