Metabolomic signatures supporting clinical stratification of Parkinson disease and atypical parkinsonism.

BACKGROUND

Accurate differential diagnosis of parkinsonian syndromes remains challenging due to overlapping clinical features between Parkinson's disease (PD) and atypical parkinsonism, including progressive supranuclear palsy (PSP) and multiple system atrophy (MSA). Systemic metabolic profiling may capture disease-related biological differences beyond clinical phenotyping.

OBJECTIVES

To investigate whether serum metabolomic signatures, analyzed through a hierarchical ensemble machine learning framework, are associated with distinct clinical phenotypes of degenerative parkinsonian syndromes.

METHODS

In this prospective single-center pilot study, 134 patients with parkinsonian syndromes (PD n = 87; PSP n = 22; MSA n = 25) and 132 healthy controls were consecutively enrolled. Untargeted serum metabolomic profiling was performed using gas chromatography-mass spectrometry. A multistep ensemble classification strategy was applied to evaluate discrimination across diagnostic categories, and model interpretability was assessed using Shapley additive explanations.

RESULTS

A total of 241 metabolites were consistently detected. Ensemble models demonstrated high internal discrimination between PD and atypical parkinsonism and satisfactory separation between PSP and MSA. Misclassifications occurred primarily at later stratification steps. Interpretability analyses identified convergent metabolic patterns across algorithms, highlighting alterations related to carbohydrate metabolism, redox balance, lipid homeostasis, and phosphate-associated pathways.

CONCLUSIONS

Serum metabolomic profiles were associated with distinct clinical phenotypes of degenerative parkinsonian syndromes in this prospective pilot cohort. These findings support the biological relevance of systemic metabolic alterations in parkinsonism and provide feasibility data for future multicenter validation studies.