Gut Dysbiosis Contributes to the Imbalance of Treg and Th17 Cells in Graves’ Disease Patients by Propionic Acid

Abstract

Background

Graves’ disease (GD) is a typical organ-specific autoimmune disease. Intestinal flora plays a pivotal role in immune homeostasis and autoimmune disease development. However, the association and mechanism between intestinal flora and GD remain elusive.

Objective

To investigate the association and mechanism between intestinal flora and GD.

Methods

We recruited 58 initially untreated GD patients and 63 healthy individuals in the study. The composition and metabolic characteristics of the intestinal flora in GD patients and the causal relationship between intestinal flora and GD pathogenesis were assessed using 16S rRNA gene sequencing, targeted/untargeted metabolomics, and fecal microbiota transplantation.

Results

The composition, metabolism, and inter-relationships of the intestinal flora were also changed, particularly the significantly reduced short-chain fatty acid (SCFA)-producing bacteria and SCFAs. The YCH46 strain of Bacteroides fragilis could produce propionic acid and increase Treg cell numbers while decreasing Th17 cell numbers. Transplanting the intestinal flora of GD patients significantly increased GD incidence in the GD mouse model. Additionally, there were 3 intestinal bacteria genera (Bacteroides, Alistipes, Prevotella) could distinguish GD patients from healthy individuals with 85% accuracy.

Conclusions

Gut dysbiosis contributes to a Treg/Th17 imbalance through the pathway regulated by propionic acid and promotes the occurrence of GD, together with other pathogenic factors. Bacteroides, Alistipes, and Prevotella have great potential to serve as adjunct markers for GD diagnosis. This study provided valuable clues for improving immune dysfunction of GD patients using B. fragilis and illuminated the prospects of microecological therapy for GD as an adjunct treatment.

Graves’ disease, intestinal flora, Treg/Th17, Bacteroides fragilis, propionic acid

As a typical organ-specific autoimmune disease, one of the main pathological features of Graves’ disease (GD) is the presence of various antithyroid antibodies, such as thyrotropin receptor antibody (TRAb) (1). Especially, the presence of thyroid-stimulating antibody (TSAb), one type of TRAb, is the unique characteristic of GD patients. TSAb is directly responsible for the elevated level of thyroxine and GD development by stimulating the TSH receptor (2). The pathogenesis of GD is very complicated. In addition to genetic factors (1), nongenetic factors, including the environment, play key roles in the destruction of immune tolerance to thyroid tissue and autoantibody production (3). However, the exact mechanism involved in these processes has not yet been fully elucidated.

Regulatory T (Treg) cell is a subset of T helper cells and can maintain immune homeostasis by secreting inhibitory cytokines (eg, IL-10 and TGF-β), playing pivotal roles in immune tolerance (4). The removal of Treg cells significantly increases the incidence and severity of autoimmune diseases, including GD and autoimmune thyroiditis, in mice (5). The number and/or function of Treg cells in the peripheral blood and thyroid of patients with GD were markedly reduced. However, some studies do not agree with this result (6). The Th17 cell is also the subset of T helper cells, determined by high secretion of IL-17 and other proinflammatory cytokines. Th17 cells are closely associated with the development of many autoimmune diseases (7–9). However, there are still conflicts about the relationship between Th17 cells and GD. Some studies showed that IL-17 and Th17 cells were increased in GD and were involved in the development of GD (10). However, there are also some contradictory findings from other studies (11). The reasons for the above discrepancy may lie in the differences in animal models, races of GD patients, treatments, stages of disease, and the assay methods and/or marker molecules used for Treg and Th17 cells. Our previous study found that inhibition of the retinoic acid signaling pathway caused by abnormal expression of microRNAs was one of the important causes of Treg cell abnormalities in GD patients (12). Nevertheless, the molecular mechanism of the dysfunction of Treg and Th17 cells in GD patients is still poorly understood.

Recently, many studies have shown that gut microbiota can regulate various immune cells, including Treg and Th17, in the intestinal mucosa and gastrointestinal-associated lymphoid tissues (13). Additionally, various intestinal bacteria and their metabolites and constituents (such as peptide polysaccharides) can enter the circulation to regulate Treg/Th17 cells and the immune system in extraintestinal tissues/organs. Moreover, the immune cells, including Treg and Th17 cells in the intestine, which are modulated by intestinal flora, can also enter the circulation to regulate immune responses in remote tissues and organs (14). Short-chain fatty acids (SCFAs) produced by some intestinal microbes like Bacteroides can regulate Treg and Th17 cells in the intestine, circulation, and extraintestinal tissues (15). Therefore, the intestinal flora is a key regulator of immune tolerance and inflammation, and it plays pivotal roles in the development of various intestinal and extraintestinal autoimmune diseases. For example, increases in the abundance of proinflammatory bacteria (Escherichia coli strains) and decreases in the abundance of anti-inflammatory bacteria (Faecalibacterium prausnitzii) are closely related to autoimmune intestinal diseases, including Crohn’s disease and ulcerative colitis (16, 17). Clostridiosis and Bacteroides fragilis (B. fragilis) have protective effects in autoimmune encephalomyelitis mice (18, 19). This implies that gut dysbiosis might be one of the most important causes of abnormalities in Treg and Th17 cells and impaired immune tolerance.

Diarrhea is an important clinical feature of GD and is associated with gut dysbiosis and infections of intestinal pathogenic bacteria (20, 21). Many studies have shown that Yersinia enterocolitica is related to GD, but mice fed only with Yersinia enterocolitica did not develop GD (22, 23). Therefore, considering the close relationship between intestinal flora and inflammation and/or autoimmunity, very interesting questions arise as to whether gut dysbiosis is associated with abnormal immune regulation (such as the abnormalities in Treg and Th17 cells) in GD patients and what the precise mechanism underlying this association is. Recently, a study reported the compositional abnormalities in intestinal flora of GD patients and speculated on the relationship between gut flora and GD development (24). However, the statistical methods used in this study should be optimized, and this descriptive study did not explore the mechanisms underlying the association between GD and intestinal flora (24). Therefore, more research is needed to answer the above-mentioned questions.

Herein, the profiles of the gut microbiome and metabolome of GD patients were investigated using 16S rRNA gene sequencing and targeted/nontargeted metabolomic techniques. Upon integrating with clinical data, we discovered the association between immune dysfunctions and abnormalities in the composition and metabolism of intestinal flora in GD patients, and revealed the mechanisms underlying these associations for the first time. The reduced abundance of SCFA-producing bacteria, including B. fragilis in the intestine of GD patients, markedly decreased the levels of SCFAs, especially propionic acid. These abnormalities lead to the decreased number of Tregs and increased the number of Th17 cells, resulting in the Treg/Th17 imbalance and inflammation in GD patients. This mechanism may play an important role in impaired immune tolerance and pathogenesis of GD.

Materials and Methods

Ethics statement

This study was approved by the Ethics Committee of Shandong Provincial Hospital and conformed to the principles of the Declaration of Helsinki (NO 2015-054). Informed consent was obtained from all participants. All procedures were performed in compliance with the relevant laws and institutional guidelines.

