We prospectively established two unique independent cohorts from the First Affiliated Hospital of Zhengzhou University and the Affiliated Hospital of Xuzhou Medical University, China. Patients (from inpatient and outpatient departments) in the cohort were diagnosed with Budd-Chiari syndrome (B-CS) using a multidetector computed tomography (CT) three-dimensional vascular reconstruction technique combined with colour Doppler ultrasound. Neither the patients nor healthy controls used antibiotics in the past 3 months without taking probiotic preparations or dietary yogurt for 1 month. In particular, none of the subjects suffered from any disease affecting intestinal microorganisms, including obesity, hypertension, diabetes, intestinal diseases, long-term diarrhoea, and constipation. All healthy controls were family members of patients with B-CS who shared the same diet and living environment. All participants provided written informed consent for participation in the study. The study protocol was approved by the local ethics committee.

Blood samples from all individuals were collected in 5 mL EP tube, then centrifuged at 4,000 rpm for 10 min at 4°C to extract serums to new Eppendorf tubes and stored at -80°C for further detection. A total of 100 μL of each thawed sample was placed in 2 mL EP tubes and extracted as previously described (Dunn et al., 2011). Then, 75 μL supernatant was transferred into a 2-mL liquid chromatography-mass spectrometry (LC/MS) glass, and 10 μL of each sample supernatant was pooled as quality control (QC) samples. Finally, a 75 μL supernatant was used for the UHPLC-QTOF-MS analysis.

Similarly, fresh faeces samples were collected in 5-mL EP tubes and stored at -80°C for further shotgun metagenomics sequence. Microbiota DNA was extracted at Novogene Bioinformatics Technology (Beijing, China) using the sodium dodecyl suplphate (SDS) method. DNA concentration and purity were assessed using 1% agarose gels, and DNA was subsequently diluted to 1 ng/μL using sterile water. DNA degradation degree and potential contamination were monitored on 1% agarose gels. The NanoPhotometer^® spectrophotometer (IMPLEN, CA, USA) was used for DNA purity (OD260/OD280 and OD260/OD230) detection. The Qubit^® dsDNA Assay Kit in Qubit^® 2.0 Fluorometer (Life Technologies, Carlsbad, CA, USA) was used for DNA concentration measurements.

LC-MS/MS analysis was performed using an Agilent 1290 Infinity UHPLC system coupled to a Triple TOF 6550 (Agilent Technologies) to acquire the full scan MS1 data, a Triple TOF 6600 mass spectrometer (AB Sciex) to acquire MS/MS spectra, and an UPLC BEH Amide column (2.1 * 100 mm, 1.7 μm, Waters) (Zhang et al., 2019). The mobile phase was composed of 25 mmol/L ammonium acetate and 25 mmol/L ammonia hydroxide in water (pH = 9.75) (A) and acetonitrile (B). The analysis elution gradient was as follows: 0 ~ 0.5 min, 95% B; 0.5 ~ 7.0 min, 95% ~ 65% B; 7.0 ~ 8.0 min, 65% ~ 40% B; 8.0 ~ 9.0 min, 40% B; 9.0 ~ 9.1 min, 40% ~ 95% B; 9.1 ~ 12.0 min, 95% B. The column temperature and auto-sampler temperature were set to 25°C and 4°C respectively. The injection volumes were 2 μL for both positive and negative modes. The parameters for full scan MS1 data acquisition in 6550 QTOF mass spectrometry (Agilent Technologies) were set as follows: the scan range was 60 – 1200 Da; gas temperature, 250°C; gas flow, 16 L/min; sheath gas temperature, 350°C; sheath gas flow, 12 L/min; nebulizer, 20 psi; fragmentor, 175 V; capillary voltage, 3,000 V, were set as the electrospray ionisation (ESI) source conditions. MS/MS spectra was obtained in an information-dependent basis (IDA) mode in Triple TOF 6600 mass spectrometer (AB Sciex) with the acquisition software (Analyst TF 1.7, AB Sciex) with the following criteria: 12 precursor ions with intensity above 100 were selected and then fragmented in each cycle (0.56 s) under the collision energy (CE) of 30 V. Gas 1, Gas 2, and Curtain Gas were 60 Psi, 60 Psi and 30 Psi, respectively, source temperature was 600°C with 5,000 V and -4,000 V of ion spray voltage floating (ISVF) for the positive and negative mode respectively, were set as the ESI sources conditions.

