An Increased Abundance of Clostridiaceae Characterizes Arthritis in Inflammatory Bowel Disease and Rheumatoid Arthritis: A Cross-sectional Study

Abstract

Background

Inflammatory bowel diseases (IBDs) are a group of heterogeneous inflammatory conditions affecting the gastrointestinal tract. Although there is considerable evidence linking the gut microbiota to intestinal inflammation, there is limited knowledge on its potential role in the development of extraintestinal manifestations of IBD.

Methods

Four groups of patients were included: IBD-associated arthropathy (IBD-A); IBD without arthropathy (IBD-N); rheumatoid arthritis (RA); and non-IBD, nonarthritis controls. DNA from stool samples was isolated and sequenced using the Illumina platform. Paired-end reads were quality-controlled using SHI7 and processed with SHOGUN. Abundance and diversity analyses were performed using QIIME, and compositional biomarker identification was performed using LEfSe.

Results

One hundred eighty patients were included in the analysis. IBD-A was associated with an increased abundance of microbial tyrosine degradation pathways when compared with IBD-N (P = 0.02), whereas IBD-A and RA patients both shared an increased abundance of Clostridiaceae when compared with controls (P = 0.045). We found that history of bowel surgery was a significant source of variability (P = 0.001) among all IBD patients and was associated with decreased alpha diversity and increased abundance of Enterobacteriaceae (P = 0.004).

Conclusions

An increased abundance of gut microbial tyrosine degradation pathways was associated with IBD-A. An increased abundance of Clostridiaceae was shared by both IBD-A and RA patients and suggests a potentially common microbial link for inflammatory arthritis. The increased abundance of Enterobacteriaceae, previously reported in IBD, may be due to the effects of previous bowel surgery and highlights the importance of controlling for this variable in future studies.

INTRODUCTION

Inflammatory bowel diseases (IBDs), namely Crohn’s disease (CD) and ulcerative colitis (UC), are a group of heterogeneous inflammatory conditions affecting the gastrointestinal tract. Inflammatory bowel disease is thought to result from inappropriate and ongoing activation of the mucosal immune system potentially driven by gut microbiota.¹ Although there is considerable evidence linking the gut microbiota to intestinal inflammation, there is limited knowledge on its potential role in the development of extra-intestinal manifestations (EIMs) of IBD.

Evidence from a variety of disciplines firmly implicates abnormal host–microbial interactions in the pathophysiology of IBD. There are distinct gut microbial changes associated with IBD in which alterations in microbial composition and diversity correlate with disease severity.² Alterations in gut microbiota, among other environmental factors, represent a factor that may increase the risk of developing IBD. Evidence provided by genome-wide association studies (GWAS) reveals that many of the susceptibility loci for IBD are implicated in the regulation of immune responses toward gut microbiota.³ Some of the disrupted mechanisms include impaired recognition of bacterial components and the inability to clear pathogenic bacteria through autophagy. Interestingly, these loci have significant overlap with other immune-mediated diseases, such as ankylosing spondylitis and psoriasis,^{3, 4} suggesting a shared pathogenesis.

About 38%–41% of patients with IBD experience extra-intestinal manifestations (EIMs).^5–7 EIMs usually affect the joints (peripheral and axial arthropathies), the skin (erythema nodosum, pyoderma gangrenosum, aphthous stomatitis), the hepatobiliary tract (primary sclerosing cholangitis), and the eyes (episcleritis, iritis, uveitis).⁸ Inflammatory bowel disease–associated arthropathies are the most common EIMs, with a prevalence between 26% and 33%^{5–7, 9} among IBD patients. Arthralgias or joint pains without evidence of inflammation are also present in a significant number of patients. Multiple studies have demonstrated gut microbial changes associated with these phenotypes as separate diseases or in the context of IBD. Studies on germ-free mice have demonstrated that segmented filamentous bacteria (SFB) are capable of inducing Th17 cells in the intestinal lamina propria, stimulating the production of autoantibodies, and rapidly causing arthritis.^{10, 11} In humans, rheumatoid arthritis has been associated with significant gut dysbiosis that correlates with disease severity and is partially resolved after treatment with disease-modifying antirheumatic drugs (DMARDs).^{12, 13} In the context of IBD, a recent study demonstrated that patients with IBD and concomitant spondyloarthritis (SpA) have significantly increased relative abundance of Proteobacteria compared with non-SpA IBD patients and significant enrichment of IgA-coated Escherichia capable of inducing Th17-mediated responses.¹⁴ Although these studies provide insight into changes in the gut microbiota associated with inflammatory conditions like IBD and rheumatoid arthritis (RA), the majority of studies do not account for potential confounders.

We conducted a cross-sectional study to investigate the compositional and functional features that characterize the gut microbiota of patients with IBD-associated arthropathy. We selected IBD patients without arthritis as controls for joint inflammation, RA patients without IBD as controls for inflammation in the gastrointestinal tract, and healthy individuals without IBD or RA as overall controls. We obtained extensive clinical metadata to allow us to make meaningful comparisons and account for potential clinical confounders. We found potential microbial features associated with IBD-associated arthropathy and features common to both RA- and IBD-associated arthropathy that are suggestive of a global role in joint inflammation.

METHODS

Patient Selection

All methods and procedures were approved by the Institutional Review Board (IRB) of the Mayo Clinic. To investigate microbial differences associated with IBD-associated arthropathy, our study design consisted of 4 groups of patients: patients diagnosed with IBD-associated arthropathy (IBD-A group), patients with IBD without arthropathy (IBD-N group), patients with RA and without IBD (RA group), and control subjects without IBD or arthropathy (control group).

