Multiple sclerosis (MS) has a complex genetic, immune and metabolic pathophysiology. Recent studies implicated the gut microbiome in MS pathogenesis. However, interactions between the microbiome and host immune system, metabolism and diet have not been studied over time in this disorder.
We performed a six-month longitudinal multi-omics study of 49 participants (24 untreated relapse remitting MS patients and 25 age, sex, race matched healthy control individuals. Gut microbiome composition and function were characterized using 16S and metagenomic shotgun sequencing. Flow cytometry was used to characterize blood immune cell populations and cytokine profiles. Circulating metabolites were profiled by untargeted UPLC-MS. A four-day food diary was recorded to capture the habitual dietary pattern of study participants.
Together with changes in blood immune cells, metagenomic analysis identified a number of gut microbiota decreased in MS patients compared to healthy controls, and microbiota positively or negatively correlated with degree of disability in MS patients. MS patients demonstrated perturbations of their blood metabolome, such as linoleate metabolic pathway, fatty acid biosynthesis, chalcone, dihydrochalcone, 4-nitrocatechol and methionine. Global correlations between multi-omics demonstrated a disrupted immune-microbiome relationship and a positive blood metabolome-microbiome correlation in MS. Specific feature association analysis identified a potential correlation network linking meat servings with decreased gut microbe B. thetaiotaomicron, increased Th17 cell and greater abundance of meat-associated blood metabolites. The microbiome and metabolome profiles remained stable over six months in MS and control individuals.
Our study identified multi-system alterations in gut microbiota, immune and blood metabolome of MS patients at global and individual feature level. Multi-OMICS data integration deciphered a potential important biological network that links meat intakes with increased meat-associated blood metabolite, decreased polysaccharides digesting bacteria, and increased circulating proinflammatory marker.
This work was supported by the Washington University in St. Louis Institute of Clinical and Translational Sciences , funded, in part, by Grant Number # UL1 TR000448 from the National Institutes of Health, National Center for Advancing Translational Sciences, Clinical and Translational Sciences Award (Zhou Y, Piccio, L, Lovett-Racke A and Tarr PI); R01 NS10263304 (Zhou Y, Piccio L); the Leon and Harriet Felman Fund for Human MS Research (Piccio L and Cross AH). Cantoni C. was supported by the National MS Society Career Transition Fellowship ( TA-180531003 ) and by donations from Whitelaw Terry, Jr. / Valerie Terry Fund. Ghezzi L. was supported by the Italian Multiple Sclerosis Society research fellowship ( FISM 2018/B/1 ) and the National Multiple Sclerosis Society Post-Doctoral Fellowship ( FG-190734474 ). Anne Cross was supported by The Manny & Rosalyn Rosenthal-Dr. John L. Trotter MS Center Chair in Neuroimmunology of the Barnes-Jewish Hospital Foundation. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
The study was approved by the Human Research Protection Office at Washington University in St. Louis School of Medicine (WUSM) (approval number: 201,502,105). All patients gave informed consent to participation.
(2) no DMT or steroid treatments in the past 3 months; (3) ages 18 to 50 years; and (4) not in clinical relapse at study enrolment. Exclusion criteria were: (1) coexistence of other chronic inflammatory (e.g., asthma, chronic hepatitis, inflammatory bowel disease, celiac disease, etc.) and autoimmune (e.g., rheumatoid arthritis, SLE, type I diabetes, etc.), or metabolic (e.g., type II diabetes, familial hypercholesterolemia, etc.) diseases; (2) antibiotics or steroid therapy in the past 3 months; (3) history of immunosuppressive or chemotherapeutic treatment; (4) history of chronic infectious disease (e.g., TBC, HIV, HBV, HCV, etc.); (5) neoplastic disease not in complete remission, and (6) pregnancy. Age, sex, BMI and ethnicity-matched healthy controls were enrolled using the same exclusion criteria. Table 1 details the demographic and clinical characteristics for all participants at enrolment.
