Metagenomics of Parkinson's disease implicates the gut microbiome in multiple disease mechanisms

Abstract

Parkinson's disease (PD) may start in the gut and spread to the brain. To investigate the role of gut microbiome, we conducted a large-scale study, at high taxonomic resolution, using uniform standardized methods from start to end. We enrolled 490 PD and 234 control individuals, conducted deep shotgun sequencing of fecal DNA, followed by metagenome-wide association studies requiring significance by two methods (ANCOM-BC and MaAsLin2) to declare disease association, network analysis to identify polymicrobial clusters, and functional profiling. Here we show that over 30% of species, genes and pathways tested have altered abundances in PD, depicting a widespread dysbiosis. PD-associated species form polymicrobial clusters that grow or shrink together, and some compete. PD microbiome is disease permissive, evidenced by overabundance of pathogens and immunogenic components, dysregulated neuroactive signaling, preponderance of molecules that induce alpha-synuclein pathology, and over-production of toxicants; with the reduction in anti-inflammatory and neuroprotective factors limiting the capacity to recover. We validate, in human PD, findings that were observed in experimental models; reconcile and resolve human PD microbiome literature; and provide a broad foundation with a wealth of concrete testable hypotheses to discern the role of the gut microbiome in PD.

Fig. 1. The STORMS flowchart.

Following reporting guidelines for human microbiome research, we show the step-by-step process by which the study was conducted, starting with subject selection, enrollment and data collection for persons with PD (a) and neurologically healthy controls (b), sequencing and bioinformatics pipeline (c), and statistical analyses (d, e).

Fig. 2. PD-associated species nominated by consensus of MaAsLin2 and ANCOM-BC.

Analysis included N = 724 biologically independent samples from 490 PD and 234 neurologically healthy control (NHC) subjects. a 257 species (denoted by circles in the plot) were tested in microbiome-wide association study (MWAS) with two statistical methods: MaAsLin2 and ANCOM-BC. The results are shown according to significance (−log10 of the FDR) achieved by MaAslin2 (Y-axis) vs ANCOM-BC (X-axis). Corresponding untransformed FDR values are provided in parentheses on the X and Y axes for easier interpretation. 84 species were nominated as PD-associated, defined by FDR < 0.05 by one method and FDR≤0.1 by the other: 68 achieved FDR < 0.05 by both methods, 10 achieved ANCOM-BC MaAsLin2 FDR < 0.05 by MaAsLin2 and FDR≤0.1 by ANCOM-BC, and 6 achieved FDR < 0.05 by ANCOM-BC and FDR≤0.1 by MaAsLin2. Blue: abundance is significantly elevated in PD. Red: abundance is significantly reduced in PD. Gray: not significantly associated with PD. Vertical and horizontal dashed lines denote points on X and Y axes that correspond to FDR=0.05. b, c Venn diagrams summarizing the overlap of species detected by MaAsLin2 (dotted circle) and ANCOM-BC (solid circle) at FDR≤0.1 (b), and at FDR < 0.05 (c).

Fig. 3. Differential abundances and effect sizes of PD-associated species.

Analysis included N = 724 biologically independent samples from 490 PD and 234 neurologically healthy control (NHC) subjects. Forty-six species that had at least 75% (and up to 750%) change in abundance in PD are shown here; for all 84 PD-associated species see Supplementary Fig. 3. a Distribution of relative abundances. Log2 transformed relative abundance values, as used in MaAsLin2, were used to generate the boxplots. Untransformed relative abundances, shown in parenthesis, are provided on the X-axis for easier interpretation of data. Boxplots show distribution of the data for PD (blue green) and NHC (orange). Each sample was plotted according to its abundance of the species. The left, middle, and right vertical boundaries of each box represents the first, second (median), and third quartiles of the data; that is, 25% of samples have abundance lower than the left border of the box, 25% of samples have abundances that are higher than the right border of the box. Absence of a box indicates 75% of samples had zero abundance. The lines extending from the two ends of each box represent 1.5x outside the interquartile range (range = (abundance value at 75% minus abundance value at 25%) x 1.5). Points beyond the lines are outlier samples. b Distribution of bias-corrected observed abundances (used in ANCOM-BC). Natural log transformed sampling-bias corrected observed abundances, as used in ANCOM-BC, were used here to generate the boxplots. Untransformed bias-corrected observed abundances, shown in parenthesis, are provided on the X-axis for easier interpretation of data. Description of boxplots are the same as for (a). c Absolute fold change in differential abundance in PD vs. NHC (squares and circles) and its 95% confidence interval (CI; solid and dashed lines), calculated from beta and standard errors estimated by MaAsLin2 (square with solid line of 95% CI), and ANCOM-BC (circle with dotted line of 95% CI). CI are truncated at 10x. Points and lines for fold changes and 95% CI were colored blue (elevated in PD) or red (reduced in PD).

