Altered Duodenal Microbiota Composition in Celiac Disease Patients Suffering From Persistent Symptoms on a Long-term Gluten-free Diet

Pirjo Wacklin PhD; Pilvi Laurikka; Katri Lindfors PhD; Pekka Collin MD; Teea Salmi MD; Marja-Leena Lähdeaho MD; Päivi Saavalainen PhD; Markku Mäki MD; Jaana Mättö PhD; Kalle Kurppa MD; Katri Kaukinen MD


Am J Gastroenterol. 2014;109(12):1933-1941. 

In This Article


Patients and Study Design

The patient recruitment and sampling were carried out at the Tampere University Hospital and University of Tampere. The enrollment was executed by inviting volunteer long-term treated (GFD ≥3 years), adult (age ≥18 years) celiac disease patients to attend a health survey. Inclusion criteria were that patients were strictly adherent to GFD, and the diagnosis had to be based on the combination of positive celiac disease autoantibodies and subtotal or total small bowel mucosal villous atrophy. In order to minimize confounding factors and to increase the probability of detecting differences in microbiota, only subjects with gastrointestinal symptoms (e.g., indigestion, diarrhea, abdominal pain) as an initial presentation of celiac disease were included.[8] Patients with extraintestinal symptoms or asymptomatic patients at the time of diagnosis were excluded, as well as those with recent or current use of medications that could remarkably affect bowel function, such as antibiotics, opioids, laxatives, or antidiarrheal drugs. All voluntary participants (n=177) underwent a thorough clinical examination, dietary assessment, gastrointestinal endoscopy with duodenal biopsies and measurements of celiac disease serology, and genetics and basic laboratory parameters. After the analyses, subjects with negative celiac antibodies and normal small bowel mucosa (n=164) were selected for further assessment of current gastrointestinal symptoms by Gastrointestinal Symptom Rating Scale (GSRS, see below in detail) and, based on the results, altogether 18 subjects with the highest GSRS total score (Persistent symptoms group) and 18 subjects with the lowest total score (No symptoms group) were chosen for the final study cohort for intestinal microbiota analyses.

Small Bowel Mucosal Biopsies

Upon upper gastrointestinal endoscopy, a minimum of seven forceps biopsy specimens were taken from the duodenum. Three bowel biopsy specimens were freshly embedded in an optimal cutting temperature compound (Tissue-Tec, Miles, Elkhart, IN), snap-frozen in liquid nitrogen, and stored at −70 °C until used for immunohistochemical stainings and for microbial DNA extraction. The remaining biopsies were stained with hematoxylin and eosin and used for morphological analyses. The small bowel mucosal villous height/crypt depth ratio was determined from multiple well-orientated hematoxylin and eosin-stained biopsy samples as previously described in detail.[10] Densities of CD3+ and γδ+ intraepithelial lymphocytes (IELs) were analyzed from the frozen samples with immunohistochemistry as follows: first, 5 μm thick biopsy cuttings were stained with monoclonal antibody Leu-4 (Becton Dickinson, San Jose, CA) for CD3+ IELs and with T-cell-receptor-γ antibody (Endogen, Woburn, MA) for γδ+ IELs. Next, the number of positive IELs was counted with a × 100 flat-field light microscope objective throughout the surface epithelium. A minimum of 30 fields measuring 1.6 mm in epithelial length were calculated. IEL density was expressed as cells in mm of epithelium.[11] The reference values were considered to be 37 cells/mm for CD3+ IELs and 4.3 cells/mm for γδ+ IELs.[11] During the endoscopy, biopsy samples were also taken from the corpus and antrum of the stomach for routine histological assessment and Helicobacter pylori staining.

Clinical and Dietary Evaluation

Besides routine clinical investigations, all patients were interviewed concerning demographic data, initial presentation of celiac disease, duration of GFD, family history for celiac disease, presence of other autoimmune or gastrointestinal diseases, and current use of medications including nonsteroid anti-inflammatory drugs and proton pump inhibitors (PPIs). An experienced dietitian analyzed the strictness of the GFD, uses of purified oats and gluten-free wheat starch, and consumption of fiber (g/day) by means of a 4-day food diary and personal interview. Body mass index was calculated as kg/m2.

The severity of current gastrointestinal symptoms was assessed quantitatively by GSRS, a structured and validated questionnaire widely used in the research of gastrointestinal diseases.[12–14] The survey covers five different gastrointestinal symptoms: indigestion, diarrhea, constipation, abdominal pain, and reflux. The scoring is based on a 7-degree Likert scale that ranges from 1 (minimal symptoms) to 7 (maximal symptoms) points. In our previous studies, the mean GSRS total score has been ~2.7 in untreated celiac disease patients, between 1.8 and 2.0 in patients on GFD, and 1.8 in healthy nonceliac population.[15,16]

