Primary sclerosing cholangitis (PSC) is a chronic inflammatory liver disease and its frequent complication with ulcerative colitis highlights the pathogenic role of epithelial barrier dysfunction. Intestinal barrier dysfunction has been implicated in the pathogenesis of PSC, yet its underlying mechanism remains unknown. Here, we identify Klebsiella pneumonia in the microbiota of patients with PSC and demonstrate that K.pneumoniae disrupts the epithelial barrier to initiate bacterial translocation and liver inflammatory responses. Gnotobiotic mice inoculated with PSC-derived microbiota exhibited T helper 17 (TH17) cell responses in the liver and increased susceptibility to hepatobiliary injuries. Bacterial culture of mesenteric lymph nodes in these mice isolated K.pneumoniae, Proteus mirabilis and Enterococcus gallinarum, which were prevalently detected in patients with PSC. A bacterial-organoid co-culture system visualized the epithelial-damaging effect of PSC-derived K.pneumoniae that was associated with bacterial translocation and susceptibility to TH17-mediated hepatobiliary injuries. We also show that antibiotic treatment ameliorated the TH17 immune response induced by PSC-derived microbiota. These results highlight the role of pathobionts in intestinal barrier dysfunction and liver inflammation, providing insights into therapeutic strategies for PSC.


Primary sclerosing cholangitis (PSC) is a chronic cholestatic liver disease characterized by the development of bile-duct strictures and the destruction of biliary trees that lead to end-stage liver cirrhosis1,2,3,4. Although the pathophysiology of PSC remains unknown, a frequent observation of portal bacteraemia5 and elevated levels of endotoxin in cholangiocytes6 suggested the implication of gut microbiota in PSC. Although gut microbiota is normally confined in the intestinal lumen, once the epithelial barrier is disrupted, the influx of gut microbiota can elicit T helper 17 (TH17) immune response, resembling the liver inflammation in PSC7. As >70% of PSC is complicated with ulcerative colitis (UC)1,2, colonic inflammation is a plausible cause of the barrier dysfunction; however, the majority of patients with UC are devoid of PSC8,9, suggesting an additional layer of complexity. Recently, it is becoming apparent that PSC-associated UC (PSC/UC) shapes a distinct disease phenotype: PSC/UC typically exhibits discontinuous right-sided colitis, whereas classical UC is characterized by the continuous lesions originating from the rectum10,11. Interestingly, the gut microbiota of PSC/UC differed from those of classical UC and was rather similar to those of PSC without UC lesions12,13,14. Thus, it has been suggested that the dysbiosis contributed to the distinct disease phenotype of PSC/UC; however, it remains unexplored whether the emergence of specific bacterial species could induce bacterial translocation and subsequent progression to hepatobiliary inflammation. In this study, we generated gnotobiotic mice using faecal samples from patients with PSC/UC and identified the specific bacteria that are responsible for pathological bacterial translocation and subsequent TH17 priming in the liver. Furthermore, using in vivo models and a two-dimensional organoid culture system, we have revealed a pathogenic mechanism by which multiple bacterial species collaboratively disrupt the intestinal epithelial barrier and induce TH17 priming in the liver.


TH17 priming is induced in the livers of gnotobiotic mice inoculated with faecal samples from patients with PSC/UC

To determine the biological effects of gut microbiota on liver inflammation, we generated gnotobiotic mice by inoculating faecal samples derived from patients with PSC/UC (PSCUC mice), patients with UC (UC mice) or healthy controls (HCs; HC mice) (Fig. 1a). Although the overall microbiota composition in human donors was not completely recapitulated in gnotobiotic mice, correlation analysis of principal coordinates from unweighted UniFrac distance indicated that dominant microbial taxonomy was conserved between human donors and corresponding gnotobiotic mice (Supplementary Fig. 1a,b). After stable colonization of the human gut microbiota, we examined T cell profiles in the livers and colons of the gnotobiotic mice (Fig. 1b,c). As previously described15, TH17 priming was induced in the colon, but not in the liver, of HC mice (Fig. 1d,e). Interestingly, three out of five PSC/UC patient-derived mouse groups exhibited potent TH17 priming in the liver (Fig. 1e). We also observed significant upregulation of inflammatory genes, including serum amyloid A and interleukin-1β (IL-1β) in both the colons and livers of PSCUC mice (Supplementary Fig. 1c,d), suggesting that the PSC/UC patient-derived microbiota induced liver inflammation.

Fig. 1: Magnitude of TH17 priming in the liver and colon of gnotobiotic mice transplanted with human faecal samples.
Fig. 1

a, Study design: faecal samples derived from healthy controls (HC mice), patients with PSC/UC (PSCUC mice) or patients with UC (UC mice) were inoculated into GF mice, followed by immunological phenotyping at day 28 post-inoculation. The yellow triangles represent the time point of analysis. b,c, Representative intracellular IFN-γ and IL-17 staining of CD3+CD4+-gated cells in the colon (b) and liver (c) of mice from the indicated groups (GF: n = 7; HC: n = 16; PSCUC: n = 21; and UC: n = 20 mice). d,e, Frequency of IL-17+ (TH17) cells (upper) and IFN-γ+ (TH1) cells (lower) in CD3+CD4+-gated cells in the colon (d) and liver (e) of mice inoculated with faecal samples from HC1–4, PSC/UC1–5 and UC1–5 individuals, compared with GF mice (GF: n = 7; H1–H4, P1, P3–P5 and U1–U5: n = 4; and P2: n = 5 mice). Data show the mean ± s.e.m. Data are combined from several independent experiments.

PSCUC mice are more susceptible to 3,5-dicarbethoxy-1,4-dihydrocollidine-induced hepatobiliary damage

Despite prominent TH17 priming, faecal transplantation was insufficient to induce histological or serological changes in the liver of PSCUC mice even after a long colonization period of 90 days (Supplementary Fig. 2a–d). Thus, we questioned whether dysbiosis could promote 3,5-diethoxycarbonyl-1,4-dihydrocollidine (DDC)-fed hepatobiliary injury that has been used as a PSC disease model16,17,18 (Fig. 2a). Although DDC induced rather mild hepatobiliary injury under germ-free conditions, gnotobiosis of HC microbiota enhanced the disease activity, suggesting an implication of gut microbiota in the progression of liver inflammatory lesions (Fig. 2b). Interestingly, DDC-fed PSCUC mice showed a higher level of serum total bilirubin and alkaline phosphatase, and TH17 response in the liver than DDC-fed germ-free (GF) or HC mice (Fig. 2b–d). Pathological assessment revealed periductal fibrosis in PSCUC mice, a characteristic feature of PSC (Fig. 2e,f). We also confirmed significant upregulation of fibrosis marker gene expression in DDC-fed PSCUC mice (Fig. 2g). These observations provided the first prospective evidence that gut microbiota from PSC/UC promotes the hepatobiliary injury along with the liver TH17 response.

Fig. 2: Increased susceptibility to DDC-induced hepatobiliary injury in PSCUC mice.
Fig. 2

a, Study design: GF mice, GF mice inoculated with HC faecal samples (HC mice) or PSC/UC faecal samples (PSCUC mice) were administered 0.05% DDC for 2 weeks from day 21 to day 35. The yellow triangles represent the time point of analysis. b, Serum total bilirubin (TB) (top) and alkaline phosphatase (ALP) (bottom) levels of mice from the indicated groups. c,d, Representative intracellular IFN-γ and IL-17 staining of CD3+CD4+-gated cells (c), and the frequency of IL-17+ cells (left) and IFN-γ+IL-17+ cells (right) in CD3+CD4+ T cells (d) in the livers of mice from the indicated groups. e, Representative photomicrographs of Sirius Red (upper) and Masson trichrome (lower) staining of the liver sections of mice. Scale bars, 100 µm. f, Quantitative Sirius Red-positive area of the liver sections of mice from the indicated groups. g, qRT–PCR analysis of Col1a1 (left) and Timp1 (right) relative to Gapdh in the whole liver of mice from the indicated groups. The numbers in each graph indicate the ratio compared to GF mice. For bg, n = 4 mice in the GF group and n = 5 mice in the other groups. For b, d, f and g, data show the mean ± s.e.m. ANOVA using Tukey’s multiple-comparison correction was applied. Data are representative of two independent experiments.

Gnotobiotic mice with PSC/UC microbiota identified specific bacterial species associated with PSC/UC

Gnotobiotic PSCUC mice showed increased levels of serum endotoxin, indicating bacterial translocation (Fig. 3a). To identify the specific pathobionts responsible for bacterial translocation, we harvested the livers, mesenteric lymph nodes (MLNs) and spleens from gnotobiotic mice (Fig. 3b). Strikingly, we successfully isolated bacterial clones from the MLNs, whereas no bacteria could be grown from the liver or spleen of PSCUC gnotobiotic mice, or from any organs of specific pathogen-free (SPF), HC or UC gnotobiotic mice (Fig. 3b). Sequencing of 16S rRNA identified these bacteria as Klebsiella pneumoniae, Proteus mirabilis and Enterococcus gallinarum. To validate the capability of bacterial translocation among these three bacteria, we reinoculated a mixture of the three bacteria (hereafter, 3-mix bacteria) into GF mice. Consistent with the result of PSCUC mice, gnotobiotic mice with the 3-mix bacteria (3-mix gnotobiotic mice) exhibited efficient bacterial translocation into the MLN but not into the liver as determined by bacterial culturing, PCR and fluorescence in situ hybridization (FISH) (Supplementary Fig. 3a,b). The failure of bacterial translocation detection in the liver was not due to a low sensitivity, because translocated bacteria were detected in the liver from mice that had been administered 3-mix bacteria by intravenous injection. These results suggested that 3-mix bacteria are sufficient to induce bacterial translocation in gnotobiotic mice.

