Diagnostic utility of FGF-21 and GDF-15 for differentiating primary mitochondrial diseases from lysosomal storage...

Differentiating primary mitochondrial disease (PMD) from disorders associated with secondary mitochondrial dysfunction may be challenging because of overlapping biochemical and clinical features. In the present study, the group designated as secondary mitochondrial dysfunction consisted specifically of lysosomal storage disorders. Single biomarkers, such as FGF-21 and GDF-15, may have limited discriminatory power in this setting. In this study, multivariate metabolite panels distinguishing PMD, lysosomal storage disorders associated with secondary mitochondrial dysfunction (SMD), and healthy controls (HC) were developed and internally validated using a leakage-controlled machine learning framework with interpretable outputs.

This prospective study included 88 participants: 28 patients with genetically confirmed PMD, 30 patients with genetically confirmed lysosomal storage disorders associated with secondary mitochondrial dysfunction, and 30 healthy controls (HC). Plasma amino acids were quantified by LC-MS/MS, urinary organic acids by GC-MS, and serum FGF-21 and GDF-15 by ELISA, yielding 77 variables. Three pairwise tasks were modeled using nested cross-validation, stability-ranked feature selection, and elastic net logistic regression. Performance was summarized using mean outer-fold AUC, accuracy, sensitivity, and specificity, with BCa confidence intervals; DeLong confidence intervals were calculated from pooled out-of-fold predictions.

The comparison between the lysosomal storage disorder group and the PMD group showed the highest discrimination within the present cohort (mean outer-fold AUC, 0.988; pooled-OOF DeLong 95% CI, 0.952-1.000; mean outer-fold accuracy, 0.948). Parsimonious panels were identified: SMD vs. PMD-asparagine, lactate, GABA, homocystine, hydroxylysine; SMD vs. HC-asparagine, pyruvate, GDF-15, histidine, urinary 4-hydroxyphenylpyruvic acid; PMD vs. HC-GDF-15 and lactate. SHAP analysis supported the consistency in the mitochondrial redox and amino acid pathways.

A multivariate metabolomic approach provided discriminatory information between genetically confirmed PMD, selected lysosomal storage disorders associated with secondary mitochondrial dysfunction, and healthy controls. The highest performance was observed between the two patient groups. However, these findings are specific to the diagnostic composition of the present cohort and should not be interpreted as establishing a universal distinction between primary and secondary mitochondrial dysfunction. External validation in larger, independent, and diagnostically broader cohorts is required before clinical implementation.