All eligible patients were selected from 3 prospective cohorts at the Mayo Clinic, Rochester, Minnesota. Each cohort study includes a biorepository of samples obtained from recruited patients intended for research use, all of which included at least stool and blood samples. IBD-N and IBD-A patients were selected from the Mayo Clinic IBD Repository, a prospective cohort of samples collected from patients with IBD. Inclusion criteria for this cohort included patients between 1 and 99 years of age diagnosed with IBD using standard diagnostic criteria.^{15, 16} Patients with RA were selected from the Mayo Clinic Rheumatology Biobank, a similar cohort for RA. Controls were selected from the Midwestern Reference Panel (MWRP), a representative cohort and biorepository of samples of the Midwestern United States (Minnesota, Iowa, North Dakota, South Dakota, and Wisconsin).¹⁷ This cohort included individuals ranging from 20 to 79 years of age who did not have gastrointestinal (GI) diseases such as IBD, irritable bowel syndrome (IBS), GI malignancies, constipation, diarrhea, Clostridium difficile infection, or celiac disease. Patients with antibiotic use within 2 weeks were also excluded from this cohort. All selected patients provided authorization to access their medical record and to share their samples with other researchers at the Mayo Clinic and other institutions.

We used the Mayo Clinic electronic health record (EHR) to ascertain cases and controls. We also revised information collected in baseline questionnaires that subjects completed upon enrollment in each cohort. This information was stored in research databases and included information about symptoms, current medications, surgical procedures, family history, and social history. All patients with GI malignancies, celiac disease, irritable bowel syndrome (IBS), and Clostridium difficile infection at the time of sample collection were excluded. In addition, we excluded patients who took any antibiotics within 1 month of sample collection. We selected 1 month as the time frame for excluding antibiotic use based on results from previous studies using different antibiotics showing that the gut microbiome returns to pretreatment composition approximately 30 days after antibiotic exposure.^18–20 Use of prebiotics and probiotics was allowed. For both the IBD-N and IBD-A groups, we selected adult patients (aged 18 years or older) from the IBD biorepository who were diagnosed with IBD using standard clinical, endoscopic, radiological, and histological criteria.^{15, 16} For the IBD-A group, we selected patients with IBD-associated arthropathy using the following diagnostic criteria: (1) fulfillment of Assessment of SpondyloArthritis international Society (ASAS) criteria for axial or peripheral spondyloarthritis^{21, 22} and (2) a recorded clinical evaluation by a rheumatologist at the Mayo Clinic to confirm the diagnosis. Diagnosis for axial spondyloarthritis (SpA) according to ASAS criteria requires the presence of back pain for at least 3 months with age of onset <45 years and 1 of the following: sacroiliitis on imaging (radiographs or magnetic resonance imaging) plus at least 1 of the SpA-associated features (inflammatory back pain, arthritis, enthesitis in the heel, uveitis, dactylitis, psoriasis, IBD, good response to nonsteroidal inflammatory drugs [NSAIDs], family history of SpA, positive HLA-B27 status, or elevated C-reactive protein) or positive HLA-B27 status plus at least 2 of the SpA-associated features. Diagnosis of peripheral SpA according to ASAS criteria requires the presence of arthritis, enthesitis, or dactylitis plus 1 or more of the following SpA-associated features: psoriasis, IBD, preceding infection, positive HLA-B27, uveitis, or sacroiliitis on radiographs or magnetic resonance imaging; or plus 2 or more of the remaining SpA-associated features: arthritis, enthesitis, dactylitis, past history of inflammatory back pain, or positive family history of SpA). We required a physical examination by a rheumatologist to confirm the presence of inflammation and reduce the number of false-positive cases, such as the presence of joint pain without inflammation. Patients with RA, any form of lupus, psoriatic arthritis, reactive arthritis, and drug-induced hypersensitivity reactions were excluded from the IBD-A group.

For the RA group, we selected adult patients from the rheumatology biorepository that fulfilled the American College of Rheumatology (ACR)/European League Against Rheumatism (EULAR) 2010 classification for diagnosis of RA.²³ These criteria require the presence of synovitis in at least 1 joint, the absence of an alternative diagnosis, and the achievement of a total score of at least 6 out of a possible 10 from individual scores in 4 domains (number of joints involved, serological abnormalities, elevated acute phase reactants, and symptom duration of at least 6 weeks). Patients with any seronegative spondyloarthropathy, drug-induced hypersensitivity reaction, any form of lupus, gout, and polymyalgia rheumatica were excluded from the RA group. For the control group, we excluded patients with concomitant IBD and those who did not fulfill the exclusion criteria for the IBD-N, IBD-A, and RA groups.

All patients were frequency-matched by age, sex, and body mass index (BMI). In addition, patients from the IBD-N and IBD-A groups were frequency-matched by diagnosis (CD, UC, and indeterminate colitis). Patients with CD and UC from both IBD groups were frequency-matched to the disease subclassification criteria included in the Montréal classification of IBD²⁴ to have a proportionate distribution of disease phenotype. Montréal criteria for CD include age at diagnosis (younger than 16 years of age, between 16 and 40 years of age, older than 40 years of age), location (ileal, colonic, ileocolonic, isolated upper GI disease), and behavior (nonstricturing/nonpenetrating, structuring, penetrating, and presence of perianal disease). Montréal criteria for UC include severity (remission, mild, moderate, and severe) and extent (proctitis, left-sided and extensive). We examined the clinical assessments recorded in the EHR from clinic visits and the information available in the research databases from each respective biobank to calculate disease severity scores for CD (Harvey-Bradshaw index),²⁵ UC (Montréal UC severity score), and RA (DAS28-CRP).²⁶ We selected these indices based on the availability and reliability of the clinical information recorded in the EHR and the databases. A Harvey-Bradshaw index of less than 5 correlates with clinical remission, an index of 5 to 7 correlates with mild disease activity, an index of 8 to 16 correlates with moderate disease activity, and an index of greater than 16 correlates with severe disease activity in CD. For the DAS28-CRP disease activity index, a score below 2.4 is classified as remission, a score of 2.4 to 2.9 is classified as low disease activity, a score 3.0 to 4.6 is classified as moderate disease activity, and a score above 4.6 is classified as high disease activity for RA. All the clinical information we collected was obtained on the same day that patients submitted their samples or from the closest available clinic visit if no changes in medications were done or new procedures were performed before sample submission. We only included clinical data obtained within 4 weeks of sample submission.