Fig. 3: Gnotobiotic mice with PSC/UC microbiota identified specific bacterial species associated with PSC/UC from the MLN.
Fig. 3

a, Serum LPS concentration. n = 3 mice in the GF group and n = 4 mice in the SPF, HC, PSCUC and UC groups. ND, not detected. Data show the mean ± s.e.m. ANOVA using Tukey’s multiple-comparison correction was applied. b, Study design: the liver, MLN and spleen were aseptically collected from the mice of each group on day 21 post-inoculation, followed by in vitro culture. The yellow triangles represent the time point of analysis. Representative photographs of the agar plates on the second day after colonization with the collected MLN are also shown. c, Amount of K.pneumoniae (upper), P.mirabilis (middle) and E.gallinarum (lower) in faecal samples from HC1–10, PSC/UC1–18 and UC1–16 individuals assessed by 16S qPCR. d, Amount of K.pneumoniae (upper), P.mirabilis (middle) and E.gallinarum (lower) in faecal samples from AIH1–9 and PBC1–10 individuals assessed by 16S qPCR.

We next determined whether these bacteria are prevalent in PSC/UC using collection of faecal samples comprising 18 patients with PSC/UC, 16 patients with classic UC and 10 HCs (Supplementary Table 1). The metagenome analyses depicted a drastic change of microbiota between HC and UC, reflecting colonic inflammation, but the difference between PSC/UC and classic UC was less evident (Supplementary Fig. 4a–c). Nevertheless, linear discriminant analysis effect size analysis identified the genera Klebsiella and Enterococcus as uniquely enriched bacterial species that corresponded to the 3-mix bacteria (Supplementary Fig. 4d). Furthermore, species-specific quantitative PCR (qPCR) analyses for K.pneumoniae, P.mirabilis and E.gallinarum revealed a clear difference of their prevalence among the groups. Notably, 17 out of 18 patients with PSC harboured K.pneumoniae, which was a significantly higher rate than that of HCs (0 out of 10) and patients with classic UC (5 out of 16), with P < 0.0001 and P < 0.0002 by Fisher’s exact test, respectively (Fig. 3c). Likewise, the detection of E.gallinarum was higher in patients PSC than in those in the other groups. The high prevalence of these bacterial species was validated in other independent cohorts of faecal samples from 27 Japanese patients with PSC19 and from 44 European patients with PSC20 (Supplementary Fig. 5). We additionally examined faecal specimens collected from patients with other autoimmune liver diseases, including autoimmune hepatitis (AIH) and primary biliary cholangitis (PBC). Of note, K.pneumoniae was detected in 3 out of 9 and 5 out of 10 of faecal specimens from patients with AIH and PBC, respectively (Fig. 3d), suggesting that the prevalence of K.pneumoniae might be related to the presence of hepatobiliary diseases. Nevertheless, our results indicated that the 3-mix bacteria isolated from patients with PSC promoted bacterial translocation in the gnotobiotic mouse model, underscoring the pathogenic role of gut microbiota in PSC.

The 3-mix bacteria cooperatively promote the hepatobiliary disease progression through the TH17 response

We next determined whether PSC-derived 3-mix bacterial strains are responsible for liver inflammation. Consistent with the result of PSCUC mice, 3-mix gnotobiotic mice showed a strong TH17 response in the colon and liver (Fig. 4a). To further determine which bacterial species is responsible for TH17 priming, we inoculated GF mice with three bacterial strains either individually or in combination. Interestingly, the mono-colonization of K.pneumoniae or dual-colonization of P.mirabilis and E.gallinarum was less efficient in TH17 induction than the 3-mix bacteria, suggesting that 3-mix bacteria cooperatively induced the liver inflammatory response (Fig. 4a). Accordingly, when stimulating naive CD4+ T cells with heat-killed 3-mix bacteria, we observed higher TH17 induction than when K.pneumoniae was used alone (Fig. 4b). These results suggest that, whether viable, the 3-mix bacteria could elicit TH17 responses in the liver.

Fig. 4: Inhibition of ROR-γt protects mice from DDC-induced hepatobiliary injury.
Fig. 4

a, Frequency of ROR-γt+IL-17+ cells in the liver (left) and colon (right) of mice inoculated with three bacterial strains either individually or in combination (n = 6 mice in the GF group and n = 4 mice in the other groups). Data show the mean ± s.e.m. Data are representative of two independent experiments. PM, P. mirabilis; EG, E. gallinarum. b, Naive CD4 T cells and dendritic cells were cultured with K.pneumoniae (KP) alone or with the 3-mix culture under the TH17-skewing condition. Left: representative dot plots of intracellular IFN-γ and IL-17 staining on CD11cCD4+-gated cells restimulated with PMA and ionomycin. Right: the frequency of IL-17A+IFN-γ cells in CD11cCD4+ cells in each condition. Data show the mean ± s.e.m. (n = 3 biological replicates). For a and b, ANOVA using Tukey’s multiple-comparison correction was applied. c, 3-mix gnotobiotic mice were intraperitoneally given either an anti-mouse IL-17A-neutralizing antibody (500 µg per mice, n = 4 mice) or mouse IgG1 isotype control (n = 5 mice) every other day for 2 weeks during DDC feeding from day 14 to day 28 post-transplantation. Serum TB (left) and ALP (right) levels are shown. d, Study design: 3-mix gnotobiotic (gnoto) mice fed with DDC were treated daily with either a RIA (10 mg per kg mice) or vehicle for 2 weeks. The yellow triangles represent the time point of analysis. e,f, Representative intracellular ROR-γt and IL-17 staining (upper) and ROR-γt and IFN-γ staining (lower) of TCR-β+CD4+-gated cells (e), and the frequency of IL-17+ cells (upper) and IFN-γ+ cells (lower) in TCR-β+CD4+ T cells (f) in the liver of mice. g, Serum TB (upper) and ALP (lower) levels of mice. h, Representative photomicrographs of Sirius Red staining (upper) and Masson trichrome staining (lower) of the liver sections of mice. Scale bars, 100 µm. i, Quantitative Sirius Red-positive area of the liver sections of mice. j, qRT–PCR analysis of Col1a1 (left) and Timp1 (right) relative to Gapdh in the whole liver of mice. The numbers in each graph indicate the ratio compared to the vehicle group. For ej, n = 5 mice per group. For c, f, g, i and j, data show the mean ± s.e.m. Two-sided Student’s t-test was applied. NS, not significant.

To address whether the liver TH17 response contributes to the disease progression, we developed a therapeutic approach that targets TH17 differentiation in the DDC-fed 3-mix gnotobiotic mouse model. As a single use of anti-IL-17A antibody failed in the amelioration of the hepatobiliary injury (Fig. 4c), we utilized RAR-related orphan receptor-γt (ROR-γt) inverse agonist (RIA) that has been previously shown to selectively inhibit TH17 differentiation21 (Fig. 4d). As expected, RIA treatment significantly reduced the number of TH17 cells without affecting the numbers of TH1 cells, confirming the specific inhibition of TH17 differentiation (Fig. 4e,f). Importantly, RIA treatment also improved hepatobiliary injury and fibrosis of DDC-fed 3-mix gnotobiotic mice, emphasizing the pathogenic role of TH17 activation in liver inflammation (Fig. 4g–j).

K.pneumoniae derived from the MLNs of PSCUC mice induces pore formation in human intestinal epithelial cells in vitro

To obtain direct evidence of whether the 3-mix bacteria penetrated the intestinal epithelial barrier, we next performed FISH analysis using intestinal mucosa of 3-mix gnotobiotic mice. Although no bacterium invaded the intestinal mucosa of SPF mice, bacterial DNA was detected underneath the intestinal epithelium in 3-mix gnotobiotic mice, indicating the presence of bacterial invasion. The mucosal bacterial DNA was also detected in single K.pneumoniae gnotobiotic mice, but not in P.mirabilis + E.gallinarum gnotobiotic mice, highlighting the essential role of K.pneumoniae in the bacterial invasion (Fig. 5a and Supplementary Fig. 6a).

