Plasma bile acids in association with Crohn’s disease

Abstract Background In addition to facilitating lipid digestions, bile acids (BA) are signalling molecules acting on receptors on immune cells and along the gastrointestinal (GI) tract. The aim of this study was to assess if altered bile acid profiles in plasma are associated with Crohn’s disease (CD). Method This cross-sectional study included individuals (aged ≥18 years) referred for colonoscopy at a tertiary centre in Stockholm between 2016 and 2019. All participants received bowel preparation, completed a lifestyle questionnaire and provided blood samples for analysis. During colonoscopy, severity of disease was graded, and biopsies were taken from colonic mucosa. In the current substudy, 88 individuals with CD and 88 age-matched controls were selected for analysis of BA in plasma with ultra performance liquid chromatography (UPLC). Linear regression models were then used to compare mean bile acid concentrations and concentration ratios between CD and controls. Results Individuals with CD had lower plasma concentrations of the majority of secondary BA compared to controls, in total CD/CC ratio 0.60 (SE 0.12), p = 0.001. The most prominent observations were lower levels of deoxycolic acid derivates and lithocolic acid derivates among participants with CD. Moreover, plasma concentration for secondary BA among participants with active CD was significantly lower compared to those with CD in remission, CD active/CD remission ratio 0.65 (SE 0.11), p < 0.002. Conclusion Crohn’s disease may be associated with altered plasma bile acid composition. The significance of colonic bacterial diversity in this context needs to be investigated in further studies. KEY SUMMARY It is known that Crohn’s disease is associated with dysbiosis in the gut microbiota and that primary bile acids are transformed to secondary bile acids by bacterial enzymes in the gut before reabsorbed and transported back to the liver. In this cross-sectional study, Crohn’s disease was associated with lower concentrations of secondary bile acids in blood plasma The findings should encourage further studies the role of the gut microbiome and bile acid metabolism in development of Crohn’s disease and bile acid profile as a biomarker for bowel inflammation.


Introduction
Crohn's disease (CD) is characterised by transmural inflammation and skip lesions which can affect the entire gastrointestinal (GI) tract.It is a severe chronic and relapsing condition that requires lifelong monitoring and symptomatic treatment [1].The pathogenesis and aetiology of CD is unknown but a combination of genetic and lifestyle factors have been suggested [1].Dysbiosis of the microbiota of the GI tract may also play an important role in the pathogenesis of CD as reduced diversity and altered phenotype composition of the gut microbiota compared to healthy subjects have been frequently reported [2,3].
Bile acids (BA) are synthesised in the liver from cholesterol by several enzymatic reactions [4].There are mainly two primary BA produced in the liver, cholic acid (CA) and chenodeoxycholic acid (CDCA).The primary BA are conjugated with glycine (G) or taurine (T) giving rise to bile salts (GCA TCA, TCDCA and GCDCA) [5].
Conjugated BA, that is, bile salts are stored in the gallbladder and secreted into the duodenum, mainly after meals, where they are transformed by bacterial enzymes to secondary BA, for example, lithocolic acid (LCA) and deoxylithocolic acid (DLCA) and tertiary BA, that is, ursodeoxycholic acid (UDCA).Thus, it is believed that gut microbiota plays a crucial role in the composition of BA.
After secretion into the duodenum most BA are then reabsorbed and transported back to the liver with the portal blood (entero-hepatic circulation).A small amount of BA spill over to the systemic circulation and is cleared in urine [5].BA are important for lipid digestion [6]; however, BA may also act as signalling molecules facilitating interplay between the host and gut microbiota through bile acid receptors (BAR) expressed on immune cells and along the GI tract [7].It has been hypothesised that the microbiome dysbiosis observed in inflammatory bowel disease (IBD) may result in decreased enzymatic capacity of the microbiota affecting the ability to metabolise primary BA to secondary BA [7].
The aim of the current study was to identify associations of bile acid profiles to CD and inflammation in the colon compared healthy controls and to investigate these metabolites as possible biomarkers for bowel inflammation.