All samples were paired-end sequenced on an Illumina NovaSeq 6000 platform (insert size 350 bp, read length 151 bp) at Novogene Bioinformatics Technology. The microbial DNA was extracted by CTAB method. Adapter and low-quality reads were discarded, and the cleaned reads were filtered from human host DNA based on the human genome reference (hg19) as previously described (Qin et al., 2012). Next, 3,511.15 Gb of high-quality pair-end reads were acquired from 221 human gut microbiome samples with an average of 15.89 Gb per sample.

Unless stated otherwise, all statistical analyses were conducted using the R and the SPSS 21.0 software (SPSS Inc., Chicago, IL, USA). The clinical characteristics of participants enrolled in the study were expressed as mean ± standard deviation and the difference between groups was evaluated using one-way analysis of variance (ANOVA).

The converted metabolomics data were analysed using the XCMS software (https://metlin.scripps.edu/) to calculate the normalised peak intensity, retention time, and exact mass. The matrix was further reduced by removing peaks with missing values (ion intensity = 0) in more than 50% samples to obtain consistent variables. The relative standard deviation (RSD) value of metabolites in the QC samples was set at a threshold of 30%, as a standard in the assessment of repeatability in metabolomics data sets (Dunn et al., 2011). Each retained peak was normalised to internal standard (IS) and the data were introduced to SIMCA14 (Umetrics, Sweden) for principal components analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) with permutation-test (n = 200). The variables with variable importance in project (variable importance in project, VIP > 1, unpaired two-tailed Wilcoxon rank-sum test P < 0.05 and fold change (FC) > 1.2 or FC < 0.83) were considered significant metabolites. Metabolomics pathway analysis (MetPA) was conducted using the online tool MetaboAnalyst with hypergeometric test for over-representation analysis and relative-betweenness centrality for pathway topology analysis (Chong et al., 2018).

Differential abundance of phyla, genera, and species between B-CS patients and healthy controls were tested using the unpaired two-tailed Wilcoxon rank-sum test. The linear discriminant analysis (LDA) effect size (LEfSe) method was used to identify the greatest differences among specific characteristics of intestinal microbiota (http://huttenhower.sph.harvard.edu/lefse/) with the LDA scores (log10) > 2.0 (Segata et al., 2011). The statistical significant differences between metabolic or microbiota profile and individuals’ phenotypes were tested using permutation multivariate analysis of variance (PERMANOVA) using Bray-Curtis distance, and the number of permutations was 999. Unless otherwise stated, P < 0.05 was considered statistically significant. To obtain the functional profile, the high-quality reads were aligned to the updated gut microbiome gene catalogue (Li et al., 2014) using SOAP2 with a threshold of more than 90% identity over 95% of the length. Sequence-based gene abundance profiling was performed as previously described (Li et al., 2014). Next, the relative abundances of Kyoto Encyclopaedia of Genes and Genomes (KEGG) orthologous groups (KOs) were summed up from the relative abundance of their respective genes. Differentially enriched KEGG pathway/modules were identified according to their reporter score (Patil and Nielsen, 2005; Feng et al., 2015) from the Z-scores of individual KOs. A reporter score of ≥ 1.96 (95% confidence according to normal distribution) was used as a detection threshold for significantly differentiating pathways. The R language was used for microbiome data analysis and the database for taxonomy annotation is metaphlan2, besides, to annotate function, reads were mapped to IGC (Intergrated Gene Catalog) to gain gene abundance, then KEGG(Kyoto Encyclopedia of Genes and Genomes) database was used to annotate KEGG function.