Recruitment of subjects and sample collection

We recruited 58 initially untreated GD patients and 63 healthy individuals from Shandong Provincial Hospital. The 2 groups of subjects were matched with respect to age, sex, and body mass index (BMI). All GD patients were diagnosed according to their medical history, physical examination, and laboratory tests, including thyrotropin (TSH), free triiodothyronine (FT3), free thyroxine (FT4), and TRAb, based on the 2018 European Thyroid Association Guideline for the Management of Graves’ Hyperthyroidism. The exclusion criteria were as follows: pregnancy, smoking, alcohol addiction, diarrhea (having loose stool 3 or more times in 1 day, or watery stool) (25, 26), hypertension, diabetes mellitus, lipid dysregulation, and BMI > 27; usage of the following drugs within 3 months: antibiotics, probiotics, prebiotics, symbiotics, hormonal medication, laxatives, proton pump inhibitors, insulin sensitizers, or Chinese herbal medicine; and medical history of autoimmune diseases, malignancy, and gastrointestinal tract surgery. Peripheral blood (5 mL) was collected from all subjects in the morning after an overnight fast (≥8 h). The serum samples were stored at -80°C for cytokine and lipopolysaccharide (LPS) detection. Peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll density centrifugation (Sigma-Aldrich, St. Louis, Missouri). Commode Specimen Collection Kits were provided to all subjects for stool collection, and the collected fecal samples were stored at -80°C after liquid nitrogen freezing.

Assays of thyroid function and thyroid-related autoantibodies

Serum levels of TSH, FT3, FT4, thyroglobulin antibody (TGAb), thyroid peroxidase antibodies (TPOAb), and TRAb of the recruited subjects were assayed using chemiluminescent immunoassays (ADVIA Centaur XP, Erlangen, Germany; Cobas E801 module, Roche Diagnostics, Basel, Switzerland) according to the manufacturer’s instructions. Reference ranges were as follows: TSH: 0.55–4.78 µIU/mL; FT4: 11.5–22.7 pmol/L; FT3: 3.5–6.5 pmol/L; TRAb: 0.00–1.58 IU/L; TPOAb: 0.00–60 IU/mL; and TGAb: 0.00–60 IU/mL. The serum levels of TRAb and total thyroxine (TT4) in mice were measured using enzyme-linked immunosorbent assay (ELISA) kits (Jianglai Biotech, Shanghai, China) and SpectraMax Microplate Reader (Molecular Devices, Sunnyvale, California) according to the manufacturer’s instructions. The normal range was defined as the mean ± 3 standard deviations (SD) of mice in the control group (27).

16S rRNA gene sequencing

Total genome DNA was extracted from the fecal samples of 38 GD patients and 37 healthy individuals using the improved CTAB (cetyl trimethylammonium bromide) method. DNA concentration was determined using the NanoDrop 2000 (Thermo Scientific, Waltham, Massachusetts) spectrophotometer. The V1-V2 regions of the 16S rRNA gene were amplified using polymerase chain reaction (PCR) and universal primers (Univ 27F/Univ 338R) (Table S1) (28, 29). Next, the amplicons were sequenced on an Illumina HiSeq 2500 system (San Diego, California) and 250 bp paired-end reads were generated.

Bioinformatic analysis of sequencing data

The raw sequencing data were merged and filtered using FastQ software. The identification and taxonomic annotation of operational taxonomic units (OTUs) were performed using QIIME software (http://qiime.org/) and the Silva 12.8 database according to the software manual (30). The linear discriminant analysis effect size (LEfSe) analysis was performed to identify differentially abundant bacterial taxa as biomarkers (LDA score > 4 and P< 0.05) (31). KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways and COG (clusters of orthologous groups) functions were analyzed using the PICRUSt (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States) software (32, 33). Spearman correlation was calculated using R package psych. To determine if taxonomic differences in the microbiota could be used to classify samples into different cohorts and the ability of taxonomic differences to distinguish between GD patients and healthy individuals, a machine learning algorithm named Random Forests was applied to analyze the genus-level abundance of gut bacteria using the R package randomForest (34).

Untargeted metabolomic analysis of stool samples

Approximately 100 mg of stool was extracted with 1 mL methanol, and 60 μL 2-Chloro-L-phenylalanine and heptadecanoic acid (0.2 mg/mL) was added as the internal standard. After vortexing, ultrasonic treatment, and centrifugation, the supernatant was transferred into a new tube and dried using vacuum concentration. Next, 60 μL methoxyamine pyridine (15 mg/mL) and N,O-bis(trimethylsilyl)trifluoro-acetamide (BSTFA) reagent were sequentially added and reacted at 37°C. After centrifugation, the supernatant was analyzed using the ACQUITY UPLC system (Waters Corporation, Milford, Connecticut) coupled with AB SCIEX Triple TOF 5600 System (AB SCIEX, Framingham, Massachusetts) in positive and negative ion modes. Data was acquired in full scan (m/z ranges from 70–1000) mode according to the manufacturer’s instructions.

Bioinformatic analysis of metabolomic data

The raw data were converted to mzXML files using MSconvent software. All metabolite ions were normalized and the ions (RSD% < 30%) were further analyzed using SIMCA 14.0 (Umetrics, Umeå, Sweden) and XCMS 1.50 software. After mean centering and unit variance scaling, orthogonal partial least-squares-discriminant analysis (OPLS-DA) was carried out to visualize the metabolic alterations among experimental groups. The peak features were selected based on the variable influence on projection values (>1) of the normalized peak areas. These selected peak features were then used to identify the metabolites with the reference material database, including Human Metabolome Database (HMDB), METLIN, and the standardized materials database that built by Dalian Chem Data Solution Information Technology Co., Ltd. Furthermore, the statistical significances between the 2 groups of metabolites were determined by t-test with false discovery rate (FDR) correction.

Targeted metabolomic analysis of SCFAs using GC–MS

All extraction procedures were performed at 4°C to protect the volatile SCFAs. After vortexing and centrifugation, 800 µL supernatant was added to 800 µL ethyl acetate, and the mixture was vortexed and centrifuged. Afterwards, isohexanoic acid was added into 600 μL upper organic phase at a final concentration of 500 μM as an internal standard. The standards of SCFAs (acetic acid, propionic acid, butyric acid, isobutyric acid, valeric acid, isovaleric acid, and caproic acid) were formulated with ethyl acetate into the concentrations of 0.1, 0.5, 1, 5, 10, 20, 50, 100, 200, and 400 (μg/mL) and were added to 500 μM isohexanoic acid. Next, a 1 μL SCFA standard or sample was analyzed by a 7890A gas chromatography system coupled to a 5975C inert XL EI/CI mass spectrometry (Agilent Technologies, Santa Clara, California). Concentrations of the SCFAs were normalized according to stool weight.

PBMC treatment and the detection of Treg and Th17 cells

The human PBMCs isolated from healthy individuals or GD patients were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum, 100 units/mL penicillin, and 100 g/mL streptomycin. Treg cells were assayed in PBMCs treated with 300 uL M2GSC medium or the supernatant of M2GSC medium culturing B. fragilis YCH46 with the stimulation of CD3/CD28 antibody (Invitrogen, Waltham, Massachusetts), TGF-β (PeproTech, Rocky Hill, New Jersey), and IL-2 (PeproTech, Rocky Hill, New Jersey). The procedure of Th17 assay was the same as above, except that IL-2 was changed to IL-6 (PeproTech, Rocky Hill, New Jersey). The cells and culture supernatants were collected for flow cytometric analysis or ELISA after 3 days of culture.

Flow cytometric analysis

For Treg cell assay, PBMCs were incubated with anti-CD4-FITC (BD Bioscience, San Jose, California), anti-CD25-APC, anti-CCR9-APC (eBioscience, Waltham, Massachusetts), or anti-CCR9-PE/Cyanine7 (BioLegend, San Diego, California) at 4°C for 30 minutes. For Th17 cell assay, after stimulation with the Cell Stimulation Cocktail/Protein Transport Inhibitor Cocktail (eBioscience, Waltham, Massachusetts) at 37°C for 4 hours, PBMCs were stained with antihuman CD4-FITC and antihuman CCR9-APC (eBioscience, Waltham, Massachusetts). After fixation and permeabilization, the PBMCs were stained with anti-FOXP3-PE or IL-17-PerCP/Cy5.5 (eBioscience, Waltham, Massachusetts). Finally, the stained cells were analyzed using a FACSCalibur or FACSAria II flow cytometer (BD Biosciences, San Jose, California) and FlowJo X software.

