Research ArticlePhysiology

A venous-specific purinergic signaling cascade initiated by Pannexin 1 regulates TNFα-induced increases in endothelial permeability

See allHide authors and affiliations

Science Signaling  02 Mar 2021:
Vol. 14, Issue 672, eaba2940
DOI: 10.1126/scisignal.aba2940

The source of the leak in sepsis

A challenge in treating sepsis, a potentially life-threatening, systemic inflammatory disorder, is the increase in endothelial permeability that leads to widespread tissue edema and immune cell infiltration. Maier-Begandt et al. uncovered a signaling pathway activated specifically in veins by TNFα, a proinflammatory cytokine whose circulating concentrations increase greatly during sepsis. TNFα treatment resulted in the activation of Panx1 channels, which stimulated a signaling pathway that led to disrupted tight junctions in veins but not in arteries. Sepsis induced less lung edema and was less fatal to Panx1-deficient mice compared to control mice. Thus, targeting this pathway may improve survival in patients experiencing sepsis.


The endothelial cell barrier regulates the passage of fluid between the bloodstream and underlying tissues, and barrier function impairment exacerbates the severity of inflammatory insults. To understand how inflammation alters vessel permeability, we studied the effects of the proinflammatory cytokine TNFα on transendothelial permeability and electrophysiology in ex vivo murine veins and arteries. We found that TNFα specifically decreased the barrier function of venous endothelium without affecting that of arterial endothelium. On the basis of RNA expression profiling and protein analysis, we found that claudin-11 (CLDN11) was the predominant claudin in venous endothelial cells and that there was little, if any, CLDN11 in arterial endothelial cells. Consistent with a difference in claudin composition, TNFα increased the permselectivity of Cl over Na+ in venous but not arterial endothelium. The vein-specific effects of TNFα also required the activation of Pannexin 1 (Panx1) channels and the CD39-mediated hydrolysis of ATP to adenosine, which subsequently stimulated A2A adenosine receptors. Moreover, the increase in vein permeability required the activation of the Ca2+ channel TRPV4 downstream of Panx1 activation. Panx1-deficient mice resisted the pathologic effects of sepsis induced by cecal ligation and puncture on life span and lung vascular permeability. These data provide a targetable pathway with the potential to promote vein barrier function and prevent the deleterious effects of vascular leak in response to inflammation.


Under normal physiological conditions, the endothelial barrier of blood vessels maintains homeostasis by regulating the passage of water, ions, and proteins between the blood and underlying tissues. Dysregulation or breakdown of the endothelial barrier is associated with pathophysiological conditions including stroke, hypertension, and sepsis (14). There are several proposed mechanisms by which endothelial barrier function becomes compromised, but there are few mechanistic studies using intact vessels and transgenic mouse models that have been underutilized in this capacity.

The vascular tree has considerable diversity, with discrete regions having different physiologic characteristics and permeability (5). In the microcirculation, it is generally appreciated that, at baseline, veins are considerably more permeable to solutes than arteries and also have lower vessel electrical resistance (6, 7). In addition, it has been known for decades that capillaries and veins are substantially more sensitive than arteries to stimuli that induce vascular leakage, such as histamine (8), platelet-activating factor (9), and tumor necrosis factor–α (TNFα) (1012). However, the molecular mechanisms that confer increased sensitivity of veins over arteries to these agents have not been determined.

TNFα regulates inflammation in part by activating the endothelium, leading to leukocyte recruitment, transmigration and, ultimately, resolution of tissue inflammation at sites of injury or infection (1315). TNFα activates Pannexin 1 (Panx1) channels on endothelial cells during inflammation by an Src family kinase–mediated pathway to release adenosine 5′-triphosphate (ATP), a leukocyte attractant and activator (16, 17). Consistent with a role for Panx1 in inflammation, these channels have been implicated in ischemia/reperfusion injury in the brain (18), kidney (19), and lung (20) microvasculature. Ischemia/reperfusion injury is frequently associated with edema (20), suggesting that Panx1 may have a role in regulating vascular barrier function. However, Panx1 is central to the recruitment of immune cells to sites of injury (16, 17), which, in turn, can damage the endothelium. Because in vivo analysis is confounded by the immune response, it has not been determined whether Panx1 has a direct mechanistic role in regulating endothelial barrier function in response to proinflammatory cytokines.

Here, we used ex vivo preparations of veins and arteries from various transgenic mouse models, as well as cultured primary human endothelial cells, to define roles for Panx1 in TNFα-induced disruption of endothelial barrier function. We found that the barrier function of venous endothelial cells was significantly more sensitive to TNFα than arteries and that the decrease in barrier function was due to stimulated Ca2+ influx through transient receptor potential vanilloid 4 (TRPV4) channels after Panx1 activation. Also, in comparison to arteries, veins were enriched for ectonucleoside triphosphate diphosphohydrolase 1 (CD39), a rate-limiting enzyme in nucleotide hydrolysis (21, 22) in the pathway hydrolyzing ATP to adenosine that, in concert with ecto-5′-nucleotidease (CD73), metabolized extracellular ATP released through Panx1 to adenosine, thus stimulating A2A adenosine receptors. CD39 and A2A were required for venous leak induced by TNFα. Moreover, we found that claudin-11 (CLDN11) was specifically expressed by venous endothelium and not by arterial endothelial cells. These data show that Panx1 has a direct role in regulating endothelial barrier function and identify downstream mediators that increase venous sensitivity to TNFα.


Vein endothelial barrier function was more sensitive to TNFα than artery endothelial barrier function

To investigate the effects of TNFα on vessel permeability, we measured vessel preparations using custom-made cassettes for a dual-channel, self-contained Ussing chamber (Fig. 1, A to D), analogous to an approach we previously used to measure intestinal permeability in situ (23). The cassettes were equipped with a 175 × 1500 μm opening that is small enough to be covered with a longitudinally sectioned vessel (Fig. 1, C and D). We isolated second- or third-order mouse mesenteric veins and arteries, cut them longitudinally, and then treated the vessels with either vehicle control or TNFα (50 ng/ml) for 30 min. Under these conditions, TNFα did not damage endothelial cells (fig. S1B). Afterward, the vessel was mounted by fully covering the opening of the custom-made chamber cassette (Fig. 1D), which was sealed and placed into the Ussing chamber, with the endothelial cell side oriented as the apical face.

Fig. 1 Custom Ussing system to measure electrophysiological characteristics of ex vivo mesenteric vessels.

(A) Schematic of a self-contained Ussing chamber system that can house two custom-made tissue cassettes. (B) Custom-made cassette for vessel analysis. Arrow indicates the horizontal opening of the cassette. Scale bar, 1 cm. (C and D) Magnified image of the cassette showing the opening with an approximate area of 0.26 mm2 (red line) (C), which can be completely covered with a longitudinally sectioned vessel (border of the vein marked in blue) (D). Scale bars, 750 μm (C) and 650 μm (D). Numerical marks reflect eyepiece magnification. (E to G) Ex vivo second- and third-order mesenteric arteries or veins of C57BL/6 mice were treated with vehicle (0.1% DMSO in PBS) or TNFα (50 ng/ml) for 30 min and mounted on Ussing chamber cassettes in physiologic Ringer’s solution. TEER (E), short-circuit current (Isc) (F), and ion permeability (G) were measured. (H) Transendothelial dilution potential measurements in veins and arteries treated with TNFα in normal Ringer’s solution or Ringer’s where NaCl was exchanged with either 140 mM Na-Asp or Lys-HCl. ***P = 0.0005 (E); #P = 0.076 (F) by unpaired two-tailed t test. *P = 0.034 by one-way ANOVA Fisher’s least significant difference (LSD) test (G). #P = 0.064 and **P = 0.0093 for both comparisons by two-way ANOVA Fisher’s LSD test (H). n = 3 to 6 venules and n = 3 to 4 arterioles for each treatment representing biological replicates. Graphs represent means ± SD. NS, not significant.

