Metagenomics of the Gut Microbiome in Parkinson's Disease: Prodromal Changes

Abstract

Objective: Prior studies on the gut microbiome in Parkinson's disease (PD) have yielded conflicting results, and few studies have focused on prodromal (premotor) PD or used shotgun metagenomic profiling to assess microbial functional potential. We conducted a nested case-control study within 2 large epidemiological cohorts to examine the role of the gut microbiome in PD.

Methods: We profiled the fecal metagenomes of 420 participants in the Nurses' Health Study and the Health Professionals Follow-up Study with recent onset PD (N = 75), with features of prodromal PD (N = 101), controls with constipation (N = 113), and healthy controls (N = 131) to identify microbial taxonomic and functional features associated with PD and features suggestive of prodromal PD. Omnibus and feature-wise analyses identified bacterial species and pathways associated with prodromal and recently onset PD.

Results: We observed depletion of several strict anaerobes associated with reduced inflammation among participants with PD or features of prodromal PD. A microbiome-based classifier had moderate accuracy (area under the curve [AUC] = 0.76 for species and 0.74 for pathways) to discriminate between recently onset PD cases and controls. These taxonomic shifts corresponded with functional shifts indicative of carbohydrate source preference. Similar, but less marked, changes were observed in participants with features of prodromal PD, in both microbial features and functions.

Interpretation: PD and features of prodromal PD were associated with similar changes in the gut microbiome. These findings suggest that changes in the microbiome could represent novel biomarkers for the earliest phases of PD. ANN NEUROL 2023;94:486-501.

Figure 1.. Parkinson’s disease gut microbiome study design.

A. STORMS flowchart outlining enrollment, data collection, and B. processing for our study of human gut metagenomes in prodromal and clinical Parkinson’s disease (PD) [86]. Since symptoms of the disease include changes to gastrointestinal transit time, both “healthy” and constipated (non-PD) control populations were also included.

Figure 2.. PD is minimally associated with overall microbiome structure.

A. Univariate R2 (Bray-Curtis PERMANOVA) explained by key covariates, with corresponding PCoA ordination overlaid with phenotype group (PD/PPS/Constipated Controls/Healthy Controls). PERMANOVA shows nominal p-values. B. Distributions of taxonomic Bray-Curtis dissimilarities by phenotype group. Plot shows distribution of Bray-Curtis dissimilarities within group (eg within Recently Diagnosed PD, within PPS, etc. as well as between groups (dissimilarities between Constipation and Health Controls, etc). Colors represent the group plotted, as indicated on legend.

Figure 2.. PD is minimally associated with overall microbiome structure.

A. Univariate R2 (Bray-Curtis PERMANOVA) explained by key covariates, with corresponding PCoA ordination overlaid with phenotype group (PD/PPS/Constipated Controls/Healthy Controls). PERMANOVA shows nominal p-values. B. Distributions of taxonomic Bray-Curtis dissimilarities by phenotype group. Plot shows distribution of Bray-Curtis dissimilarities within group (eg within Recently Diagnosed PD, within PPS, etc. as well as between groups (dissimilarities between Constipation and Health Controls, etc). Colors represent the group plotted, as indicated on legend.

Figure 3.. Individual taxa significantly associated with PD phenotypes.

The most significantly enriched and depleted taxa and pathways associated with phenotype group identified via feature-wise analyses (MaAsLin 2), These generally exhibit a gradient across the phenotype groups. All analyses consider healthy controls as the reference group, and all p-values are presented with FDR correction.

Figure 4.. The PPS microbiome lies on a taxonomic and functional continuum between healthy controls and PD.

A. Taxa and B. pathways are similarly related to PD and PPS, as identified by beta coefficients from feature-wise analyses (MaAsLin 2). X axis indicates coefficients from PD vs. control comparison, Y axis PPS vs. control. Multivariate models were adjusted for batch, age, sex, adherence to the Mediterranean Dietary Pattern (Med diet), pack years smoking and Body Mass Index (BMI).

Figure 5.. Overall gut microbial community structure moderately differentiates PD patients.

Receiver operating characteristic (ROC) performance of a random forest (RF) classifier based on microbial taxa and pathways discriminating between phenotype groups. Classification was performed for the (PD vs. control) comparison.

Figure 6.. Abundance of strict anaerobes associated with reduced inflammation by study group.

Total (sum) abundance, stratified by phenotype group, of strict anaerobes associated with production of short-chain fatty acids. Sums include abundance of Eubacterium rectale, Roseburia inulinivorans, Roseburia intestinalis, Eubacterium hallii, Anaerostipes hadrus, and Faecalibacterium prausnitzii [61]. P-values (nominal) are reported for pair-wise Wilcoxon tests. We tested for linear trend across phenotype groups by fitting a logistic regression model to ordinal phenotype (healthy, constipation, PPS, PD) in relation to summed species abundance adjusting for sequencing batch, age, pack-years smoking, BMI, and MED diet adherence.

Figure 7.. A-priori targeted species selected based on association with PD in prior published studies.

We focused on species available in our dataset within genera previously associated with PD: Akkermansia[, , , , –32], Bifidobacterium[, , , , , , –33], Hungatella[62], Lactobacillus[15, 23, 32], Blautia[10, 15, 26, 30], Oscillibacter, Prevotella[9, 10, 19, 22, 33], Eubacterium[9], Roseburia[10, 18, 26, 30, 32], Lachnospira[18, 63], and Faecalibacterium[15, 19, 26, 34]. Based on the distribution among the a-priori species considered, species abundance was categorized into tertiles, or binary presence/absence if tertile categorization was not possible due to low abundance. Plotted are OR and 95% based on logistic regression adjusted for batch, age, sex, adherence to the Mediterranean Dietary Pattern (Med diet), pack years smoking and Body Mass Index (BMI). Based on the distribution among the a-priori species considered, species abundance was categorized into tertiles (top vs. bottom plotted), or binary presence/absence if tertial categorization was not possible due to low abundance.