Changes in Microbiota and Metabolites Are Associated With Effects of Bile Acid Sequestrant on Icteric Primary Biliary Cholangitis


 BackgroundIncreasing data suggests an interaction between bile acids (BAs) and intestinal microbiota in the pathogenesis of primary biliary cholangitis (PBC). Bile acid sequestrants are widely used to bind BA in the intestinal lumen and therefore are posited to impact gut bacteria. Herein we aimed to investigate the effects of cholestyramine on the BA profile and gut microbiome in a cohort of icteric PBC patients. Results Thirty-three PBC patients were treated with cholestyramine and serum and stool samples were collected at baseline, 4 and 16 weeks. Shotgun metagenomic sequencing, targeted metabolomic profiling were performed. Following cholestyramine administration, patients exhibited a high inter-personal variability in remission of cholestasis, and were therefore dichotomized according to the decrease of total bilirubin. In general, BA profiling revealed a shift toward a more hydrophilic milieu, along with decreased circulating fibroblast growth factor 19 and increased active glucagon-like peptide 1. Gut microbial co-abundance networks showed distinct taxa interactions in subjects with superior remission (SR) and those with inferior remission (IR) at baseline. Compositional shifts of the microbiome in SR group were characterized with enrichment of two Lachnospiraceae species, typically producing short-chain fatty acids (SCFAs). In contrast, Klebsiella pneumonia, a commensal pathobiont, was only increased in IR group. Correspondingly, patients with SR, but not IR, were marked by elevations of SCFAs including valeric acid and caproic acid. Finally, integrative analysis identified robust associations between the variations of microbiota, metabolites, and inflammatory cytokines in SR group, indicating potential mechanistic links.ConclusionsBeneficial responses caused by cholestyramine were closely related with compositional and functional alterations in gut commensal, highlighting the possibility of exploring bile acids-microbiota interactions for treating PBC.


Introduction
Primary biliary cholangitis (PBC) is an important but uncommon autoimmune liver disease characterized by the presence of antimichondrial antibodies (AMA) and progressive destruction of interlobular bile ducts [1] . Ursodeoxycholic acid (UDCA), a choleretic bile acid, is successful in approximately two thirds of early-stage PBC patients and improves life expectancy without additional therapies [2,3] . Emerging evidence suggests a critical role for the excessive toxic bile acids (BAs) and gut dysbiosis in the pathogenesis of PBC. [1,3] Interactions between bile acids and gut microbiome have been well estabilished. By acting on farnesoid X receptor (FXR) and Takeda G protein-coupled receptor 5 (TGR5), BAs are involved in multiple signaling pathways, including metabolism, brosis and immune homeostasis [4,5] . Gut microbiota metabolize BAs with de ned enzymes, and thereby impact the BA signalings [4] . For example, supplementation of Lactobacillus rhamnosus prevented liver brosis in a mouse model of cholestasis through upregulating intestinal FXR signaling [6] . In turn, dynamics of BAs also exert a profound impact on the intestinal microbiome. The use of a BA analogue, obeticholic acid (OCA), led to a reversible induction of grampositive bacteria in rodents and human [7] . Of note, our previous work revealed the presence of a microbial perturbation in naïve PBC patients; and UDCA treatment can partially reverse this dysbiosis [8,9] . In addition, levels of secondary BAs were inversely correlated with PBC-enriched gut bacteria [10] .
Nevertheless, how alterations of BA modulate gut microbiota in PBC remains elusive.
Cholestyramine is one of the BA sequestrants capable of binding intestinal lumen BAs and have been used to treat cholestatic pruritus with a good safety pro le and accessibility [11] . In the Mdr2 knockout mouse model, BA sequestrants alleviate cholestatic liver and bile duct injury [12] and, in humans with primary sclerosing cholangitis (PSC), there are case reports of improvement of cholestasis by cholestyramine [13] . Here, we performed a 16-week longitudinal study in icteric PBC subjects using cholestyramine to characterize the compositional and functional responses of gut microbiota to alterations in endogenous BA levels. Multi-omic analysis including shotgun metagenomic sequencing and targeted metabolomic pro ling were utilized. We further investigated whether these changes in BAs and microbiota could explain the bene cial effects of cholestyramine on cholestasis.

