Hepatobiliary phenotype of individuals with chronic intestinal disorders

Despite the known functional relationship between the gut and the liver, the clinical consequences of this circuit remain unclear. We assessed the hepatobiliary phenotype of cohorts with celiac disease (CeD), Crohn´s disease (CD) and ulcerative colitis (UC). Baseline liver function tests and the frequency of hepatobiliary diseases were analyzed in 2377 CeD, 1738 CD, 3684 UC subjects and 488,941 controls from the population-based UK Biobank cohort. In this cohort study associations were adjusted for age, sex, BMI, diabetes, and alcohol consumption. Compared to controls, cohorts with CeD, but not CD/UC displayed higher AST/ALT values. Subjects with CD/UC but not CeD had increased GGT levels. Elevated ALP and cholelithiasis were significantly more common in all intestinal disorders. Non-alcoholic steatohepatitis and hepatocellular carcinoma (HCC) were enriched in CeD and CD (NASH: taOR = 4.9 [2.2–11.0] in CeD, aOR = 4.2 [1.7–10.3] in CD, HCC: aOR = 4.8 [1.8–13.0] in CeD, aOR = 5.9 [2.2–16.1] in CD), while cholangitis was more common in the CD/UC cohorts (aOR = 11.7 [9.1–15.0] in UC, aOR = 3.5 [1.8–6.8] in CD). Chronic hepatitis, autoimmune hepatitis (AIH) and cirrhosis were more prevalent in all intestinal disorders. In UC/CD, a history of intestinal surgery was associated with elevated liver enzymes and increased occurrence of gallstones (UC: aOR = 2.9 [2.1–4.1], CD: 1.7 [1.2–2.3]). Our data demonstrate that different intestinal disorders predispose to distinct hepatobiliary phenotypes. An increased occurrence of liver cirrhosis, NASH, AIH and HCC and the impact of surgery warrant further exploration.


Methods
Population-based UK biobank participants. The 'UK biobank' (UKB) is a population-based cohort study built up in the United Kingdom from 2006 to 2010. In this period approximately 500,000 individuals from across the United Kingdom, aged 37 to 73 years at baseline, were recruited and registered with the UK National Health Service. At baseline visit, all participants gave informed consent for genotyping and data linkage to medical reports. They provided socio-demographic and clinical data, blood samples and physiological measures in an initial examination, which was the basis for our study. ICD-10 codes (international classification of diseases, 10th revision) were obtained from medical records from the year 1996 on to identify diagnoses. Participants with chronic hepatitis B or C, pathological alcohol consumption (> 60 g/d in men or > 40 g/d in women) or coexistence of IBD/CeD (n = 578) were excluded (in total n = 5771, Supplementary Fig. 1).
2377 individuals with celiac disease (CeD), 3684 with ulcerative colitis (UC) and 1738 with Crohn's disease (CD) were included in our study. We compared liver enzymes in the blood as well as liver-related ICD-codes between cohorts with CeD, UC, CD and controls. For each disease entity, we compared cohorts with and without cirrhosis to assess the underlying risk factors. Bowel resection was defined as operation codes 1464, 1459, 1461, 1462, and 1465. The presence of the following primary ICD10 codes was evaluated: Celiac disease (K90.0), ulcerative colitis (K51.0-9), Crohn's disease (K50.0-9), cirrhosis (K74.6), non-alcoholic steatohepatitis (NASH) (K75.8), chronic hepatitis (K73), primary liver cancer (C22.0), cholelithiasis, cholecystitis (K80, K81) and AIH (K75.4). The study has been approved by the UKB Access Committee (Project #47527). The manuscript is based solely on the analysis of pseudonymized data obtained from the UK Biobank Resource under Application Number 47527. The authors were not in contact with the described individuals nor had they access to their personal data. The data were reported as described by the STROBE guidelines.
Statistical analysis. Kolmogorov-Smirnov-test was used to assess normal distribution. All continuous variables were analyzed by unpaired, two-tailed t-tests or Mann-Whitney U test, and by a multivariable model to account for relevant confounders. As a result, all these variables were shown as mean ± standard deviation (normal distribution) or median [IQR] (non-normal distribution). All categorical variables were displayed as relative (%) frequencies and the corresponding contingency tables were analyzed using the Chi-square test.
All analyses were adjusted for age, sex, BMI, presence of diabetes mellitus, and mean alcohol consumption via multivariable logistic or linear regression. Odds ratios (ORs) were presented with their corresponding

Results
Characterization of study cohort. Among 497,404 participants in the UK biobank, we identified 2377 individuals with CeD, 1738 with CD and 3684 with UC ( Supplementary Fig. 1). The CeD or UC cohorts were slightly older than controls and the CD cohort. 65% of individuals with CeD and 48% of the UC cohort were female compared to 54% of controls (Table 1). Participants from all disease subgroups reported lower average alcohol consumption than controls. Individuals with CeD and CD had a lower average BMI than controls (Table 1).

