Clinical characteristics of PD and control groups
The demographic and clinical information of 80 patients with PD and 77 control subjects enrolled in the first step of the study are summarized in Additional file 2: Table S1. A higher proportion of the PD group reported constipation than the controls.
(66.3% vs 12.3%, P < 0.01). The mean UPDRS part III scores and Hoehn-Yahr stages during the “on” and “off” state as well as NMSS scores are also included for patients with PD. There was no difference in age or sex between the two groups. There were more medical comorbidities including diabetes mellitus and hypertension in controls than in patients with PD. Of the clinical variables, the intensity of constipation was higher in PD patients with more advanced Hoehn-Yahr stages than in PD patients at early motor stages (P < 0.001 by one-way ANOVA).
Patients with PD have altered and more diverse gut microbiota
We then characterized the bacterial gut microbiota associated with PD by high-throughput sequencing of the V3–V4 region of the 16S ribosomal RNA gene. We measured the bacterial richness within each sample from both the PD and control groups using three different methods, the observed number of operational taxonomic units (OTUs), the Chao1 diversity index, and the Shannon entropy index. The bacterial gut microbiota from patients with PD was more diverse than those from controls by all the three estimators (P = 0.08 for the Chao1 diversity index, P = 0.02 for the observed species diversity index, and P < 0.001 for the Shannon entropy index; by Wilcoxon rank-sum test, Fig. 2a).
We calculated the β-diversity of the samples using the weighted UniFrac distances and the Bray-Curtis dissimilarity to identify possible differences between the bacterial components in the gut microbiota of patients with PD and controls. The principal coordinates analysis (PCoA) revealed that the gut microbiota of patients with PD was distinct from those of the controls (P < 0.001 by a Permutational MANOVA (PERMANOVA) implementation using Uni-Frac distances and Bray-Curtis dissimilarity, Fig. 2b). These findings indicate that the richness and diversity of the gut microbiota in patients with PD are significantly different from that of controls.
Alteration in gut microbiota between PD and control groups
A supervised comparison of the microbiota between PD and control groups was performed by linear discriminant analysis (LDA) effect size (LEfSe) analysis without any adjustments, which is often used to identify the presence and effect size of region-specific OTUs among different groups (Additional file 1: Supplementary Methods) [19]. We used a logarithmic LDA score cutoff of 2.0 to identify important taxonomic differences between the PD and control groups and found a notable difference in fecal microbiota between the PD and control groups based on LDA LEfSe analysis (Fig. 3a). We observed that the relative abundance of the Prevotella genus was higher in the control group than in the PD group, while the relative abundances of Parabacteroides, Verrucomicrobia, Akkermansia, Butyricimonas, Veillonella, Odoribacter, Mucispirillum, Bilophila, Enterococcus, and Lactobacillus were higher in patients with PD than in controls (LDA score (log10) > 2, Fig. 3a, b). GLMs with negative binomial distribution for bacterial abundances defined by sequence counts were used to model genera that were significantly different between the two groups after controlling for possible confounding factors such as age, sex, and diet (the mean daily intake amount of protein, carbohydrates, total fat, and dietary fiber). Microbiome differences between the groups were associated with Verrucomicrobia, Prevotella, Mucispirillum, Porphyromonas, Lactobacillus, and Parabacteroides (P < 0.05, Additional file 2: Table S2), suggesting an association between these genera and PD. The mean abundance of Prevotella was reduced by 46.6% in patients with PD compared to controls (Fig. 3c). The other genera were more abundant in patients with PD than in controls, but the absolute differences between the groups were smaller than that observed for Prevotella (data not shown).
Association between fecal microbiota and PD clinical characteristics
We next examined the association between fecal microbiota and PD clinical subtypes. Of 80 PD patients, 47 (58.8%) were initially presented with tremor while the others presented with non-tremor subtypes, including the akinetic-rigidity or postural instability and gait difficulty (PIGD) subtypes. We found a significant difference in fecal microbiota between the tremor and non-tremor subtypes based on LDA LEfSe analysis without specific confounder adjustment (Fig. 4a). GLMs with negative binomial distribution for bacterial abundances defined by sequence counts were used to model genera that were significantly different between tremor and non-tremor subtypes after controlling for possible confounding factors such as age, sex, and diet (the mean daily intake amount of protein, carbohydrates, total fat, and dietary fiber). The relative abundances of Clostridium, Verrucomicrobia, and Akkermansia were higher in the tremor subtype than in the non-tremor subtypes, whereas the relative abundances of Propionibacterium, Bacteroidia, Flavobacterium, Mogibacterium, Sutterella, Alcaligenacea Cupriavidus, and Desulfovibrio were higher in the non-tremor subtype (LDA score (log10) > 2, Fig. 4a, b). Of these classes and genera, the mean abundance of Bacteroides genus from Bacteroidia class was increased by 41.6% in patients with non-tremor subtype PD compared to patients with the tremor-subtype (Fig. 4c). Bacteroides abundance also correlated with motor symptom severity as defined by UPDRS part III motor scores (rho = 0.637 [95% confidence interval 0.474 to 0.758], P < 0.01 by Spearman correlation analysis).
Altered cytokine responses in PD patients with alterations in taxonomic compositions of the gut microbiota
Because gut microbial dysbioses are often accompanied by abnormal production of inflammatory cytokines such as IL-1β, IL-22, and IFNγ [15, 20], we next investigated whether there are specific cytokine responses correlated with the relative abundances of the candidate genera associated with risk for or motor severity of PD, including Verrucomicrobia, Prevotella, Mucispirillum, Porphyromonas, Lactobacillus, Parabacteroides, and Bacteroides. Among the nine cytokines we examined in the human Th1/Th2 cytokine panel, we found there was a correlation between Bacteroides and plasma concentrations of TNFα (rho = 0.638 [95% CI 0.102–0.887], P = 0.02, Fig. 5a); and a correlation between Verrucomicrobia abundance and plasma concentrations of IFNγ (rho = 0.545 [95% CI − 0.043–0.852], P = 0.05, Fig. 5b). Consistently, the plasma levels of TNFα and IFNγ were increased in the PD group than controls (TNFα: PD vs. control: 10.71 ± 3.89 pg/ml vs. 6.04 ± 2.73 pg/ml, P = 0.02; and IFNγ: PD vs. control: 37.22 ± 9.15 pg/ml vs. 28.39 ± 10.21 pg/ml, P < 0.01) (Fig. 5c).
We then evaluated plasma cytokine responses in an additional independent 240 participants, including 120 patients with PD (62.3 ± 7.8 years, 62.5% men) and 120 age and sex-matched controls (61.8 ± 8.3 years, 60.0% men) to confirm the altered cytokine responses in the PD disease state. The plasma concentrations of TNFα and IFNγ were consistently and significantly higher in patients with PD than in controls (TNFα: PD vs. control: 8.51 ± 4.63 pg/ml vs. 4.82 ± 2.23 pg/ml, P < 0.01; and IFNγ: PD vs. control: 38.45 ± 7.12 pg/ml vs. 32.79 ± 8.03 pg/ml, P = 0.03; both were analyzed by one-way ANOVA) (Fig. 5). Additionally, the plasma level of IL-13 showed a marked increment in PD compared to the control group (12.49 ± 6.55 pg/ml vs. 2.93 ± 0.52 pg/ml, P < 0.01; by one-way ANOVA, Additional file 3). These results suggest that the altered microbiota seen in patients with PD may be associated with systemic inflammatory responses that contribute to PD development (Additional file 3).