Study subjects
All subjects were enrolled from the outpatient clinical of Renji Hospital, a liated to Shanghai Jiao Tong University School of Medicine between January 2017 and ending March 2018. The mean age of the subjects was 48.8 years (SD 9.1), and 27/33 (81.8%) were women. All patients took a standard dose of 13-15mg/kg/d UDCA at baseline and throughout the study. The mean baseline total bilirubin (TB) level was 95.09 μmol/L (Supplementary Table 1).
Patients were enrolled on a consecutive basis if eligible. All patients were adults with a con rmed diagnosis of PBC and severe cholestasis and all had been treated with standard UDCA therapy for at least 6 months. Exclusion criteria included malignancy, renal dysfunction, pregnancy or lactation. None of the patients had previously received cholestyramine. Patients were administrated twice-daily cholestyramine at a dose of 8g each time prior to meals, with a 4-hour window before administration of other medications, to avoid drug interference in intestinal absorption of UDCA. Liver biochemistries were performed at baseline and weeks 4 and 16. Serum and stool samples were collected on each visit. Thirtythree patients received cholestyramine treatment for 4 weeks, 28/33 received cholestyramine for 16 weeks, and 5/33 patients discontinued the study voluntarily after the 4 week sample period because they were unable to tolerate the "taste" of the cholestyramine.

Strati cation of the subjects
For further analysis, patients were strati ed into two groups according to changes of total bilirubin following 16-week treatment of cholestyramine. Subjects with a delta percent (△%) of TB higher than the median were allocated to the group with superior remission (SR), while subjects with △% of TB lower than the median were assigned to the inferior remission (IR) group.

Sample preparation
Patients were required to fast overnight before collection of blood samples on the morning of each visit.
Blood samples were centrifuged at 2800 rpm for 15 minutes at 4°C and serum were aliquoted and stored at -80°C until analysis. Fecal samples were freshly collected and immediately frozen at -80°C.
Brie y, 180 μL of acetonitrile/methanol (8:2) containing 10 internal standards was added to 20 μL of serum samples in a 96-well plate. The mixture was then vortexed at 1500 rpm for 2 min at 10 °C and centrifuged at 13000 rpm for 20 min at 4 °C. The supernatant was transferred to another plate and vacuum-dried. The residues were reconstituted with equal volume of acetonitrile/methanol (8:2) and water, and then centrifuged at 13000 rpm for 20 min at 4 °C. After centrifugation, the supernatant (5 μL) was injected into the system for analysis. UPLC-MS/MS raw data were obtained and processed using MassLynx 4.1 software (Waters Corp., Milford, MA, USA).

UPLS-MS/MS pro ling of serum microbiota-derived metabolites
Quanti cation of the microbial metabolites was performed by Metabo-pro le (Shanghai, China) using a UPLC-MS/MS system. All targeted standards were obtained from Sigma-Aldrich (St. Louis, MO, USA). The serum samples were processed as previously described [16] . Brie y, 25 μL serum samples were extracted with 100 μL cold methanol by centrifuging at 4000g for 30 min. Then 30 μL supernatant underwent derivatization at 30 °C for 60 min and was subsequently diluted with cold 50% methanol and stored at -20 °C for 20 min. After that, the mixtrure was centrifuged at 4000g for 30 min, and 135 μL of supernatant was transferred to a new 96-well plate with 15 μL internal standards and nally subjected to LC-MS analysis. To ensure reproducibility, the quality control samples were prepared with the test samples and injected at every 14 test samples throughout the process.

DNA extraction and metagenomic sequencing
Genomic DNA was extracted from feces using QIAamp PowerFecal DNA Kit (Qiagen, USA). DNA concentration was quantitated using a Quantus Fluorometer (Promega, CA, USA) and quality checked on a 1% agarose gels electrophoresis system. Whole-genome shotgun sequencing libraries were prepared using a TruSeq TM DNA Sample Prep Kit (Illumina, San Diego, CA, USA). Individual libraries were pooled and then sequenced on a HiSeq 4000 platform (Illumina, CA, USA), using a 150-bp paired-end read protocol.
Metagenomic sequencing data processing Raw sequencing reads were rst quality ltered using KneadData (version 0.7.2). Brie y, low quality reads were trimmed with Trimmomatic, setting the minimum length to 50% of the total input read length. Human DNA reads were removed with Bowtie 2. Taxonomic pro les of quality-ltered metagenomes were generated using MetaPhlAn2 (version 2.7.7) with default parameters [17] . In our study, species-level data were considered and reported as relative abundance. Only species with relative abundance higher than 0.001% and present in at least 20% of the total samples were kept for further analysis. Functional pro ling was performed using HUMAnN2 (version 0.11.2) and summarized as KEGG (Kyoto Encyclopedia of Genes and Genomes Statistical analysis) pathways, which were normalized to counts per million (CPM) [18] . Pathways present in less than 20% of the samples were not included in the analysis.