Serum liver enzyme concentrations in chronic intestinal disorders.
Mean alanine aminotransferase (ALT) and aspartate aminotransferase (AST) concentrations in the CeD cohort were significantly higher than in controls (Supplementary Table 1, Fig. 1A Fig. 2). While the CD or UC cohorts were also more likely to have AST elevations above the upper limit of normal than controls (Supplementary Table 1), the corresponding odds ratios  Fig. 2D). In the vast majority, only mild elevations of AST/ALT/ALP were seen (i.e. ≤ 2 × ULN), whereas moderately elevated GGT levels (i.e. ≥ 2 × ULN) were detected in 4-6% of all individuals and were more common in the UC/CD cohorts (Supplementary Table 2). While the number of patients with elevated total serum bilirubin was comparable in all groups, serum albumin concentrations were significantly lower in all disease cohorts with lowest concentrations seen in the CD cohort. (Supplementary Table 1, Fig. 1E,F).
Liver enzyme serum concentrations in selected subgroups. Next, we analyzed factors associated with elevated liver enzymes. In all investigated intestinal diseases, females more often displayed elevated transaminase serum concentrations than males. Individuals with IBD who had a BMI > 30 kg/m 2 more frequently demonstrated elevated AST/ALT levels than their non-obese counterparts. Somewhat surprisingly, diabetes was associated with elevated AST/ALT values in the UC cohort, but not in the CD or CeD cohorts (Fig. 3). Individuals who had diabetes, were obese or of higher age, and more frequently displayed elevated GGT serum concentrations irrespective of the underlying intestinal disease (Fig. 3). Females and participants aged 50 years  www.nature.com/scientificreports/ or older often presented with elevated ALP concentrations, whereas the impact of diabetes and obesity was less evident (Fig. 3). Notably, the observed changes reflected mostly the alterations seen in the control group (Fig. 3).

Liver-related diagnoses in cohorts with chronic intestinal disorders.
To determine, whether the differences in liver-related parameters translate into clinically relevant diseases, we analyzed the occurrence of the most relevant hepatobiliary ICD codes in the described cohorts (Table 1). Among the hepatic disorders, individuals from all disease subgroups more frequently displayed cirrhosis, chronic hepatitis or autoimmune hepatitis (Table 1, Fig. 4). While the odds ratios for developing cirrhosis ranged between 3 and 4 compared to controls, even higher odds were seen for chronic hepatitis (ORs between 4 and 9) and autoimmune hepatitis (ORs between 5 and 8).
Among the biliary diseases investigated, cholelithiasis was more common in all disease subgroups compared to controls and was particularly frequent in the CD cohort (Table 1, Fig. 4D). As another evidence highlighting the importance of small bowel in the pathogenesis of gallstone formation, cholelithiasis was nearly twice as common in a CD subcohort with isolated small bowel affection compared to the subcohort with isolated affection of the colon (Supplementary Table 3). Notably, none of the other assessed parameters differ between the subcohorts (Supplementary Table 3). Cholecystitis was much less common, but displayed a similar distribution pattern. In line with the published data, the ICD code cholangitis (that includes primary sclerosing cholangitis) was most common in UCs, but also significantly overrepresented in CDs when compared to controls (Table 1, Fig. 4E). Likely due to its dismal prognosis, the occurrence of cholangiocarcinoma (CCA) in UK participants was very low and when compared between the subgroups, the ICD code was significantly elevated only in the CD cohort, (CD: OR = 3.26[1.04-10.19]; p = 0.042, Table 1). In contrast, the diagnosis of HCC was significantly more common in the CD and CeD cohort, but not in UC participants (CeD: aOR = 4.79[1.77-12.96]; CD: aOR = 5.93 [2.20-16.05]; both p < 0.01) (Fig. 4F, Table 1).

The association of bowel resection with liver phenotypes in cohorts with UC and CD.
Intestinal resection constitutes an established therapeutic strategy for complications of CD as well as for refractory UC. Therefore, we examined the association between previous intestinal surgery and liver enzyme levels as well as the occurrence of biliary diseases. As expected, bowel resection was less common in the UC than in the CD cohort (32% vs. 10%) and associated with lower BMI and lower reported alcohol consumption (Tables 2 and 3).

