Research ArticleLiver disease

The Liver May Act as a Firewall Mediating Mutualism Between the Host and Its Gut Commensal Microbiota

See allHide authors and affiliations

Science Translational Medicine  21 May 2014:
Vol. 6, Issue 237, pp. 237ra66
DOI: 10.1126/scitranslmed.3008618


A prerequisite for establishment of mutualism between the host and the microbial community that inhabits the large intestine is the stringent mucosal compartmentalization of microorganisms. Microbe-loaded dendritic cells trafficking through lymphatics are arrested at the mesenteric lymph nodes, which constitute the firewall of the intestinal lymphatic circulation. We show in different mouse models that the liver, which receives the intestinal venous blood circulation, forms a vascular firewall that captures gut commensal bacteria entering the bloodstream during intestinal pathology. Phagocytic Kupffer cells in the liver of mice clear commensals from the systemic vasculature independently of the spleen through the liver’s own arterial supply. Damage to the liver firewall in mice impairs functional clearance of commensals from blood, despite heightened innate immunity, resulting in spontaneous priming of nonmucosal immune responses through increased systemic exposure to gut commensals. Systemic immune responses consistent with increased extraintestinal commensal exposure were found in humans with liver disease (nonalcoholic steatohepatitis). The liver may act as a functional vascular firewall that clears commensals that have penetrated either intestinal or systemic vascular circuits.


Liver disease is markedly increasing in incidence in the developed world, secondary not only to alcohol abuse and viral infections but also to progressive hepatic fatty changes and damage arising from metabolic syndrome. All forms of chronic liver disease, including nonalcoholic fatty liver disease (NAFLD), potentially lead to cirrhosis and liver failure. Nevertheless, even with very mild liver disease, such as excess fatty accumulation in the liver, the structure of the sinusoids and Kupffer cell function are compromised (13). Cirrhosis imposes markedly increased resistance to blood flow in the hepatic portal vein. The resulting increased pressure (portal hypertension) causes enlargement of the spleen, damage to the intestine, and abnormal vascular channels (shunts), where the portal circulation bypasses the liver and drains directly into the inferior vena cava returning blood to the heart. Despite the fearsome metabolic complications of liver failure, the commonest cause of death in patients with terminal cirrhosis is from infections (4). Many of these infections derive from oral or intestinal commensal organisms, resulting in peritonitis, spontaneous bacterial empyema, or bacteremia (5); so, understanding the impact of liver function and dysfunction on host-microbial mutualism is highly relevant to human health.

Mesenteric lymph nodes (MLNs) are known to act as a firewall during the induction of mucosal immunity through the dendritic cells (DCs)/lymphatic route in healthy mice with a normal intestinal mucosa (6). We have previously shown that live commensal bacteria are sampled by intestinal DCs, which can traffic to the MLNs (6, 7). These intestinal DCs normally do not penetrate further to reach systemic secondary lymphoid structures (6); thus, induction of intestinal mucosal immune responses is compartmentalized to the intestinal mucosal immune system.

The liver receives both venous blood from the intestine via the hepatic portal vein, and arterial blood from the hepatic artery arising from the celiac trunk of the aorta. Although the liver is ideally positioned as a second firewall for commensals that succeed in reaching the mesenteric or systemic vasculature, and dynamic studies have shown that its abundant Kupffer macrophage population can capture pathogens from the blood (8, 9), it has been unclear whether the liver has an obligatory role in limiting penetration of commensal microbes that are sampled in the course of inducing normal mucosal immune responses. Here, we show that the liver remains sterile when benign intestinal commensals are sampled through the lymphatic route in the healthy mouse intestine. However, the liver becomes important as an independent second firewall when commensal microbes penetrate the vasculature of the inflamed intestine or when commensal microbes have to be cleared from the systemic circulation. In both animal models and human patients with liver disease, loss of this central firewall function progressively disrupts host-microbial mutualism by increasing systemic exposure and systemic immune activation to intestinal commensals.


To determine whether the liver is also acting as a firewall during sampling of intestinal commensal bacteria in healthy wild-type mice, we compared numbers of culturable bacteria in the MLNs and the liver over time after gavaging germ-free wild-type mice with the reversible colonization strain Escherichia coli HA107 (10) or its parent strain JM83. In each case, live commensal bacteria penetrated the MLNs via lymphatic trafficking of intestinal DCs carrying commensals sampled at the intestinal epithelial surface (6, 7) (Fig. 1A). In contrast, the liver remained entirely sterile throughout the experiment (Fig. 1A). Similarly, in experiments designed to detect portal bacteremia at earlier times, live bacteria were never recovered from the portal blood in these mice (fig. S1A), suggesting that extremely efficient bacterial killing in the liver does not explain the lack of live bacterial recovery. Despite no penetration into the liver by live bacteria or induction of systemic immune responses (6, 10) after intestinal challenge in healthy mice, we could demonstrate drainage of bacterial breakdown products to hepatic tissues after gavage of E. coli JS219 metabolically labeled during growth with [14C]glucose (Fig. 1B). Therefore, penetration of the MLNs by live organisms during induction of mucosal immunity can be uncoupled from penetration to either the liver or the spleen in a healthy animal, despite substantial accumulation of metabolites originating from intestinal bacteria in both organs. This suggests that in healthy animals, the liver is not a required part of the firewall that compartmentalizes lymphatic-based mucosal immune induction, but rather functions for processing and detoxification of bacterial-derived material, or as a second firewall for those microbes that enter the mesenteric blood vessels as a result of increased epithelial penetration or during mucosal damage.

Fig. 1. Firewall function of the liver in host-microbial mutualism.