Random forest (RF) analysis was used to screen potential metabolic and microbiota biomarkers for B-CS. RF models were used to rank each type of profiles independently and validated by 10-fold stratified cross-validation testing with default parameters in RF function of R package RF (ntree = 1,000). The optimal number of discriminatory markers for each type of profiles was determined using the recursive feature elimination method with default parameters using five different random seeds. The risk probability of B-CS disease for each subject was computed by the selected metabolic or microbiota features, and receiver operating characteristic (ROC) curves were constructed using pROC (Robin et al., 2011). The optimised RF model was further tested on the test datasets. Besides, Spearman’s rank-order correlation was used to determine the strength and direction of the monotonic relationships among variables [metabolites, gut microbiota (phyla, genera and species), KOs, and pathways].

We collected 214 B-CS patient samples and 41 healthy control samples from 2018 from the two units. The baseline information [average age, sex, and basal metabolic index (BMI)] of the patients and healthy controls were comparable. Specimens with less abundant samples or those that changed during storage were excluded. We performed faecal metagenomics sequencing for 191 patients and 30 controls as well as serum metabolomics sequencing for 176 patients and 39 controls. We divided patients into untreated (n = 93) and treated (n = 121) groups based on whether invasive treatment (percutaneous recanalization) was performed at the time of sample collection. Simultaneously, according to our previous grouping pattern based on imaging findings and clinical manifestations (Fu et al., 2011), we subdivided the patients into four groups: IVCHT-PHT- (n = 17), IVCHT+PHT- (n = 23), IVCHT-PHT+ (n = 71), and IVCHT+PHT+ (n = 103) ( Table 1 ).

In this study, diversity was evaluated using the richness measure (number of microbial taxa with nonzero counts) and Shannon index. The α diversities at species and gene levels did not differ significantly among the groups (P > 0.05). Furthermore, the results of principal coordinate analysis (PCoA) for β diversity at species, phylum, genus, and gene levels significantly varied among groups (P > 0.05). We examined the differences in microbial composition of the different groups by decomposing the total variance based on the distance matrix. We found that the results of PERMANOVA analysis (Bray-Curtis distance, n = 999) was not significantly different: the P values for species, genus, phylum and genes were 0.694, 0.758, 0.548, and 0.543, respectively. ( Supplementary Figure 1 ).

However, gut microbiota was different in untreated B-CS patients compared with that in the control group. The diversity varied as follows: at the phylum level, Ascomycota was significantly decreased ( Figure 1A ); at the genus level, Campylobacter was increased, while Comamonas, Saccharomyces, Deinococcus, Thiomonas, and Burkholderiales noname were decreased ( Figure 1B ); at the species level, Prevotella stercorea, Campylobacter gracilis, Streptococcus anginosus, Veillonella atypical, and Veillonella unclassified were increased, and Burkholderiales bacterium_1_1_47, Lachnospiraceae bacterium_6_1_63FAA, Deinococcus unclassified, Thiomonas unclassified, and Saccharomyces cerevisiae were decreased ( Figure 1C ). The results showed dysbiosis of gut microbiota in untreated B-CS.

To assess the functional consequences of changes in microbial communities, we performed enrichment analysis using Kyoto Encyclopedia of Genes and Genomes (KEGG) annotations. Compared to that in the healthy controls, twenty-four functions were reduced in the untreated B-CS patients, including phosphotransferase system (PTS), flagellar assembly, ABC transporters, benzoic acid degradation, nitrogen metabolism, galactose metabolism, bacterial chemotaxis, and lipopolysaccharide biosynthesis. Eighteen functions involving phosphoinositide metabolism, secondary bile acid biosynthesis, valine, leucine and isoleucine degradation, proteasome function, N-glycan biosynthesis, RNA transport, glycosaminoglycan degradation, and methane metabolism were enriched in untreated B-CS ( Figure 1D ). Overall, these results illustrated changes at the level of the microbial community structure, and functional potential.