Bacterial selective culture and species identification

Bacterial isolation was performed using the fecal samples from healthy adult males. One-gram fecal samples were homogenized in 9 mL anaerobic M2GSC medium. After a 10-fold serial dilution, 0.3 mL suspension was inoculated on the 1.5% agar plates containing M2GSC medium at 37°C for 48 hours in Anaeropack Jars. Single colonies were picked and grown overnight in 10 mL M2GSC medium. Next, the gDNA of the isolated bacterium was extracted and the full length 16S rRNA gene was amplified with PrimeSTAR Max Premix (TaKaRa, Kusatsu, Japan) using universal 16S primers (Univ 27F/Univ 1492R) (Table S1) (29). PCR products were purified and sequenced using Sanger sequencing. For species identification, the Sanger sequencing data were blasted using the BLASTN online tool (http://blast.ncbi.nlm.nih.gov/Blast.cgi).

Cytokine and LPS assays

The serum levels of cytokine IL-2, IL-18, and sCD25 were assayed using Multiplex Human Premixed Multi-Analyte Kit (R&D Systems, Minneapolis, Minnesota) on Luminex 200 (Luminex, Austin, Texas) according to the manufacturer’s instructions. The cytokine concentrations were calculated using a standard curve of recombinant cytokine standards. The concentrations of other cytokines, including IL-10, IL-17A (eBioscience, Waltham, Massachusetts; Jianglai Biotech, Shanghai, China); sCD14, TGF-β (MultiSciences, Hangzhou, China); IL-6, IL-12 (Jianglai Biotech, Shanghai, China), and IL-1β (OmnimAbs, Alhambra, California) in human serum, mouse serum, or culture supernatant were assayed using ELISA kits. The plates were read at 450 nm using SpectraMax Microplate Reader (Molecular Devices, Sunnyvale, California). LPS, the gram-negative bacterial endotoxin, was quantified using the EC Endotoxin Test Kit (Bioendo, Xiamen, China) according to the manufacturer’s instructions.

PCR amplification of 16S rRNA gene of Yersinia enterocolitica

For detection of Yersinia enterocolitica, the stool samples were collected from Graves’ patients with or without diarrhea and heathy individuals, and a primer set specific to its 16S rRNA gene was used to amplify a 330 bp product (Table S1) (29). The exclusion criteria of diarrhea were as follows: diarrhea caused by food intolerance and allergy, adverse reaction to a medication, viral infection, parasitic infection, intestinal disease, or gallbladder or stomach surgery (25, 26). The gDNA was extracted from stool samples using the improved CTAB method. PCR reactions were performed in a DNA Thermal Cycler (Biometra TAdvanced, Jena, Germany). Cycling conditions were as follows: denaturation at 94°C for 1 minute, 35 subsequent cycles consisting of heat denaturation at 98°C for 10 seconds, primer annealing at 55°C for 15 seconds, and extension at 72°C for 40 seconds. PCR products were visualized on 1% agarose gels stained with GelRed.

Quantitative real-time PCR

Total RNA was extracted using the TRIzol reagent (Invitrogen, Waltham, Massachusetts) and was reverse-transcribed using the PrimeScript RT Reagent Kit (Takara, Kusatsu, Japan). Additionally, gDNA was extracted from stool samples using the improved CTAB method. Next, quantitative PCR was performed using TB Green Premix Ex Taq (Takara, Kusatsu, Japan) and specific primers (Table S1) (29) on a CFX96 Real-Time PCR System (Bio-Rad, Hercules, California). Relative gene expression was calculated by the 2^-∆∆Ct method using 16S rRNA gene (Univ 337F/Univ 518R) or ACTB gene as an internal control.

Graves’ disease mice model construction and fecal microbiota transplantation

The adenovirus containing TSHR amino acid residues 1–289 (Ad-TSHR289) and control adenovirus (Ad-Control) were purchased from Hanbio Biotechnology Co., Ltd (Shanghai, China). Female specific pathogen-free (SPF) BALB/c mice at age 6–8 weeks were purchased from the SPF Biotechnology Co., Ltd (Beijing, China) and housed in SPF conditions. All experiments were conducted in accordance with the principles and procedures outlined in the Guideline for the Care and Use of Laboratory Animals and have been approved by the Ethics Committee of Shandong Provincial Hospital of Shandong University (NO. 2019–155). After 3 days of adaptive feeding, all mice except 5 mice from the control group were administered a combination of 4 antibiotics: ampicillin, neomycin-sulfate, metronidazole, and vancomycin (Sigma-Aldrich, St. Louis, Missouri) via oral gavage for 5 days (10 mg of each antibiotic per mouse per day), followed by ad libitum administration in drinking water (ampicillin, neomycin, and metronidazole, 1 g/L; vancomycin, 500 mg/L) (35). The antibiotic treatment lasted 2 weeks (35). The weight of the mice was measured during the antibiotic treatment. The depletion of gut flora was confirmed using 16S rRNA gene sequencing. The mice were then transplanted with the intestinal flora of GD patients or healthy individuals. The colonization of gut microbiota in recipient mice was verified using 16S rRNA gene sequencing after fecal microbiota transplantation (FMT). In detail, the fresh fecal samples collected from healthy individuals or GD patients were mixed with sterile normal saline and then homogenized. The homogenates were then passed through a 20 μm pore filter to remove the particulate and fibrous matter. The filtrates (400 mg/mL) were used for transplantation (36). Mice were administered 200 μL of the filtrates via oral gavage once a day for a week, followed by once every 2 days for the remaining 6 weeks. After the first week of FMT, the mice transplanted with different intestinal flora were respectively injected in their quadriceps with 50 uL phosphate-buffered saline (PBS) containing 10⁹ particles of Ad-TSHR289 or control adenovirus twice at 3-week intervals (37, 38). During the experiment, the control group mice were orally gavaged or injected with PBS by following the same procedure for the mice of experimental groups. Blood was harvested 3 weeks after the second immunization from the tail vein for corresponding analysis.

Statistical analysis

Data are presented as the mean ± SD or median (interquartile range [IQR]). The Mann-Whitney U test or unpaired ttest with appropriate correction for multiple comparisons was used to detect significances for continuous outcomes. For categorical variables, the Chi-square test or Fisher exact test were performed. P < 0.05 was considered as statistically significant. False discovery rate–corrected P-value (q-value) < 0.05 was deemed to be significant for some specific analysis. All experiments were repeated independently at least 3 times.

Results

Significant abnormalities in peripheral cytokines and Treg and Th17 cells of GD patients in association with the intestine

In this study, 58 initial patients with GD (GD group) and 63 healthy individuals (control group) were recruited. The demographic characteristics of all the subjects are presented in Table 1. Compared with healthy individuals, the percentage of Treg (CD4+CD25+FOXP3+) cells in CD4+ T cells was significantly lower, while the percentage of Th17 (CD4+IL17+) cells in CD4+ T cells was markedly higher, resulting in a significantly decreased ratio of Treg/Th17 in the peripheral blood of GD patients (Fig. 1A–1C). CCR9 is one of the most important intestinal homing markers. CCR9-positive cells indicate that they are derived from the intestine or are closely related to the intestine (39, 40). We found that the proportion of CCR9+ Treg cells in CD4+ T cells or in Treg cells was significantly decreased (Fig. 1D and 1E), while the proportion of CCR9+ Th17 cells in CD4+ T cells or in Th17 cells was significantly increased in the peripheral blood of GD patients (Fig. 1F and I G). Furthermore, compared to healthy individuals, the levels of serum anti-inflammatory cytokines (TGF-β and IL-10) were markedly decreased, while the levels of serum proinflammatory cytokines, including sCD14, sCD25, IL-2, IL-17A, IL-1β, IL-6, and IL-12, were significantly increased in GD patients (Fig. 1H–1P). Additionally, the serum IL-18 level, which plays a pivotal role in intestinal inflammation, was also significantly elevated in GD patients (Fig. 1Q). All these results suggested that the immune and inflammatory state of the GD patients were significantly changed, and these changes may be associated with the intestine.

