Rho and Rac but not Cdc42 regulate endothelial cell permeability

Members of the Rho family of GTPases regulate both actin cytoskeletal organization and the integrity of intercellular junctions (Hall, 1998). Rho itself induces the formation of actin stress fibres in fibroblasts (Ridley and Hall, 1992), epithelial cells (Ridley et al., 1995) and endothelial cells (Wójciak-Stothard et al., 1998), and acts at least in part by increasing the phosphorylation of myosin light chain (MLC), which in non-muscle cells leads to actomyosin contraction (Ridley, 1999). Rho-induced phosphorylation of MLC is mediated by Rho kinases (also known as ROCKs or ROKs), which phosphorylate and inhibit the activity of MLC phosphatase. Rac and Cdc42 also regulate actin organization, inducing the extension of lamellipodia and filopodia, respectively (Ridley et al., 1992; Nobes and Hall, 1995). In addition, Rho, Rac and Cdc42 affect the assembly of E-cadherin-containing adherens junctions in epithelial cells, and inhibition of either Rho, Rac or Cdc42 activity leads to a reduction in E-cadherin-mediated cell adhesion (Braga et al., 1997; Hordijk et al., 1997; Takaishi et al., 1997; Jou and Nelson, 1998; Fukata et al., 1999; reviewed in Kaibuchi et al., 1999). One mechanism whereby Rac and Cdc42 may enhance adherens junction formation is by repressing the antiadhesive activity of their downstream target IQGAP1 (Fukata et al., 1999). In endothelial cells, however, inhibiting Rho or Rac does not perturb the localization of endothelial-specific VE-cadherin to intercellular junctions (Braga et al., 1999), suggesting that the regulation of adherens junctions differs according to context or cell type. Expression of either constitutively activated or dominant-negative forms of Rho and Rac also perturbs tight junction structure and function in epithelial cells (Nusrat et al., 1995; Jou et al., 1998).

There is accumulating evidence that Rho regulates endothelial permeability, which depends on the integrity of intercellular junctions and actomyosin contractility (Hordijk et al., 1999; Essler et al., 1998a,b; Essler et al., 1999). The vasoactive mediators thrombin and histamine have long been known to increase vascular permeability in vivo (Majno and Palade, 1961; Haraldsson et al., 1986; Wu and Baldwin, 1992) and in vitro (Killackey et al., 1986; Lum et al., 1992; Ehringer et al., 1996). Increased vascular permeability is essential for inflammatory responses, but can also contribute to the development of pathological conditions such as atherosclerosis (Stemerman et al., 1986; Stender and Hjelms, 1987; Ross, 1993; Raines and Ross, 1996) and is accompanied by alterations to cell-cell junctions and the actin cytoskeleton. Thrombin-induced endothelial permeability has been attributed to stress fibre formation and subsequent actomyosin-mediated contraction of cells (Lum and Malik, 1996; Van Hinsbergh, 1997; Essler et al., 1998a) as well as to changes in the distribution and phosphorylation of intercellular adhesion proteins such as cadherins and catenins (Rabiet et al., 1996; Drenckhahn and Ness, 1997; Lampugnani and Dejana, 1997). The relative contributions however, of these two responses to the increase in endothelial permeability are not clear.

In order to investigate the contribution of different signalling pathways to the regulation of endothelial cell permeability, it is essential to be able to inhibit the activity of specific signalling proteins in the majority of an endothelial cell population. As normal endothelial cells are not transfected efficiently, this has been achieved primarily by adding cell-permeable pharmacological inhibitors to cells. For example, an inhibitor of Rho kinase can inhibit thrombin-induced permeability (Essler et al., 1998a). By using C3 transferase, a bacterial exoenzyme that can enter cells and modify Rho, it has been possible to implicate Rho in both thrombin-induced permeability and stress fibre formation in endothelial cells (Essler et al., 1998a; Van Nieuw Amerongen et al., 1998; Carbajal and Schaeffer, 1999), but the roles of Cdc42 and Rac in regulating permeability have not been described. To investigate the effects of Rho, Rac and Cdc42 on endothelial cell permeability we have developed recombinant adenoviruses as an efficient means to express Rho, Rac and Cdc42 proteins in endothelial cells. This has allowed us to correlate changes in permeability with alterations in both the actin cytoskeleton and intercellular junctions, using immunofluorescence and transmission electron microscopy.

We have shown that inhibition of Rho or Rho kinase prevents thrombin- and histamine-induced loss of tight junctions and adherens junctions, as well as inhibiting stress fibre formation, and that Rho is required for both thrombin- and histamine-induced vascular permeability. Inhibiting Rac activity by itself leads to increased permeability and enhances thrombin-induced permeability, reflecting Rac-induced changes to intercellular junctions and the appearance of intercellular gaps. In contrast, inhibition of Cdc42 has no effect on endothelial permeability, despite its effects on the actin cytoskeleton.