Following pre-process, 1,866 peaks from positive mode and 810 from negative mode were log-transformed and merged for subsequent analyses. Three-dimensional data, including the sample name, peak number, and normalized peak area, were analysed using SIMCA 14 for principal components analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA). The PCA scores plot distinguished the control group from the four B-CS groups (IVCHT+PHT-, IVCHT+PHT+, IVCHT-PHT+, and IVCHT-PHT-) ( Figure 2A ). Supervised OPLS-DA (7-fold cross validation) model was applied to better understand the variables responsible for classification of B-CS groups vs. healthy controls. The parameters for the classification were R2Y = 0.996 and Q2 = 0.985 (untreated IVCHT+PHT- vs. control groups), R2Y = 0.995 and Q2 = 0.989 (untreated IVCHT+PHT+ vs. control groups), R2Y = 0.996 and Q2 = 0.993 (untreated IVCHT-PHT+ vs. control groups), and R2Y = 0.998 and Q2 = 0.991 (untreated IVCHT-PHT- vs. control groups), which were stable and reliable for further prediction ( Figure 2B ). The four untreated B-CS groups also were clearly separated from the control group in the OPLS-DA scores plot ( Figure 2B ). These results indicated that B-CS was notably different from healthy controls at the serum metabolic level.

Compared with the serum metabolomics of controls, we identified 30 differential metabolites in untreated IVCHT+PHT-, IVCHT+PHT+, IVCHT-PHT+ and IVCHT-PHT- B-CS patients ( Figure 2C ). The levels of partially gut microbiota engaged synthesized metabolites, such as bile acids (e.g., taurocholate, glycocholic acid, glycochenodeoxycholate, tauroursodeoxycholic acid, and glycycodeoxycholic acid), short-chain aromatic fatty acid derivatives (e.g., (R)−3−hydroxybutyric acid), phenylalanine and derivatives (e.g., L−phenylalanine, phenyllactic acid, and 4-methoxyphenylacetic acid) were significantly increased, while purine metabolites (e.g., inosine, guanosine, and uracil), branch-chain amino acid (e.g., L-valine), tryptophan derivatives (e.g., serotonin) were significantly decreased in B-CS patients compared with that in healthy controls. These serum metabolites were significantly different at untreated B-CS, suggesting serum metabolites might be novel diagnostic biomarkers for the identification of B-CS and indicating there were sophisticated interactions among gut microbiota, serum metabolites and B-CS manifestations.

To understand the mechanism underlying B-CS, we further applied metabolomics pathway analysis (MetPA), and found that purine metabolism, ketone body synthesis and degradation, primary bile acid biosynthesis were significantly altered in untreated B-CS ( Figure 2B ). The results suggest that large scale pathway changes of metabolism were associated with the pathogenesis of B-CS.

The samples of The First Affiliated Hospital of Zhengzhou University were used as the training datasets, while the other was used as the test datasets. Training samples were used to obtain differential expressed metabolites and gut bacteria between B-CS patients and healthy controls to build RF models. For metagenomics, the sample size of training is 163, while the size of test is 52, while for metabolomics, the sample size of training is 158,while the size of test is 63. The optimized combinations of gut bacteria panel or metabolites panel were obtained ( Figures 3A, E ). The metabolomics results showed that both the training datasets and test datasets performed perfectly (area under the curve, AUC = 100%, 95% confidence interval, CI: 100% – 100%) ( Figures 3F–H ). We also found that metagenomics performed relatively poorly on test datasets (AUC = 58.48%, 95% CI: 38.46% – 78.5%) ( Figures 3C, D ), although the training datasets showed good performance (AUC = 100%, 95% CI: 100% – 100%) ( Figures 3B, D ). This indicates that the detection of serum metabolites is more advantageous for the identification of B-CSs.

In terms of variable importance, intestinal bacteria, including Clostridium leptum (mean decrease Gini = 3.505), Coprobacillus_sp_29_1 (mean decrease Gini = 0.794), and Parasutterella_excrementihominis (mean decrease Gini = 2.572), were the most associated with B-CS. Among the metabolomic variables, the highest ranked variables were taurocholate (mean decrease Gini = 9.723), azelaic acid (mean decrease Gini = 8.143) and (R)-3-hydroxybutyric acid (mean decrease Gini = 6.734). These findings indicate that significantly different gut microbiome and serum metabolites might contribute to B-CS progress.