Table 1.

Demographic and clinical characteristics of GD patients and healthy individuals.

	Control	GD	P-value
Number	63	58	–
Sex (M/F)	28/35	23/35	ns
Age (Y)	43.86 ± 9.20	42.07 ± 10.22	ns
BMI (kg/m2)	22.49 ± 2.25	22.48 ± 2.48	ns
FT3 (pml/L)	4.85 (0.58)	16.94 (21.71)	****
FT4 (pmol/L)	15.83 (2.18)	48.89 (54.23)	****
TSH (μIU/mL)	1.92 (1.29)	0.005 (0.005)	****
TRAb (IU/L)	0.54 (0.47)	11.98 (22.94)	****
TGAb (IU/mL)	23.20 (9.50)	67.70 (379.37)	***
TPOAb (IU/mL)	33.10 (8.60)	206.75 (1253.15)	****

Data are shown as the mean ± SD and median (interquartile range). The Chi-squared test (sex), ttest (age and BMI), and Mann-Whitney U test (FT3, FT4, TSH, TRAb, TGAb, and TPOAb) were used to detect significant changes; control = healthy individuals; *** p < 0.001, **** p < 0.0001.

Abbreviations: BMI, body mass index; F, female; FT3, free T3; FT4, free T4; GD, Graves’ disease patients; M, male; ns, not significant; TGAb, thyroglobulin antibody; TPOAb, thyroperoxidase antibody; TRAb, thyrotropin receptor antibody; TSH, thyrotropin; Y, year.

Figure 1.

Abnormalities of cytokines, Treg, and Th17 cells in the peripheral blood of GD patients. A, B: The percentages of Treg (CD4+CD25+FOXP3+) cells (A) and Th17 (CD4+IL17+) cells (B) in CD4+ T cells of healthy individuals (control) and GD patients (GD). C: The ratio of Treg/Th17 cells in the peripheral blood of control and GD groups. D, E: The percentages of CCR9+ Treg cells (CCR9+CD4+CD25+FOXP3+) in CD4+ T cells (D) or in Treg cells (CD4+CD25+FOXP3+) (E) in the peripheral blood of control and GD groups. F, G: The percentages of CCR9+ Th17 cells (CCR9+CD4+IL17+) in CD4+ T cells (F) or in Th17 cells (CD4+IL17+) (G) in the peripheral blood of control and GD groups. H–Q: The serum levels of anti-inflammatory cytokines TGF-β (H), IL-10 (I), and proinflammatory cytokines sCD14 (J), sCD25 (K), IL-2 (L), IL-17A (M), IL-1β (N), IL-6 (O), IL-12 (P), and IL-18 (Q) in control and GD groups were assayed using Luminex and ELISA. Data are presented as the mean ± SD or median (IQR). The ttest (I, K, N) and Mann-Whitney U test (A–H, J, L–Q) were used to detect significant differences. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001. Abbreviations: GD, Graves’ disease; IQR, interquartile range, SD, standard deviation.

Gut microbiota of GD patients markedly differed from those of healthy individuals

To clarify the roles of gut microbiota in the pathogenesis of GD, fecal samples were analyzed using 16S rRNA gene sequencing. The good’s coverage index of each sample, which was used to evaluate the sequencing depth, was over 0.99. This result indicated that the sequencing depth had been saturated. Moreover, the sequencing depth was also not significantly different in either the GD and control groups (Fig. 2A). The parameters of richness (Ace and Chao1 indexes) and diversity (Shannon and Simpson indexes) of the intestinal flora were all significantly lower in GD patients than in the healthy controls, indicating a markedly decreased α diversity (Fig. 2B). Moreover, the Pielou and Simpson indexes were significantly negatively correlated with the serum levels of TRAb, FT3, and FT4 and positively correlated with the serum TSH level, indicating a significant association between the intensity of the disease and the intestinal flora (Fig. 2C). The Bray-Curtis distance-based community analysis, which is an indicator that characterizes the similarity of microbial composition between individuals, showed a noticeable separation between the samples of healthy individuals and GD patients, revealing that the microbiota composition differed significantly between the 2 groups (Fig. 2D). At the phylum level, Firmicutes were less abundant, while Bacteroidetes were more abundant (Fig. 2E), which was further confirmed with quantitative PCR (qPCR) (Fig. 2F); thereby, the ratio of Firmicutes/Bacteroidetes decreased significantly in GD patients (Fig. 2G). Meanwhile, the abundance of many other bacterial phyla, including Proteobacteria, Saccharibacteria, and Verrucomicrobia in GD patients, were also markedly changed compared with those in the healthy individuals (Fig. 2E). At the genus level of the intestinal bacteria (abundance > 0.2%), the abundance of 33 bacterial genera were significantly decreased, while 7 bacterial genera were markedly increased in GD patients, and they all belonged to Firmicutes, Proteobacteria, and Bacteroidetes (Fig. 2H). LEfSe analysis demonstrated that the genus Bacteroides (g_Bacteroides), enriched in healthy individuals, was one of the top 4 biomarkers. However, the phylum Bacteroidetes (p_Bacteroidetes), order Bacteroidales (o_Bacteroidales), and class Bacteroidia (c_Bacteroidia) were enriched in GD patients (Fig. 2I and 2J). The random forest analysis also showed that 3 intestinal bacteria (Bacteroides, Alistipes, Prevotella) could distinguish GD patients from healthy individuals with 85% accuracy (Fig. 3A and 3B). These findings indicated that Bacteroides might play an important role in the occurrence of GD. Additionally, Yersinia enterocolitica is an enteropathogenic bacterium and is frequently reported in GD patients. However, it was not detected in our 16S rRNA gene sequencing data. To explain this result, qPCR was performed on fecal samples from different individuals and showed that the detection rate of Yersinia enterocolitica was significantly higher in GD patients with diarrhea than in GD patients without diarrhea and healthy individuals (Fig. 4A and 4B).

Figure 2.

Significant changes in the gut microbiota composition of GD patients. A: Average number of sequencing reads in healthy individuals (control) and GD patients (GD). B: The α diversity indices (Ace, Chao1, Shannon, and Simpson indexes) of the intestinal flora in healthy individuals (control) and GD patients (GD). C: Heatmap of the correlations between major clinical indicators of GD and the Pielou and Simpson indexes of intestinal flora. The color bar with numbers indicates the correlation coefficients. The FDR-adjusted P-values by ttest were shown by asterisks (*** corrected P < 0.001). D: NMDS plot based on Bray-Curtis distance matrix of control and GD groups. Each point represents a sample. The ellipses do not represent statistical significance, but rather serve as a visual guide to group differences. E: The relative abundances of intestinal bacteria at the phylum level in the control and GD groups. F: The fold changes in the abundances of Firmicutes and Bacteroidetes in intestinal flora of GD group compared with that of control group. They were determined using qPCR. G: The ratio of Firmicutes/Bacteroidetes in GD patients and healthy individuals was calculated using the 16S rRNA gene sequences data. H: Heatmap of the relative abundances of intestinal bacteria with significant differences (corrected P < 0.05 by Mann-Whitney U test) at the genus level in control and GD groups. The color bar indicates z score, which represents the relative abundance. Z score < 0 (>0) means the relative abundance was lower (higher) than the mean. The color bar on the left represents the phyla of intestinal bacteria: pink, Bacteroidetes; green, Firmicutes; brick red, Proteobacteria. I: Cladogram generated by LEfSe analysis. The significantly differential bacterial clades or taxa are highlighted. Red: increased abundance in GD group. Green: increased abundance in control group. J: Linear discriminant analysis (LDA) scores of the intestinal bacteria in control and GD groups at different taxonomic levels generated using LEfSe analysis. LDA score > 4 or < -4 means bacterial taxa are significantly enriched in control group (green) or GD group (red) (P < 0.05). Data are presented as the median (IQR). The ttest (A) and Mann-Whitney U test (B, E, F, G, I, J) were used to detect significant changes. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001. Abbreviations: c, class; f, family; FDR, false discovery rate; FT3, free triiodothyronine; FT4, free thyroxine; g, genus; GD, Graves’ disease; IQR, interquartile range; LDA, linear discriminant analysis; NMDS, nonmetric multidimensional scaling; ns, not significant; o, order; p, phylum; s, species; TRAb, thyrotropin receptor antibody; TSH, thyrotropin.