Stool Sample Collection, Processing, and Metagenomic DNA Sequencing

Stool collection kits with instructions were given to patients upon recruitment in each respective cohort study. All stool samples from all 3 cohorts were returned using the same protocol. Patients collected their stool specimens in a paper container, placed a marble-sized sample in a collection tube using the spoon provided, and put the collection tube on a frozen gel pack within an expanded polystyrene container. Patients returned their samples within 24 hours by mail or by dropping them off personally to our study personnel at the Mayo Clinic. Patients who preferred to return their samples by mail were given same-day delivery labels to ensure that the samples arrived within 24 hours.

Returned stool samples were processed within 1 hour of collection. A spoonful of stool was obtained from the collection tube and placed in a sterile urine cup. Using a sterile spatula, a small amount of stool was placed in a sterile 0.5-cc Eppendorf tube. All tubes were stored in a –80°C refrigerator until further processing for DNA extraction.

DNA isolation from stool samples was performed using the Mo Bio PowerSoil DNA Isolation Kit (Mo Bio Laboratories, Carlsbad, CA, USA) after bead beating. DNA content was quantified for each sample using the Quant-iT PicoGreen Assay to ensure a minimum concentration of 0.2 ng of DNA per µL (250 pg of isolated DNA). Each sample was processed at the University of Minnesota Genomics Center using the Nextera XT Library Prep Kit (Illumina, Inc., San Diego, CA, USA). DNA sequencing was performed with the Illumina HiSeq 2500 Sequencing System using the v4 reagent kit to produce 2 × 125 bp (paired-end) reads in high-output mode. One hundred ninety samples were originally submitted for sequencing (26 IBD-A, 75 IBD-N, 25 RA, and 64 controls). Two IBD-N samples failed quality control due to insufficient DNA content and were excluded from further analysis and processing. To preserve matching of age, sex, BMI, and IBD subtype, severity, and extent, another 7 IBD-N samples and 1 IBD-A sample were excluded from sequencing. A total of 180 samples were finally sequenced and run over 2 lanes of 95 samples each and included in the downstream analysis. Two hundred fifteen libraries were created using one-quarter reaction Nextera XT preparations followed by custom barcoding polymerase chain reaction (PCR). At least 220 million pass-filter reads were generated for each of the 2 lanes. The average quality score for each read was 36.72 (see Supplementary Table 1 for a detailed quality report). The median sequencing depth (interquartile range) was 42,025 (25,258–73,418). Sequencing and associated metadata were submitted to the Sequence Read Archive (SRA Study: SRP158781).

Compositional, Diversity, and Functional Analyses

Quality control for the paired-end whole-genome shotgun (WGS) sequences, including trimming and stitching, was completed using the SHI7 pipeline.²⁷ The quality-controlled sequences were annotated for both taxonomic and functional profiles using the SHOGUN pipeline²⁸ and the Reference Sequence (RefSeq, version 81) database.²⁹ The SHOGUN pipeline was utilized because of its fully exhaustive alignment algorithms specifically calibrated for the depth of the sequencing used in this study. To tabulate taxonomic and functional profiles, each of the sequencing reads was fully and exhaustively aligned against all reference genomes in RefSeq above 98% alignment identification, with the best match being selected for species and Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthology (KO) counts.³⁰ The KO gene families and pathways were summarized from the KO count table using the SHOGUN pipeline mappings. Diversity analyses were performed using the Quantitative Insights Into Microbial Ecology (QIIME; version 1.9.1) analysis pipeline.³¹ Functional profiles on gene families and pathways were obtained using the SHOGUN pipeline. Functional biomarker identification was performed using LEfSe.³²

Statistical Analyses

Statistical analyses were performed using R software (The R Foundation). Alpha diversity was calculated using the observed species metric, Shannon index, Chao1 estimator,³³ and the inverse Simpson index. Beta diversity was calculated on the rarefied count table using the Bray-Curtis dissimilarity metric³⁴ and the Jaccard index.³⁵ Statistical significance was defined by a P value <0.05. Multiple linear regression was used to test the association between the covariates and alpha diversity while adjusting for potential confounders, and a Wald-type test was used for assessing statistical significance. To test the association between the covariates and beta diversity measures, we used permutational analysis of variance (PERMANOVA),³⁶ a distance-based analysis of variance method based on permutation (1000 permutations, “adonis” function in the R “vegan” package). We tested the association of alpha and beta diversity with the following clinical covariates: age, sex, body mass index (BMI), IBD diagnosis (CD, UC, and indeterminate colitis), disease duration (for IBD and RA), disease severity (using respective severity scores for RA, CD, and UC), aminosalicylates (5-ASA), steroids (both luminal and systemic), methotrexate, azathioprine, 6-mercaptopurine, biologics (TNF-alpha inhibitors, anti-integrin antibodies, and ustekinumab), DMARDs (hydroxychloroquine, leflunomide, abatacept, rituximab, tofacitinib), probiotic use, inflammatory markers (erythrocyte sedimentation rate and C-reactive protein), serologic markers of RA (rheumatoid factor and anticitrullinated peptide antibodies), and history of bowel surgery. Ordination plots were generated using principal coordinate analysis (PCoA; “cmdscale” function in the R “vegan” package) for visualizing the association of covariates with the beta diversity measures. Differential abundance analysis was performed at the phylum, class, order, family, and genus levels; taxa with prevalence rates of less than 10% or with a maximum proportion of less than 0.2% were excluded from testing to reduce the number of tests. The count data were normalized into relative abundances (proportions) and square-root transformed. To identify differentially abundant taxa associated with covariates and account for the non-normality of the abundance data, a permutation-based approach was used based on the F-statistics of a linear model (transformed relative abundance as the response variable), and residual permutation was used to account for potential confounders.³⁷ False discovery rate (FDR) control (Benjamini-Hochberg procedure,³⁸ ‘p.adjust’ in standard R packages) was used to correct for multiple testing at each taxonomical level. The adjusted P values are reported.

RESULTS

Patient Characteristics

A total of 180 patients were included in this study: 25 IBD-A, 66 IBD-N, 25 RA, and 64 controls. The demographic characteristics of our study population are shown in Table 1. Among the 180 patients, 74 (41.11%) were men and 106 (58.89%) were women. The mean age was 49.22 years (95% confidence interval [CI], 47.26 to 51.18), and the mean BMI was 27.49 (95% CI, 26.72 to 28.27). The distribution of age, sex, and BMI was similar across all groups.