Statistical analysis
Statistical analyses were performed using R packages (version 3.5.2) or SPSS (version 24). Alphadiversity was reported as Shannon index, using diversity function from vegan package. Beta-diversity was based on Bray-Curtis dissimilarities of taxonomic species assessed by vegdist function from the vegan package. A two-tailed Wilcoxon signed-rank test was used to analyze paired groups for bile acids, microbial metabolites and clinical data. Longitudinal generalized linear mixed models (GLMMs) were constructed for metagenomic taxa and cytokines using the lmer function from package lmerTest. Time (week) was set as xed effect and subject as a random effect in the longitudinal GLMMs. The relative abundances of species were centered log ratio transformed before analysis. For correlation analysis, we applied repeated measures correlation (rmcorr) methods to test for associations between variables (bile acids, microbial taxa, metabolites and cytokines) within each subject. The Benjamini-Hochberg method was used to adjust p values for multiple testing.

Co-abundance network analysis
We constructed microbial co-abundance network for samples at baseline and after 16-weeks of cholestyramine, respectively. SparCC was applied to compute the microbial correlations between species with relative abundance higher than 0.001% and present in at least 20% of the samples [19] . The correlation values larger than 0.3 or smaller than -0.3 with p value <0.05 were retained for network visualization in Cytoscape [20] .

Variability in biochemical responses to the BA sequestrant
Patients were administrated cholestyramine for 16 weeks and provided serum and fecal samples at the timepoints of week 0, 4, 16 ( Figure 1A). A signi cant reduction of TB was observed at 4 weeks (paired Wilcoxon rank-sum test, p<0.001, 4 weeks vs baseline; Figure 1B) and the TB level further declined during follow-up visits (paired Wilcoxon rank-sum test, p<0.0001,16 weeks vs Baseline; p<0.01, 16 weeks vs 4 weeks; Figure 1B). In parallel, levels of alkaline phosphatase (AKP) and γ-glutamyl transpeptidase (GGT) declined compared with baseline, to a modest but signi cant extent ( Figure 1C, D). Other details of liver biochemistry changes are shown in Supplemantary Table 1.
We observed a high inter-personal variability in the amelioration of hyperbilirubinemia, which suggested a heterogeneous responses to the treatment of cholestyramine. As such, subjects were further strati ed according to the median decrease of bilirubin at 16 weeks, i.e., group with superior remission of cholestasis (SR, n=14) and group with inferior remission (IR, n=14). The baseline comparison of these two groups is shown in Supplementary Table 2. As expected, patients in SR group exhibited more reduction of TB, AKP and GGT comparing with patients in IR group (Figure1 E-G). A concordant decrease of peripheral in ammatory cytokines was observed in patients in SR group following treatment of BA sequestrant ( Figure 1H-K). These cytokines, including monocyte chemoattractant protein-1 (MCP-1), interleukin-6 (IL-6), interleukin-8 (IL-8) and interleukin-18 (IL-18), are overexpressed in PBC and contribute to the reactive phenotype of cholangiocyte [21] .
BA pro le and canonical BA signaling were altered Overall, treatment of cholestyramine markedly reduced circulating BA level and altered its composition (Figure 2A, B). Of note, the reduction of serum total BA is more prominent in the SR group, in line with the better amelioration of cholestasis ( Figure 2C). The ratio of hydrophobic/hydrophilic BA declined in both subgroups ( Figure 2D), due to the prior sequestration of hydrophobic BAs including TCDCA and GCDCA [22] . In addition to shifting the polarity of the circulating BA pool, cholestyramine treatment also lowered the ratio of taurine/glycine BA in two groups ( Figure 2E). The serum ratios of unconjugated/conjugated BA or secondary/primary BA were essentially unchanged (data not shown).
Next, canonical signaling of BA was examined. The level of FGF19 reduced profoundly following the BA sequestrant treatment in all the subjects (paired Wilcoxon rank-sum test, p<0.0001, 4 weeks vs baseline; p<0.001, 16 weeks vs baseline, Figure 2F). A reduction in circulating FGF19 level indicates an impaired activation of the ileal FXR signaling and thereby less suppression of CYP7A1, the key rate-limiting enzyme of BA synthesis in liver. In contrast to FGF19, we observed an increase in the secretion of active glucagon-like peptide-1 (GLP-1) (paired Wilcoxon rank-sum test, p<0.01, 16 weeks vs baseline, Figure 2G). GLP-1 is one of the two identi ed incretins maintaining glucose homeostasis and is likely produced by enteroendocrine L cells residing in the distal intestine. Activation of TGR5 on enteroendocrine L cell triggers the release of GLP-1 [23] . In line with that, increase of GLP-1 has also been observed in Mdr2 -/mice treated with BA sequenstrant and conferred the cholangioprotective effects [12] . Indeed, GLP-1 and its analogue extendin-4 were proposed to prevent cholangiocyte from apoptosis and facilitate its proliferative repair in response to cholestasis [24,25] . Patients in the two subgroups exhibited similar changes in BA signaling in response to cholestyramine ( Figure 2H, J), suggesting a comparable sequestration of BA in the intestine.

