Metabolic modeling links gut microbiota to metabolic markers of Parkinson's disease

Abstract

Human gut microbiota have been implicated in metabolic disruptions in Parkinson's disease (PD). However, the underlying mechanisms linking gut microbiota to these disease-related metabolic changes remain largely unknown. In this study, we applied constraint-based metabolic modeling to identify potential causal links between compositional shifts in gut microbiota in PD and metabolic blood markers of PD. We personalized in silico whole-body metabolic models with gut metagenomics of 435 PD patients and 219 healthy controls and profiled in silico gut microbiome influences on 116 blood metabolites with replicated associations with PD diagnosis. Our analysis identified a reduced capacity of the PD host-microbiome co-metabolism to produce L-leucine and leucylleucine in blood. These metabolic predictions were traced back to lower L-leucine production of Roseburia intestinalis and higher L-leucine consumption by Methanobrevibacter smithii in PD microbiomes. We further predicted reduced host-microbiome production capacities of butyrate, myristic acid, and pantothenate in the blood of PD patients and linked these associations to reduced relative abundances of Faecalibacterium prausnitzii. Finally, lower nicotinic acid production capacities were predicted in PD patients, which were associated with increased relative abundances and increased nicotinic acid consumption of Ruthenibacterium lactatiformans in PD. In conclusion, we predicted that the gut microbiome can drive altered blood levels of six metabolites in PD and identified candidate microbial species that may influence these metabolic alterations. These findings may facilitate the development of novel therapies targeting the gut-brain axis in PD.

Keywords: Gut microbiome; biomarkers; host-microbiome co-metabolism; metabolic modeling; parkinson’s disease.

Graphical abstract

Figure 1.

Schematic overview of the study design created with BioRender. a) Gut metagenomic reads were mapped onto the microbial metabolic reconstructions in the AGORA2 and APOLLO resources. The corresponding microbial reconstructions were then combined into microbiome community models and personalized with the species-level relative read abundances. The personalized microbiome community models were joined with male and female WBMs corresponding to the sex of the microbiome sample donor and were given an average Western metabolic diet. Metabolic blood fluxes were predicted for 116 metabolites with previously reported metabolomic PD associations and were statistically analyzed to find metabolites whose alterations may be driven by compositional changes in PD microbiomes. b) For the metabolites with altered metabolic blood flux predictions, we identified metabolite-associated subsets of the gut microbiomes by calculating which microbial species in the WBMs could have contributed to the changed predicted metabolic fluxes. The species relative abundances of these microbial subsets were then correlated with their corresponding metabolic flux predictions to identify potential key microbes for metabolites with altered metabolite production potentials in PD patients.

Figure 2.

Predicted host-microbiome metabolite production potentials in blood of PD patients and neurologically healthy controls. (a-f) Predicted fluxes in PD patients (red) and neurologically healthy controls (blue) for selected metabolites with the strongest associations with PD status. The shown predicted fluxes in mmol/day/person are log2 transformed and normalized via z-transformation. Identical associations with PD status are shown for leucylleucine and L-leucine, indicating that the predicted gut microbiome influences on L-leucine and leucylleucine fluxes in blood occur through a shared metabolic pathway. G. Table summarizing the logistic regression results of the selected metabolites in a-f. The number of samples for which fluxes could be predicted is shown for PD patients and controls. The regression $\mathit{\beta }$ shows the log odds ratios for PD status against the predicted fluxes. The negative values $\mathit{\beta }$ indicate that the predicted blood flux potentials of PD patients are associated with lower predicted fluxes compared to WBMs of controls. The 2.5% and 95% CI further represent the 95% confidence intervals.

Figure 3.

(a-f). Sensitivity of the predicted fluxes toward removing a microbial species. The bar plots show the mean average reduction in predicted fluxes (x-axis) upon removing a microbial species (y-axis). The error bars indicate the 95% confidence intervals for the means. The confidence intervals and the mean averages were calculated from a population of 50,000 bootstrapped samples. The color annotations represent the previously found shifts in relative abundances in the analyzed PD microbiomes. Blue bars indicate microbial species for which no shift in PD microbiomes has been detected, while red bars represent the microbial species with decreased relative abundances. Lastly, the microbial species with higher relative abundances in PD microbiomes are represented by the green bars. The largest reductions in fluxes were found when removing B. uniformis, B. vulgatus, and F. prausnitzii for all shown metabolites. However, only F. prausnitzii has been previously found to associate with PD status. The bar plots show only the top 20 most flux-sensitive microbial species. (g). Spearman correlations between predicted metabolic fluxes in blood and relative microbial species abundances. Correlations are shown for six selected metabolites with lower predicted fluxes in PD patients and microbial species that contributed to the predicted fluxes. Correlation coefficients of zero represent microbial species that could not contribute to the predicted blood fluxes of the associated metabolites. Shown are the sets of microbial species that could together contribute to, on average, 95% of the total potential gut microbiome contributions to their corresponding metabolite blood fluxes.

Figure 4.

Microbial combinations of metabolic contributors to the predicted blood fluxes that best correlated with the predicted fluxes. Shown are only the largest sets of microbial combinations that improved the Spearman correlation coefficients with the predicted blood fluxes by $|\mathit{\rho }|>0.05$ compared to any smaller combination of microbial species. a. shows the microbial species that together best predicted the blood fluxes, the associated Spearman correlation, and the shift in relative abundances in PD microbiomes. The microbial combinations for butyrate and myristic acid strongly correlated with their associated blood fluxes, meaning that most of the variance in the predicted butyrate and myristic acid blood fluxes could be explained by these microbial species. b. visualizes the identified associations between the shown microbial species and the blood production potentials of the selected metabolites. The size of each stratum shows the number of connections for the microbial species (left stratum) and the selected metabolites (right stratum).