Figure 3.

Random forest analysis. A: Results of the Random Forests analysis. The bacterial genera that could significantly discriminate between GD patients (GD) and healthy individuals (control) were presented in descending order. Three bacterial genera that could accurately distinguish GD patients and healthy individuals with 85% accuracy were indicated by rectangle. B: Graphic representation of the classifier accuracy per sample group/cohort. Abbreviation: GD, Graves’ disease.

Figure 4.

Detection rate of Yersinia enterocolitica in healthy individuals and GD patients with and without diarrhea. A: The agarose electrophoresis of polymerase chain reaction (PCR) products of Yersinia enterocolitica amplified in fecal samples from healthy individuals (lanes 1, 2, 3, 4, and 5), GD patients without diarrhea (lanes 6, 7, 8, 9, 10, and 11), and GD patients with diarrhea (lanes 12, 13, 14, 15, 16, 17, 18, and 19). Lane M: DNA marker. B: Comparisons of the detection rate of Yersinia enterocolitica in the intestinal flora of healthy individuals and GD patients with and without diarrhea. Statistical analysis was performed using Fisher’s exact test. The detection rate of Yersinia enterocolitica was significantly higher in GD patients with diarrhea than in healthy individuals and GD patients without diarrhea. Abbreviation: GD, Graves’ disease; PCR, Polymerase chain reaction.

Gut dysbiosis can seriously interfere with the interactions among intestinal bacteria. The correlations between the abundance of bacteria are a good reflection of the interactions among bacteria. Our results demonstrated that the correlations among many intestinal bacteria of GD patients were significantly changed. In healthy individuals, the probiotic Lactobacillus, which is considered to play an important role in physical health and is considered to be involved in autoimmune diseases (41, 42), was positively correlated with SCFA-producing bacteria (such as Desulfovibrio and Veillonella), but these correlations were weakened or even disappeared in GD patients. Additionally, the probiotic Bacteroides, which could produce SCFAs, was negatively correlated with the pathogenic bacteria Prevotella_2 in healthy individuals, while this relationship disappeared in GD patients. Another genus producing propionic acid, Dialister, was negatively correlated with Streptococcus in healthy individuals, but this correlation became positive in GD patients (Fig. 5A). These changes of correlations among the microflora might be closely related to the development of GD, but more research is still needed to reveal their significance and mechanism.

Figure 5.

Correlation network of gut microbiota and correlation analysis of clinical indicators and intestinal bacteria, as well as serum LPS level. A: Correlation network of gut microbiota in healthy individuals (control) and GD patients (GD). The significant strong correlations (|r| > 0.3 and corrected P < 0.05 by ttest) among the intestinal bacterial genera are presented in the network. The red and green edges represent the positive correlations and negative correlations, respectively. The spot colors represent different bacterial phyla. The thickness of the edges represents the strength of the correlation. The dotted rectangles indicate that the bacteria are emphasized in the Results section. B: Heatmap of correlations between six clinical indicators of GD and the abundances of intestinal bacterial phyla. The bacterial phyla emphasized in text are marked with rectangles. C: Heatmap of the correlations between 6 clinical indicators of GD and the top 40 most abundant bacterial genera whose abundance significantly changed in GD patients. In (B) and (C), the color bar with numbers indicates the correlation coefficients, and the FDR-adjusted P-values using ttest are shown by asterisks (* corrected P < 0.05; ** corrected P < 0.01; *** corrected P < 0.001). D: Comparisons of the relative abundances of the top 10 most abundant bacterial genera, which were significantly correlated with the clinical indicators of GD. E: Serum levels of LPS in GD patients (GD) and healthy individuals (control) measured using the LAL test. Data are presented as the median (IQR). The Mann-Whitney U test was used to detect significant changes for (D) and (E). * P < 0.05; ** P < 0.01; *** P < 0.001; **** P < 0.0001. Abbreviations: FT3, free triiodothyronine; FT4, free thyroxine; GD, Graves’ disease; IQR, interquartile range; TGAb, thyroglobulin antibody; TPOAb, thyroperoxidase antibody; TRAb, thyrotropin receptor antibody; TSH: thyrotropin.

To clarify the pathological significance of these significantly altered intestinal bacteria in GD patients, the correlations between their relative abundances and the clinical features of GD (serum levels of FT3, FT4, TPOAb, TGAb, TRAb, and TSH) were analyzed. The results showed that Proteobacteria, Tenericutes, Lentisphaerae, Verrucomicrobia, and Synergistetes were negatively correlated with the serum levels of FT3, FT4, TPOAb, and TRAb, while positively correlated with the serum TSH level. Hydrogenedentes, Bacteroidetes, Saccharibacteria, and Spirochaetae demonstrated the exact opposite of the above results (Fig. 5B). At the genus level, intestinal bacteria, except for Streptococcus, Alloprevotella, Veillonella, Neisseria, Acinetobacter, Prevotella, and Prevotella_7, were significantly negatively correlated with the serum levels of TPOAb, TRAb, FT3, and FT4, but positively correlated with the serum TSH level (Fig. 5C). Among the altered bacteria with significant association with the clinical indicators of GD, the abundance of Bacteroides was considerably higher than that of other genera (Fig. 5D). Additionally, the serum level of LPS was significantly increased in GD patients (Fig. 5E), implying that intestinal dysbiosis in GD patients might result in intestinal barrier disruption.

Gut metabolites, including SCFAs, were significantly changed in GD patients

Gut metabolites are important bridges for intestinal flora to regulate the host immune system. Therefore, metabolomic analysis was performed to explore the intestinal metabolic profiles of GD patients. A total of 540 and 578 peak features were identified, respectively, in positive ion mode (ES+) and negative ion mode (ES-). These peak features were clustered by orthogonal, partial, least squares discriminant analysis (OPLS-DA), showing that the samples of the control and GD groups were clearly separated from each other (Fig. 6A and 6B). A total of 292 (112 upregulated peaks and 180 downregulated peaks) and 225 (134 upregulated peaks and 91 downregulated peaks) peak features were significantly changed in the ES+ and ES- modes, respectively (Fig. 6C). These findings indicated that the intestinal metabolic profiles in GD patients changed markedly. The significantly changed peak features were analyzed using MS/MS, and many metabolites were identified by the combination of precise molecular weight and structural information from the compound structure database. These metabolites included amino acid derivatives, lipids, fatty acids, and bile acids, and many of them were significantly changed in GD patients (Fig. 6D and 6E).

Figure 6.

