Disease severity and proton pump inhibitor use impact strongest on faecal microbiome composition in liver cirrhosis.

Abstract Background & Aims Compositional changes of the faecal microbiome in cirrhosis are well described and have been associated with complications and prognosis. However, it is less well known, which disease or treatment‐related factors affect microbiome composition most distinctively. Methods 16S rDNA sequencing data of 88 cirrhotic outpatients were investigated. Factors influencing microbiome composition were analysed by univariate and multivariate redundancy analysis. The association of the identified factors with changes in diversity and taxonomic composition was studied in depth using analysis of composition of microbiome, LDA‐effect size and least absolute shrinkage and selection operator regularized regression. Results Disease severity and aetiology, proton pump inhibitor (PPI) use, nutritional status, age and C‐reactive protein are significant explanatory variables for faecal microbiome composition in liver cirrhosis. Despite some taxonomic overlaps especially between disease severity and PPI use, we could show that the effects of disease severity, aetiology, PPI use and age are independent factors influencing microbiome composition also in subgroup analyses. Conclusion Our cross sectional system biology study identifies disease severity, aetiology, PPI use and age as independent factors that influence microbiome composition in liver cirrhosis. In chronic diseases with high morbidity, such as liver cirrhosis, precise patient metadata documentation is of utmost importance in microbiome analysis. Further studies with a higher sample size are necessary to validate this finding. Trial Registration Number: NCT01607528


| INTRODUC TI ON
Liver cirrhosis is an increasingly common disease with high complication rates. It leads to reduced quality of life and a high burden of disease for patients, their family and the healthcare system. 1 Cirrhosis is associated with changes in the structure and functionality of the microbiome of the gut [2][3][4][5] and other body sites, such as skin or the mouth. 6 Compared to healthy individuals, a decrease in faecal microbial diversity and an imbalance between commensal and pathogenic taxa is seen in patients with cirrhosis. Furthermore, cirrhosis-associated dysbiosis goes hand in hand with increased gut permeability, intestinal bacterial translocation, intestinal and systemic inflammation leading to complications of cirrhosis and an increased mortality. 3,[7][8][9][10][11] Dysbiosis in general and especially in cirrhosis may be caused by host and environmental factors, which shape the microbiome. These factors include alcohol consumption, aetiology and severity of liver disease, diet composition and medication. 5,[12][13][14][15][16][17][18] Drug intake has emerged as one of the most important drivers of dysbiosis. It has recently been shown in vitro that, apart from classic antimicrobials, many other drugs have an extensive impact on human gut bacteria. 19 A population-based deep sequencing study of the faecal microbiome revealed that proton pump inhibitors (PPI) were associated with the most profound microbiome changes, followed by statins, antibiotics, laxatives and beta blockers. 20 Nearly 50% of older adults take one or more medications that are not medically indicated and 45% of patients over the age of 75 take five or more drugs per day. 21 The known consequences of polypharmacy and over-medication are increased healthcare costs, adverse drug reactions, increased rates of drug-drug interactions, decreased performance status of the patients, cognitive impairment, higher risk of falls and non-compliance. Interventions to reduce polypharmacy are difficult to implement. 22,23 From the currently available data it is not known which of the above described factors may have the strongest impact on faecal microbiome composition in cirrhosis and would therefore be the most promising therapeutic goal.
We therefore conducted a systems biology analysis on a large dataset of cirrhotic patients of different aetiology and disease severity to analyse, which factors have the strongest influence on faecal microbiome composition and predicted metagenomics in patients with liver cirrhosis.