Fig. 5: K.pneumoniae derived from the MLNs of PSCUC mice induces epithelial pore formation of colonic epithelial cells.
Fig. 5

a, Representative FISH staining of the ileum of mice from the indicated groups (n = 4 mice per group) to identify the bacterial 16S rRNA genes (red), co-stained with phalloidin (F-actin; green) and 4′,6-diamidino-2-phenylindole (DAPI; blue). Insets show higher magnification. Scale bars, 50 µm in the upper panels and 10 µm in the lower panels. b, Schematic drawing of the two-dimensional organoid microbiome interaction system used in this experiment. c,d, Representative scanning electron microscopy images (c) and the quantitative scoring of damaged intestinal epithelium with pores (>10 µm; epithelial pore) (d) of the intestinal epithelium after 8 h of culture of KP-P1 or KP-P5, enterohemorrhagic E.coli O157, a non-pore-forming K.pneumoniae strain JCM 1662T provided from RIKEN (KP-1662) and the PBS control. The middle and lower panels in c show each upper image at a higher magnification. Arrowheads indicate collapsed colonic monolayers. Microorganisms are indicated by the arrows. Scale bars, 20 µm in the top and middle panels and 2 µm in the lower panels. For c and d, n = 16 independent samples in the PBS group, n = 8 samples in the KP-P1, KP-P5 and KP-1662 groups and n = 24 samples in the O157 group. For d, data show the mean ± s.e.m. ****P < 0.0001 by ANOVA using Tukey’s multiple-comparison correction. e, Intestinal epithelial cells co-cultured with the indicated bacterial strain were stained with the anti-active caspase-3 antibody (green) and co-stained with phalloidin (F-actin; pink). Arrowheads represent epithelial pores, and the inset shows a higher magnification of an epithelial pore. Scale bars, 100 µm. Data are representative of four independent experiments. f, Comparative analysis of the whole genomes of the tested 13 K.pneumoniae strains revealed that 97 orthologous genes correlated with the pore-forming capacity, defined as present (orange) in more than 4 of 9 pore-forming K.pneumoniae strains (+) and defective (yellow) in all 4 non-pore-forming K.pneumoniae strains (−). Detailed information about the gene ID for each row is listed in Supplementary Table 4. ROS, reactive oxygen species; T6SS, type VI secretion system.

To clarify the mechanism of how microbiota interacts with the intestinal epithelial barrier, we established a monolayered culture system using human intestinal organoids. Contrary to the closed apical sides in conventional organoids, the monolayered organoids directly exposed their apical sides to culture medium, allowing interactions between the administered bacteria and the epithelial surface22,23 (Fig. 5b). We obtained two different K.pneumoniae strains, KP-P1 and KP-P5, which were derived from the MLN of PSCUC mice generated from two independent patients. Interestingly, upon administering KP-P1 and KP-P5 on the monolayered organoids, both bacteria induced the formation of epithelial pores in the monolayer within 8 h (Fig. 5c,d). A similar phenotype was observed using a strain of enterohemorrhagic Escherichia coli O157:H7 (Fig. 5c,d). Immunostaining illustrated that epithelial pores were filled with cleaved caspase-3+ apoptotic cells and encircled by viable epithelial cells (Fig. 5e). Furthermore, RNA sequencing (RNA-seq) analysis revealed marked upregulation of genes related to apoptotic and inflammatory pathways in KP-P1-stimulated epithelial cells compared to JCM 1662-stimulated epithelial cells (Supplementary Fig. 6b,c and Supplementary Table 2). Heat-killed K.pneumoniae failed to induce epithelial pore formation (Supplementary Fig. 6d). These results suggested that K.pneumoniae may induce epithelial pore formation through apoptotic and inflammatory signalling in a contact-dependent manner. Interestingly, the epithelial pore-inducing capacity of K.pneumoniae varied among the strains, of which seven strains induced pore formation, whereas four did not (Fig. 5c and Supplementary Fig. 6e).

To gain genetic insights into epithelial pore-forming capacity of K.pneumoniae, we obtained metagenome data from 13 strains of K.pneumoniae. There was no correlation between epithelial pore-forming capacity and multilocus sequencing typing based on virulence genes or capsular (K) typing (Supplementary Table 3). Interestingly, the whole-genome phylogenetic analysis revealed that K.pneumoniae strains with epithelial pore-forming capacity were widely distributed, whereas those without epithelial pore-forming capacity tended to be closely related (Supplementary Fig. 7). Furthermore, 97 orthologous genes associated with epithelial pore-forming capacity were identified by comparative analysis of the whole-genome sequencing of 13 K.pneumoniae strains, including genes involved in the type VI secretion system24,25 and reactive oxygen species degradation (Fig. 5f and Supplementary Table 4). These results suggested that the epithelial pore-forming capacity of K.pneumoniae is determined by strain-specific genotype.

Epithelial pore-forming capacity in a K.pneumoniae strain was responsible for bacterial translocation and subsequent TH17 priming in the liver

To determine whether the in vitro epithelial pore-forming capacity correlates to in vivo barrier disruption, we performed fluorescein isothiocyanate (FITC)-conjugated dextran (FITC-dextran) leakage analysis. In gnotobiote with epithelial pore-inducing K.pneumoniae, 4 kDa FITC-dextran, but not 70 kDa FITC-dextran, leaked to the systemic circulation, indicating the presence of barrier disruption that allows the penetration of small particles into the mucosa (Fig. 6a). Of note, the increased leakage was not detected in gnotobiote with non-epithelial pore-inducing K.pneumoniae (JCM 1662T), indicating that the epithelial pore-forming capacity contributes to in vivo barrier disruption. The association between K.pneumoniae and intestinal barrier function prompted us to investigate whether the epithelial-damaging effect of K.pneumoniae contributed to liver inflammation. Thus, we generated gnotobiotic mice with modified 3-mix strains (m3-mix gnotobiotic mice) in which K.pneumoniae was replaced with the non-pore-forming strain (Fig. 6b). In contrast to 3-mix gnotobiotic mice, m3-mix gnotobiotic mice were devoid of mucosal barrier invasion (Fig. 6c) and showed lower serum endotoxin levels (Fig. 6d). Furthermore, m3-mix gnotobiotic mice exhibited considerably lower levels of liver TH17 priming (Fig. 6e,f). These results suggested that the epithelial-damaging effect of K.pneumoniae is crucial for the induction of the liver TH17 response.

Fig. 6: Bacterial translocation and TH17 priming in the liver are dependent on K.pneumoniae strains.
Fig. 6

a, Serum FITC-dextran (4 kDa (left) and 70 kDa (right)) concentration calculated 8 h following mono-colonization with KP-P1 or KP-1662 to GF mice (n = 4 mice per group for 4 kDa and n = 3 mice per group for 70 kDa). b, Study design: 3-mix strains derived from the MLNs of PSCUC1 mice or m3-mix were inoculated into GF mice, followed by immunological phenotyping at day 21. The yellow triangles represent the time point of analysis. c, Representative FISH staining of the ileum of mice (n = 3 mice per group) to identify the bacterial 16S rRNA genes, co-stained with phalloidin (F-actin) and DAPI (nucleus). Scale bars, 20 µm. d, Serum LPS concentration of mice. e,f, Representative intracellular ROR-γt and IL-17 staining on CD3+CD4+-gated cells (e), and the frequency of ROR-γt+IL-17+ cells (f) in the liver of mice. For df, n = 5 mice in the GF group and n = 4 mice in the other groups. g, Mice inoculated with the indicated bacteria were administered 0.05% DDC for 2 weeks from day 14 to day 28. Serum TB levels are shown (n = 4 mice in the GF and m3-mix groups, and n = 5 mice in 3-mix group). h, Study design: mice in the indicated groups were administered 0.5% TCA for 2 weeks from day 14 to day 28. i, Serum TB (left) and ALP (right) levels. j,k, Representative intracellular IFN-γ and IL-17 staining of TCR-β+CD4+-gated cells (j), and the frequency of IL-17+ cells (left) and IFN-γ+ cells (right) in TCR-β+CD4+ T cells (k) in the liver of mice. l, Representative photomicrographs of haematoxylin and eosin staining of the liver sections of mice. Insets show higher magnification. The arrows and arrowheads indicate interlobular bile ducts and larger bile ducts, respectively. Scale bars, 100 µm. m, qRT–PCR analysis of Col1a1 (left) and Timp1 (right) relative to Gapdh in the whole liver. The numbers in each graph indicate the ratio compared to GF mice. For im, n = 5 mice in the GF group and n = 4 mice in the other groups. For a, d, f, g, i, k and m, data show the mean ± s.e.m. ANOVA using Tukey’s multiple-comparison correction was applied.

We next determined whether the presence of pore-forming capacity affects liver inflammation in the DDC model. Interestingly, m3-mix gnotobiotic mice showed significantly decreased liver dysfunction compared to that in 3-mix gnotobiotic mice, highlighting the pathogenic effect of pore-forming capacity (Fig. 6g). Because the microbiota may affect the cholestatic phenotype by modulating the DDC metabolism, we decided to employ taurocholic acid (TCA)-induced hepatobiliary injury, as another disease model for PSC (Fig. 6h). Interestingly, TCA feeding caused intrahepatic cholestasis-mediated liver enzyme level elevation in SPF mice, but the hepatobiliary dysfunction was minimal in GF mice, suggesting the involvement of gut microbiota in this model. Consistent with the DDC model, 3-mix gnotobiotic mice, but not m3-mix gnotobiotic mice, developed severe hepatobiliary injury along with increased TH17 response (Fig. 6i–m). Pathological assessment revealed periductal inflammation and fibrotic change in interlobular bile ducts in 3-mix gnotobiotic mice. Importantly, we noted that the infiltration of inflammatory cells and fibrotic response were involved with larger bile ducts, which is a characteristic of PSC (Fig. 6l). These results further reinforced the biological effect of pore-forming K.pneumoniae on hepatobiliary injuries.