Human fibronectin, FITC-phalloidin and histamine were obtained from Sigma-Aldrich (Poole, UK); Y-27632 was from Welfide Corporation (Osaka, Japan); FITC-dextran, tetramethylrhodamine isothiocyanate (TRITC)-dextran, BODIPY-, fluorescein- and TRITC-labelled goat anti-mouse secondary antibodies were from Molecular Probes (Leiden, The Netherlands); mouse monoclonal anti-human cadherin-5 (VE-cadherin) antibody was from BDPharmingen (San Diego, CA, USA); rabbit anti-occludin antibody was from Zymed (San Francisco, CA, USA); anti-c-myc monoclonal antibody (9E10) was from Santa Cruz Biotechnology (Santa Cruz, CA); Moviol was obtained from Calbiochem (Nottingham, UK).

To purify human umbilical vein endothelial cells (HUVECs), umbilical veins were flushed twice with PBS, then filled with 0.1% collagenase in PBS (Sigma) prewarmed to 37°C, and incubated for 20 minutes at 37°C. The collagenase solution was then collected and the veins were washed twice with 20 ml of Medium 199 (Life Technologies, Paisley, UK) containing 20% fetal calf serum (FCS) to collect the remaining cells. The collagenase solution and the washes were combined together and centrifuged. The cell pellet was resuspended in 20 ml of Medium 199/20% FCS and cells plated into a 75 ml culture flask (Nunclon) coated with 10 μg/ml fibronectin. After 3 hours the medium was changed to remove non-adherent cells. Cells were cultured in Medium 199 containing 20% FCS, 100 μg/ml endothelial cell growth supplement (Sigma), 100 μg/ml heparin (Sigma), 1% Nutridoma NS (Boehringer Mannheim Ltd, Lewes, UK) in flasks coated with 10 μg/ml human fibronectin, as previously described (Wojciak-Stothard et al., 1998). To determine the purity of the cell population, the uptake of low-density lipoprotein (LDL) was measured by adding DiI-labelled LDL (Molecular Probes, Leiden, The Netherlands) to cells at 2.5 μg/ml. 98-99% of cells in the third passage labelled positive for diI-LDL. For some experiments, HUVECs kindly provided by Ruggero Pardi (DIBIT, Milan, Italy) were used. For experiments, HUVECs were used between 2 and 5 passages. Human thrombin was added to the culture medium at a final concentration of 1 U/ml, and histamine at 10 μM.

For confocal microscopy, HUVECs were grown on glass coverslips coated with 10 μg/ml human fibronectin until confluent. To obtain quiescent HUVECs, the culture medium was replaced by medium containing 10% FCS but no heparin or other growth factors, as previously described (Wojciak-Stothard et al., 1998), and the cells were incubated in this new medium for 16-20 hours before microinjection and/or treatment with different agents.

For permeability studies, HUVECs were grown on fibronectin-coated polyester Transwell-Clear filters (3 μm pore size, 12 mm diameter; Corning Costar Corporation, Costar Ltd, High Wycombe, UK). The cells were plated at 3×10⁵ cells/well, and after 4 hours non-adherent cells were removed. Filters were used for experiments 4 days after plating. The Rho kinase inhibitor Y-27632 (5 μ?? was added to the upper chamber 30 minutes before the addition of histamine or thrombin and C3 transferase (15 μg/ml) 1 hour before stimulation. Although the cells grown on filters were visible under the light microscope and were examined for confluency before the experiment, all filters were fixed and then stained with Coomassie Blue after the experiment to reveal the general appearance of the cell monolayer and to visualize gaps between cells.

The recombinant proteins, V12Rac1, N17Rac1, N17Cdc42 and C3 transferase, were expressed in Escherichia coli from the pGEX-2T vector as glutathione S-transferase fusion proteins and purified as previously described (Ridley et al., 1992). Protein concentrations were estimated by Bradford assay using a protein assay kit (Bio-Rad).

Recombinant proteins were microinjected into the cytoplasm of HUVECs together with TRITC-dextran (MW 10,000) (5 mg/ml) to identify injected cells. C3 transferase was microinjected at a concentration of 4 μg/ml, V12Rac1 at approximately 0.75 mg/ml, N17Cdc42 at 1 mg/ml and N17Rac1 at 2 mg/ml. Histamine and thrombin were added 15 minutes after microinjection.

To generate recombinant adenoviruses, cDNAs encoding amino-terminal myc-tagged N19RhoA, N17Cdc42, N17Rac1 and V12Rac1 were subcloned into the admid transfer vectors pCR259 and pCR244. The admid transfer vectors were then transformed into an E. coli strain containing a vector encoding a transposase and an Ad5-based adenovirus vector deleted in the E1 and E3 genes (Ad5 ΔE1ΔE3). Transposition of the Rho cDNAs from the admid transfer vectors into Ad5 ΔE1ΔE3 created the adenoviral vectors Ad-N19RhoA, Ad-N17Rac1, Ad-N17Cdc42 and Ad-V12Rac1, where the transgenes are under the control of a CMV promotor.