BA sequestrant modulated gut microbiota
We performed whole-genome shotgun sequencing on fecal samples and obtained an average of 25.5 million paired-end reads per sample (min: 19.6 million, max: 30.7 million) after quality ltering.
Longitudinal generalized linear mixed model (GLMM) was constructed for microbiome analysis. First, alpha diversity did not alter in response to the treatment of cholestyramine ( Figure 3A). Principal coordinate analysis (PCoA) of Bray-Curtis dissimilarity was used to evaluate the global compositional shift of the gut microbiome. Compared to the baseline, there were no overall alterations in the microbial composition after the intervention (PERMANOVA, p>0.05; Figure 3B), possibly due to the similar abundance of the most abundant species among these groups (Supplementary gure 1). Nevertheless, four species suggested time-dependent alterations in response to cholestyramine, using longitudinal generalized linear mixed models (GLMMs, fdr<0.2, Figure 3C, Supplementary table 3). These included an increase in two Lachnospiraceae species (3146FAA and 1157FAA), known to be involved in production of short-chain fatty acids (SCFA), as well as a decrease in Roseburia intestinalis, which was reported to express mimotopes and trigger autoimmunity via cross-reactivity. Interestingly, we found a negative correlation between the changes of total bilirubin and these two Lachnospiraceae species in our PBC cohort (Supplementary Figure 2). Surprisingly, the abundance of Klebsiella pneumoniae, found increased in UDCA-naïve PBC in our previous study, signi cantly increased following cholestyramine treatment. We also examined functional changes in the gut microbiome. However, no pathways achieved statistical signi cance after correction for multiple comparisons (data not shown).

Divergent alterations of gut microbiota between patients with superior remissioin (SR) and inferior remission (IR)
Given the high inter-individual variability in the gut microbial taxa, we sought to determine whether the divergent responses to cholestyramine were associated with differential alterations in gut microbiome.
No obvious difference in alpha or beta diversity before and after intervention was found in patients of group SR or IR (Supplementary gure 3). At the taxonomic level, the species Lachnospiraceae 3146FAA, which demonstrated the greatest alteration in response to cholestyramine in the whole group, increased in patients of SR group (GLMMs, P=0.00039), whereas it remained unchanged in IR group ( Figure 3D). Likewise, the alterations of Lachnospiraceae 1157FAA and Roseburia intestinalis were detected only in SR group (P<0.05, Figure 3D). In contrary, the signi cant increase of Klebsiella pneumoniae found in all the participants occurred in IR group (P<0.05, Figure 3D), but not SR group.
Gut microbial co-abundance network suggested difference between SR and IR groups We next performed co-abundance network analysis using SPARCC to interrogate the microbial relationships in the cohort ( Figure 3E). Interestingly, very few co-abundance associations were found in the ecosystem of all the samples before and after treatment (Supplementary Figure 4), but stratifying the participants into two subgroups resulted in considerably more complex networks ( Figure 3E). More importantly, distinct patterns of microbial interactions existed between the SR and IR groups. At baseline, the connectivity in SR group was dominated by taxa in the phylum Bacteriodetes, whereas enhanced asscociations between taxa in Firmicutes were present in IR group. After intervention, the density of the networks decreased in both groups with a lower average degree, while SR group appeared to have more nodes and connections in comparison with IR group. Additionally, the species with the highest degree centrality in the networks, including Bacteroides thetaiotaomicron and Prevotella copri, were found to be differentially altered in response to the cholestyramine between the two subgroups (Supplementary Figure 5). P. copri, associated with several chronic in ammatory diseases such as rheumatoid arthritis, were increased only in IR group. Taken together, these ndings indicate that the discrepancies in species co-abundances may underlie the variability of responsiveness to the resin.

