Consistent alterations in faecal microbiomes of patients with primary sclerosing cholangitis independent of associated colitis

Summary Background Single‐centre studies reported alterations of faecal microbiota in patients with primary sclerosing cholangitis (PSC). As regional factors may affect microbial communities, it is unclear if a microbial signature of PSC exists across different geographical regions. Aim To identify a robust microbial signature of PSC independent of geography and environmental influences. Methods We included 388 individuals (median age, 47 years; range, 15‐78) from Germany and Norway in the study, 137 patients with PSC (n = 75 with colitis), 118 with ulcerative colitis (UC) and 133 healthy controls. Faecal microbiomes were analysed by 16S rRNA gene sequencing (V1‐V2). Differences in relative abundances of single taxa were subjected to a meta‐analysis. Results In both cohorts, microbiota composition (beta‐diversity) differed between PSC patients and controls (P < 0.001). Random forests classification discriminated PSC patients from controls in both geographical cohorts with an average area under the curve of 0.88. Compared to healthy controls, many new cohort‐spanning alterations were identified in PSC, such as an increase of Proteobacteria and the bile‐tolerant genus Parabacteroides, which were detected independent from geographical region. Associated colitis only had minor effects on microbiota composition, suggesting that PSC itself drives the faecal microbiota changes observed. Conclusion Compared to healthy controls, numerous microbiota alterations are reproducible in PSC patients across geographical regions, clearly pointing towards a microbiota composition that is shaped by the disease itself and not by environmental factors. These reproducibly altered microbial populations might provide future insights into the pathophysiology of PSC.


| INTRODUC TI ON
Primary sclerosing cholangitis (PSC) is a chronic progressive disease of the biliary system. 1 There is no medical treatment available and patients have an increased risk of both hepatobiliary and bowel malignancies. 2,3 As a result, most patients decease or receive liver transplantation 15-20 years after diagnosis. 4 The pathogenesis is still unclear, although a dysregulated immune reaction at the epithelial barrier in the intestine and the biliary system may play a key role. Cooccurrence of inflammatory bowel disease (IBD), typically in form of a mild pancolitis, has been reported in up to 80% of patients with PSC, and at least 2%-7.5% of IBD-patients develop PSC. 1,[5][6][7] It has been postulated that PSC-associated colitis (PSC-IBD) represents a distinct IBD entity alongside Crohn's disease and UC. 8,9 Several genetic risk factors have been identified that are associated with PSC, but collectively they only contribute to a small fraction of disease susceptibility. 1 Given the minor role of genetic variation, environmental factors likely play a major role in the pathogenesis of PSC. Of these, the microbiome has emerged as one of the most important potential environmental players in chronic inflammatory diseases. Gut microbiome alterations have been identified in several metabolic and inflammatory diseases, [10][11][12] and recent reports have demonstrated alterations in both the faecal and mucosal microbiota in patients with PSC. 13,14 The faecal microbiota of PSC patients has been characterised by a distinct profile compared to healthy controls and UC patients, including an overabundance of Veillonella in a Norwegian and a Czech cohort. 13,15 However, a classification between PSC and UC in a Norwegian cohort based on the abundance of Veillonella could not be validated in a German population. 16 In a Belgian cohort, stool samples of PSC patients showed an increased abundance of the genera Fusobacterium, Enterococcus, Lactobacillus and Streptococcus compared to healthy individuals. 17 In summary, it is unclear whether the microbial signature so far described in PSC single-centre cohorts is centre-specific or if a PSCspecific microbial signature across different geographical regions exists. Cross-regional analysis of faecal microbiota of patients with PSC might reveal general patterns of microbial perturbation, which could elucidate the role of the microbiome in PSC and PSC-IBD and provide the basis for a better understanding of their possible pathogenetic significance in further mechanistic and clinical longitudinal studies.

