Faecal volatile organic compounds differ according to inflammatory bowel disease sub‐type, severity, and response to treatment in paediatric patients

Abstract Background Faecal volatile organic compounds (VOCs) differ with disease sub‐type and activity in adults with established inflammatory bowel disease (IBD) taking therapy. Objective To describe patterns of faecal VOCs in children newly presented with IBD according to disease sub‐type, severity, and response to treatment. Methods Children presenting with suspected IBD were recruited from three UK hospitals. Children in whom IBD was diagnosed were matched with a non‐IBD child for age, sex, and recruitment site. Faecal VOCs were characterised by gas chromatography–mass spectrometry at presentation and 3 months later in children with IBD. Results In 132 case/control pairs, median (inter‐quartile range) age in IBD was 13.3 years (10.2–14.7) and 38.6% were female. Compared with controls, the mean abundance of 27/62 (43.6%) faecal VOCs was statistically significantly decreased in Crohn's disease (CD), ulcerative colitis (UC) or both especially amongst ketones/diketones, fatty acids, and alcohols (p < 0.05). Short‐chain, medium chain, and branched chain fatty acids were markedly reduced in severe colitis (p < 0.05). Despite clinical improvement in many children with IBD, the number and abundance of almost all VOCs did not increase following treatment, suggesting persistent dysbiosis. Oct‐1‐en‐3‐ol was increased in CD (p = 0.001) and UC (p = 0.012) compared with controls and decreased following treatment in UC (p = 0.01). In CD, propan‐1‐ol was significantly greater than controls (p < 0.001) and extensive colitis (p = 0.001) and fell with treatment (p = 0.05). Phenol was significantly greater in CD (p < 0.001) and fell with treatment in both CD (p = 0.02) and UC (p = 0.01). Conclusion Characterisation of faecal VOCs in an inception cohort of children with IBD reveals patterns associated with diagnosis, disease activity, and extent. Further work should investigate the relationship between VOCs and the microbiome in IBD and their role in diagnosis and disease monitoring.


INTRODUCTION
Volatile organic compounds (VOCs) are carbon-based, low molecular mass molecules that contribute to the odour of biological samples.In the gut, VOCs result mainly from the metabolism of the gut microbiota and the intestinal mucosa and their abundance changes in intestinal disease. 1 In inflammatory bowel disease (IBD), a loss of gut microbial diversity is correlated with disease severity and with greater derangement in Crohn's disease (CD) than ulcerative colitis (UC). 2 Faecal VOCs are relatively stable over time within and between individuals despite day-to-day variations in diet and are not affected by optimized freezing/storage of stool samples. 3In adults, most of whom were taking medication at the time of sampling, faecal VOCs differ between diarrhoea-predominant irritable bowel syndrome and active CD and UC, between Crohn's colitis and UC 4 and between active and inactive CD and UC. 5 In small studies in children, faecal 6 and urinary 7 VOCs were altered in active IBD.
We characterized faecal VOCs in children newly presenting with suspected IBD.The objectives were to compare IBD with controls and assess how faecal VOCs were related to disease sub-type, disease severity, the region of bowel affected and response to treatment in IBD.We also consider how characterisation of faecal VOCs may provide insights into pathogenesis in IBD.

Study design and settings
We undertook a prospective, case-control study of children attending paediatric gastroenterology clinics in Alder Hey Children's Hospital, Liverpool; Bristol Royal Hospital for Children, Bristol; and Birmingham Children's Hospital, Birmingham, UK.

Screening, recruitment, sample collection and clinical management
Children with suspected IBD were identified from referral letters and asked to bring a stool sample when attending the clinic (Table S1 and Supplementary Methods).If suspected IBD was confirmed following clinical assessment, a research nurse provided written and verbal information about the study.Children were excluded if they had an established diagnosis of IBD or another significant intestinal disorder or had already started treatment for IBD including exclusive enteral nutrition.Signed, informed consent was secured from young people aged 16 years, or the parent or guardian in younger children.

Clinical diagnosis was assessed at 3 months follow-up by review
of medical records and a further stool sample requested in those with IBD.IBD was diagnosed, 8 managed, 9,10 and disease distribution 11 and activity 9 assessed following established clinical guidelines.Similarly, appropriate investigations were performed to diagnose non-IBD disorders (e.g., endoscopy and small intestinal biopsy in coeliac disease (Supplementary Methods)).Children in whom a diagnosis other than IBD was made served as controls.To maximise relevance to clinical practice, each IBD case was matched by age (�6 months), sex and recruitment site to a non-IBD control.Children in whom the diagnosis of IBD could not be either confirmed or refuted at follow-up were excluded from the analysis.