Recombinant adenoviral DNA was purified from E. coli and transfected into 293 human embryonic kidney cells, which express E1 genes, to allow purification of adenoviral particles. Ad-p-Gal, encoding the p-galactosidase protein, was a kind gift from Carolyn Dent (GlaxoWellcome, Stevenage, UK). The titer of adenovirus stocks was determined by titration on 911 cells (Fallaux et al., 1996).

Confluent HUVECs grown on Transwell filters or glass coverslips were infected with adenoviruses at a multiplicity of infection (MOI) of 1500 in medium containing 10% FCS. Adenoviruses were added only into the top chamber of Transwells. For transmission electron microscopy, HUVECs were grown on a flexible transparent substratum, melanex (ICI), coated with fibronectin. After incubation for 4 hours the virus-containing medium was removed and fresh medium containing 10% FCS was added. The cells were then incubated for a further 18 hours before the addition of thrombin or histamine.

To detect p-galactosidase activity, Ad-p-Gal-infected cells were fixed with 3.7% formaldehyde for 20 minutes. After washing twice with PBS, 1 mg/ml of X-gal (5-bromo-4-chloro-3-indoyl galactoside) substrate (Promega, Southhampton, UK) in PBS containing 5 mM potassium ferricyanide, 5 mM potassium ferrocyanide and 2 mM MgCl₂ was added and cells were incubated at 37°C for 1-2 hours. Blue cells were counted to monitor expression of p-galactosidase.

Transendothelial permeability was measured as previously described (Draijer et al., 1995). HUVECs in Transwell chambers were incubated in Medium 199 with 1% bovine serum albumin (BSA) for 1 hour before adding 1 mg/ml FITC-dextran (MW 42,000) in the same medium together with thrombin and histamine. Samples were taken at 15 minutes, 30 minutes, 1 hour, 2 hours and 3 hours after stimulation from the lower compartment of the Transwell chambers and an equal volume of Medium 199 with 1% BSA was re-added to the lower chamber. The amount of FITC-dextran in the upper and lower wells was determined with a fluorometer (RF-5301PC, Shimadzu Corp. Kyoto, Japan), using an excitation wavelength of 492 nm, and detecting emission at 520 nm. In experiments with adenovirally infected cells, the flux of FITC-dextran (μg/hour/cm²) was compared with control cells over the first hour after addition of thrombin or 30 minutes after addition of histamine, as at these time points the highest accumulation of FITC-dextran was obtained in the lower wells of the Transwell chambers. The results in Fig. 2A represent mean values of fluorescence ± s.d. of three independent experiments and are presented as percentage of the fluorescence in controls.

The electrical resistance of HUVEC monolayers cultured on Transwell filters was measured with a Millicell-ERS instrument (Millipore Continental Water Systems, Bedford, MA, USA). Resistance measurements were taken before and 5 minutes after stimulation with thrombin or histamine. The measured potential difference between the upper and lower wells was used to calculate the electrical resistance in Ω cm². Transendothelial resistance (TEER) values were then calculated by subtracting the contribution of the filter and the bathing solution. To normalize the results of different experiments, the TEER for monolayers of control cells (uninfected) was taken as 100%. Changes in TEER are presented as percentage of the control value. Experiments were performed in triplicate and values are means ± s.d.

To visualise the distributions of VE-cadherin, β-catenin, ZO-1, vinculin and F-actin, HUVECs were fixed with 4% formaldehyde dissolved in PBS for 10 minutes at room temperature and permeabilised for 6 minutes with 0.2% Triton X-100. Cells were incubated in 0.5% BSA in PBS for 45 minutes to block nonspecific antibody binding and then incubated with 1 μg/ml FITC-phalloidin for 45 minutes to stain actin filaments or with mouse monoclonal anti-human cadherin-5 (VE-cadherin) antibody (1:400), rabbit anti-ZO-1 antibody (1:200), mouse monoclonal anti-β-catenin antibody (1:200) or mouse monoclonal anti-vinculin antibody VIN-11-5 (1:200). For occludin staining, cells were pre-extracted, fixed in 95% ethanol for 30 minutes, then rehydrated and incubated with rabbit anti-occludin antibody (1:20) as described (Balda et al., 1996). Primary antibodies were visualised by incubation with BODIPY-, fluorescein-or TRITC-labelled goat anti-mouse or goat anti-rabbit secondary antibodies for 1 hour at room temperature and the specimens mounted in Moviol.

Confocal laser scanning microscopy was carried out with an LSM 510 (Zeiss, Welwyn Garden City, UK), using a ×10 eyepiece, and either a ×40 NA 1.3 or a ×63 NA 1.4 oil immersion objective (Zeiss). FITC and TRITC were excited at 488 nm and 543 nm and visualised with a 540±25 and a 608±32 nm bandpass filter, respectively, where the levels of interchannel cross-talk were insignificant. Image files were collected as a matrix of 1024×1024 pixels describing the average of 8 frames scanned at 0.062 Hz.