Study design and setting
The cross-sectional study design included individuals (aged ≥ 18 years) referred for colonoscopy at Danderyd Hospital (a tertiary centre in Stockholm) between 1st November 2016 and 1st July 2019.All participants completed a comprehensive lifestyle questionnaire after treatment with bowel preparation, Movprep®.Before colonoscopy, blood samples were provided for analysis of haemoglobin and C-reactive protein (CRP) and were then stored in a biobank.Biopsies were taken from healthy and diseased colonic mucosa for later DNA-sequencing of colonic microbiota.In the current substudy, 88 individuals with CD and 88 age-matched controls were selected for analysis of BA in plasma with ultra-performance liquid chromatography (UPLC).

Participants
The study flowchart including the number of participants is shown in Figure 1.Out of 2,395 invited individuals, 1,259 (52.6%) agreed on participating.Thereafter, two individuals were excluded due to interrupted colonoscopy.All included individuals (n = 1,257) gave written informed consent before participating in the study.
Within the scope for the current study, there were 141 individuals with CD and 213 individuals with no findings on colonoscopy and no history of colorectal cancer or polyps, that is, 'Clean Colon' (CC).From this group, we selected 88 individuals with age-matching from CD and CC without antibiotic treatment within three months before colonoscopy for analysis.The indication for colonoscopy for the included individuals is shown in Supplementary Table 1.

Exposures
The examining endoscopist diagnosed and evaluated CC and CD, and severity of CD was graded in accordance with simplified endoscopic activity score point system for CD (SES-CD) [8].For subgroup analysis, CD was further divided into 'CD remission' (SES-CD 0-2) and 'CD active disease' (mild (SES-CD 3-6), moderate (SES-CD 5-15), or severe disease (SES-CD > 15).In addition, the CD group was also analysed in subgroups according to history of bowel resection (none/ ileocolic resection/colon or rectum resection/colectomy/ Roux-en-y gastric bypass (RyGB).

Bile acids
Plasma samples were mixed with cold methanol containing IS mix (10 deuterated bile acids), vortexed and then centrifuged.The supernatants were evaporated to dryness and reconstituted with 10% acetonitrile in ammonium acetate solution.Quality controls (QC) of human plasma containing all measured substances were prepared similarly.Six external calibrators containing a mixture of 20 bile acids used for quantification in the range of 50-5000 nM were prepared.
Extracted plasma and standards were separated on an Acquity HSS T3 column using a Waters Acquity UPLC (Ultra Performance Liquid Chromatography) system running a gradient elution with ammonium acetate solution and acetonitrile.
Mass spectrometry (MS) data were collected using a Waters Xevo TQS™ mass spectrometers (triple quadrupole), electrospray ionisation (ESI) in negative ion mode and multiple reaction monitoring (MRM) for 25 known and 38 tentative bile acid.Peak integration was performed using Target Lynx (Waters).The result was reported as the ratio of the sample response (sample peak area/IS peak area) by the QC response.
All 63 BA were analysed in the machine learning algorithms, while only the 25 BA with known structure were analysed in remaining analysis.

The lifestyle questionnaire
Study participants completed a validated 277-item questionnaire [9] to collect information on basic characteristics including data on sex, age and body mass index (BMI), dietary intake, lifestyle data including smoking (smoker/ non-smoker), snuff use (user/non-user), education level and bowel habits (including Bristol Stool Scale (BSS)).A total of 176 (98.9%) of the 178 participants in the study completed the lifestyle questionnaire.Educational level was classified into ≤9 years, 10-12 years and ≥13 years of schooling.Bristol Stool Scale was categorised into the following groups: slow (BSS 1-4), normal (BSS 3-4 only), rapid (BSS 3-7) and varied (BSS both 1-2 and 5-7).Antibiotic use during the past three months was also registered.

Dietary components
Data on nutritional and dietary intake were collected from the lifestyle questionnaire.Average daily intakes (grams/ day) of total fruits, vegetables, whole grains (crisp bread, whole grain bread, oatmeal, other porridge, müsli, quinoa, bulgur, and wheat bran), dairy products, fermented dairy products, red and processed meat, fish, refined grains, and sugarsweetened beverages were estimated.All nutrients were energy-adjusted to the mean energy intake in the study population using the residual method [10,11].The questionnaire was used to calculate alternate Healthy Eating Index (AHEI) [12].The AHEI is a dietary score which includes 10 components, where each of the following items was given a score between 0 and 10 proportionally to dietary intake: (1) vegetables, (2) fruit, (3) whole grains, (4) nuts and legumes, (5) saturated fatty acids, (6) polyunsaturated fatty acids), (7) alcohol consumption, (8) sugar-sweetened drinks and fruit juices, (9) red and processed meat, and (10) sodium.A higher score indicates a healthier diet.

