The first 2 authors contributed equally to this work.
Twin studies reveal specific imbalances in the mucosa-associated microbiota of patients with ileal Crohn's disease
Article first published online: 20 NOV 2008
Copyright © 2009 Crohn's & Colitis Foundation of America, Inc.
Inflammatory Bowel Diseases
Volume 15, Issue 5, pages 653–660, May 2009
How to Cite
Willing, B., Halfvarson, J., Dicksved, J., Rosenquist, M., Järnerot, G., Engstrand, L., Tysk, C. and Jansson, J. K. (2009), Twin studies reveal specific imbalances in the mucosa-associated microbiota of patients with ileal Crohn's disease. Inflamm Bowel Dis, 15: 653–660. doi: 10.1002/ibd.20783
- Issue published online: 7 APR 2009
- Article first published online: 20 NOV 2008
- Manuscript Accepted: 9 SEP 2008
- Manuscript Received: 27 AUG 2008
- Örebro University Hospital Research Foundation
- Örebro County Research Foundation
- Bengt Ihre's Foundation
- Uppsala BioX Micprof project
- Swedish University of Agricultural Sciences
- US Department of Energy Contract. Grant Number: DE-AC02-05CH11231
- Lawrence Berkeley National Laboratory
- monozygotic discordant twins;
- ileal Crohn's disease;
- Faecalibacterium prausnitzii;
- Escherichia coli;
- mucosa-associated microbiota
Background: Large interindividual variation in the composition of the intestinal microbiota between unrelated individuals has made it challenging to identify specific aspects of dysbiosis that lead to Crohn's disease (CD).
Methods: To reduce variations in exposure during establishment of the gut flora and the influence of genotype, we studied the mucosa-associated microbiota of monozygotic twin pairs that were discordant (n = 6) or concordant (n = 4) for CD. DNA was extracted from biopsies collected from 5 locations between the ileum and rectum. Bacterial 16S ribosomal RNA genes were amplified and community composition assessed by terminal-restriction fragment length polymorphism, cloning and sequencing, and quantitative real-time polymerase chain reaction (PCR).
Results: The microbial compositions at all biopsy locations for each individual were similar, regardless of disease state, but there were differences between individuals. In particular, individuals with predominantly ileal CD had a dramatically lower abundance (P < 0.001) of Faecalibacterium prausnitzii and increased abundance (P < 0.03) of Escherichia coli compared to healthy co-twins and those with CD localized in the colon. This dysbiosis was significantly correlated to the disease phenotype rather than genotype.
Conclusions: The reduced abundance of F. prausnitzii and increased abundance of E. coli are indicative of an ileal CD phenotype, distinct from colonic CD, and the relative abundances of these specific bacterial populations are promising biomarker candidates for differential diagnosis of CD and eventually customized treatment.
(Inflamm Bowel Dis 2009)
Defining the etiology of Crohn's disease (CD) has proven difficult due to the complicated interplay of environmental and genetic factors leading to disease development. Increasing evidence suggests that CD is ultimately the result of a breakdown in the détente relationship between commensal bacteria in the intestine and the host immune system.1, 2 A disease concordance rate of ≈50% in monozygotic twins3 identifies genetics as a major contributing factor toward CD development. However, an equal discordance rate in monozygotic twins3 points to the importance of environmental factors, including the composition of the intestinal microbiota.
The current hypothesis is that sufficient antigenic stimulation to cause disease is provided by an imbalance of beneficial and detrimental commensal organisms or “dysbiosis.”4–6 Recent reports indicate increased abundances of Bacteroides7 and Enterobacteriaceae,8–10 particularly Escherichia coli,11 and a reduction in microbial diversity12–14 in CD. However, no single bacterial species has been convincingly indicated as a biomarker of CD to date. Since host genotype plays a substantial role in shaping microbial population, as evidenced by similarities in monozygotic twins,15–17 it is unclear whether microbial imbalance is associated with host genotype or disease. This can be clarified by studying monozygotic twin pairs that are discordant (1 is healthy and 1 is sick) for disease.