(A) Bacterial counts per gram of organ (liver, triangles; MLNs, dots) are shown at the indicated time points after oral gavage of NMRI (Naval Medical Research Institute) mice with 1010 E. coli K-12 JM83 (filled symbols) or reversible colonizing E. coli HA107 (open symbols). (B) Bacterial products and live bacteria in the MLNs (dots) and liver (triangles) after monocolonization of germ-free C57BL/6 mice with 14C-labeled E. coli JS219 at the indicated time points after gavage. Radioactive counts as a measure of bacterial products in liver and MLNs are shown as DPM on the left y axis (open dots and triangles), and live bacteria are shown as CFU per gram tissue on the right y axis (filled symbols). (C) Bacterial counts in the liver (triangles) and spleen (dots) 8 hours after intramesenteric vein injection of phosphate-buffered saline (PBS; open symbols) or 104 E. coli JM83 (filled symbols). (D) Liver CFU in 14-week-old C57BL/6 mice after a DSS treatment regimen of 1, 1.5, or 2% DSS in drinking water for 5 days each, followed by a normal water interval of 4 days each. Control mice received normal water throughout the experiment. Livers were then aseptically removed after the last drinking water period, and CFU were assessed after anaerobic and aerobic culture on blood agar or Luria broth (LB) plates, respectively. Shown are the pooled counts from both types of plates. (E) Bacterial counts in the liver of splenectomized (squares) or sham-operated mice (dots) 18 hours after intravenous (i.v.) injection of 107 E. coli JM83 (filled symbols) or PBS (open symbols) into the tail vein. (F) Bacterial counts per liver in C57BL/6 mice 48 hours after ligation of the hepatic artery (open circles) or sham operation (filled dots). All mice were intravenously challenged with 107 E. coli K-12 and sacrificed after 18 hours for plating of livers and spleens. All mice were 8 to 10 weeks of age; each data point represents a single animal from one experiment, and horizontal lines show means. The dashed lines show the detection limit. Unpaired t test was used to compare the groups.

Venous blood from the intestine drains mainly to the liver via the hepatic portal vein, which is formed by the confluence of the superior mesenteric vein and the splenic/inferior mesenteric vein (fig. S1B). To test the role of the liver as a firewall for the mesenteric vasculature, commensals were deliberately injected into the hepatic portal venous system: these were primarily cleared by the liver and, to a minor extent, by the spleen (Fig. 1C). Because model commensal microbes do not penetrate the liver when given as intestinal challenge doses to healthy pathogen-free mice, we next tested the effect of inducing intestinal inflammation. After treatment with dextran sodium sulfate (DSS), an intestinal dose of bacteria that would normally be found only within DCs in the MLNs, now consistently reached the liver (Fig. 1D) and, in some animals, could be detected in the spleen and peripheral blood (fig. S2A). This suggested that whereas the liver is not part of the firewall required to compartmentalize the mucosal lymphatics in healthy specific pathogen–free (SPF) mice, it may act as a vascular firewall for the mesenteric circulation under conditions of weakened mucosal-blood barrier function.

We next considered the role of the liver in clearing commensal microbes from the general systemic vasculature (not just the mesenteric/portal intestinal blood vessels). It is well documented that the spleen filters microorganisms within the general blood circulation. We have previously shown that live commensals given into the tail vein are cleared by the spleen, with only small numbers reaching the lymph nodes (6). However, splenectomy can be carried out in humans or experimental animals with relatively little disturbance in host-microbial mutualism, provided that there has been preimmunization against encapsulated bacteria (11). The main vascular exchange structures of the liver—the hepatic sinusoids—are a capillary-level confluence of branches of the hepatic portal vein (which receives the splenic vein as a tributary) and the hepatic artery (fig. S1B) (12). This anatomical arrangement caused us to question whether the liver functions at the heart of host-commensal mutualism by clearing microbes within the general systemic vasculature through two possible routes: (i) directly through filtration of blood from the hepatic artery, or (ii) indirectly by filtering any overflow from the spleen or transit through mesenteric vessels via the splenic→hepatic portal vein. Liver sinusoids are lined by enormous numbers of Kupffer cell phagocytes, which constitute more than 80% of all tissue macrophages in the body and about 20% of all hepatic cells and are present in normal numbers in germ-free animals (13, 14). These Kupffer cells are capable of taking up and killing microorganisms that they encounter (8, 9, 15, 16). Indeed, we found that doses of commensals given intravenously caused culturable commensals to be detected in the liver of wild-type mice, and this effect was not significantly different in splenectomized mice (Fig. 1E), verifying an effective systemic vascular clearance by the liver, independently of the spleen, probably mostly through the hepatic blood supply. We therefore examined the role of the hepatic arterial supply in delivering commensals to the liver of animals where the spleen is present. After ligating the hepatic artery (a nonessential vascular supply of the liver) before administering the venous dose of commensals, uptake by the liver was significantly reduced compared with that in sham-operated animals, in which the artery was exposed but not ligated (Fig. 1F) and clearance from the blood was also delayed [3 hours: ligated 203 ± 69 colony-forming units (CFU)/50 μl versus sham 41 ± 26 CFU/50 μl, mean ± SD, n = 5, P < 0.01], indicating that not only the spleen but also the liver is important to clear systemic bacteremia. The liver therefore forms a critical part of the filtration system that clears systemic commensal bacteremia through both the arterial and portal venous circuits.

