A bidirectional association between the gut microbiota and CNS disease in a biphasic murine model of multiple sclerosis

Induction of EAE modifies the composition of the gut microbiota in NOD mice at early stages of disease

We induced active EAE in 10 week-old female NOD mice as described previously.^3-5 EAE in NOD mice shows an incidence of disease of approximately 75%. In our study, 31 of 45 total EAE-induced mice (68.9%) exhibited disease progression, which is similar to what was described previously (Fig. 1A, Table 1). Mice initially developed a short phase of disease with reduced severity that subsequently progressed on to a severe form of EAE (severe EAE group). When averaging the clinical scores of the remaining 14 mice (31.1% of EAE-induced mice), the resulting pattern of EAE clinical scores showed a continuation of mild EAE disease throughout the duration of the experiment. Although one of the 14 mice developed a score of 2 (day 34), these mice did not progress into a severe form of EAE and, therefore, were grouped as ‘mild EAE’ in the study (Fig. 1A, Table 1). EAE scores recorded for both mild and severe EAE groups showed an initial phase of disease that persisted from approximately day 9 through day 35 with scores averaging between 0.5 and 1. At day 35, the clinical disease scores for the severe EAE group progressed into greater disability of EAE with no apparent remissions, which was not observed in the mice grouped as mild EAE (Fig. 1A). A third group of control mice were not immunized with MOG_35–55/CFA and did not develop any spontaneous symptoms of EAE (no EAE group) (Fig. 1A). The area under the curve for the severe EAE mice was significantly elevated compared with the mild EAE mice; this difference also occurred between the severe EAE mice versus no EAE mice (Fig. 1B). In contrast, no significant differences were observed in the onset of disease between mild and severe EAE mice (Fig. 1C). Also, no significant differences were observed in the area under the curves for mild EAE mice vs. no EAE mice (Fig. 1B). The body weights of mice with EAE and controls were monitored weekly and no significant differences were observed when all EAE mice were grouped together (not shown). However, when the percentage of the initial weight was compared based on the form of disease that developed (mild vs. severe), we observed a significant decrease in weight gain in severe EAE vs. no EAE control mice (Fig. 1D).

To evaluate the effects of EAE induction on the composition of the gut microbiota at different stages of the disease, we grouped the mice to be analyzed at the end of the experiment when an overall assessment of disease severity could be established for individual mice (Fig. 1C). The samples were compared longitudinally from control mice without EAE (no EAE), mice that developed a mild form of the disease (scores 0.5 or 1; mild EAE), and mice that developed disease that corresponded with EAE scores equal to or above 1.5 for at least 2 consecutive days (severe EAE). Stool samples were collected at the day of EAE induction (day 0), day 14, day 30 and day 58. The clinical scores at the times of the collection are shown in Figure 2A. In total, stool samples were isolated and compared from 15 out of the 70 mice used (Fig. 1 and described in Table 1). Mice within each group were never housed with mice from other groups. An analysis of the gut microbiota was performed by amplicon 16S rRNA (rRNA) gene sequencing to evaluate the relative abundances of the taxonomic groups and the total numbers of taxonomical levels: phylum, class, order, family, genus, and species. To determine whether the stage of disease affected the composition of the microbiome, we compared the level of similarity in the relative abundances of each taxonomic level identified at different time points throughout the course of EAE disease. To ascertain the effects of the genetic background of the stool microbiota of each individual animal compared, we first analyzed all samples prior the initiation of the experiment (day 0). The analysis was performed at the genus level by ordination analysis using nonmetric dimensional scaling (NMDS) in 2 dimensions (Fig. 2B). We used the NMDS combination with the lowest stress, i.e. the result having the smallest difference between the original dissimilarity of samples and generated after the simplified combination of the original taxa levels. After NMDS analysis, we applied the ADONIS multivariate non-parametric ANOVA and ran the ADONIS analysis with 10,000 permutations to obtain a distribution of p values per comparison. By using the NMDS method, we aimed to obtain the optimal combination of the original taxa levels that illustrated the composition of the microbial communities across samples. No significant differences were observed among individuals grouped as no EAE, mild and severe EAE at day 0 (p = 0.895) (Fig. 2 and Table 2). When groups were compared individually (No EAE vs. Mild, No EAE vs. Severe and Mild vs. Severe EAE), no significant differences were observed (Table 2).