The mucosal microbiota differs from that of the digesta,18, 19 and is of particular interest because the mucosal surface is the site of microbial recognition by the host that if improperly regulated could lead to the inflammatory responses typical of CD.20, 21 Therefore, we studied biopsy samples collected at colonoscopy from previously studied monozygotic twin pairs22 that were either discordant or concordant for CD. Our hypothesis was that studying discordant monozygotic twins in depth using a combination of molecular approaches would facilitate the identification of specific bacteria that are correlated with CD incidence rather than host genotype, including those not yet cultivated.
MATERIALS AND METHODS
A total of 10 monozygotic twin pairs obtained from a previously described Swedish twin population3 were studied: 2 pairs discordant for predominantly ileal CD (ICD), 4 pairs discordant for predominantly colonic CD (CCD), 2 pairs concordant for ICD, and 2 pairs concordant for CCD. Table 1 summarizes relevant clinical data for patients and responses to a questionnaire regarding the usage of antibiotics, nonsteroidal antiinflammatory drugs within the preceding 12 months, gastroenteritis within the last 3 months, and specific dietary habits that have been previously reported.22 Subjects were born between 1936 and 1986 with patient groups, as defined by disease status, being similar in age (mean ± SD) CCD 49.0 ± 18.5 (n = 8), healthy (HC) 52.8 ± 17.2 (n = 6), ICD 50.8 ± 4.5 (n = 6). According to the Harvey–Bradshaw score,23 all individuals were in remission with the exception of 2 (10b and 15a), who had previously undergone ileocecal resection. However, none of these had postoperative recurrence, endoscopic recurrence score <2,24 at colonoscopy, strongly suggesting that their Harvey–Bradshaw score reflected a coexisting functional disorder or possibly bile acid malabsorption and not active CD. The use of human subjects for this study was approved by the Örebro County Ethical Committee (Dnr167/03).
|ID||Status||Birth Year||Sex||Nod2 Status||Gastroenteritis||Disease Duration||Antibiotics||NSAID||Surgery|
|10a||ICD||1962||F||wt||Yes||22||No||Yes||ileal res + righ hemi|
|13a||CCD||1943||F||wt||No||18||No||Yes||segm colonic rers|
|15a||ICD||1953||M||snp 8||No||31||No||No||ileal res|
|15b||ICD||1953||M||snp 8||No||31||No||No||ileocec res|
|16a||ICD||1954||F||wt||No||33||No||Yes||Ileo res + right hemi|
|18a||ICD||1953||M||wt||No||34||No||No||ileal res + right hemi|
Patients received standard bowel cleansing with the polyethylene glycol preparation, Laxabon (BioPhausia, Stockholm, Sweden) the evening before colonoscopy. Biopsies were taken from distal ileum, ascending, transverse and descending colon, and rectum, if technically possible. Two biopsies were collected from each location in each subject using disposable biopsy forceps (Olympus, Center Valley, PA). Biopsies were immediately placed in 0.5 mL freezing buffer (5 mM K2HPO4, 1.3 mM KH2PO4, 2 mM Na3C6H5O7, 1 mM MgSO4x7H2O, and 17.2% glycerol) in CryoPlus tubes (Sarstedt, Germany), frozen on dry ice, and stored at −70°C.
Biopsies were thawed on ice and centrifuged at 15,000g for 5 minutes. Freezing buffer was removed by aspiration and DNA was isolated from the entire biopsy using the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany). Extraction was performed according to the manufacturer's instructions, with the addition of 2 45-second bead-beating steps at level 5 on a FastPrep-24 (MP Biomedicals, Solon, OH) at the beginning of the protocol.
Terminal-restriction Fragment Length Polymorphism (T-RFLP)
To obtain a profile of mucosa-associated bacteria, the total extracted DNA from each biopsy was analyzed by T-RFLP using a protocol previously developed for DNA extracted from feces.22 Briefly, bacterial 16S rRNA genes were specifically amplified by polymerase chain reaction (PCR) using broad-range bacterial primers Bact-8F,25 5′-end labeled with 6-carboxyfluorescein, and 926r26 (Table 2). The resulting PCR products were digested with HaeIII (New England BioLabs, Ipswich, MA) and separated on an ABI 3730 capillary sequencer (Applied Biosystems, Foster City, CA). Electrophoregrams were processed using Peak Scanner v1.0 (Applied Biosystems) and relative peak areas of terminal restriction fragments (TRFs) corresponding to sizes between 50 and 500 bp were calculated by dividing individual peak area by total peak area within this size constraint. Only peaks above a threshold of 0.5% were included in further analyses. A minimum threshold of 104 16S rRNA gene copies, as measured by quantitative (q)PCR (see below), was set to avoid bias due to low total abundance.