Laboratory Parameters and Celiac Disease Genetics

Serum IgA class endomysial antibodies were investigated by an indirect immunofluorescence method using human umbilical cord as substrate as previously described in detail.[17] A titer of 1:≥5 for endomysial antibodies was considered positive. Serum IgA class transglutaminase 2 antibodies were measured by an enzyme-linked immunosorbent assay (Celikey; Phadia, Freiburg, Germany) according to the manufacturer's instructions. Transglutaminase 2 antibody values of ≥5.0 U were considered positive. Blood hemoglobin (reference values: men 134–167 g/l; women 117–155 g/l), serum total iron (reference values 9–34 μmol/l), and erythrocyte folic acid (reference values 200–700 nmol/l) were measured using standard laboratory methods. The presence of celiac disease-associated human leukocyte antigen DQB1*02 and DQB1*0302 alleles (DQ2 and DQ8) was investigated by using either the DELFIA Celiac Disease Hybridization Assay (PerkinElmer Life and Analytical Sciences, Wallac Oy, Turku, Finland) or the SSP DQB1 low-resolution kit (Olerup SSP AB, Saltsjöbaden, Sweden/Qiagen Vertriebs GmbH, Vienna, Austria) according to the manufacturer's instructions. Both tests were available in our laboratory at the time of study and were equally appropriate to test celiac disease genetics.

DNA Extraction

The total DNA was extracted from the frozen biopsy samples using mechanical lysis of bacterial cells combined with the QIAamp Mini Kit (Qiagen, Valencia, CA). Briefly, the biopsy samples were lysed by incubating the sample in ATL lysis buffer with proteinase K overnight at 56 °C and following mechanical lysis with Fastprep instrument (MP Biomedicals, Carlsbad, CA) for 1 min at the level of 6.0 m/s, purified with spin columns, and eluted with 400 μl of buffer AE. The DNA concentrations were determined with Nano-Drop 1000 (Thermo Scientific, Wilmington, DE). The extracted DNA was stored at −20 °C.

The 16S rRNA Gene Pyrosequencing

The bar-coded pyrosequencing method was used to profile duodenal microbiota. The V4–V6 region of the 16S rRNA gene was PCR amplified in three replicates using a bacterial primer pair (F515 5′-TGYCAGCMGCCGCGGTA-3′ and 1061R 5′-TCACGRCACGAGCTGACG-3′). The V4–V6 variable region was chosen for this study based on pretesting results of PCR amplification. In the pretesting, the primers amplifying V4–V6 region showed significantly better amplification (PCR products analyzed by agarose gel electrophoresis) in comparison with the other tested primer pair for V1–V3. Nested PCR protocol would have been necessary for proper amplification with the primer pair for V1–V3 region that would have caused more bias to the results because of larger number of amplification cycles. The PCR products were purified, quantified, and pooled in equal amounts and sequenced using a Genome sequencer FLX Titanium (Roche, Branford, CT) in the Eurofins MWG (Ebersberg, Germany). The raw sequences were trimmed using Mothur v.1.31.2.[18] The sequences that had an averaged quality score of over 25 within a 50-bp window, length of over 250 bases, maximum 2 mismatches to barcode tags and the forward primer, no ambiguous bases, no homopolymers longer than 8 bp, and were nonchimeric according to the Uchime (Robert C. Edgar, were included in the analysis. Two samples with a low sequence gain were excluded from the further analysis. The high-quality sequences were aligned using Greengenes as the reference database (May 2013 release, To avoid the bias of varying sequencing coverage in samples, 960 sequences were randomly subsampled from each sample using Mothur. This subsampled data set (32,640 sequences) was used in further analysis. The sequences were binned into operational taxonomic units (OTUs) applying a sequence similarity of over 97%. In addition, the sequences were classified into bacterial taxa using the Wang approach in Mothur and SILVA, release 111, as a reference database (July 2012),[19] and threshold of certainty over 60%.


In the clinical, serological, and histological analyses, quantitative data were expressed as medians with ranges or means with 95% confidence intervals. Normality of the variables was analyzed by Kolmogorov–Smirnov test and differences between mean values of the study groups were analyzed with independent sample t-test in normally distributed variables or with Mann–Whitney U-test in nonparametric variables. Cross-tabulation with Pearson's χ2 test or Fisher's exact test was used to assess differences between categorical variables.

Inverse Simpson and Shannon microbial diversity indices that reflect the number, abundance, and evenness of species, and species richness (number of detected microbial species) were calculated using R version 3.0.2[20] and its extension package Vegan.[21] Average microbial richness was compared by c2m randomization test with 9,999 permutations using R package rich.[22] The difference in mean diversity and in the relative abundance of taxa between the study groups was assessed by analysis of variance in R. Sequence coverage was estimated using a rarefaction analysis and Good's coverage calculation in Mothur. Redundancy analysis based on Hellinger transformed data and multidimensional scaling based on Bray–Curtis distances were assessed in R and its extension package Vegan. Analysis of variance for redundancy analysis clustering was performed with a full model using 999 permutations in R. Weighted and unweighted Unifrac[23] and hierarchical clustering applying a parsimony method with 1,000 replications was calculated as implemented in Mothur. To exclude the possible confounding effect of other gastrointestinal diseases, all analyses were repeated with a data set containing only samples of the patients (16 in No symptoms group and 9 in Persistent symptoms group) who did not report any other gastrointestinal disease than celiac disease.

Ethical Considerations

The study protocol was approved by the Regional Ethics Committee of Tampere University Hospital District. All study participants gave written informed consent.