Meta-analysis identifies common gut microbiota associated with multiple sclerosis

Abstract

Background: Previous studies have identified a diverse group of microbial taxa that differ between patients with multiple sclerosis (MS) and the healthy population. However, interpreting findings on MS-associated microbiota is challenging, as there is no true consensus. It is unclear whether there is gut microbiota commonly altered in MS across studies.

Methods: To answer this, we performed a meta-analysis based on the 16S rRNA gene sequencing data from seven geographically and technically diverse studies comprising a total of 524 adult subjects (257 MS and 267 healthy controls). Analysis was conducted for each individual study after reprocessing the data and also by combining all data together. The blocked Wilcoxon rank-sum test and linear mixed-effects regression were used to identify differences in microbial composition and diversity between MS and healthy controls. Network analysis was conducted to identify bacterial correlations. A leave-one-out sensitivity analysis was performed to ensure the robustness of the findings.

Results: The microbiome community structure was significantly different between studies. Re-analysis of data from individual studies revealed a lower relative abundance of Prevotella in MS across studies, compared to controls. Meta-analysis found that although alpha and beta diversity did not differ between MS and controls, a higher abundance of Actinomyces and a lower abundance of Faecalibacterium were reproducibly associated with MS. Additionally, network analysis revealed that the recognized negative Bacteroides-Prevotella correlation in controls was disrupted in patients with MS.

Conclusions: Our meta-analysis identified common gut microbiota associated with MS across geographically and technically diverse studies.

Keywords: Faecalibacterium; Prevotella; Meta-analysis; Microbiota; Multiple sclerosis.

Fig. 1

Common microbiota revealed by individual study analysis. a Summary of microbiota analysis at the microbial community level for each study. Significant differences are demonstrated by different colors. The sizes of the circles represent scale of -log10 p-value. b Genera that are significantly different in MS vs controls by Wilcoxon rank-sum test or DESeq2 test. Significant genera identified in at least two studies are shown in descending order from top to bottom by its shared frequency. For each individual study, the sizes of the circles represent mean relative abundance of the genera, and the colors of the circles represent statistically higher (blue) or lower (orange) of the relative abundance of the genera in MS, compared to controls. Taxonomy at the class level for each genus is indicated on the class column. Different shapes in the class column indicate the abundance of the genera are consistently (triangles) and inconsistently (circles) higher or lower in MS in at least two studies, compared to controls

Fig. 2

Microbiota variation across all studies. a Abundance and prevalence of 652 genera identified in 7 studies. Left Y-axis shows number of studies that detect a given genus (frequency); right Y-axis represents the relative abundance of each genus in a study. Several relatively high abundant genera are labeled in the plot. b Principal component analysis of samples from all seven studies based on Bray-Curtis distance; different studies are color-coded and group (MS vs controls) is indicated by different shapes. Boxplots at the bottom and at the right show PC1 and PC2 loadings for different studies or group (MS vs controls). Studies are significantly different in both PC1 and PC2 (p < 0.001). Group (MS and controls) is significantly different in PC2 after controlling for study effect (p = 0.0002). c Percent of variance that is significantly contributed by each factor

Fig. 3

Common microbiota revealed by meta-analysis. a Alpha and beta diversity in MS and controls. Diversity is not statistically different between MS and controls (p = 0.72 for Bray-Curtis dissimilarity; p = 0.57 for richness; p = 0.63 for Shannon diversity). b Significantly different genera between MS and controls identified by meta-analysis. Fifteen genera are significantly different between MS and controls after controlling for the “study” factor. Mean and standard error for each genus are illustrated. Class level taxonomy for each genus is indicated next to the genus. c Results of leave-one-out sensitivity analysis

Fig. 4

Disruption of Bacteroides-Prevotella correlative network in MS. The network is constructed using correlations with p-value less than 0.05 and correlation coefficient larger than 0.2. Red lines represent positive correlations and blue lines represent negative correlations; The width of line varies by absolute value of correlation coefficient, and the nodes size represents relative abundance of genera. a, b Correlations shared between MS and controls. Nine positive correlations are shared in MS (a) and controls (b). c Correlations that are unique in controls. d Correlations that are unique in MS