For transmission electron microscopy (TEM), HUVECs were fixed with 2% paraformaldehyde/2% glutaraldehyde in PBS for 30 minutes. Cells were prepared for analysis by TEM by Mark Tumaine (Department of Anatomy and Developmental Biology, University College London, UK). After fixation, the cells were washed three times in 0.1 M sodium cacodylate, then incubated for 10 minutes in 0.1 M sodium cacodylate and 1% osmium tetroxide, then washed again in 0.1 M sodium cacodylate and distilled water. The samples were dehydrated by subsequent washes in 25%, 50% and 100% ethanol, then embedded in epoxy resin and viewed in TEM Jeol 1010. In order to identify tight junctions and adherens junctions, the stage goniometer was used to tilt the specimen and view cross sections of cell membranes at different angles. The quantitative analysis of the number of tight junctions and adherens junctions on lateral membranes by TEM was based on the methodology of Burns et al. (Burns et al., 2000). Results are presented as means ± s.d. from 2-3 separate experiments where 20-25 lateral membranes/monolayer were analysed in three monolayers/treatment group.

Data are presented as means ± s.d. Comparisons between >2 groups were made using a one-way ANOVA test followed by Tukey post-test for multiple comparisons (TEER and TEM data) or Kruskal-Wallis test followed by Dunn’s post-test (endothelial permeability data). The choice between parametric and non-parametric tests was based on the Bartlett’s test for homogeneity of variances. Statistical significance was accepted for P?0.05 and all tests were performed with GraphPad Instat version 2.01.

To investigate the contributions of Rho, Rac and Cdc42 to thrombin- and histamine-induced endothelial permeability, and correlate this with changes in the actin cytoskeleton and cell-cell junctions, we first characterized the effects of thrombin and histamine under precisely controlled conditions of HUVEC culture. 4 days after reaching confluence, HUVECs were maintained in 10% FCS without growth supplements for 24 hours prior to stimulation. Under these conditions, HUVECs were polygonal in shape and had few stress fibres traversing the body of cells, although there were some stress fibres localised at the cell periphery (Fig. 1A). VE-cadherin, the major transmembrane component of endothelial adherens junctions (Lampugnani and Dejana, 1997), as well as occludin, a transmembrane component of tight junctions (Balda and Matter, 1998), were localized along the cell margins as previously reported (Lampugnani et al., 1995; Burns et al., 2000) (Fig. 1D,G).

Thrombin induced rapid assembly of stress fibres (Fig. 1B) and vinculin-containing focal contacts (data not shown) in quiescent HUVECs. The peak in stress fibre formation was observed within 3-5 minutes of thrombin addition, and the earliest changes in F-actin distribution were observed after 1-2 minutes, consistent with previous observations on F-actin (Goeckeler and Wysolmerski, 1995) and on the time course of myosin light chain phosphorylation (Van Nieuw Amerongen et al., 1998). Thrombin also induced disappearance of occludin from cell margins (Fig. 1E) and disruption of the linear VE-cadherin staining at intercellular junctions (Fig. 1H). Although VE-cadherin was still localized to areas of cell-cell contact following thrombin stimulation, the junctions became segmented and discontinuous. This in part reflected the appearance of gaps between cells (e.g. arrows in Fig. 1B), and some cells clearly retracted and became rounded. Thrombin induced similar changes in the actin cytoskeleton, cell retraction and intercellular gap formation when endothelial cells were grown on fibronectin-coated Transwell filters (data not shown), demonstrating that the cells behave similarly on glass and on filters.

Like thrombin, histamine promoted the formation of stress fibres within 5 minutes of addition (Fig. 1C). The earliest changes in the distribution of actin filaments were observed after 1 minute of histamine treatment, consistent with reports by others (Niimi et al., 1992). In contrast to thrombin, histamine-treated cells did not retract and although gaps were observed between some cells (Fig. 1C, arrow), they were considerably less abundant and generally much smaller than those observed with thrombin (Fig. 1B). Histamine caused fragmentation of occludin and VE-cadherin staining (Fig. 1F,I). In contrast to the response to TNF-α (Wójciak-Stothard et al., 1998), no membrane ruffling or formation of lamellipodia or filopodia was detected in either thrombin-or histamine-treated cells.