Blood haemoglobin and C-reactive protein
Before colonoscopy blood samples were taken for analysis on haemoglobin and C-reactive protein.Blood samples were then stored in a biobank for further analyses.

Statistical methods
Results are presented as mean ± standard deviation (SD) for continuous variables and proportions for categorical variables.Student's t-test was used for crude group comparisons of continuous variables and chi-square test was used for comparisons of categorical variables.A p-value <0.05 was considered significant when describing the basic characteristics of the cohort.
For association, a univariate linear regression analysis was performed presenting mean ± standard error (SE) for each exposure (CC, CD) and outcome (BA) and ratio between means ± standard error (SE), (CD/CC ratio).After Bonferroni correction [13,14] a p-value <0.002 (0.05/25) was considered statistically significant.Analyses were performed using STATA 16.1 (Stata Corp.) All machine learning algorithms were trained in R version 4.1.2.The following algorithms were run using the caret package version 6.0-93 [15,16] svmRadial, ranger, pcaNNet, knn, glmnet, and xgbTree.Partial Least Squares Discriminant Analysis (PLS-DA) was performed with the mixOmics package version 6.19.1 [17].After replacing zeros in the data set by half of the lowest level, the data were autoscaled using the function prep.autoscale from mdatools version 0.13.0 [18].
The age-matched samples (n = 176) were split 70/30 into train/test set where the train set (n = 124) was used to train each algorithm.In caret, tuning and cross validation was performed using the trainControl function with method = 'repeatedcv' , number = 10, repeats = 10, search = 'grid' , and a tuneLength of 18 in the actual train function.For ranger, the additional arguments importance = 'permutation' and num.trees = 5000 were given.For mixomics, Mfold-validation with 8-fold, nrepeats = 22, dist = 'centroids.dist'was chosen for the perf function.The optimal PLS-DA model was achieved with 19 features in component 1 and 5 features in component 2.
The best performing resulting model of each method was then applied to the test set samples (n = 52), which were never seen by the algorithm during training, to classify these samples as either 'clean colon' or 'Crohn's disease' .The results were summarised in a confusion matrix and the percent accuracy of classification of the test set was calculated as the sum of the correctly classified samples of either class divided by the sum of all classes where 100% indicated that all samples were correctly classified.This percent accuracy of the test set was used to compare the models with each other and the model with the highest accuracy was chosen.

Ethical approval
The research protocol was approved 2016-02-03 by Stockholm ethical review board (2015/2138-31/2) and carried out in accordance with the Declaration of Helsinki of the World Association (1989) (clinical trial number: NCT03302715, www.clinicaltrials.gov).

Study population characteristics
Population characteristics are shown in Table 1.After exclusion of individuals with use of antibiotics within three months prior to the colonoscopy, the final cohort consisted of 88 individuals with CD and 88 individuals with CC.There was no difference in sex, age distribution and BMI between CD and CC.Grading of stools according to the Bristol stool scale showed a significantly higher proportion of individuals with rapid bowel emptying in CD (58%) compared to CC (29%), p < 0.05.
Among individuals with CD, 53 (60%) were in remission while 21 (24%), 14 (16%) and 0 (0%) had mild, moderate, and severe disease respectively at the time for colonoscopy.A larger proportion of CD individuals had previously undergone surgery.
Results from the evaluation of dietary components, AHEI-score and consumption of single dietary components showed similar proportions between groups.