We performed association analysis of intestinal bacteria and serum metabolites with phyla and species ( Supplementary Figure 2 ). The results showed that in the presence of hypertension in untreated B-CS individuals, including those in IVCHT+ PHT-, IVCHT- PHT+, and IVCHT+ PHT+ groups, the Campylobacter concisus population significantly increased and showed a significant positive correlation with taurocholate. In particular, taurocholate showed significant positive correlation with K00925: acetate kinase [EC: 2.7.2.1] of Campylobacter concisus in the IVCHT+ PHT+ population ( Figure 4 ). Further analysis of KEGG pathways revealed that taurocholate and dysregulated Campylobacter concisus are mainly associated with primary bile acid biosynthesis, fatty acid biosynthesis, and fatty acid metabolism, which are the key metabolic pathways involved in the development of liver diseases ( Supplementary Figure 3 ). Thus, Campylobacter concisus may affect taurocholate in serum, which in turn is involved in the development of hypertensive symptoms in patients with B-CS.

The clinical manifestations of B-CS primarily originate from IVCHT and PHT. However, the IVCHT-PHT- group had no corresponding clinical manifestations because of the opening of collateral circulation. We compared the IVCHT-PHT- group with other B-CS groups; the metabolomics analysis showed no significant differences in metabolites between the groups ( Supplementary Table 1 ). However, the results of metagenomics showed significant difference in gut microbiota levels between IVCHT-PHT- and IVCHT-PHT+ patients (PERMANOVA P = 0.049), and between IVCHT-PHT- and IVCHT+PHT- patients (PERMANOVA P = 0.020) ( Supplementary Table 2 ). At the phylum level, Proteobacteria were significantly enriched in IVCHT-PHT+ and IVCHT+PHT- patients. At the genus level, Lactobacillus was significantly enriched in IVCHT+PHT- patients, while Citrobacter was significantly enriched in IVCHT-PHT+ patients. At the species level, Bacteroides massiliensis was significantly enriched but Dorea formicigenerans was significantly decreased in IVCHT-PHT+, IVCHT+PHT- and IVCHT+PHT+ patients. Enterobacter cloacae was enriched, while Campylobacter hominis, Gemella unclassified and Succinatimonas hippei were decreased in IVCHT-PHT+ and IVCHT+PHT+ patients. Lachnospiraceae bacterium_3_1_57FAA_CT1 was prominently enriched but Lachnospiraceae bacterium_1_4_56_FAA was prominently decreased in IVCHT+PHT- and IVCHT+PHT+ patients ( Figure 5A ). The results demonstrate that further dysbiosis of the gut microbiota occurred with the emergence of hypertension in patients with B-CS.

To further investigate the relationship between B-CS hypertensive symptoms and intestinal microbiota dysbiosis, we compared patients with IVCHT+PHT+ with those with IVCHT-PHT+ and IVCHT+PHT-, respectively. At the phylum level, no significant difference was observed between the groups. At the genus level, Aggregatibacter was significantly enriched in patients with IVCHT+PHT+ than in patients with IVCHT-PHT+, Coprococcus and Bifidobacterium were significantly enriched in patients with IVCHT+PHT+ than in patients with IVCHT+PHT-. Catenibacterium was significantly decreased in patients with IVCHT+PHT+. At the species level, Haemophilus haemolyticus, Coprococcus comes, and Enterobacter mori were significantly enriched in patients with IVCHT+PHT+ than in patients with IVCHT-PHT+. Compared to that in patients with IVCHT+PHT-, Bifidobacterium longum, Eubacterium ventriosum, Turicibacter sanguinis, Aggregatibacter unclassified, and Lachnospiraceae bacterium_5_1_63FAA were significantly enriched in patients with IVCHT+PHT+. Catenibacterium mitsuokai was significantly decreased in patients with IVCHT+PHT+ ( Figure 5B ). To identify the microbiota associated with B-CS, we performed an intergroup gut microbiota analysis using with the linear discriminant analysis (LDA) effective size (LEfSe) method. The maximum difference in surtax was determined based on the LDA score ( Figure 5C ). At the same time, the microbial structure and main bacteria between the two were compared to draw an evolutionary branch diagram ( Figure 5D ). These results confirm that the composition of the intestinal microbiota of B-CS patients with different types of hypertension varied.

