Gut microbial interaction networks control autoimmunity to neuroretina

Abstract

The gut microbiome influences the development of immune-mediated inflammatory diseases. One such condition is autoimmune uveitis, a sight-threatening ocular inflammation driven by retina-specific T cells¹. Using a model of spontaneous experimental autoimmune uveitis (sEAU) we showed that gut commensals provide innate and adaptive immune stimuli that trigger the disease². Here we report that uveitis-promoting microbes are present in human gut flora and that colonization of germ-free (GF) mice with commensal flora from healthy human donors was sufficient to provoke disease. Severity of sEAU correlated with expansion of Akkermansia and contraction of short-chain fatty acid (SCFA)-producing Firmicutes, as well as decreased SCFA levels and a dominant gut Th1 effector response. Mechanistic gain-of-function experiments, enriching GF sEAU mice with Akkermansia, reproduced these microbiome, metabolite and immune phenotype shifts, and exacerbated disease. We propose that Akkermansia promotes autoimmunity by outcompeting SCFA-producers and enhancing Th1-type responses. Notably, an inverse correlation between Akkermansia (Verrucomicrobia) and Firmicutes was also present in fecal microbiome of patients with uveitis, multiple sclerosis and Crohn's disease. These findings reveal a stereotypic gut microbial interaction network that regulates systemic immune balance, and may represent an ecologically conserved mechanism through which the gut microbiome modulates autoimmune and inflammatory diseases.

Keywords: T cell receptor transgenic mice; autoimmune uveitis; fecal microbiota reconstitution; humanized gnotobiotics; microbial interaction; short chain fatty acids.

Fig. 1 |. Human gut commensals support development of spontaneous experimental autoimmune uveitis (sEAU).

a, Schematic representation of human fecal microbiota transfer experiments and data collection. b, Beta diversity based on Bray–Curtis dissimilarity showing differences in fecal microbiome composition between recipient groups U (n = 50), V (n = 59), W (n = 64). Encircled with dashed lines are the respective human donor (original) fecal samples. c, Disease scores (histology) of age-matched R161H mice (4–5 months old) from human flora (U, V, W, n = 28, 38, 18), SPF (n = 22) and GF (n = 10) groups (Kruskal–Wallis test between mouse recipient groups, p = 0.0002). d, Immunological clusters of CD4 T cells from colon lamina propria (LP) by the microbiome groups. PCA analysis performed on immune signature matrix by frequencies of CD4 T cells expressing each cell surface or intracellular marker (see Extended Data Fig. 1e). e–g, Frequencies of memory (CD44^highCD62L^low, e), IFN-γ- or IL-17A-producing (f), and regulatory (Foxp3-positive or IL-10-producing, g) CD4 T cells in colon LP of HuFl, SPF and GF R161H mice. d–g, Compiled data from 12 experiments, n = 15, 19, 11, 11, 27 for GF, U, V, W and SPF, respectively. Kruskal–Wallis tests show p < 0.0001 (e–g). P-values adjusted for post hoc multiple comparisons are displayed as asterisks. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and bar plots showing mean values ± SEMs (c, e–g).

Fig. 2 |. Human flora (HuFl) sEAU tracks with a dominant Th1 phenotype and abundance of Akkermansia (Verrucomicrobia).

a, Mean relative frequency of colon LP CD4 T cells of each cell surface or intracellular marker arranged in a circular plot, colored by histology category: high (score ≥ 1.5, n = 10), mid (0.75 < score < 1.5, n = 8), low (score ≤ 0.75, n = 17). b, Correlational heatmap of histology scores with relative frequencies of each cell surface or intracellular marker in CD4 T cells across six tissue sites. LP, lamina propria; LN, lymph nodes; Sm Int, small intestine; Mes, mesenteric; Subm, submandibular; Periph, peripheral. ^△0.05 < P < 0.08, *P < 0.05, **P < 0.01, ***P < 0.001, modeled by simple linear regression. c–d, Representative trend lines with 95% confidence interval (shaded area) show that disease scores of HuFl mice are positively associated with IFN-γ-producing, but not IL-17A-producing, CD4 T cells, in colon LP (c) and small intestine LP (d). Data in a–d are aggregated from 8 experiments (UVW combined, sample sizes ranged from n = 27 to 39 across tissues following quality filtering). e, Correlational heatmap showing major fecal bacterial taxa (out of top 20 most abundant families) that correlate with disease scores in adult HuFl R161H (UVW combined, n = 105). P-values assessed after FDR (false discovery rate) correction following Pearson correlation. *P < 0.05. f–g, More Verrucomicrobia (f), and fewer Firmicutes (g) are present in the gut of HuFl mice with high disease scores. Sample sizes are n = 36 and 43 for high and low histology categories, respectively. Data in bar plots are presented as mean values ± SEMs, and P-values from Mann-Whitney U tests are shown in plots.