In the presence of physiological Ringer’s solution, transendothelial electrical resistance (TEER) was measured from either control or TNFα-treated mouse veins and arteries. Control veins showed a ~2.4-fold decrease in TEER 30 min after TNFα treatment (Fig. 1E). By contrast, arteries did not show a significant change in TEER between vehicle control and TNFα treatment. Consistent with the TEER measurements, short circuit currents (Isc) were specifically decreased in TNFα-treated veins as opposed to arteries, where Isc was essentially unchanged by exposure to TNFα (Fig. 1F). Thus, vein barrier function was specifically altered by TNFα.

Ussing chambers allow detailed electrophysiological characterization of samples. To determine whether TNFα affected ion permeability in native blood vessels, we performed dilution potential measurements to calculate the relative permeability of Na+ to Cl (24, 25). Untreated veins and arteries both showed comparable Na+ and Cl permeability, where PNa/PCl was nearly 1 (Fig. 1G). However, TNFα treatment specifically increased venous Cl permeability, as reflected by a decrease in PNa/PCl. To further confirm changes in vessel ion permselectivity due to TNFα, we also examined the effect of TNFα on vessel transendothelial potential when the current was clamped to 0 mA. TNFα induced a change in vein polarity from 1.3 mV under control conditions to −1.1 mV in normal Ringer’s buffer (Fig. 1H). Vein potential showed a comparable TNFα-dependent change in polarity when aspartate was substituted for Cl or lysine was substituted for Na+, suggesting that the effect of TNFα on transendothelial potential involved changes in both Na+ and Cl permeability. Arteries did not show this effect, regardless of buffer composition. Thus, ex vivo Ussing analysis revealed a venous-specific effect of TNFα on paracellular ion channel homeostasis.

To determine whether the vein-specific effects of TNFα were due to endothelial cell function, we examined primary human venous cells [human umbilical vein endothelial cells (HUVECs) and human saphenous vein endothelial cells (HSaVECs)] and arterial cells [human aortic endothelial cells (HAoECs) and human coronary arterial endothelial cells (HCoECs)] cultured on Transwell inserts and measured the effect of TNFα on TEER with an EVOM voltmeter. Consistent with our observations from intact vessels in the Ussing chamber (Fig. 1, E to H), TNFα treatment induced a decrease in the TEER of HUVECs and HSaVECs that was not observed for either HAoECs or HCoECs (Fig. 2, A and B). These results suggest that venous endothelial cells may be more sensitive to the effects of the proinflammatory cytokine TNFα on TEER than arterial endothelial cells.

Fig. 2 Effects of TNFα on barrier function and differential claudin expression by venous endothelium.

(A and B) TEER in ohm×square centimeter (A) or as percentage change (B) to control of untreated or TNFα-treated (10 and 100 ng/ml) human venous endothelial cells (HSaVEC and HUVEC) and human arterial endothelial cells (HAoEC and HCoEC) cultured on Transwell permeable supports (n = 3 different human donors). (C and D) Quantification of CLDN1, CLDN3, CLDN5, CLDN12 (C), and CLDN11 (D) gene expression by real-time PCR in human vein endothelial cells (HSaVEC and HUVEC) and human arterial endothelial cells (HAoEC and HCoEC) (n = 2 to 3 independent RNA preparations). (E) Scatterplot showing microarray transcription profiling analysis comparing venous endothelial cells (EC; HSaVEC and HUVEC) with arterial endothelial cells (EC; HAoEC and HCoEC). The tight junction protein gene CLDN11 and ENTPD1 (which encodes CD39) are highlighted. (F) Representative immunoblot (F) of CLDN11 expression in cell lysates from HSaVEC, HUVEC, HAoEC, and HCoEC. Protein expression of CLDN11 is normalized to the amount of GAPDH (glyceraldehyde-3-phosphate dehydrogenase) protein expression (100%) (n = 2 to 3 independent protein preparations). (G) Representative microscopic images of second-order mesenteric veins and arteries from C57BL/6 mice presented en face immunostained for Cldn5, Cldn11, or IgG control for labeling specificity (red). Nuclei were stained with DAPI (blue). Scale bars, 100 μm. **P = 0.0015, ***P = 0.0009, and ****P < 0.0001 (A); **P = 0.0012, 0.0017, and 0.0011 and ***P = 0.0008 (B); **P = 0.0069, 0.0028, and 0.0054 and #P = 0.067 (C); **P = 0.0020 and 0.0023 by two-way ANOVA Fisher’s LSD test (F). *P = 0.034 and 0.037 by one-way ANOVA Fisher’s LSD test (D). Graphs represent means ± SD.

Given the differential sensitivity of venous versus arterial endothelial cells to TNFα, we used quantitative real-time polymerase chain reaction (qRT-PCR) to screen mRNA isolated from HUVECs, HSaVECs, HAoECs, and HCoECs for differences in claudin (CLDN) gene expression, focusing on claudin-encoding genes expressed by vascular endothelium, namely CLDN1, CLDN3, CLDN5, CLDN11, and CLDN12 (2628). Of these claudins, CLDN11 showed the greatest difference in expression when comparing vein to artery mRNA, suggesting that CLDN11 could be considered a marker for venous endothelium (Fig. 2, C and D). Because CLDN5 is considered to be the most prominent claudin expressed by endothelial cells, it was unexpected that CLDN11 was expressed at a ~4-fold higher level compared to CLDN5 by venous endothelial cells (Fig. 2, C and D).

Moreover, gene expression profiling comparing mRNA isolated from HUVECs, HSaVECs, HAoECs, and HCoECs demonstrated that CLDN11 was the most highly enriched mRNA when comparing venous endothelial cells with arterial endothelial cells (Fig. 2E, figs. S2 and S3, and data file S1). We further analyzed Cldn11 expression at the protein level, in primary cultures of human endothelial cells (Fig. 2F) and in murine en face (Fig. 2G) and transverse (fig. S4A) preparations of third-order mesenteric veins and arteries. Although Cldn5 was consistently expressed by both venous and arterial endothelium, Cldn11 was preferentially expressed by venous endothelium. Together, these data suggest that CLDN11 mRNA and Cldn11 protein expression can be used as a marker to distinguish venous endothelium from arterial endothelium.

TNFα-induced changes in venous endothelial barrier function were regulated by TRPV4

TRPV4 channels regulate endothelial barrier function and can be activated by multiple stimuli (2933). TRPV4 channels are also activated by TNFα in odontoblasts (34). Thus, we tested whether endothelial TRPV4 channels were activated by TNFα and, given the vein-specific increase in permeability in response to TNFα, whether TRPV4 channel activation was enhanced in veins as compared with arteries.

Initially, we performed Ca2+ imaging experiments by quantifying local Ca2+ influx, “Ca2+ sparklets,” arising through distinct regions of the plasma membrane containing TRPV4 channels. Third-order mesenteric veins and arteries from C57BL/6 mice cut transversely in en face preparation were incubated with the Ca2+-binding dye Fluo-4 AM and imaged using confocal microscopy (Fig. 3A). The TRPV4 agonist GSK1016790A was added to identify healthy tissue and regions of interest with activatable TRPV4 channels. We found that the addition of TNFα (10 ng/ml) produced an increase in Ca2+ sparklets over baseline in both veins and arteries, where the TNFα-induced activation of TRPV4 channels induced significantly higher sparklet activity in veins than in arteries (Fig. 3, B and C). Moreover, addition of the TRPV4 inhibitor GSK2193784 inhibited sparklets induced by TNFα, confirming that the Ca2+ influx was due to TRPV4 channel activity.