| Assessment of dietary patterns
Dietary data were collected for 220 individuals (Table 1)

| Stool sample processing and sequencing
Samples were collected and subjected to DNA extraction as previously described for the respective cohorts. 13,22 The amplification and library preparation of the V1-V2 region of the 16S rRNA gene using dualindexing was performed in a single facility (Methods S1). We chose V1-V2 as target amplicon, as 2 300 bp sequencing covers the amplicon of 300-320 bp almost entirely twice, thus assuring high quality data readout. Sequencing data were subjected to quality control and data processing to obtain count-based relative abundance tables for operational taxonomic units (OTUs) and taxonomic levels from phylum to genus (Methods S1).

| Data analysis
Data analyses were performed with R statistical programming language (v3.4.3). 23 Differences in dietary patterns were evaluated using the logtransformed average intake of the primary macronutrients protein, fat and carbohydrates (g/day), as well as fibre, water (both g/day), and total energy intake (kJ/day). To assess dietary differences of microbiota composition that is shaped by the disease itself and not by environmental factors. These reproducibly altered microbial populations might provide future insights into the pathophysiology of PSC. JH is funded by the Research Council of Norway (no. 240787/F20). MKu is funded by the Regional Health Authority of South-Eastern Norway (no. 2016067).   Shannon index, as a measure of within-sample diversity (alphadiversity), and PERMANOVA on Bray-Curtis dissimilarity, as a measure for beta-diversity, were applied to investigate differences according to disease state and other co-variables (Methods S1).
All abundances are based on a normalised number of counts, thus being relative abundances, this is always the case when the term 'abundance' is used throughout the text. To assess differential taxa abundances, we tested microbes that were identified as differentially abundant in previous studies in a first step (Methods S1), followed by a second step where we aimed to discover new associations. Details on the applied regression models and meta-analysis are provided in the Methods S1. Taxa with significant signals in both respective geographical cohorts and the meta-analysis were re-analysed with inclusion of dietary co-variables in the German cohort, for which food frequency questionnaire data was available, to correct for dietary effects.
Predictive performance of the identified taxonomic signature was evaluated using random forests classification 24 (Methods S1).
To evaluate model performance, receiver operating characteristic analysis was used. Additionally, F1 score and Matthews Correlation Coefficient (MCC) as weighted measures of true and false positives rates were calculated (Methods S1).

| RE SULTS
In total, we applied 16S rRNA gene sequencing to 257 German (n = 95 controls, n = 37 PSC only, n = 37 PSC-IBD and n = 88 UC) and 131 Norwegian samples (n = 38 controls, n = 25 PSC only, n = 38 PSC-IBD and n = 30 UC). We applied the same amplification and library preparation standard operating procedure within a single facility to all samples. For 220 of the 257 (85.6%) German study participants dietary data, assessed by standardised food frequency questionnaires, was available (Table 1).

| Healthy Norwegian and German subjects share similar core microbiota
To determine the baseline similarities and differences between the faecal microbiota of German and Norwegian healthy volunteers, we compared the healthy controls of both cohorts. We found large proportions of the core microbiota (122 of 144 taxa, 84.7%) shared by healthy controls from both populations (Figures S1-S3). The healthy Norwegian cohort displayed a significantly lower intra-individual (alpha) diversity compared to the healthy German population (P = 0.006). We observed slight but significant differences in between-sample diversity (beta-diversity) between German and Norwegian controls (P = 0.01; R 2 = 0.021).

| The faecal microbiota of patients with PSC is significantly different from both healthy controls and patients with UC
In both cohorts, between-sample diversity (beta-diversity) was significantly different between patients with PSC and controls (P < 0.001, respectively; R 2 GER = 0.028; R 2 NOR = 0.042). Differences in beta-diversity between patients with PSC and UC were also significant but less pronounced in both cohorts (P NOR = 0.016, R 2 NOR = 0.027; P GER = 0.013, R 2 GER = 0.015). In the Norwegian cohort, mean within-sample diversity (alphadiversity) of patients with PSC was reduced compared to controls (P = 0.001) and comparable to patients with UC (P > 0.05). In the German cohort however, alpha-diversity of patients with PSC was comparable to controls (P > 0.05) and significantly increased in contrast with patients with UC (P = 0.01) (Figure 1).