To understand the relationship more clearly, we performed KEGG functional prediction. Compared to those in IVCHT-PHT- patients, the pentose phosphate pathway, butanoate metabolism, glycolysis/gluconeogenesis, dioxin degradation, ABC transporters, glycine, serine and threonine metabolism related to carbohydrate, xenobiotics biodegradation and metabolism, membrane transport, and amino acid metabolism were significantly enriched in IVCHT-PHT+, IVCHT+PHT-, and IVCHT+PHT+ patients. Interestingly, compared to those in patients with IVCHT+PHT- and IVCHT-PHT+, pathways associated with methane metabolism, starch and sucrose metabolism, and drug-other metabolism and energy metabolism, carbohydrate metabolism, xenobiotics biodegradation and metabolism enzymes were significantly enriched in patients with IVCHT+PHT+ ( Figure 5E ). These findings further indicate that intestinal microbiota dysregulation deteriorates with disease progression.

We performed a comparative analysis of the pre- and post-invasive treatments of patients with B-CS (IVCHT-PHT+, IVCHT+PHT- and IVCHT+PHT+). We found no significant differences in the results of metabolomics ( Supplementary Table 3 and Figure 2A ) and metagenomics analyses ( Supplementary Table 4 and Supplementary Figure 1 ). However, certain differential bacteria were present in the intestinal microbiota of patients with B-CS. At the phylum level, Verrucomicrobia was significantly enriched in patients with IVCHT-PHT+ following treatment; Actinobacteria and Proteobacteria were significantly enriched in patients with IVCHT-PHT+ before treatment ( Figure 1A ). At the genus level, Akkermansia was significantly enriched in patients with IVCHT-PHT+ following treatment. Olsenella, Oxalobacter, Gordonibacter, Collinsella, Sutterella, Eikenella, and Selenomonas were significantly enriched in patients with IVCHT-PHT+ prior to treatment. Similarly, Burkholderiales noname, Blautia, and Clostridium were significantly enriched in patients with IVCHT+PHT- following treatment; Fusobacterium and Odoribacter were enriched in patients with IVCHT-PHT+ before treatment. Faecalibacterium, Ruminococcaceae noname, Peptostreptococcus, and Porphyromonas were significantly decreased in the IVCHT+PHT+ patients following treatment ( Figure 1B ). We observed that B-CS patients have high levels of beneficial bacteria, low levels of harmful bacteria, and improved gut microbiota following invasive treatment.

To determine the function of dysregulated bacteria, we performed a comparative analysis of untreated and treated B-CS patients using KEGG pathway analysis. Reduction in flagellar assembly, proteasome, peptidoglycan biosynthesis, Vibrio cholerae infection and pathogenic cycle, and terpenoid backbone biosynthesis were observed in treated patients with B-CS. At the same time, two-component system, propionate metabolism, phenylalanine metabolism, galactose metabolism, glyceride metabolism, fructose and mannose metabolism, amino sugar and nucleotide sugar metabolism, benzoate degradation, and PTS were decreased in treated patients with B-CS. Notably, flagellar assembly and terpenoid backbone biosynthesis were reduced in untreated patients compared to those in treated patients and healthy controls, while RNA polymerase and other glycan degradation were increased ( Figure 1D ). In addition, prior clinical data from treated-groups showed worse liver function during last invasive procedure (data not shown). These findings suggest that although current invasive treatments do effect the functional recovery of B-CSs with hyperbaric symptoms, some degree of intestinal dysbacteriosis persisted following B-CS treatment.

At the same time, we also focused on the pre- and post-treatment conditions of patients with B-CS in the IVCHT-PHT- group. Clinically asymptomatic individuals who did not require invasive treatment, for the time being, are partially administered invasive treatment. No significant differential bacteria were observed at the phylum level. At the genus and species levels, fewer discrepant microbes were detected compared to those in the other groups. This suggests that although invasive treatments improved the gut microbiota of the patients in the IVCHT-PHT- group, gut dysbiosis persisted following treatment.