Table 1Demographics and clinical characteristics of MS patients and healthy controls.
Data are provided as n (%), mean (SD) or range; # including cigars, pipes or electronic cigarettes. BMI: body mass index; EDSS: expanded disability status scale; OCB: oligoclonal bands. DMTs: disease modifying therapies; SD=standard deviation.
Differences between groups were compared using Pearson chi-square test for categorical data and the t-test or non-parametric Mann-Whitney test as appropriate for continuous data.
MS and healthy control participants were studied again six months after baseline. No relapse was reported in any of the MS patients during the study. Although DMTs commencement was strongly recommended to the 24 MS patients by their clinicians, only 8 started DMT treatment during the six-month study period. The DMTs started were glatiramer acetate and fingolimod (n=1 each), interferon-β1a (n=3), and dimethyl fumarate (n=3).
Stool and blood of all participants were collected at the time of enrolment and six months later. Collection of stool and blood samples were done concomitantly (less than 24 h difference). Stools were self-collected and placed on frozen gel packs and shipped overnight to the research laboratory. Upon receipt, stools were immediately stored at -80 °C until further processing. Stools from baseline and six months were processed at the same time for DNA extraction and microbiome sequencing to minimize batch effects among the specimens.
Blood was collected in heparinized tubes, insulated, and shipped at room temperature overnight to Ohio State University for immunophenotyping. Peripheral blood mononuclear cells (PBMCs) were isolated immediately on arrival and analyzed by flow cytometry. Serum samples were stored frozen at -80 °C and sent to University of Massachusetts for metabolomics.
Stool DNA extraction and microbiome sequencing
16S rRNA gene sequencing permits deep microbiota profiling, especially of low abundance taxa. Metagenomic whole genome shotgun sequencing (mWGS) classifies to species levels but may not enumerate low abundance bacteria. We applied these complementary platforms to sequencing the gut microbiome.
Briefly, stool DNAs were extracted using the MOBIO PowerSoil DNA Extraction kit. For 16S rRNA gene sequencing, hyper-variable regions V1–V3 of 16S gene were amplified using primers 27F and 534R (27F: 5′-AGAGTTTGATCCTGGCTCAG-3′ and 534R: 5′- ATTACCGCGGCTGCTGG-3′). 16S libraries were prepared and sequenced on the Illumina MiSeq sequencing platform using a V3 2 × 300 bp paired end sequencing protocol with a target read depth of 10,000 reads/sample. Illumina’s software handles the initial processing of all the raw sequencing data. One mismatch in primer and zero mismatch in barcodes were applied to sample deconvolution. Reads were further processed by removing sequences with low quality (average qual ftp://ftp.ncbi.nlm.nih.gov/pub/agarwala/bmtagger); (b) removing duplicated reads using GATK-Picard 4·1·0 (MarkDuplicates); (c) trimming low-quality bases and low-complexity screening using PRINSEQ (v0·20·4). We used MetaPhlAn2 to classify species and HUMAnN2 for microbiome functional potential inference (gene ortholog or KEGG ortholog and metabolic pathways),
and Lefse (v1·0) to determine the statistical difference of KOs or metabolic potentials between MS patients and controls.
We performed sequencing and processed data at The Jackson Laboratory for Genomic Medicine.
Blood sample preparation and processing for non-targeted metabolome
Short-chain fatty acids determination by GC chromatography
SCFAs in stool samples were extracted by rigorous vortex with 0·05% phosphoric acid at the ratio of 1:15 (w/v), followed by centrifugation. Supernatants were mixed with ethyl acetate at the ratio of 1:1 (v/v), the SCFAs were transferred to ethyl acetate, and the supernatants were obtained after centrifugation. SCFAs were measured by gas chromatography with a flame ionization detector (Schimadzu GC-QP2010 SE, Tokyo, Japan). Acetate, propionate, butyrate, isobutyrate, valerate, isovalerate were used as standards for the identification, and 2-ethylbutyrate was used as the internal standard.