Figure 2.
Odds ratios to present with elevated serum ALT, AST, GGT or ALP levels in UK Biobank cohorts with celiac disease, Crohn's disease, and ulcerative colitis compared to controls. Adjusted odds ratios (OR) with their corresponding 95% confidence intervals (CI) are shown for alanine aminotransferase (ALT; A), aspartate aminotransferase (AST; B), gamma glutamyl transferase (GGT; C) and alkaline phosphatase (ALP; D). The risk to display levels higher than the corresponding sex-dependent upper limit of normal (ULN) was compared to the respective controls. Odds ratios were adjusted for age, sex, BMI, alcohol consumption and diabetes mellitus.

Subjects with cirrhosis.
Since cirrhosis is the major cause of hepatobiliary mortality, we took a closer look at all individuals with this diagnosis. Although the majority of cirrhotics in UK biobank were male, 64% of CD cirrhotics were female (Table 4). Accordingly, females with CD displayed a particularly high odds for HCC compared to female controls (aOR = OR = 7.79 [3.98-15.23]; p < 0.001). Liver enzymes in serum did not significantly differ between cirrhotics with and without the analyzed intestinal disorders ( Table 4). The ICD code cholangitis was markedly more common in UC and CD cirrhotics (UC: OR = 37.69 [15.69-90.55]; p < 0.001; CD: OR = 7.37 [1.43-37.94]; p = 0.017), while the diagnosis autoimmune hepatitis was significantly more frequent in UC cirrhotics. Finally, the diagnosis chronic hepatitis was more common in CeD cirrhotics (Table 4).
To further characterize the factors associated with the development of cirrhosis in different intestinal disease entities, we compared cirrhotics with non-cirrhotics. Among the CeD cohort, cirrhotics more frequently displayed NASH (OR = 98. 29 Table 6). In line with the prominent role of NASH, the CD cirrhotics more frequently had diabetes and were obese (Supplementary Table 6). The analysis of the control cohort reflected the well-established factors associated with development of cirrhosis, i.e. higher age, male sex, alcohol consumption obesity, diabetes mellitus as well as presence of liver co-morbidities (Supplementary Table 7).