Fig. 3 TNFα-stimulated venous barrier permeability required TRPV4.

(A and B) Representative grayscale image of Fluo-4–loaded, en face mesenteric vein of C57BL/6 mice (A). Scale bar, 10 μm. Square boxes (a, b, and c) indicate regions of interest placed at Ca2+ sparklet sites to record fractional fluorescence (F/F0) traces (B). Experiments were performed in the presence of cyclopiazonic acid (CPA; 20 μM) to eliminate Ca2+ release from the endoplasmic reticulum and also in the presence of the TRPV4 channel agonist GSK1016790A (GSK101; 6 nM). (C) Ca2+ sparklet activity was recorded before and after stimulation with TNFα (10 ng/ml). TNFα-induced sparklet activity in the absence or presence of the TRPV4 inhibitor GSK2193784 (GSK219) was quantified (n = 5 vessels for each treatment). (D) Image of a vein cannulated and sutured onto glass micropipettes for ex vivo analysis. Scale bar, 100 μm. (E) Schematic of the perfusion system, to measure vessel permeability ex vivo, adapted from (16). Vessels were perfused through the lumen with fluorescein (10 μg/ml) in Krebs/0.1% BSA containing TNFα (50 ng/ml), and permeability was measured. (F) Permeability induced by TNFα in venous (n = 5) and arterial (n = 4) mesenteric vessels from C57BL/6 mice. (G) Permeability induced by TNFα (10 ng/ml) in veins from Trpv4+/+ mice (control) or from Trpv4−/− mice (n = 5 to 8 veins for each genotype) or veins from C57BL/6 mice treated with the TRPV4 agonist GSK1016790A (GSK101; 100 nM) (n = 4 veins). **P = 0.0040 and ****P < 0.0001 (C); **P = 0.0015 and ***P = 0.0005 and 0.0009 (F); *P = 0.027 and **P = 0.0097 (G) by one-way ANOVA with Tukey’s multiple comparisons test. Graphs represent means ± SD.

To link TRPV4 channel activity to the effects of TNFα on endothelial permeability ex vivo, we adapted a previously described approach to measure the permeability of the fluorescent tracer molecule fluorescein in dissected, cannulated murine mesenteric veins and arteries (Fig. 3, D and E) (35). TNFα significantly increased fluorescein permeability in veins as opposed to arteries, which were significantly less sensitive to TNFα (Fig. 3F and fig. S1A). This result underscores that, by two different measures in intact vessels, TEER (Fig. 1E) and transvessel flux of a fluorescent tracer (Fig. 3F and fig. S1A), veins became leakier and were significantly more responsive to TNFα than arteries.

We then used this perfusion system to measure TNFα-induced permeability in mesenteric veins isolated from Trpv4+/+ and Trpv4−/− mice. Unlike Trpv4+/+ wild-type (WT) mice, where addition of TNFα led to a significant increase in permeability, veins isolated from Trpv4−/− mice were resistant to TNFα and showed no increase in permeability (Fig. 3G). Moreover, veins from WT C57BL/6 mice perfused with the TRPV4 agonist GSK1016790A also showed a significant increase in permeability. Together, these results confirm a role for TRPV4 channels downstream from TNFα in regulating venous endothelial barrier function.

TNFα-induced changes in venous endothelial barrier function were regulated by pannexin channels

We have previously shown that Panx1 channels are specifically activated by TNFα in the venous endothelium, leading to the extracellular release of ATP (16), supporting the hypothesis that TRPV4 activation was downstream of Panx1 in endothelial cells. To test this hypothesis, we used pharmacological inhibition and genetic ablation of Panx1. We previously established that spironolactone can act as an acute inhibitor of Panx1 in addition to being a mineralocorticoid receptor antagonist (36). Consistent with a role for Panx1 in activating TRPV4 in venous endothelial cells, addition of spironolactone to veins isolated from C57BL/6 mice significantly decreased the ability of TNFα to enhance Ca2+ sparklet activity (Fig. 4A). To place Panx1 upstream in a TRPV4-dependent vein permeability pathway, we examined the effect of spironolactone on vein barrier function. Pharmacologic blockade of Panx1 protected veins from TNFα-induced barrier weakening (Fig. 4B). Critically, TNFα-induced permeability to fluorescein in veins isolated from human donors was also inhibited by spironolactone treatment (Fig. 4C). In addition, veins from an inducible, endothelial-specific Panx1 knockout mouse developed by our group (Cdh5-CreERT2+/Panx1fl/fl) (16) showed fewer TRPV4 Ca2+ sparklets (Fig. 4D) and also were resistant to TNFα-induced increases in paracellular fluorescein permeability (Fig. 4E). Spironolactone and genetic ablation provided two complementary methods demonstrating that Panx1 is upstream of TNFα increases in permeability and Ca2+ sparklet activity.

Fig. 4 TNFα-stimulated venous barrier permeability required Panx1.

(A) Effect of spironolactone (Spiro; 80 μM) on TRPV4 sparklets induced by TNFα (10 ng/ml) in en face mesenteric veins of C57BL/6 mice (n = 4 veins for each treatment). (B and C) Effect of spironolactone (Spiro; 80 μM) on TNFα-induced permeability of veins from C57BL/6 mice (B) or from human adipose tissue biopsy samples (C) (n = 4 veins for each treatment). (D) Effect of the TRPV4 inhibitor GSK2193784 (GSK219) on TRPV4 sparklets induced by TNFα (10 ng/ml) in second-order mesenteric veins from Cdh5-CreERT2+/Panx1fl/fl mice injected with tamoxifen (+TMX) to induce Panx1 deficiency or veins from control mice (−TMX) (n = 4 veins for each treatment). (E) TNFα-induced permeability of mesenteric veins from Cdh5-CreERT2+/Panx1fl/fl mice injected with tamoxifen (+TMX) or veins isolated from mice treated with vehicle control alone (−TMX; n = 4 veins for each treatment). (F and G) Representative microscopic images of TNFα-treated (10 ng/ml), en face second-order mesenteric veins either pretreated with spironolactone (15 min; Spiro; 10 μM) from C57BL/6 mice or from Cdh5-CreERT2+/Panx1fl/fl mice injected with tamoxifen. Cldn11 was immunostained with a specific antibody (red), and nuclei were stained with DAPI (blue). The amount of Cldn11 localized to endothelial tight junctions was measured by the presence of junctional discontinuities (G). Scale bar, 100 μm. n = 3 to 4 fields imaged of vessel preparations from three mice for each treatment. (H and I) Representative immunoblot (H) and quantitative analysis (I) of whole-cell lysates of either spironolactone-pretreated and TNFα-treated (10 ng/ml) second-order mesenteric veins from C57BL/6 mice or of second-order mesenteric veins from Cdh5-CreERT2+/Panx1fl/fl mice injected with tamoxifen or vehicle control and treated with TNFα (10 ng/ml). Total protein expression was used as loading control. n = 3 protein isolates from biological replicates. ***P = 0.0004 (A); **P = 0.0071 (B); **P = 0.0026 (C); **P = 0.0012 (E) by unpaired two-tailed t test. ***P = 0.001 and **P = 0.0097 (D); *P = 0.034 and 0.017 (G); **P = 0.0054 and #P = 0.12 (versus +Spiro) (I) by one-way ANOVA with Fisher’s LSD test. Graphs represent means ± SD.

