Comparison of the Metabolomic Profiles of Irritable Bowel Syndrome Patients With Ulcerative Colitis Patients and Healthy Controls

New Insights Into Pathophysiology and Potential Biomarkers

Ammar Hassanzadeh Keshteli; Karen L. Madsen; Rupasri Mandal; Guy E. Boeckxstaens; Premysl Bercik; Giada De Palma; David E. Reed; David Wishart; Stephen Vanner; Levinus A. Dieleman


Aliment Pharmacol Ther. 2019;49(6):723-732. 

In This Article

Materials and Methods


This cross-sectional study examined adult patients with IBS, ulcerative colitis (UC) in clinical remission and healthy controls. All participants gave informed consent and the study was approved by the Health Research Ethics Board-Biomedical Panel, University of Alberta (Pro00035413) and the Health Sciences Research Ethics Board at Queen's University (DMED-1443-11).

Thirty-nine IBS patients meeting the Rome III criteria[13] were recruited from adult outpatient clinics at a single-centre academic teaching hospital in Kingston, Ontario, Canada. This cohort was previously reported in a diet intervention study[14] and only baseline urine samples were used for the current study.

Fifty-three UC patients in clinical remission were recruited from the IBD clinic at the University of Alberta, Edmonton, Canada. UC diagnosis was confirmed by previously established clinical, radiological and endoscopic criteria as well as histological findings. Patients were included if they were in clinical remission at the time of enrollment (partial Mayo score <3).[15] Subjects were excluded if they had used oral corticosteroids or antibiotics in the previous four weeks or had a history of colectomy.

A group of healthy adult volunteers (n = 21) with no evidence of coronary artery disease, diabetes mellitus, inflammatory or autoimmune conditions was also examined.

Collection of Demographic and Clinical Information

A self-administered questionnaire was used to collect demographic information. Clinical information (eg medications, disease severity) was collected through face-to-face interview and reviewing medical files of participants. IBS severity was assessed using the previously validated IBS symptom severity system (IBS-SSS).[16] The IBS-SSS contains five questions that are rated on a 100-point visual analog scale including the severity of abdominal pain, the frequency of abdominal pain, the severity of abdominal distention, dissatisfaction with bowel habits and interference with the quality of life. The range total IBS-SSS score varies from 0 to 500, with a higher score indicating a worse condition. In UC patients, partial Mayo scoring was used to assess disease activity.

Sample Collection

First morning urine samples were collected from participants in sterile urine specimen cups. After being centrifuged at 4000 g for 10 minutes to remove particulate matter, urine aliquots were stored at −80°C until analysis.

Metabolomic Assays

Urine samples were assayed using a combined direct infusion (DI-)/liquid chromatography (LC-) tandem mass spectrometry (MS/MS) (AbsolutIDQ p180 kit, Biocrates Life Sciences AG, Innsbruck, Austria) and gas-chromatography (GC-) MS assay. All metabolomic measurements were conducted at the Metabolomics Innovation Center (Edmonton, AB, Canada) following a previously described protocol[17] (Data S1).

Gut Microbial Composition

Genomic DNA was extracted from stool samples from UC and IBS patients for sequencing. Details of 16S rRNA sequencing in IBS patients were presented previously.[14] In UC patients, genomic DNA was extracted from stool samples using FastDNA Spin Kit for faeces (MP Biomedicals, Lachine, QC, Canada) and quantified using PicoGreen DNA quantification kit (Invitrogen, Carlsbad, CA, USA). Microbial composition was assessed using Illumina's established 16S rRNA amplicon sequencing method and the MiSeq sequencing platform. No deviations from the manufacturer's protocol were used. Briefly, a segment of the V3 and V4 region of the 16S gene was amplified with gene specific primers (aligning to 341 and 805 bp in the gene) that also include an adapter sequence overhang: Bact_16s_ILL1_341mF 5-TCG TCG GCA GCG TCA GAT GTG TAT AAG AGA CAG CCT ACG GGN GGC WGC AG-3, Bact_16s_ILL1_805mR 5- GTC TCG TGG GCT CGG AGA TGT GTA TAA GAG ACA GGA CTA CHV GGG TAT CTA ATC C-3. This PCR reaction was cycled 25 times and the resulting reaction was purified using bead-based clean-up followed by an 8 cycle PCR reaction using Illumina's proprietary bar-coding primers that also align to the adapter sequence. After a second clean-up the bar-coded libraries were diluted, denatured, pooled and run using a V3 300 bp reagent cartridge on the MiSeq system. Bacterial composition was estimated from the data using Quantitative Insights into Microbial Ecology (QIIME 1.9.1) pipelines.[18] In brief, QIIME was used to analyse for phylogenetic and operational taxonomic unit (OTU). It was used to de-multiplex the barcoded reads and perform chimera filtering. Filtered sequence reads were grouped into OTUs at a sequence similarity level of 97%, which approximates species-level phylotypes. Taxonomy of the OTUs was assigned and sequences were aligned with RDP classifier and Pynast.