Discussion
In our study, we analyzed the hepatobiliary phenotype of the most common inflammatory intestinal diseases using the well-characterized community sample available in the UK Biobank. By this approach, we demonstrated elevated transaminases in CeD compared to controls. This is not surprising, since CeD subjects were shown previously to more frequently display elevated transaminases than the general population even when adhering to a strict gluten-free diet 30 . For example, Castilo et al. reported elevated transaminases to be ~ 1.5 times more  www.nature.com/scientificreports/ www.nature.com/scientificreports/ common in individuals with CeD even 1.5 years after the start of gluten-free diet compared to matched controls 31 .
Notably, the frequency of elevated transaminases in the CeD cohort reported herein is lower than that in the previous studies (i.e. < 10%), which might be due to the facts that (1) we excluded subjects with various liver comorbidities and (2) the UK biobank is enriched for healthy individuals 29 . On the other hand, our observation that average transaminase levels do not substantially differ between the IBD cohort and healthy controls is novel, since no robust, community-based data exist on this topic. This is unexpected, since several studies demonstrated that abnormal liver tests are common in patients with CD and UC 32 and several liver diseases are overrepresented in individuals with IBD 28,33 . While these data add the population-based perspective and strengthen the importance of an appropriate clinical work-up in IBD individuals with elevated serum transaminases, they also have several important limitations. Because of that, further studies are needed to define the values in phases of active inflammation vs. remission, the impact of different treatment regimen and many more.
In contrast to serum transaminases, the CeD cohort did not display elevated GGT levels. This is interesting, since CeD was previously shown to increase the risk of liver steatosis 26,34 and in our study, the CeD cohort more frequently harbored NASH. However, the CeD cohort also had lower alcohol consumption, lower BMI values and relatively low percentage of diabetic subjects, which are all conditions associated with lower GGT values 35,36 . Another unexpected finding were the elevated ALP levels in the CeD cohort compared to controls. In this respect, several studies suggested that increased ALP values are uncommon in CeD and might be related to bone disease rather than cholestatic disorders 30 . Although bone affection might be the key determinant of elevated ALP values in CeD individuals, in our study, the CeD cohort also more frequently suffered cholelithiasis. While the Table 4. Liver phenotype in cirrhotic UK Biobank cohorts with celiac disease, Crohn's disease or ulcerative colitis compared to controls. Quantitative measures are expressed as median with IQR or relative frequency (%). ALT alanine aminotransferase, ALP alkaline phosphatase, AST aspartate aminotransferase, BMI body mass index, GGT gamma-glutamyl transferase, NASH Non-alcoholic steatohepatitis, ULN upper limit of normal (sex-specific). a OR = 0.24[0.08-0.74]; b OR = 8.28 [3.82-20.2 www.nature.com/scientificreports/ association between gallstones and CeD has not been established previously 26 , CeD subjects were reported to have impaired gallbladder motility, which constitutes a well-established factor predisposing to gallstone formation 37 . Finally, CeD individuals are at a higher risk for primary biliary cirrhosis 26 and this established association might also be in part responsible for elevated ALP levels.
In patients with CD/UC, we observed elevated GGT and ALP concentrations. A likely explanation for the former finding is that steatosis is overrepresented in both disorders and was shown to correlate with severity of colitis and duration of disease 26,33,38 . This is likely in part due to steroid use. With regard to the elevated ALP levels, a simultaneous occurrence of PSC is particularly important in UC and may be seen in up to 8% of cases, whereas it is less common in CD 39,40 . On the other hand, gallstones are more prevalent in CD individuals, which is well in line with our observations 28 . Notably, we also saw a higher rate of gallstones in UC individuals, however, this finding is controversial 26,28 and needs to be confirmed by future studies.
Beyond looking at the importance of individual diseases, we assessed the impact of previous intestinal resections. In CD, this event was associated with elevated AST, ALT and ALP levels as well as higher occurrence of gallstones. This is not surprising, since inflammation and/or removal of terminal ileum leads to loss of bile acids that are crucial to prevent cholesterol precipitation 19,21 . In UC, the impact of previous surgery was even more striking and was associated with higher AST, ALT, GGT and ALP concentrations as well as higher occurrence of gallstones and cholangitis, presumably referring to PSC. This is in line with a previous report that demonstrated high frequency of abnormal liver enzymes in individuals with ileal pouch-anal anastomosis 41 . Several reasons might be responsible for this observation. First, surgery is substantially less common in UC than CD and accordingly, it indicates a small subset of patients with a severe, more generalized disease. Moreover, a simultaneous presence of PSC and UC is associated with more severe colitis 28 , that likely accounts for higher surgery rates in individuals suffering from both diseases. Finally, colectomy was shown to alter the biochemical composition of the bile and this mechanism might be responsible for the higher frequency of gallstones 42 .
While the liver enzyme values differed between the analyzed intestinal disorders, all three conditions resulted in a comparably increased occurrence of liver cirrhosis. In the case of CeD, the adjusted OR of ~ 3.6 seen in this study is well in line with previously published data suggesting that CeD subjects display a three times increased hepatic mortality 43 . Two other studies also found at least twice elevated prevalence of liver fibrosis/cirrhosis in CeD individuals compared to age-and sex-matched controls 12,44 . This is likely at least in part due to increased occurrence of chronic hepatitis, non-alcoholic steatohepatitis and autoimmune hepatitis that were found both in our study and previous reports 26,34,44 . Conversely, increased cirrhosis rates in UC are primarily due to the increased prevalence of PSC and to lesser extent to autoimmune hepatitis and steatosis 28 . This may also explain the higher prevalence of HCCs in the CeD but not in the UC cohort 39 . Vice versa, compared to UCs, liver cirrhosis in CDs might be more related to liver steatosis and less to PSC, which may explain the higher occurrence of HCCs. In contrast to our findings, a recent analysis described a similarly increased risk of HCC in CD and UC individuals 45 . Given that the design of UK biobank cohort does not allow a careful analysis of the medical charts of the individual patients and since the diagnosis of HCC and CCA might be difficult to discern, further studies are needed to corroborate our findings. Finally, drug-induced liver injury is known to play a role in both CD and UC 28 , however, cannot be reliably assessed with the data available in the UK Biobank.
Our study has both significant strengths and limitations. Its cross-sectional design precludes an identification of causal relationships and is not well-suited for assessing rapidly progressive disorders such as cholangiocellular carcinoma. In addition, the diagnosis of the studied diseases is based on UK hospital admission codes (ICD10), which may miss some patients, in particular in the case of CeD. However, previous analyses used the same approach and saw a similar performance as case-control studies 46,47 .
A major advantage of the UK Biobank cohort is its community-based setting that closely mimics the general population and minimizes a selection bias seen in single-center studies. Moreover, it allows side-by-side comparison of the different intestinal disorders, which is otherwise difficult to accomplish. Moreover, our study quantifies the previously suggested association between chronic intestinal disorders and the occurrence of endstage liver disease. This association should promote a more thorough hepatologic monitoring of individuals with these intestinal disorders, especially in situations with recurrently elevated liver enzymes and/or presence of additional risk factors, such as obesity, diabetes or metabolic syndrome.

Data availability
The data analyzed in this article are property of UK Biobank and can be obtained through a procedure described at http:// www. ukbio bank. ac. uk/ using-the-resou rce/.