Treatment with metronidazole or vancomycin ameliorated TH17 responses in the liver

To determine whether antibiotic treatment could modulate liver immune responses in our experimental model, we generated gnotobiotic mice using faecal samples from a patient with PSC (P5) that harboured K.pneumoniae and E.gallinarum, and treated the mice with metronidazole or vancomycin that possess antibacterial potential against K.pneumoniae and E.gallinarum, respectively26,27 (Supplementary Fig. 8a,b). Notably, the TH17 response was induced in P5-PSCUC gnotobiotic mice and the immune response was significantly decreased by metronidazole or vancomycin treatment (Supplementary Fig. 8c,d). The fact that vancomycin treatment decreased the TH17 response suggests that, in addition to K.pneumoniae, there may be vancomycin-sensitive microbiota that possess epithelium-disrupting capability. Nevertheless, these results not only underscored the involvement of K.pneumoniae and E.gallinarum in hepatobiliary inflammation but also provide preclinical evidence for a feasible pathobionts-targeting therapeutic strategy for PSC.


Recent microbiome studies revealed an association between dysbiosis and PSC; however, the mechanistic understanding of how gut microbiota contribute to the pathogenesis of PSC remained elusive. In this study, by combining gnotobiote and classical bacterial culturing, we successfully isolated K.pneumoniae, P.mirabilis and E.gallinarum from the MLN of gnotobiotic mice. These 3-mix bacteria were found to be prevalent in patients with PSC and responsible for bacterial translocation and subsequent hepatobiliary inflammation in the gnotobiotic mouse models.

It has been reported that bacterial translocation is frequently associated with hepatobiliary diseases28,29, and intestinal barrier dysfunction has been considered as its pathological basis30. In this study, we first demonstrated that PSC-derived K.pneumoniae possess epithelial-damaging abilities that contribute to bacterial translocation. Interestingly, members of the Enterobacteriaceae family, including K.pneumoniae, were often observed in the microbiota of individuals with hepatobiliary diseases, such as liver cirrhosis28,31 and PBC32. Furthermore, the abundance of the bacteria in PBC was associated with bacterial invasion of epithelial cells32. These results collectively highlighted the previously unappreciated link between bacterial translocation and the Enterobacteriaceae family in hepatobiliary diseases. Our monolayered-organoid culture system further revealed that PSC/UC-derived K.pneumoniae can induce epithelial pore formation in the epithelial layer. In accordance with this, the epithelial pore-forming capacity was associated with intestinal barrier dysfunction in vivo, providing insights into the interplay between gut microbiota and bacterial translocation in PSC. Interestingly, the pore-forming capacity of K.pneumoniae was strain specific and enabled a comparative whole-genome analysis. The genetic analysis did not identify conserved virulent toxins, but listed several candidate genes responsible for pore-forming capacity, including the type VI secretion system. The type VI secretion system is a contact-dependent protein secretion system known to transfer various effector proteins to target host cells33,34, and thus may contribute to the K.pneumoniae-induced epithelial pore formation demonstrated in this study. Further studies will be required to identify and confirm the genetic factors responsible for the epithelial pore-forming capacity in K.pneumoniae.

Although we successfully isolated bacteria from the MLN of PSCUC mice, no bacteria could be grown from the liver. This was in contrast to the frequent observation of translocated bacteria in the liver of patients with PSC7. There is a possibility that the translocated bacteria were immediately eliminated by immune cells upon reaching the liver; however, this scenario seems unlikely because we failed to detect any bacterial signal using FISH and PCR. A more plausible explanation is that a ‘gut–vascular barrier’35 blocked the invaded bacteria from translocating into the liver. Indeed, 4 kDa but not 70 kDa FITC-dextran was able to reach the systemic circulation, supporting the intact gut–vascular barrier hypothesis in the 3-mix gnotobiotic mice. It is tempting to speculate that disruption of this second firewall by comorbid colitis allows escalation of bacterial translocation and in turn activates the liver TH17 response in PSC. Interestingly, the gnotobiotic mice with PSC/UC-derived microbiota or 3-mix bacteria establish hepatic TH17 responses even in the absence of bacterial translocation to the liver. Instead, we observed increased endotoxin levels in the serum of the mice, suggesting that a disrupted epithelial barrier promoted an influx of microbial components and/or by-products to the liver. As heat-killed 3-mix bacteria were able to promote TH17 differentiation, the delivery of the 3-mix bacteria-derived products to the liver may be sufficient to induce hepatic TH17 responses. Taken together, our findings suggest that the gut microbiota is responsible for the disruption of the first intestinal epithelial barrier, and a second trigger, such as colitis or hepatobiliary injury, further consolidates the TH17-mediated disease progression in PSC.

The infiltration of TH17 cells in the PSC liver was previously reported7, yet the activation mechanism of TH17 differentiation has not been determined. In this study, using a gnotobiotic system, we demonstrated that PSC/UC microbiota was able to induce TH17 differentiation in the liver. The comparable level of TH17 induction was achieved by a colonization of 3-mix bacteria but not by the mono-colonization of K.pneumoniae, suggesting that pathobionts cooperatively induced liver inflammation. Interestingly, we demonstrated that elimination of these pathobionts was effective in reducing the number of TH17 cells in the liver. Consistent with our results, a recent report demonstrated that antibiotics-based eradication of E.gallinarum alleviated the progression of extra-intestinal autoimmune diseases36. Furthermore, one pilot clinical trial demonstrated the therapeutic effect of metronidazole and vancomycin on PSC37. These results collectively suggested that the gut microbiota is implicated in hepatobiliary inflammation, and larger clinical trials will be warranted to establish the antibiotics-based therapeutic approach to PSC.

We also found that the gut microbiota-inducible TH17 response was crucial for hepatobiliary disease progression and was therapeutically targetable in PSC disease models. We noted that antibody-based inhibition of IL-17A was insufficient to block the disease progression of the DDC-fed PSCUC model, reminiscent of a failure in a clinical trial for IBD38,39. This failure may be attributed to the residual production of other cytokines from TH17 cells, such as IL-17F and granulocyte–macrophage colony-stimulating factor40. To circumvent this complexity, we employed therapeutic targeting of ROR-γt, an essential master transcription factor for TH17 differentiation41,42,43. Interestingly, the RIA efficiently blocked the pathobiont-induced hepatic TH17 response and ameliorated the DDC-induced hepatobiliary injury in PSCUC mice. These results provided evidence that pharmacological inhibition of TH17 differentiation could be a therapeutic approach for PSC.

In summary, the current study identified disease-modulating pathobionts from patients with PSC/UC. One of the key functions of these pathobionts is to disrupt the intestinal epithelial barrier through the strain-specific epithelial-damaging effect of K.pneumoniae, which enables collateral translocation of other pathobionts and elicits subsequent TH17 priming in the MLN and liver. Our results collectively provide insights into the implication of the gut microbiota in the disease pathogenesis of PSC.



HCs (n = 10) and patients with complications of PSC/UC (n = 18), UC (n = 16), AIH (n = 9) and PBC (n = 10) were included in this study. The diagnosis of PSC was made according to clinical guidelines and typical findings on cholangiography (endoscopic retrograde cholangiography and/or magnetic resonance cholangiopancreatography) or liver biopsy. The diagnosis of UC was based on a combination of endoscopy, histopathology and radiological and serological investigations. The diagnoses of AIH and PBC were made according to clinical guidelines and typical histological findings. The ethics committee at Keio University School of Medicine (Tokyo, Japan) approved the protocol (no. 20140211). The study was conducted according to the principles of the Helsinki Declaration II, and written informed consent was obtained from all study participants. The study is registered at the University Hospital Medical Information Network (UMIN) clinical trial registration system (UMIN 000018068).


Male GF mice (C57BL/6 background strain, 6–8-weeks old) were purchased from Sankyo Lab Service Corporation and were kept in the GF Facility of Keio University School of Medicine. Male C57BL/6 mice (6–8-weeks old) were purchased from Japan CLEA (Tokyo, Japan) and maintained under SPF conditions in Central Laboratories for Experimental Animals (Kawasaki, Japan) and the Animal Care Facility of Keio University School of Medicine. All experiments were approved by regional animal study committees and were performed during March 2015 to September 2018 according to the institutional guidelines and Home Office regulations.

