Gut microbiome is associated with multiple sclerosis activity in children

Abstract

Objective: To identify features of the gut microbiome associated with multiple sclerosis activity over time.

Methods: We used 16S ribosomal RNA sequencing from stool of 55 recently diagnosed pediatric-onset multiple sclerosis patients. Microbiome features included the abundance of individual microbes and networks identified from weighted genetic correlation network analyses. Prentice-Williams-Peterson Cox proportional hazards models estimated the associations between features and three disease activity outcomes: clinical relapses and both new/enlarging T2 lesions and new gadolinium-enhancing lesions on brain MRI. Analyses were adjusted for age, sex, and disease-modifying therapies.

Results: Participants were followed, on average, 2.1 years. Five microbes were nominally associated with all three disease activity outcomes after multiple testing correction. These included butyrate producers Odoribacter (relapse hazard ratio = 0.46, 95% confidence interval: 0.24, 0.88) and Butyricicoccus (relapse hazard ratio = 0.49, 95% confidence interval: 0.28, 0.88). Two networks of co-occurring gut microbes were significantly associated with a higher hazard of both MRI outcomes (gadolinium-enhancing lesion hazard ratios (95% confidence intervals) for Modules 32 and 33 were 1.29 (1.08, 1.54) and 1.42 (1.18, 1.71), respectively; T2 lesion hazard ratios (95% confidence intervals) for Modules 32 and 33 were 1.34 (1.15, 1.56) and 1.41 (1.21, 1.64), respectively). Metagenomic predictions of these networks demonstrated enrichment for amino acid biosynthesis pathways.

Interpretation: Both individual and networks of gut microbes were associated with longitudinal multiple sclerosis activity. Known functions and metagenomic predictions of these microbes suggest the important role of butyrate and amino acid biosynthesis pathways. This provides strong support for future development of personalized microbiome interventions to modify multiple sclerosis disease activity.

Figure 1

Example of survival analyses for relapse and MRI outcomes. (A) For relapse analyses, time to each relapse (an “event” in panel A) started on the day stool was collected (d₀) and ended on the day of each respective relapse or the last study visit where relapse status was known (d_x). A 30‐day period was subtracted from the at‐risk period following a relapse because, by definition, a new relapse must be at least 30 days after the previous. (B) For MRI analyses, an “event” was defined as a brain MRI that indicated a new or enlarging lesion compared to the prior MRI. Specifically, we used two MRI outcomes considered separately: new gadolinium‐enhancing lesion and new or enlarging T2 lesion. Because the timing of MRI varies in clinical practice and the specific time of lesion activity is unknown, midpoint survival analyses were used. For the first MRI event after stool was collected, time at risk started on the date stool was collected and ended on the midpoint between the first MRI event and the prior MRI where an event did not occur (or the MRI that preceded baseline). For subsequent MRI events, time at risk started on the date of the previous MRI with an event and ended on the midpoint between the respective MRI event and the prior MRI where an event did or did not occur. Individuals were censored at the date of their last MRI (with or without an MRI event) before the study end.

Figure 2

Microbial alpha diversity was not associated with clinical relapses or MRI outcomes in pediatric‐onset multiple sclerosis. (A) The Chao1 microbial richness estimator was not associated with relapse (HR = 1.00; 95% CI: 0.99, 1.00; p = 0.56), new gadolinium‐enhancing lesion on MRI (HR = 0.99; 95% CI: 0.99, 1.00; p = 0.16), or new or enlarging T2 hyperintense lesion on MRI (HR = 1.00; 95% CI: 0.99, 1.01; p = 0.84). (B) The Faith’s phylogenetic diversity microbial richness estimator was not associated with relapse (HR = 0.98; 95% CI: 0.94, 1.02; p = 0.29), new gadolinium‐enhancing lesion on MRI (HR = 0.96; 95% CI: 0.92, 1.01; p = 0.15), or new or enlarging T2 hyperintense lesion on MRI (HR = 0.99; 95% CI: 0.94, 1.04; p = 0.77). (C) Microbial evenness (Pielou estimator) was not associated with relapse (HR = 1.35; 95% CI: 0.34, 5.32; p = 0.67), new gadolinium‐enhancing lesions on MRI (HR = 0.64; 95% CI: 0.13, 3.08; p = 0.58), or new or enlarging T2 hyperintense lesions on MRI (HR = 0.95; 95% CI: 0.19, 4.84; p = 0.95). Beta coefficients and related HRs and 95% CIs for evenness were scaled to represent a 0.1‐unit change in evenness. Regression models adjusted for sex, age, and disease‐modifying therapy use.

