Nutritional intake and gut microbiome composition predict Parkinson’s disease


Background: Models to predict Parkinson’s disease (PD) incorporating alterations of gut microbiome (GM) composition have been reported with varying success.

Objective: To assess the utility of GM compositional changes combined with macronutrient intake to develop a predictive model of PD.

Methods: We performed a cross-sectional analysis of the GM and nutritional intake in 103 PD patients and 81 household controls (HCs). GM composition was determined by 16S amplicon sequencing of the V3-V4 region of bacterial ribosomal DNA isolated from stool. To determine multivariate disease-discriminant associations, we developed two models using Random Forest and support-vector machine (SVM) methodologies.

Results: Using updated taxonomic reference, we identified significant compositional differences in the GM profiles of PD patients in association with a variety of clinical PD characteristics. Six genera were overrepresented and eight underrepresented in PD patients relative to HCs, with the largest difference being overrepresentation of Lactobacillaceae at family taxonomic level. Correlation analyses highlighted multiple associations between clinical characteristics and select taxa, whilst constipation severity, physical activity and pharmacological therapies associated with changes in beta diversity. The random forest model of PD, incorporating taxonomic data at the genus level and carbohydrate contribution to total energy demonstrated the best predictive capacity [Area under the ROC Curve (AUC) of 0.74].

Conclusion: The notable differences in GM diversity and composition when combined with clinical measures and nutritional data enabled the development of a predictive model to identify PD. These findings support the combination of GM and nutritional data as a potentially useful biomarker of PD to improve diagnosis and guide clinical management.


Parkinson’s disease, gut microbiota, gastrointestinal microbiome, dysbiosis, medication, biomarker, prediction model

Link to Publisher Version (URL)


This document is currently not available here.

Find in your library