The Shannon diversity index, indicative of richness and evenness of the microbial community structure, of the stool samples isolated from mild EAE mice on day 14 was elevated when compared with no EAE mice (Fig. 2C: p = 0.0267). Although also increased in severe EAE mice, the results did not show any significant differences. Interestingly, when comparing the observed taxa (species level) of severe EAE mice vs. no EAE mice, a significant reduction was observed at day 58, indicating a lower number of species identified in EAE mice at the end of the experiment (Fig. 2C; p = 0.0234).

We next compared the overall composition of the microbiome of the 3 different groups on days 14, 30 and 58 at the genus level by 2-dimensional NMDS (Fig. 3A and Table 2). We first compared the microbiome composition of the control group (no EAE) at day 0 vs. days 14, 30 and 58 and observed no significant differences (day 0 vs. 14: p = 0.0961; day 0 vs. 30: p = 0.0960, day 0 vs. 58: p = 0.0961). Similarly, no significant differences were observed among the composition of the microbiome on day 14 vs. 30 (p = 0.0962), day 14 vs. 58 (p = 0.0962) and day 30 vs. 58 (p = 0.0959) (graphs not shown). When comparing the taxa grouped at the genus level on day 14, we observed an overall significant difference between the composition of the gut microbiome of controls without EAE, mild and severe EAE on days 14 (p = 0.0034) and day 30 (p = 0.0096), but not on day 58 (p = 0.1915) (Fig. 3A and Table 2).

On day 14 after disease induction, the gut microbiome of EAE mice was significantly different from the gut microbiome of control mice; No EAE mice vs. mild EAE (p = 0.0406) and no EAE vs. severe EAE (p = 0.0167). Furthermore, by comparing mild vs. severe EAE mice we observed that the extent of the disease also significantly affected the structural composition of the intestinal microbiome (p = 0.0327; Table 2) (Fig. 3A).

On day 30 post-EAE induction, significant differences were observed between mice with mild vs. severe EAE (p = 0.0407) and between mice with mild EAE and controls (p = 0.0168) (Table 2). In contrast, no overall significant differences were observed in the composition of the gut microbiome between the 3 experimental groups on day 58 post-EAE induction (Fig. 3A). Similar NMDS analysis and statistical comparison was performed at the phylum, class, order and family level, but no significant differences were observed when comparing all taxonomic levels at once (not shown). Based on these results, we postulate that EAE disease does in fact alter the composition of the gut microbiome of NOD mice. Furthermore, retrospective comparison indicates that animals that ultimately developed a severe form of EAE showed more profound alterations in the composition of the gut microbiome at early stages of disease.

At the phylum level, no significant differences were found when the identified sequenced counts were compared among groups (Fig. S1 for individual mice, Fig. S2 for averages per group, and Table S1). We next compared the effects of disease progression and severity on the relative abundances of members of genera among all major phyla. The relative abundances at the genus level are shown in Figure 3B (average relative abundances for n = 5 per group and timepoint) and Figure S6 (for individual samples per group and timepoint). The remaining taxonomic levels are represented in supplemental figures (Phylum: Figs. S1–2; class: Fig. S3; Order: Fig. S4; Family: Fig. S5). Since the disease progression and severity impacted the overall microbiota structure (Fig. 3A and Table 2), we performed side-by-side comparisons among the sequence counts at the genus level among the 3 experimental groups (Tables S2–5, for day 0, 14, 30, and 58). A total of 39 out of the 158 genus level taxa showed raw p values below 0.05 when the 3 experimental groups were compared individually at any timepoint evaluated (Tables S2–5). However, the analysis showed no significant differences among genera when analyzing the adjusted p values obtained. We observed changes in the genus-level sequence counts of genera composition of the gut microbiota that resulted from the induction of disease, primarily noticed in the early stage of disease. An undetermined genus of the family Ruminococcaceae and the genus Akkermansia (Fam. Verrucomicrobiaceae) were increased in severe EAE mice vs. no EAE mice (Fig. S7). The genus Turicibacter was also increased in severe and mild EAE when compared with controls, although at day 0 a raw p = 0.008 / adjusted: p = 0.206 was observed when comparing no EAE vs. mild EAE mice. The increased counts of undefined genus of Ruminococcaceae and genus Akkermansia in EAE mice were progressively reduced with disease. The numbers of Akkermansia were reduced in severe EAE mice to the point that at day 58 a raw p = 0.008 / adjusted: p = 0.21 was observed when compared with no EAE mice (Fig. S7).