To address the function of the liver to clear, as opposed to merely capture, commensals that have reached the systemic circulation of wild-type mice, we carried out trajectory experiments to measure the presence of commensals in the peripheral general circulation over time after an experimental intravenous dose. We started by asking whether depletion of Kupffer cells in the liver would impair the clearance of commensals from the general vasculature independently of the spleen. To do this, clodronate liposomes were injected into C57BL/6 mice to deplete phagocyte populations, after previous splenectomy (17). Nevertheless, even in the absence of the spleen, clodronate treatment resulted in delayed clearance of bacteremia from a challenge dose of E. coli K-12 (Fig. 2A and fig. S2B) given 3 days later into the tail vein. Although there is probably a small effect of macrophages outside the liver, hepatic Kupffer cells form the majority of body macrophages and culture experiments showed that >86% of culturable model commensal bacteria were recovered from either the liver or the spleen after intravenous challenge in wild-type unmanipulated animals. Given that there is also delayed clearance of commensals from the peripheral general vasculature in the setting of liver disease (Fig. 2B and fig. S3, A to C) and that mice with liver disease show increased liver bacterial burdens after intravenous bacterial challenge (fig. S3D), we concluded that bacterial clearance by Kupffer cells in the liver is an important mechanism for the removal of live commensals from the blood, mirroring their role in dynamic studies of pathogen clearance.

Fig. 2. Liver firewall function is compromised in liver disease independently of alterations in innate immunity.

(A) Bacterial counts in peripheral blood at the indicated time points after intravenous injection of 107 live E. coli into the tail vein of splenectomized and Kupffer cell–depleted (clodronate, filled dots, solid line) or healthy control mice (open circles, dashed line). Kupffer cells were depleted 3 weeks after splenectomy in C57BL/6 mice by injecting clodronate liposomes 3 days before live bacterial challenge. Lines connect the means of each group. The horizontal dashed line represents the detection limit. (B) Bacterial counts in peripheral blood at the indicated time points after intravenous injection of 107 live E. coli into the tail vein of BDL (filled circles, solid lines) or sham-operated mice (open circles, dashed lines), 3 weeks after surgery. Shown are pooled data from two independent experiments. Lines connect the means of each group. The horizontal dashed line represents the detection limit. (C) Bacterial counts per spleen of the same mice as in (B) 6 hours after live bacterial challenge. Horizontal lines show means. (D) Bacterial counts in the spleen of CCl4-treated (black dots) or healthy control mice (open circles) 6 hours after live bacterial challenge with 107 E. coli into the tail vein. C57BL/6 mice were intraperitoneally injected two times per week with CCl4 or olive oil, respectively, for 16 weeks. Horizontal lines show means. (E and F) Bacterial counts in peripheral blood (E) and spleen (F) in PBS-injected (open circles) or LPS-injected (filled dots) mice 3 hours after injection of 107 live E. coli into the tail vein. All mice were treated three times per week with PBS alone or with 1 μg of LPS from Salmonella typhimurium intraperitoneally and analyzed 48 hours after the final treatment. All mice were then intravenously injected with 107 live E. coli. Bacterial counts were determined per 50 μl of blood or organ. All mice were 8 to 10 weeks of age; each data point represents a single animal from one experiment; horizontal lines show means. Unpaired t test was used to compare the groups.

To examine the effects of liver dysfunction on host-microbial mutualism specifically, we studied two independent models of liver disease. First, we measured the trajectory of clearance of E. coli K-12 challenge doses in fibrotic mice after bile duct–ligated (BDL) compared with sham-operated controls. This showed significantly reduced efficiency of clearance of commensal bacteria from the blood (P = 0.007 at 3 hours, P = 0.005 at 6 hours) and spleen (P = 0.03) of fibrotic mice (Fig. 2, B and C). This was observed despite a marked increase in the number of blood granulocytes in the circulation of BDL mice (fig. S4A) and evidence of stronger release of acute-phase proteins (lipocalin 2) into the serum (fig. S4B). Therefore, despite heightened inflammatory responses to commensal bacterial challenge, liver disease was associated with a strong reduction in bacterial clearance ability of the mice. Because BDL in mice induces cholestasis and thus altered bile salt recycling, and some degree of intestinal pathology, we used an additional model of liver disease that allowed us to study the trajectory of host-microbiota mutualism in liver disease. Clearance of an intravenous challenge dose of bacteria from the blood, liver, and spleen was also delayed if the liver was damaged using carbon tetrachloride (CCl4) treatment (causing less severe hepatic fibrosis at an earlier stage) (18) (Fig. 2D and figs. S3 and S5). CCl4-treated mice showed a normal distribution of liver Kupffer cells in histological specimens, supporting the concept of altered phagocytic capacity rather than simply disrupted liver architecture, reduced numbers of immune cells, and fibrotic replacement in liver disease (fig. S5D).