Blood sample preparation for immune profiling
Briefly, 1 × 106 cells were washed and resuspended in cold PBS/1% BSA, incubated with FcR Blocking Reagent (Miltenyi) for 10 min, and incubated with antibody cocktails for 30 min. Samples were washed and fixed with either PFA (0·5%), Cytofix/Cytoperm Solution Kit (BD Biosciences), or Foxp3 Transcription Factor Staining Set (eBioscience). Intracellular molecules were stained in their appropriate permeabilization wash buffers for 30 min (cytokines) or 45 min (Foxp3). After a final wash with PBS/1% BSA, we acquired and analyzed cytometric data using BD FACSCanto II, FACSDiva, and FlowJo Software v.10·7 (Becton, Dickinson and Company, 2019). Table S2 and Figure S5 describe the gating strategy for each panel, and the percentage of each cell subset population within the parent gate.
Food diary and conversion to food serving
The participants were asked to continue their usual eating habits before the initial stool collection and for the next six months.
This system assigns foods in the Nutrition Data System for Research (NDSR) database (2017 version) to food groups that fit within nine major categories. Eighteen food groups including average daily servings of fruit, vegetable, whole grain, refined grain, meat, poultry, fish and shellfish, cold cuts and sausage, eggs, nuts and seeds, butter and animal fats, plain and flavored cow’s milk, dairy cheese, yogurt live active cultures, vegetable oils, salad dressings, beer and liquor and wine, sugar sweetened soft drinks (soda, punch, tea), and their daily serving sizes were used for our study. Serving sizes are assigned to each NDSR food based on the 2000 Dietary Guidelines for Americans recommendations when available.
For foods not included in recommendations (e.g., cookies, fruit drinks), United States Food and Drug Administration (FDA) serving sizes are used.
Statistical analysis of the microbiome data
All P values were two sided. Adjusted P values with a False Discovery Rate (FDR) of
we further inspected the results by plotting raw and relative abundance data. We removed results that are low in relative abundance (
We tested age, BMI, smoking, family history of autoimmune diseases, probiotics intake and diagnosis (MS vs controls) individually in the PERMANOVA model. Note the total variance explained by each variable was calculated independent of other variables and should therefore be considered the total variance explainable by that variable.
Statistical analysis of immune profile
We present immune cell populations from the flow cytometry as proportions and scaled for PCA analysis to estimate the overall similarity of the immune profile among all participants. Wilcoxon Rank Sum Test was used to identify significantly different immune cell populations between MS and controls. P values were further adjusted by FDR approach. All p-values were two-tailed.
Statistical analysis of blood metabolome
P values were further adjusted by FDR approach. All p-values were two-tailed.
Mantel correlation and multi-OMIC feature-feature correlation
We quantified covariation between multi-omics using Mantel tests (Pearson correlation between distances of two matrices). A pair-wised inter-participant variation/distance matrix was first computed for each OMIC dataset, with Bray-Curtis dissimilarity for the microbiome data and Euclidean distance for the immune profile, blood metabolome and food intakes. Inter-participant dissimilarity matrices were then compared using the mantel function in the vegan package. Mantel correlation analysis was also conducted in similar manner to quantify longitudinal covariation for two given omics data. The significance of the statistic is produced by permuting rows and columns of the first dissimilarity matrix for 1000 times.
We performed feature-feature correlations within and between omics datasets using Pearson correlation with cor.test function in the stats package in R. Because of potential for different interactions in MS patients and controls, all correlations were performed separately for the two groups, accounting for BMI and age. P values were corrected based on FDR approach. FDR <0·2 was considered significant for correlations among features from the microbiome, immune profile and food intakes and FDR <0·05 was considered significant for correlations with features from the blood metabolome. To further remove the false positive or negative correlations that were driven by single values, we generated x-y scatter plots all statistically significant correlations, and confirmed them by manual inspection. The resulting correlations with absolute correlation coefficient >0·7 were considered strong correlations and the network of correlation was illustrated using Cytoscape. A hub in the correlation network was defined as nodes with at least 20 connections. All correlation results including before and after FDR corrections and after manual inspections are summarized in Table S4.
