Meta-analysis of the Parkinson's disease gut microbiome suggests alterations linked to intestinal inflammation

Abstract

The gut microbiota is emerging as an important modulator of neurodegenerative diseases, and accumulating evidence has linked gut microbes to Parkinson's disease (PD) symptomatology and pathophysiology. PD is often preceded by gastrointestinal symptoms and alterations of the enteric nervous system accompany the disease. Several studies have analyzed the gut microbiome in PD, but a consensus on the features of the PD-specific microbiota is missing. Here, we conduct a meta-analysis re-analyzing the ten currently available 16S microbiome datasets to investigate whether common alterations in the gut microbiota of PD patients exist across cohorts. We found significant alterations in the PD-associated microbiome, which are robust to study-specific technical heterogeneities, although differences in microbiome structure between PD and controls are small. Enrichment of the genera Lactobacillus, Akkermansia, and Bifidobacterium and depletion of bacteria belonging to the Lachnospiraceae family and the Faecalibacterium genus, both important short-chain fatty acids producers, emerged as the most consistent PD gut microbiome alterations. This dysbiosis might result in a pro-inflammatory status which could be linked to the recurrent gastrointestinal symptoms affecting PD patients.

Fig. 1. Sample distribution across studies and bioinformatic workflow adopted in our study.

a The number of control and PD samples refers to the data that could be recovered from the Sequence Read Archive (SRA) or the European Nucleotide Archive (ENA) and passed the quality filtering we applied. b Overview of the bioinformatic workflow adopted in our study. TSS total sum scaling, VST variance stabilizing transformation, CLR centered log ratios, WMW Wilcoxon–Mann–Whitney test, ANCOM analysis of the composition of microbiomes, lm linear models, genodds Agresti’s generalized odd ratios, meta random-effect meta-analysis.

Fig. 2. Alpha-diversity indices are significantly different between PD patients and controls.

Indices were calculated at the species level for each dataset. Results were then combined using a random-effect meta-analysis approach. The log-generalized Odds Ratios indicate the degree of variation of each index between controls and PD. The richness of the samples was estimated using the observed number of species and the indices Chao1, ACE, and Fisher’s alpha. To estimate evenness, which indicates how different the species abundances in a community are from each other, we used the Bulla and Simpson indices. Finally, we estimated dominance, which describes how much one or few species dominate the community, and rarity, which assesses the number of species with low abundance in the samples. The data suggest that the gut microbiota of PD patients is more diverse (higher richness) than controls and this is likely a consequence of an increase in rare taxa (rarity).

Fig. 3. The gut microbiome structure differs significantly between PD patients and controls.

Data were normalized using three independent approaches (VST variance stabilizing transformation, TSS total sum scaling, CLR centered log-ratio) and beta-diversity was estimated using three indices (Bray–Curtis, BC; Jensen–Shannon divergence, JSD; Euclidean). The effect of the disease status on the clustering of the data was assessed using a permutational analysis of variance (PERMANOVA). In the majority of the studies and approaches considered, and across all taxonomic ranks (a, b, c), the gut microbiome of PD patients resulted significantly different from the one of controls. The disease status explains only a small fraction of the data variability (<13% R²), indicating that other environmental factors might have a stronger role in shaping the bacterial communities. The dataset obtained by pooling all ten studies is referred to as “Combined” in the figure.

Fig. 4. Most important species driving the divergence of the gut microbiome between PD patients and controls.

Distance-based redundancy analysis (dbRDA) was performed on Jensen–Shannon divergence (JSD) calculated on data normalized through total sum scaling (TSS). dbRDA was conditioned (blocked) by study and constrained by disease status. Data refer to species abundances. The limited proportion of data variability explained by the axis constrained for disease status (CAP1) indicates that environmental factors have a major influence in shaping the bacterial communities. However, the influence of the disease status on the community structure is statistically significant (ANOVA-like permutation test). Only taxa showing a significant association with the clustering of the samples and the strongest abundance variation between conditions are reported.

Fig. 5. Genera showing a significant difference in abundance between PD patients and controls.

The relative abundances of the genera retrieved from the rarefied pooled data are reported in panel a. Effect sizes were estimated via the mean difference in CLR (panel b) using a random-effect meta-analysis approach (Pooled results approach). This was calculated for all taxa resulting differentially abundant in the Pooled results or Pooled data approaches. The color of the dots indicates which of the two above approaches detected the taxa differentially abundant. Taxa more abundant in controls have an effect size shifted to the left, whereas taxa more abundant in PD have an effect size shifted to the right. Panel c shows the number of times each genus was detected differentially abundant between PD patients and control samples across studies (diamonds) and approaches (bars). We used ten studies and three approaches, hence the maximum number of times a taxon can be detected differentially abundant is 30.

Fig. 6. Metabolic pathways showing a significant difference in abundance between PD patients and controls.

Only selected relevant pathways are shown (a full overview is reported in Supplementary Fig. 14). The relative abundances of the pathways retrieved from the rarefied pooled data are reported in panel a. Effect sizes were estimated via the mean difference in CLR (panel b) using a random-effect meta-analysis (Pooled Results approach). This was calculated for all pathways resulting differentially abundant in the Pooled results or Pooled data approaches. The color of the dots indicates which of the two above approaches detected the pathway differentially abundant. Pathways more abundant in controls have an effect size shifted to the left, whereas pathways more abundant in PD have an effect size shifted to the right. Panel c shows the number of times each pathway was detected differentially abundant between PD patients and controls across studies (diamonds) and approaches (bars). We used ten studies and three approaches, hence the maximum number of times a pathway can be detected differentially abundant is 30.