Gnotobiotic study design

Faecal samples from PSC/UC, UC and HC individuals were collected after having obtained written informed consent. Faecal samples were suspended in equal volumes (w/v) of PBS containing 40% glycerol, snap-frozen and stored at −80 °C until use. The frozen stocks were thawed, suspended in 6-fold volumes of PBS and passed through a 70-μm cell strainer. GF mice were orally inoculated with 200 µl of the suspensions using a stainless-steel feeding needle. Experiments were conducted after a colonization period of 3–4 weeks or after a longer period of 12 weeks. We initially tested faecal samples obtained from each disease group (4–5 per group) to confirm that the composition of microbiota used for the gnotobiotic studies is representative of each group by the unweighted UniFrac analysis. In some experiments, GF mice transplanted with human stool samples were given either metronidazole (1 g l−1) or vancomycin (500 mg l−1) in drinking water 1 day after the inoculation for 21 days. To confirm the existence of any bacteria outside the intestine of SPF mice or humanized gnotobiotic mice, the liver, MLN and spleen were carefully harvested and anaerobically cultured on blood liver, BHI and reinforced clostridial medium agar plates. The isolates were identified with 16S rRNA gene sequencing. To examine the effects of each of the bacterial species, 1 × 108 colony-forming units of a single species were suspended in 200 µl of medium or a combination of mixed bacteria was suspended and then orally or intravenously administered to GF mice.

DDC-induced or TCA-induced experimental hepatobiliary inflammation and liver fibrosis

To induce hepatobiliary inflammation and liver fibrosis, GF mice or humanized gnotobiotic mice were freely fed a 0.05% DDC-enriched (Sigma-Aldrich) or 0.5% TCA-enriched (Tokyo Chemical industry) diet for 14 days, followed by serological, histological and immunological assessment. To inhibit TH17 responses, a RIA, GSK805 (10 mg per kg mice; no. HY-12776, Medchemexpress) dissolved in 10% DMSO/90% corn oil, was orally administered to DDC-fed mice daily for 14 days21,44. To neutralize IL-17A, an anti-mouse IL-17A-neutralizing antibody (500 µg per mice; clone 17F3, no. BE0173, BioXcell) or mouse IgG1 isotype control (500 µg per mice; no. BE0083, BioXcell) was intraperitoneally administered to DDC-fed mice every other day for 14 days.

Bacterial strains and culture

K.pneumoniae JCM 1662T, JCM 1663, JCM1664, JCM 20034, JCM 20348, JCM 20507 and JCM 20694 were obtained from the Japan Collection of Microorganisms (RIKEN BioResource Center, Wako, Japan). K.pneumoniae ATCC 700603, ATCC BAA1705, ATCC BAA2552 and ATCC 700721 were kindly provided by K.H., Keio University. E.coli O157:H7 was grown overnight in LB broth. K.pneumoniae, P.mirabilis and E.gallinarum were anaerobically cultured in BHI broth (BD). To prepare heat-killed bacteria, the bacteria were collected by centrifugation and washed three times with PBS. After washing, the bacteria were heat-killed at 95 °C for 30 min.

Human intestinal organoid culture

Normal colonic organoids were established from human patients and maintained as described in our previous work45. Long-term healthy intestinal organoids were embedded into Matrigel (Corning) and cultured in medium composed of Advanced DMEM/F12 supplemented with penicillin–streptomycin, 10 mM HEPES, 2 mM GlutaMAX, 1×B27 (Life Technologies), 1 mM N-acetylcysteine (Wako Pure Chemical Industries), 10 nM GastrinI (Sigma), 50 ng ml−1 human recombinant epidermal growth factor, 0.5 µM A83-01, 3 μM SB202190, 50% afamin–Wnt3a (Afm-W) complex condition medium (v/v)46, R-spondin 1-condition medium (10% v/v) and noggin-condition medium (10% v/v).

Preparation of two-dimensional organoid microbiome interaction system

For the monolayer organoid culture, three-dimensional cultured human colonic organoids were maintained in Afm-W, R-spondin 1, noggin, epidermal growth factor, A83-01, SB202190 with 10 μM Y-27632 for at least 1 day before dissociation and then seeded onto 0.4-μm pore ThinCert 24-well plates (Greiner) coated with 10% Cellmatrix type I-C (Nitta Gelatin). At 2–3 days after seeding, the medium of the monolayer colonic organoids was replaced with differentiation condition medium without Afm-W, SB202190 and Y-27632. Prior to infection with bacteria, the colonic epithelium was washed in advanced DMEM/F12 and antibiotic-free differentiation condition medium was added. For the two-dimensional organoid microbiome interaction system, intestinal epithelia were co-cultured with 1 × 105 colony-forming units of each bacterium for 8 h.

Sample collection and DNA extraction

Fresh faecal samples were collected using stool collection tubes and an anaerobiosis generator was added to the samples to favour the preservation of anaerobic bacteria at the outpatient clinic of Keio University Hospital. The samples were processed immediately and frozen at −80 °C for bacterial preservation. Bacterial DNA was isolated as described previously47. In brief, bacterial DNA was isolated by the enzymatic lysis method using lysozyme (Sigma-Aldrich) and achromopeptidase (Wako). DNA samples were then purified by treating with ribonuclease A (Wako), followed by precipitation with 20% polyethylene glycol solution (PEG6000 in 2.5 M sodium chloride). DNA was then pelleted by centrifugation, rinsed with 75% ethanol and dissolved in Tris-EDTA buffer.

Faecal microbiota quantification by qPCR

Quantification of K.pneumoniae was performed using the K.pneumonia-EASY genesig kit (Primerdesign), PrecisionPLUS qPCR Master Mix (Primerdesign) and the Thermal Cycler Dice Real Time System II (Takara), according to the manufacturer’s instructions. Quantification of E.gallinarum and P.mirabilis was performed using SYBR Premix EX Taq II (Takara) and the Thermal Cycler Dice Real Time System II. The primer sets used in this study were as follows: E.gallinarum forward 5′-TTACTTGCTGATTTTGATTCG-3′ and reverse 5′-TGAATTCTTCTTTGAAATCAG-3′ (ref. 48); P.mirabilis forward 5′-GTTATTCGTGATGGTATGGG-3′ and reverse 5′-ATAAAGGTGGTTACGCCAGA-3′ (ref. 49).

16S rRNA metagenomic analysis

The hypervariable V3–V4 region of the 16S gene was amplified using Ex Taq Hot Start (Takara Bio) and subsequently purified using AMPure XP (Beckman Coulter). Mixed samples were prepared by pooling approximately equal amounts of each amplified DNA and sequenced using the Miseq Reagent Kit V3 (600 Cycle) and Miseq sequencer (Illumina), according to the manufacturer’s instructions. Sequences were analysed using the QIIME software package version 1.9.1 (refs. 50,51). Paired-end sequences were joined using a fastq-join tool in the ea-utils software package (https://doi.org/10.2174/1875036201307010001). High-quality sequences per sample (15,000) were randomly chosen from the quality filter-passed sequences. After trimming off both primer sequences using cutadapt (https://doi.org/10.14806/ej.17.1.200) followed by chimeras detection by the USEARCH52 de novo method, the sequences were assigned to operational taxonomic units using the UCLUST algorithm53 with a sequence identity threshold of 96%. Taxonomic assignments of each operational taxonomic unit were made by similarity searching against the publicly available 16S (RDP version 10.27 and CORE update 2 September 2012) and the NCBI genome database using the GLSEARCH program. The data were rarefied to 10,000 sequences per sample, as determined by the rarefaction curves. Relative abundances of the community members were determined using the rarefied data. UniFrac analysis was performed as described previously54. To determine bacterial taxonomy that explained differences between conditions, the linear discriminant analysis effect size method was used55.

Localization of bacteria by FISH

Colon and ileal tissues containing faecal material were fixed with Carnoy’s solution for 3 h. Paraffin-embedded sections were de-waxed and hydrated. The hybridization step was performed at 50 °C overnight with an EUB338 probe (5′-GCTGCCTCCCGTAGGAGT-3′, with an Alexa 555 label) diluted to a final concentration of 10 µg ml−1 in hybridization buffer (20 mM Tris-HCl, pH 7.4, 0.9 M NaCl, 0.1% SDS and 20% formamide). After washing for 10 min in wash buffer (20 mM Tris-HCl, pH 7.4, and 0.9 M NaCl) and three times for 10 min in PBS, samples were stained with phalloidin-iFluor 488 (Abcam). After washing three times for 10 min in PBS, slides were mounted using Prolong anti-fade mounting media with DAPI (Life Technologies). Microscopic observations were performed using a BIO-REVO BZ-9000 fluorescence microscope (Keyence).