We speculated whether this altered bacterial clearance ability during murine liver disease could be due to “exhaustion” of the innate immune system by accumulated bacterial products in the systemic circulation. To investigate this possibility, we serially pretreated mice with increasing doses of Salmonella lipopolysaccharide (LPS) over 21 days (to match the timing of bacterial challenge in BDL mice after surgery). However, when these preconditioned mice were challenged with an intravenous dose of a nonpathogenic E. coli, they displayed greatly enhanced bacterial clearance compared to control animals, despite non–cross-reactivity between the LPS used for pretreatment and that of the challenge strain (Fig. 2, E and F, and fig. S6, A and B). Therefore, increased exposure to microbial products alone was insufficient to cause the observed commensal clearance defects, in the absence of liver disease. We also considered whether the integrity of the MLN firewall was impaired in the setting of liver disease. We verified that an intestinal challenge dose of E. coli reached the MLNs equivalently whether there was hepatocellular damage in the CCl4 model, but there was no penetration into thoracic duct lymph, showing that the mesenteric firewall remained intact (fig. S6C). These defects of commensal bacterial clearance from the systemic circulation could be detected in acute CCl4 liver disease in mice colonized with a strict altered Schaedler flora (ASF; a very simple benign microbiota consisting of eight microbial species) and with no evidence for dysbiosis or defects in MLN barrier function or intestinal integrity (figs. S6C and S7A). More advanced liver disease (such as observed in BDL) certainly causes secondary histological damage to the intestine (fig. S7B) as well as low-grade intestinal inflammation measured through the fecal content of the neutrophil activation marker lipocalin 2 (fig. S7C) as a result of increased pressure in the portal circulation. Therefore, in the face of advancing liver disease, there is a combination of an increased probability that intestinal microbes translocate into the vasculature and a decreased ability to clear these microbes from portal or systemic blood.

To follow the functional consequences of liver dysfunction for host-microbial immune mutualism in both mouse and human, we needed to interrogate low levels of systemic exposure to intestinal commensal microbes independently of intestinal or systemic challenge doses both in murine models of liver disease and in human patients. To do this, we made use of immunoglobulin (Ig) isotype differences that occur when commensal bacteria induce immune responses in different body compartments (19). We have previously demonstrated that the response to commensals typically found in SPF mice is compartmentalized within the mucosa and, hence, to mucosal IgA (6, 19). Serum IgG against commensals can be experimentally induced by deliberate injection of commensal organisms into the tail vein, but is absent from SPF wild-type mice (19). Anti-commensal serum IgG also occurs spontaneously in mice, which inefficiently clear commensals (defective Toll-like receptor signaling resulting from MyD88/TRIF adaptor deficiency or in mice deficient in the enzyme pathways required to generate oxygen or nitric oxide radicals for biocidal activity in phagocytes), because live commensals can then reach and induce immunity in systemic secondary lymphoid structures (20). Therefore, the presence of high titers of specific serum IgG directed against commensal microbes can be used as a surrogate marker of commensal penetration of the systemic immune system.

To validate the concept of examining induction of specific serum IgG against dominant intestinal commensals, we used a trajectory of exposure to CCl4 as a sequential progressive model of liver disease. In early liver disease, specific serum IgG against commensals of the gut microbiota was absent, but appeared as the liver disease progressed. The antibodies were measured using a specific fluorescence-activated cell sorting (FACS)–based assay that assesses specific surface Ig binding to the endogenous microbes (20) normalized for total IgG concentration (Fig. 3, A to F). Specific serum IgG against commensal microbes, such as autologous Lactobacillus sp. or bulk cultured aerobes, was also increased in mice or rats with a more diverse microbiota that had developed fibrosis after BDL (Fig. 3, E and F, and fig. S8).

Fig. 3. Increased systemic immune priming by intestinal commensal microbes in animal models with liver dysfunction.

(A to D) Serum IgG1 (A and C) or IgG2b (B and D) titers against bacteria isolated from the mice’s own feces (Clostridium sp.) in healthy control (Ctrl) (black dots) and fibrotic C57BL/10 mice (open circles) as determined by bacterial FACS. Liver fibrosis was induced by 6 (A and B) or 12 (C and D) weeks of treatment with CCl4, intraperitoneally administered two times per week, or olive oil as control. (E and F) Serum IgG1 (E) and IgG2b (F) titers of the mice shown in (A) to (D). IgG titers were calculated by fitting four-parameter logistic curves to each sample and determining the concentration of IgG required to give a geometric mean fluorescence intensity (FI) binding above background staining. The inverse of this IgG concentration (μg−1 ml) is shown for ease of interpretation. Unpaired t test was used to compare the groups. (G) Serum IgG1 titers against an autologous commensal bacterium (Lactobacillus murinus) isolated from the feces in BDL (open circles) or sham-operated control (black dots) mice 21 days after surgery. (H) Serum IgG titers against an anaerobic bulk culture inoculated from the animals’ feces 28 days after BDL (filled dots) or sham surgery (open circles) in rats. (A to H) Dose titrations of serum were incubated with the bacterial strain and specific IgG1 or IgG2b binding visualized by FACS analysis. Data were then normalized to the total amount of IgG1 or IgG2b present in serum (x axis) as determined by enzyme-linked immunosorbent assay (ELISA). All curves represent individual mice from one of two independent experiments with 7 to 10 mice per group.

A critical issue in modeling human disease is that rodent models of liver fibrosis and cirrhosis develop over weeks and months, as compared with human disease that can develop over decades. The question is whether early liver disease in humans with its associated microvascular and phagocytic abnormalities (1, 3) is associated with abnormalities of handling and compartmentalizing intestinal commensals, or whether disturbance of host-microbial mutualism only develops with very advanced liver pathology. To address this question, we assembled two separate cohorts of patients from different university hospitals in Italy and Switzerland. In the first cohort, we compared patients with NAFLD and matched control patients. In the second, we were able to examine patients with liver steatosis, nonalcoholic steatohepatitis (NASH), and cirrhosis compared with controls. In most cases of liver steatosis or NASH, the patient has no symptoms of the disease and there is no metabolic liver failure, although some cases will later progress to cirrhosis. Strikingly, in both cohorts, these early cases of human liver disease also had evidence of increased serum IgG and IgA responses against both aerobic and anaerobic nonpathogenic intestinal commensals independently of the disease stage (Fig. 4A and fig. S9A). Euclidean heatmaps of the responses were also discriminatory for diagnosis in unbiased cluster analysis (Fig. 4B and fig. S9B).