Figure 3

Variance in fecal microbiota composition was not associated with pediatric‐onset multiple sclerosis clinical relapse and MRI outcomes. Having, on average, more than 0.5 relapses per year was not associated with beta diversity using (A) weighted UniFrac (PERMANOVA R²  = 0.01, p = 0.78) or (B) unweighted UniFrac distance matrices (PERMANOVA R²  = 0.02, p = 0.38). Having any new gadolinium‐enhancing lesions over the study period was not associated with beta diversity using (C) weighted UniFrac (PERMANOVA R²  = 0.01, p = 0.78) or (D) unweighted UniFrac distance matrices (PERMANOVA R²  = 0.02, p = 0.42). Having any new or enlarging T2 hyperintense lesions over the study period was not associated with beta diversity using (E) weighted UniFrac (PERMANOVA R²  = 0.02, p = 0.43) or (F) unweighted UniFrac distance matrices (PERMANOVA R²  = 0.02, p = 0.63). PERMANOVA models adjusted for sex, age, and disease‐modifying therapy use. The first two principal coordinates from principal coordinate analysis were plotted.

Figure 4

No species of gut microbes were significantly (FDR q < 0.05) associated with pediatric‐onset multiple sclerosis activity outcomes. The long‐dashed line indicated an FDR q cutoff of 0.05 and the small dotted line indicated a less conservative threshold FDR q = 0.2. Each point was an ASV. The genus and species (or lowest known taxonomy) of ASVs associated with a respective outcome with FDR q < 0.2 were labeled. Regression models were adjusted for sex, age, and disease‐modifying therapy use. (A) Adjusted log‐hazard ratios for relapse; (B) adjusted log‐hazard ratios for new gadolinium‐enhancing lesions on MRI; and (C) adjusted log‐hazard ratios for new or enlarging T2 hyperintense lesions on MRI.

Figure 5

Five gut microbes were associated with all three pediatric‐onset multiple sclerosis activity outcomes. Each row was an ASV that was associated with either relapse, new gadolinium‐enhancing lesions, or new or enlarging T2 hyperintense lesions at p < 0.05. Hazard ratios, adjusted for sex, age, and disease‐modifying therapy use, were shown for each significant (p < 0.05) ASV‐outcome association. Gray indicated an ASV–outcome association was not significant. “NA” indicated an association was not estimated because the ASV was not in at least 20% of the respective sample. Rows were labeled with ASV ID and the lowest known taxonomic classification. Rows were arranged by taxonomic order.

Figure 6

Five networks of gut microbes were significantly associated with MRI‐related multiple sclerosis activity. The long‐dashed line indicated an FDR q cutoff of 0.05 and the small dotted line indicated an FDR q cutoff of 0.2. Significant modules were labeled with their respective module name. Regression coefficients were scaled to a 0.1‐unit increase in module eigengenes because the standard 1‐unit increase would represent nearly the entire range of eigengene values. Regression models were adjusted for sex, age, and disease‐modifying therapy use. (A) Adjusted log‐hazard ratios for relapse; (B) adjusted log‐hazard ratios for new gadolinium‐enhancing lesions on MRI; and (C) adjusted log‐hazard ratios for new or enlarging T2 hyperintense lesions on MRI.