The opposite pattern was observed for an undetermined genus of the family Christensenellaceae (Order Clostridiales) being increased in no EAE mice on day 14 (no EAE vs. mild EAE: raw p = 0.008 / adjusted p = 0.17; no EAE vs. severe EAE: raw p = 0.008 / adjusted p = 0.22) and day 30 (no EAE vs. mild EAE: raw p = 0.008 / adjusted p = 0.20; no EAE vs. severe EAE: raw p = 0.008 / adjusted p = 0.21). The increased counts of Christensenellaceae remained elevated on day 58 although no significant raw (and adjusted) p values were detected. A similar pattern was observed in members of the genus Lactobacillus (fam. Lactobacillaceae, Order Lactobacillales). Lactobacillus spp. were reduced in mice with severe EAE when compared with control mice on day 14 (no EAE vs. severe EAE: raw p = 0.008 / adjusted p = 0.17) (Fig. S7).

Taken together, our results obtained from a retrospective analysis of stool samples from mice that were grouped based on the severity of their disease (mild vs. severe) suggested that mice that develop the most severe disease exhibit significant variations in gut bacterial genera at early stages of disease but that many of these variations are not long-lived and can continually change throughout the course of the disease (Fig. 3). Our findings, using an extended non-acute EAE model, indicate that different time points during the disease may differentially affect the gut microbiome. When comparing the individual genera, our findings show different patterns of change while disease progresses. However, our preliminary approach was unable to identify specific significant changes when adjusting the p values to the total number of taxa observed.

Early alteration of the gut microbiota results in reduced disease progression and severity of EAE during the chronic stage

We and others have previously documented the effects of the gut microbiota in regulating the progression of acute EAE when SJL and B6 mice were treated orally with broad spectrum antibiotics before the induction of disease.^6,17 Our work showed that treatment with antibiotics affected the balance of regulatory and proinflammatory responses in mice with EAE and that regulatory T cells (Tregs) were, at least in part, responsible for the protection achieved by modulation of the microbiota.⁶ Similar observations were reported by Yokote and colleagues when altering the gut microbiota using a different combination of antibiotics,¹⁷ but the authors proposed a protective mechanism dependent on invariant NKT cells. Studies in germ free mice have further confirmed the relevance of the microbiota in the establishment and progression of EAE disease.^18,19 In this current work, we evaluated whether oral antibiotic treatment would affect the progression of the secondary chronic stage of EAE in NOD mice. NOD mice were treated orally with neomycin, vancomycin, metronidazole, and ampicillin in the drinking water for 2 weeks (from day 0 of EAE induction to day 14). Control mice received normal drinking water. When comparing the course of the disease, the average clinical scores were significantly reduced in mice treated with antibiotics at multiple time-points after day 40 (Fig. 4A). An analysis of the area under the curves showed a significant reduction in the clinical disease scores of mice treated with antibiotics compared with untreated mice (Fig. 4B; p = 0.0162). Furthermore, treatment with antibiotics significantly delayed the onset of EAE (Fig. 4C; p = 0.0005) and reduced the overall severity index compared with untreated mice (Fig. 4D; p = 0.008). Because of the potential impact of antibiotic treatment on the overall health of the animals, we monitored the body weights in both groups weekly. At day 6 of the treatment, we observed a significant reduction in body weights of treated mice when compared with controls (Fig. 4E). However, the weights were restored by day 10, and overall no significant differences were observed by comparing the area under the curves (not shown; p = 0.7295). Flow cytometric analysis indicated that during EAE the frequency of Foxp3⁺ Tregs increased in the Peyer's patches (Fig. 4F). We also observed similar increases in CD39⁺ T cells and Foxp3⁺CD39⁻ Tregs.