Fig. 4. Disturbance of host-microbial mutualism in human patients with liver dysfunction.

(A) Serum IgG titers against the indicated bacteria in NAFLD patients (open symbols) compared to age- and sex-matched healthy controls (filled dots). Pure cultures of the indicated bacteria were stained with dose titrations of serum from patients or controls. Serum antibody coating of bacteria was visualized using monoclonal DyLight-conjugated anti-human IgG and quantified per bacterium by flow cytometry. Resulting geometric mean fluorescence intensity (FI) was plotted against total IgG added to the assay as determined by ELISA, and IgG titers were calculated by fitting four-parameter logistic curves to each donor and determining the concentration of IgG required to give a geometric mean fluorescence intensity binding of 80. The inverse of this IgG concentration (μg−1 ml) is shown for ease of interpretation. Each point represents an individual subject, and lines show means. Unpaired t test or one-way analysis of variance (ANOVA) and Tukey posttest were used to compare the groups; *P ≤ 0.05, **P ≤ 0.01. (B) Cluster analysis from Basel NAFLD patients with different stages of liver disease (steatosis, blue; NASH, yellow; cirrhosis, red) and age- and sex-matched healthy controls (green). Heatmaps were generated using a Euclidean distance function with complete linkage clustering in the statistical package R using the package “gplots version 2.8.0,” function “heatmap.2.” Red indicates increased and blue indicates decreased titers compared to the mean of the entire population.


We have found that compartmentalization of commensal intestinal microbes is defective in animal models of liver disease. This was not due to loss of function in the well-described primary lymphatic firewall that functions during normal mucosal sampling of intestinal bacteria, but due to failure of a hepatic vascular firewall that is required to clear blood-borne commensals from the mesenteric and systemic vasculature efficiently. Disturbed compartmentalization during murine liver dysfunction resulted in spontaneously increased systemic immune responses to resident commensals without intestinal challenge or overt intestinal pathology. In mice with liver disease, it is possible to show directly that commensal bacteria entering the circulation are inefficiently cleared despite heightened innate immunity. There are increased serum Ig responses to intestinal commensals in mice with liver dysfunction, probably as a consequence of increased exposure of their systemic lymphoid structures to commensal bacteria (19).

Because murine and human liver disease is associated with increased complement activity (21), impaired liver bactericidal capacity is likely to be a major contributor to increased systemic commensal bacterial spread in liver disease. In human patients, similar systemic immune responses to commensals are observed very early in the course of the human condition long before cirrhosis has developed. Nevertheless, although it was not possible to collect patient fecal samples to normalize systemic anti-microbiota immune responses to each patient’s microbiota composition, we found systemic immune responses against a selection of nonpathogenic commensal bacteria to be discriminatory for diagnosis in unbiased cluster analysis. It is known that dysbiosis of intestinal microbes can precede and trigger liver damage (22, 23). This is also likely to be an initiating event in at least some human patients, although it is not required per se for the disturbance in host-microbial mutualism because in our mouse model, the very simple ASF microbiota was tightly controlled and liver damage was provoked by other methods. Thus, we were able to separate dysbiosis from the direct effects of liver disease on systemic bacterial handling. This revealed a critical requirement for liver ultrastructure and function to clear bacteria. Our observations in human liver disease highlight that such mechanisms are essential to maintain normal host-microbiota homeostasis in the context of frequent “real-world” challenges to intestinal integrity. It is possible that monitoring anti-microbiota IgA and IgG titers in liver patients may identify those with the highest risk of infectious complications and may thus have diagnostic value. Failed compartmentalization and vascular clearance of commensals are likely to be important factors resulting in catastrophic infective complications in patients with end-stage liver disease such as spontaneous bacterial peritonitis and sepsis, which are predominantly caused by organisms of intestinal origin.


Study design

Mice. In hypothesis-driven experimental designs, we addressed the role of the liver in host-microbiota mutualism. Mice harboring defined intestinal floras were studied to determine the functional role of the liver in host-microbial mutualism. The experimental techniques included (i) intramesenteric injection, (ii) hepatic artery ligation and splenectomy, and (iii) disturbing liver architecture and Kupffer cell function (CCl4 treatment, BDL, and clodronate liposome depletion). These animals were then challenged with commensal bacteria applied through different vascular routes, and their immune responses against commensal microbes were monitored.

Humans. In a retrospective cohort design (using two independent cohorts), frozen serum samples from patients with different stages of fatty liver disease and age- and sex-matched healthy controls were used to analyze systemic antibody responses to commensal bacteria.

Animal experiments

All animal experiments were carried out with permission in accordance with local rules for the care and use of laboratory animals.

Liver fibrosis models. For analysis of serum anti-microbiota antibodies, BDL treatments were performed at Ghent University Hospital, Belgium, in C57BL/10 mice as described previously (24). Briefly, a midline abdominal incision was made, and the common bile duct was isolated and occluded with a double ligature of a nonresorbable suture. The first ligature was made below the junction of the hepatic ducts, and the second was made above the entrance of the pancreatic duct. The common bile duct was sectioned between the two ligatures, and mice were euthanized 6 weeks after BDL. In sham-operated mice, ligatures were placed identically, but no section was made. CCl4 was administered intraperitoneally in C57BL/6 mice colonized with an ASF consisting of eight bacterial species [Lactobacillus acidophilus (strain ASF 360), Lactobacillus salivarius (strain ASF 361), Bacteroides distasonis (strain ASF 519), Flexistipes sp. (ASF 457), and Clostridium cluster XIV group (ASF 356, ASF 492, ASF 500, and ASF 502)] twice weekly (1:4 dissolved in olive oil; 60 μl per animal), whereas control animals received 60 μl of olive oil. Mice were analyzed after 6, 12, and 16 weeks.

