A gut-vascular barrier controls the systemic dissemination of bacteria

See allHide authors and affiliations

Science  13 Nov 2015:
Vol. 350, Issue 6262, pp. 830-834
DOI: 10.1126/science.aad0135

A gut bacterial containment system

Trillions of bacteria selectively inhabit our guts, but how do our bodies keep them contained? Spadoni et al. describe a “gut-vascular barrier” that prevents intestinal microbes from accessing the liver and the bloodstream in mice (see the Perspective by Bouziat and Jabri). Studies with human samples and in mice revealed that the cell biology of the gut-vascular barrier shares similarities with the blood-brain barrier of the central nervous system. Pathogenic bacteria such as Salmonella typhimurium could penetrate the gut-vascular barrier in mice, gaining access to the liver and bloodstream, in a manner dependent on the Salmonella pathogenicity island 2–type III secretion system.

Science, this issue p. 830; see also p. 742


In healthy individuals, the intestinal microbiota cannot access the liver, spleen, or other peripheral tissues. Some pathogenic bacteria can reach these sites, however, and can induce a systemic immune response. How such compartmentalization is achieved is unknown. We identify a gut-vascular barrier (GVB) in mice and humans that controls the translocation of antigens into the blood stream and prohibits entry of the microbiota. Salmonella typhimurium can penetrate the GVB in a manner dependent on its pathogenicity island (Spi) 2–encoded type III secretion system and on decreased β-catenin–dependent signaling in gut endothelial cells. The GVB is modified in celiac disease patients with elevated serum transaminases, which indicates that GVB dismantling may be responsible for liver damage in these patients. Understanding the GVB may provide new insights into the regulation of the gut-liver axis.

Upon ingestion, food antigens can access the lymphatics to reach the mesenteric lymph nodes (MLNs) (1) and the blood stream (portal vein) to reach the liver (2). In contrast, the microbiota cannot access the liver (3) and can reach the spleen only when the MLNs are excised (4). This suggests that the microbiota are actively excluded from the bloodstream and that the MLNs provide a firewall for the systemic dissemination of the microbiota from the lymphatics (5). What determines antigen access to the bloodstream is unknown.

We hypothesized the existence of a gut-vascular barrier (GVB) that might control the type of antigens that are translocated across blood endothelial cells (ECs) to reach the portal vein. To evaluate the presence of such a barrier, we injected mice intravenously with fluorescein isothiocyanate (FITC)–dextran of different molecular sizes and analyzed leakage of the dye into the intestine. We observed that there is an endothelial barrier that discriminates between differently sized particles of the same nature. FITC-dextran of 4 kD freely diffused through the ECs, whereas FITC-dextran of 70 kD could not (Fig. 1A; fig. S1, A and B; and movies S1 and S2). However, after oral infection with Salmonella enterica serovar Typhimurium, which disseminates systemically in mice (6), 70 kD dextran was readily released from the bloodstream (Fig. 1A, fig. S1, and movies S3 and S4). This was not due to increased blood flow during infection, as 1-μm microspheres were retained within the vessels even after Salmonella infection (fig. S1, C and D, and movies S5 and S6). These results suggest the existence of a GVB that prevents the translocation of molecules of around 70 kD and that can be disrupted by Salmonella.

Fig. 1 Characterization of the GVB.

(A) Intestinal blood vessels permeability to 4 kD and 70 kD FITC-dextran was analyzed by intravital two-photon microscopy in untreated or Salmonella-infected mice. Images were acquired every 30 s for 10 min. Arrowheads indicate fluorophore extravasation. Scale bar, 20 μm. Images are representative of three mice per group. (B) Localization of indicated TJ proteins (green) on endothelial cells (CD31, red) in the mouse intestine. Scale bars: (i and iv) 10 μm; (ii and iii) 20 μm. (C) AJ proteins expression on intestinal vessels. Scale bars, 10 μm. (D) Confocal images of C57BL/6J mice intestine stained with CD31 (red) and GFAP (green) (scale bar, 30 μm) or CD31 (gray), α-SMA (green), and desmin (red). Scale bars: (right) 20 μm, (left) 10 μm. In each section, actin filaments or nuclei were stained with phalloidin or 4′,6′-diamidino-2-phenylindole (DAPI) (blue), respectively. For panels (B) to (D), images are representative of six different mice.