Isolation of lymphocytes and flow cytometry

Liver and spleen mononuclear cells were separated as described previously56. Intestinal lamina propria mononuclear cells and MLN mononuclear cells were prepared as described previously57. After blocking with anti-Fc receptor (CD16/32, BD Pharmingen) for 20 min, the cells were incubated with the specific fluorescence-labelled monoclonal antibody at 4 °C for 30 min. The following monoclonal antibodies were used: anti-mouse anti-T cell receptor-β (TCR-β; no. 109228, Biolegend, peridinin chlorphyll protein–Cy5.5 conjugate, clone H57-597), anti-CD3e (no. 552774, BD Bioscience, phycoerythrin (PE)–Cy7 conjugate, clone 145-2C11), anti-CD4 (no. 563106, BD Bioscience, BV510 conjugate, clone RM4-5), anti-CD11b (no. 101245, Biolegend, BV510 conjugate, clone M1/70), anti-CD11c (no.558079, BD Bioscience, PE–Cy7 conjugate, clone HL3), anti-F4/80 (no. 123131, Biolegend, BV421 conjugate, clone BM8) and fixable viability dye eFluor780 (no. 65-0864-14, eBioscience). Events acquired with FACS Canto II (Becton Dickinson) were analysed with FlowJo software (Tree Star). For intracellular cytokine staining, cells were stimulated for 4 h with lipopolysaccharide (LPS; from E.coli B5; Sigma) or PMA (50 ng ml−1; Sigma-Aldrich) and ionomycin (500 ng ml−1; Sigma-Aldrich) in the presence of brefeldin A (10 µg ml−1; BD Bioscience) or Golgistop (10 µg ml−1; BD Bioscience), followed by surface staining, permeabilization and intracellular staining with anti-mouse anti-interferon-γ (IFN-γ; no. 554412, BD Bioscience, PE conjugate, clone XMG1.2; no. 11-7311-82, eBioscience, FITC conjugate, clone XMG1.2), anti-tumour necrosis factor-α (no. 554419, BD Bioscience, PE conjugate, clone MP6-XT22), anti-IL-1β (no. 17-7114-80, eBioscience, allophycocyanin (APC) conjugate, clone NJTEN3), anti-IL-17A (no. 560221, eBioscience, Alexa Fluor 488 conjugate, clone TC11-18H10), anti-IL-22 (no. 17-7222-82, eBioscience, APC conjugate, clone IL22JOP) and anti-ROR-γt (no. 562894, BD Bioscience, BV421 conjugate, clone Q31-378).

Serological and histological analysis

Serum levels of alanine aminotransferase, alkaline phosphatase and total bilirubin were measured using a LDH-UV kinetic method (SRL). Livers were fixed in 10% formalin, embedded in paraffin. Sections were stained with haematoxylin and eosin, Masson trichrome or Sirius Red for collagen staining and quantitative analysis of the fibrotic area was performed using the Image J software. Colon and ileal tissues were fixed in 10% formalin or Carnoy’s solution and embedded in paraffin.

qPCR analysis

Total RNA was extracted from liver homogenates using the RNeasy Mini Kit (Qiagen). Complementary DNA was synthesized from 1 µg total RNA by reverse transcription. To determine the gene expression, PCR was performed using AmpliTaq Gold Fast PCR MasterMix (Applied Biosystems) and the following predesigned primers: Gapdh (Mm03302249_g1), Saa1 (Mm00656927_g1), Saa2 (Mm04208126_mH), Saa3 (Mm00441203_m1), Fut2 (Mm01205565_m1), Reg3b (Mm00440616_g1), Reg3g (Mm00441127_m1), Il1b (Mm01336189_g1), Col1a1 (Mm008016666_g1) and Timp1 (Mm01341361_m1). To quantify the products, real-time PCR was performed using the TaqMan Universal Master Mix and StepOne Plus systems (Applied Biosystems). The level of target gene expression was normalized to Gapdh expression in each sample.

TH17 differentiation from mice naive CD4 T cells

Bacterial culture supernatants were sterilized through a 0.22-μm pore size membrane filter. Naive CD4 T cells were prepared from the spleen and superficial lymph nodes using the CD4+ T Cell Isolation Kit for mouse (Miltenyi Biotec). Naive CD4 T cells were then isolated using CD62L MicroBeads, mouse (Miltenyi Biotec). For preparation of dendritic cells, following lysis of red blood cells, splenocytes were blocked with anti-CD16/32 antibody (2.4G2) followed by incubation with CD11c MicroBeads, mouse (Miltenyi Biotec). Splenic dendritic cells were then isolated using the autoMACS system. Naive CD4 T cells (5 × 104 cells) and dendritic cells (2.5 × 103 cells) were cultured in a round-bottomed 96-well plate with bacterial culture supernatants (3% v/v of culture medium), in the presence of soluble anti-CD3e antibody under the TH17-skewing condition (anti-IFN-γ: 5 mg ml−1, anti-IL-4: 5 mg ml−1, anti-IL-2: 2 mg ml−1, IL-6: 20 ng ml−1 and transforming-growth factor-β: 0.5 ng ml−1) in RPMI 1640 supplemented with 10% FBS, 1% penicillin–streptomycin mixed solution, 1× non-essential amino acids solution and 0.05 mM 2-ME. Four days later, cells were restimulated with PMA or ionomycin in the presence of brefeldin A for an additional 5 h. Following staining of cell surface CD4 and CD11c, cells were fixed and permeabilized and stained for intracellular IL-17A. Data were acquired by FACS Canto II and analysed with FlowJo software (Tree Star).

RNA-seq analysis

Total RNA was isolated from organoids using TRIzol reagent (Life Technologies). For RNA-seq, RNA library preparation was performed using a TruSeq RNA Sample Prep Kit v2 (Illumina) according to the manufacturer’s instructions. Libraries were sequenced by Illumina HiSeq 4000 on the 100-bp paired-end mode. The sequenced reads were mapped to the human reference genome (NCBI 37.2) and normalized to fragments per kilobase per million reads (FPKM) values using the Tophat2 and Cufflinks software pipeline (https://doi.org/10.1186/gb-2013-14-4-r36 and https://doi.org/10.1038/nprot.2012.016). Gene Ontology and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was done with the DAVID tool (https://doi.org/10.1038/nprot.2008.211) on genes that were significantly differentially expressed.

Scanning electron microscopy analysis

Co-cultured intestinal epithelium with bacteria were fixed with 2.5% glutaraldehyde (TCI Chemicals) in PBS overnight at 4 °C. These samples were post-fixed for 1 h with 1% osmium tetroxide (TCI Chemicals) dissolved in PBS, dehydrated in a graded series of ethanol, coated with gold sputtering and observed under a VHX-D510 scanning electron microscope in the high vacuum mode (Keyence). In every experiment, the morphology of four independent positions in an intestinal epithelium was examined by scanning electron microscopy and the number of epithelial pores (>10 µm) were counted in each experiment. These experiments were repeated 3–6 times for statistical analysis.


Monolayered colonic epithelium was co-cultured with bacteria for 8 h in a 24-well transwell system. After incubation with bacteria or PBS control, cells were triple stained with cleaved caspase-3 antibody (1:400, Cell Signaling), the cell membrane dye (filamentous actin) phalloidin (1:300, Thermo Fisher) and the DNA dye Hoechst 33324 (1:1,000, Thermo Fisher), and apoptotic cells were observed using a confocal microscope (SP5, Leica).

Intestinal permeability in vivo

Mice were orally administered with 4 kDa FITC-dextran or 70 kDa FITC-dextran (Sigma-Aldrich) via a nasogastric tube after being fasted for 6 h. Blood samples were collected from the tail vein at 1 and 4 h post-administration. The concentration of FITC in the collected serum was measured using a spectrophotometer, according to the manufacturer’s instructions. Serum from non-administered mice was used to determine the background.

Measurement of LPS concentration

ToxinSensor Chromogenic Limulus Amebocyte Lysate Endotoxin Assay Kit (GenScript) was used for determination of serum LPS concentrations, following the manufacturer’s instructions. Briefly, samples were diluted 5-fold with endotoxin-free water. Limulus amebocyte lysate reagents were added to serum and incubated at 37 °C for 40 min. Subsequently, the absorbance at 540 nm was measured using the FilterMax F3 Multi-Mode Microplate Reader (Molecular Devices).

Bacterial genome sequencing

The genome sequences of Klebsiella strains were determined by the whole-genome shotgun strategy using PacBio RSII and Illumina MiSeq sequencers. The genomic DNA was sheared to obtain DNA fragments. Template DNA was prepared according to each supplier’s protocol. Obtained RSII reads were subjected to de novo assembly using HGAP3. MiSeq reads (2 × 300 nt) were mapped onto the RSII-assembled contigs to correct low-quality regions. Phylogenetic trees were constructed based on the Mash distance58 using the neighbour-joining method.

Multilocus sequencing typing, wzi and wzc sequence typing

The genomic sequences of each strain were aligned against the multilocus sequencing typing database (http://bigsdb.pasteur.fr/klebsiella/klebsiella.html). Sequence-based capsular (K) typing was carried out based on the sequencing of wzi or wzc genes59,60.

Statistical analysis

Statistical analyses were performed using GraphPad Prism software version 6.0 (GraphPad software Inc.). Differences between two groups were evaluated using a two-sided Student’s t-test or Wilcoxon rank-sum test. Comparison of more than two groups was performed by one-way analysis of variance (ANOVA), followed by Tukey’s multiple-comparison test. Correlations were tested for significance by the Spearman rank correlation test. Proportions between the two groups were compared with Fisher’s exact test. Differences at P < 0.05 were considered significant. Sample sizes were determined based on previous studies from our group and publication in this field. All animals were age and sex matched and randomly allocated to each group, and the investigators were not blinded during group allocation.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability

The data supporting the findings of this study are available from the corresponding authors upon request. 16S rRNA sequencing and whole-genome sequencing data have been deposited in the DNA Data Bank of Japan (DDBJ) database with the accession numbers DRA007475 and DRA007476 for 16S rRNA sequencing and PRJDB7545 for whole-genome sequencing. RNA sequencing data have been deposited in the European Genome-phenome Archive (EGA) database with the accession number EGAS00001003332.