Distinct changes of microbial metabolites in SR and IR groups
The gut microbiota constantly produce large amounts of metabolites, which can enter the circulation and act as important signaling moleculars at the extraintestinal organs. To further understand how bacteria impact host physiology, we performed microbiota-related metabolomics analysis of serum samples at baseline and after 16-week treatment of cholestyramine. Overall, distinguishable shifts in the composition of metabolites were observed before and after treatment (R 2 Y=0.65, Q 2 Y=0.165, Figure 4A). Speci cally, 14 metabolites were upregulated and 4 were downregulated (median fold change>1.25 or <0.8, p<0.05, Figure 4B). It was worth noting that several SCFAs were increased in response to the intervention ( Figure 4B, 4C). More importantly, patients in two subgroups exhibited differential metabolic alterations (Supplementary Figure 6). In particular, concentrations of valeric acid and caproic acid were selectively upregulated in SR group, consistent with the enrichment of SCFA-producing bacteria observed in this group ( Figure 4D, E). These metabolic difference between the two subgroups and its consistency with taxonomic changes provide further evidence that the microbial alterations are biologically relevant with the cholestyramine treatment.
Changes of bile acids, gut microbiome, metabolites and in ammation were differentially correlated in SR and IR group We next performed a repeated measure correlation analysis (rmcorr) to explore the potential connections of the alterations in bile acids, microbial taxonomy, metabolism and host in ammation following cholestyramine treatment. Multiple associations were identi ed when analyzing patients in SR group, while few connections were observed in IR group. (Figure 5A). In particular, we found strong associations between changes of bile acids and gut bacteria in SR group ( Figure 5A). Furthermore, signi cant increases of Lachnospiraceae species, known for producing SCFAs, correlated with increased levels of valeric acids and caproic acids, as well as reductions of in ammatory markers including IL-18 and MCP-1 in this group ( Figure 5B). In addition, SCFAs including valeric acids and isovaleric acids inversely correlated with systemic in ammation only in SR group ( Figure 5B).

