Gut Microbiome Associated With Graves Disease and Graves Orbitopathy: The INDIGO Multicenter European Study

Abstract

Context: Gut bacteria can influence host immune responses but little is known about their role in tolerance-loss mechanisms in Graves disease (GD; hyperthyroidism caused by autoantibodies, TRAb, to the thyrotropin receptor, TSHR) and its progression to Graves orbitopathy (GO).

Objective: This work aimed to compare the fecal microbiota in GD patients, with GO of varying severity, and healthy controls (HCs).

Methods: Patients were recruited from 4 European countries (105 GD patients, 41 HCs) for an observational study with cross-sectional and longitudinal components.

Results: At recruitment, when patients were hyperthyroid and TRAb positive, Actinobacteria were significantly increased and Bacteroidetes significantly decreased in GD/GO compared with HCs. The Firmicutes to Bacteroidetes (F:B) ratio was significantly higher in GD/GO than in HCs. Differential abundance of 15 genera was observed in patients, being most skewed in mild GO. Bacteroides displayed positive and negative correlations with TSH and free thyroxine, respectively, and was also significantly associated with smoking in GO; smoking is a risk factor for GO but not GD. Longitudinal analyses revealed that the presence of certain bacteria (Clostridiales) at diagnosis correlated with the persistence of TRAb more than 200 days after commencing antithyroid drug treatment.

Conclusion: The increased F:B ratio observed in GD/GO mirrors our finding in a murine model comparing TSHR-immunized with control mice. We defined a microbiome signature and identified changes associated with autoimmunity as distinct from those due to hyperthyroidism. Persistence of TRAb is predictive of relapse; identification of these patients at diagnosis, via their microbiome, could improve management with potential to eradicate Clostridiales.

Keywords: Firmicutes:Bacteroidetes ratio; Graves disease; Graves orbitopathy; autoimmunity; gut microbiota; hyperthyroidism.

Figure 1.

Gut microbiota composition in Graves disease (GD) patients and healthy controls at recruitment (T0). A, Distributions of the main phyla Bacteroidetes, Firmicutes, and Actinobacteria in controls (HCs), GD, and Graves orbitopathy (GO) patients. P values are generated from a pairwise t test with Benjamini-Hochberg (FDR) correction. Bootstrapped distribution of the Firmicutes to Bacteroidetes (F:B) ratio from 500 replicates of the data set for B, diagnosis groups (controls, GO and GD) and C, controls vs cases (GD + GO samples).

Figure 2.

Significantly different genera across eye-disease severity (distributed as control, no sign/Graves disease [GD] only, mild, moderate-severe according to the EUGOGO guidelines). A, Bacteroides spp; B, Bifidobacterium spp; C, Fusicatenibacter spp, and D, Roseburia spp are shown. P values are generated from a pairwise t test with Benjamini-Hochberg (FDR) correction: *P less than .05; **P less than .01, and ***P less than .001.

Figure 3.

Analysis of fecal water metabolites associated with Graves orbitopathy (GO). A, Orthogonal projection to latent structure discriminant analysis of the fecal metabolites significantly discriminate between GO patients (both severe and sight-threatening; n = 10) and healthy controls (n = 11). B, GO fecal metabolites were enriched in short-chain fatty acids (butyrate, propionate, etc) compared to those of healthy controls.

Figure 4.

Random forest (RF) prediction of diagnosis based on the gut microbiome at genus level. A, Confusion matrix with the per-class classification for diagnosis (Graves disease [GD], Graves orbitopathy [GO], and healthy controls). Columns represent the true classification while the rows represent the predicted classification. B, Top-12 variable importance for diagnosis classification according to the mean decrease accuracy. The model included the thyroid status, nation of provenance, age, and sex as predicting variables. Thyroid status and nation of provenance were also identified among the most important variables. C, Confusion matrix with the per-class classification of the eye disease (no signs, mild, moderate-severe compared to healthy controls) and D, top-12 variable importance for eye disease classification according to the mean decrease accuracy.

Figure 5.

Correlations between thyroid function and the Graves' disease (GD) or Graves' orbitopathy (GO) gut biomarkers. Pearson's correlation coefficient was used to correlate the relative abundances of the selected bacterial biomarkers to the standardized levels of free-thyroxine (fT4), thyroid-stimulating hormone (TSH) and TRAB) in either GD or GO patients. All correlation coefficients are shown. Significant correlations are shown according to t-test P-value: * P<0.05; ** P<0.01 and ***P<0.001. Mean and standard deviation of each bacterial biomarker was shown for GD and GO groups.

Figure 6.

Differences in gut microbiota associated with thyroid autoantibody levels (TRAb). A, Rationale of the longitudinal study. N = 20 patients enrolled at baseline (T0; all TRAb positive per diagnosis) provided a follow-up fecal sample between 218 and 644 days (median 471 days) and here referred to as “200+ days” (T1). At follow-up (T1), 11 of 20 patients remained TRAb positive while 9 of 20 became TRAb negative. Long-TRAb refers to TRAb measurements taken at 200+ days. B, Heat map of the mean values of differentially abundant genera between TRAB-positive and TRAB-negative patients either at baseline (T0) or at T1. Fold change was calculated as the (mean TRAb positive – mean TRAb negative)/mean TRAb-negative values. C, Paired differences of Dorea sp between T0 and T1 in those patients who remained TRAb positive (n = 11) and D, paired differences of Butyrivibrio sp between T0 and T1 in those patients who became TRAb negative (n = 9). Paired t test corrected for Benjamini-Hochberg multiple tests.

Figure 7.

The INDIGO cohort gut microbiota signature attempted to identify operational taxonomic units (OTUs; out of necessity this covers all taxonomic levels from species to family) associated with disease status, autoantibody titers (TRAb), and free thyroxine (fT4) levels. We listed differentially abundant OTUs (from the discriminant analysis models) and/or important variables (from the random forest models) associated with TRAb, diagnosis (Graves disease [GD], Graves orbitopathy [GO], or controls), and thyroid status (whether hyperthyroid or euthyroid). We noted that all genera that associated uniquely with TRAb were Firmicutes of the Clostridiales family (the human equivalent of segmented filamentous bacteria implicated in several autoimmune conditions) (3). Similarly, genera uniquely associated with thyroid status were predominantly Clostridiales. In contrast, genera uniquely associated with disease status coincided predominantly with changes in genera from the Bacteroidetes, Proteobacteria, and Actinobacteria phyla. Intersections of the Venn plots represent the common significant genera between groups. Clostridiales are common both to TRAB and disease status and predominate in the overlap between thyroid and disease status along with members of the Bacteroidetes and Bacillota phyla. There was no significant overlap between autoantibodies analysis and thyroid status analysis. Owing to the nature of the experimental/sequencing procedure and analyses, this analysis is not intended as a mechanistic explanation of the microbiota role in the outcome of GD/GO.