MS classification using machine learning models
The models were trained by each individual omics to determine the importance of a given omics data in classification performance (Figure 3a, b, c), or by the combination of all the omics to determine whether it achieves a better classification performance (Figure 3d). For the microbiome data, OTUs with 0 abundances were firstly replaced by a small value 1e-5. Then the centred log-ratio transformation was applied,
so that the transformed data obey the Euclidean geometry. The number of raw features could greatly exceed the sample size and many of them are either redundant or irrelevant in distinguishing MS patients and controls. Hence, we reduced the size of the feature set before fitting any machine learning model, by applying a statistical marginal screening procedure through multiple hypothesis testing with false discovery rate control.
Such a hybrid “marginal screening + machine learning” approach facilitated the model training and consistently improved the performance of the classifiers in our study.
For RF, we set the maximum number of features allowed to try in an individual tree as the square root of the number of features, and the number of trees as 30,000, a sufficiently large number. The feature importance in the RF is measured by the Mean Decrease Gini. After building a classifier using the training samples, we used the testing samples to compute its pairs of out-of-sample true positive rate (TPR) and false positive rate (FPR), based on which we constructed the sample receiver operating characteristic (ROC) curve and calculated the corresponding area under the ROC curve (AUC) value. By aggregating these results from 200 random splits, we draw the average ROC curve and computed the average AUC and its 90% confidence interval for each classifier. Missing data were imputed based on group mean.
Justification for statistical approaches used in single and multi-OMICS analysis
All the analysis were conducted using a complete case analysis except for the multi-omics predictive modelling analysis, in which multi-OMICS data were combined to evaluate classification accuracy for MS and controls. There was no missing in microbiome and metabolome data. In the immune data, 36.0% of the subjects have missing values. This is largely attributed by one immune feature that has very low level of immune cells. In the metadata, 14.0% of subjects have missing values. In the nutrition data, 27 % of subjects have missing values. In the multi-omics predictive modeling analysis, as much as 82.7% of subjects have at least one missing value. Therefore, due to the already limited sample size, it is not practical to run a complete case study. On the other hand, the overall missing rate (total missing values among all value) of the multi-omics data is very low: it is 4.6%. We thus chose the simple imputation method of using group means (by MS/Control).
For our immune data analysis, some immune cell populations are not normally distributed, thus we chose a robust non-parametric Wilcoxon Rank Sum test for two group comparisons. For metabolome data, because of large variation of the data, log transformation is conventionally encouraged and performed, followed by Welch’s t-test. Mantel correlations between two matrices were based on continuous data derived from Euclidean distance or Bray-Curtis dissimilarity and showed good linear relationships as demonstrated by scatter plots. Generally, if the scatter plot shows linear trend, instead of other curvature trend, ex quadratic trend, the data satisfies the linearity assumption. Due to the limited sample size, our general strategy is to use parametric methods to gain more power when their underlying assumptions are not violated, otherwise non-parametric tests are adopted to ensure robustness.
Role of funding source
The funders had no role in the conceptualization, study design, data collection, analysis, interpretation of data, in writing the paper, or in the decision to submit the paper for publication.