| Targeted analysis of taxa with previously reported association with PSC
A total of nine genera that were previously identified as differentially abundant 13,17 were analysed in a targeted approach aiming to reproduce the taxonomic signals. A detailed summary is provided in the Tables S1 and S2.
In both cohorts, an increased relative abundance in patients with PSC was displayed by Veillonella and Streptococcus (both P META < 0.0001). In addition, an increased prevalence in patients with PSC was confirmed for Lactobacillus and Enterococcus (both P META < 0.0001, respectively). Other taxa either showed inconsistent distribution patterns between cohorts or could not be recovered (with sufficient prevalence) in our samples.

| Extensive microbiota alterations in patients with PSC
In the next step, we aimed to discover new robust taxonomic distribution patterns between patients with PSC and controls across all taxonomic hierarchy levels. A detailed summary is provided in Regarding microbial (binary) prevalence patterns, we observed an extensive depletion in patients with PSC compared to controls affecting 32 taxa. Among these were the genera Holdemanella and Desulfovibrio as well as OTUs classified as Faecalibacterium and Clostridium IV (both Q META < 0.0001).

| PSC and UC are both characterised by altered microbiota, but cannot be differentiated by single taxa
We investigated if the observed difference of beta-diversity between patients with PSC and UC can be traced to robust differential distribution of individual taxa (Tables S9 and S10). In both cohorts, the phylum Firmicutes was significantly increased in patients with PSC compared to patients with UC (Q META = 0.011). We found no cohort-spanning differences of lower hierarchy level taxa except for one OTU assigned to the genus Ruminococcus (OTU_59; Q META < 0.01; Figure 3). Effect size (β)

| The faecal microbiota profile can predict the diagnosis of PSC across different geographical regions
In order to investigate, whether faecal microbiota can be used to predict the presence of disease, we applied random forests classifica-  Figure 4C). Further in-depth exploration of the classification is provided in the Supporting information.

| Diet has minor impact on microbial community alterations in PSC
Univariate comparisons of disease groups to healthy individuals showed only a minor reduction in daily water intake in PSC patients (P adj = 0.049).
Comparing multivariate differences in major dietary patterns in 220 samples of the German cohort, no significant differences were seen according to age (P > 0.05, R 2 = 0.002), BMI (P > 0.05, R 2 = 0.008) and diagnosis (P > 0.05, R 2 = 0.013). Gender, however, showed a strong impact (P < 0.001, R 2 = 0.2), as expected. Using the mentioned macronutrients and questionnaire derived intake variables as covariates in the analysis of disease-associated shifts in microbial beta-diversity yielded no significant associations (P > 0.05, respectively) and also revealed only minor effects on the still highly significant change in community composition associated with PSC (P = 0.003, R 2 = 0.019). membranes, such as Enterobacteriaceae. 25 Multiple lines of evidence point to LPS as a common co-factor of liver injury. 26 Individual variation in Gammaproteobacteria has been directly linked with susceptibility to fatty liver disease. 27 The increase in Parabacteroides is intriguing, as it has been demonstrated that this bile-tolerant taxon is linked to changes in cholesterol and bile acid metabolism. 28,29 For eight taxa across different taxonomic levels we demonstrated a consistently increased abundance in patients with PSC. These findings confirm previously found associations, eg for the genera Veillonella 13,16 and Streptococcus. 17 Increased abundance of both were previously found in patients with primary biliary cholangitis (PBC) and patients with liver cirrhosis of different origins. 12,30 This indicates that these alterations might be rather unspecific features of chronic liver disease. Additionally, we confirmed a markedly reduced abundance of one OTU belonging to the genus Coprococcus in patients with PSC. This taxon has previously been found to be decreased in patients with UC, 31 which is consistent with our own findings (Supporting information). This supports the notion of Coprococcus as an indicator of general gut integrity. Other previously described associations of bacterial taxa with PSC could not be confirmed, such as the recently described association with the genus Klebsiella. 32 We could demonstrate an extensive alteration of the microbial com-  In previous studies of the mucosal microbiota in UC patients with and without PSC, city of origin was the main determinant of gut microbiota profile despite identical handling of all samples and no consistent differences between UC and PSC with colitis were observed. 35 This is in line with the present findings, as significant differences were detected between PSC and UC in each cohort separately, however, only few of them were consistently observed across both cohorts. We can therefore not answer conclusively if microbiota profiling is also sufficient to distinguish PSC from UC in a geographically independent manner.