Plasma BA levels
Mean levels of all 25 BA and conjugates with known structure for each exposure (CC, CD) and the ratio between CD and CC (CD/CC ratio) are shown in are shown in Table 2.A summary of the difference in total secondary bile acids is shown in Figure 2.
Primary bile acids: Levels of Cholic Acid (CA) were similar between groups, CD/CC ratio = 1.00 (SE 0.18) and although levels of all other primary BA were higher in CD compared to CC, CD/CC ratio 1.22-2.73,the differences were not statistically significant.In total, the amount of primary BA was increased in CD, CD/CC ratio 1.52 (SE 0.33), however, not statistically significant (Table 2).
Secondary bile acids: Levels of all Deoxycholic acids were lower in CD compared to CC, CD/CC ratio 0.20 − 0.73.The differences were statistically significant for DCA, CD/CC ratio 0.47 (SE 0.11), p < 0.001 and HDCA, CD/CC ratio 0.20 (SE 0.07), p < 0.001.Levels of all Lithocolic acids were significantly lower in CD, CD/CC ratio 0.40-0.52,p < 0.001.There were no statistically significant differences in levels of Hyocholic acids and Ursodeoxycholic acids except TUDCA, CD/CC ratio 0.28 (SE 0.21), p < 0.001 and no differences in levels of MCA, TaMCA and C4.The total amount of secondary BA was lower in CD, CD/CC ratio 0.60 (SE 0.12), p = 0.001, Table 2 and Figure 2.
The ratio between total amount of secondary BA and primary BA (Secondary BA/Primary BA) for was 6.50 (SE 0.51) and 12.05 (SE 1.50) for CD and CC respectively, Table 2.
Subgroup analysis of the statistically significant alterations in secondary BA levels in the main analysis is shown in Tables 3 and 4.
The plasma concentration for secondary BA in CD active disease was significantly lower compared to CD in remission, CD active/CD remission ratio 0.65 (SE 0.11), p < 0.002.Significantly lower levels of HDCA, ILCA and TLCA-3S were shown in CD in remission versus CC and significantly lower levels of DCA, HDCA, LCA, LCA.3S ILCA, GLCA, TLCA and TLCA.3S were shown in CD with active disease vs. CC, (Table 3).
Another subgroup analysis for CD individuals with or without previous ileocolic bowel resection vs. individuals with CC without previous bowel resection is shown in Table 4.
Significantly decreased levels of DCA, HDCA, LCA, LCA.3S and ILCA were shown in non-resected CD vs non-resected CC and significantly decreased levels of HDCA, LCA, ILCA, TLCA and TLCA-3S were shown in ileocolic resected CD vs. non-resected CC.

Machine learning algorithms for identification of potential novel biomarkers
The performance of the seven algorithms used to build classifiers for Crohn's disease diagnosis based on plasma bile acid profiling was evaluated by the percentage of correct classification of a validation test set.The results are summarised in Supplementary Table 1.PLS-DA achieved the highest accuracy of the algorithms tested, 77% and was thus used for further analysis.
The sparse PLS-DA with 19 features in component 1 and 5 features in component 2 showed some separation across the x-axis for Crohn's and clean colon (Figure 3(a)).The features responsible for this separation were extracted and  ranked by importance (Figure 3(b)).Interestingly, only eight out of 19 of the important variables had confirmed structures and were included in the univariate analysis.Thus, the majority of the bile acids contributing to the classification model have tentative structures requiring confirmation.Here, BA41 and BA42 are of particular interest as novel bile acids associated with CD.

Main results
In this cross-sectional study comparing blood plasma concentrations of BA among 176 age-matched individuals with CD and CC, we observed significantly lower concentrations of most secondary BA in individuals with CD compared to controls.The most prominent observations were lower levels of DCA, HDCA, LCA, LCA.3S, ILCA, GLCA, TLCA and TLCA.3S among participants with CD.Furthermore, the plasma concentration for selected secondary BA in active CD disease was significantly lower compared to CD in remission.Although a tendency of higher plasma concentrations of primary bile acids was shown among CD vs CC individuals, the difference was not significant.