Endothelial barriers are characterized by the presence of elaborate junctional complexes that include tight junction (TJ) and adherens junction (AJ), which strictly control paracellular trafficking of solutes and fluids (7, 8). Other cell types, such as pericytes or fibroblasts, can be found associated with the microvasculature and are involved in the maintenance of the vascular barrier, where they form a vascular unit (9). To study GVB characteristics, we analyzed the composition of TJ and AJ in gut ECs. We found that enteric ECs have TJ formed by occludin, zonula occludens-1 ZO-1, cingulin, and junctional adhesion molecule-A JAM-A (Fig. 1B and fig. S2) and AJ formed by vascular endothelial cadherin (VE-cadherin) and β-catenin (Fig. 1C and fig. S3). Claudin-5 was expressed primarily in lymphatic endothelial TJ (fig. S4). Claudin-12 was associated with other cell types in the lamina propria (fig. S2), which probably reflects a function in ion transport rather than in sealing (10). Finally, we found that gut ECs were surrounded by enteric glial cells expressing the intermediate filament glial fibrillary acidic protein (GFAP) and by pericytes expressing desmin (Fig. 1D and fig. S5). Thus, gut ECs are organized in a gut-vascular unit and express TJ and AJ proteins.

We then analyzed the expression of plasmalemma vesicle–associated protein-1 (PV1) (11), a marker of EC permeability (12, 13). We found that gut blood ECs in the lamina propria (but not in the submucosa) did not express PV1 (Fig. 2A). We then hypothesized that PV1 expression could be modulated upon Salmonella challenge, which would reflect the increased vessel permeability. We observed a peak of PV1 up-regulation on blood but not lymphatic vessels in the jejunum and ileum 6 hours after Salmonella infection (Fig. 2, A and B). This correlated with Salmonella dissemination to the liver and spleen (Fig. 2C) and with liver damage (Fig. 2D). In the blood-brain barrier (BBB), which is the most selective vascular barrier, the down-regulation of PV1 during maturation is paralleled by the up-regulation of claudin-3 (12) and claudin-5 (14). However, expression of claudin-3 and -5 were only marginally detected on vascular ECs, and claudin-3 was modified only in epithelial cells upon infection (fig. S6).

Fig. 2 PV1 expression correlates with Salmonella systemic dissemination.

(A) C57BL/6J mice were orally infected or not with S. typhimurium ΔaroA, and PV1 (red) expression was evaluated in small intestinal (SI) blood vessels stained with CD34 (green). Cell nuclei were stained with DAPI (blue). Scale bars, 50 μm. (B) Mean fluorescence intensity of PV1 was measured to quantify the up-regulation of the protein in the intestinal vessels. AU, arbitrary units. Statistical comparisons were based on one-way analysis of variance (ANOVA). *P < 0.001. Each part of the SI was compared with its nontreated counterpart; #,§P < 0.001 ileum (I) was compared with duodenum (D) (#) or with jejunum (J) (§) at the same time point. (C and D) S. typhimurium dissemination in PPs, MLNs, spleen, and liver (C) and ALT in the serum (D) were evaluated at the indicated time points after infection. Two independent experiments are shown, each data point representing an individual mouse (n = 4 to 8). Error bars represent SEM. Student’s unpaired t test was used to determine statistical significance. NT, not treated.

To confirm increased barrier permeability, we injected FITC-dextran in intestinal loops at different time points. We detected the dye in the serum only if mice were infected with Salmonella (fig. S7). The kinetics of FITC-dextran translocation into the bloodstream was different throughout the intestinal tract. Maximal leakage was observed in the duodenum and jejunum loops at 4 hours and in the ileum at 24 hours after Salmonella infection (fig. S7).

To distinguish between direct gut-liver dissemination via the portal circulation and a systemic dissemination through the lymphatics-thoracic duct, we evaluated the presence of 70 kD FITC-dextran in the liver and spleen after intestinal injection. We found that 60 min after injection, the dye was detected in the liver but not the spleen of mice previously infected with Salmonella (fig. S8). We could not detect any dye in either tissue at an earlier time point (20 min), which suggests that the dye reaches the liver before the spleen, using the portal circulation, consistent with a report showing that liver and spleen infections of Salmonella originate from a different pool of bacterial colonies compared with those of Peyer's patches (PP) and MLNs (15).