Table 1.

Demographic Characteristics of the Study Population

	IBD-A(n = 25)	IBD-N(n = 66)	RA(n = 25)	Control(n = 64)
Age, mean, y	48.28	48.65	51.12	49.44
Female, %	56.00	60.61	60.00	57.81
BMI, mean, kg/m2	26.97	27.54	28.41	27.29
Smoking, No. (%)	4 (16.00)	6 (9.09)	5 (20.00)	6 (9.38)
Bowel surgery, No. (%)	12 (48.00)	26 (39.39)	0 (0.00)	1 (1.61)
Small bowel resection	10 (40.00)	14 (21.21)	0 (0.00)	0 (0.00)
Colectomy	10 (40.00)	21 (31.82)	0 (0.00)	1 (1.61)
Medications, No. (%)
5-ASA	4 (16.00)	18 (27.27)	1 (4.00)	0 (0.00)
Steroids	8 (32.00)	10 (15.15)	7 (28.00)	0 (0.00)
Topical steroids (budesonide)	1 (4.00)	3 (4.55)	0 (0.00)	0 (0.00)
Systemic steroidsa	7 (28.00)	8 (12.12)	7 (28.00)	0 (0.00)
Methotrexate	3 (12.00)	1 (1.52)	12 (48.00)	0 (0.00)
AZA or 6-MP	3 (12.00)	20 (30.30)	1 (4.00)	0 (0.00)
Probiotics	4 (16.00)	7 (10.61)	1 (4.00)	0 (0.00)
Biologics, No. (%)	15 (60.00)	31 (46.97)	12 (48.00)	0 (0.00)
Infliximab	3 (12.00)	17 (25.76)	2 (8.00)	0 (0.00)
Adalimumab	6 (24.00)	8 (12.12)	2 (8.00)	0 (0.00)
Certolizumab pegol	2 (8.00)	3 (4.55)	3 (12.00)	0 (0.00)
Vedolizumab	3 (12.00)	3 (4.55)	0 (0.00)	0 (0.00)
Ustekinumab	1 (4.00)	0 (0.00)	0 (0.00)	0 (0.00)
Etanercept	0 (0.00)	0 (0.00)	3 (12.00)	0 (0.00)
Disease duration, mean, y	15.48	14.81	10.28
Disease severity, No. (%)b
Remission	12 (48.00)	32 (48.48)	12 (48.00)
Mild	7 (28.00)	18 (27.27)	0 (0.00)
Moderate	5 (20.00)	13 (19.70)	7 (28.00)
Severe	1 (4.00)	3 (4.55)	2 (8.00)

Abbreviations: 6-MP, 6-mercaptopurine; AZA, azathioprine.

^aAll patients on systemic steroids were on prednisone, except 1 patient from the IBD-N group (hydrocortisone).

^bBowel disease severity is shown for IBD-A and IBD-N patients (CDAI for CD patients and Montréal criteria for UC patients). DAS28-CRP was used for the 21 RA patients with severity data available.

In the IBD groups, 55 patients had CD (38 IBD-N and 17 IBD-A), 31 patients had UC (24 IBD-N and 7 IBD-A), and 5 patients had indeterminate colitis (4 IBD-N and 1 IBD-A). Disease characteristics of the IBD population included in the IBD-N and IBD-A groups are summarized in Table 2. For an expanded table on demographic, clinical, and laboratory metadata collected for each patient, see Supplementary Table 2. The mean Harvey-Bradshaw index was 7.25 (95% CI, 4.51 to 9.99) for CD patients in the IBD-N group and 7.63 (95% CI, 1.97 to 13.28) for CD patients in the IBD-A group. Among the patients with UC in the IBD-N group, 46% were in remission, 33% had mild disease, and 21% had moderate disease according the Montréal UC severity classification. Among UC patients in the IBD-A group, 71% were in remission, 29% had mild disease, and none had moderate disease. There was no significant difference in UC severity between IBD-N and IBD-A patients (P = 0.34). No patients with severe UC were included in the study. The mean disease activity score (DAS28-CRP) for RA patients was 2.52 (95% CI, 1.84 to 3.20), which lies within the classification of low disease activity.²⁶ Of the 25 RA patients, 20 patients were both rheumatoid factor (RF) and anticitrullinated peptide antibody (ACPA) positive at diagnosis, 3 patients were RF positive and ACPA negative, and 3 patients had unknown RF and ACPA status at the time of diagnosis. The mean RF titer among RF-positive patients was 157.09 IU/mL (95% CI, 41.56 to 272.62), and the mean ACPA titer among ACPA-positive patients was 148.63 U/mL (95% CI, 103.68 to 193.57).

Table 2.

IBD Population Characteristics

	IBD-A(n = 25)	IBD-N(n = 66)
IBD diagnosis, No. (%)
Crohn’s disease	17 (68.00)	38 (57.58)
Ulcerative colitis	7 (28.00)	24 (36.36)
Indeterminate colitis	1 (4.00)	4 (6.06)
Harvey-Bradshaw index, mean	7.63	7.25
Montréal UC severity, No. (%)
Remission (S0)	5 (71.43)	11 (45.83)
Mild (S1)	2 (28.57)	8 (33.33)
Moderate (S2)	0 (0.00)	5 (20.83)

As established in our inclusion and exclusion criteria, patients with known antibiotic exposure per the EHR within 4 weeks of sample submission were excluded. A total of 12 patients were on probiotics during sample submission: 4 IBD-A patients (16%), 7 IBD-N patients (11%), and 1 control (4%). The majority of these patients (86%) were on generic probiotic complexes containing strains of Bifidobacterium and Lactobacillus, whereas 1 IBD-N patient was on VSL #3.