Baseline characteristics of the study population
Overall gut microbiota profile and factors underlying microbial variation in MS patients and controls
Specific gut microbiota associated with MS
mWGS data identified six species that are significantly lower in abundance in MS patients than in controls, three of which have known immunomodulatory properties (Bifidobacterium longum, Clostridium leptum, Faecalibacterium prausnitzii) (FDR
Bacteroides thetaiotaomicron and two unclassified Parabacteroides and Escherichia species were also significantly different between two groups (Figure 1e). Thus, lower abundance of Faecalibacterium species was consistently detected by both sequencing technologies in MS patients compared to healthy controls. The average relative abundance of B. fragilis, which is protective in the EAE model,
was 0·34% at baseline, with no statistically significant differences between MS and controls by univariate analysis (Figure S2). In summary, decreased relative abundances of bacteria with immunomodulatory properties seem to characterize gut microbiome changes in MS patients vs. controls.
Sutterella, a Gram-negative genus from Proteobacteria, has been associated with various diseases including autism and inflammatory bowel disease (IBD).
Eubacterium siraeum (r=-0·47, FDR=0·049) was negatively correlated with EDSS. These findings suggest that specific gut microbes may be associated with the degree of disability in MS patients.
Next, we inferred the metabolic potentials of the gut microbiome for all study participants using mWGS data by HUMAnN2 and LEfSe. Sixty-six metabolic pathways and 276 gene Ortholog or KEGG ortholog (KOs) significantly differed between MS patients and controls before adjusting for multiple comparisons (Table S1). Interestingly, most differentiating pathways (64/66=96·9%), which included glycolysis, glutamate degradation, fermentation pathways or phospholipid biosynthesis and KOs (244/276=88·4%), were under-represented in MS patients compared to controls. However, after adjusting for multiple comparisons, no KO or pathway differed significantly between the two groups (all FDR > 0·3).
Additionally, we measured concentrations of the short chain fatty acids (SCFAs) including acetic acid and butyric acid in stool by gas chromatography/mass spectrometry (GC-MS) and found a trend of lower level of SCFAs in stools of MS patients than in those of controls (Figure S4), but the trend did not attain statistical significance (P>0·05, Wilcoxon Rank Sum Test) and the inter-individual variations of SCFAs in both groups were quite high.
Loss of the microbiome-immune homeostatic interaction and establishment of an immune-metabolome association in MS
dihydrochalcone, 3-mercaptolactate guanine 1-naphthaldehyde, riboflavin, 4-hydroxy lauric acid and chalcone were significantly enriched in the MS patients. Blood bile acid metabolism has been reported to be decreased in MS.
Our study did not find statistically significant differences in primary and secondary bile acid metabolism between MS and controls, though we did identify a trend towards decreased glycocholate, taurodeoxycholate, glycochenodeoxycholate, and taurohyocholate, and increased deoxycholic acid and other bile acid metabolites in MS. Pathway analysis showed that pathways involved in linoleate metabolism, fatty acid metabolism were altered in MS patients (Figure 2d).
Overall dietary patterns did not differ significantly by PERMANOVA between MS patients and controls. The MS group had a higher median meat intake compared to controls before (P=0·006, Wilcoxon Rank Sum Test) (Figure S6), but not after FDR adjustment (FDR=0·10).
Together, our data infer distinct, diverse, and cross-system interrelationships of key pathways in MS patients and controls, providing a compendium of potential targets for future studies of pathogenic mechanisms underlying MS.
A potential pathway linking meat serving, gut microbiome, Th17 cells and blood metabolites
B. thetaiotaomicron was strongly negatively correlated with proportions of circulating Th17 cells (r=-0·40, p=0·01), while Th17 cells were positively correlated with meat servings (r=0·50, p=0·003). As diet, the gut microbiome and immune response all potentially affect blood metabolites, we next correlate blood metabolites with B. thetaiotaomicron, Th17 cells and meat servings and found five blood metabolites significantly correlated with all three measurements (Figure 4b, Figure S8). Interestingly, metabolite mz34, annotated as SAM, was positively correlated to meat servings (r=0·42, p=0·01), Th17 cell proportion (r=0·75, p=1·7e-07), and negatively correlated with relative abundance of B. thetaiotaomicron (r=-0·40, p=0·006) (Figure 4a). Taken together, our multi-omics analysis suggests a correlation network involving dietary meat serving, gut microbiome, Th17 cells and blood metabolites. However, our analyses do not indicate the directionality of regulation between each of the aforementioned correlative pair. These results highlight a discovery process driven by omics analysis and provide an interesting hypothesis that now warrants further validation.