Additional information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.


  1. 1.

    Lazaridis, K. N. & LaRusso, N. F. Primary sclerosing cholangitis. N. Engl. J. Med. 375, 1161–1170 (2016).

  2. 2.

    Horsley-Silva, J. L., Carey, E. J. & Lindor, K. D. Advances in primary sclerosing cholangitis. Lancet Gastroenterol. Hepatol. 1, 68–77 (2016).

  3. 3.

    Hirschfield, G. M., Karlsen, T. H., Lindor, K. D. & Adams, D. H. Primary sclerosing cholangitis. Lancet 382, 1587–1599 (2013).

  4. 4.

    Dyson, J. K., Beuers, U., Jones, D. E. J., Lohse, A. W. & Hudson, M. Primary sclerosing cholangitis. Lancet 391, 2547–2559 (2018).

  5. 5.

    O’Toole, A. et al. Primary sclerosing cholangitis and disease distribution in inflammatory bowel disease. Clin. Gastroenterol. Hepatol. 10, 439–441 (2012).

  6. 6.

    Sasatomi, K., Noguchi, K., Sakisaka, S., Sata, M. & Tanikawa, K. Abnormal accumulation of endotoxin in biliary epithelial cells in primary biliary cirrhosis and primary sclerosing cholangitis. J. Hepatol. 29, 409–416 (1998).

  7. 7.

    Katt, J. et al. Increased T helper type 17 response to pathogen stimulation in patients with primary sclerosing cholangitis. Hepatology 58, 1084–1093 (2013).

  8. 8.

    Loftus, E. V. Jr, Sandborn, W. J., Lindor, K. D. & Larusso, N. F. Interactions between chronic liver disease and inflammatory bowel disease. Inflamm. Bowel Dis. 3, 288–302 (1997).

  9. 9.

    Loftus, E. V. et al. PSC-IBD: a unique form of inflammatory bowel disease associated with primary sclerosing cholangitis. Gut 54, 91–96 (2005).

  10. 10.

    Claessen, M. M. et al. More right-sided IBD-associated colorectal cancer in patients with primary sclerosing cholangitis. Inflamm. Bowel Dis. 15, 1331–1336 (2009).

  11. 11.

    Karlsen, T. H. & Boberg, K. M. Update on primary sclerosing cholangitis. J. Hepatol. 59, 571–582 (2013).

  12. 12.

    Sabino, J. et al. Primary sclerosing cholangitis is characterised by intestinal dysbiosis independent from IBD. Gut 65, 1681–1689 (2016).

  13. 13.

    Kummen, M. et al. The gut microbial profile in patients with primary sclerosing cholangitis is distinct from patients with ulcerative colitis without biliary disease and healthy controls. Gut 66, 611–619 (2017).

  14. 14.

    Iwasawa, K. et al. Characterisation of the faecal microbiota in Japanese patients with paediatric-onset primary sclerosing cholangitis. Gut 66, 1344–1346 (2017).

  15. 15.

    Atarashi, K. et al. Th17 cell induction by adhesion of microbes to intestinal epithelial cells. Cell 163, 367–380 (2015).

  16. 16.

    Pollheimer, M. J., Trauner, M. & Fickert, P. Will we ever model PSC?—“It’s hard to be a PSC model!”. Clin. Res. Hepatol. Gastroenterol. 35, 792–804 (2011).

  17. 17.

    Tephly, T. R., Gibbs, A. H. & De Matteis, F. Studies on the mechanism of experimental porphyria produced by 3,5-diethoxycarbonyl-1,4-dihydrocollidine. Role of a porphyrin-like inhibitor of protohaem ferro-lyase. Biochem. J. 180, 241–244 (1979).

  18. 18.

    Fickert, P. et al. A new xenobiotic-induced mouse model of sclerosing cholangitis and biliary fibrosis. Am. J. Pathol. 171, 525–536 (2007).

  19. 19.

    Atarashi, K. et al. Ectopic colonization of oral bacteria in the intestine drives TH1 cell induction and inflammation. Science 358, 359–365 (2017).

  20. 20.

    Bajer, L. et al. Distinct gut microbiota profiles in patients with primary sclerosing cholangitis and ulcerative colitis. World J. Gastroenterol. 23, 4548–4558 (2017).

  21. 21.

    Withers, D. R. et al. Transient inhibition of ROR-γt therapeutically limits intestinal inflammation by reducing TH17 cells and preserving group 3 innate lymphoid cells. Nat. Med. 22, 319–323 (2016).

  22. 22.

    In, J. et al. Enterohemorrhagic Escherichia coli reduce mucus and intermicrovillar bridges in human stemcell-derived colonoids. Cell. Mol. Gastroenterol. Hepatol. 2, 48–62 (2016).

  23. 23.

    Sato, T. et al. Long-term expansion of epithelial organoids from human colon, adenoma, adenocarcinoma, and Barrett’s epithelium. Gastroenterology 141, 1762–1772 (2011).

  24. 24.

    Shneider, M. M. et al. PAAR-repeat proteins sharpen and diversify the type VI secretion system spike. Nature 500, 350–353 (2013).

  25. 25.

    Russell, A. B., Peterson, S. B. & Mougous, J. D. Type VI secretion system effectors: poisons with a purpose. Nat. Rev. Microbiol. 12, 137–148 (2014).

  26. 26.

    Siu, L. K., Yeh, K. M., Lin, J. C., Fung, C. P. & Chang, F. Y. Klebsiella pneumoniae liver abscess: a new invasive syndrome. Lancet Infect. Dis. 12, 881–887 (2012).

  27. 27.

    Murray, B. E. The life and times of the Enterococcus. Clin. Microbiol. Rev. 3, 46–65 (1990).

  28. 28.

    Wiest, R., Lawson, M. & Geuking, M. Pathological bacterial translocation in liver cirrhosis. J. Hepatol. 60, 197–209 (2014).

  29. 29.

    Llorente, C. & Schnabl, B. The gut microbiota and liver disease. Cell. Mol. Gastroenterol. Hepatol. 1, 275–284 (2015).

  30. 30.

    Miele, L. et al. Increased intestinal permeability and tight junction alterations in nonalcoholic fatty liver disease. Hepatology 49, 1877–1887 (2009).

  31. 31.

    Steffen, E. K., Berg, R. D. & Deitch, E. A. Comparison of translocation rates of various indigenous bacteria from the gastrointestinal tract to the mesenteric lymph node. J. Infect. Dis. 157, 1032–1038 (1988).

  32. 32.

    Tang, R. et al. Gut microbial profile is altered in primary biliary cholangitis and partially restored after UDCA therapy. Gut 67, 534–541 (2018).

  33. 33.

    Jiang, F., Waterfield, N. R., Yang, J., Yang, G. & Jin, Q. A Pseudomonas aeruginosa type VI secretion phospholipase D effector targets both prokaryotic and eukaryotic cells. Cell Host Microbe 15, 600–610 (2014).

  34. 34.

    Alcoforado Diniz, J., Liu, Y. C. & Coulthurst, S. J. Molecular weaponry: diverse effectors delivered by the type VI secretion system. Cell. Microbiol. 17, 1742–1751 (2015).

  35. 35.

    Spadoni, I. et al. A gut–vascular barrier controls the systemic dissemination of bacteria. Science 350, 830–834 (2015).

  36. 36.

    Manfredo Vieira, S. et al. Translocation of a gut pathobiont drives autoimmunity in mice and humans. Science 359, 1156–1161 (2018).

  37. 37.

    Tabibian, J. H. et al. Randomised clinical trial: vancomycin or metronidazole in patients with primary sclerosing cholangitis—a pilot study. Aliment. Pharmacol. Ther. 37, 604–612 (2013).

  38. 38.

    Hueber, W. et al. Secukinumab, a human anti-IL-17A monoclonal antibody, for moderate to severe Crohn’s disease: unexpected results of a randomised, double-blind placebo-controlled trial. Gut 61, 1693–1700 (2012).

  39. 39.

    Colombel, J. F., Sendid, B., Jouault, T. & Poulain, D. Secukinumab failure in Crohn’s disease: the yeast connection? Gut 62, 800–801 (2013).

  40. 40.

    McGeachy, M. J. GM-CSF: the secret weapon in the TH17 arsenal. Nat. Immunol. 12, 521–522 (2011).

  41. 41.

    Korn, T., Bettelli, E., Oukka, M. & Kuchroo, V. K. IL-17 and Th17 cells. Annu. Rev. Immunol. 27, 485–517 (2009).

  42. 42.

    Ivanov, I. I. et al. The orphan nuclear receptor RORγt directs the differentiation program of proinflammatory IL-17+ T helper cells. Cell 126, 1121–1133 (2006).

  43. 43.