For bacterial clearance measurements, A.d.G. performed BDL treatments in gnotobiotic NMRI mice in Bern, as described above. Mice were analyzed 3 weeks after surgery.

Rat BDL was performed as described previously (25). Seven-week-old male Wistar rats (Charles River Laboratories) were housed in individually ventilated cages at the University of Bern.

LPS preconditioning experiments. Swiss Webster mice housed under ultraclean conditions at the Department of Microbiology, ETH Zürich, were treated three times per week with PBS alone or with 1 μg of LPS from S. typhimurium intraperitoneally and analyzed 48 hours after the final treatment.

DSS colitis induction. Fourteen-week-old SPF C57BL/6 mice were treated with three cycles of DSS (molecular weight 36,000 to 50,000, MP Biomedicals) in drinking water for 5 days (week 1: 1% DSS, week 2: 1.5% DSS, and week 3: 2% DSS) followed by 4 days with normal drinking water. All mice were anesthetized with CO2 before terminal bleeding, and livers were aseptically removed for bacterial analysis.

Hepatic artery ligations. After midline abdominal incision, the hepatic artery was prepared and two ligatures were placed, followed by sectioning of the hepatic artery. In sham-operated animals, the hepatic artery was only exposed and the ligatures were placed but not sectioned.

Splenectomies. Splenectomy was performed in 8- to 10-week-old mice. After midline abdominal incision, the spleen was exposed, and splenic artery and vein were ligated before removal of the spleen and suture of the peritoneum and skin.

Intramesenteric vein bacterial challenge. After midline abdominal incision, 107 E. coli JM83 or PBS in a volume of 100 μl was injected into the mesenteric veins of 8- to 10-week-old mice with an insulin syringe followed by compression of the vessel for 2 min to stop bleeding. The peritoneum and skin were then closed with a suture. Animals were euthanized 8 hours after challenge, and organs were aseptically removed for plating and counting of CFUs.

Preparation of clodronate liposomes. Clodronate liposomes were generated as described previously (17), and the effectiveness of depletion was validated by FACS. Briefly, a dry lipid mixture was solubilized with clodronate (Ostac, Boehringer Mannheim), and the resulting multilammelar vesicles were filter-extruded. Unencapsulated clodronate was then removed by ultrafiltration followed by size exclusion chromatography. Liposomes were then reconcentrated (3 to 5 mg/ml) and sterile-filtered. C57BL/6 mice were intravenously injected with liposomes containing 2 to 2.5 mg of clodronate 3 weeks after splenectomy and challenged with live E. coli K-12 into the tail vein 3 days later to determine bacterial clearance capacity.

Peripheral blood granulocyte numbers. Determination of granulocyte counts per microliter was performed with a Vet ABC animal blood counter (Medical Solution GmbH).

From all animal models, fresh fecal pellets and serum were collected and used immediately or stored at −80°C before use.

All animal experiments were approved by the local Animal Care Committee.

Colonization with 14C-labeled E. coli K-12

Germ-free C57BL/6 mice were monocolonized with 14C-labeled E. coli JS219, and liver biopsies were taken at 7, 24, 48, and 126 hours after colonization for culture on LB agar and to measure levels of radioactivity. Samples for radioactivity were dissolved in 1 ml of NCS II Tissue Solubilizer (GE Healthcare) for a minimum of 1 hour at 56°C, at which time the pH level was neutralized with 100 μl of 100% glacial acetic acid, and the sample was mixed with 18 ml of Ultima Gold liquid scintillation cocktail (PerkinElmer) before measuring radioactivity. 14C radioactivity levels in each sample were counted in a Tri-Carb 2300TR Liquid Scintillation Analyzer (Packard) for a maximum time of 5 min. Colorimetric quench curves were used to ensure accurate measuring of samples. Background levels of radioactivity were determined by measuring the levels of 14C in MLNs and livers of germ-free C57BL/6 mice (n = 2).

E. coli HA107/JM83 monocolonization

Colonization was performed as described previously (10). Briefly, d-Ala (200 μg/ml)/m-DAP (50 μg/ml)–supplemented LB cultures were aseptically inoculated from single colonies of E. coli strains HA107 or JM83 and incubated with shaking at 160 rpm at 37°C for 18 hours. Bacteria were harvested under axenic conditions by centrifugation (15 min, 3500g, 4°C), washed in sterile PBS, and concentrated to a density of 2 × 1010 CFU/ml in PBS. The bacterial suspensions were sealed in sterile tubes and imported into flexible film isolators, where 500 μl (1010 CFU) was gavaged into the stomachs of germ-free Swiss Webster mice. Fecal samples exported from the isolator were bacteriologically analyzed to monitor E. coli shedding and microbiological status of the inoculated mice. The same protocol was used to prepare bacteria for intravenous injections.


Animals were euthanized, and livers, spleens, and MLNs were removed aseptically. For thoracic duct lymph sampling, animals were gavaged with olive oil 2 hours before analysis. The thoracic duct was then exposed and the lymph was sampled with a sterile microcapillary for plating on agar plates. Organs were homogenized in 0.5% Tergitol/PBS using a TissueLyser (Retsch MM400, 25 Hz) and sterile stainless steel ball-bearing beads, and homogenates were plated on supplemented agar plates for overnight culture at 37°C and CFU quantification.