| Patients
We included faecal 16S microbiome sequencing data from cirrhotic patients recruited at the outpatient clinic of the Department of Gastroenterology and Hepatology, University Hospital of Graz, Austria who were screened for an intervention study 10 between July 2012 and September 2013 in this post-hoc analysis. All patients gave written informed consent. Diagnosis of cirrhosis was based on liver histopathological examinations, or a combination of clinical, radiological and/or labouratory features. Patients with a Child-Pugh score of 12 or higher, alcohol consumption within 2 weeks prior to inclusion, active infection at screening, gastrointestinal haemorrhage within 2 weeks prior to inclusion, immuno-modulating drugs, hepatic encephalopathy stage two or higher, renal failure (creatinine over 1.7 mg/ dL), other severe diseases unrelated to cirrhosis, malignancy or pregnancy were excluded. Stool and blood samples analysed in this study were taken before patients received any study medication. The study protocol was approved by the institutional review board (ethics committee) in Graz (23-

| Statistical analysis
For microbiome analysis, demultipexed FASTQ files were processed using Qiime2 tools implemented in Galaxy (https ://galaxy. medun igraz.at). Denoising (removing primers, quality filtering, correcting errors in marginal sequences, removing chimeric sequences, removing singletons, joining paired-end reads and dereplication) were done with DADA2. 28 Taxonomy assignement was based on Silva 132 database release at 99% OTU level and trained using a Naïve Bayes classifier. After pre-processing, an average of 59 369 reads per sample could be reached. All analyses, except diversity analysis, were done on an unrarefied feature table. For normalization Hellinger transformation was used.
Rare taxa with a relative abundance of less than 0.01% across all samples were filtered. Chloroplast and cyanobacteria filtering was performed to remove contaminants. Alpha diversity analysis was performed using Chao1 on a rarefied feature table (sequencing depth 14 086). Beta diversity was analysed with Redundancy Analysis (RDA) 29 based on Bray Curtis dissimilarity. Differentially abundant taxa were identified with Analysis of Composition of Microbiomes (ANCOM). 30 As machine learning methods to select genera associated with the explanatory variables, LDAeffect size (LEfSe) 31  Score matching was performed without replacement based on logistic regression, using R 3.6.1 33 package "MatchIt." 34,35 The web-based software Calypso version 7.14 (http://cgeno me.net/ calyp so/) was used for analyses of microbiome data. 36 For nonmicrobiome analyses SPSS V25.0.0.1 (IBM, Armonk, NY, USA) was used (descriptive statistics, group comparisions, Spearman Rho correlation, Collinearity analysis) Visualization was performed in R 3.6.1 33 package "ggplot2". 37 Sequencing data have been made publically available at the NCBI Sequence Read Archive (accession number PRJNA390475). A formal samples size calculation was not performed.

| Study cohort
Microbiome sequence data of 88 cirrhotic patients were analysed.
Patient characteristics are presented in Table 1. We analysed the impact of age, sex, smoking status, aetiology of cirrhosis, severity of liver disease, comorbidities, nutritional status, drug intake, intestinal permeability, intestinal and systemic inflammation on faecal microbiome composition. Drug classes that were taken by more than 15% of the study population were included with the exception of lactulose, which was used by only 11% of the study population but was included into univariate analysis due to its supposed microbiome modulating properties. Antibiotics (chinolons) for prophylaxis of spontaneous bacterial peritonitis were only taken by 2 out of 88 patients and rifaximin was not taken by any patients.
Moderate malnutrition was more frequent in hepatitis C cirrhosis (43.8%) compared to alcoholic cirrhosis (19.1%) or other aetilologies (4%, P = .006). No other significant differences between categorical variables were found. Spearman correlation showed a weak, but significant correlation between CRP and disease severity (Spearman Rho r = 0.380, P < .001). No significant correlation between other variables was found. Testing for collinearity showed that all combinations of explanatory variables had a VIF <1.5, excluding a strong codependence.

| Differences in microbiome composition between groups (beta diversity)
Univariate RDA revealed that severity of liver disease (assessed by  Figure 1).