The rapid time course of stress fibre formation and cell junction changes induced by histamine and thrombin correlated with a decrease in endothelial barrier function, measured either by transendothelial electrical resistance (TEER) or passage of FITC-dextran through HUVEC monolayers grown on Transwell filters. Thrombin and histamine induced a decrease in TEER within the first 5 minutes of stimulation (Fig. 2B). These small but consistently observed changes in TEER were in the range reported by others (Langeler and van Hinsbergh, 1991; Westendorp et al., 1994). Loss of endothelial barrier function was also assayed by measuring endothelial permeability to FITC-dextran. FITC-dextran was added to the upper chamber of Transwells, and the amount of FITC-dextran that had accumulated in the lower chamber was determined at different time points after stimulation. The passage rate of FITC-dextran through control, unstimulated HUVECs was 20-25 μg/hour/cm² (measured over the first hour after addition of FITC-dextran), and was 10-to 15-fold lower than through filters without cells. Thrombin induced an increase in FITC-dextran flux, and this was highest (60-70 μg/hour/cm²) when measured over the first hour after stimulation (Fig. 2A, insert a), reflecting the fact that thrombin induces a rapid decrease in endothelial barrier function, which then gradually returns to basal levels (Rabiet et al., 1996; Van Nieuw Amerongen et al., 1998; Burns et al., 2000). With histamine, the maximal rate of FITC-dextran flux (30-40 μg/hour/cm²) was observed during the first 30 minutes after stimulation (Fig. 2A, insert b), consistent with the more transient increase in permeability reported previously for histamine compared to thrombin (Rabiet et al., 1996; Van Hinsbergh et al., 1997; Andriopoulou et al., 1999). As changes in FITC-dextran flux were larger and more easily measurable than the relatively small changes in TEER, this was used as a measure of endothelial permeability in the majority of subsequent experiments.

In order to assess the roles of Rho, Rac1 and Cdc42 in regulating thrombin- and histamine-induced changes in endothelial permeability, confluent HUVECs on Transwell filters were infected with the recombinant adenoviruses to express the dominant-negative mutants N19RhoA, N17Rac1 and N17Cdc42. In addition, we constructed an adenovirus containing constitutively active V12Rac1, as expression of V12Rac1 has been reported to enhance adherens junction assembly and/or stability in epithelial cells (Hordijk et al., 1997; Takaishi et al., 1997; Sander et al., 1998; Stöffler et al., 1998), but to perturb epithelial tight junctions (Jou et al., 1998). All proteins were tagged at the amino terminus with a myc epitope to facilitate analysis of their expression and localization. The infection efficiency of HUVECs was initially evaluated using Ad-p-gal, encoding the reporter protein p-galactosidase. By 18 hours after adenoviral infection 80±10% of cells expressed detectable levels of p-galactosidase (data not shown). Expression of the Rho GTPases was subsequently monitored by western blotting and immunofluorescence, and was detectable at 6 hours after infection, although it increased several-fold by 18 hours (data not shown).

Infection with Ad-p-gal had no effect on endothelial permeability in either unstimulated or stimulated cells (Fig. 2). Both the thrombin- and histamine-induced increase in permeability was significantly reduced in cells infected with Ad-N19RhoA (P<0.05) (Fig. 2A). Incubation of cells with C3 transferase, an exoenzyme from Clostridium botulinum that specifically inhibits Rho by ADP-ribosylating it (Machesky and Hall, 1996), also inhibited histamine- and thrombin-induced permeability (P<0.05) (Fig. 2A). Similarly, Y-27632, an inhibitor of Rho kinase acting downstream of Rho, significantly inhibited thrombin- and histamine-induced permeability (P<0.05).

Ad-V12Rac1 and Ad-N17Rac1 both significantly increased endothelial permeability in unstimulated cells at 18 hours after infection (P<0.05), and Ad-N17Rac1 enhanced the response to thrombin (P<0.05) (Fig. 2A). In contrast, expression of N17Cdc42 had no significant effect on control or thrombin/ histamine-induced permeability. Taken together, these results indicate that Rho and Rho kinase activity is required for thrombin- and histamine-induced increased permeability whereas perturbation of Rac activity in itself induces increased permeability.

In addition to inhibiting thrombin- and histamine-induced endothelial permeability as measured by FITC-dextran flux, expression of N19RhoA or pretreatment of cells with C3 transferase or Y-27632 prevented the transient drop in TEER induced by thrombin and histamine (Fig. 2B). N17Rac1 and N17Cdc42 did not have a significant effect on TEER (data not shown). These data are consistent with a role for Rho in mediating thrombin- and histamine-induced decreases in endothelial barrier function.

As inhibition of Rho function using either N19RhoA, C3 transferase or the Rho kinase inhibitor Y-27632 reduced thrombin- and histamine-induced permeability, we investigated the effects of these Rho inhibitors on F-actin organization and intercellular junctions in thrombin- and histamine-stimulated cells. As expected, microinjection of C3 transferase into HUVECs completely prevented thrombin- and histamine-induced formation of stress fibres (Fig. 3A-C and data not shown). The microinjected cells remained well spread and were almost completely devoid of stress fibres or actin bundles, although F-actin was still associated with intercellular junctions. Similarly, infection of cells with Ad-N19RhoA inhibited thrombin- and histamine-induced stress fibre formation (Fig. 3E,G), and prevented thrombin-induced contractility so that intercellular gaps were not observed (Fig. 3E, compare with Fig. 3A). In addition, C3 transferase and N19RhoA inhibited the fragmentation of VE-cadherin staining induced by thrombin and histamine (Fig. 3D,F, compare with Fig. 1H,I). Control Ad-p-gal had no effect on thrombin and histamine responses. Y-27632 acted similarly to Ad-N19RhoA and C3 transferase in inhibiting thrombin- and histamine-induced changes to the actin cytoskeleton and adherens junctions (data not shown). Y-27632 also prevented loss of occludin from intercellular borders induced by thrombin (Fig. 3I,J). Ad-p-gal and Ad-N19RhoA did not affect the distribution of F-actin or junctional proteins in unstimulated cells (Fig. 5A,B and data not shown).