Isolation of fecal bacteria

Fresh fecal pellets were dissolved in 1 ml of sterile PBS and streaked out onto LB and blood agar plates. Plates were then incubated aerobically (LB) and anaerobically (blood agar) for 24 or 48 to 72 hours, respectively. Single colonies were grown in liquid brain-heart infusion medium (Oxoid, CM0225) and used for bacterial FACS and 16S polymerase chain reaction (PCR) to identify the bacterial species.

16S ribosomal RNA sequencing

One milliliter of a 5-ml overnight culture was pelleted by centrifugation at 7000 rpm for 3 min. Pellets were then dissolved in 250 μl of direct PCR lysis reagent (Viagen Biotech Inc., [102-T]) containing 5 μl of proteinase K (Roche, 3115887001). Samples were then incubated for 1 hour at 55°C followed by 1 hour at 85°C to inactivate proteinase K. DNA template (5 μl) was then used for PCR, using 1 μl of forward 10 μM primer FD1 (AGAGTTTGATCCTGGCTCAG), 1 μl of forward 10 μM primer FD2 (AGAGTTTGATCATGGCTCAG), 1 μl of reverse 10 μM primer RP1 (CGGTTACCTTGTTACGACTT), 1 μl of deoxynucleotide triphosphate (Promega, U1515), 0.4 μl of Taq polymerase (Promega, M3175), 10 μl of 5× buffer, and 30.6 μl of molecular water per reaction. PCR was performed according to the following setup: 94°C, 5 min; 94°C, 1 min; 43°C, 1 min; 72°C, 2 min, repeat 35×; 72°C, 7 min; 10°C, forever. PCR purification was then performed with the Miniprep Kit (Qiagen, 28004) and DNA eluted in 10 μl of water. DNA concentration was measured with a NanoDrop ND-1000 (Thermo Scientific), and 1 μl of DNA plus 1 μl of FD1 primer in 18 of μl molecular water was sent for 16S sequencing (Microsynth).

NAFLD cohorts and controls

Serum samples from two different patient cohorts of NAFLD were collected. The Basel cohort consisted of 28 patients with elevated aminotransferase levels and drinking less than 40 g (males) or 20 g (females) of alcohol per week with biopsy-proven NAFLD [age (mean ± SD): 49 ± 16 years, sex (male/female): 18/10], and the Rome cohort consisted of 16 patients with biopsy-proven NASH drinking less than 20 g of alcohol per day [age (mean ± SD): 50 ± 13 years, sex (male/female): 2/14] (26). All patients had biopsy-proven NAFLD, and alcohol intake was assessed by patient history. For Rome patients, also one close relative was interviewed, and an Alcohol Use Disorders Identification Test (AUDIT) test was performed and had to be less than 7 (27). Patients from the Basel cohort were further divided into a steatosis, NASH, and cirrhosis group on the basis of the NAFLD activity score [NAS (28)]. Frozen serum samples were stored at −80°C before analysis. Healthy control samples were obtained from individuals interviewed to exclude acute illness or previous liver or intestinal disease. A selection of 4 to 16 precisely age- and sex-matched controls was used for each patient cohort. All experiments were performed in accordance with the local ethical committee approval and guidelines.

Enzyme-linked immunosorbent assays

Total concentrations of antibody isotypes in mouse or human serum were determined by sandwich ELISA. Mice: Coating antibodies were goat anti-mouse IgG1, IgG2b, IgA, and IgM (Serotec), and detection antibodies were horseradish peroxidase (HRP)–conjugated anti-mouse IgG, IgM, or IgA (Sigma). Standards were myeloma-derived purified IgG1, IgG2b, IgM (Invitrogen), and IgA (BD Pharmingen). Humans: IgG and IgA antibodies were determined using the Human IgG and IgA ELISA Quantitation Kit (LuBio Science, E80-104, E80-102).

Bacterial FACS analysis

Bacterial FACS analysis was performed as described previously (10, 20). Briefly, 5 ml of LB cultures was inoculated from single colonies of plated bacteria and cultured overnight at 37°C without shaking. One milliliter of culture was gently pelleted for 3 min at 7000 rpm in an Eppendorf minifuge and washed three times with sterile-filtered PBS/2% bovine serum albumin (BSA)/azide before resuspending at about 107 bacteria/ml. Mouse or human serum was diluted 1:10 in PBS/2%BSA/azide and heat-inactivated at 60°C for 30 min. The serum solution was then spun at 13,000 rpm in an Eppendorf minifuge for 10 min to remove any bacteria-sized contaminants, and the supernatant was used to perform serial dilutions. Serum solution (25 μl) and 25 μl of bacterial suspension were then mixed and incubated at 4°C for 1 hour. Bacteria were washed twice before resuspending in monoclonal FITC (fluorescein isothiocyanate) anti-mouse IgG2b or IgA, phycoerythrin (PE) anti-mouse IgG1, and allophycocyanin (APC) anti-mouse IgM (BD Pharmingen, 559354, 550083, 553395, and 550676), or DyLight 647 anti-human IgG, FITC anti-human IgA (Jackson ImmunoResearch). After a further hour of incubation, the bacteria were washed once with PBS/2% BSA/azide and then resuspended in 2% paraformaldehyde (PFA)/PBS for acquisition by FACSArray using FSc (forward scatter) and SSc (side scatter) parameters in logarithmic mode. Data were analyzed using FlowJo software (Tree Star), and titers were calculated by fitting four-parameter logistic curves to each donor and determining the concentration of IgG or IgA required to give a geometric mean fluorescence intensity binding significantly above background. The inverse of this IgG or IgA concentration (μg−1 ml) is shown, for ease of interpretation.