| Species diversity (alpha diversity) and taxonomic differences
We analysed differences in alpha diversity and taxonomic composition in relation to the six variables (severity of disease, aetiology, PPI use, nutritional status, age and CRP) that significantly affected beta diversity of the stool microbiome composition.
For alpha diversity analyses the feature

| Machine learning and network analysis
To further understand the association of microbiome composition with the factors that were identified to significantly influence beta propensity score matching, disease severity, aetiology, PPI use and age were still significant explanatory variables of microbiome composition on multivariate RDA whereas CRP and nutritional status did not influence microbiome composition significantly any more (see Figure S1).
Feature selection by LEfSe in the propensity score matched cohort showed comparable results as obtained from the original, non-matched dataset, indicating that the microbiome effects are true effects and not caused by other confounders (see Figure S2). To confirm the influence of PPI use independent of severity of liver disease, we additionally performed a subgroup analysis for Child A cirrhosis and Child B/C cirrhosis separately. When performing multivariate RDA analysis separately for severity groups, we observed that PPI use (P = .035) in Child A cirrhosis and aetiology (P = .015), PPI use (P = .038) and age (P = .014) in the Child B/C group were still predictive for microbiome composition. Due to low sample size in the subgroups the results have to be interpreted with caution.

| D ISCUSS I ON
Faecal microbiome composition has been associated with various diseases, but not only disease severity and comorbidities, but also other factors such as medication, diet or lifestyle habits may influence its composition. It is therefore important to discern the effect of different influencing variables as a first step from association to causality. We performed a system biology analysis of a well-characterized cohort of patients with liver cirrhosis and were able to show by multivariate RDA that severity and aetiology of Cirrhosis is a complex disease requiring long-term drug treatment with several drug classes. Many medically approved drugs influence microbiome composition. 19 In liver cirrhosis, PPI use has been described to alter microbiome composition, increase the rate of complications and negatively impact prognosis. 5,46-49 We recently expanded this knowledge by describing the consequences of PPI-induced dysbiosis and oralization of the faecal microbiome on inflammation, intestinal permeability and outcome in cirrhosis. 11 In the present study PPI use also had a strong impact on the faecal microbiome, being associated with an increased abundance of oral bacteria and potential pathogens, such as Streptococcus species and Veillonella. Although PPI use and disease severity were partially linked in our cohort, both were still independent factors influencing microbiome composition in subgroup analyses, however low sample size in this analysis weakens the explanatory power.
Also other drugs commonly administered in cirrhosis, may impact the faecal microbiome. However, in our cohort, other drugs influenced microbiome composition to a lesser extent. We were interested in the effect of lactulose on the cirrhotic microbiome, since available data are conflicting, showing no major alterations of the faecal microbiome between lactulose user and non-user, but withdrawal of lactulose leads to the loss of beneficial species. 16,50,51 In our cohort only 11% of the cirrhotic patients used lactulose. On univariate analysis we found that lactulose had a weak, but significant, effect on microbiome composition, which did not remain significant in multivariate analysis. Therefore our data support the notion that lactulose has no major effect on the taxonomic composition of the faecal microbiome in cirrhosis and may therefore exert its function through functional microbiome changes. Other drug classes did not have a major impact on microbiome composition.
Liver cirrhosis is also a heterogeneous disease from an aetiological perspective. Aetiology of liver disease as a factor to explain differences in microbiome composition has already been studied. Bajaj et al describe alcohol and non-alcoholic steatohepatitis driven changes in microbiome composition. 45 Hepatitis C alters faecal microbiome which has been implicated in the pathogenesis of HCV-induced chronic liver disease. However, Hepatitis C induced dysbiosis seems to be stable over different disease stages. 52 Chronic cholestatic diseases such as primary biliary and primary sclerosing cholangitis are also associated with distinct changes in microbiome composition. 53 Although aetiology of liver disease was identified as a significant factor by multivariate RDA in our data set, the taxonomic differences of the microbiomes from different aetiologies were surprisingly small in our cohort.