To further link TNFα and Panx1 to changes in barrier function, we examined the effects of TNFα on Cldn11. Veins treated with TNFα (50 ng/ml) showed disrupted localization of Cldn11 to endothelial tight junctions, as reflected by an increase in junctional discontinuities, an effect that was reversed by spironolactone (Fig. 4, F to I). Moreover, TNFα treatment of veins from Cdh5-CreERT2+/Panx1fl/fl mice had little effect on Cldn11 as opposed to WT veins (Fig. 4, F to I, and fig. S4B). Together, these data demonstrate that TNFα-activated Panx1 channels initiate a TRPV4 signaling response that leads to an increase in venous permeability that is partially due to a decrease in tight junction–associated Cldn11.

Purinergic signaling mediated TNFα-induced impairment of venous endothelial barrier function

In response to TNFα, Panx1 releases ATP that plays a critical role as a signaling molecule in venous endothelial cells (16). We therefore measured ATP in perfusate released by veins from WT mice in response to TNFα (Fig. 5A). Although there was detectable ATP in vein perfusates, 79.7 ± 3.6% of the purinergic species detected were ATP metabolites, specifically adenosine 5′-diphosphate (ADP), adenosine 5′-monophosphate (AMP), and adenosine. This finding suggested that ATP released by Panx1 was rapidly hydrolyzed by endothelial ectonucleotidases. Thus, we performed an immunohistological screen of cross sections of mesenteric veins and arteries from C57BL/6 mice for two prominent ectonucleotideases implicated in inflammation, CD39 and CD73 (Fig. 5B) (37). Cultured primary endothelial cells were also screened by qRT-PCR (Fig. 5C). We found that, although CD73 was expressed in both arteries and veins, as well as primary human cultures of arterial and venous endothelial cells, veins and venous endothelial cells were enriched for CD39 expression as compared with arteries and arterial endothelial cells (Fig. 5, B and C). Microarray analysis of human endothelial cells also demonstrated a significant enrichment of ENTPD1 mRNA (which encodes CD39) by venous endothelial cells as compared with arterial endothelial cells (Fig. 2E and fig. S2).

Fig. 5 Venous endothelial cells expressed higher amounts of the ecto-nucleoside CD39 than arterial endothelial cells.

(A) Veins were isolated from CD39+/+ and CD39−/− mice, incubated with TNFα (10 ng/ml), and adenosine metabolites were measured (n = 3 independent vessel perfusates per group). (B) Representative images from immunohistochemical staining of CD39 and CD73 using specific antibodies to label mesenteric artery (a) and vein (v) pairs in sections from C57BL/6 mice. Labeling specificity was validated by IgG and secondary-only controls. CD73 and CD39 are labeled with secondary antibodies (red), and nuclei are stained with DAPI (blue). Scale bar, 100 μm. (C) Expression analysis of ecto-nucleoside triphosphate diphosphohydrolase (ENTPD1; CD39) (top) and ecto-5′-nucleotidase (NT5E; CD73) (bottom) by qRT-PCR in human vein endothelial cells (HSaVEC and HUVEC) and human arterial endothelial cells (HAoEC and HCoEC) compared with the housekeeping gene B2M (n = 2 to 3 different human donors). (D and E) TNFα-induced permeability was measured in veins from C57BL/6 mice in the presence of the CD39 antagonist ARL67156 (300 μM; n = 4 veins) (D) or in veins from CD39−/− mice (n = 3 veins) (E). *P = 0.018 and ****P < 0.0001 by two-way ANOVA with Sidak’s multiple comparisons test (A); *P = 0.022 and 0.040 and #P = 0.11 and 0.059 by one-way ANOVA with Fisher’s LSD test (C). ***P = 0.0001 (D) and *P = 0.32 (E) by unpaired two-tailed t test (D and E). Graphs are means ± SD.

Confirming a key role for CD39 in venous-specific ATP metabolism, TNFα-stimulated veins from CD39−/− mice showed ~3.7-fold more ATP present in the perfusate than WT controls (Fig. 5A). Moreover, pretreatment of WT veins with the ecto-ATPase CD39 inhibitor ARL67156 or genetic ablation of CD39 inhibited the ability of TNFα to increase venous leak (Fig. 5, D and E), further implicating CD39 as a downstream component required for the effects of Panx1 on venous barrier function.

After ATP has been enzymatically processed by CD39 and CD73, the free adenosine produced can bind and stimulate adenosine receptors at the apical cellular surface (38). We performed qRT-PCR expression analysis to profile adenosine receptor expression and found that the predominant subtypes present in both primary human venous and arterial endothelial cells were A2A (ADORA2A) and A2B (ADORA2B) (Fig. 6A). By contrast, the expression of genes encoding the A1 and A3 subtypes was undetectable. Differences in adenosine receptor transcript levels between venous and arterial cells were modest (Fig. 6A). However, in cross sections of mesenteric veins and arteries from WT C57BL/6 mice analyzed by immunohistochemistry, there was a more prominent A2A signal present in the apical surface of venous endothelial cells as compared with arterial endothelium (Fig. 6B, arrow), whereas A2B was comparably distributed in both veins and arteries.

Fig. 6 A2A receptors were required for TNFα-stimulated increases in venous endothelial permeability.

(A) Expression analysis of adenosine receptors A2A (ADORA2A) and A2B (ADORA2B) by real-time PCR in human venous endothelial cells (HSaVEC and HUVEC) and human arterial endothelial cells (HAoEC and HCoEC) as compared to housekeeping gene B2M (n = 2 to 3 different human donors). (B) Representative images of immunohistochemical labeling of A2A and A2B in mesenteric sections from C57BL/6 mice containing arteries or veins. Asterisks indicate vessel lumens. A2A and A2B were labeled with a secondary antibody (red), and vessels were marked by autofluorescence of the internal elastic lamina (IEL) (green). Arrows indicate the location of the endothelial layer. Representative examples of n = 3 independent experiments. Scale bar, 50 μm. (C) TNFα-induced permeability was measured in veins of C57BL/6 mice pretreated with either the A2A antagonist SCH58261 (100 nM; n = 4 independent vessel preparations), A2A/A2B antagonist ZM241385 (100 nM; n = 5 independent vessel preparations), or treated with the A2A agonist CGS21680 (100 nM) in the absence of TNFα (n = 4 independent vessel preparations. (D) TNFα-induced permeability was measured in veins from A2A+/+ (control) or A2A−/− mice (n = 4 independent vessel preparations). *P = 0.042 and 0.013 (A); ***P = 0.0003, *P = 0.048 and 0.022, and ****P < 0.0001 (C) one-way ANOVA with Fisher’s LSD test. **P = 0.0015 by unpaired two-tailed t test (D). Graphs are means ± SD.

Because there was differential CD39 and A2A expression between arteries and veins, we assessed whether adenosine receptors present in veins contributed to TNFα-induced permeability. WT mesenteric veins pretreated with either the A2A antagonist SCH58261 or the A2A/A2B antagonist ZM241385 eliminated the TNFα-induced increase in permeability (Fig. 6C). Moreover, veins from A2A-deficient mice also were not sensitive to the effects of TNFα on permeability (Fig. 6D). By contrast, direct stimulation of WT veins using the A2A agonist CGS21680 caused an increase in vein permeability that was comparable to that of veins stimulated by TNFα (Fig. 6C), further underscoring a role for A2A in regulating vein permeability.