|Target (Purpose)||Sequence (5′-3′)||Annealing Temp (°C)||Reference|
|Bacterial 16S||F: TCCTACGGGAGGCAGCAGT||60||27|
|rRNA gene (qPCR)||R: GGACTACCAGGGTATCTAATCCTGTT|
|F. prausnitzii (qPCR)||F: GATGGCCTCGCGTCCGATTAG||58||28|
|uidA (qPCR)||F: AGCGTTGAACTGCGTGATGC||58||this study|
|ipaH (PCR)||F: TGGAAAAACTCAGTGCCTCTGCGG||60||29|
|β-globin (qPCR)||F: CAACTTCATCCACGTTCACC||58||this study|
|16S rRNA gene||F: AGAGTTTGATCCTGGCTCAG||55||25|
Cloning and Sequencing
16S rRNA genes were cloned from selected DNA samples (11a, 11b, 13a, and 13b) to aid in the identification of bacteria corresponding to TRF sizes of interest based on the analysis of T-RFLP data. Extracted DNA samples were amplified using primers Bact-8F and 926r.25, 26 PCR products were gel-purified (Qiagen), cloned into TOPO TA pCR 4.0 vector (Invitrogen, Carlsbad, CA), and transformed into E. coli TOP 10 chemically competent cells (Invitrogen). Transformants were screened by PCR amplification with M13 primers (Invitrogen) and appropriate PCR product sizes were determined by agarose gel electrophoresis. The resulting PCR products were then analyzed by T-RFLP as described above and products that resulted in TRF lengths corresponding to peaks of interest were sequenced. Sequences obtained were subjected to phylogenetic assignment using the RDP Naïve Bayesian rRNA Classifier v. 2.0. Unique sequences were deposited with GenBank at NCBI under accession numbers EU668957 to EU668961.
All qPCR assays were performed in duplicate 25 μL reactions in 96-well plates using an iQ5 Real-time Detection System (BioRad, Hercules, CA). Total 16S rRNA gene copies for bacteria were determined for each sample with the general bacterial primers (Table 2) using iQ supermix (Biorad). Human cell numbers were determined for biopsies from the ileum and descending colon using self-designed primers specific for the β-globin gene (Table 2), which were designed using MacVector 9.5.4 (Accelrys Software, San Diego CA). 16S rRNA gene copies for the specific bacterium Faecalibacterium prausnitzii were also determined using specific primers (Table 2). Escherichia coli was quantified using self-designed primers that amplify uidA (Table 2), encoding β-glucuronidase, which is specific to E. coli and Shigella. Specificity to E. coli was confirmed by negative PCR for Shigella specific ipaH.29 qPCR assays for β-globin, uidA, and F. prausnitzii were performed using SYBR GreenER qPCR Supermix for iCycler (Invitrogen). The resulting melting curves were visually inspected to ensure specificity of product detection. For quantification of target DNA copy number, standard curves were generated for each gene of interest using gel-extracted PCR products quantified using a Qubit Fluorometer (Invitrogen).
To assess the diversity of the microbial community, defined by evenness and richness of TRFs, Simpson's index of diversity30 was applied to the T-RFLP data. Similarity indices were calculated within individuals, comparing each location to the consensus of all locations, using Manhattan metrics.31 Differences between discordant twins were calculated using Manhattan metrics and significance tested by analysis of variance (ANOVA; SPSS, Chicago, IL). qPCR data and abundance data for individual TRFs were analyzed as a 1-way ANOVA using the general linear model (SPSS). Group means were separated by REGWF with significance of P < 0.05. Correlation coefficients were calculated using the Pearson Correlation Procedure (SPSS).