Noticeable changes in the metabolic profiles of gut microbiota in GD patients. A: The plot of OPLS-DA score of all peak features from the untargeted metabolomics analysis of stool samples of healthy individuals (control) and GD patients (GD). B: Validation of the orthogonal partial least-squares-discriminant analysis (OPLS-DA) model via permutation test (times = 200). The criterion for model validity was that the regression line of the Q2-points (blue dotted line) intersected the vertical solid line (on the left) below zero. C: Cloud plots of the peak features of intestinal metabolites which significantly changed in GD patients in positive ion mode (ES+, upper panel) and negative ion mode (ES-, under panel). Red (blue) circles indicate the significantly increased (decreased) metabolites (fold change > 1.5, P < 0.05) in the stool samples of GD patients. The color tone indicates P-values: the darker the color tone, the larger the P-value. The radius of circle indicates the fold changes of corresponding peak features. D: Heatmap of the relative abundance of the top 50 most abundant metabolites which significantly changed in GD group (corrected P < 0.05 by t-test). The color bar indicates z score, which represents the relative abundance. Z score < 0 (>0) meant the relative abundance was lower (higher) than the mean. E: Comparisons of the relative amounts of some representative intestinal metabolites in GD and control groups. F: The significant differences in metagenomics functions in GD patients (GD) compared to that in healthy individuals (control) (corrected P < 0.05 and confidence intervals = 95%). This analysis was performed using PICRUSt software. G: The concentrations of propionic acid, butyric acid, and acetic acid in fecal samples of control and GD groups were determined by gas chromatography–mass spectrometry (GC–MS). H: The concentrations of isobutyric acid, isovaleric acid, and valeric acid in the fecal samples from healthy individuals (control) and GD patients (GD). Data are presented as the mean ± SD or median (IQR). The ttest (C, E, F) and Mann-Whitney U test (G, H) were used to detect significant differences. * P < 0.05; ** P < 0.01; *** P < 0.001; **** P < 0.0001. Abbreviations: GC-MS, gas chromatography–mass spectrometry; GD, Graves’ disease; IQR, interquartile range; ns, not significant; SD, standard deviation; OPLS-DA, orthogonal partial least-squares-discriminant analysis.

The results of PICRUSt showed that the predicted relative abundances of butanoate and propanoate metabolism were significantly lower in GD patients (Fig. 6F). Therefore, the volatile SCFAs in the intestine were detected by gas chromatography–mass spectrometry (GC–MS), and the results showed that propionic acid and butyric acid, 2 important immunomodulatory SCFAs, were significantly decreased in GD patients compared with those in healthy individuals. Moreover, the reduction of propionic acid was the most significant in these SCFAs (Fig. 6G). The intestinal levels of other SCFAs, including acetic, isobutyric, valeric acid, and isovaleric acids in GD, were decreased without statistical significance (Fig. 6G and 6H).

Close correlations among clinical features of GD, SCFAs, and SCFA-producing bacteria

To explore the significance of the perturbations in intestinal flora and metabolites, 52 differential metabolites with the most statistical significance were selected, and the Spearman correlation coefficients between them (a part of them) and the clinical features of GD (the significantly changed bacteria) were calculated. Many altered metabolites were significantly correlated with the changed intestinal flora (|r| > 0.5, corrected P < 0.05). Propionate and butyrate were significantly positively correlated with many SCFA-producing bacteria, including Roseburia, Bacteroides, Phascolarctobacterium, Alistipes, Lachnospiraceae NK4A136, Christensenellaceae R-7, Ruminococcaceae_UCG-005, Ruminococcaceae_UCG-002, Turicibacter, and Ruminococcus (Fig. 7A). In addition, many remarkably changed intestinal metabolites were significantly correlated with the clinical indicators of GD. Propionate, which significantly decreased in the gut of GD patients, was strongly negatively correlated with the serum levels of TPOAb, FT3, FT4, and TRAb, while strongly positively correlated with the serum level of TSH, which suggested that propionate was closely related to the condition of patients with GD. Moreover, dihydroxyacetone, representing high metabolic rate, was significantly positively correlated with the serum levels of FT3 and TRAb (Fig. 7B).

Figure 7.

Correlations among intestinal bacteria, metabolites, and clinical indicators of GD, as well as the changes in short-chain fatty acids (SCFA)-producing genera and key enzymes for producing butyric acid and propionic acid. A: Heatmap of correlations between the significantly changed bacterial genera and 8 metabolites with important functions and significant differences. The color bar with numbers indicates the correlation coefficients. The important bacterial genera are marked with rectangles. B: Heatmap of correlations between 6 clinical indicators of GD and the remarkably changed intestinal metabolites in GD patients. The color bar with numbers indicates the correlation coefficients. The important metabolites are marked with rectangles. C: The abundance changes in the key enzymes for producing butyric acid and propionic acid in the intestinal flora of GD patients (GD) and healthy individuals (control). ButA, Butyryl-CoA transferase; LcdA, lactoyl-CoA dehydratase; PduP, propionaldehyde dehydrogenase; MmdA, methylmalonyl-CoA decarboxylase. D: Heatmap of correlations between 6 clinical indicators of GD and the abundance of the 9 representative SCFA-producing bacteria, which significantly changed in GD patients. The color bar with numbers indicates the correlation coefficients. E: The abundance changes in the 9 representative SCFA-producing bacteria in intestinal flora of GD patients (GD) and healthy individuals (control). They were determined using quantitative polymerase chain reaction (qPCR). Data are presented as the median (IQR). The false discovery rate (FDR)-adjusted P-values by ttest are shown by asterisks for (A, B, D). * corrected P < 0.05; ** corrected P < 0.01; *** corrected P < 0.001 in (A, B, D). The Mann-Whitney U test was used to detect significant differences for (C, E). ** P < 0.01; *** P < 0.001. Abbreviations: FDR, false discovery rate; GD, Graves’ disease; IQR, interquartile range; ns, not significant; qPCR, quantitative polymerase chain reaction; SCFA, short-chain fatty acids.

ButA (butyryl-CoA transferase), LcdA (lactoyl-CoA dehydratase), PduP (propionaldehyde dehydrogenase), and MmdA (methylmalonyl-CoA decarboxylase) are the key enzymes to produce butyric acid and propionic acid in gut flora. In the intestinal flora of GD patients, the abundance of these key enzyme-encoding genes were all significantly decreased, confirming the significant reduction in their SCFA-producing ability and resulting in the lack of SCFAs in GD patients (Fig. 7C). The qPCR using the primers specific for the representative bacteria of SCFA-producing genera showed that the abundance of Bacteroides fragilis species was the most closely related to the levels of TRAb, FT3, FT4, and TSH and decreased with the most statistical significance in the intestine of GD patients (Fig. 7D and 7E).

Bacteroides fragilis modulates Treg/Th17 balance through the pathway regulated by propionic acid

To elucidate the role and mechanism of B. fragilis in immune dysfunction, such as Treg/Th17 imbalance in GD patients, the strains of B. fragilis were isolated from human feces using a selective culture medium (M2GSC) in anaerobic conditions and confirmed as B. fragilis YCH46 strain with 100% identity using Sanger sequencing. qPCR also showed that the abundance of B. fragilis YCH46 strain in GD patients was significantly reduced compared to that in healthy controls (Fig. 8A). Compared to M2GSC medium, the supernatant of M2GSC medium culturing B. fragilis YCH46 (B.f.S) could significantly increase the percentage of CD4+CD25+FOXP3+ Treg cells and IL-10 level, while reducing the percentage of CD4+IL17+ Th17 cells and IL-17A level in PBMCs from healthy individuals (Fig. 8B–8D). Moreover, B.f.S could improve the imbalance of Treg/Th17 in patients with GD (Fig. 8E–8G). Propionic acid is well known as the important metabolite produced by B. fragilis. Our results showed that the expression of FFAR2 (a propionic acid receptor) mRNA was markedly upregulated, while the expression of histone deacetylase 6 and histone deacetylase 9 mRNA, which are the downstream molecules of FFAR2 and could suppress Treg cells, were downregulated in PBMCs from both healthy individuals and GD patients treated with B.f.S (Fig. 8H and 8I). These results indicated that B. fragilis YCH46 could influence the differentiation and/or proliferation of Treg and Th17 cells through the pathway regulated by propionic acid, and then modulate the Treg/Th17 balance.