Panx1 as a target to reduce severity of sepsis

Sepsis is a life-threatening condition due to an uncontrolled infection leading to increased TNFα and venous permeability (39). To identify roles for Panx1 in the response to sepsis, we compared the response of Cdh5-CreERT2+/Panx1fl/fl to control mice that were injured using the cecal ligation and puncture (CLP) injury model. In contrast to control mice that showed little survival 2 days after CLP injury, mice with an endothelial cell–specific depletion of Panx1 were largely viable 5 days after CLP (Fig. 7A). Comparable results were obtained using WT C57BL/6 mice treated with spironolactone (40 mg/kg), which showed a trend toward improved viability (Fig. 7B). Using the lung wet/dry weight ratio as a measure of pulmonary edema, we found that Cdh5-CreERT2+/Panx1fl/fl mice and mice treated with spironolactone had significantly lower edema in response to CLP than controls, consistent with a preservation of endothelial barrier function in Panx1-deficient mice and Panx1-inhibited mice (Fig. 7, C and D). There also was less renal injury in septic Panx1-deficient mice or in mice treated with spironolactone, as assessed by blood urea nitrogen (fig. S5, A and B); however, there was little effect on heart rate (fig. S5, C and D).

Fig. 7 Panx1 inhibition improved survival from systemic sepsis.

(A and B) Survival after cecal ligation and puncture (CLP) protocol to induce systemic sepsis was assessed in Cdh5-CreERT2+/Panx1fl/fl mice injected with tamoxifen (+TMX) to induce Panx1 deletion, noninduced Cdh5-CreERT2+/Panx1fl/fl mice, −TMX-treated Cre mice expressing Panx1 (A), or WT C57BL/6 mice treated or not with spironolactone (+Spiro) (B). n = 8 mice per group. (C and D) Twenty-four hours after CLP injury, lungs were isolated, and the wet/dry weight ratio was determined as a measure of pulmonary edema in the indicated groups. n = 5 mice per group. Kaplan-Meier curves were assessed by log-rank Mantel-Cox test. (A). *P = 0.038 and 0.011 (C) and *P = 0.010 and ****P < 0.0001 (D) by one-way ANOVA with Fisher’s LSD test. Graphs are means ± SD.

Together, these data support a model in which the enhanced sensitivity of veins to TNFα-induced vessel leak is primarily due to differential expression of the ectonucleoside triphosphate diphosphohydrolase CD39 and A2A adenosine receptors. This differential expression provides a pathway in which ATP released by Panx1 channels is hydrolyzed to adenosine that acts as either an autocrine or paracrine stimulus to decrease barrier function in a vein-specific manner (Fig. 8).

Fig. 8 Model depicting the pathway stimulated by TNFα to increase venous endothelial permeability.

TNFα-R1 induces opening of Panx1 channels at the plasma membrane (through c-SRC) that release ATP from the cytosol to the extracellular environment. ATP is hydrolyzed by CD39 and CD73 to adenosine, which then stimulates A2A/A2B receptors. TRPV4 channel is activated, resulting in an increase in paracellular leak through venous tight junction remodeling.


There are several lines of evidence linking Panx1 channels to inflammation (1620). Among the pathologic consequences of inflammation are vascular leak and tissue edema, suggesting a role for Panx1 in regulating vascular barrier function. However, because Panx1 channels are broadly expressed by cells such as leukocytes and interstitial cells, in vivo studies are unable to assign a direct role for Panx1 in regulating endothelial barrier function. Here, we used perfused, isolated vessel preparations and cultured endothelial cells to demonstrate that Panx1 activation by TNFα initiates a cascade that preferentially increases the paracellular permeability of venous endothelial cells. Critically, dissected vessels cannulated onto glass micropipettes allow the investigation of venous permeability in living intact vessels under flow that maintain the complex structure of vessels, including endothelial cells, pericytes, the basal membrane that separates the two cell types, and the glycocalyx that lines the apical surface of endothelial cells (40). At the same time, perfused vessels enabled us to define the Panx1-dependent pathways that promote vessel leakage by endothelial cells in a native context while avoiding the confounding effects of Panx1 channels expressed by other cells activated by inflammation.

The barrier function of capillaries and postcapillary veins is substantially more sensitive to TNFα than arteries in vivo (1012). However, the preferential effect of TNFα on venous endothelium is not absolute. For instance, TNFα increases the permeability of bovine pulmonary arterial endothelial cells, although this requires at least 4 hours of treatment with TNFα (20 to 100 ng/ml) (41, 42). Bovine aortic endothelial cells also show increased leak after a 45-min TNFα treatment period (43). We therefore compared mesenteric venular and arteriolar leak after a 30-min treatment period to specifically examine differences in early responses to TNFα between these two classes of vessels. Using this protocol, we revealed that venous blood vessels were more sensitive to the proinflammatory stimulus TNFα than arterial vessels.

On the basis of transcriptional profiling of cultured human endothelial cells, we found that CLDN11 is the most enriched gene by venous endothelial cells as compared with arterial endothelial cells. Cldn11 protein expression paralleled the pattern of mRNA expression, such as human venous endothelial cells and murine veins that were enriched for Cldn11 as compared with comparable arterial endothelium. Although there are previous reports that Cldn11 is highly expressed by the microvasculature (2628) and corpus cavernosum (44), we found that Cldn11 was specifically enriched in venous endothelial cells, whereas arterial endothelial cells are Cldn11 deficient. Considering that there are few markers to distinguish venous from arterial endothelial cells, these data suggest that Cldn11 expression is likely to be a useful marker to identify venous endothelial cells.

Enhanced Cldn11 expression by venous endothelial cells also has functional ramifications. At baseline, Cldn11-expressing venous endothelial cells have a higher dilution potential (Fig. 1H) than arterial endothelial cells that lack Cldn11, suggesting that Cldn11 fine-tunes resting endothelial cell paracellular ion permeability (45). Moreover, when veins were treated with TNFα (Fig. 1, E and G), there was both a decrease in net barrier function (as measured by TEER) and a change in tight junction permselectivity such that anion permeability was increased to a greater extent than cation permeability. These TNFα-induced changes in permselectivity further underscore the effect of altered tight junction composition on venous barrier function due to decreased Cldn11 expression and assembly into tight junctions (Fig. 4, E to G).

Venous endothelial cells were more sensitive to the effects of TNFα-stimulated Panx1 channels and showed a decrease in barrier function as compared with arterial endothelium. Because Panx1 is ubiquitously expressed by most endothelial cells, our findings indicated that venous endothelial cells have unique elements downstream from Panx1 that are not shared with arterial endothelial cells. Specifically, we found that veins expressed more of the ectonucleoside CD39, which is the rate-limiting enzyme required in conjunction with the ecto-5′-nucleotidase CD73 to fully hydrolyze ATP to adenosine (37). Also, the CD39 inhibitor ARL67156 or CD39 deficiency protected veins from TNFα-induced leak, further verifying a role for endothelial CD39 in this effect.

In addition to constitutive expression by venous endothelium, CD39 expression also increases in response to injurious stimuli such as endotoxin and hypoxia (46, 47). Thus, CD39 could potentially play a role in regulating permeability in vessels where it is not constitutively expressed at baseline, such as in arteries. Whether this is the case remains to be determined. Nonetheless, control of venous permeability by localized, constitutive expression of CD39 by veins is consistent with their physiological role as sites for leukocyte adhesion and extravasation.