Mucosa-associated Community Profiles
Of the 94 biopsies collected, 8 (3 CCD, 2 ICD, 3 HC) were not included in T-RFLP analysis because of low bacterial numbers in those samples (<104 16S rRNA gene copies). However, a minimum of 3 biopsies was included from each subject for further analyses. A total of 111 TRFs, representing different ribotypes, were found in the 20 individuals studied. The average number of TRFs ± SE per biopsy was 28.2 ± 2.94, 27.4 ± 1.67, and 23.6 ± 5.99 for CCD, HC, and ICD respectively.
Bacterial numbers were not affected by disease phenotype (P = 0.376) or sample location (P = 0.295). Mean 16S rRNA gene copy numbers per copy β-globin ± SE were 35.5 ± 15.6 and 25.2 ± 15.3 for CCD, 27.4 ± 15.7 and 11.4 ± 5.7 for HC, 17.0 ± 10.3 and 10.6 ± 5.3 for ICD, for the ileum and distal colon, respectively.
Microbial diversity (mean ± SE), measured as a combination of TRF richness and evenness, was lower (P < 0.004) in individuals with ICD (0.798 ± 0.05) than those with CCD (0.850 ± 0.010) or HC (0.877 ± 0.018). This result may be confounded by surgery; however, the microbial diversity in CCD individuals who had undergone surgery was not different (P = 0.67) from CCD individuals who had not undergone surgery.
Consistency Between Locations
Cluster analysis of T-RFLP profiles, using Bray Curtis metrics, showed that the microbial compositions in samples collected from an individual grouped together, regardless of biopsy location (Fig. 1). Similarity scores for biopsies, using the Manhattan index, as compared to the mean of all biopsies within an individual were generally higher than 80%, with an average of 88.5%. No difference in the microbial community structures at the different sampled locations within individuals was observed in the different sampled groups (P = 0.25); with similarity scores (mean ± SE): HC (86.0 ± 5.3), ICD (88.8 ± 3.3), and CCD (90.3 ± 1.7) groups.
Similarity Between Twins
The average microbial community structures from all biopsy locations for each individual were compared between matched twin pairs. The discordant twin pairs, with ICD, had lower similarity scores (36%–53%), measured using abundance data (Manhattan index), compared to the discordant twins with CCD (52%–66%). This was also reflected in lower average similarities between concordant twins pairs with ICD (48%–62%) compared to concordant twins with CCD (56%–67%). These data show a trend (P < 0.07) for individuals with ICD to be less similar to their twin than CCD individuals; however, concordant twins were not more similar (P = 0.29) than discordant twins.
Phenotypic Differences in Mucosa-associated Microbiota
Analysis of T-RFLP data identified TRF223 to be abundant in all HC and CCD individuals, but scarce or absent in subjects with ICD. This TRF corresponded to F. prausnitzii (98.2%–99.5% similarity) based on screening of 16S rRNA gene clone sequences obtained from the same individuals. It was noteworthy that all sequences were more similar to uncultured clones deposited in databases than to the cultured F. prausnitzii type strain.
We found that F. prausnitzii, measured by qPCR as a percent of total 16S rRNA genes (Fig. 2A) was dramatically lower in individuals with ICD (P < 0.001) than those with CCD or HC, with respective values of (mean ± SE) 0.4 ± 0.89 (ICD), 11.3 ± 3.36 (CCD), and 8.7 ± 2.49 (HC). Discordant ICD twins (pairs 16 and 18) segregated to disease phenotype rather than twin pair for F. prausnitzii abundance.
We designed novel primers targeting the uidA gene for specific quantitation of E. coli by real-time PCR (Fig. 2B). E. coli was significantly more abundant (P < 0.03) in biopsies at all locations from ICD patients than in HC or those with CCD. In 8 of 9 positive diseased individuals, whether CCD or ICD, when E. coli was present at detectable levels it was found at all locations (1.3–4.5 log uidA per 106 16S rRNA genes). E. coli was also detected in some healthy individuals (11b, 12a, and 14a); however, only in a maximum of 2 locations, and at a level of 1.2–2.5 log uidA per 106 16S rRNA genes. However, in 1 pair of ICD concordant twins (pair 10) E. coli was only detected in 1 of the individuals, but at all locations. Similar to the situation we observed for F. prausnitzii, the discordant twins with 1 individual having CD localized in the ileum segregated according to disease phenotype rather than twin pair with respect to E. coli abundance.
