Blood Test for Oral Cancer? MMP Enzymes Show Promise as Prognostic Biomarkers

Oral squamous cell carcinoma (OSCC) is one of the deadliest cancers in South Asia, and a major reason is simple: it is usually caught too late. Now, a research team from Vellore Institute of Technology and Government Royapettah Hospital in Chennai, India, has taken a systematic look at an entire family of enzymes implicated in tumor spread — the matrix metalloproteinases (MMPs) — and validated three of them as minimally invasive, serum-based biomarkers in a South Indian cohort. The study, published in Frontiers in Bioinformatics, combined large-scale computational screening with clinical ELISA measurements in 170 participants, and the results point toward a blood test that could help detect OSCC earlier and predict patient outcomes more accurately.

MMPs are a family of 23 zinc- and calcium-dependent proteolytic enzymes in humans that degrade the extracellular matrix (ECM) — the structural scaffold of collagen, fibronectin, elastin, and laminin that confines cells within tissues. When tumor cells or stromal cells in the tumor microenvironment upregulate MMP expression, they weaken the basement membrane, liberate growth factors, and create physical space for tumor expansion, migration, and intravasation into lymphatic vessels. This is the molecular machinery of invasion and metastasis. In OSCC specifically, cancer-associated fibroblasts, inflammatory cells, and endothelial cells secrete cytokines such as tumor necrosis factor-alpha and interleukins, along with growth factors like EGF and TGF-beta, which upregulate MMP gene expression and accelerate oral tumorigenesis. Despite decades of research on individual MMPs like MMP-2 and MMP-9, no study had systematically evaluated the collective influence of all 23 human MMPs on OSCC progression — until now.

The team, led by P. Sakthi Subhashini and supervised by S. Asha Devi at VIT, along with surgical oncologists T. Bharanidharan and D. Suresh Kumar, retrieved gene expression data from 12 microarray datasets in the NCBI Gene Expression Omnibus (GEO) and supplemented this with RNA-seq data from 258 OSCC tumor samples and 18 normal controls in The Cancer Genome Atlas (TCGA). Using the limma and edgeR R packages, they identified differentially expressed genes (DEGs) with a fold change greater than 2 and an adjusted p-value below 0.05. Batch effects across multiple microarray platforms were corrected using the SVA package. From this integrated transcriptomic landscape, six MMP family members emerged as consistently upregulated in OSCC tumor tissue: MMP-1, MMP-10, MMP-3, MMP-12, MMP-13, and MMP-9. Notably, MMP-1 was upregulated in all 13 datasets, followed by MMP-10 and MMP-3 in 12 datasets each, MMP-12 in 11, MMP-13 in 8, and MMP-9 in 7.

To understand what these enzymes are doing biologically, the researchers performed Gene Set Enrichment Analysis (GSEA) on TCGA tumor samples stratified by an MMP-based risk score. The results were striking: rather than pointing primarily at core signaling cascades, the analysis revealed that upregulated MMPs are deeply intertwined with metabolic reprogramming — a key emerging cancer hallmark. Oxidative phosphorylation (NES = 2.7160), fatty acid metabolism (NES = 2.0731), adipogenesis (NES = 1.73), and bile acid metabolism (NES = 1.71) were among the top enriched hallmark pathways. In KEGG gene sets, oxidative phosphorylation, the ribosomal pathway, carbon metabolism, and glyoxylate metabolism were significantly upregulated. Oncogenic signature analysis also highlighted E2F3 transcriptional pathways, STK33-associated signatures, and NRF2-oxidative stress response pathways. This suggests that MMP dysregulation in OSCC is not merely about ECM degradation — it is coupled to enhanced energy production, biosynthetic support for tissue remodeling, and proliferation-related signaling that together drive tumor progression.

The team then turned to survival modeling using Cox proportional hazards regression on the TCGA cohort, which provided the most comprehensive clinicopathological data. Univariate analysis revealed that tumor location, tobacco use, TNM stage III-IV, T stage, and N stage were significantly associated with survival. However, after adjusting for confounders in multivariate Cox regression, only tumor location remained an independent prognostic factor (HR = 1.787, 95% CI = 1.097–2.912, p = 0.0198). Interestingly, none of the six individual MMPs emerged as statistically significant independent prognostic factors on their own — a finding that underscores the complexity of cancer prognosis and the limitations of single-gene biomarkers.