Inflammation and injury are more severe when the CD39/CD73 complex is impaired, and CD73–up-regulating agents have been tested as a therapeutic target for patients with acute respiratory distress syndrome (4851). A protective effect of CD39 (and CD73) is in apparent contrast with our finding that CD39 is required for TNFα to induce venous barrier leak. However, any increase in severity of inflammation in mice with a global CD39 deficiency is likely to accompany increased leukocyte adhesion to sites of vessel inflammation because adenosine inhibits recruitment in response to proinflammatory stimuli (52, 53). Whether this is due to endothelial CD39 is not known because platelets, neutrophils, and other leukocytes also express CD39 (54). By contrast, ATP is a potent stimulus to recruit immune cells in response to inflammation (16). Because veins do not leak in response to ATP, this raises the possibility that ATP release may be an early event to attract immune cells to sites of damage that remain relatively intact until they are subsequently rendered leaky by ATP hydrolysis to adenosine, which also releases immune cells from strong attachment to the endothelium (55, 56).

A2A receptors were required for veins to be sensitive to TNFα because vessels from A2A-deficient mice remained intact and the effects of TNFα were antagonized by the A2 receptor inhibitors SCH58261 and ZM241385. Consistent with a role for A2A in vessel permeability, the A2A agonist Lexiscan increases the permeability of human brain microvascular endothelial cells and also permeabilizes the blood-brain barrier in mice, an effect that is diminished in A2A−/− mice (57, 58). Human lung microvascular endothelial cells, which lack A2A receptors, retain barrier function in response to adenosine, thus underscoring a role for A2A receptors in endothelial leak (59).

Given the established role for TRPV4 channels in increasing endothelial cell barrier permeability (29, 30, 60), we examined the effects of TNFα on TRPV4 channels in venous endothelium. Veins showed greater TRPV4 channel activation as compared with arteries in response to TNFα, consistent with our observations that veins were rendered more permeable by TNFα than arteries. Moreover, veins from TRPV4-deficient mice resisted the effects of TNFα on endothelial barrier function.

Sepsis is a life-threatening condition that currently lacks targeted therapeutic treatment and instead relies on antibiotics, fluid resuscitation, and supportive care (61). More than 250,000 patients die of sepsis each year, largely as a result of septic shock (62). Here, we found in transgenic mouse models that endothelial Panx1 played a key role in the severity of sepsis. Critically, spironolactone had a protective effect in CLP injured mice, suggesting that Panx1 is a potential therapeutic target in the treatment of sepsis. The venous-specific signaling pathway that we have defined downstream from Panx1 provides additional targets that may provide improved therapeutic options in the management of sepsis.

Our study provides evidence that TRPV4 activation was downstream from Panx1 because TNFα activation of Ca2+ sparklets was diminished in vessels treated with spironolactone. Also, vessels deficient for endothelial Panx1 had a level of Ca2+ sparklet activity that was comparable to WT vessels treated with the TRPV4 antagonist GSK219. TRPV4 has the capacity to activate Panx1 channels (6365). In these instances, it is likely that Panx1 is specifically activated by localized TRPV4 sparklets because a sustained increase in cytosolic Ca2+ has no effect on Panx1 (66, 67). One intriguing possibility is that Panx1 and TRPV4 channels in a complex may form a feed-forward cycle that amplifies the effects of proinflammatory cytokines on venous endothelial cells, a notion supported by the ability of TRPV4 activation to induce TNFα release (68). If this is the case, then agents targeting the reciprocal activation of Panx1 and TRPV4 channels would have utility in preventing the deleterious effects of vascular leak that accompany inflammation.



All mice were on a C57BL/6J genetic background, male, 10 to 14 weeks of age, and were cared for under the provisions of the University of Virginia and Emory Animal Care and Use Committees and followed the National Institutes of Health (NIH) guidelines for the care and use of laboratory animals. Mice were anaesthetized with carbon dioxide or isoflurane (2.5%) and sacrificed by cervical dislocation. The inducible, endothelial cell–specific Panx1 knockout mice using VE-cadherin (Cdh5)–specific cre recombinase expression (Cdh5-CreERT2+/Panx1fl/fl) were generated as previously described (16). Panx1 deletion was induced specifically in the vascular endothelium by intraperitoneal injection of tamoxifen (TMX; 1 mg in 0.1 ml of peanut oil vehicle) for 10 consecutive days into Cdh5-CreERT2+/Panx1fl/fl mice; vehicle-injected mice served as controls. CD39−/− mice (69), adenosine receptor subtype A2A−/− mice (70, 71), and TRPV4−/− mice (31, 72) were generated as previously described.

Human vascular biopsies

Vessels from human adipose tissue biopsy samples were aseptically obtained from the gluteal region as supervised by E. Barrett at the University of Virginia outpatient clinic (Institutional Review Board approval 20408), following the Declaration of Helsinki Principles. For research purposes, biopsies were from nonobese adult patients ranging in age from 40 to 60 years old and that were not being treated with adrenergic blocking agents (66).

Electrophysiological vessel measurements

Second- and third-order veins and arteries were isolated from mouse mesenteric circulation in a Ringer’s solution [140 mM NaCl, 2 mM CaCl2, 1 mM MgCl2, 10 mM glucose, and 10 mM Hepes (pH 7.3)]. Veins or arteries 2 to 3 mm in length were dissected, and the adventitia was removed and cut longitudinally. The vessels were treated for 30 min at 37°C with TNFα (50 ng/ml) (R&D Systems, recombinant murine or human TNFα was used accordingly) or control vehicle [0.1% dimethyl sulfoxide (DMSO) in phosphate-buffered saline (PBS)] in a rotary incubator.

Tissues were oriented in custom-made cassettes (Department of Physics, Emory University) designed to fit a dual-channel, self-contained Ussing chamber (U2500, Warner Instruments) with the endothelial layer positioned facing the apical compartment. Cut, positioned veins were sealed in the cassettes with vacuum grease and a parafilm gasket, placed in the U2500, and then equilibrated with Ringer’s solution at 37°C using a temperature-controlled water bath. Solutions were bubbled with a 5% CO2/95% O2 medical-grade gas mixture to maintain oxygenation and pH. Ussing chamber data were collected with a VCC MC-8 Multichannel Voltage/Current Clamp controller and analyzed using Acquire & Analyze software (Physiological Instruments) (23).

For transendothelial dilution potential measurements, veins were equilibrated in either Ringer’s saline, Ringer’s lysine [140 mM lysine-HCl, 2 mM CaCl2, 1 mM MgCl2, 10 mM glucose, and 10 mM Hepes (pH 7.3)], or Ringer’s aspartate [140 mM Na-aspartate, 2 mM CaCl2, 1 mM MgCl2, 10 mM glucose, and 10 mM Hepes (pH 7.3)]. Next, medium on the apical side of the Ussing chamber was replaced with an osmotically balanced 1:3 dilution [diluted with iso-osmolar mannitol solution: 280 mM mannitol, 2 mM CaCl2, 1 mM MgCl2, 10 mM glucose, and 10 mM Hepes (pH 7.3)] and equilibrated for 5 min in each solution. The measurements were performed when current was clamped to 0 A with voltage measured in millivolts (24, 73). Tissue resistance and voltage measurements were made in Ringer’s solution and calculated in ohm×square centimeter. The ratio of Na permeability to Cl permeability (PNa/PCl) was calculated by the method of Hou et al. (74). All measurements were area-corrected.

Ex vivo vessel dye permeability measurements

Ex vivo permeability assays were developed from ex vivo perfusion assay as previously described (16). Second- and third-order veins or arteries were isolated from mouse mesenteric circulation in a Krebs solution containing 118.4 mM NaCl, 4.7 mM KCl, 1.2 mM MgSO4, 4 mM NaHCO3, 1.2 mM KH2PO4, 10 mM Hepes, 6 mM glucose, and 2 mM CaCl2. Veins and arteries with 2 to 3 mm in length were dissected, and the adventitia was removed, cannulated onto glass micropipettes, and sealed with sutures. To apply a constant flow to the lumen of the blood vessels, the micropipettes were connected to a perfusion pump system with a three-way stopcock enabling different media or agents to be perfused through the vessel using Krebs solution supplemented with 1% bovine serum albumin (BSA) (Krebs/BSA). Before permeability measurements, cannulated vessels were equilibrated for 20 min with Krebs/BSA.