Data extraction
Demographic, clinical and laboratory information was extracted from medical records and collected by questionnaire.For diet, we asked for major dietary modifications (e.g., vegetarian diet).

Sample storage and analysis
Stool aliquots were stored at −80°C and shipped frozen to the University of Liverpool.Faecal VOCs were characterized by gas chromatography-mass spectrometry (GC-MS).Faecal calprotectin (FC) was measured at each hospital according to their usual practice (Supplementary Methods).Laboratory staff were blinded to the patient's diagnosis and response to treatment in those with IBD.

Sample size calculation
This study analysed all case/control pairs from a study evaluating an electronic nose in the differentiation of IBD from other gastrointestinal disorders (to be reported separately), which aimed to recruit a total of 286 children with suspected IBD.T-tests and one-way analysis of variance were used to compare VOC abundance according to disease severity, distribution and response to treatment.To maximise the number of observations, we combined UC and IBD-unspecified (IBD-U) for the analysis of disease severity and distribution.A paired t-test was used to compare IBD baseline and follow-up samples (online software tool Metaboanalyst version 5.0, https://www.metaboanalyst.ca).Logistic least absolute shrinkage selection operator (LASSO) was performed using the R package to assess the association between VOCs abundance and FC at baseline and dietary therapy in CD at follow-up.

Data processing
Given the similar clinical presentation and response to treatment in children with IBD in both sexes, 12 we have not reported findings disaggregated by sex.In this exploratory study, p < 0.05 was accepted to show statistically significant differences with no correction for multiple comparisons.

RESULTS
Between June 2017 and June 2020, 432/616, children attending gastroenterology clinics with suspected IBD were recruited (Figure 1).Twenty-six children were withdrawn, mainly because of inability to obtain a stool sample.In the remaining children, 132/140 children diagnosed with IBD and with sufficient stool for VOC analysis were matched with a non-IBD control.Demographic, clinical, and environmental variables were broadly similar amongst IBD and non-IBD children and IBD sub-types although there was a greater proportion of South Asian children amongst the IBD cases and those recruited in Birmingham (Table 1 and Tables S2   and S3).Amongst the IBD cases, 78/132 (59%) had CD, 38 (29%) had UC and 16 (12%) had IBD-U (Table S4).Common diagnoses amongst the non-IBD controls were functional gastrointestinal disorders (68, 51.5% children: 25 functional abdominal pain, 24 irritable bowel syndrome, 19 functional constipation) and coeliac disease (7; 5.3%).Across all samples, 62/152 identified faecal VOCs remained after filtering for sparse features.Ketones (n = 13), fatty acids (11), and aldehydes (11) were the most common VOCs identified (Table S5).
There was no clustering for hospital in a principal component analysis of VOC abundances (Figure S1).VOC abundances were not significantly different between hospitals (Kruskal-Wallis, p > 0.05).The numbers of VOCs detected were significantly lower in samples recruited from Alder Hey Children's Hospital (median 37, IQR 11) compared to those recruited from Bristol Children's Hospital only (median 40, IQR 10, Kruskal-Wallis, p = 0.034) (Table S6).Collectively, these results suggest minimal differences in VOCs between hospital sites.

Faecal VOCs in Crohn's disease
The number of VOCs detected in children with CD (median 39, IQR 13) and matched controls (median 40, IQR 11) was similar (Wilcoxon signed-rank test p = 0.282).However, the CV was higher for CD (28%) than controls (20%; Figure 2a).5/62 (8.1%) VOCs had higher 1 Children who had intended to provide a stool sample were withdrawn if an inadequate or no sample was provided.and 9 (14.5%) had lower mean abundance in CD compared to controls (p < 0.05; Figure 2b; Table S7).
Although the number of VOCs did not differ significantly according to disease distribution (Figure 2a; Table S4, Kruskal-Wallis test p = 0.090), the abundance of 8 VOCs varied according to distribution with 6 VOCs associated with ileal (LI) disease (Figure 2d, Table S10; Fifty-three children provided a stool sample after 3 or more months (Table S4).The mean number of VOCs at follow-up (38, SD 11; paired t-test p = 0.42) and the variability in VOC numbers (CV = 25%) were similar to baseline (mean 37, SD 9, CV = 25%).The abundance of 4-ethylphenol was significantly increased (paired t-test p = 0.02), but four VOCs significantly reduced at follow-up (Table S11; p < 0.05).
Propan-1-ol, phenol and ethanol significantly increased at baseline compared to controls (paired t-test p ≤ 0.001, <0.001, 0.002, respectively, Table S7), subsequently fell to levels similar to, or below, that of controls post-treatment (Figure 2e).Propan-1-ol reduced regardless of disease activity at follow-up and phenol showed a reducing trend in remission (p = 0.12; Figure 2e; Table S11).
At follow-up, 19 (35.8%) children with CD were receiving nutritional therapy; some ketones, medium chain fatty acids, and an alcohol had a weak negative correlation with dietary therapy (Table S12).
The abundance of 11 (17.7%)VOCs varied significantly with disease extent, with roughly half associated with distal disease and half with extensive disease (Figure 4b; Table S15; p < 0.05).
In the follow-up of 27 children with UC, the number of VOCs (mean 34, SD 12, paired t-test, p = 0.96) and the variability of VOC numbers (CV = 35%) was similar to baseline (mean 34, SD 12).
Oct-1-en-3-ol was significantly increased in UC compared to controls at baseline (p = 0.012; Table S13) and decreased to levels similar to controls for active disease and remission at follow-up (Figure 5).Similarly, propan-1-ol tended to be higher in cases than controls at baseline (p = 0.054; Table S13) and at follow-up decreased significantly in remission (10 pairs, paired t-test, p = 0.046) but not in active disease (17 pairs, paired t-test, p = 0.41; Figure 5).The abundance of phenol fell significantly from baseline and was lower in abundance at follow-up even in children with active disease (Paired t-test p = 0.02; Table S16).No children with UC were receiving therapeutic feeds at follow-up.