To assess whether the capacity to cross the epithelial barrier would be sufficient for a bacterium to reach the endothelial barrier and cross it, we constructed a nonpathogenic strain of Escherichia coli with the Yersinia enterocolitica Inv protein (E. coli pInv), which enabled it to cross the epithelial barrier and penetrate the different segments of the gut (fig. S9). However, E. coli pInv was unable to disseminate to the liver and spleen (fig. S9), even if we administered 10 times the dose used for Salmonella, which suggested that the ability of Salmonella to disseminate systemically is not simply due to its capacity to cross the epithelium and face the endothelial barrier but rather results from an active process.

BBB maturation is controlled by the activation of the canonical Wnt/β-catenin signaling pathway (12, 16). We hypothesized that this pathway might be important also for GVB maturation and that Salmonella might interfere with it. We generated primary lung ECs as we were unable to culture primary enteric ECs (17) and confirmed that they can be infected by Salmonella (fig. S10A). We treated these cells with different Salmonella mutants or with Wnt3a to stimulate β-catenin activation (12). We found that Axin2 expression, a β-catenin target gene (18), was reduced upon wild-type (WT) Salmonella infection, even after Wnt3a stimulation (Fig. 3A), which suggested that Salmonella affects the GVB by inhibiting canonical Wnt signaling. Salmonella is equipped with type III secretion systems that are encoded by the Salmonella pathogenicity island 1 and 2 (Spi1 and Spi2) (6). We found that the Salmonella Spi2 mutant (ΔssaV) could not down-regulate Axin2 expression (Fig. 3A), even though the levels of Spi2-encoding genes were higher in ECs than in epithelial cells (fig. S10B). Together, these results show that Salmonella infection interferes with β-catenin activation in ECs via the Spi2.

Fig. 3 Salmonella interferes with activation of the Wnt/β-catenin signaling pathway via a Spi2-mediated mechanism.

(A) Primary lung endothelial cells were infected with WT S. typhimurium, a Spi2 mutant strain (ΔssaV), or a Spi1 mutant strain (ΔinvA). Alternatively, cells were treated with recombinant Wnt3a as a positive control. The expression of Axin2 was assessed by quantitative reverse transcription polymerase chain reaction. Results are pooled from three independent experiments. Error bars represent SEM. Statistical significance between untreated and treated samples was evaluated using one-way ANOVA. Error bars represent SEM. (B) β-Catenin gain-of-function mice (black symbols) or β-cateninlox(ex3)/lox(ex3) control mice (gray symbols) were orally infected with S. typhimurium ΔaroA. Colony-forming units (CFUs) in PPs, MLNs, spleen, and liver were determined at the indicated time points. Two independent experiments are pooled; each data point represents an individual mouse (n = 3 to 9). (C) Bacterial counts in PPs, MLNs, liver, spleen, SI, and colon at the indicated time points after C57BL/6J mice infection with 1010 S. typhimurium ΔssaV. NT, not treated. Each dot represents one mouse (n = 5 to 8). Error bars represent SEM. Statistical significance was evaluated by Student’s unpaired t test. *P < 0.05 **P < 0.01.

We then asked whether Salmonella could still disseminate if we induced β-catenin transcriptional activation only in ECs in mice carrying an inducible degradation-resistant β-catenin. Cre+ (with a degradation-resistant form of β-catenin in ECs) and Cre (control) mice contained equivalent bacterial burdens in the different intestinal tracts, MLNs, and PPs after infection (fig. S11A). By contrast, in Cre+ mice, Salmonella lost the ability to reach the liver or the spleen (Fig. 3B). Moreover, we could not detect any PV1 up-regulation at 2 or 6 hours and only a slight up-regulation at 24hours in Cre+ mice (fig. S12). Gut permeability measured by FITC-dextran leakage was also affected in Cre+ Salmonella-infected mice (fig. S13). We excluded that Cre+ mice had an impaired ability to initiate inflammation by testing the production of several inflammatory mediators including S100a9, Ccl2, Cxcl1, Il1β, and Il6 in PPs after Salmonella infection (fig. S14). We also ensured that liver dissemination of Salmonella upon intravenous infection was similar in Cre+ and Cre mice (fig. S15). Together, these results indicate that Salmonella cannot penetrate the GVB when we force β-catenin activation. It remains to be established whether modulation of β-catenin signaling is the target of Salmonella for its dissemination in vivo.