It is known that the host genotype partly determines the microbial community composition in the human gut.15–17 Therefore, our hypothesis was that by studying the mucosa-associated microbiota of identical twins it would be easier to untangle the respective contributions of host genetics and intestinal bacteria toward CD etiology. We found that discordant ICD twins segregated according to their disease phenotype rather than twin pair with respect to the abundance of some specific bacterial populations. These results suggest that the bacterial composition in the intestine is associated with disease rather than host genotype.
We used 2 molecular approaches to quantify the abundance of specific bacterial populations in the biopsy samples. Both T-RFLP and qPCR approaches provided significant indications that specific bacterial species were more or less abundant in individuals with ICD compared to CCD or healthy individuals. In particular, TRF223, corresponding to F. prausnitzii, was abundant in all HC and CCD individuals, but scarce or absent in subjects with ICD. We previously detected the same TRF in fecal samples collected from 21 out of 22 healthy individuals, including some of the same individuals included in the present study22 and in 89 out of 90 healthy children in a separate study,32 but at that time we did not have the sequence data to match this TRF to a particular species. The abundant and consistent presence of F. prausnitzii in healthy individuals suggests that it is an important member of the mucosal microbiota. F. prausnitzii produces butyrate, formate, and D-lactate as a product of fermentation,33 providing the major energy source for colonic epithelial cells and having effects on epithelial barrier integrity and immune modulation.34, 35 The functional activity of F. prausnitzii in affecting host physiology was recently demonstrated by Li et al,36 where they showed that its population variation was associated with 8 urinary metabolites of diverse structure, including butyrate isoforms. The loss of F. prausnitzii, and thus reduced butyrate production, may trigger gut inflammation in genetically susceptible individuals. The absence of this organism may also leave a gap that can be filled by other organisms resulting in a less stable population.
Using qPCR the abundance of F. prausnitzii was confirmed to be significantly lower in all biopsies collected from ICD patients compared to healthy controls and those with disease location in the colon. Recently, Faecalibacteria and Subdoligranula were found to represent a combined total of 0.4%, 15%, and 26% of bacterial 16S rRNA gene clones from ileal biopsies in ICD, healthy controls, and CCD subjects, respectively.11 Martinez-Medina et al37 also observed F. prausnitzii more frequently in healthy (87%) compared to CD (53%) subjects; however, they did not separate CD phenotypes in their analysis or compare abundance data. Vasquez et al20 detected F. prausnitzii by fluorescence in situ hybridization (FISH) in only 5 of 22 CD patients that had undergone ileal resection, and at low levels (1.8 ± 0.5%) in positive individuals. Additional reports of reduced abundance and diversity of Firmicutes,12, 13 particularly from the Clostridium leptum group, of which F. prausnitzii is a member,38 coincide with our findings; however, F. prausnitzii was not specifically mentioned in those studies.
Swidsinski et al39 recently reported that low fecal levels of F. prausnitzii, along with fecal leukocyte counts, differentiated active CD from UC. Although they did not indicate disease location in CD patients, they found that treatment with high-dose cortisol therapy or infliximab restored F. prausnitzii levels, indicating that this bacterium is being suppressed by the host immune system.39 Intriguingly, all enrolled twins were in endoscopic remission, yet differences in F. prausnitzii were observed in ICD.
Another bacterial population that differentiated according to disease phenotype was E. coli. Increased abundance of E. coli in biopsies from CD patients has previously been reported.8–10, 12 However, our study and 1 other11 identify a particular increase in E. coli levels in subjects with ICD compared to HC. We found a consistent presence of E. coli in CD patients, spanning from the ileum to the rectum. Previously, an increased E. coli abundance in rectum biopsies was found in CD patients.40 Although the E. coli were not isolated and typed in this experiment, E. coli isolated from CD biopsies in previous studies8, 11 have shown pathogen-like behavior in vitro, including survival within macrophages, and may play a role in the inflammatory process.41 Reduced production of α-defensins and reduced antimicrobial activity against E. coli has been observed in ICD but not CCD as compared to healthy mucosa.42 This phenomenon would directly explain the increased E. coli abundance in ICD observed in this and another study.11 Intriguingly, butyrate induces the expression of some antimicrobial peptides43; thus, a reduced abundance of butyrate producing F. prausnitzii may ultimately result in reduced secretion of antimicrobial peptides and proteins allowing E. coli to proliferate.