Transmission electron microscopy (TEM) indicated that the cells infected with Ad-p-gal as well as uninfected cells had intercellular junctions but there were few tight junctions (Fig. 4A), as previously reported in HUVECs in the absence of growth supplements (Burns et al., 1997). Tight junctions were identified as points where the outer leaflets of lateral membranes between adjacent endothelial cells appeared to fuse (Fig. 4F; Burns et al., 1997). In control cells, tight junctions were detected in 20±2% of lateral membranes between adjacent cells, and this level was not significantly different from that in cells treated with Y-27632 or in cells expressing N19RhoA (Table 1). Adherens junctions lacked membrane fusion and were identified as areas of close apposition of cell membranes associated with electron-dense material. The percentage of lateral membranes where adherens junctions were detected was also similar in control, N19RhoA-expressing and Y-27632-pretreated cells (Table 1).

TEM analysis of regions of intercellular contact between thrombin-stimulated endothelial cells showed that there were often wide gaps between cells, and intercellular junctions were rarely observed (Fig. 4C, Table 1). Pretreatment of cells with Y-27632 prevented the loss of tight junctions and adherens junctions induced by thrombin (Table 1, Fig. 4D,F). Similarly, Y-27632 prevented histamine-induced loss of tight junctions and adherens junctions (Table 1).

Rho and its downstream target Rho kinase are therefore required both for changes in the actin cytoskeleton and to intercellular junctions induced by thrombin and histamine. In particular, the ability of N19RhoA and Y-27632 to prevent loss of tight junctions in thrombin- and histamine-stimulated cells provides an explanation for their inhibitory effect on increased endothelial permeability.

As expression of both N17Rac1 and V12Rac1 increased endothelial permeability even in unstimulated HUVECs, we analysed their effects on intercellular junctions. VE-cadherin and β-catenin in cells infected with Ad-N17Rac1 localized to intercellular borders, but in many places formed a meshwork-like pattern (Fig. 5E,F). Though also occasionally observed in control cells (e.g. Figs 1G, 5A), this pattern of staining was much more widespread in N17Rac1-expressing cells. Levels of ZO-1 at intercellular junctions appeared reduced in most N17Rac1-expressing cells, and ZO-1 was absent or had a fragmented localization along some intercellular borders (Fig. 5F, compare to Fig. 5B), suggesting loss of tight junctions. Intercellular gaps were observed between some cells (e.g. arrow in Fig. 5E).

TEM analysis of HUVECs expressing N17Rac1 revealed that lateral membranes from adjacent cells often came into close apposition in only a few places and between these there were large intercellular gaps (Fig. 5D), which were not observed in control cells (Fig. 4A). N17Rac1-expressing cells showed a significant decrease in the percentage of endothelial cell borders with tight junctions and adherens junctions (Table 1). Taken together, these results indicate that adherens junctions and tight junctions were weakened in cells expressing N17Rac1. This is likely to be responsible for the increased permeability observed in these cells.

We have previously observed that microinjection of V12Rac1 protein rapidly induces lamellipodium extension in HUVECs (Wójciak-Stothard et al., 1998). In contrast, AdV12Rac1-infected HUVECs did not show lamellipodia at 18 hours after infection, probably because lamellipodium extension is a transient response to Rac in confluent quiescent cells. However, some V12Rac1-expressing HUVECs had increased levels of stress fibres (data not shown), as also observed at later time points after V12Rac1 protein injection (Wójciak-Stothard et al., 1998). Some cells expressing high levels of V12Rac1 retracted to leave intercellular gaps, and in these places VE-cadherin and β-catenin were not detected at the plasma membrane (Fig. 5G, arrow). ZO-1 staining at intercellular junctions was considerably more fragmented than in control cells (Fig. 5H), indicating disruption of tight junctions. Consistent with this, analysis of TEM micrographs revealed that cells expressing V12Rac1 had significantly fewer tight junctions and adherens junctions than control cells (Table 1, Fig. 5C). In some cases, lateral membranes of adjacent cells had only occasional sites of close contact (data not shown) as in N17Rac1-expressing cells (Fig. 5D). Overall, these results suggest that the increase in endothelial permeability observed in V12Rac1-expressing cells is a consequence of loss of tight junctions and intercellular gap formation.

