송현호 (서울대학교 식품대사체 연구실)

석기태 (한림대학교 소화기연구소)

이도엽 (서울대학교 식품대사체 연구실)

송현호 (서울대학교 식품대사체 연구실)

최지은 (서울대학교 식품대사체 연구실)

김대환 (서울대학교 식품대사체 연구실)

석기태 (한림대학교 소화기연구소)

이도엽 (서울대학교 식품대사체 연구실)

Metabolic dysfunction-associated steatotic liver disease (MASLD) lacks reliable non-invasive biomarkers. We investigated whether disease-specific shifts in fecal metabolite correlations could yield superior diagnostic classifiers. Untargeted LC-MS profiling of fecal samples from healthy controls (NC, n=30) and MASLD patients (n=18) identified 331 metabolites. Group-specific pearson correlation matrices were compared using Fisher Z-tests, with 269 metabolite pairs remaining significant after FDR correction (FDR < 0.15). Six metabolites with the largest mean differences between NC and MASLD (Mean_Diff panel) and the top correlation-shifted pair, M-180/M-263, were evaluated in machine learning models (70/30 train-test split). External validation was performed using public plasma datasets from MASLD patients and healthy adults .

In healthy controls, M-180 and M-263 showed no correlation (r = -0.06, p = 0.76), but in MASLD patients, they exhibited a strong positive correlation (r = 0.99, p < 0.001). Individually, these metabolites showed minimal group differences. However, the interaction term of the M-180/M-263 pair (Group_Interaction model) achieved perfect discrimination (AUC = 1.00, sensitivity = 1.00, specificity = 1.00) without overfitting, outperforming the Mean_Diff panel (AUC = 0.72) and five other abundance-based models. In MASLD plasma, the M-180/M-263 correlation was consistently significant across three molecular subtypes (all p-value < 0.01) and absent in healthy plasma. These gut-to-blood signatures suggest a gut-liver axis mechanism, where microbial dysbiosis-derived M-180 increases concurrently with mitochondria-stress-associated M-263.

The gain-of-correlation between these two metabolites represents a robust, cross-matrix biomarker for MASLD, surpassing traditional abundance-based approaches. Correlation‑centric modeling highlights unseen yet important disease signals that conventional methods overlook.