Patients with alcoholic cirrhosis had a higher abundance of the genus Erysipelatoclostridium and the group of patients with "other" aetiologies of liver cirrhosis had a higher abundance of two yet uncultured bacteria of the family Lachnospiraceae and of the genus Blautia. Since liver cirrhosis aetiology is varying in different geographical regions 54 and geographical region itself is a factor that impacts diversity and composition of the microbiome, 55 57 Collinsella aerofaciens, Alistipes onderdonkii and the genus Slackia were differentially abundant between age groups in our cirrhosis cohort.
However, none of these have been previously described to be associated with age related changes in microbiome composition.
Collinsella aerofaciens has been associated with diet and intestinal inflammatory diseases, [58][59][60] Slackia with equol production, 61 whereas Alistipes onderdonkii has not been associated with any human condition yet. 62 Nutritional status is an important prognostic factor in liver cirrhosis. Especially sarcopenia is a frequent complication of malnutrition in cirrhosis and is associated with adverse outcome. 63,64 Assessing nutritional status in cirrhosis can be challenging. We used the well-established SGA, that is also recommended by the A loss of Akkermansia muciniphila has been associated with metabolic diseases and the abundance may be influenced by diet. 68 Changes in microbiome composition in cirrhosis impact intestinal barrier function and lead to intestinal and systemic inflammation due to translocation of bacterial products to the liver and also to the systemic circulation. This concept of a crosstalk between gut, liver and immune system -the so called gut-liver axis-is widely implicated in the pathogenesis of liver disease and a promising therapeutic target. 69,70 We assessed a panel of biomarkers of the gut-liver axis in our cohort and CRP as a marker of inflammation was found to be a significantly explanatory variable in the multivariate analysis, however the consistency of the effect remains unclear in our dataset. CRP is a well-known biomarker in liver cirrhosis and is predictive for complications and outcome. 71,72 Increased abundance of Veillonella and Streptococcus species but also Faecalibacterium species were associated with higher CRP levels. While the former support the link between microbiome composition and inflammation in liver cirrhosis, Faecalibacterium species are usually associated with anti-inflammatory properties, making a firm conclusion difficult.
Interestingly, gender did not impact on microbiome composition in our study. In studies in obese individuals, gender seems to cause taxonomic differences, whereas data on changes on alpha and beta diversity are still conflicting. 73,74 In liver cirrhosis, gender differences have not been associated with changes in beta diversity so far. The male predominance in liver cirrhosis may be a reason for difficulties to detect consistent gender differences.
In summary our cross sectional system biology study shows that disease severity and PPI use are the main factors explaining variation of the faecal microbiome in cirrhosis. Aetiology of liver disease, age, nutritional status and inflammation (CRP levels) are further explanatory variables. The limitation of our study is the single-center design that does not allow to account for geographical differences in the microbiome composition, the cross sectional design that does not allow to draw any conclusions on causality as well as the low sample size in the subgroup analyses. The strength of our study is the thorough characterization of our study patients, that allows detailed analysis of influencing variables. For future studies, we strongly suggest to increase sample size and expand the minimal set of metadata as suggested by the IHMS consortium 75 to include detailed information on disease aetiology, disease severity, drug intake and also information on further essential biomarkers, depending on the disease studied. This will open up new paths to understand the crosstalk between the faecal microbiome and the human body in disease states.

| Lay summary
The composition of gut bacteria-the gut microbiome-is altered in many diseases. In chronic liver diseases, such as liver cirrhosis, the gut microbiome is severely disturbed. We were able to show in this study which factors explain this disturbed microbiome composition.
These factors were: The cause of liver disease, the severity of liver disease, intake of acid-blockers (proton pump inhibitors), age and inflammation. Our study also shows the importance of collecting sufficient data on the disease and drug intake to be able to assess the effects of different factors on the gut microbiome.

CO N FLI C T O F I NTE R E S T
None of the authors declares any conflict of interest.