To assess vessel permeability, fluorescein (10 μg/ml) (Sigma-Aldrich, MW 376.27) in Krebs/BSA (Krebs/BSA/fluorescein) was perfused through the lumen of the vessel, and diffusion into the abluminal compartment (containing Krebs/BSA) was collected at time intervals and measured. Quantification of fluorescein fluorescence (excitation, 490 nm; emission, 514 nm) of the collected abluminal fractions was carried out with a FLUOstar Omega microplate reader (BMG Labtech). Initially, nontreated, equilibrated vessels were measured for 30 min to obtain a baseline permeability measurement, then the abluminal buffer was supplemented with control vehicle (0.1% BSA in PBS) or TNFα (50 ng/ml) for additional 30 min, with abluminal buffer collected for fluorometric analysis.

Vessel permeability (Perm) was defined as the ratio of the rate of dye flux from the vessel lumen into the abluminal side under experimental conditions divided by the rate of flux at baseline (Perm = F/Fbaseline). TNFα-induced increases in permeability are presented as relative fluorescent units (RFU). To test the effects of different agents on permeability, vessels were treated during both baseline measurement (first 30 min of incubation) and TNFα treatment or control conditions (second 30 min of incubation). These agents included a Panx1 inhibitor (spironolactone, Sigma-Aldrich), an ecto-nucleotidase inhibitor (ARL67156, Tocris Bioscience), and A2A and A2 receptor antagonists (SCH58261 and ZM241385, respectively, Tocris Bioscience). Agonists, including A2A receptor agonist (CGS21680, Tocris Bioscience) and TRPV4 channel agonist [GSK1016790A (GSK101), Tocris Bioscience], were applied during the second 30 min of incubation instead of TNFα. Vessels treated with different agents were compared to vehicle-treated controls (0.1% DMSO).

Ca2+ imaging and analysis of TRPV4 Ca2+ sparklets

Third-order mesenteric vessel was isolated from C57BL/6 mice, cut transversely, and pinned down on a SYLGARD block in en face preparation as previously described (31, 75). Vessels were incubated with fluo-4 AM (10 μM) at 30°C for 30 min. All the experiments were performed at 37°C in bicarbonate physiological salt solution. Vessels were incubated with cyclopiazonic acid (CPA; 20 μM) at 37°C for 15 min to deplete internal Ca2+ stores and then with a low level of the TRPV4 agonist GSK101 (6 nM) for 5 min to confirm a healthy endothelium and to detect sites where sparklets will occur. To determine the effect of TNFα on TRPV4 sparklets, TNFα (10 ng/ml) was applied for 5 min in the presence or absence of spironolactone (80 μM). Vessels were then treated with the TRPV4 inhibitor GSK2193784 (GSK219, 100 nM) for 10 min to confirm that the TNFα-induced increase in sparklets was due to TRPV4 channels.

Ca2+ signals were detected in 1.7-μm2 regions of interest and using an increase in fluorescence over the averaged image obtained from 10 images, where TRPV4 Ca2+ sparklets were measured as previously described (31, 75). The average TRPV4 channel activity is defined as NPO (where N is the number of TRPV4 channels per site and PO is the open-state probability of the channel). NPO was calculated using the Single Channel Search module of Clampfit, quantal amplitudes derived from all-points histograms (ΔF/F0 of 0.29 for fluo-4–loaded vessels), and the following equation: NPO = (Tlevel1 + 2Tlevel2 + 3Tlevel3 + 4Tlevel4) / Ttotal, where T represents the dwell time at each quantal level and Ttotal is the total recording duration. Average NPO per site was obtained by averaging the NPO for all the sites in a field. When a signal crossed the threshold and came back to the baseline, it was counted as one event. All the events in a field during the recording duration were combined to obtain the total number of events per field of view. All the sites in a field of view during the recording duration were combined to obtain the total number of sites per field where the fold change in sparklet activity was normalized to untreated baseline values.

Immunohistochemical staining

For immunohistochemical staining of transverse sections, C57BL/6 mice were euthanized with carbon dioxide and sacrificed by cervical dislocation followed by perfusion through the heart with 4% paraformaldehyde (PFA) in PBS for fixation. Mouse mesenteric tissue isolates containing second-order artery and vein pairs were isolated and fixed in 4% PFA for 30 min, then incubated in 70% ethanol and subjected to paraffin embedding and sectioning (4 to 5 μm in thickness), and mounted on slides. For immunohistochemistry, sections on slides were deparaffinized and processed as previously described (16).

To analyze CD39 and CD73 expression, vessel sections were incubated with monoclonal antibodies directed against murine CD39 [clone EPR3678(2), Abcam] or CD73 (clone D7F9A, New England Biolabs), respectively. Expression of A2A and A2B adenosine receptors in cells of artery/vein pairs was determined with polyclonal antibodies directed against murine A2A (Thermo Fisher Scientific, PA1-042) or A2B (Thermo Fisher Scientific, PA5-33322), respectively. Antibody specificity was confirmed by secondary antibody–only control or by exchange of primary antibodies with the appropriate immunoglobulin G (IgG) control antibody.

For immunohistochemical staining of vessels en face, freshly isolated second- or third-order mesenteric arteries and veins were fixed with either 4% PFA at 4°C for 30 min (for labeling with CLDN5, Thermo Fisher Scientific, 34-1600) or with a 1:1 methanol acetone mixture at room temperature for 10 min (for labeling with CLDN11, Abcam, ab53041). In some cases, vessels were treated ex vivo with TNFα (10 ng/ml in Krebs/BSA) for 30 min at room temperature in the presence or absence of spironolactone pretreatment (10 μM in Krebs/BSA) for 15 min at room temperature before fixation. After PBS washes, vessels were cut en face and pinned out with tungsten wire (0.013 mm in diameter) on Sylgard. Next, en face preparations were permeabilized in 0.2% NP-40/PBS at room temperature for 30 min, incubated in 5% goat serum at room temperature for 1 hour (in 0.2% NP-40/PBS), and incubated with a primary antibody at 4°C overnight (1:100 in 5% serum/0.2% NP-40/PBS). The next day, en face preparations were washed in PBS and incubated with a secondary antibody (Thermo Fisher Scientific, A21069) at room temperature for 1 hour (1:400 in 5% goat serum/0.2% NP-40/PBS). After PBS washes, they were mounted on microscope slides with DAPI (4′,6-diamidino-2-phenylindole) mounting media; images were obtained using an Olympus Fluoview 1000 microscope.

To quantify the effect of TNFα on tight junction morphology, images were thresholded using ImageJ and scored for areas of cell-cell contact lacking CLDN11 that were 5 μm or greater in length as a junctional discontinuity. For each preparation, three fields of view containing at least eight endothelial cells were scored using vein preparations from at least three different mice.