But when the researchers combined the six-MMP expression signature with the conventional TNM staging system, predictive accuracy improved. The standalone MMP signature model achieved a concordance index (C-index) of 0.552, the TNM stage model 0.585, and the integrated combination model 0.609. The area under the ROC curve (AUC) for the combined model reached 0.63, compared to 0.58 for the MMP signature alone and 0.57 for TNM staging alone. Kaplan-Meier survival analysis showed that patients stratified into high-risk groups using the median MMP-based risk score had significantly reduced survival probability (p = 0.045), while TNM staging alone showed only borderline significance (p = 0.088). A nomogram integrating the MMP signature with TNM staging yielded AUC values of 0.655 for 3-year survival and 0.612 for 5-year survival, with calibration graphs confirming strong consistency between predicted and observed outcomes.

The most compelling part of the study came next: clinical validation. The team recruited 170 participants from the Department of Surgical Oncology at Government Royapettah Hospital in Chennai — 79 histologically confirmed, treatment-naive OSCC patients and 91 age- and ethnicity-matched healthy controls. From the six computationally identified MMPs, they prioritized MMP-1, MMP-10, and MMP-9 for serum validation based on their consistent cross-dataset manifestation and documented roles in ECM remodeling and oral tumor invasion. Using sandwich ELISA kits from CUSABIO, they quantified serum protein concentrations and found dramatic elevations in OSCC patients. Median serum MMP-1 was 19.70 ng/mL in OSCC versus 4.290 ng/mL in controls (p < 0.0001). Median serum MMP-10 was 1141 pg/mL in OSCC versus 259.5 pg/mL in controls (p < 0.0001). Median serum MMP-9 was 3.311 ng/mL in OSCC versus 2.379 ng/mL in controls (p = 0.0002). All three comparisons were highly statistically significant.

ROC curve analysis further demonstrated the diagnostic potential of these serum biomarkers. MMP-10 yielded the highest AUC at 0.8843, followed by MMP-1 at 0.8614 and MMP-9 at 0.6653. An AUC above 0.85 is generally considered excellent discrimination between diseased and healthy populations, placing MMP-1 and MMP-10 in that category. The OSCC cohort was predominantly male (74.68%), aged 35–54 (69.6%), with buccal mucosa as the most common tumor site (81.01%), and a striking 87.34% had a history of tobacco usage including cigarette smoking and betel nut chewing — reflecting the lifestyle-related risk factors driving OSCC incidence in South India.

The implications are significant for a region where oral cancer accounts for 55.6% of global cases and 56.5% of related deaths, with India alone representing 85.6% of OSCC incidence among medium-human-development-index nations. Healthcare disparities and delayed diagnosis remain major contributors to advanced-stage presentation and high mortality. A minimally invasive blood test measuring serum MMP-1, MMP-10, and MMP-9 could complement existing screening approaches, particularly in resource-constrained settings where tissue biopsies and imaging may not be readily accessible. The researchers emphasize that these biomarkers are intended as adjuncts to — not replacements for — conventional clinicopathological staging, and that the modest concordance values of the combined prognostic model indicate the need for further validation in larger, multicentre cohorts.

Limitations remain. The computational approach relied on retrospective gene expression data from public repositories, which may introduce population-related bias and dataset heterogeneity. The TCGA cohort may not fully represent the South Indian OSCC population. Serum validation was performed for only three of the six candidate MMPs, and the lack of matched tissue samples prevented tissue-level validation. The study evaluated diagnostic potential rather than therapeutic efficacy. Nevertheless, this is the first study to integrate computational prediction with experimental validation of MMPs as biomarkers in South Indian OSCC patients, and it adds to growing evidence from Iranian, Taiwanese, Indian, Egyptian, and American cohorts linking elevated MMP protein concentrations to oral tumor progression. As the search for early-detection tools for one of Asia’s deadliest cancers intensifies, these zinc-dependent molecular scissors — floating freely in patient serum — may prove to be exactly the minimally invasive window clinicians have been looking for.

Subject of Research: MMP biomarkers for oral squamous cell carcinoma

Article Title: Profiling the MMP family genes as prognostic biomarkers in oral squamous cell carcinoma: from computational discovery to clinical validation in a South Indian cohort

Article References: Sakthi Subhashini, P., Bharanidharan, T., Suresh Kumar, D., & Asha Devi, S. (2026). Profiling the MMP family genes as prognostic biomarkers in oral squamous cell carcinoma: from computational discovery to clinical validation in a South Indian cohort. Frontiers in Bioinformatics, 6, Article 1955795. https://doi.org/10.3389/fbinf.2026.1955795

Image Credits: AI Generated

DOI: 10.3389/fbinf.2026.1955795

Keywords: oral cancer, OSCC, MMP, biomarkers, ELISA, bioinformatics, South India, serum biomarkers, prognosis, early detection, Profiling, family