Host-microbiome multi-omics capacity to classify MS patients and controls
Longitudinal changes of the gut microbiome and host peripheral immune and metabolome profiles in MS patients and controls
indicating temporal stability. Compared to between-participant variation, within-participant variations of the microbiome and metabolome were significantly lower for all MS patients and controls (Figs. 6a, c), and the within-participant variations of the immune profile were also significantly lower than between-participant variations in controls (Figure 6b). This suggests a relatively stable overall microbiome and metabolome for MS patients and controls as well as the immune profile in controls during the study period. In contrast, between- and within-participant variations of the immune profile in MS patients who received treatment with DMTs were not statistically different (P=0·28, Wilcoxon-rank test, Figure 6b). To determine if DMTs altered the immune profile and subsequently affect the within-participant dissimilarity, we performed Wilcoxon Rank Sum test for MS patients before and after these interventions. We found that proportions of memory Th17 cells and GM-CSF+ memory T Cells were significantly reduced at six months compared to baseline in MS patients who initiated DMTs (FDR=0·05, Figure S11).
We further performed a Mantel correlation to test if between-participant similarities were maintained over the study interval based on distance measures of gut microbiome, blood immune cell and metabolome profiles. We found significant correlations of the gut microbiome (r=0·69, P=0·001 for MS; r=0·4, P=0·005 for controls) and blood metabolome (r=0·39, p=0·001 for MS; r=0·27, p=0·005 for controls) between baseline and six-month samples. In contrast, the blood immune profile showed no correlation between baseline and six-month for MS patients (r=0·05, P=0·32), while controls demonstrated significant correlation (r=0·33, p=0·01). These findings suggest that between-participant relationships were maintained over the course of the study in gut microbiome and metabolome profiles for both MS and controls, and in the immune profiles for controls, but not for MS participants.
We did not identify specific gut microbiome, metabolome or food servings that significantly changed between the beginning of the study and six-month follow-up in the MS patients. We also did not identify microbiota that differed between treated and untreated MS patients at six months. Strikingly, 41·9% of the blood metabolites that were significantly different between MS patients and controls at baseline still maintained similar differences at six months follow-up. Therefore, the analysis of the data obtained at six months follow-up validated our baseline findings. Accordingly, machine learning models constructed using six-months follow-up data consistently showed the best classification accuracy in differentiating MS patients from controls based on metabolome features, compared to those based on immune and microbiome profiles. In summary, host peripheral immune profile, blood metabolites and the gut microbiome of MS participants remain relatively stable over six months in those patients who remained untreated, while initiation of treatment with DMTs affected specific immune cell populations.
We also extended the microbiome analyses by seeking factors that might drive microbiome variations between MS patients and controls, and validated the small impact from the disease itself. Interestingly, BMI contributed significantly to microbiome variation among all tested variables, and positively correlated with EDSS in MS patients. Hence, BMI should be considered in future inter-group microbiome comparisons in MS studies.
and is biologically plausible. Faecalibacteria, Lachnospiraceae, and Anaerostipes produce butyrate, which acts via G-protein coupled receptors activation and histone deacetylase inhibition to suppress CNS demyelination,
the main pathological feature in MS. Indeed, concentrations of fecal SCFAs (i.e., acetate, butyrate and propionate) were decreased in RRMS patients, compared to healthy controls.
Blood SCFAs were significantly decreased in long-term active progressive MS patients.
Propionic acid, but not butyrate and acetate, was significantly reduced in blood and stool in MS patients with all disease subtypes, particular after relapse. Supplementation of propionic acid promoted Treg cell function, and in long-term administration reduced relapse rate, disability and brain atrophy.