Fig. 3 |. SCFA signatures are enriched in low disease HuFl mice or healthy subjects.

a, A simplified pathway map highlighting functional genes involved in the main SCFA synthesis and metabolism. Dashed arrows indicate hidden intermediate steps. Abbreviations of enzyme symbols: pta, Phosphotransacetylase; ackA, Acetate kinase; acs, Acetyl-CoA synthetase; but, Butyryl-CoA:acetate CoA-transferase; ptb, Phosphate butyryltransferase; buk, Butyrate kinase; crt, Crotonase; bcd, Butyryl-CoA dehydrogenase. b, Enzymes in SCFA pathways (annotated with InterPro and Gene Ontology databases) are enriched in HuFl mouse gut with low histology scores (UVW combined, weanling and adult samples included). PPM, parts per million, a measure of relative abundance based on normalized base count for each feature in a metagenomic sample. c, Spearman correlation (visualized with a linear trend line) showing higher disease scores associated with lower SCFA concentrations both locally (in gut content) and systemically (in serum) in representative cohort U mice (n = 9). d, Enzymes in SCFA pathways (annotated with Gene Ontology and Pfam databases) are enriched in fecal samples from healthy humans compared to those from treatment-naïve uveitis patients, from a cohort of autoimmune uveitis (ocular-restricted, including idiopathic uveitis, VKH/SO and BCR). Sex, age, race and diet are comparably represented in patient and control groups. Normalized gene abundance is reported in PPM. e, SCFA levels from uveitis patients and healthy controls (same cohort as in d, but n = 6 available from the patient group). For b, d & e, Data are presented as mean values ± SEMs. Statistical significance was determined by Mann-Whitney U test.

Fig. 4 |. Akkermansia negatively correlates with Firmicutes in both sEAU and clinical uveitis.

a, Simple linear regression showing a negative correlation of microbial relative abundance between Akkermansia (equivalent to Verrucomicrobia in this dataset) and Firmicutes in HuFl R161H mice (n = 234, UVW combined, weanling and adult samples included). b, Fecal bacterial association network exemplified in HuFl cohort V (n = 79, weanling and adult samples included). Network constructed at the taxonomic level of the bacterial family. Correlations with a threshold of Pearson r > 0.5 were transformed into dissimilarities via the “unsigned” distance metric, on which edge weights (strength of correlations) are based. Blue edges correspond to positive associations and orange edges to negative ones. Yellow and purple nodes represent taxa enriched in high vs. low disease categories, with Firmicutes labeled as bolded taxon names. Bigger nodes are more influential in the network (higher eigenvector centrality). c, Spearman correlation (visualized with a linear trend line) showing a negative association between relative Akkermansia abundance and SCFA concentrations in R161H gut (n = 35, UVW and SPF combined). Akkermansia was selected following multivariable association between SCFA concentrations and microbial features using MaAsLin2. d, Class level association networks in treatment-naïve uveitis patients (left, n = 8) and healthy controls (right, n = 12) from the same cohort of autoimmune uveitis. Correlations with a threshold of Pearson r > 0.3 are shown. Node and edge color schemes same as in b. Shaded ellipse in the patient network highlights the negative correlation between Verrucomicrobiae (Akkermansia) and Clostridia (Firmicutes), and this relationship is absent from the healthy human gut.

Fig. 5 |. Akkermansia promotes uveitis by outcompeting SCFA producers and decreasing circulating SCFA levels.

a, Study design of microbe reconstitution experiments. Previously GF R161H mice gavaged with a baseline human gut microbiota suspension (HuFl) plus A. muciniphila culture (AKK, 10⁸ CFU) or sterile PBS. b, Relative abundance of Akkermansia and Firmicutes in feces from one representative of 3 experiments (n = 7 and 6 for Akk-rich and control, respectively). c, Left: Fundoscopy score increments from starting disease scores, combined from three experiments. Right: Histology scores and example images from 2 out of 3 experiments (n = 11 and 7 for Akk-rich and control, respectively). Mean values ± SEM. P-value determined by Welch-corrected t test. d, A negative correlation of relative abundance between Akkermansia and Firmicutes (n = 52, AKK and PBS groups combined, multiple time points included) by simple linear regression. e, Serum concentrations of main SCFAs are significantly lower in Akk-rich (n = 13) mice than in PBS controls (n = 10). Serum harvested 3 – 4 weeks post gavage, three experiments combined. f, Correlational heatmap of SCFA levels with cytokine concentrations in serum of engrafted mice. *P < 0.05, modeled by simple linear regression as well as Pearson correlation. g, Spearman correlation (visualized with a linear trend line) showing Akkermansia relative abundance associated positively, and serum acetic acid associated negatively, with the frequency of IFN-γ-producing CD4 T cells in colonic LP. h, Akk-enriched (n = 18) mice had higher serum IFN-γ concentrations compared to PBS control mice (n = 14), five experiments combined, of which two experiments used the minimal flora model. i, Representative trend lines show that main SCFA concentrations are negatively associated with IFN-γ levels in serum. Data in g, i are from 2 experiments (n = 12) using the minimal flora model.

Fig. 6.. The inverse relationship between Verrucomicrobia and Firmicutes extends to clinical uveitis, multiple sclerosis (MS) and Crohn’s disease.

a, Graphical summary of the working model. b, A positive correlation between V/F (Verrucomicrobia/Firmicutes) ratios and histology scores of HuFl mice by simple linear regression, and V/F ratios compared between high and low disease of HuFl mice. c, V/F and F/B (Firmicutes/ Bacteroidetes) ratios compared between patients from uveitis, MS and IBD and their respective healthy controls. Analyzed datasets of MS and IBD (Crohn’s disease and UC) and healthy controls were from public sources. Uveitis and MS cohorts included those established on treatment. For F/B ratio in the IBD cohort, samples with <0.1% relative abundance of either Bacteroidetes or Firmicutes were removed from the analysis. Data are presented as mean values ± SEM. Statistical significance was determined by Mann-Whitney U test (one-tailed analysis was applied for uveitis and MS) and Kruskal–Wallis test with P-values adjusted for post hoc multiple comparisons (for IBD).