We processed the WGS sequences using SHOGUN to characterize microbial taxonomic and functional profiles. We obtained 988 observations at the species level (median [range], 41,388 [192–504,992] reads per sample). Observations were classified into 21 phyla, 206 families, and 451 genera. Subjects within the 4 groups were matched for age, sex, and BMI. Before performing the compositional analyses, we analyzed clinical and demographic factors that contribute to variability in microbial composition and diversity. We used the EHR to extract clinical data on medications (5-aminosalicylates, steroids, immunomodulators, biologics), probiotics, disease severity indexes, antibody titers (RF, ACPA), inflammatory markers (erythrocyte sedimentation rate, C-reactive protein), and bowel surgery (small bowel resection, colectomy). We used the Bray-Curtis dissimilarity metric, the Jaccard index, and the PERMANOVA test³⁹ to identify sources of variability.

Compositional Profile of our IBD Population

We analyzed diversity and compositional differences that characterized our IBD cohort (both IBD-A and IBD-N) in comparison with controls (Supplementary Table 3). We also performed multiple linear regression to test the association between clinical covariates and measures of alpha and beta diversity. Among the covariates analyzed, we found that a history of bowel surgery (P < 0.001) and disease severity (P < 0.001) were significant sources of taxonomic variability in our samples. We did not find IBD severity to be a significant confounder after adjusting for history of bowel surgery. Medications, including 5-ASAs, steroids, methotrexate, azathioprine, 6-mercaptopurine, biologics (TNF-alpha inhibitors, anti-integrin antibodies, and ustekinumab), DMARDs (hydroxychloroquine, leflunomide, abatacept, rituximab, tofacitinib), and probiotics, were also not associated with taxonomic variability after adjusting for history of bowel surgery. There was no significant difference in alpha diversity when comparing all IBD patients against controls, as calculated using the observed species (P = 0.727), Shannon index (P = 0.265), Chao1 (P = 0.839), and inverse Simpson index (P = 0.164). On the other hand, beta diversity was significantly different among all IBD patients compared with controls, as calculated using the Bray-Curtis dissimilarity index (P = 0.01, PERMANOVA) and Jaccard distances (P = 0.004, PERMANOVA). These results are consistent with previous studies.⁴⁰

At the phylum level, we found a significant reduction in Euryarchaeota in our IBD cohort compared with controls (0.08% vs 1.6%, P = 0.009) (Supplementary Fig. 1A). A decrease in Methanobrevibacter (Supplementary Fig. 1B) among IBD patients was the largest contributor to the reduction in Euryarchaeota (0.07% vs 1.6%, P = 0.01). Among Proteobacteria, the abundance of Desulfovibrio was also decreased among IBD patients (0.02% vs 0.3%, P = 0.01). On the other hand, there were significant increases in several bacterial genera within the Actinobacteria, including Actinomyces (0.07% vs 0.02%, P = 0.01) and Cutibacterium (0.2% vs 0.01%, P = 0.01); and within Firmicutes, including Lachnoclostridium (1.2% vs 0.8%, P = 0.01). As previously described, every result was adjusted for multiple comparisons using FDR, and only adjusted P values are reported.

Gut bacterial composition and diversity are similar between IBD-N and IBD-A patients

We analyzed compositional and diversity differences between IBD-A patients and IBD-N patients. There were no significant differences in alpha diversity between the 4 groups of patients (IBD-N, IBD-A, controls, RA) as tested with analysis of variance (ANOVA) using the observed species (P = 0.88), Shannon index (P = 0.09), and Chao1 (P = 0.80) (Supplementary Fig. 2). For taxonomic abundance comparisons between all 4 groups of patients, see Supplementary Table 4. However, there was a significant difference in alpha diversity when using the inverse Simpson index (P = 0.01), but this difference was due to an increase in alpha diversity among RA patients when compared with controls (P = 0.02). There were no significant differences in alpha diversity between IBD-N and IBD-A patients. There were also no significant differences in beta diversity between IBD-N and IBD-A patients, as measured using the Bray-Curtis dissimilarity (P = 0.874, PERMANOVA) and Jaccard index (P = 0.262, PERMANOVA) (Supplementary Table 5).

We used the Linear discriminant analysis (LDA) effect size (LEfSe) tool to identify bacterial signatures that may serve as biomarkers of IBD-A by assessing biological consistency in each sample within each group through a series of pairwise comparisons (Wilcoxon tests). No bacterial signatures were identified as potential biomarkers for IBD-N or IBD-A patients. In contrast to previous reports,¹⁴ we did not detect increased abundances of Proteobacteria associated with IBD-A.

An increased abundance of gut microbial genes encoding metabolic pathways of tyrosine degradation is associated with IBD-A

We performed the functional analysis of metagenomics data using SHOGUN. We identified 305 metabolic pathways using the KO database.³⁰ We analyzed the difference in abundance of KEGG metabolic pathways between IBD-A and IBD-N patients. We proceeded with LEfSe to detect potential biomarkers that differentiate IBD-A from IBD-N. The results were adjusted for bowel surgery effects as performed for previous analyses. An increased abundance of tyrosine metabolism pathways (ko00350, LDA 2.38, P = 0.02) and limonene and pine degradation pathways (ko0093, LDA 2.21, P = 0.04) was associated with IBD-A (Fig. 1). On the other hand, an increase in phosphonate and phosphinate metabolism pathways (ko00440, LDA 2.30, P = 0.02) was associated with IBD-N (for calculated effect sizes, see Supplementary Table 6).

FIGURE 1.

LEfSe barplot showing KEGG biomarkers associated with IBD-A (n = 25, red) and IBD-N (n = 66, green). An increased abundance of metabolic pathways of tyrosine metabolism (ko00350, LDA 2.38, P = 0.02) and limonene and pine degradation (ko0093, LDA 2.21, P = 0.04) was associated with IBD-A. On the other hand, phosphonate and phosphinate metabolism (ko00440, LDA 2.30, P = 0.02) were identified as biomarkers for IBD-N.

To ensure that medications were not a confounding factor, we analyzed if medications had any effect on the abundance of these metabolic pathways. We included the following medications and medication categories: 5-aminosalicylates, steroids, methotrexate, azathioprine, 6-mercaptopurine, TNF-alpha inhibitors, probiotics, DMARDs, in our analysis. DMARDs included hydroxychloroquine, methotrexate, leflunomide, sulfasalazine, TNF-alpha inhibitors, abatacept, rituximab, and tofacitinib. Sources of variability associated with these medications were determined through a beta-diversity analysis using the Bray-Curtis dissimilarity and PERMANOVA tests, where a P value <0.05 was deemed significant. As described in the “Methods,” FDR control was used to correct for multiple testing. None of these medications was associated with significant variability on microbial gene abundance (Supplementary Table 7).