We were unable to identify any TRFs that clearly differentiated CCD individuals from HC, although there were a few TRFs that were more common in CCD than HC and ICD. However, an increasing number of reports have indicated differences between ICD and CCD phenotypes, including beneficial effects of antibiotic therapy in CCD, but not ICD.44, 45 The reduced effectiveness of antibiotic therapy in ICD has not been explained; however, it may be a result of the differences in microbial composition in these 2 disease phenotypes, and the intracellular localization of E. coli within macrophages.46 When we previously studied fecal samples from these subjects, we observed that community profiles of twins with CCD were similar to those of healthy individuals, while those with ileal involvement separated from the others.22
The consistency of the bacterial composition along the intestinal tract in each individual, ranging from the distal ileum to the rectum, suggests that dysbiosis associated with CD is not localized to the area of disease. This consistency between locations regardless of disease phenotype confirms previous reports in healthy and IBD individuals.19, 37, 47
Reduced bacterial diversity in CD individuals compared to their healthy co-twins agrees with numerous other reports in non-twin populations,12–14, 48, 49 but this is the first finding that specifically points to a lower diversity for a specific disease phenotype (i.e., ICD). Reports of increased bacterial numbers associated with biopsies from CD patients as compared to healthy controls have been inconsistent.47, 50, 51 We did not observe any differences in bacterial numbers associated with disease phenotype; however, this may be due to the nature of the control patients, as individuals related to IBD patients may also have disrupted mucosal surfaces although they do not suffer from disease.52, 53
We acknowledge that these results are subject to biases inherent in PCR amplification and that we may have missed differences in some members of the microbial community; however, biases were consistent in all groups. In this investigation we studied patients who were in remission and not diseased, although the gut microbiota is altered in a diseased compared to a quiescent state.48 Confounding factors including surgery and medication in ICD may also give rise to some skepticism; however, similar surgeries were performed in individuals with CCD (12b and 13a), who had normal levels of F. prausnitzii and an absence of E. coli. Frank et al51 also found that appendectomy was not correlated with the subsequent microbial composition in colonic biopsies of IBD patients.
In conclusion, we identified specific bacterial species that are significantly increased or decreased in abundance in individuals with CD in the ileum compared to individuals with CD in the colon and to healthy individuals. Specifically, we propose that lower amounts of F. prausnitzii and higher amounts of E. coli are indicative of ileal CD and can be used to distinguish the ileal disease phenotype from colonic CD and from healthy individuals. The data from discordant twins indicate that the host genotype is not solely responsible for the observed differences in the gut microbiota, but instead support the hypothesis that ICD and CCD are different subgroups of the disease. Therefore, we propose that the relative abundances of F. prausnitzii and E. coli can serve as indicators to distinguish CD phenotype, and may eventually support customized treatment.
We thank Kerstin Eriksson for assistance with collection of biological material.
- 11Culture independent analysis of ileal mucosa reveals a selective increase in Escherichia coli of novel phylogeny relative to depletion of Clostridiales in Crohn's disease involving the ileum. ISME J. 2007; 1–16., , , et al.
- 16The resident fecal flora is determined by genetic-characteristics of the host — implications for Crohns disease. J Microbiol. 1983; 49: 119–124., , .
- 17The host genotype affects the bacterial community in the human gastrointestinal tract. Microb Ecol Health Dis. 2001; 13: 129–134., , , et al.
- 20Patchy distribution of mucosal lesions in ileal Crohn's disease is not linked to differences in the dominant mucosa-associated bacteria: a study using fluorescence in situ hybridization and temporal temperature gradient gel electrophoresis. Inflamm Bowel Dis. 2007; 13: 684–692., , , et al.
- 30Ecology: From Individuals to Ecosystems. 4th ed. Oxford: Blackwell; 2006: 471–472., , .
- 31Numerical Ecology. 2nd English ed. Amsterdam: Elsevier Science; 1998., .