Fig. 4. Network analysis reveals polymicrobial clusters of correlated species in the PD metagenome.

Analysis included N = 490 biologically independent samples from 490 PD cases. (For neurologically healthy control (NHC) network, N = 234, see Supplementary Fig. 4). a All species detected in PD gut metagenome were tested for correlation with one another using SparCC correlations and plotted in a network if their abundance correlated with at least one other species (i.e., |r| > 0.2 and uncorrected permuted P-value <0.05). Clusters were defined by Louvain algorithm and were randomly assigned a color and a number. Each circle (node) denotes a species and the curved lines (edges) connect correlated species. b PD-associated species that were identified via MWAS were mapped to the network, and highlighted in blue if elevated in PD, or red if reduced in PD. Correlation among PD-associated species were often positive, indicating abundances tended to rise together like Escherichia coli, Klebsiella pneumoniae, and Klebsiella quasipneumoniae in cluster #6, or decline together like the polysaccharide metabolizing species in cluster #13. Sometimes the correlation was negative indicating increase in abundance of one species correlated with decrease in the other, e.g., in cluster #2 the rise in abundances of some Streptococcus species in PD microbiome correlated with a decline in the abundances of other Streptococcus species. Enlargements with representative species labeled for (c) cluster #17, (d) cluster #2, and (e) cluster #8 labeled to the left, cluster #13 in the middle, cluster #8 on the upper left, and cluster #6 on the lower left (E. coli and Klebsiella species). For (c–e), boxes around species names are colored to represent their algorithm defined cluster shown in (a).

Fig. 5. Microbial gene-families and pathways with functional relevance to PD.

Analysis included N = 724 biologically independent samples from 490 PD and 234 neurologically healthy control (NHC) subjects. Overall, 15% of microbial gene-families (KO groups) and 30% of pathways (MetaCyc) tested were elevated or depleted in PD, a conservative estimate derived from consensus at FDR < 0.05 by two statistical methods (MaAsLin2 and ANCOM-BC) (Supplementary Data 9, 10). Examples are shown here, grouped by inferred functional relevance to PD (left panel). Data show increased levels of microbial activities that could contribute to PD pathogenesis (immunogenicity, alpha-synuclein aggregation, and creation of toxic metabolites), and reduced levels of protective mechanisms (anti-inflammation, and neuroactive and neuroprotective molecules). a Relative abundances (used in MaAsLin2). b Bias-corrected observed abundances (used in ANCOM-BC). c Fold change in PD vs. NHC as estimated by MaAsLin2 and ANCOM-BC. Y-axis: KO groups (identifiers begin with “K”, gene symbol in parentheses) and pathways (“PWY”). X-axis: Log2 transformed relative abundances (used in MaAsLin2) (a), or natural log transformed bias-corrected observed abundances (used in ANCOM-BC) (b), with untransformed relative abundances in parenthesis for easier interpretation; (c) fold change in differential abundance in PD vs. NHC (squares and circles) with 95% confidence interval (CI; solid and dashed lines), calculated from beta and standard errors estimated by MaAsLin2 and ANCOM-BC. Content in (a and b): Boxplots show frequency distribution of each KO and pathway in PD (blue green) and NHC (orange) metagenomes. Left, middle, and right vertical boundaries of each box represent first, second (median), and third quartiles of the data. The absence of a box indicates >75% of samples had zero abundance. The lines extending from the two ends of each box represent 1.5x outside the interquartile range (range=(abundance value at 75% minus abundance value at 25%)x1.5). Points beyond the lines are outlier samples. Content in (c): Fold change difference between PD and NHC in log2-transformed relative abundance (square with solid line of 95% CI, MaAsLin2), or natural log transformed bias-corrected observed abundances (circle with dotted line of 95% CI, ANCOM-BC). Blue: elevated in PD, red: reduced in PD. LPS: Lipopolysaccharide. LTA: Lipoteichoic acid. BLP: murein/bacterial lipoprotein. SCFA: short-chain fatty-acid. TMA: Trimethylamine.