An expansion of Clostridiaceae characterizes arthritis

In addition to looking for bacterial signatures that differentiate IBD-A from IBD-N, we were also interested in bacterial signatures that characterize arthritis and are common between IBD-A and RA patients (arthritis phenotype). The goal was to identify gut microbiota that may have a role in the pathophysiology of arthritis. An increase in the relative abundance in the Clostridiaceae family of bacteria was found to be characteristic of the arthritis phenotype (1.6% vs 0.14%, P = 0.045) (Fig. 2), which was almost entirely composed of Clostridium (99.9%). No other significant compositional differences between the arthritis phenotype and controls were found. All P values were FDR-corrected for multiple comparisons.

FIGURE 2.

Barplot comparing the abundance of Clostridiaceae and Clostridrium between patients in the arthritis group (A, 25 IBD-A patients + 25 RA patients) and controls (C, n = 64). Results were adjusted for surgical and IBD effects. The increased abundance of Clostridiaceae was found to be statistically significant (P = 0.045).

We also looked into compositional differences that characterize our RA cohort in comparison with controls. We found a significant increase in Epsilonproteobacteria (0.3% vs 0.2%, P = 0.03) and Campylobacterales (0.3% vs 0.2%, P = 0.04), 2 classes within the Proteobacteria phylum, among RA patients. Alpha diversity was significantly increased in RA when compared with controls, as measured by the inverse Simpson index (P = 0.017), which indicated increased species evenness in RA. Alpha diversity was not significantly different when measured using the observed species (P = 0.583), Shannon index (P = 0.100), and Chao1 (P = 0.327). Beta diversity was not significantly different between both groups as measured using the Bray-Curtis dissimilarity index (P = 0.142, PERMANOVA) and Jaccard distances (P = 0.528, PERMANOVA). A previous study that characterized the RA-associated fecal microbiota and included patients taking immunosuppressive medications identified an expansion of Actinobacteria, particularly Eggerthella, when compared with controls.¹³ In addition, there were also increases in Eubacterium, a genus within the Clostridiales family. We did not identify an increase in Actinobacteria in our study, but the expansion of Eubacterium is consistent with the increased abundance of Clostridiales shared between our IBD-A and RA patients. Bacterial clusters within the Clostridiales family have also been previously identified in treatment-naïve RA patients, particularly Clostridium aspariforme.¹² The expansion of Prevotella copri previously reported in new-onset treatment-naïve RA was not detected in our cohort.⁴¹ All P values were FDR-corrected for multiple comparisons.

Bowel surgery and disease severity significantly alter microbial composition in IBD

We found that a history of bowel surgery (P < 0.001) and disease severity (P < 0.001) were significant sources of variability in our samples (Supplementary Table 8). A summary of the compositional and diversity changes associated with a history of bowel surgery among IBD patients is shown in Figure 3. After adjusting for the effects of bowel surgery, the effects of disease severity were not significant. Twenty-eight IBD patients (42%) had a history of bowel surgery (39% IBD-N, 48% IBD-A). There were no significant differences in the proportion of IBD patients with a history of bowel surgery between the IBD-N and IBD-A groups (P = 0.46). In the IBD-N group, 12 patients had partial colectomies, 9 patients had total colectomies, and 14 patients had small bowel resections with an average of 36.03 cm resected (95% CI, –5.11 to 77.18). In the IBD-A group, 9 patients had partial colectomies, 2 patients had total colectomies, and 10 patients had small bowel resections with an average of 32.53 cm resected (95% CI, 10.41 to 54.65). The average time since the last surgical intervention before sample submission was 9.88 years (95% CI, 6.67 to 13.08). Time since last surgery was similar between IBD-N and IBD-A patients. Alpha diversity was significantly decreased in IBD patients with a history of bowel surgery compared with IBD patients without a history of bowel surgery (Fig. 3A), as measured using observed species (P = 0.003), Shannon index (P = 0.001), Chao1 (P = 0.003), and the inverse Simpson index (P = 0.002). Beta diversity was also significantly different among IBD patients with a history of bowel surgery (Fig. 3B), as measured by Bray-Curtis dissimilarity (P = 0.001) and the Jaccard index (P = 0.001).

FIGURE 3.

Compositional and diversity changes associated with history of bowel surgery among IBD patients. A, Alpha diversity (observed species) was significantly decreased among all 91 IBD patients (both IBD-N and IBD-A included) with a history of bowel surgery (P = 0.003, Wald-type test; error bars represent standard error). B, Beta diversity (Bray-Curtis dissimilarity) was significantly different among IBD patients with a history of bowel surgery (P = 0.001, PERMANOVA). C, An increased abundance of Proteobacteria (P = 0.009) and Fusobacteria (P = 0.01) and a decrease in Firmicutes (P = 0.009), Bacteroidetes (P = 0.01), and Spirochaetes (P = 0.009) were found in IBD patients with a history of bowel surgery.

Among bacterial phyla, we detected a significant increase in Proteobacteria (36.7% vs 6.3%, P = 0.009) and Fusobacteria (1.6% vs 0.1%, P = 0.01) in IBD patients with a history of bowel surgery compared with IBD patients without any history of bowel surgery (Fig. 3C). On the other hand, the abundance of Firmicutes (23.9% vs 40.4%, P = 0.009), Bacteroidetes (29.3% vs 43.4%, P = 0.01), and Spirochaetes (0.04% vs 0.06, P = 0.009) was decreased among IBD patients with a history of bowel surgery. In terms of magnitude, the increase in Proteobacteria appears to be most significant, with an increase in Enterobacteriaceae as the greatest contributor. Within the Enterobacteriaceae family, the abundance of Escherichia in IBD patients with history of bowel surgery is 21.1% compared with 4.3% in IBD patients without history of bowel surgery (P = 0.004). Other notable genera that were also found to be increased among IBD patients with a history of bowel surgery include Klebsiella (10.9% vs 1.0%, P = 0.004), Salmonella (0.1% vs 0.004%, P = 0.007), and Shigella (0.005% vs 0.001%, P = 0.01). On the other hand, there was a decrease in Enterobacter (0.7% vs 0.04%, P = 0.004) among IBD patients with a history of bowel surgery.