| D ISCUSS I ON
Our results strongly suggest that the increase of abundance or prevalence in PSC in contrast with controls is specific for certain taxa.
Therefore, their potential role in the pathogenesis of PSC should be investigated. In addition, the extensive loss of many taxonomic groups is relatively unspecific and may potentially be explained by general effects of chronic inflammation rather than by disease-specific etiology.
Only marginal differences were identified in our study between patients with PSC only and PSC-IBD. This suggests that the liver disease and not the colitis is the primary driving force behind the observed gut microbial dysbiosis and that, as the differences in inflammation-localisation in the colon already suggest, PSC-IBD displays significant pathophysiological differences to UC.
Our study has several strengths and shortcomings. Notably, even though the German and Norwegian samples were processed with kits from different manufacturers, we could show extensive overlap in biological signals. Analysis of dietary patterns, that were available for the German cohort, showed no general differences between individuals based on disease status. Only minor influences on beta diversity were observed, which did not affect the clear disease-associated shift in microbial communities. Previous studies investigating the influence of dietary patterns on the microbiota could show, that indeed diet can influence microbiome compostion. 36,37 However, these were performed on larger study populations (n > 1000) and still only found small individual effects, thus to really investigate diet-disease interactions, larger, well-typed PSC cohorts are needed, which are currently not available.
Although facilitating large-scale analysis, amplicon-based marker gene surveys always come with trade-offs regarding fragment-specific biases in amplification and taxonomic assignment quality, which may in part explain differences between studies. Ultimately, the intestine must be regarded as an open, highly dynamic and spatially heterogeneous system, and faecal samples can only serve as a proxy for the actual state. Nevertheless, we here present the largest microbiota study performed in PSC to date and show robust and geography-spanning alterations in the faecal microbiota, in spite of sampling and technical differences between the centres. This strongly supports the conclusion that the observed microbiota changes are disease specific, and not primarily driven by environmental factors.
In summary, patients with PSC from different geographical regions display shared differences in gut microbial composition compared to healthy controls, independent from the presence of concurrent IBD and the use of common medication. The PSC specific gut microbiota signature might have diagnostic potential and provides a strong rationale for further microbiome meta-analyses with even larger international cohorts. These should also utilise metagenomic sequencing, profiling of microbial metabolites as well as longitudinal designs to further define the role of the gut microbiota in the pathogenesis and clinical care in PSC. Furthermore, additional patients with chronic or cholestatic liver disease should be included for validation of disease specific effects.
This study focused on broad, disease-specific signals, as these are likely to be directly connected to the largely unknown etiology of PSC, independent of environmental influences. Ultimately, this may lay the foundation for new therapies targeting the gut microbiota in PSC.

ACK N OWLED G EM ENTS
We thank Ms. Ilona Urbach, Ms. Ines Wulf and Mr. Tonio Hauptmann of the IKMB microbiome laboratory and the staff of the IKMB sequencing facilities for excellent technical support.
Declaration of personal interests: None.

AUTH O R S H I P
Guarantor of the article: None.