Discussion
By modulating endogeneous bile acid through cholestyramine in a cohort of patients with progressive PBC, we identi ed potential links between bile acids dynamics, compositional and metabolic changes of gut commensal, and amelioratioin of in ammation and PBC-related cholestasis. A high variability in the remission of cholestasis in PBC following intervention was related to the divergent alterations in microbial taxonomy and metabolites. Of note, subjects with relatively good remission exhibited enrichment of Lachnospiraceae species and concordant elevations of SCFAs, which mediated, in part, the anti-cholestatic and anti-in ammatory effects of cholestyramine.
In addition to the decrease of total serum BAs upon treatment, serum BA pro ling showed a signi cant shift towards a more hydrophilic con guration. Hydrophobic bile acids are known to elicit apoptosis and senescence of biliary epithelial cells [1] . Sequestration of intestinal BAs downregulated the levels of FGF19, leading to derepression of CYP7A1 and upregulation of BA synthesis, which is in part evidenced by the reduced levels of cholesterol in our PBC cohort. This is also supported by a study of healthy volunteers in which treatment for 3 weeks with cholestyramine led to reduced FGF19 levels but increased levels of C4, a well-established serum marker of BA synthesis in liver [11] .
Previous work on both rodents and humans consistently suggests that one pharmalogical effect of BA sequestrant is to augment intestinal activation of the BA receptor, TGR5, and thus lead to the release of the downstream hormone GLP-1 [12,23,[26][27][28] . As expected, treatment of cholestyramine induced the release of GLP-1 in our PBC cohort. Mechanistically, although minimal levels of BAs are capable of reaching the colon, sequestration by a resin would allow an ectopic concentration of BAs in colon, where TGR5 and GLP-1-producing enteroendocrine L-cells are highly expressed [26,28] . GLP-1 and its analogue extendin-4 can prevent cholangiocyte from apoptosis and facilitate its proliferative repair in response to cholestasis [24,25] . Therefore, BA sequestrant applied in this PBC cohort probably act to augments intestinal TGR5 signaling and colonic production of GLP-1, which then exert cholangioprotective effects. However, changes in the two canonical BA signalings as well as the "hydrophilisation" of the BA were comparable in both SR and IR groups, thus the divergence of the biochemical responses can not be fully explained by BA alterations.
Impacts of BAs on the intestinal microbe are complex. The modest alteration at taxonomic level and no differences in the alpha and beta diversity indicated a relatively stable gut microbial community of the samples during the treatment, which was consistent with the results obtained from Mdr2 knockout mice treated with BA sequestrants [12] . Nonetheless, further stratifying the patients according to their clinical responses provided more clues. It is worth noting that the abundance of two species in Lachnospiraceae family were exclusively enriched in SR group. The Lachnospiraceae family is known for producing SCFAs, which is generally recognized as one of the most important immunoregulatory molecules metabolized by microbiota [29] . On the contrary, Klebsiella pneumoniae, the pathobiontfound to cause gut barrier damage and T helper 17 (Th17) cell immune response in PSC [30] , was elevated in IR group.
It was suggested that subtle changes in microbial community were able to shift its function profoundly [31] . As microbial metabolites were regarded as readouts of their function, we further performed targeted metabonomic pro ling of microbiota-derived metabolites. Interestingly, SCFAs including valeric acid and caproic acid, were increased in SR group. Similar to butyric acid, valeric acids and caproic acids also provide energy for intestinal epithelium and exert anti-in ammtory effects in intestinal and systemic immune diseases [32][33][34] . It has been recently reported that valerate provides protection against colitis and multiple sclerosis via promoting Breg differentiation and meanwhile suppressing Th17 cells [32] .
With regards to the associations among the variations of microbiota, metabolites, and in ammatory cytokines, substantial differences were found between the two subgroups. SR group possessed a considerably larger number of connections, compared with IR group. Of particular interest was the nding that the enrichment of the Lachnospiraceae species was strongly correlated with the increase of circulating valeric acid and caproic acid, as well as the accompanying decrease of in ammatory cytokines, providing further evidence that SCFAs produced by the altered microbiota in SR group act to mitigate the in ammation and cholestasis.
The indiscriminative shift of BA pro le and its downstream signaling between the two subgroups could not explain the heterogeneity of the responses. Although there was no obvious difference in baseline bacterial taxonomy in our study, we noticed that the microbial relationships distinguished between the SR and IR groups, which was evidenced by the co-abundance networks of taxa associations. Moreover, the strati ed subgroups showed substantially more complex networks than the whole group, implying the microbial ecosystems were virtually not identical among all the subjects. The human intestine harbours a huge number of microbes interacting intricately with each other. Previous studies have suggested that microbial interactions are not only essential for maintaining healthy ecology, but also implicated in disease-associated states [35][36][37][38][39] . In this regard, the differential microbial networks in SR and IR group may help explain the variable adaptation to the BA uctuation. Therefore, in addition to the microbial composition, it is critical to characterize the interactions between the microbes and determine the underlying mechanisms.
There are several limitations in the study. First, cholestyramine is a traditional BA sequestrant with a taste intolerant for part of patients. However, second generation of BA sequestrants are currently unavailable in China and cholestyramine was thus employed to sequestrate intestinal BAs in an effort to investigate the BA-microbiota crosstalk. Second, the study is of relatively modest sample size. Nonetheless, PBC is not a common disease and the UDCA treatment is effective in more than half of patients. Third, ndings in this work are correlational, as the differential responses of human gut microbiome to cholestyramine provide a potential explanation for the inconsistent clinical effects. Future study is needed to elucidate the underlying causality.

Conclusion
In conclusion, we provide a unique perspective into the dynamic changes of the gut microbiome in response to bile acid modulation of PBC-related cholestasis. Patients administrated cholestyramine demonstrated heterogeneous but overall advantageous responses, which were largely mediated by gut commensal. Given the suboptimal therapeutic strategies for progressive PBC, this real-world study highlighted the possibilities for implementation of microbiota or its metabolites-targeted treatment in future.    Distinct changes of microbial metabolites between SR and IR groups. (A) OPLA-DA score plots depicted the shift of circulating bacterial metabolites before and after the 16-week treatment of cholestyramine (R2Y=0.65, Q2Y=0.165). (B) Volcano plots showed individual metabolites altered in response to the intervention (p<0.05). Dots in red and blue denote the up-regulated and down-regulated metabolites, respectively. (C) Heatmap showed the median fold changes of short-chain fatty acids (SCFAs) concentrations in SR and IR groups before and after the 16-week treatment. # p< 0.05 by paired Wilcoxon rank-sum test within group SR. * p< 0.05, ** p< 0.01 by paired Wilcoxon rank-sum test in all subjects. (D,E) Differential changes of valeric acid and caproic acid in SR and IR groups. * p< 0.05 by paired Wilcoxon rank-sum test within group SR or IR. Figure 5