We next compared the ability of oral antibiotics to reduce the severity of EAE disease when administered therapeutically after the onset of disease symptoms. This could only be accomplished using mice that survived the initial wave of CNS inflammation. Groups of mice were treated with antibiotics from day 30 to day 44 or day 69 to day 83 post-EAE induction. Control mice continued to receive normal drinking water. Overall, the mice that received antibiotics during the later stages of the disease showed a slight improvement in the clinical scores, however, no significant differences between antibiotic-treated and control groups were observed when all mice were pooled (Fig. S8). These findings suggest that early interventions targeting gut microbiota composition can affect the progression of disease in the long-lasting biphasic model of EAE in NOD mice.

Mice and treatments

Ten-week old female NOD ShiLt (NOD/ShiLt) mice were obtained from the Jackson Laboratories. All animal care and procedures were in accordance with at Eastern Washington University institutional policies for animal health and well-being.

Mice subjected to oral antibiotic treatment were given drinking water with neomycin (1 g/L) (Fisher Bioreagents, BP2669–25), vancomycin (0.5 g/L) Fisher Bioreagents, BP2958–1), metronidazole (1 g/L) (Acros Organics, 210340050) and ampicillin (1 g/L) (Fisher Bioreagents, BP1760–25) for 2 weeks. Control mice received standard drinking water. Three different treatment windows were used depending on the experiment: 1) day 0 of EAE induction to day 14, 2) day 30 to day 44, or 3) day 69 to day 83. Body weights were measured on day 0 of EAE induction (at the initiation of treatment with antibiotics), every 3 d during antibiotic treatment, and weekly following the cessation of antibiotic treatment.

EAE induction

EAE was induced with a Hooke Kit™ for EAE Induction (Hooke Laboratories, EK-2110) containing sufficient material per mouse for a single subcutaneous (s.c.) challenge with 200 µg MOG_35–55 emulsified in 200 µl of complete Freund's adjuvant (CFA). On days 0 and 1 post-challenge, mice received 400 ng of Bordetella pertussis toxin intraperitoneally (i.p.) (List Biological Laboratories, Campbell, CA; provided with Hooke Kit™ for EAE Induction). Mice were monitored and scored daily for disease progression as previously shown:⁹ 0, clinically normal; 0.5, limp tip of tail (when picked up by base of tail, the tail has tension except for the tip); 1, limp tail; 1.5, limp tail and hind leg inhibition, slightly wobbly walking; 2, limp tail, hind leg weakness, wobbly walking, poor balance; 2.5, limp tail and hind legs dragging, poor balance; 3, limp tail, hind legs show complete paralysis, or limp tail with paralysis of one front leg and one hind leg; 3.5, limp tail, complete hind legs paralysis, and mouse moving but when placed on its side is unable to right itself; 4, limp tail, complete hind limb paralysis, partial front leg paralysis (mouse is minimally moving but appear to be alert and feeding); 5, mouse with minimal movement in front legs. No response to contact. Euthanized mouse due to severe paralysis. Per IACUC guidelines, mice were killed at scores of 3.5 and higher. The first of 2 concurrent days of scores of 0.5 or higher was documented as the onset of disease. NOD mice develop spontaneous type 1 diabetes (T1D) at week 14–20 of age³ unless administered CFA, which prevents the induction of T1D,^28-30 and this approach is currently used to maintain NOD breeding colonies. We have performed multiple studies, including those described in this work. Among the approximate total of 120 mice used, only one mouse, at late stages of disease, showed clinical symptoms of T1D, with weight loss, excessive urination, and blood glucose levels above 300 mg/dL.