Figure 8.

Treg/Th17 balance is modulated by Bacteroides fragilis through the pathway regulated by propionic acid. A: Changes in the abundance of Bacteroides fragilis YCH46 strain in the intestinal flora of the control and GD groups was determined using qPCR. B, C: Peripheral blood mononuclear cells (PBMCs) from healthy individuals were treated with M2GSC medium (M2GSC) or the supernatant of M2GSC medium culturing B. fragilis YCH46 (B.f.S) for 72 hours. Then, the percentage of Treg (CD4+CD25+FOXP3+) and Th17 cells (CD4+IL17+) in CD4+ T cells were analyzed using flow cytometry. D: The concentrations of IL-10 and IL-17A in the culture medium of healthy individuals’ PBMCs treated with M2GSC or B.f.S. E, F: PBMCs from GD patients were treated as described above. The percentage of Treg (CD4+CD25+FOXP3+) and Th17 cells (CD4+IL17+) in CD4+ T cells were analyzed by flow cytometry. G: The concentrations of IL-10 and IL-17A in the culture medium of GD patients’ PBMCs treated with M2GSC or B.f.S. H, I: The fold changes in mRNA expression of FFAR2, HDAC6, and HDAC9 in PBMCs from healthy individuals (H) or GD patients (I) treated with B.f.S compared to that treated with M2GSC. The experiment was repeated at least 3 times independently. Data are presented as the median (IQR). The Mann-Whitney U test was used to detect significant differences. * P < 0.05; ** P < 0.01; *** P < 0.001; **** P < 0.0001. Abbreviations: B.f.S, the supernatant of M2GSC medium culturing B. fragilis YCH46; GD, Graves’ disease; IQR, interquartile range; PBMCs, peripheral blood mononuclear cells.

Intestinal flora contributed to the development of GD, revealed by fecal microbiota transplantation in mouse model

To clarify the causal relationship between gut dysbiosis and GD pathogenesis, FMT was performed using a mouse model of GD. The experimental procedure was shown in Fig. 9A. The results showed that the broad-spectrum antibiotic treatment effectively removed most of the gut bacteria in mice, while did not significantly affect mice body weight (Fig. S2A, B) (29). The principal coordinates analysis (PCoA) results showed that the sample points representing the intestinal flora from the donor individuals and the corresponding recipient mice were almost completely overlapped, indicating that the bacteria colonization was successful after FMT (Fig. S2C, D) (29). The mouse model was constructed by transfecting the adenovirus containing TSHR amino acid residues 1–289 (Ad-TSHR289) (37, 43). The results showed that transfection of the blank adenovirus vector (Ad-control) had no significant effect on the thyroid function (serum total T4 level) and immunoinflammation (serum levels of TRAb, IL-17A, and IL-10) in the mice. On the contrary, transfecting Ad-TSHR289 significantly increased the serum levels of TT4, TRAb, and IL-17A, and decreased serum IL-10 level in the mice (Fig. 9B–9E). These results indicated that the GD mouse model was successfully established. Moreover, FMT showed that the transplantation of fecal microbiota from healthy individuals or GD patients had similar effects on thyroid function and immunoinflammation of the mice transfected with Ad-control (Fig. 9F–9I). However, in mice transfected with Ad-TSHR289, the transplantation of intestinal flora of GD patients significantly elevated the serum levels of TT4, TRAb, and IL-17A, decreased the serum IL-10 level compared with transplanting the intestinal flora of healthy individuals (Fig. 9J–9M), and increased the GD incidence from 28.6% to 73.3% (Fig. 9N). These results indicate that abnormal intestinal flora alone may not be sufficient to cause thyroid dysfunction and GD development. However, it can significantly promote the occurrence and development of GD together with other pathogenic factors. Thus, intestinal flora abnormality is one of the important pathogenic factors of GD, not just its consequence or accompanying phenomenon.

Figure 9.

Fecal microbiota transplantation (FMT) in the mouse model of GD. A: The schematic diagram of the experimental process including FMT and immunization. B–M: Serum levels of total T4 (TT4; B, F, J), TRAb (C, G, K), IL-17A (D, H, L), and IL-10 (E, I, M) were measured at 3 weeks after the second immunization. N: Comparison of the incidences of thyroid dysfunction in different groups. The levels of serum TT4 and TRAb more than the values, namely their corresponding mean + 3SD of the mice immunized with Ad-Control (control group) were considered as abnormal and as GD development. In (B–M), the mice in the control group were treated as described in the Materials and Methods section. The mice transfected with Ad-Control were Ad-Control group (Ad-Control). The mice transfected with Ad-TSHR289 were Ad-TSHR289 group (Ad-TSHR289). Transplanting the fecal microbiota from healthy individuals (GD patients) to the mice transfected with Ad-Control was FMT-C+Ad-Control (FMT-GD+Ad-Control) group. Transplanting the fecal microbiota from healthy individuals (GD patients) to the mice transfected with Ad-TSHR289 was FMT-C+Ad-TSHR289 (FMT-GD+Ad-TSHR289) group. Data are presented as the median (IQR). Kruskal-Wallis test with Steel Dwass in (B–M) and Fisher’s exact test in (N) were used to detect significant changes. * corrected P < 0.05; ** corrected P < 0.01; *** corrected P < 0.001; **** corrected P < 0.0001. Abbreviations: Ad-Control, control adenovirus; Ad-TSHR289, adenovirus containing TSHR amino acid residues 1–289; FMT-C, transplanting fecal microbiota from healthy individuals; FMT-GD, transplanting fecal microbiota from GD patients; GD, Graves’ disease; GD+, mice with GD; GD-, mice without GD; IQR, interquartile range; ns, not significant.

Discussion

In this study, we revealed for the first time that intestinal flora abnormality is one of the important pathogenic factors in the development and in immune abnormalities of GD. The composition, metabolism, and inter-relationships of intestinal flora in GD patients were considerably reshaped, with significant relations to the clinical features of GD, including the decreased abundance of SCFA-producing bacteria and SCFAs. Intestinal B. fragilis could modulate the balance between Treg and Th17 cells through its metabolite propionic acid. The reduction in B. fragilis contributed to the imbalance of Th17 and Treg cells, thereby promoting the pathogenesis of GD.

Until now, many studies have demonstrated the complex interactions between the gut flora and the host (44). Recently, the concept of the gut–thyroid axis has been proposed (45). However, the regulatory mechanism between the gut and thyroid has still not been fully elucidated, and the gut bacteria involved in the development of GD have rarely been reported. A recent descriptive study reported changes in the composition of intestinal flora in GD patients but did not explore the mechanism and causal relationship between these changes and GD (24). Another study also reported the association between the serum TRAb level and the intestinal microbiota in patients with Graves’ orbitopathy (GO) (46). Our FMT results, for the first time, proved that although abnormal intestinal flora alone may not be sufficient to cause thyroid dysfunction and GD development, it could markedly promote the occurrence and development of GD together with other pathogenic factors. Thus, intestinal flora abnormality is one of the important mechanisms for GD pathogenesis. With regard to the mechanism through which the gut regulates thyroid function, most previous studies were focused on the molecular mimicry of bacteria (22). Our results imply that in addition to molecular mimicry, the immune regulation of intestinal flora is a novel mechanism through which the gut regulates thyroid function, namely how the gut–thyroid axis works. Additionally, although Yersinia enterocolitica is considered to be involved in the development of GD through different mechanisms, including molecular mimicry, we did not detect it using 16S rRNA gene sequencing. The main reason may be that in this study, we excluded GD patients with diarrhea (similar to other studies on gut microbiota using 16S rRNA gene sequencing), because diarrhea can markedly change the composition of the intestinal flora. In fact, we found that Yersinia enterocolitica was easy to detect only in GD patients with diarrhea. Another reason may be the differences in the assay methods; serological assay was used often in previous studies in which Yersinia enterocolitica was detected in GD patients, while 16S rRNA gene sequencing was used in this study. Yersinia enterocolitica was also not detected in recently published studies, which analyzed the intestinal flora of GD and GO patients using 16S rRNA gene sequencing (24, 46).