In our study, we found a trend toward decreased concentrations of butyrate in the stools of MS patients, consistent with decreased SCFA-producing bacteria in MS. Notably, SCFA levels can also be affected by diet, and higher meat servings in MS patients may also contribute to the observed reduction of SCFAs.
We found reduced abundance of a different Prevotella species, P. copri, in stools of MS patients. P. copri is a dominant Prevotella species in healthy American adults,
and is more prevalent in non-western populations.
Moreover, P. copri was highly correlated with IL10+ memory B cells in our control group, providing a potential novel microbiome-driven immune pathway to test in future.
However, it is possible that the same household controls may decrease the sensitivity to detect MS associated microbes, because individuals from the same household tend to share gut microbes, and a shared microbe may still influence MS development in genetically predisposed individuals. Changes in gut microbiota observed in MS participants in our study resembled previous studies conducted in different geographical locations, offering credence to our findings. MS-associated microbes are also over-represented in other autoimmune and metabolic diseases as well as cancer that are associated with inflammation.
This finding argues against a unique microbiome signature for MS patients but supports a common microbiome dysbiosis indicator for extra-intestinal pathophysiology associated with inflammation.
and our findings now link gut microbiome and systemic cellular immune profiles in healthy human adults. Our data recapitulate in a human cohort the inter-relatedness of the gut microbiome, diet, immune system and host metabolome, a relation that has been reported mostly in germ-free mice.
Specifically, we now demonstrate an association between the gut microbiome and peripheral immune phenotype in healthy participants, implying that healthy people with similar gut microbiome tend to have similar immune phenotype. In contrast, gut microbiome-immune haemostatic interactions were disrupted in the MS cohort we studied. Immune cell phenotypes of MS patients significantly differed from healthy controls in our dataset. However, we wish to note that the degree of dysbiosis of gut microbes in MS was modest. The discordance of the changes in immune phenotypes and microbiome may explain the lack of correlation between the microbiome and peripheral blood immune profiles in MS. It will be of great interest to elucidate the directionality and time-to-response of microbiome-immune regulation in MS. Future work might also be directed towards microbiota and immune response at the gut mucosa,
as the site of the systemic changes that ultimately affect the CNS.
and dysfunctional lipid metabolism.
The observed enrichment of circulating novel metabolites and multiple pathways in MS patients requires future validation to address their role in MS. We found that methionine was significantly enriched in MS patients, which is consistent with higher meat consumption in the patients we studied. Methionine drives T cell proliferation and differentiation.
Our data are also in accord with recent findings that methionine activates Th17 cells through epigenetic modification.
Methionine is an essential metabolite for methyl donor SAM synthesis, and SAM promotes Th17 cell activation through methylation.
Reduction of dietary methionine ameliorated EAE through reprogramming pathogenic Th17 cells. Our data, together with Roy et al., prompt the hypotheses that meat or methionine restriction might beneficially decrease the number of circulating inflammatory Th17 cells in MS patients. Future studies of metabolites in MS should consider specific dietary nutrients and gut microbiome derived metabolites, as these factors play large roles in ordaining human metabolism.
we identified their connections with a common gut commensal bacterium, B. thetaiotaomicron. This opens new research directions to elucidate regulatory pathways among diet, metabolites, microbiome, and immune response, and may identify therapeutic targets for MS.
However, differential microbiome responses to dietary intake in different individuals might explain the lack of strong correlation between diet and gut microbiome in human studies.
We also found that food composition did not correlate with circulating immune phenotypes. Nonetheless, lack of systemic associations does not exclude the possibility of specific feature-feature correlations, as we have identified associations between food compositions and the microbiome, as well as food compositions and immune cell populations. Feature-feature correlations were reported conservatively, and only significant associations after multiple comparison corrections were included in this final report, thereby strengthening confidence in the correlations.