Expression of N17Rac1 not only induced permeability of quiescent HUVECs, but also enhanced the permeability of thrombin-stimulated HUVECs (Fig. 2A). To determine whether this reflects changes in the actin cytoskeleton and/or to intercellular junctions, cells were infected with Ad-N17Rac1 or microinjected with N17Rac1 protein before stimulation with thrombin or histamine. Interestingly, thrombin- and histamine-induced stress fibre formation was inhibited both in cells injected with N17Rac1 protein (Fig. 6A,B) and in Ad-N17Rac1-infected cells (Fig. 6C,E). Rac1 is best-characterized for its role in regulating membrane ruffling and lamellipodium formation, but in some situations can act upstream of Rho to mediate stress fibre formation, for example in PDGF-stimulated fibroblasts (Ridley et al., 1992) and TNF-α-stimulated endothelial cells (Wójciak-Stothard et al., 1998). In thrombin-stimulated cells expressing N17Rac1, gaps between cells were still occasionally detected (Fig. 6C, arrow) although they were less pronounced than in thrombin-stimulated control cells. VE-cadherin staining in N17Rac1-expressing cells was less fragmented than in uninfected thrombin-or histamine-treated cells but showed a meshwork-like pattern similar to that in unstimulated N17Rac1-expressing cells (Fig. 6D,F; compare with Fig. 5E). These results suggest that Rac1 acts upstream or in parallel with Rho to mediate thrombin- and histamine-induced stress fibre formation. However, in contrast to the effects of inhibiting Rho, inhibition of stress fibre formation by N17Rac1 does not correlate with decreased endothelial permeability (Fig. 2). Instead, the increased permeability induced by N17Rac1 correlates with loss of intercellular junctions.

Although expression of V12Rac1 increased the permeability of quiescent HUVECs, it did not alter histamine-or thrombin-induced permeability (Fig. 2). Consistent with this, V12Rac1 did not prevent thrombin-or histamine-induced fragmentation of adherens junctions or stress fibre formation, and gaps were still observed between cells (Fig. 6G,H; and data not shown).

Expression of N17Cdc42 did not affect endothelial permeability (Fig. 2A), but we have previously shown that microinjection of N17Cdc42 protein prevents TNF-α-induced stress fibre formation and the appearance of intercellular gaps in HUVECs (Wójciak-Stothard et al., 1998). Similarly, microinjection of HUVECs with N17Cdc42 protein or infection with Ad-N17Cdc42 inhibited thrombin-induced contractility and reduced stress fibre formation (Fig. 7A,C,E). Ad-N17Cdc42 did not alter the distribution of F-actin or junctional proteins in quiescent, unstimulated HUVECs (data not shown), consistent with the lack of effect of N17Cdc42 on endothelial barrier function (Fig. 2A). In contrast, N17Cdc42 did not appear to affect histamine-induced stress fibre formation or alterations to adherens junctions (Fig. 7B,D,G,H). These results are consistent with a model where Cdc42 acts to enhance contractility and intercellular gap formation in response to thrombin. As the response of HUVECs to histamine differs from that to thrombin in that histamine does not induce cell contraction or the appearance of large intercellular gaps (Fig. 1B,C), this may explain why Cdc42 is not involved in histamine-induced changes to F-actin and adherens junctions.

Vasoactive agents such as thrombin and histamine rapidly induce vascular permeability, and this has been attributed to increased actomyosin contractility leading to the formation of intercellular gaps and/or to modification of intercellular junctions (van Hinsbergh, 1997; Lampugnani and Dejana, 1997). Rho, Rac and Cdc42 are prime candidates for intracellular signalling molecules regulating endothelial permeability, as they influence both actin cytoskeletal organization and the integrity of intercellular junctions. We have found that inhibition of Rho and Rac but not Cdc42 significantly affects thrombin- and histamine-induced endothelial permeability, and that this correlates with effects on intercellular junctions.

We have observed that inhibition of Rho using either C3 transferase or dominant-negative RhoA expression, or of Rho kinase using Y-27632, prevented thrombin-induced permeability, consistent with previous observations (Essler et al., 1998; Carbajal and Schaeffer, 1999). That Rho and Rho kinase are also required for increased permeability induced by histamine (our observations), Pasteurella toxin (Essler et al., 1998b), and oxidized low-density lipoprotein (Essler et al., 1999), demonstrates that Rho and Rho kinase are important generally for regulating endothelial permeability. Tight junctions play a central role in regulating permeability of epithelial and endothelial cell monolayers, and we demonstrate for the first time that inhibition of RhoA and Rho kinase prevents the loss of tight junctions induced by thrombin and histamine. The requirement for Rho and Rho kinase in mediating increased endothelial permeability also correlates with their involvement in stress fibre formation (Essler et al., 1998; Vouret-Craviari et al., 1998; Wójciak-Stothard et al., 1998; Ridley, 1999), and Rho may therefore play a dual role in endothelial cells, concomitantly regulating stress fibre formation and destabilizing intercellular junctions.