Measurement of adenosine metabolites

Vessels were perfused with Krebs/BSA solution for 30 min. The perfusate was collected and diluted to a final proportion of 1:1 in inhibitor solution [Krebs/BSA containing 6 mM EDTA, 10 μM EHNA (erythro-9-(2-Hydroxy-3-nonyl)adenine hydrochloride; Sigma-Aldrich), 40 μM dipyridamole (Sigma-Aldrich), 5 nM NBTI (S-(4-Nitrobenzyl)-6-thioinosine; Sigma-Aldrich), 10 μM forskolin (Sigma-Aldrich), 100 μM IBMX (3-isobutyl-1-methylxanthine; Calbiochem), and 10 μM S-methyl-ITU (S-Methylisothiourea; Sigma-Aldrich)] (76). To extract metabolites, 1 ml of perfusate was diluted with 4 ml of methanol, acetone, and acetonitrile (1:1:1), and the resulting supernatant was analyzed by liquid chromatography–mass spectrometry and compared with calibration standards. Individual adenosine-containing metabolites were normalized to the total of ATP + ADP + AMP + adenosine present in the sample.

Cell culture

Primary HUVECs were purchased from Cell Applications (200K-05), primary HSaVECs and HAoECs from PromoCell (C-12231 and C-12271, respectively), and HCoECs from Lonza (CC-2585). All cells were cultured in Microvascular Endothelial Cell Growth Medium-2 (complete medium, EGM-2MV, Lonza) and used for no more than five passages (16).

In vitro measurement of TEER

For in vitro TEER measurements, cells were seeded on a fibronectin (25 μg/ml; Sigma-Aldrich)–coated Transwell polycarbonate membrane (6.5 mm in diameter, 3.0 μm pore size, Corning Costar) at 1 × 105 cells per well (100 μl apical volume), in a 24-well plate tissue culture dish containing 600 μl of medium (abluminal compartment). After 24 hours of culture, cells were used for TEER experiments. At 30 min before TEER measurements, TNFα (10 and 100 ng/ml, respectively; R&D Systems) or vehicle (0.1% BSA in PBS, control) was added to the cells in the upper compartment. Subsequently, cells were placed in an EndOhm chamber (World Precision Instruments), and TEER was measured using the Millicell-ERS (Electrical Resistance System, Millipore). A Transwell chamber without cells was used as a blank.

Quantitative real-time PCR and microarray analysis

RNA was isolated from HUVEC, HAoEC, HSaVEC, and HCoEC using the RNeasy mini kit (Qiagen, catalog no. 74104) with respective deoxyribonuclease (DNase) digestion (ribonuclease-free DNase set, Qiagen, catalog no. 79254). The isolated RNA was then reverse-transcribed into complementary DNA (cDNA) using random hexamers and SuperScript III (Thermo Fisher Scientific, catalog no. 18080093). To analyze transcript expression, cDNA was used as a template for qRT-PCR analysis using the TaqMan Gene Expression Master Mix (Thermo Fisher Scientific, catalog no. 4369016). Each transcript was measured in duplicates using gene-specific TaqMan probes, and the relative transcript expression was determined compared to housekeeping gene B2M by using the X0 method (77). Target TaqMan probes used were CD39 Hs00969559_m1; CD73 Hs00159686_m1; ADORA 2A Hs00169123_m1; ADORA 2B Hs00386497_m1; ADORA 1 Hs00379752_m1; ADORA 3 Hs00252933_m1; B2M Hs00984230_m1.

For microarray analysis, cytoplasmic RNA was isolated as described above and analyzed on an Agilent 2100 Bioanalyzer (Agilent, Palo Alto, CA) to assess RNA integrity. Double-stranded cDNAs were synthesized from total RNA by using the cDNA Synthesis System (Invitrogen, Carlsbad, CA). Biotin-labeled complementary RNA was generated and used to probe Affymetrix HuGene 1.0 ST v1 Arrays according to the manufacturer’s recommendations (Expression Analysis Technical Manual, 2001, Affymetrix, Santa Clara, CA). For each Affymetrix dataset, raw celfiles were normalized using invariant set normalization, and the perfect match - mismatch (PM-MM) model of expression was used to construct gene expression measures and SEs for each probe set (dCHIP). An N = 3 for each cell type allowed for weighted mean and variance calculations. Probe sets in the top 5% with a fold difference greater than ±2.0 were analyzed. Probe sets with an absolute difference less than 100 expression units relative to control were removed for all Affymetrix array analyses. This filtering allowed for identification of genes well above background levels. Absent calls were also excluded. Microarray data were analyzed using Transcriptome Analysis Console Software (Applied Biosystems, Thermo Fisher Scientific) to produce expression scatterplots and for hierarchical cluster analysis. Microarray data were uploaded onto the NIH Gene Expression Omnibus website, accession no. GSE137041.

Sepsis model

Mice were subjected to the CLP model of peritonitis-induced sepsis (78). Mice were anesthetized using isoflurane, and then a midline abdominal incision was made, and the cecum was ligated immediately distal to the ileocecal valve. The cecum was then punctured with an 18-gauge needle, a small amount of stool was extruded, and the gut was returned to the abdominal cavity that was closed in layers. All septic mice were injected subcutaneously with 1 ml of normal saline to balance fluid losses that occurred during surgery. For functional analysis, mice received antibiotics (25 mg/kg of ceftriaxone + 12.5 mg/kg of metronidazole, ip) at 3 and 15 hours postoperatively and then were euthanized 24 hours after CLP. Mice that were followed for 5 days for survival studies received antibiotics at 48 hours.


All graphs and statistics were produced or calculated using GraphPad Prism. Unpaired two-tailed t tests and one-way and two-way analysis of variance (ANOVA) tests were used to determine statistical significance as specified in the figure legends. qRT-PCR data were analyzed for statistical significance by one-way ANOVA followed by Tukey’s post hoc test. Results are presented as means ± SD or means ± SEM as indicated in the figure legends.


Fig. S1. Dose-response curves for the effect of TNFα on vessel permeability and endothelial cell viability.

Fig. S2. Hierarchical cluster diagram comparing venous and arterial endothelial cells.

Fig. S3. Hierarchical cluster diagram showing mRNAs of venous and arterial endothelial cells that differ fourfold or more in magnitude.

Fig. S4. Cldn11 is preferentially expressed by venous endothelium.

Fig. S5. Panx1 inhibition reduced the severity of renal injury in response to systemic sepsis.

Data file S1. List of genes differentially expressed by venous and arterial endothelial cells.


Acknowledgments: A2A/ mice were a gift from V. E. Laubach (University of Virginia School of Medicine). We thank the Emory Department of Physics machine shop for design and fabrication of custom cassettes for the Ussing chamber. Funding: This work was supported by R01-HL137112 (to B.E.I. and M.K.), R01-AA025854 (to M.K.), R01-HL120840 (to B.E.I.), R56-HL138496 (to S.K.S.), R00-HL143165 (to M.E.G.), K12-GM000680 (to H.S.C.), AHA-CDA 19CDA34630036 (to S.R.J.), DFG SFB914 A02 (to D.M.-B.), and DFG IRTG 1902 (to N.K.). Author contributions: D.M.-B. performed permeability assays. D.M.-B., S.A.M., and M.K. created and analyzed data from Ussing chambers. N.K., H.M.B., M.E.G., L.J.D., L.A.B., A.W.L., B.E.I., and S.R.J. used endothelial cells and performed mRNA analysis and Western blots. A.W.L. and M.K. analyzed microarray data. Y.-L.C., K.H., and S.K.S. performed and analyzed calcium data. D.M.-B., H.M.B., S.R.J., H.S.C., and C.A.R. performed imaging. X.C. and Z.Y. performed and analyzed sepsis experiments. D.M.-B., S.A.M., M.K., and B.E.I. participated in experimental design. D.M.-B., M.K., and B.E.I. wrote and edited the manuscript. B.E.I. and M.K. managed the project and financial backing. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The microarray data have been deposited in the NCBI Gene Expression Omnibus and are accessible through GEO Series accession number GSE137041. All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.

Stay Connected to Science Signaling

Navigate This Article