We next assessed whether the Salmonella Spi2 was responsible for conferring GVB-disrupting properties to Salmonella in vivo. We challenged the mice with ΔssaV Salmonella and analyzed bacterial dissemination. We used 10 times as many ΔssaV bacterial cells to overcome their reduced ability to survive intracellularly. Salmonella ΔssaV entered the gut and reached the PPs but not the liver or spleen (Fig. 3C). This correlated with the inability to induce PV1 up-regulation in any of the intestinal tracts (fig. S11, B and C). Hence, Salmonella is capable of disseminating systemically through, presumably, the down-modulation of β-catenin activation via a Spi2-mediated mechanism that leads to PV1 up-regulation.

Finally, we evaluated whether we could identify a GVB in the human intestine. We found that human intestinal ECs were ensheathed by GFAP+ enteric glial cells and were in close proximity to α-SMA+ and desmin+ pericytes (Fig. 4A). In addition, blood ECs displayed both AJ and TJ proteins (fig. S16), and similarly to the mouse, claudin-5 was expressed primarily in lymphatic vessels (fig. S16). We detected PV1 up-regulation both in the ileum and in the colon upon apical infection with Salmonella enterica serovar Typhi (fig. S17, A and B). These results indicate that the human gut also harbors a GVB and that it can be disrupted by Salmonella infection.

Fig. 4 GVB is present in the human intestine and is disrupted in patients with celiac disease.

(A) Confocal images and three-dimensional reconstruction of human healthy ileum stained with CD31 (green) and GFAP (red). Tissue sections were also stained for pericyte markers α-SMA (antibody against smooth muscle actin) (left) and desmin (right). (B and C) Immunofluorescence of three representative duodenal biopsies from celiac disease patients with high (B) or normal (C) levels of ALT and aspartate transaminase (AST). PV1 (green) stainings are shown. All sections are counterstained with DAPI (blue) to visualize cell nuclei. Scale bars, 50 μm. (D) PV1 expression quantified in 6 samples from celiac disease patients with high ALT and AST, 9 from patients with low ALT and AST, and 10 from healthy individuals. One-way ANOVA and t test were used to determine statistical significance. *P < 0.05.

To assess whether disruption of the GVB in the human gut may lead to liver damage, we analyzed the tissue of celiac disease patients following a gluten-free diet and displaying increased alanine transaminase (ALT) serum levels, in whom no gut histopathology was observed and other possible causes of liver damage were excluded. Celiac disease patients with high ALT displayed a higher expression of PV1 compared with those with normal serum ALT (Fig. 4, B to D). To address whether liver damage can disrupt the GVB, we intravenously injected WT mice with concanavalin A (Con A) to induce liver inflammation. As expected, we found a strong increase of serum transaminases (fig. S18A); however, we could not detect any significant up-regulation of intestinal PV1, which suggested that GVB impairment is not secondary to liver damage (fig. S18, B and C).

Our results demonstrate the existence of a GVB with morphological and functional characteristics similar to those of the BBB, particularly characteristic of pial vessels in the subarachnoid space, which are prone to damage during infection (19) but have some distinct features. Although the BBB has a size exclusion of 500 daltons, in the GVB, molecules as large as 4 kD can easily diffuse. This is probably because the GVB must allow passage of larger molecules for nutrient exploitation and tolerance induction and may be due to differences in EC-TJ composition. Similarly to the BBB, β-catenin signaling in the GVB inhibits vascular permeability and bacterial penetration. Previous work is also consistent with a role of Wnt/β-catenin signaling in controlling the endothelial barrier, as the administration of R-spondin3 during ischemia reperfusion of the gut induced tightening of the endothelium and reduced leakage (20).

Our data from celiac disease patients with elevated serum transaminases, independently of an epithelial barrier defect, indicate that endothelial barrier modifications may be responsible for liver damage and that endothelial and epithelial barriers are independent entities.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S18

Tables S1 and S2

References (2128)

Movies S1 to S6


  1. ACKNOWLEDGMENTS: This work was funded by the European Research Council (M.R., Dendroworld and HomeoGUT; E.D., Wnt for Brain), the Italian Association for Cancer Research (AIRC) and the Italian Ministry of Health (Ricerca finalizzata). I.S. is the recipient of a FIRC fellowship. The authors declare no competing financial interests.
View Abstract

Stay Connected to Science

Navigate This Article