Cross-Cohort Discovery and Independent Validation of Transcriptomic Programs and Candidate Genes in Oral Squamous Cell Carcinoma

Abstract

Background: Cross-cohort validation can distinguish reproducible oral squamous cell carcinoma (OSCC) expression signals from cohort- and platform-specific findings.

Objective: This study aimed to identify genes and pathways that replicate across independent OSCC microarray cohorts.

Methods: GSE30784 (discovery) and GSE25099 and paired GSE37991 (validation) comprised 264 OSCC and 107 non-malignant samples. Cohorts were normalized, annotated, and modeled separately using robust empirical-Bayes linear models, with 12,708 shared genes eligible for testing. Discovery required FDR < 0.05 and |log₂ fold change| ≥ 1.0. Replication required concordant direction, validation FDR < 0.05, and |log₂ fold change| ≥ 0.5 in both validation cohorts. Direction-specific GO and KEGG enrichment used the common gene universe; QC-flagged arrays were excluded in sensitivity analyses.

Results: GSE30784 yielded 1,167 DEGs (583 upregulated; 584 downregulated), of which 504 replicated in both validation cohorts (273 upregulated; 231 downregulated). CRISP3, MMP10, MMP13, MMP1, FAM3B, KRT4, TMPRSS11B, TYRP1, MMP12, and SERPINE1 had the largest minimum cross-cohort effects. Replicated upregulated genes were enriched in cytokine, ECM-receptor, integrin, IL-17, focal-adhesion, and PI3K-Akt pathways; downregulated genes were enriched in cornified-envelope and metabolic pathways. Sensitivity analysis retained 488 of 504 primary calls, including all top 20 candidates.

Conclusion: Independent validation identified reproducible OSCC transcriptional programs centered on matrix remodeling, adhesion, inflammation, and reduced epithelial-metabolic functions. These expression-based candidates require orthogonal and functional validation.