Interpretation
Previous studies have investigated concentrations of BA in faeces in individuals with Crohn's disease.A cross-sectional study including 107 patients with UC, 173 with CD and 42 healthy controls showed lower concentrations of the primary bile acid, CDCA, and secondary bile acids LCA, TLCA in faeces in both CD and UC [19].Another study that also included faecal samples only, showed higher abundance of primary BA (CA and CDCA) in IBD patients and lower abundance of secondary BA (LCA and DCA) in CD patients [20].
In line with our results, a small cross-sectional study measuring BA in both faeces and plasma among 42 patients with IBD (12 CD and 30 UC) and 29 healthy controls observed lower proportion of secondary BA in plasma and faeces and a higher proportion of conjugated and sulphated BA in faeces among IBD patients in comparison to healthy controls [21].Another study on BA in plasma including 358 IBD patients and 310 healthy controls and showed decreased levels of LCA but increased levels of HDCA, both secondary BA in CD and UC [22].In a recent study, including 27 treatment-naive individuals with CD, elevated levels of the secondary BA, including DCA, GDCA, TDCA, GLCA-3S and isoLCA were found to be associated with slower disease progress [23].Another recent study including 39 individuals with CD and 14 healthy controls, showed results consistent with our findings.This study reported a decrease in levels of secondary BA, DCA and LCA in both feces and serum of CD patients.Moreover, an increase in primary BA and a reduced ratio of secondary to primary BA were observed in the faces of CD patients [24].Limitations in previous studies on plasma BA in association with CD has for example been lack of data on dietary intake [21][22][23][24] or inclusion of individuals treated with antibiotics [22].Secondary bile acids.data are presented as mean (Se) of ratio of the sample response (sample peak area/iS peak area) by the Quality control response (arbitrary unit) and the ratio between cd and cc (cd/cc ratio).level of significance p < 0.002.cd = crohn's disease.cc = clean colon.* = Wald test.The current study, with data on several potential confounders compared to previous studies, confirms earlier findings in some of previous publications, that is, higher proportion of primary and lower proportion of secondary bile acids in CD in comparison to CC.In agreement with previous studies, the current study also showed lower levels of LCA and DCA in individuals with CD.The novelty that this study adds to previous knowledge is that plasma concentrations for selected secondary BA in CD active disease were significantly lower compared to CD in remission.This strengthens the previously more uncertain association between CD and low plasma concentrations of BA.
To the best of our knowledge, this is the first study to include estimation of dietary intake in relation to CD and plasma BA.There was no statistically significant difference in dietary intake when comparing CD to healthy controls.However, we cannot exclude that small differences in intake exist and that our results were limited by small sample size.
The effect of diet on BA-secretion and the role of BA in lipid digestion is widely known [6,25].Thus, differences in dietary habits between groups is a potential source of bias in previous studies.In addition, no individuals with antibiotic treatment within three months prior to colonoscopy were included in the analysis which reduces potential impact of antibiotics on the gut microbiota [26,27] and consequently also on the BA composition.
A lower content of secondary bile acids in the colon among CD patients can be explained by several mechanisms: First, as the terminal ileum is the primary site for BA reabsorption, impaired reabsorption of secondary BA due to ileitis or ileal resection is a common feature in CD [28] although reduced uptake of BA has also been observed in CD patients with normal ileal histology [29].In this study, a more prominent reduction of secondary BA in CD with active disease was shown which indicates an association between impaired BA composition in plasma and concurrent ileitis.The possible increase of primary BA in plasma among CD individuals, although not statistically significant, could be explained by compensatory de novo BA synthesis due to reduced entero-hepatic circulation.The subgroup analysis comparing non-resected CD, ileocolic resected CD and nonresected CC, showed similar changes in BA levels in both CD groups.This may indicate that the observed results are driven by the disease itself rather than surgery.
Secondly, as the primary BA are metabolised to secondary BA by the gut microbiota, the dysbiosis frequently reported in individuals with IBD [2] could result in differences in BA composition in IBD in comparison to healthy individuals in both faeces and plasma.
An example of the importance of bacteria in this context is that the first step of the formation of secondary BA, LCA and DCA, conducted by bacteria is through bile salt hydroxylase (BSH) and 7-α-hydroxylase.While BSH is found in all major bacterial phyla of the gut microbiota, 7-α-hydroxylase is expressed by specific microbial taxa such as Clostridium and Eubacterium [7].Decreased abundance of Firmicutes, including e.g.Clostridium and Eubacterium rectale have been reported in CD [7], which indicates a possible link between dysbiosis, CD and altered bile acid composition.
In addition to previously mentioned explanatory mechanisms, an increase of secondary BA in plasma has been observed in CD patients after anti-TNFα treatment [30] suggesting the use of BA profile as a marker for disease activity.
Bile acids have multiple biological functions.In addition to facilitating lipid absorption, BA acts as signalling molecules on bile acid-activated receptors (BARs) mainly expressed along the GI tract and on immune cells i.e.Farensoid x-receptors (FXR), Pregnane x-receptor (PXR) [7].FXR is involved in immunological functions such as inflammatory response, maintaining the integrity and function of the intestinal barrier, and regulating the growth of the intestinal microbiota [7].BA conjugates have varying affinity to the receptors.PXR is most strongly activated by LCA and DCA, while FXR is most strongly activated CDCA and CA [31].A disruption in the BA composition could therefore contribute to the immune dysfunction seen in IBD patients [7].(B) ranking of the variables contributing to the cluster separation along the x-axis in a based on their importance, where the higher the positive value on the x-axis indicates higher importance in cc and the lower the negative value the higher the importance in cd.Ba with known structure are presented with their names.tentative Ba are presented as Ba-XX.