Faecal VOCs in Crohn's disease versus ulcerative colitis at baseline
The number of VOCs was higher in CD (median 39, IQR 13) than in UC (median 35, IQR 18; Wilcoxon rank sum test p = 0.013; Figure 6).
The abundance of 15 VOCs, mainly ketones and fatty acids, was higher in CD, whereas 3-methylbutanal was increased in UC (Student's t-test, p < 0.05, Table S17).The number in each dietary category was similar in each group (P = not significant).

Faecal VOCs in small versus all large bowel disease
Twenty-four VOCs were increased in the small bowel and oct-1-en-3-ol was increased in the large bowel disease (Student's t-test p = 0.009; Table S18).Figure 7 represents VOCs which altered across the chemical classes.
A summary of the VOCs that differed significantly in abundance according to clinical parameters in IBD is shown in Table S19.

DISCUSSION
This is the largest GC-MS study of faecal VOCs in an inception cohort of children with IBD and matched controls with non-IBD gastrointestinal disorders.There were clear differences between IBD and controls in the number of VOCs in UC, and in the abundance of many VOCs in both UC and CD.Some of these differences were associated with disease activity and, to a lesser extent, disease distribution, and also with response to treatment in IBD cases.The sources of VOCs highlighted in our study and evidence from other studies of IBD are summarised in Table S20.

Differences between IBD and matched controls at baseline
The mean number of faecal VOCs was significantly lower in UC than in controls.The abundance of 12 VOCs was significantly lower in UC, three in CD and six in both CD and UC.Given that the metabolism of the gut microbiota is the main source of faecal VOCs, 2 these findings are consistent with the reduced microbial diversity and dysbiosis reported in IBD in adults 2 and children. 13

Disease severity
At baseline in UC, both the number and abundance of VOCs were reduced in more severe disease.In both CD and UC, several fatty acids were less abundant in more severe disease.Fatty acids were similar in IBD and controls at baseline, indicating that the change in these compounds may be related to disease severity; a decrease in SCFA-producing bacteria with increased severity in new-onset paediatric UC has been reported previously. 16In CD, differences in short and medium chain fatty acids have been observed between active and inactive cases 17 but the stepwise reduction with increasing disease severity that we have reported for several VOCs is a new finding.We observed lower abundance of BCFA in UC and CD and

Disease distribution in IBD
Studies evaluating faecal VOCs according to disease distribution are scarce.The abundance of 25 VOCs differed significantly between small and large bowel disease, with 24 VOCs (predominantly ketones/diketones, fatty acids, some aldehydes and aromatic compounds) increasing in small bowel disease.In contrast, the alcohol oct-1-en-3-ol was increased in large bowel disease.Previous metabolomic studies were able to distinguish between predominantly ileal and colonic CD and indicated that metabolites of fatty acids, bile acid, tyrosine and phenylalanine biosynthesis may be of importance in the pathogenesis of CD. 19 Recently, Notararigo et al. reported Faecal VOCs in colitis at baseline (UC and IBD unclassified combined).(a) VOCs which altered significantly with disease activity (as in Figure 3a) in colitis at baseline (one-way ANOVA, p < 0.05); results of statistical comparisons are listed in Table S14.(b) VOCs which altered significantly with disease distribution (as in Figure 3a) at baseline (one-way ANOVA, p < 0.05); results of statistical comparisons are listed in Table S15.IBD, inflammatory bowel disease; ANOVA, analysis of variance; UC, ulcerative colitis; VOCs, volatile organic compounds.