Our longitudinal study design provided a unique opportunity to evaluate the stability of multi-omics data in MS patients over a period of six months, during which most DMTs take effect. Baseline and six-month follow-up measures for microbiome, metabolome components, immune-phenotypes and diet were similar, except for MS patients who began DMTs, which showed a decrease in memory Th17 cells and GM-CSF+ T cells at the six-month time point. These findings suggest an overall stability of the different systems in this defined interval in humans without strong exogenous influences. In addition, the consistent results between baseline and six-months strengthen our findings related to differential features between MS and controls, suggesting that these differences are not likely spurious.
While our longitudinal study offers a highly textured view of microbial-host interactions in MS, we acknowledge several limitations. First, the relatively small sample size increases the risk of type II error. To avert this, we carefully chose analytical tools and statistical tests suitable for high dimension data analysis. Second, food diary in our study was self-recorded, which potentially pose selection bias since it is less likely that all food consumptions are completed by all participants. Third, the majority of study participants are female because MS is more prevalent in women. It would be interesting to research next whether these findings still hold true in male MS patients. Lastly, we could not determine the causal connection or directionality of feature-feature interactions. Nonetheless, we provide data on multiple novel molecules that could be plausibly implicated in MS pathogenesis that obligate a more in-depth analysis in the future. A larger study with multiple sampling points that capture both relapse and remitting stages of MS will demonstrate a more complete dynamic picture of multiple-omics interactions in MS.
Declaration of interests
Dr. Evans has been a paid consultant and/or speaker for the following: Biogen, EMD Serono, National MS Society, Genentech/Roche, Novartis, Sanofi/Genzyme & Teva.
Dr. Cross has done paid consulting for: Biogen, Celgene, EMD Serono, Genentech/Roche, Greenwich Biosciences, Janssen and Novartis, and has contracted research funded by EMD Serono and Genentech.
Dr. Tarr is a consultant to, a member of the Scientific Advisory Board of, and a holder equity in, MediBeacon Inc, which is developing a method to test intestinal permeability in humans. He might receive royalty payments if the product generates revenues.
YZ, LP, ALR designed the study. YZ, CC, QL, LG, ZL, YP, KC, YH, HX, AL performed the omics analysis. ALR, MG, YL, CC, LG, performed flow cytometry analysis. YD, SB, HP, CS, EE, LD, RB, KO, PN contributed to the collection and assembly of data. YZ, LP, ALR, PIT, YW, AS, AHC, contributed to the writing of the grant proposal and/or data interpretation. YZ wrote the manuscript, PIT, CC, LG, LP and YD edit the manuscript. YZ and ZL verified underlying data. All authors read and approved the final manuscript.
We thank the Microbial Genomic Services Core at Jackson Laboratory and Metabolome core at University of Massachusetts for data generation.
This work was supported by the Washington University in St. Louis Institute of Clinical and Translational Sciences , funded, in part, by Grant Number # UL1 TR000448 from the National Institutes of Health, National Center for Advancing Translational Sciences, Clinical and Translational Sciences Award (Zhou Y, Piccio, L, Lovett-Racke A and Tarr PI); R01 NS102633–04 (Zhou Y, Piccio L); the Leon and Harriet Felman Fund for Human MS Research (Piccio L and Cross AH). Cantoni C. was supported by the National MS Society Career Transition Fellowship ( TA-1805–31003 ) and by donations from Whitelaw Terry, Jr. / Valerie Terry Fund. Ghezzi L. was supported by the Italian Multiple Sclerosis Society research fellowship ( FISM 2018/B/1 ) and the National Multiple Sclerosis Society Post-Doctoral Fellowship ( FG-1907–34474 ). Anne Cross was supported by The Manny & Rosalyn Rosenthal-Dr. John L. Trotter MS Center Chair in Neuroimmunology of the Barnes-Jewish Hospital Foundation.
The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Data sharing statement
The raw reads of the microbiome data were deposited in the short reads archive database (SRA) (accession no. PRJNA634779).