    Gaffen, S. L., Jain, R., Garg, A. V. & Cua, D. J. The IL-23–IL-17 immune axis: from mechanisms to therapeutic testing. Nat. Rev. Immunol. 14, 585–600 (2014).

  44. 44.

    Xiao, S. et al. Small-molecule RORγt antagonists inhibit T helper 17 cell transcriptional network by divergent mechanisms. Immunity 40, 477–489 (2014).

  45. 45.

    Fujii, M. et al. A colorectal tumor organoid library demonstrates progressive loss of niche factor requirements during tumorigenesis. Cell Stem Cell 18, 827–838 (2016).

  46. 46.

    Mihara, E. et al. Active and water-soluble form of lipidated Wnt protein is maintained by a serum glycoprotein afamin/alpha-albumin. eLife 5, e11621 (2016).

  47. 47.

    Nishijima, S. et al. The gut microbiome of healthy Japanese and its microbial and functional uniqueness. DNA Res. 23, 125–133 (2016).

  48. 48.

    Jackson, C. R., Fedorka-Cray, P. J. & Barrett, J. B. Use of a genus- and species-specific multiplex PCR for identification of enterococci. J. Clin. Microbiol. 42, 3558–3565 (2004).

  49. 49.

    Stankowska, D., Kwinkowski, M. & Kaca, W. Quantification of Proteus mirabilis virulence factors and modulation by acylated homoserine lactones. J. Microbiol. Immunol. Infect. 41, 243–253 (2008).

  50. 50.

    Caporaso, J. G. et al. QIIME allows analysis of high-throughput community sequencing data. Nat. Methods 7, 335–336 (2010).

  51. 51.

    Kuczynski, J. et al. Using QIIME to analyze 16S rRNA gene sequences from microbial communities. Curr. Protoc. Bioinformatics 36, 10.7.1–10.7.20 (2011).

  52. 52.

    Edgar, R. C., Haas, B. J., Clemente, J. C., Quince, C. & Knight, R. UCHIME improves sensitivity and speed of chimera detection. Bioinformatics 27, 2194–2200 (2011).

  53. 53.

    Edgar, R. C. Search and clustering orders of magnitude faster than BLAST. Bioinformatics 26, 2460–2461 (2010).

  54. 54.

    Tsuda, A. et al. Influence of proton-pump inhibitors on the luminal microbiota in the gastrointestinal tract. Clin. Transl Gastroenterol. 6, e89 (2015).

  55. 55.

    Segata, N. et al. Metagenomic biomarker discovery and explanation. Genome Biol. 12, R60 (2011).

  56. 56.

    Ojiro, K. et al. MyD88-dependent pathway accelerates the liver damage of concanavalin A-induced hepatitis. Biochem. Biophys. Res. Commun. 399, 744–749 (2010).

  57. 57.

    Hayashi, A. et al. A single strain of Clostridium butyricum induces intestinal IL-10-producing macrophages to suppress acute experimental colitis in mice. Cell Host Microbe 13, 711–722 (2013).

  58. 58.

    Ondov, B. D. et al. Mash: fast genome and metagenome distance estimation using MinHash. Genome Biol. 17, 132 (2016).

  59. 59.

    Brisse, S. et al. wzi Gene sequencing, a rapid method for determination of capsular type for Klebsiella strains. J. Clin. Microbiol. 51, 4073–4078 (2013).

  60. 60.

    Pan, Y. J. et al. Capsular types of Klebsiella pneumoniae revisited by wzc sequencing. PLoS ONE 8, e80670 (2013).

Download references


We thank S. Chiba, S. Shiba, R. Morikawa, T. Katayama, A. Ikura, Y. Mikami and T. Sujino (Division of Gastroenterology and Hepatology, Keio University) for technical assistance and critical reading of this manuscript. This study was supported by the Japan Society for the Promotion of Science (JSPS) KAKENHI Grant-in-Aid (C) 16K09374 and (A) 15H02534, the Advanced Research and Development Programs for Medical Innovation (AMED-CREST; 16gm1010003h0001), the TAKEDA Science Fund, Ezaki Glico Co. Ltd and Keio University Medical Fund. K.H. was funded through AMED LEAP under grant number JP17gm0010003. A. Yoshimura was supported by the JSPS KAKENHI Grant-in-Aid (S) JP17H06175, Challenging Research (P) JP18H05376 and AMED-CREST JP18gm0510019 and JP18gm1110009.

Author information


  1. Division of Gastroenterology and Hepatology, Department of Internal Medicine, Keio University School of Medicine, Shinanomachi, Tokyo, Japan

    • Nobuhiro Nakamoto
    • , Nobuo Sasaki
    • , Ryo Aoki
    • , Kentaro Miyamoto
    • , Toshiaki Teratani
    • , Takahiro Suzuki
    • , Yuzo Koda
    • , Po-Sung Chu
    • , Nobuhito Taniki
    • , Akihiro Yamaguchi
    • , Toshiro Sato
    •  & Takanori Kanai
  2. AMED-CREST, Japan Agency for Medical Research and Development, Tokyo, Japan

    • Nobuo Sasaki
    •  & Takanori Kanai
  3. Institute of Health Science, Ezaki Glico Co., Ltd, Osaka, Japan

    • Ryo Aoki
  4. Miyarisan Pharmaceutical Co., Ltd, Tokyo, Japan

    • Kentaro Miyamoto
    •  & Takahiro Suzuki
  5. Department of Microbiology and Immunology, Keio University School of Medicine, Shinanomachi, Tokyo, Japan

    • Wataru Suda
    • , Mitsuhiro Kanamori
    • , Koji Atarashi
    • , Seiko Narushima
    • , Akihiko Yoshimura
    •  & Kenya Honda
  6. Laboratory of Metagenomics, Department of Computational Biology and Medical Sciences, The University of Tokyo, Chiba, Japan

    • Wataru Suda
    •  & Masahira Hattori
  7. Research Unit/Immunology & Inflammation, Sohyaku Innovative Research Division, Mitsubishi Tanabe Pharma Corporation, Kanagawa, Japan

    • Yuzo Koda
  8. Division of Gastroenterology, Department of Internal Medicine, University of Michigan, Ann Arbor, MI, USA

    • Nobuhiko Kamada
  9. Cooperative Major in Advanced Health Science, Graduate School of Advanced Science and Engineering, Faculty of Science and Engineering, Waseda University, Tokyo, Japan

    • Masahira Hattori
  10. Department of Bacterial Infection and Host Response, Graduate School of Medical and Dental Sciences, Tokyo Medical and Dental University, Tokyo, Japan

    • Hiroshi Ashida
  11. Department of Pathology, Keio University School of Medicine, Shinanomachi, Tokyo, Japan

    • Michiie Sakamoto
  12. RIKEN Center for Integrative Medical Sciences, Laboratory for Gut Homeostasis, Kanagawa, Japan

    • Koji Atarashi
    • , Seiko Narushima
    •  & Kenya Honda


  1. Search for Nobuhiro Nakamoto in:

  2. Search for Nobuo Sasaki in:

  3. Search for Ryo Aoki in:

  4. Search for Kentaro Miyamoto in:

  5. Search for Wataru Suda in:

  6. Search for Toshiaki Teratani in:

  7. Search for Takahiro Suzuki in:

  8. Search for Yuzo Koda in:

  9. Search for Po-Sung Chu in:

  10. Search for Nobuhito Taniki in:

  11. Search for Akihiro Yamaguchi in:

  12. Search for Mitsuhiro Kanamori in:

  13. Search for Nobuhiko Kamada in:

  14. Search for Masahira Hattori in:

  15. Search for Hiroshi Ashida in:

  16. Search for Michiie Sakamoto in:

  17. Search for Koji Atarashi in:

  18. Search for Seiko Narushima in:

  19. Search for Akihiko Yoshimura in:

  20. Search for Kenya Honda in:

  21. Search for Toshiro Sato in:

  22. Search for Takanori Kanai in:


N.N. and T.K. designed the project. N.N., N.S., R.A., K.M., T.T., Takahiro S., Y.K., P.-S.C., N.T., Akihiro Y., M.K. and H.A. performed the experiments. W.S. and M.H. performed the bacterial sequence, microbiome analyses and contributed to data discussions. K.A., S.N. and K.H. provided essential materials and contributed to data discussions. N.N., N.K., M.S., Akihiko Y., Toshiro S. and T.K. interpreted the experimental data. N.N., N.K. and Toshiro S. wrote the manuscript. T.K. critically revised the manuscript and supervised the study.

Competing interests

The authors declare no competing interests.

Corresponding authors

Correspondence to Toshiro Sato or Takanori Kanai.

Supplementary information

  1. Supplementary Information

    Supplementary Figures 1–8.

  2. Reporting Summary

  3. Supplementary Table 1

    Demographic characteristics of study subjects.

  4. Supplementary Table 2

    Fpkm values in RNA-seq analysis.

  5. Supplementary Table 3

    Phylogenetic relationship among K. pneumoniae strains used in this study.

  6. Supplementary Table 4

    Genes positively correlated with the epithelial pore-forming capacity.

About this article

Publication history