788
-UNITED EUROPEAN GASTROENTEROLOGY JOURNAL higher levels of homoserine-methionine and isobutyrate (2-methylpropanoic acid) in serum in ileocolonic CD. 20 The abundance of specific VOCs may have clinical utility in identifying disease distribution in IBD.

Differentiating IBD sub-type
Strikingly, 17 VOCs differed significantly in abundance between CD and UC, with most increased in CD.These were mainly ketones/ diketones, fatty acids and aromatic compounds.The diketone 6methylhept-5-en-2-one and the aromatic compound 4-methylphenol were lower in abundance in UC than controls at baseline and in severe and extensive colitis and may be particularly helpful in differentiating disease subtypes.Furthermore, 3-methylbutanal was increased in UC compared with CD and was not significantly associated with any other comparisons; therefore, it may also be a useful marker to distinguish between subtypes.

Response to treatment in IBD
The reduced number and abundance of many VOCs at baseline persisted at follow-up, suggesting persistent dysbiosis despite clinical improvement in many children.In paediatric IBD in the Netherlands, treatment did not result in microbiota recovery to levels in healthy controls. 21The weak associations between VOCs and dietary therapy in CD suggest that the changes in VOCs following treatment reflect differences in disease status rather than a change in gut flora resulting from nutritional therapy.Changes in specific VOCs to levels found in controls may have utility as treatment targets in IBD.

Specific VOCs of interest
The difference in abundance compared to controls at baseline, association with disease severity, distribution and sub-type and change in abundance following treatment support the role of some specific VOCs in pathogenesis and monitoring response to treatment.
The abundance of the alcohol propan-1-ol was significantly increased in CD and possibly also in UC (p = 0.054) and with more extensive disease in UC.Levels reduced to those similar to controls after treatment in CD and showed a similar trend in UC in remission.
Resulting from the degradation of the amino acid threonine by Escherichia coli, other Enterobacteriaceae 22 and Clostridium sp, 23 propan-1-ol is considered to be damaging to the gut. 24Found to be related to active IBD previously, 25 propan-1-ol is a promising compound in breath, faeces and urine for both the diagnosis and monitoring of IBD. 4,25,26other alcohol, oct-1-en-3-ol, was notable as the only VOC in greater abundance in UC and was also raised in CD and in large bowel disease at baseline.Post-treatment, oct-1-en-3-ol reduced significantly in UC but not in CD.In a study in adults, Ahmed et al.
also reported raised oct-1-en-3-ol in CD. 5 We also observed that ethanol was in significantly greater abundance in CD and severe colitis at baseline and fell following treatment in CD.8][29] Although Ott et al. reported a greater diversity in fungi in colonic biopsies in CD compared with controls, 30 CD in clinical remission was associated with greater fungal diversity in faeces where fungal profiles clustered with Candida spp. 31 Our findings support further research regarding the role of intestinal fungi in IBD.
Phenol was in greater abundance in CD and fell with treatment to levels seen in controls in both CD and UC, with levels tending to be lower in IBD cases that achieved remission.Increased phenol abundance in stools has been reported in adults with CD, UC and also irritable bowel syndrome. 25Phenol arises from the microbial degradation of tyrosine and tryptophan 24 and is produced by many intestinal bacterial species including Enterobacteriaceae and Clostridium clusters I, XI, and XIVa. 32Phenol is considered to be damaging to the gut 24 exhibiting cytotoxicity and increased paracellular permeability in vitro and impaired integrity of the intestinal epithelium and the viability of intestinal epithelial cells. 33,34

Strengths and limitations
Recruiting children with non-IBD gastrointestinal disorders as the comparison group, rather than healthy controls, increase the relevance of our findings to clinical practice.The matching of the case and controls by age, sex and recruitment site and the similarity in demographic variables, including diet, and exposure to antibiotics amongst the cases and controls, suggest that the differences in faecal VOCs were due to disease status rather than environmental factors influencing the gut microbiota.The recruitment of children at initial presentation excluded the effect of IBD drug treatment on VOCs, which may occur in studies of established disease.
We were not able to test our findings in a validation cohort.Some children did not provide follow-up stool samples, limiting our ability to evaluate the effects of response to treatment.Although we have identified faecal VOCs that may cause or be a consequence of intestinal damage, further work is needed to quantify these compounds to assess their importance in IBD pathogenesis.Using a single assay for FC with a wide range of measurements may have improved our assessments of the association between VOCs and disease severity.Aligning metabolomic studies and therapeutic trials would help explain how VOCs are regulated and hold the potential for developing a biomarker to predict clinical response to specific IBD therapies as we have proposed for irritable bowel syndrome. 35ture work should also investigate the source of the VOCs of interest using genomics tools.
A further replication cohort would add strength to the current work.Future studies might target VOCs that changed most with disease activity and might serve as biomarkers, and the role of fungi in UC, given our findings regarding octen-3-ol.Finally, longer followup to assess how VOCs change between remission and later relapse will strengthen the associations that we have reported.