Strengths and limitations
A strength of this study is the use of an extensive bile acid targeted platform covering 63 bile acids and derivatives based on biomarker discovery efforts in previous preclinical and clinical studies.Both univariate and multivariate methods have been used to analyse the bile acid data.
Unlike previous reports, our analyses also include an extensive questionnaire with a compliance rate of 98.9%, with detailed data on the participants dietary-, lifestyle-and bowel habits, which together with other objectively collected clinical important data, for example, endoscopic disease activity status and previous surgery, all strengthen the study.
However, there are also some potential flaws in the study.Before the blood samples were taken, all individuals were fasting and underwent bowel preparation with Movprep®.This could impact plasma BA concentrations and hide potential differences in BA between study participants.Also, the individuals in the control group were chosen due to lack of pathological findings in colonoscopy.But since they were referred to colonoscopy for some reason, they cannot be considered completely healthy controls.In addition, there was a slight imbalance between groups in terms of sex and age.While these differences did not reach statistical significance, they have the potential to impact the results.Furthermore, our methodology allowed analysis of 63 different BA but only 25 BA could be identified with a complete structure.Thus, we might have missed other potential differences in BA composition between CC and CD.Finally, the IBD patients in this cohort were not treatment-naive and the effects of medications used to treat IBD are unknown.However, the study population did not include any individuals treated with antibiotics within three months before colonoscopy, which is important since a major impact of antibiotics on the gut microbiota have been shown in several previous studies [26,27].
In summary, in this cross-sectional study assessing plasma BA association with CD, we found significantly lower levels of secondary BA in plasma in CD in comparison to CC.The plasma levels for secondary BA in active CD were significantly lower compared to CD in remission which strengthens the hypothesis that plasma bile acid concentrations are linked to the grade of inflammation in the gut.The findings should encourage further studies the role of the gut microbiome and BA metabolism in the development of CD and BA profile as a biomarker for bowel inflammation.

Figure 1 .
Figure 1.consort diagram of the study population: the main cohort consisted of 1257 individuals including 141 individuals with cd and 213 individuals with cc. for this substudy, 88 age-matched individuals with cd and cc without history of antibiotic treatment within three months prior to inclusion were selected.cd = crohn's disease.cc = clean colon.

Figure 2 .
Figure 2. a summary of the difference in total amount of secondary bile acids.(arbitrary unit on y-axis).

Figure 3 .
Figure 3. Sparse Partial least-Squares discriminant analysis (sPlS-da) of all 63 bile acids measured.(a) two-dimensional clustering of individuals with clean colon (blue) and crohn's disease (orange).(B)ranking of the variables contributing to the cluster separation along the x-axis in a based on their importance, where the higher the positive value on the x-axis indicates higher importance in cc and the lower the negative value the higher the importance in cd.Ba with known structure are presented with their names.tentative Ba are presented as Ba-XX.

Table 1 .
Baseline characteristics for cd and cc.Simplified endoscopic activity score point system for crohn's disease.aHei = alternate healthy eating index.*=grams per day.

Table 2 .
Serum concentration of Ba with known structure.

Table 4 .
Subgroup analysis.Plasma concentration of significant secondary Ba.

Table 3 .
Subgroup analysis.Plasma concentration of significant secondary Ba.Se) of ratio of the sample response (sample peak area/iS peak area) by the Quality control response (arbitrary unit) and the ratio between cd and cc (cd/cc ratio).level of significance p < 0.002.cdrem = cdremission, cdact = cd active disease.cc = clean colon * = Wald test.