Isolation of fecal samples and 16S rRNA (rRNA) sequencing

Stool samples were collected aseptically on days 0, 14, 30 and 58 and stored at −80°C. Samples were sent to AKESOgen (Norcross, GA) for 16S rRNA analysis of the microbiome. Qiagen DNA stool extraction kits were used for DNA isolation. One ng/ml of DNA aliquots were analyzed by PCR using primers specific to the variable region 4 (V4) of prokaryotic 16S rRNA gene. Library preparation and sequencing for V4 amplicon sequencing was performed on the Illumina MiSeq platform. Modified protocol with Nextera XT kit was used for library prep and sequenced using MiSeq V2(2 × 250bp) chemistry. AKESOgen used a protocol that combined the 2-steps in 1-step of amplification with forward primer(515F) and indexed-reverse primer(806R). Once the sequencing was performed, we used a cloud-based web application for analyzing microbiome data from the Office of Cyber Infrastructure and Computational Biology (OCICB), National Institute of Allergy and Infectious Diseases (NIAID). Nephele. http://nephele.niaid.nih.gov/ (2016). QIIME was used for analysis³¹ and R for statistical analysis. A total of 60 samples were compared using the QIIME FASTQ paired end protocol. Samples were pre-processed with a Phred quality score of 19 (99% base pair accuracy). Quality filters were applied to obtain a median sequence length of 253.0 per sample. 23435 total observations were found. The minimum and maximum number of counts obtained per sample were 174,345.0 and 573,715.0 respectively, with a mean +/− standard deviation of 287,059.183 +/− 75305.628. Reads that were demultiplexed were clustered into OTUs open reference approach by comparison with the Greengenes database allowing sequences clustered at 97% similarity. Analysis included the identification of chimeras (6,119 chimeras found) and removal using UCHIME. The OUT Table was rarified to 174,344 sequences per sample.

The abundance of each taxa were analyzed using the phyloseq package in R.³² We visualized the compositional heterogeneity of the microbial community of each sample at every time point and at each taxonomic level using non-metric multidimensional scaling (NMDS³³), the ordinate function in the phyloseq package and the metaMDS function in the vegan package,³⁴ and the Bray-Curtis dissimilarity index. We used 6 dimensions as it produced the lowest amount of stress, defined as the difference between the original dissimilarity and the dissimilarity based on a simplified combination of the original taxa.

Public access to raw sequences can be found at Sequence Read Archive (SRA) at NCBI (BioProject ID: PRJNA383155).

Cell preparation and flow cytometry

Single cell lymphocyte preparations from Peyer's patches were stained using conventional methods. A live/dead fixable fluorescence-labeled viability dye (BD Biosciences, 564407) was used in all staining protocols, and samples were gated on viable cells for subsequent analysis. Cell subsets were analyzed using fluorochrome-conjugated mAbs against T cell antigens CD3 (BD PharMingen, 553062), CD4 (BD PharMingen, 553052), and CD39 (BioLegend, 143804). Intracellular staining for Foxp3 was performed using fluorochrome labeled anti-Foxp3 mAb (clone FJK-16s; eBioscience, 175773–82). Samples were acquired using a BD Accuri C6 (BD Biosciences, San Jose, CA). Data was analyzed with FloJo software (FloJO LLC, Ashland, OR).

Statistical analysis

Parametric and non-parametric t-tests and one-way ANOVA followed by Kruskal-Wallis comparison of multiple groups was applied to show differences in flow cytometric analysis. Area under the curve analysis followed by one-way ANOVA and multiple comparison tests were applied to show differences in EAE scores of no EAE, mild EAE, and severe EAE mice. Two-way ANOVA followed by multiple comparisons test was used to compare the scores of EAE and EAE-ABX mice. In addition, AUC followed by Mann-Whitney test was used to compare scores of EAE and EAE-ABX mice. Weights were compared by two-way ANOVA followed by multiple comparisons test. To compare severity index, Mann-Whitney test was used. To compare the onset of disease Mann-Whitney test was used. p-values < 0.05, < 0.01, < 0.001 and < 0.0001 are indicated. Disease effects on the overall microbiota structure was visualized by plotting the results by NMDS in 2 dimensions. Plots were obtained using the R vegan package. After multiple iterations the one showing the lowest stress was used to generate visuals and for statistical permutational Multivariate Analysis of Variance Using Distance Matrices, ADONIS,³⁵ a permutational form of a multivariate analysis of variance (MANOVA). Since it is a permutational method, it produces a slightly different p-value each time. To obtain reliable p-values, we repeated the ADONIS method 10,000 times and reported the average of those p-values for each taxonomic level, time-point and comparison (Table 2). Genus-level counts were compared using the non-parametric Wilcoxon Rank-Sum test (when comparing 2 sample or matched samples) (Supplemental Tables 1–5, and supplemental figure 7). Differences were considered significant with a p-value < 0.05. The graphs and the statistical analysis were performed using Prism 7 (GraphPad Software Inc., San Diego, CA).