�
the established knowledge on this subject � The pathogenesis of inflammatory bowel disease (IBD) remains poorly understood.� Investigations for the diagnosis and monitoring of IBD in children can cause distress and delay the start of treatment.Measurement of volatile organic compounds (VOCs) in stool is simple and inexpensive.What are the significant and/or new findings of this study?� In 132 IBD/non-IBD gastrointestinal diseases matched pairs, 27/62 (43.6%) faecal VOCs were statistically significantly decreased in Crohn's disease (CD), ulcerative colitis (UC) or both.� Propan-1-ol was increased in CD and extensive colitis and decreased with treatment.Phenol was increased in CD and fell with treatment in CD and UC.Oct-1-en-3-ol was increased in CD and UC and fell with treatment in UC. � Measurement of faecal VOCs provides insights into disease pathogenesis and has potential for the non-invasive diagnosis and monitoring of IBD.BELNOUR ET AL.

F I G U R E 2
Faecal VOCs in CD.(a) Number of VOCs at baseline in matched controls and CD according to disease activity (defined by wPCDAI) and distribution (defined by the Paris classification: L1-distal 1/3 ileum with or without limited cecal disease; L2-colonic disease; L3 -ileocolonic disease; C = Control).(b) Abundance of VOCs at baseline that differed significantly between CD and matched controls.(c) VOCs which altered significantly with disease activity (as in panel (a)) in CD at baseline (one-way ANOVA, p < 0.05); results of statistical comparisons are listed in Table S8.(d) VOCs which altered significantly with disease distribution (as in panel (a)) at baseline (one-way ANOVA, p < 0.05); results of statistical comparisons are listed in TableS10.(e) Control (n = 53), baseline (BL, n = 53), active at follow-up (AFUP, n = 29), remission at follow-up (RFUP, n = 24).Propan-1-ol, phenol and ethanol were significantly raised in CD compared to control at baseline and reduced posttreatment (p < 0.05, paired t-test).ANOVA, analysis of variance; CD, Crohn's disease; VOCs, volatile organic compounds.IBD, four VOCs were in greater abundance in CD than in controls (three alcohols and phenol).One of these alcohols, oct-1-en-3-ol, was also in greater abundance in UC suggesting an increase in some bacterial taxa in IBD14 or the presence of fungi which frequently produce 8-carbon compounds.15F I G U R E 2(Continued)

also 4 -
methylphenol and phenol in UC, all products of amino acid fermentation.BCFAs are produced mainly by Bacteroides and Clostridium fermentation of branched-chain amino acids (valine, leucine and isoleucine18 ).De Preter et al. also noted a reduction in the aromatic compounds 4-methylphenol and methyl-indole in active UC.17 These findings are consistent with a change in the metabolism or composition of the intestinal microbial community in active IBD.Although not associated with disease activity in CD, the positive association between phenol and FC at baseline is consistent with the reduction in phenol abundance with clinical improvement at followup.6-methylhept-5-en-2-one was lower in more severe colitis and also negatively associated with FC.These two VOCs may have a role in monitoring disease activity.F I G U R E 3 Faecal VOCs in UC.(a) Number of VOCs at baseline in matched controls and UC according to disease activity defined by PUCAI and distribution defined by the Paris classification: E1E2 = left-sided colonic inflammation, E3E4 = more extensive colonic inflammation; C = Control.(b) Abundance of VOCs at baseline that differed significantly between UC and matched controls.PUCAI, paediatric ulcerative colitis activity index; UC, ulcerative colitis; VOCs, volatile organic compounds.
Characterisation of faecal VOCs in IBD provides insights into the pathogenesis that may inform the development of novel interventions.Despite the challenges often faced in the F I G U R E 7 Faecal VOCs comparisons of small and large bowel disease.In total, 25 VOCs were significantly different (Student's t-test); a subset of 12 are shown to represent VOCs which altered across the chemical classes.VOCs, volatile organic compounds.
Demographic variables and diet according to diagnosis.