How Rho affects intercellular junctions is not known, but it is interesting that in epithelial cells inhibition of Rho reduces tight junction function (Nusrat et al., 1995; Jou et al., 1998,) whereas it preserves tight junction function in endothelial cells. This indicates that the regulation of tight junction integrity differs between epithelial cells and endothelial cells, similar to what has been observed with adherens junctions, where inhibition of Rho or Rac leads to loss of cadherins from junctions in epithelial cells but not in endothelial cells (Braga et al., 1999). Rho might directly affect tight and adherens junction assembly/disassembly by regulating the phosphorylation status of junctional proteins via Rho kinase. Rho could also affect intercellular junctions indirectly through its effects on actomyosin contractility. It has been suggested that Rho might affect tight junction structure and function in epithelial cells by altering the contractility of the cortical actin belt (Nusrat et al., 1995; Jou et al., 1998; Madara, 1998). In endothelial cells, Rho-mediated stress fibre assembly induced by thrombin and histamine would lead to increased tension exerted on junctional regions, and this could contribute to junction disassembly.

Both thrombin and histamine rapidly induce the assembly of stress fibres and focal contacts in HUVECs. Thrombin, however, induces the appearance of consistently larger intercellular gaps compared to histamine, and the staining of VE-cadherin is fragmented or even lost from intercellular junctions in thrombin-stimulated cells, whereas in histamine-stimulated cells interjunctional staining of VE-cadherin is less disrupted and cells do not retract. This is consistent with differences in endothelial cell responses to histamine and thrombin reported previously. For example, thrombin but not histamine was shown to increase isometric tension and contractility, and thrombin induces higher levels of myosin light chain (MLC) phosphorylation and causes a more prolonged increase in endothelial permeability than histamine (Boswell et al., 1992; Moy et al., 1993; Moy et al., 1996; van Hinsbergh, 1997). The molecular basis for these differences between thrombin and histamine responses is not known, but it is interesting in this respect that dominant-negative Cdc42 has no effect on histamine responses but reduces thrombin-induced stress fibre formation and prevents the appearance of intercellular gaps. This suggests that Cdc42 might contribute to the increased cell contractility observed in thrombin-compared to histamine-treated cells (Moy et al., 1996). One possibility is that Cdc42 enhances and/or prolongs the activation of Rho and thereby MLC phosphorylation. Indeed, introduction of constitutively activated Cdc42 induces stress fibre formation and gap formation in HUVECs, indicating that Cdc42 can induce Rho activation (Wójciak-Stothard et al., 1998). Cdc42 may also activate actomyosin-based contractility independently of Rho, as PAKs and myotonic-dystrophy related kinase MRCKα, which are downstream targets of Rac and Cdc42, can induce increased MLC phosphorylation (Leung et al., 1998; Kiosses et al., 1999; Sells et al., 1999). However, the fact that dominant-negative Cdc42 prevents gap formation between cells treated with thrombin yet has no effect on permeability supports a model where permeability can increase even when there are no microscopically detectable gaps between cells (Andriopoulou et al., 1999).

Rac1 has previously been shown to regulate the integrity of adherens and tight junctions in epithelial cells (reviewed in Kaibuchi et al., 1999), and it is interesting that in endothelial cells either inhibition or activation of Rac1 increases permeability. This is likely to be a consequence of changes to intercellular junctions rather than to stress fibres, as dominant-negative Rac1 and constitutively active Rac1 have opposing effects on stress fibres. With dominant-negative Rac1, although VE-cadherin is still localized to intercellular borders, as previously reported (Braga et al., 1999), TEM analysis reveals that junctional areas are often reorganized so that small areas of close contact are interspersed by large intercellular gaps, and far fewer tight junctions are present. This suggests that both adherens junctions and tight junctions in endothelial cells are weakened by expression of dominant-negative Rac1, as observed in epithelial cells (Takaishi et al., 1997; Braga et al., 1997; Jou et al., 1998). A reduction in tight junctions and adherens junctions is also observed with constitutively active Rac1, suggesting that Rac activity needs to be precisely controlled to maintain the integrity of endothelial cell junctions.

In summary, our observations suggest that in endothelial cells Rho activation is important for thrombin- and histamine-induced endothelial cell permeability and that this correlates with the ability of Rho to mediate disassembly of adherens and tight junctions. Although increased permeability generally correlates with increased stress fibre formation (van Hinsbergh, 1997), the fact that dominant-negative Rac1 enhances endothelial permeability yet blocks stress fibre formation indicates that the two responses are separable. Through its effects on endothelial permeability, the Rho signalling pathway represents a potential target for the treatment of pathological conditions involving the impairment of endothelial barrier function.

This work was supported by a European Community Concerted Action grant no. QLG1-1999-01036 and the British Heart Foundation. We are particularly grateful to Ritu Garg for purification of recombinant proteins, to Mark Turmaine (University College London, UK) for processing HUVECs for electron microscopy, to Laurence Pearl (Institute of Cancer Research, London, UK) for use of his fluorimeter, to Caroline Dent (GlaxoWellcome Research Centre, Stevenage, UK) for providing the p-gal adenovirus and for expert tuition in working with adenoviruses and to Welfide Corporation for providing the Rho kinase inhibitor, Y-27632. We thank Ruggero Pardi (DIBIT-Scientific Institute San Rafaele, Milan, Italy) for providing HUVECs for some experiments, and Peter Clarke, J. Anthony Firth (Imperial College, London, UK) and Paul Martin (University College, London, UK) for advice on electron micrographs.