Serum metabolomic profiling reveals load-specific adaptations to resistance training

Resistance training (RT) imposes repeated mechanical stimuli triggering systemic metabolic reprogramming whose molecular underpinnings remain incompletely characterized. Univariate approaches fail to capture pathway-level shifts defining the metabolomic response to RT in protocols differing in load.

To evaluate whether high-load (HL) and low-load (LL) RT protocols performed to volitional failure induce serum biochemical patterns specific to training status and loading condition, undetectable by univariate approaches.

Seventeen healthy young men completed an 8-week RT intervention (HL: 80% 1-RM, n = 9; LL: 30% 1-RM, n = 8). Fasting serum profiles were acquired by untargeted ¹H-NMR spectroscopy. Univariate comparisons used paired t-tests and one-way ANOVA with Benjamini-Hochberg correction. Random Forest models were validated by stratified 5 × 5-fold cross-validation and 1000-iteration permutation testing, with group separation assessed by sensitivity, specificity, and AUC. Discriminant metabolites were independently mapped onto established metabolic pathways to assess biochemical plausibility.

Of 10 metabolites significantly altered, five were consistently modulated across both protocols (3-hydroxyisovalerate, 3-hydroxybutyrate, acetone, isobutyrate, and lactate), reflecting shared adaptations in amino acid turnover and ketone body metabolism. Training-status separation achieved AUC = 1.00 (p < 0.001), with 3-hydroxyisovalerate as the dominant feature. Load-specific divergence was captured only by an exploratory multivariate signature of choline, glucose, and alanine (AUC = 0.94; p = 0.008), none of which was individually significant in univariate testing. Pathway integration demonstrated that discriminant metabolites are consistently related to well-established metabolic pathways: leucine catabolism, ketone body turnover, and glycolytic-oxidative rebalancing.

RT induces biologically coherent, load-modulated serum metabolomic shifts detectable only through multivariate analysis. These findings are hypothesis-generating and require external validation in independent cohorts before applied implementation.