Heatmap analysis

Heatmaps and correlation analyses were performed with R software ( Heatmaps were generated for scaled, normalized titer data using a Euclidean distance function with complete linkage clustering in R using the package gplots version 2.8.0, function heatmap.2.

Serum lipocalin 2 ELISA

Lipocalin 2 values, as a marker of granulocyte activation and an acute-phase response, were determined by ELISA according to the manufacturer’s instructions with a few modifications (human: R&D Systems, DY1757, murine: R&D Systems, DY1857). Nunc-Immuno Plates C96 MaxiSorp (Milian, 430341) were coated with 50 μl of capture antibody (1:200 in PBS) overnight at 4°C in a humidified chamber. After washing in PBS/Tween 0.05% (Sigma, P2287-500ML) and blocking in 150 μl of PBS/1% BSA for 15 min at room temperature, samples and standards were added in threefold dose titrations starting at 1:10 (serum and standard) or neat (fecal pellets, resuspended in 1 ml of PBS) and incubated overnight at 4°C in a humidified chamber. After washing in PBS/0.05% Tween, 50 μl of detection antibody (1:200 in PBS/2% BSA) was added and plates were incubated for 1 hour at room temperature. Plates were then washed in PBS/0.05% Tween, and 100 μl of HRP-streptavidin (BioLegend, 405210, 1:1000 in PBS) was added for 1 hour. Plates were then washed and developed with 100 μl of substrate (10 ml of substrate buffer, 1 mg of ABTS, 10 μl of H2O2). Optical density was measured at 415 nm, and four-parameter curves were generated to compare EC50 (median effective concentration) values of samples and standards.

Histological analysis

Intestinal tissue samples were embedded in optimal cutting temperature, snap-frozen in liquid nitrogen, and stored at −80°C. Cryosections (5 μm) were mounted on glass slides, air-dried for 2 hours at room temperature, and stained with hematoxylin and eosin (H&E). Livers were fixed in 4% formaldehyde for 24 hours followed by 70% ethanol for 48 hours. Samples were then paraffin-embedded, cut, and stained with H&E, F4/80, or Masson’s trichrome, respectively.

Statistical analysis

Differences were analyzed for statistical significance using Prism 4 for Macintosh (GraphPad Software Inc.). The details of the test carried out are indicated in figure legends. Where data were approximately normally distributed, values were compared using either a Student’s t test for single variable or two-way ANOVA for two variables. Approximate P values were computed for two-way ANOVA. Where data were not normally distributed (for example, including bacterial CFU counts close to or equal to zero), nonparametric two-tailed Mann-Whitney U tests were applied. In all cases, P < 0.05 was considered significant.


Fig. S1. The complexity of the liver vascular supply and its exchange structures.

Fig. S2. Bacteremia in DSS-treated animals and efficiency of clodronate depletion.

Fig. S3. Trajectory of bacterial clearance of intravenous challenge doses in CCl4-treated mice.

Fig. S4. Peripheral blood granulocytes and serum lipocalin 2 in BDL mice.

Fig. S5. Liver histology in BDL and CCl4-treated mice.

Fig. S6. Clearance of intravenous bacterial challenge doses in LPS-preconditioned mice and mesenteric firewall function in CCl4-treated mice.

Fig. S7. Intestinal histology and fecal lipocalin 2 in BDL and CCl4-treated mice.

Fig. S8. Serum IgG titers in BDL mice and rats.

Fig. S9. Serum IgA responses in human liver patients and healthy controls.


  1. Acknowledgments: We thank J. Kirundi, B. Blomme, S. Rupp, J. Cahenzli, B. Flogerzi, C. Furer, M. Terrazos, H. Sägesser, J. Reichen, and L. Idrissova for technical support; R. A. Schwendener for generation of clodronate liposomes; and R. Blumberg, R. Germain, M. R. Thursz, C. Müller, and R. Wiest for helpful comments. Funding: The Swiss National Science Foundation (310030-124732 and 313600-123736 to A.J.M.), the Canadian Institutes of Health Research, and the Genaxen Foundation. The human NAFLD cohort from Basel was supported by Astra-Altana research funding. M.L.B. was supported by the Swiss Cancer League and the Swiss National Foundation (grant 313600-123736/1). S.H. and K.D.M. were supported by an ERC Starting Grant from the European Research Council under the European Union’s Seventh Framework Programme (FP/2007-2013), ERC Grant Agreements 281904 (to S.H.) and 281785 (to K.D.M.). N.D. was supported by the Swiss National Foundation (grant 138392). Author contributions: M.L.B. designed, performed, and analyzed all experiments and wrote the paper. A.J.M. designed and performed experiments shown in Figs. 1 (C, E, and F) and 2A, and figs. S1A and S2B, provided funding and supervisory support, and wrote the paper; E.S. designed and analyzed all experiments and performed experiments shown in Fig. 2 (E and F) and fig. S6 (A and B), provided supervisory support, and wrote the paper; S.H. designed, performed, and analyzed experiments shown in Fig. 1A and figs. S5 and S7. K.D.M. performed experiments shown in fig. S6C and provided technical and supervisory support; A.d.G. performed BDLs; M.A.E.L. performed experiments shown in Fig. 1B. R.F. and F.R. performed CCl4 treatments and assisted in generating data for Fig. 2D and fig. S3, N.D. performed DSS treatments, D.S. performed histological sections and stainings, and M.W. performed CCl4 treatments. L.M., A.G., H.V.V., N.P., C.B., and M.H.H. helped collect human NAFLD samples. Competing interests: The authors declare that they have no competing interests.
View Abstract

Navigate This Article