DISCUSSION

The role of the gut microbiota in the pathophysiology of extraintestinal manifestations of IBD is not clearly understood. This cross-sectional study aimed to characterize the compositional and functional differences in the gut microbiota of patients with IBD-associated arthropathy, with the objective of identifying potential biomarkers of disease and targets of treatment in this population. We found that IBD-A was associated with a higher abundance of bacterial genes encoding pathways of tyrosine metabolism. In addition, our 4-group design allowed us to determine that an increased abundance of Clostridiaceae in the gut is associated with the arthritis phenotype, irrespective of IBD-A or RA. Last, but not least, we also found that a history of bowel surgery is a significant confounder not usually accounted for in microbiome studies.

Evidence from previous human studies shows that immune cells from the lamina propria in the gut of IBD patients are capable of producing antibodies against host proteins and epitopes.^{42, 43} Some of these targeted epitopes are shared between intestinal bacteria and the joints and eyes of IBD patients.⁴⁴ Subsequent studies on mice models have demonstrated that the gut microbiota drive the differentiation and migration of T follicular (Tfh) helper cells into systemic lymphoid tissues, an essential process in the production of autoantibodies that cause arthritis.^{10, 11, 45} Evidence collected from genome-wide association studies and the gut microbiome in IBD patients suggests a pathophysiologic link between intestinal inflammation and arthritis. The independent identification of variants of the IL23R gene as being strongly associated with IBD⁴⁶ and ankylosing spondylitis⁴⁷ has supported the development of monoclonal antibodies directed against this pathway for treatment.^{48, 49} More recently, IgA-coated Escherichia coli have been shown to be selectively enriched in the gut of patients with CD-associated SpA when compared with CD alone.¹⁴ Our study contributes to the available knowledge by providing evidence on the functional characteristics that differentiate the gut microbiota of patients with IBD-A. Specifically, we found a higher abundance of bacterial genes encoding pathways of tyrosine metabolism among IBD-A patients compared with IBD patients without arthropathy. Gut microbiota are crucial for the breakdown of tyrosine and other amino acids, as shown in previous animal studies, where serum concentrations of sulfated direct bacterial metabolites of tyrosine have been demonstrated to be significantly higher in conventional mice compared with germ-free mice.⁵⁰ Phenols and p-cresol account for approximately 70% of the luminal products of aromatic amino acid metabolism in the gut, which are predominantly produced in the distal colon by tyrosine-metabolizing anaerobes.⁵¹ Among these metabolites, p-cresol has been studied the most because of its potential implication in a variety of human diseases, particularly chronic kidney disease.⁵²P-cresol might have a role in increasing the risk of cardiovascular disease among uremic patients with chronic kidney disease by inhibiting endothelial proliferation and repair⁵³ while also contributing to increased susceptibility to infection in these patients by suppressing Th1-type cellular responses.⁵⁴ At the cellular level, P-cresol is also known for inhibiting mitochondrial respiration and increasing superoxide anion production in intestinal epithelial cells.⁵⁵ At high concentrations, p-cresol uncouples ATP synthesis and increases DNA damage. In the context of IBD, p-cresol levels have been found to be significantly decreased in urine samples from patients with CD when compared with UC and controls.⁵⁶ On the other hand, p-cresol appears to be increased in CD during remission,⁵⁷ suggesting some value in discriminating between IBD subtypes. Whether the association of increased tyrosine metabolism in the gut of patients with IBD-A results in increased absorption of tyrosine metabolites is not known and will need further confirmation through metabolomics. The capacity of p-cresol to suppress Th1 activity and cause epithelial cell damage suggests that it may have a role in altering interactions between the gut microbiota and the mucosal immune system.

The association between increased microbial tyrosine metabolism in the gut of patients with IBD-A may provide some potential for studying dietary interventions that target this pathway. Although diet is known to cause significant shifts in gut microbiome composition and metabolism,⁵⁸ evidence supporting specific dietary interventions to improve outcomes in IBD is limited and standard guidelines have not been established.⁵⁹ A systematic review associated diets high in total fat, polyunsaturated fatty acids, omega-6 fatty acids, and meat with increased risk of developing CD and UC.⁶⁰ A prospective cohort study found that consumption of red and processed meat was associated with higher risk of relapse from UC.⁶¹ Interestingly, there is some overlap between foods high in tyrosine (cheese, soybeans, beef, pork, lamb, salmon, chicken, eggs) and foods associated with increased risk of IBD. Future studies should explore whether short-term dietary interventions with low tyrosine content result in any improvements in clinical outcomes in patients with IBD-A.

In addition to the functional differences identified in IBD-A, we were also able to identify compositional changes shared between IBD-A and RA that are significantly different from controls. Our compositional analyses showed that an increased abundance of Clostridiaceae is associated with the arthritis phenotype that IBD-A and RA patients have. Previous studies have associated Clostridium species with various forms of arthritis, including RA with high disease activity.^{12, 62}Clostridium bacteria are also among the group of bacteria known to metabolize aromatic amino acids, such as tyrosine, which are then absorbed in the gut and eventually excreted in the urine.^{63, 64} An increased abundance of Clostridium among patients with arthritis may correlate with the increased activity of tyrosine degradation found in the gut of IBD-A patients. The relative contribution of Clostridiaceae family members and their tyrosine-derived metabolites in the pathophysiology of extra-intestinal manifestations and arthritis remains to be further elucidated.