The decrease in Treg cells and increase in Th17 cells in the peripheral blood of GD patients have been reported previously (47). However, the mechanism of these abnormalities has not been fully elucidated. CCR9 is one of the most important gut homing markers and CCR9+ cells are likely to be derived from, or closely related to, the intestine (39, 40). In this study, we found that a high proportion of the peripheral Treg and Th17 cells were CCR9 positive. Moreover, the changes in CCR9+Treg and CCR9+Th17 cells were almost the same as the changes in total Treg and Th17 cells in the peripheral blood of GD patients. This implied that the abnormalities of peripheral Treg and Th17 cells in GD patients were mainly caused by the cells derived from the intestine and were closely related to the intestine and intestinal flora. This also suggested that in GD patients, the gut flora might regulate the Treg and Th17 cells mainly in the intestine, and then the regulated cells enter the circulation and play immunomodulatory roles outside the gut.

B. fragilis is one of the most frequent species at the human intestinal mucosal surface, and plays pivotal roles in the development of the host immune system (48). We found that its abundance was significantly decreased in the gut of GD patients. Moreover, the supernatant of the medium culturing the YCH46 strain of B. fragilis (B.f.S) containing a high concentration of propionic acid could increase the ratio of Treg/Th17 in PBMCs of both healthy individuals and GD patients. This indicated that B. fragilis could produce propionic acid, which might be the important mechanism of its immunomodulatory effects (49). Some studies have reported that propionic acid could regulate immune cell activation, proliferation, and differentiation through G-protein coupled receptors (such as GPR41 and GPR43) and histone deacetylase (HDAC) (15). We found that the expression of GPR43 (FFAR2) mRNA was significantly upregulated, while the expression of HDAC6 and HDAC9 mRNAs was significantly downregulated in PBMCs of both healthy individuals and GD patients treated with B.f.S. This suggested that propionic acid produced by B. fragilis might inhibit HDACs through GPR43 (FFAR2) and regulate Treg and Th17 cells. In addition to directly regulating the inflammatory and immune responses, SCFAs, including propionic acid, can also affect gut barrier integrity (50). Our study showed that the plasma level of bacterial LPS was markedly increased in GD patients, implying gut barrier dysfunction, which is closely related to the immune abnormalities of GD patients (51). Its mechanism may be that the reduction in SCFA-producing bacteria leads to a decrease in SCFAs, which impairs gut barrier integrity.

Due to the lack of effective early warning indicators, most GD patients are diagnosed and treated after noticeable clinical symptoms appeared (52). At this moment, the best time window for intervention has been missed and it is difficult to eliminate the immune abnormalities that cause GD. Our study identified a few intestinal bacteria that significantly changed in the gut of GD patients. The result of random forest showed that a group of intestinal bacteria (Bacteroides, Alistipes, Prevotella) could accurately differentiate GD patients from healthy individuals with 85% accuracy. Since gut dysbacteriosis is one of the important causes of GD, the abnormalities of intestinal flora are likely to occur before the clinical symptoms and/or classic indicators of GD appear. Therefore, these 3 intestinal bacteria have potential as markers for diagnosis of GD.

Recurrence is a major problem for GD therapy and 40%–60% of GD patients are reported to relapse after treatment (53). Its main cause might be that the factors causing immune dysfunction have not been effectively removed and the immune function has not been restored. Intestinal flora is a key regulator of immune function and can be used as a therapeutic target for improving immune function in GD patients. In recent years, the efficacy of microecological therapy, which targets the intestinal flora and includes fecal microbiota transplantation, probiotic, or prebiotic treatments, is increasingly being recognized in the treatment of chronic metabolic diseases and autoimmune diseases (54). According to our findings, YCH46 strain of B. fragilis was a natural inhibitor of Th17 cells and activator of Treg cells, and could ameliorate the immune dysfunction in GD patients in vitro, indicating it could be used as an immunomodulator and an auxiliary treatment for GD patients to reduce the recurrence rate. The therapeutic potential of B. fragilis YCH46 warrants a comprehensive study to assess its efficacy and safety.

Conclusions

In summary, our study revealed the role and mechanism of gut microbiota in the immune dysfunction of GD (Fig. 10) and identified 3 intestinal bacteria, which have potential as adjunct markers for GD diagnosis. This study also provided valuable insights for improving immune dysfunction in GD patients using B. fragilis and illuminated the prospect of microecological therapy for GD. The combination therapy of microecology and drugs could be a powerful candidate for reducing the recurrence of GD.

Figure 10.

Schematic representation of the mechanism through which gut microbiota modulates Treg/Th17 balance and contributes to the development of GD. The blue downward arrows indicate downregulation/decrease. The red upward arrows indicate upregulation/increase. Abbreviation: Graves’ disease.

Abbreviations

Abbreviations
 
• 16S rRNA

  16S ribosomal RNA

 
• Ad-TSHR289

  adenovirus containing TSHR amino acid residues 1–289

 
• B.f.S

  the supernatant of M2GSC medium culturing B. fragilis YCH46

 
• BMI

  body mass index

 
• ButA

  butyryl-CoA transferase

 
• CTAB

  cetyl trimethylammonium bromide

 
• ELISA

  enzyme-linked immunosorbent assay

 
• FDR

  false discovery rate

 
• FFAR2

  free fatty acid receptor 2

 
• FMT

  fecal microbiota transplantation

 
• FT3

  free triiodothyronine

 
• FT4

  free thyroxine

 
• GC–MS

  gas chromatography–mass spectrometry

 
• GD

  Graves’disease

 
• GO

  Graves’ophthalmopathy

 
• HDAC

  histone deacetylase

 
• LcdA

  lactoyl-CoA dehydratase

 
• LEfSe

  the linear discriminant analysis effect size

 
• LPS

  lipopolysaccharides

 
• MmdA

  methylmalonyl-CoA decarboxylase

 
• OPLS-DA

  orthogonal partial least-squares-discriminant analysis

 
• OUTs

  operational taxonomy units

 
• PBMCs

  peripheral blood mononuclear cells

 
• PBS

  phosphate buffered saline

 
• PduP

  propionaldehyde dehydrogenase

 
• PICRUSt

  Phylogenetic Investigation of Communities by Reconstruction of Unobserved States

 
• SCFAs

  short-chain fatty acids

 
• TGAb

  thyroglobulin antibody

 
• Th17

  T helper 17 cell

 
• TPOAb

  thyroperoxidase antibody

 
• TRAb

  thyrotropin receptor antibody

 
• Treg

  regulatory T cell

 
• TSAb

  thyroid stimulating antibody

 
• TSH

  thyrotropin

 
• TSHR

  TSH receptor

 
• TT4

  total thyroxine

Acknowledgments

We would like to thank Xinjie Zhang for her help in the experiments.

Financial Support: This work was supported by the National Natural Science Foundation of China (NO. 81570712 and 81670247), the Natural Science Outstanding Youth Foundation of Shandong Province (NO. JQ201519), Major Science and Technology Innovation Project of Shandong Province (NO. 2018CXGC1218), Jinan Clinical Medical Science and Technology Innovation Program (NO. 201805055, NO. 201704105), and Taishan Scholar Project of Shandong Province (NO. ts201712092).

Additional Information

Disclosure Summary: The authors declare that there are no competing interests associated with the manuscript.

Data Availability

The datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.