Other Laboratory Tests

In UC patients, faecal calprotectin (FCP) was measured in stool samples using an enzyme-linked immunosorbent assay with monoclonal antibodies specific to calprotectin (Bühlmann Laboratories AG, Basel, Switzerland). In addition, C-reactive protein (CRP) was measured in serum samples of UC patients.

Statistical Analysis

Continuous and categorical variables are presented as mean ± SD or median (interquartile range) and number (%) respectively. Kolmogorov-Smirnov test was used to test the normality of the distribution for continuous variables. For normally distributed variables, one-way analysis of variance or Student's t test was used. For non-normally distributed variables Mann-Whitney U test was applied. Categorical variables were compared between two groups using Chi-square test. To explore the correlation between levels of metabolites and severity of IBS, Spearman's rank correlation was used. SPSS version 20.0 software (IBM, Armonk, NY, USA) was used for statistical analysis and P < 0.05 were considered statistically significant.

For metabolomic analysis, metabolites with at least 50% missing values were removed from analysis. Otherwise, missing values were replaced by half of the minimum positive values in the original dataset. Concentrations of urinary metabolites (μmol/L) were normalised to creatinine (mmol/L) and reported as the ratio (μmol/mmol). Multivariate statistical analysis was performed using principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA). For multivariate analysis, identified metabolites were normalised using logarithmic transformation and pareto scaling. Permutation analysis using random resampling (n = 2000) of the two groups of subjects (ie IBS vs UC, IBS vs controls and mixed IBS [IBS-M] vs diarrhoea predominant IBS [IBS-D]) was conducted to determine if the probability of the observed separation was a result of chance or not and a P value was reported. Since age and gender were significantly different between the three groups of participants, we used age and gender adjusted values of metabolites (The residuals of metabolites were computed from linear models with sex and age as independent variables). To identify major metabolites responsible for the discrimination between different groups of participants, variable importance in projection (VIP) plots were generated. The VIP score indicates the contribution of each feature to the regression model. Higher values of VIP scores indicate greater contribution of the metabolites to the group separation. Correction for multiple comparisons was performed by testing the false discovery rate[19] and Q value was reported. Metabolites with VIP scores above 1.5 and false discovery rate (Q value) less than 0.05 were considered to play major roles for the discrimination between two groups of participants. To perform additional validation of metabolomic algorithms, we split the dataset randomly into a discovery (training) set (two thirds of samples) and a validation (test) set (one third of samples). The Least Absolute Shrinkage and Selection Operator technique using 10-fold cross validation was used for variable selection in the regression. Stepwise variable selection using 10-fold cross validation was used to optimise the logistic regression model. To determine the performance of logistical regression models, area under the receiver operating characteristics (ROC) curve (AUC) as well as sensitivity and specificity values were calculated validation set. For metabolomic-related multivariate statistical analysis MetaboAnalyst 3.0[20] was used.

To investigate and visualsze the correlations between metabolites and gut bacterial composition (order level), we used debiased sparse partial correlation (DSPC) algorithm option of the Metscape v3.1.3[21] which is a plug-in for Cytoscape.[22] DSPC is an extension of Gaussian graphical model. Under the assumption that the number of true connections among the metabolites is much smaller than the available sample size, DSPC reconstructs a graphical model and provides partial correlation coefficients and P-values for every pair of metabolic features in the dataset. The results can be visualised as weighted networks where nodes represent metabolites and edges represent partial correlation coefficients or the associated P-values.[21] In the present study, the correlations between gut microbial composition and metabolites were filtered using |r| > 0.3 and subsequently correlation network was built.