As part of our aim of identifying microbial signatures that characterize the arthritis phenotype, we analyzed taxonomic differences associated with RA. We found a significant increase in Epsilonproteobacteria and Campylobacterales, 2 classes within the Proteobacteria phylum, among RA patients when compared with controls. Species evenness was also increased in RA. Our findings are partially consistent with those from a previous report by Chen et al., in which fecal samples from a similar cohort of RA patients were analyzed.¹³ Although an expansion of Actinobacteria was not identified in our study, the expansion of Eubacterium is taxonomically consistent with the increased abundance of Clostridiaceae shared between IBD-A and RA in our study. Other members of the Clostridiaceae family, such as SFB, have also been associated with arthritis in mice^{10, 11} and treatment-naïve RA patients.¹² Additional differences in composition and diversity identified in our RA cohort are likely due to differences in disease duration and RF titers previously associated with decreased species richness,¹³ which were lower in our study compared with those reported by Chen et al. In addition, a lower sample size of RA patients may have limited our ability to detect significant changes in Eubacterium. In summary, there seems to be a consistent association between bacteria within the Clostridiales family and other phylogenetically related taxa and arthritis, which should be further confirmed through analysis of mucosal samples in larger cohorts.

We also found that a history of bowel surgery was associated with significant microbial changes in our IBD cohort, characterized by an expansion of Proteobacteria as a consequence of increased abundance of Escherichia and Klebsiella within the Enterobacteriaceae family. Our cohort included patients who previously underwent small bowel resection for stricturing or fistulizing CD and partial or total colectomy for UC. The expansion of Proteobacteria was consistent regardless of the type or location of the intestinal surgery. Intestinal surgery is common among patients with a longstanding history of IBD, as 70%–80% of patients with CD and 20%–30% of patients with UC will require surgical intervention during their lifetimes.⁶⁵ Unfortunately, disease recurrence is still common after bowel resection. The reduction in recurrence rates associated with the use of antibiotic prophylaxis after ileocecal resection (ICR) in CD^{66, 67} suggests that bacteria may be implicated. Ileocecal resection, the predominant surgical intervention in our CD patients, has been shown to induce colonization of Bacteroides, E. coli, Fusobacterium, and nitrate-reducing bacteria in the neoterminal ileum after 1 year.⁶⁸ Mouse models of non-IBD ICR have demonstrated similar findings of significantly increased colonic Enterobacteriaceae abundance and dramatically decreased colonic Bacteroides, Porphyromonas, and Prevotella 28 days post-ICR.⁶⁹ Although these studies were focused on ICR, the microbial changes were similar to those found in our patients who previously underwent bowel surgery that did not involve ICR. Our study also showed that the expansion of Enterobacteriaceae is persistent among patients who underwent surgery more than 8 years before. It is important to note that increased abundance of Proteobacteria and Escherichia has also been found in previous studies with non-treatment-naïve patients with CD, especially in patients with ileal CD.^{14, 70} Furthermore, significant changes in beta diversity have been attributed to shifts in the abundance of Proteobacteria in patients with CD and ICR.⁷¹ In the study by Viladomiu et al., Proteobacteria and Enterobacteriaceae were significantly increased in CD-associated SpA. However, patients with previous intestinal surgery other than ICR were not explicitly excluded. We did not detect significant differences in Proteobacteria among IBD-A patients after adjusting for surgical effects in our study. Based on our results, the increased abundance of Enterobacteriaceae in non-treatment-naïve patients with CD is mostly a consequence of the effects of intestinal surgery on the gut microbiota. We suggest that future studies involving microbiome analysis of non-treatment-naïve IBD patients with history of bowel surgery should correct for surgical effects to avoid confounding results.

Our study presents several limitations. First, our cohort was composed of predominantly white patients from the Midwestern United States. African Americans have a higher incidence of IBD-associated arthropathy and ophthalmological complications compared with whites with Crohn’s disease within the same geographical area.^{72, 73} These differences in incidence may be due to differences in genetic susceptibility and unequal access to health care services. However, environmental variables such as diet and lifestyle may play a larger role in gut microbial composition and function than race.⁷⁴ Second, we were not able to control for diet in our study, a major factor that affects gut microbial composition and function. However, dietary variations within the same geographical region are likely to be less significant than diet-associated changes across continents and cultures.⁷⁵ Third, our analyses were performed on stool samples and not mucosal samples. Previous studies have shown that dysbiosis in IBD is more pronounced in the mucosa-associated microbiota compared with the fecal microbiota.² It is possible that compositional differences between IBD-N and IBD-A may be more pronounced at the mucosal level and may be more relevant in the pathophysiology of EIMs and arthritis due to the close interaction with the mucosal immune system. For this reason, the results from this study should be corroborated with analysis of mucosal samples. Finally, it is possible that the lack of taxonomic differences between IBD-A and IBD-N in our study may be due to a small sample size of IBD-A patients. Our sample size was calculated based on known bacterial differences between RA patients and controls, as the effect size of microbial differences between IBD-A and IBD-N is not known. In the study by Viladomiu et al., a sample size of 25 patients with CD-associated SpA was sufficient to detect taxonomic differences in Proteobacteria.¹⁴ In our study, expansion of Proteobacteria was associated with history of bowel surgery and not IBD-A. For this reason, it is more likely that the use of fecal samples instead of mucosal samples may have played a larger role in the lack of taxonomic differences between IBD-A and IBD-N than the sample size used in our study.

CONCLUSIONS

An increased abundance of gut microbial genes encoding pathways of tyrosine degradation was associated with IBD-A. An increased abundance of Clostridiaceae was shared by both IBD-A and RA patients and suggests a potentially common microbial link for inflammatory arthritis. The increased abundance of Enterobacteriaceae previously reported in IBD may be due to the effects of previous bowel surgery; it is important to control for this variable in future studies.

Conflicts of interest: D.K. serves as CEO and holds equity in CoreBiome, a company involved in the commercialization of microbiome analysis. The University of Minnesota also has financial interests in CoreBiome under the terms of a license agreement with CoreBiome. These interests have been reviewed and managed by the University of Minnesota in accordance with its Conflict-of-Interest policies. No conflicts of interest to disclose from other authors.

Supported by: This publication was supported by funding from the National Center for Advancing Translational Sciences grant number UL1 TR002377, from the National Center for Advancing Translational Sciences, grant number DK100638 (P.C.K.), grant number DK111850 (P.C.K.), grant number DK114007 (P.C.K.), and support from the Center for Individualized Medicine, Mayo Clinic, Rochester, Minnesota (P.C.K.).

Disclaimer: The contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH.