Introduction
VEGF was first described as a potent vascular permeability factor (VPF) secreted by tumour cells that stimulates a rapid and reversible increase in microvascular permeability without mast cell degranulation or endothelial cell damage1, 2. This tumour-secreted VPF was later shown to be a highly selective and remarkably potent growth factor for endothelial cells3. Since then, VEGF has been shown to promote the migration, growth and survival of endothelial cells4. VEGF is essential for vasculogenesis (the establishment of a primitive network of blood vessels during embryonic development) and for the subsequent formation of new vessels from pre-existing ones, a process known as angiogenesis4, 5. In adults, deregulated expression of VEGF is involved in a variety of disease states, ranging from inflammation, oedema and thrombotic reactions to cancer and metastasis6, 7. VEGF expression is rapidly increased in hypoxic tissues, providing a mechanism to assist tissue re-oxygenation by stimulating angiogenesis8. This physiological process is often co-opted by tumour cells to build a new vascular network dedicated to supply oxygen and nutrients to the cancerous cells, thereby enabling them to proliferate and metastasize8. Remarkable progress has been made towards the elucidation of the intracellular signalling pathways by which VEGF promotes endothelial cell growth and survival5. However, the mechanism by which VEGF promotes endothelial-cell permeability and vascular leakage is still elusive.
Different studies have provided compelling evidence that the non-receptor tyrosine kinases of the Src family (SFK) are involved in the increase in permeability induced by VEGF, but not by inflammatory cytokines9, 10, 11. These findings may have direct clinical implications as, for example, stroke-induced vascular leakage through the brain–endothelium barrier can be prevented by the use of SFK inhibitors in mouse models of ischemia12. However, VEGF stimulation of endothelial-cell permeability through Src remains a poorly understood process. It is known that the VEGF receptor 2 (VEGFR-2) interacts with VE-cadherin13, 14, an endothelial specific cell–cell adhesion molecule15, and this complex is disrupted on VEGF stimulation by a Src-dependent mechanism11. Accordingly, genetic deletion of VE-cadherin, or inhibition of its adhesive function, results in increased vessel permeability13, 16, 17, whereas enhancing VE-cadherin-dependent adhesion can protect the integrity of endothelial barrier18. Thus, we hypothesized that the effects of VEGF on endothelial-cell permeability may result from the transient disruption of the VE-cadherin-based cell–cell adhesion.
Results
VEGF promotes the endocytosis of VE-cadherin
We investigated VE-cadherin localization in response to VEGF, taking advantage of a monoclonal antibody (BV6) which recognizes the extracellular domain of human VE-cadherin (hVE-cad; Fig. 1a and see Supplementary Information, Fig. S1). All cells displayed cell-surface staining that was highly sensitive to a mild acid wash (Fig. 1a and see Supplementary Information, Fig. S1). In contrast, VEGF stimulation led to the accumulation of an acid-resistant hVE-cad intracellular staining. These vesicle patterns were observed as early as 2 min after VEGF stimulation (data not shown), and in more than 50% of the cells after 30 min (Fig. 1b). After VEGF removal, internalized VE-cadherin relocalized to the cell surface and participated in cell–cell contact, and only a few sparse internal vesicles remained (see Supplementary Information, Fig. S1), suggesting that VE-cadherin endocytosis is reversible. The accumulation of these VE-cadherin-containing vesicles was independent of the cell confluence conditions, as the acid-resistant VE-cadherin intracellular staining was observed after VEGF stimulation of isolated cells (data not shown), subconfluent cells and in cells engaged in multiple cell–cell contacts within cell monolayers (Fig. 1a, b). The anti-hVE-cad uptake induced by VEGF could be prevented by pretreatment with a VEGFR-2 inhibitor (Fig. 1b), and VEGF increased the rapid internalization of VE-cadherin, but not N-cadherin, in an intracellular compartment that is protected from trypsin-induced degradation (Fig. 1c). Moreover, most of these hVE-cad antibody-containing vesicles colocalized with EEA1, dynamin II, Rab5 and clathrin as endosomal markers, but not with caveolin (Fig. 1d). These VE-cadherin-stained vesicles did not colocalize with cathepsin D, a lysosomal marker (see Supplementary Information, Fig. S2), suggesting that VEGF induces the endocytosis of VE-cadherin into a clathrin-containing endosomal compartment.
Figure 1: VEGF promotes VE-cadherin endocytosis.
(a) Human endothelial cells (HUVECs) were incubated with BV6 antibodies at 4 °C to block cellular trafficking. The VE-cad internalization was monitored by the uptake of BV6 antibodies at 37 °C and then visualized in fixed cells using secondary fluorescent antibodies (green). Membrane-bound BV6 antibodies, observed in no wash conditions, were removed by a mild acid wash before fixation, which preserved internalized antibodies (acid wash). Arrows point to hVE-cad in internal vesicle-like compartments. Confocal microscopy pictures of untreated cells (control) show the organized pattern of VE-cadherin staining in endothelial cell monolayers, and reveal a disorganized pattern of VE-cadherin at the plasma membrane and its internalization after VEGF treatment (30 min, 50 ng ml-1) in HUVECs under both subconfluent and monolayer culture conditions. Nuclei are shown in blue. The scale bars represent 10 
m. (b) The internalization of endogenous hVE-cad was quantified by the percentage of HUVECs cells showing internal acid-resistant vesicles, as described in methods, in control or VEGF stimulated cells, pre-incubated or not with su1498 (VEGFR-2 inhibitor) under subconfluent or monolayer conditions. ANOVA test on three independent experiments: n >450 cells; three asterisks indicate P <0.001, with respect to unstimulated cells. (c) Monolayers of mouse endothelial cells (SVECs) were starved overnight and then treated with VEGF. At the indicated times, cells were treated with trypsin, pelleted and lysed for further analysis of VE-cadherin (VE-cad), N-cadherin (N-cad) and 
-catenin (-cat) contents present in these trypsin-resistant fractions. Total cell lysates (TCL) without trypsin pretreatment were used as controls. (d) Acid-resistant labelling of VE-cadherin (internal hVE-cad, green) in HUVECs after VEGF stimulation (30 min), further stained for EEA1 (early endosome antigen 1; a selected enlargement of this image is shown below), dynamin II, caveolin, Rab5b and clathrin, as indicated (all in red). Colocalization is displayed in yellow in these overlay confocal microscopy sections (arrows). The scale bars represent 10 m. All scanned films and confocal microscopy images are representative of 3–5 independent experiments.
VEGF promotes VE-cadherin endocytosis by regulating Vav2 through Src
We have previously observed that Vav2, a GEF, could be phosphorylated in response to platelet-derived growth factor (PDGF) through Src19. As PDGF and VEGF receptors share structural and functional similarities20, we asked whether Vav2 could be an integral component of VEGFR-2–SFK signalling9, 11. VEGF stimulation results in the enhanced tyrosine phosphorylation of Vav2, together with Src and VEGFR-2, which was abolished by VEGFR-2 and SFK inhibitors (Fig. 2 and see Supplementary Information, Figs S2,S3,S4,S5). The role of Src in Vav2 phosphorylation downstream of VEGFR-2 was further confirmed by the knockdown of Src by short interfering RNA (siRNA). Under these conditions, phospho-Src was barely detectable and Src siRNA prevented the enhanced Vav2 tyrosine phosphorylation on VEGF stimulation (Fig. 2a). These results were supported by the opposite effects on VEGF-dependent Vav2 phosphorylation provoked by the expression of Src dominant negative (YF/KM) and active (YF) mutants (see Supplementary Information, Fig. S3). Furthermore, the blockade of VEGFR-2 and SFK by pharmacological inhibitors, and the inhibition of Src by the expression of its dominant-negative mutant or its specific knockdown with siRNA, diminished the ability of VEGF to induce endothelial permeability, whereas the expression of a Src active mutant enhanced endothelial permeability under basal conditions (Fig. 2b). These data prompted us to assess the contribution of Vav2 to VEGF-induced permeability. Short hairpin RNAs (shRNAs) that decreased Vav2 protein-expression levels strongly reduced VEGF-induced permeability, compared with mock or control shRNA-infected SV-40 immortalized mouse vascular endothelial cells (SVECs; Fig. 2c, d). Similar results were obtained by the expression of a dominant-negative form of Vav2 (ref. 19 and data not shown). Thus, Vav2 may represent a biologically relevan t component of the signalling pathway by which VEGF stimulates endothelial permeability through Src.
Figure 2: VEGF induces the endocytosis of VE-cadherin by phosphorylating Vav2 through Src.
(a) SVECs were treated with DMSO, su1498, or su6656 30 min before VEGF stimulation (10 min) and lysed for western blot analysis of total Src and Vav2, as well as activated Src, using anti-phospho-Tyr 418 antibodies (P-Src) in total cell lysates (TCL), and Vav2 tyrosine phosphorylation (PY) in Vav2 immunoprecipitates (IP). Similar analysis in control and Src siRNA (50 nM) transfected SVECs, 3 days post-transfection is also shown. (b) FITC–dextran permeability was determined in mature SVEC monolayers in cells pretreated with the anti-hVE-cad antibody (BV6), su1498 and su6656 30 min before VEGF treatment, or in cells transfected with SrcYF, SrcYF/KM and Src siRNA. FITC–dextran permeability is expressed as fold increase 
s.e.m. with respect to untreated cells (control). (c) Vav2 levels were analysed 5 days after SVEC infection with lentiviruses containing mock (-), control sequence or Vav2 specific shRNAs sequences sh1 and sh2. Tubulin served as a loading control. (d) Monolayers of virally infected cells were used for permeability assays with or without VEGF treatment (30 min). (e) SVECs expressing hVE-cad were used to specifically monitor its endocytosis using BV6 antibody, in control vector or active Src mutant (SrcYF) transfected cells. Alternatively, SVECs were infected with Vav2 shRNA lentiviruses (sh2) and transfected with hVE-cad 4 days later. Cells were processed as described in the Methods, but the acid-wash step was avoided to reveal all hVE-cad-transfected cells. Arrows indicate VE-cad in vesicular structures. hVE-cad, green; nuclei, blue. The scale bars represent 10 m. (f–g) The number of cells exhibiting vesicle-like staining was quantified in su6656 treated-cells, Src siRNA transfected cells, negative (YF/KM) and active (YF) Src mutant-expressing cells, and in mock (-), control sh, and Vav2 sh1 and sh2 lentivirus infected cells, and expressed as the percentage (mean s.e.m.) with respect to control cells. All scanned films and confocal microscopy pictures are representative of 3–5 independent experiments. ANOVA test was on three independent experiments. The double asterisk indicates P <0.01 with respect to corresponding unstimulated control cells (-, white bars) or VEGF-stimulated cells (VEGF, black bars).
We then explored whether the Src–Vav2 pathway participates in VE-cadherin endocytosis in response to VEGF. Similarly to endogenous hVE-cad in human umbilical vein endothelial cells (HUVECs), VEGF induced the internalization of transfected hVE-cad in SVECs (Fig. 2e) into a clathrin, Rab5 and EEA1-containing compartment (data not shown). The SFK inhibitor, Src siRNA and the negative mutant of Src blocked the VEGF-induced VE-cadherin internalization, whereas the active form was sufficient to enhance the basal level of VE-cadherin-containing vesicles (Fig. 2e, f). Furthermore, the knockdown of Vav2 protein levels prevented VEGF from inducing VE-cadherin endocytosis (Fig. 2e–g). These findings suggested that VEGF may regulate endothelial permeability and VE-cadherin endocytosis through the sequential activation of VEGFR-2, Src and Vav2.
VEGF induces VE-cadherin endocytosis through Rac
Because we observed that Vav2 is involved in VEGF-induced permeability, we investigated whether this GEF functions in the activation of Rho GTPases19. Rac1 and Cdc42 were transiently activated by VEGF, but RhoA was not (see Supplementary Information, Fig. S3). The expression of dominant-negative form of Src (see Supplementary Information, Fig. S3,S4,S5) or the pretreatment of SVECs by inhibitors of VEGFR-2 and SFK strongly reduced the level of GTP-bound Rac (Fig. 3a). In agreement with these observations, the active form of Src raised the basal level of GTP-bound Rac (see Supplementary Information, Fig. S3). Moreover, when Vav2 protein expression was reduced by shRNA, the VEGF-dependent Rac activation was lost (Fig. 3b). Based on these results, and the fact that Rac activation regulates cell–cell junctions21, 22, we next explored whether Rac functions in VEGF-triggered VE-cadherin endocytosis using both dominant-negative approaches (RacN17 and wild-type PAK-N) and RNA interference (RNAi; Fig. 3c). The expression of RacN17, wild-type PAK-N and the knockdown of Rac1 blocked VEGF-induced hVE-cad internalization, whereas expression of a constitutively active form of Rac (RacQL), was sufficient to promote the internalization of hVE-cad into intracellular vesicles (Fig. 3d).
Figure 3: A highly conserved SVR motif in human VE-cadherin is a potential target for PAK phosphorylation.
(a) Rac activation was determined by GST-pulldown in control SVECs or stimulated by serum (10 min), or by VEGF (+, 5 min) in cells pretreated for 30 min with su1498 or su6656. (b) Rac activation (Rac–GTP) in response to VEGF (+, 5 min) was examined in SVEC infected with mock (-), control shRNA, or Vav2 shRNA sequences 1 or 2, 5 days after infection. Anti-Vav2 and anti-Rac western blots were performed on total cell lysates, and anti-Rac in pulldown fractions. (c) SVECs were transfected with Rac siRNA (25 nM) and analysed by western blot 3 days later. Tubulin served as a loading control. (d) The number of cells exhibiting internal vesicle-like staining was quantified in SVECs transfected with hVE-cad expressing either RacQL, RacN17, PAK-NL2, wild-type PAK-N (PAK-N WT), or previously transfected by Rac siRNA and expressed as the percentage (mean s.e.m.) with respect to control (-) cells. (e) The human VE-cadherin (amino acids 621–674) was compared with human E-cadherin and N-cadherin, and to simian, bovine, porcine, mouse, rat and chicken VE-cadherins. Identical amino acids are shown in red, similar in blue. (f) Coomassie staining of purified bacterially expressed His–ICD of hVE-cad wild-type or hVE-cadS665D or its mock Ni-based purification (-). The same preparations were subjected to PAK phosphorylation in presence of 32P-
-ATP. (g) HUVECs were pretreated for 30 min with control (DMSO <0.1%), su1498, su6656 before VEGF stimulation (10 min) and lysed for western blot analysis of phospho-Ser 665 VE-cadherin (pS665) and VE-cad signals. (h) SVECs were transfected with the auto-inhibitory domain of PAK (PID–GFP) or with GFP for 24 h, starved and treated with VEGF (+, 10 min). TCL were blotted against GFP, phospho-Ser 665 VE-cadherin (pS665) and VE-cad. (i) Quantification of the number of cells exhibiting vesicle-like staining, assayed for anti-hVE-cad antibody, BV6, uptake and expressed as the mean percentage of transfected cells s.e.m. in SVEC transfected with hVE-cad and expressing PID–GFP or GFP. All scanned images are representative of at least three independent experiments. ANOVA test on three independent experiments: double asterisk indicates P <0.01, with respect to corresponding unstimulated control cells (-, white bars) or VEGF-stimulated cells (VEGF, black bars).
A highly conserved serine residue within VE-cadherin represents a potential target for PAK
To examine the molecular mechanisms linking VEGF stimulation to VE-cadherin endocytosis in more detail, we tested whether its intracellular domain (ICD), which is involved in VE-cadherin trafficking23, contributes to VE-cadherin internalization. Amino-acid sequence analysis of the hVE-cad ICD revealed the presence of an area adjacent to the p120 binding region that is highly conserved among VE-cadherins, but distinct from that of other classical cadherins, such as E- and N-cadherins (Fig. 3e). This conserved region included a serine–threonine cluster upstream from a Ser–Val–Arg motif, which may represent a putative phospho-acceptor signal for Ser–Thr kinases (http://ca.expasy.org/prosite). Therefore, we evaluated the role of PAK, a Ser–Thr kinase that acts downstream from Rac. Kinase assays showed that PAK1 could phosphorylate in vitro the recombinant hVE-cad ICD but not its mutant lacking the Ser 665 residue (Fig. 3f). To study whether VEGF can trigger phosphorylation of VE-cadherin at this residue in vivo, we raised a phospho-specific antibody against the phospho-SVR motif (see Supplementary Information, Fig. S4) and observed that the pSer 665-VE-cad signal was specifically enhanced in HUVECs in response to VEGF, in a VEGFR-2- and Src-dependent fashion (Fig. 3g). We then expressed the auto-inhibitory domain of PAK1 (PID), which acts as a dominant-negative for PAK24 to address the participation of PAK in this VEGF-dependent phosphorylation of VE-cadherin. PID expression did not affect upstream VEGF signalling, as determined by Src activation (data not shown), but blocked the VEGF-induced increase of phospho-Ser 665 VE-cadherin (Fig. 3h). Furthermore, PID expression, which inhibited VEGF-induced permeability24, also reduced the internalization of VE-cadherin (Fig. 3i). Thus, our data strongly suggest that PAK is involved in the VEGF-mediated VE-cadherin endocytosis, most likely through the phosphorylation of VE-cadherin at Ser 665.
The highly conserved Ser 665 of hVE-cadherin regulates VEGF-dependent endocytosis
To address whether Ser 665 phosphorylation could participate in the regulation of the endothelium barrier function, hVE-cad mutants that were non-phosphorylable (S665V) or that mimicked the phosphorylated state (S665D) were engineered. On VEGF stimulation wild-type hVE-cad was found in internal vesicles, whereas hVE-cadS665V remained mostly at the cell borders (Fig. 4a, b). In contrast, hVE-cadS665D seemed to be internalized spontaneously, independently of VEGF signalling (Fig. 4a, b). Thus, mimicking Ser 665 phosphorylation may be sufficient to induce hVE-cad internalization. N-cadherin–GFP was used as a control and did not show internal vesicles on stimulation with VEGF, nor did it affect the VEGF-induced permeability, indicating that this response is cadherin-subtype specific (Fig. 4a–c). Moreover the VEGF-induced permeability was significantly reduced, albeit not completely, in hVE-cadS665V-expressing cells, whereas cells expressing the hVE-cadS665D mutant exhibited higher permeability, even in the absence of VEGF (Fig. 4c). To address the contribution of hVE-cad Ser 665 phosphorylation to endothelial-cell permeability in the absence of the influence of endogenous VE-cadherin, VE-cadherin was knocked down in SVECs and then reconstituted to almost basal levels using wild-type hVE-cad and its mutants (Fig. 4d). Interestingly, the reduction of approximately 80% of VE-cadherin levels disrupted the barrier function of endothelial monolayers (Fig. 4d, e), as previously suggested25. In addition, the reconstitution experiments confirmed the critical role of Ser 665 in the regulation of VEGF-dependent endothelial permeability (Fig. 4e). Finally, the availability of the S665D mutant enabled investigation of the interplay between VE-cadherin endocytosis and the organizational stability of the endothelial tight junctions, as judged by the pattern of ZO-1 distribution15, 26. As a control, the expression of wild-type hVE-cad did not affect the organized pattern of ZO-1 staining in resting cells, nor its re-organization on VEGF stimulation (Fig. 4f). However, in the absence of VEGF stimulation hVE-cadS665D-expressing cells already displayed a disorganized pattern of ZO-1, which was also perturbed in adjacent endothelial cells but retained in distant cells that served as internal controls (Fig. 4f).
Figure 4: Ser 665 regulates VE-cadherin endocytosis and VEGF-induced permeability.
(a) SVECs were transfected with wild-type, hVE-cadS665V, hVE-cadS665D and subjected to internalization assays as described above. Confocal microscopy analysis showed the localization of BV6 staining, recognizing human VE-cadherin, at the cell surface and/or in vesicles in control or VEGF stimulated (30 min) cells. Alternatively, SVECs were transfected with chicken N-cadherin–GFP (cN-cad–GFP) and localization of the GFP-fusion protein was assessed by confocal microscopy analysis in control or VEGF treated cells. Green, hVE-cad or cN-cad–GFP; nuclei, blue. The scale bars represent 10 m. (b) Quantification of the number of cells exhibiting vesicle-like staining and expressed as the mean percentage of transfected cells s.e.m. in SVEC transfected with wild-type, hVE-cadS665V, hVE-cadS665D or N-cad. (c) FITC–dextran permeability was determined in starved 2 day-old SVEC monolayers in cells transfected with wild-type, hVE-cadS665V, hVE-cadS665D or N-cad–GFP. Unstimulated wild-type hVE-cad transfected SVECs were used to normalize the FITC–dextran fluorescence intensity. (d) Mouse SVECs were transfected with non-silencing (control) or mVE-cadherin siRNA on day 0, then transfected on day 1 with siRNA and mock (–),wild-type, hVE-cadS665V, hVE-cadS665D plasmids. On day 2, cells were distributed for permeability assays or western blot analysis against pan-VE-cadherin, hVE-cad or -catenin as a loading control. (e) Cells were treated as in d and allowed to form a nonpermeable monolayer for 1 day more before evaluation of FITC–dextran passage, as described above. (f) SVECs were transfected with wild-type and hVE-cadS665D and subjected to internalization assay, as described above. Acid wash revealed the internal hVE-cad (green) and tight junctions are visualized by ZO-1 staining (red) in overnight starved cells (control) or stimulated by VEGF for 30 min (VEGF). DAPI staining (blue) is shown in the insert for each panel. The scale bars represent 10 m. All confocal microscopy images and scanned blots are representative of three independent experiments. In each case, ANOVA test was performed on three independent experiments: n >300 cells; three asterisks represent P <0.001, and a single asterisk represents P <0.05, with respect to corresponding unstimulated control cells (-, white bars) or VEGF-stimulated cells (VEGF, black bars).
-arrestin2 mediates the endocytosis of VE-cadherin and regulates VEGF-induced permeability
We next explored how the VEGF-dependent Ser 665 phosphorylation of VE-cadherin can lead to its internalization. The best example of phospho-serine–threonine-targeted endocytosis is provided by the ligand-dependent internalization of phosphorylated G-protein coupled receptors (GPCRs), as previously shown for the -adrenergic receptor27, on their interaction with -arrestins. The localization of -arrestin2–GFP in endothelial cells was evaluated. VEGF induced a dramatic change in -arrestin2–GFP distribution, which accumulated into vesicle-like clusters within few minutes after the addition of VEGF (Fig. 5a). Unexpectedly, VEGF induced clustering of -arrestin2–GFP together with hVE-cad-containing vesicles (Fig. 5a). Confocal microscopy analysis showed a very limited corecruitment of VEGFR-2, together with internal VE-cadherin and -arrestin2–GFP, but VEGFR-2 did partially colocalize with Src–GFP (see Supplementary Information, Fig. S4), suggesting that the VEGF-induced clustering of -arrestin may be associated with VE-cadherin-containing vesicles rather than VEGFR-2-containing vesicles. The expression of hVE-cadS665V reduced the VEGF-dependent -arrestin2 redistribution. In contrast, a strong clustering of -arrestin2 was observed when hVE-cadS665D was expressed, which was independent of VEGF signalling (Fig. 5b, c) and thus, suggested that the redistribution of -arrestin2–GFP induced by VEGF stimulation may require the integrity of the Ser 665 site.
Figure 5: A role for Ser 665 in VEGF induced coclustering of 
-arrestin2 with VE-cadherin.
![Figure 5 : A role for Ser 665 in VEGF induced coclustering of |[beta]|-arrestin2 with VE-cadherin.](/ncb/journal/v8/n11/thumbs/ncb1486-f5.jpg)
(a) SVECs were cotransfected with -arrestin2–GFP together with human VE-cadherin (hVE-cad) and studied for BV6 antibody uptake after VEGF stimulation. Samples were further analysed by confocal microscopy for GFP (green) and VE-cadherin (red). Nuclei are shown in blue in the overlaid images. -arrestin2–GFP was widely distributed in the cytosol and small intracellular vesicles in unstimulated endothelial cells, without any particular accumulation at cell–cell contacts. In contrast, on stimulation with VEGF, -arrestin2–GFP was concentrated in vesicle-like clusters. The scale bars represent 10 m. (b) Confocal microscopy analysis of SVECs cotransfected with -arrestin2–GFP together with wild-type hVE-cad, hVE-cadS665V or hVE-cadS665D. VEGF-induced -arrestin2–GFP clustering occurred within less than 5 min of VEGF stimulation, in mock (data not shown) or hVE-cad transfected SVECs. The scale bar represents 20 m. (c) Quantification of -arrestin2–GFP clustering in vesicle-like structures in wild-type hVE-cad, hVE-cadS665V or hVE-cadS665D transfected SVECs in response to VEGF stimulation. ANOVA test on three independent experiments: n >250 cells; double asterisk represents P <0.01, with respect to corresponding unstimulated control cells (-, white bars) or VEGF-stimulated cells (VEGF, black bars). All confocal microscopy images are representative of three independent experiments.
We next examined whether VE-cadherin interacts with endogenous -arrestin2, the most abundant arrestin isoform in endothelial cells at the RNA level (data not shown). VE-cadherin coimmunoprecipitated with VEGFR-2, and VEGF stimulation reduced their association, as previously reported11, 13 (Fig. 6a). In contrast, -arrestin2 coimmunoprecipitated more efficiently with VE-cadherin following VEGF stimulation (Fig. 6a), indicating that VEGF stimulation enhances the association of VE-cadherin and -arrestin2. -arrestin2 was not found in complexes containing VEGFR-2 (Fig. 6a). To investigate whether the -arrestin2–VE-cadherin interaction involves the Ser 665 phospho-acceptor site, we examined whether the S665V mutant interacts with -arrestin2. The association of -arrestin2 with the non-phosphorylable VE-cadherin mutant was much more limited than the association with wild-type hVE-cad in response to VEGF (Fig. 6b). As this coimmunoprecipitation may be caused by the cis-dimerization between the hVE-cad mutant and the endogenous mVE-cad, we assessed the ability of recombinant wild-type, hVE-cadS665V and hVE-cadS665D ICD to pulldown -arrestin2–GFP in vitro. Wild-type and hVE-cadS665V ICD barely pulled down -arrestin2, and this interaction was enhanced if the hVE-cad ICD was first phosphorylated in vitro by PAK1 (Fig. 6c). In contrast, hVE-cadS665D ICD could affinity precipitate -arrestin2 effectively, even without prior PAK1 phosphorylation (Fig. 6c). Moreover, -arrestin2 did not associate with hVE-cad or hVE-cadS665V when overexpressed in HEK-293T cells, which do not express endogenous VE-cadherin, but associated spontaneously with hVE-cadS665D (Fig. 6d). Thus, these results are in agreement with the hypothesis that VEGF induces the recruitment of -arrestin2 on the phospho-SVR motif within the VE-cadherin ICD.
Figure 6: -arrestin2 is involved in VEGF-induced VE-cadherin endocytosis and permeability.
![Figure 6 : |[beta]|-arrestin2 is involved in VEGF-induced VE-cadherin endocytosis and permeability.](/ncb/journal/v8/n11/thumbs/ncb1486-f6.jpg)
(a) HUVECs were plated at confluence for 2 days and starved for 4 h before stimulation with VEGF (+, 10 min). Total cell lysates, in addition to goat or rabbit immunoglobulin controls (IP, -), VE-cadherin (IP, VE-cad) or VEGFR-2 (IP, VEGFR-2) immunoprecipitates were subjected to western blot analysis against VEGFR-2, VE-cad or -arrestin2. (b) Murine SVECs were transfected with wild-type or hVE-cadS665V and subjected to human VE-cadherin (hVE-cad) immunoprecipitations 24 h later. Immunoprecipitations were further analysed by western blot for -arrestin2 and hVE-cad. (c) Ni beads conjugated to mock (-), recombinant His-tagged wild-type hVE-cad (WT), His–hVE-cad phosphorylated by PAK (WT*), His–hVE-cadS665D or His–hVE-cadS665V ICDs were used to pulldown lysates from HeLa cells. His pulldowns were analysed by western blot for -arrestin2 and the His tag. (d) Mock (-), wild-type, hVE-cadS665D or hVE-cadS665V transfected HEK-293T cells were used for immunoprecipitation with anti-VE-cad antibodies and then blotted for VE-cad or -arrestin2. (e) SVECs were mock (-) infected or infected with lentiviruses expressing control shRNA, -arrestin2 shRNA sh1 or sh2 lentiviruses. 
-arrestin2 expression levels were analysed 5 days later by western blot. Levels of -arrestin1 were used as loading controls and to assess shRNA specificity. (f) SVECs were infected with mock (-), control shRNA, -arrestin2 shRNA sh1 or sh2 lentiviruses and transfected with hVE-cad 4 days later. Internalization assay of hVE-cad using BV6 antibody was performed 1 day later and quantified as the ratio between the number of cells with internal vesicles and total BV6-positive cells. (g) Control shRNA and -arrestin2 shRNA sh1 or sh2 infected SVECs were subjected to FITC–dextran permeability assays. Unstimulated starved mock transfected cells (-, control) were used to normalize the relative fluorescence units. All scanned images are representative of three independent experiments. ANOVA test on three independent experiments: double asterisk indicates P <0.01 with respect to corresponding unstimulated control cells (-, white bars) or VEGF-stimulated cells (VEGF, black bars).
To address whether -arrestin2 is required for VEGF-induced VE-cadherin endocytosis, we designed specific shRNA for the knockdown of -arrestin2. Lentiviral infections with two independent shRNAs effectively decreased -arrestin2 expression without affecting -arrestin1 levels (Fig. 6e), and prevented the internalization of hVE-cad on exposure to VEGF (Fig. 6f). The knockdown of -arrestin2 also protected against VEGF-induced permeability in endothelial monolayers (Fig. 6g). Thus, -arrestin2 is required for VEGF-induced permeability, most likely by interacting with the phosphorylated VE-cadherin ICD, thereby promoting the endocytosis of VE-cadherin.
Discussion
VEGF has a central role in both vasculogenesis and angiogenesis, and its deregulated expression contributes to the aberrant growth of solid tumours by promoting their neo-vascularization and also contributes to many other human diseases that are characterized by abnormal angiogenesis and enhanced vascular permeability7. Whereas recent studies have begun to uncover the underlying mechanisms by which VEGF promotes the proliferation, migration and survival of endothelial cells5, how VEGF induces vessel leakiness, its first described function1, 2, is still far from being fully understood. Here, we observed that VEGF regulates the availability of VE-cadherin at the cell surface by promoting its endocytosis through a VEGFR-2–Src–Vav2–Rac–PAK signalling axis. Furthermore, we found that the VEGF-induced internalization of VE-cadherin involves the phosphorylation of a conserved motif within the VE-cadherin ICD and the consequent recruitment of -arrestin2, which is best known for its involvement in the GPCR ligand-dependent endocytosis28, 29.
The architecture of endothelial junctions is rapidly modified in response to VEGF, allowing the extravasation of macromolecules and metastatic cells10. Whereas the barrier function of the endothelium is supported by different cell–cell adhesion systems (mostly adherens junctions involving VE-cadherin and tight junctions7, 15), the disruption of VE-cadherin-mediated adhesion is sufficient to disrupt these intercellular junctions13, 16, 17. In agreement with these observations, VE-cadherin is required to prevent the disassembly of the blood-vessel walls17, 30 and to coordinate the passage of macromolecules through the endothelium18, 25. In this regard, we observed that VEGF stimulation resulted in the rapid internalization of VE-cadherin in a clathrin-containing vesicular compartment, and that the expression of a VE-cadherin mutant (S665D) that is persistently internalized is sufficient to provoke the disruption of the tight junctions and to enhance endothelial cell permeability, even in the absence of VEGF stimulation. These findings raised the possibility that VEGF could cause the disruption of cell–cell junctions by promoting the localized removal of VE-cadherin from the cell surface by an endocytotic process.
Genetic and biochemical data support the requirement of Src kinase family for the vascular permeability response to VEGF7, 9. However, the precise mechanism by which VEGF receptors act through Src to promote permeability is not fully understood. For example, VEGF-induced Src activation can result in the destabilization of VEGFR-2–VE-cadherin complexes and the tyrosine-phosphorylation of junctional proteins10, 11, 31. Nevertheless, in some cases, tyrosine dephosphorylation rather than phosphorylation of VE-cadherin-associated proteins was observed on VEGF stimulation32, 33. Alternatively, a crosstalk between Src and the focal adhesion kinase can regulate vascular permeability by interfering with integrin adhesion and signalling34, 35. Our findings suggest that Src has an important function in linking VEGFR-2 activation to the stimulation of Vav2, thereby activating Rac and resulting in the endocytosis of VE-cadherin and the disruption of the endothelial junctions.
Recent studies have revealed that the serine–threonine kinase PAK, a direct downstream effector of Rac, regulates endothelial-cell permeability in response to VEGF and other cytokines and inflammatory mediators24. We observed that VEGF stimulation of endothelial cells leads to the rapid increase in PAK phosphorylation (data not shown) and that PAK-inhibitory mutants prevent the internalization of VE-cadherin caused by VEGF. Taken together, we can hypothesize that a variety of cell-surface receptors may converge to regulate Rac, thereby initiating a PAK-dependent increase in junctional permeability by promoting the endocytosis of VE-cadherin, and through its effects on the actin cytoskeleton and contractility. In this regard, we identified a specific motif within the cytoplasmic tail of VE-cadherin (Ser–Val–Arg (SVR) in position 665–667 of hVE-cad) that may represent a downstream target for PAK. Despite the high degree of homology between classical cadherins, this SVR motif is not found in N- or E-cadherins, although it is highly conserved in VE-cadherins from mammals to birds. Thus, although distinct species-specific VE-cadherin endocytic mechanisms may exist, the high conservation of this SVR motif may provide a shared structural feature by which growth factors may specifically promote VE-cadherin endocytosis in endothelial cells.
Based on the role of VEGF-dependent phosphorylation within the SVR motif, we explored whether -arrestin, which mediates the ligand-dependent endocytosis of GPCRs on serine-phosphorylation by GRKs28, 29, contributes to target VE-cadherin to the endosomal compartment. Surprisingly, we observed that VEGF stimulation leads to the association of VE-cadherin to endogenous -arrestin2, and that GFP–-arrestin2 colocalizes with internalized VE-cadherin, but not with VEGFR-2. In agreement with a key function for -arrestin2 in the endocytosis of VE-cadherin, the knockdown of -arrestin2 inhibits the effect of VEGF on both VE-cadherin endocytosis and endothelial permeability. Taken together, our results support the notion that the VE-cadherin SVR motif represents a docking site for -arrestin2 when it is phosphorylated by PAK. Of interest, -arrestin2 may also aid VE-cadherin endocytosis based on its ability to interact with Src36. In this scenario, -arrestin2 may recruit Src to the vicinity of VE-cadherin, thus facilitating the Src-dependent phosphorylation of cadherin–catenin complexes11, 37. This unexpected role for -arrestin in VE-cadherin endocytosis, together with the recently described internalization of Frizzled4 (ref. 38), TGF-R III (ref. 39), Smoothened40 and LDL-R41, may now broaden the repertoire of cell-surface molecules that can undergo -arrestin-dependent endocytosis (RADE), in addition to GPCRs.
The emerging picture from our study is that VEGF promotes the endocytosis of VE-cadherin through a novel signalling pathway that involves the Src-dependent activation of Vav2 by VEGFR-2 and the consequent activation of Rac. This, in turn, leads to the phosphorylation of VE-cadherin on a conserved SVR motif, likely by PAK, and the binding of -arrestin2 to this phosphorylated VE-cadherin (Fig. 7). This RADE mechanism triggers the internalization of VE-cadherin into clathrin-coated early endosomes and the consequent removal of VE-cadherin from the cell surface, resulting in the disruption of endothelial-cell junctions. Ultimately, these new insights into the basic molecular mechanisms by which VEGF induces vascular permeability through the endocytosis of VE-cadherin may have broad spectrum clinical implications, as they may help identifying novel therapeutic targets for the treatment of many human diseases that involve pathological vessel leakiness — including acute and chronic inflammation, tissue damage following stroke and myocardial infarction, diabetic retinopathy, macular degeneration, and even tumour-induced angiogenesis7.
Figure 7: Schematic representation of a model of the molecular mechanisms by which VEGF promotes VE-cadherin internalization and vascular permeability by VE-cadherin receptor -arrestin-dependent endocytosis (RADE) on the activation of the Vav2–Rac–PAK signalling pathway through VEGFR-2 and Src.
![Figure 7 : Schematic representation of a model of the molecular mechanisms by which VEGF promotes VE-cadherin internalization and vascular permeability by VE-cadherin receptor |[beta]|-arrestin-dependent endocytosis (RADE) on the activation of the Vav2|[ndash]|Rac|[ndash]|PAK signalling pathway through VEGFR-2 and Src.](/ncb/journal/v8/n11/thumbs/ncb1486-f7.jpg)
The endothelial barrier is maintained by the integrity of the adherens junctions through homophilic interactions between VE-cadherin expressed in adjacent endothelial cells and the interaction of VE-cadherin with the actin-based cytoskeleton through catenins. VEGFR-2 associates with VE-cadherin, and when activated by VEGF, this receptor dimerizes and causes the sequential activation of Vav2, Rac and PAK, through Src. This results in the serine phosphorylation of a conserved motif in the cytoplasmic tail of VE-cadherin (SVR 665–667) by PAK, which is likely coordinated with the tyrosine phosphorylation of the VE-cadherin–catenin complexes (dashed arrow) by Src. Serine-phosphorylated VE-cadherin recruits -arrestin2, which promotes the consequent internalization of VE-cadherin into clathrin-coated pits. Interestingly, the SVR motif is adjacent to the p120 binding region, which is conserved among all classical cadherins, thus raising the possibility that the association/dissociation of p120 with VE-cadherin may regulate the status of phosphorylation of the SVR motif within the VE-cadherin ICD and/or its interaction with -arrestin. This provides also a molecular framework by which tyrosine phosphorylation of p120 and VE-cadherin may contribute to the enhanced permeability caused by VEGF11, 31, 48, 50. Therefore, the tyrosine phosphorylation of VE-cadherin and its associated molecules may be coordinated with the Src-dependent activation of Vav2 and Rac to regulate the dynamic disassembly and reassembly of adherens junctions. This process leads to the disassembly of endothelial-cell junctions, resulting in the enhanced permeability of the blood-vessel wall. Endosomal VE-cadherin may be recycled to the cell surface, thus participating in the dynamic reorganization of adherens junctions during vessel remodelling.
Methods
Cell culture, transfections and siRNA.
SV-40 immortalized mouse vascular endothelial cells (SVECs)42, HeLa and human embryonic kidney (HEK) 293T cells were cultured in DMEM plus 10% foetal bovine serum (Sigma, St Louis, MO). HUVECs were grown in endothelial cell medium (EGM2; Cambrex, Walkwersille, MD). HEK-293T cells were transfected using LipofectAMINE (Invitrogen, Carlsbad, CA) or PolyFect (Qiagen, Santa Clarita, CA). SVECs were transfected by ExGen500 (Fermentas Inc., Hanover, MD), achieving an efficiency of about 40%, as judged by fluorescence from GFP-tagged expression vector. siRNA transfections were performed using the Hiperfect transfection reagent (Qiagen) and the following double-strand oligonucletides: mRac1 (CAGACAGACGUGUUCUUAAUUUGCU; Invitrogen), mSrc (CAGCACGAGGGUUGCCAUCAA; Qiagen) and mVE-cadherin designed-oligonucleotides (Santa Cruz Biotechnologies, Santa Cruz, CA). Transfection efficiency of siRNA was judged to be greater than 80% using fluorescent-labelled oligos.
Reagents and antibodies.
Recombinant vascular endothelial growth factor (VEGF; 50 ng ml-1) was from PropSpec-Tany TechnoGene Ltd (Rehovot, Israel). The inhibitors of Src family kinase su6656, and the inhibitor of VEGFR-2 su1498 (1 M and 5 M, respectively; Calbiochem, La Jolla, CA) were added 30 min before VEGF stimulation. The following antibodies were used: mouse anti-human VE-cadherin (BV6 clone; Research Diagnostics Inc., Flanders, NJ); mouse anti-GFP, HA and His (Covance Research Products Inc., Denver, CO); mouse anti-Rac1, clathrin heavy chain, EEA1, N-cadherin and caveolin (BD-Pharmingen, Palo Alto, CA); mouse anti-phospho-tyrosine (4G10 clone; Upstate Biotech, Waltham, MA); mouse anti-tubulin (Santa Cruz); rabbit anti-c-Src, Rab5b, VEGFR-2 and Vav2 (Santa Cruz); rabbit anti-Z01 (Zymed, Invitrogen); rabbit anti--catenin (Sigma); goat anti-VE-cadherin and dynamin II (Santa Cruz); rabbit anti-VE-cadherin (Cayman, Ann Arbor, MI); rabbit anti--catenin (Sigma); rabbit anti-VEGFR-2 and phosphor-Tyr 418-Src (Biosource QCB, Camarillo, CA): and rabbit -arrestin1 and -arrestin2 antibodies43. The rabbit phospho-Ser 665-antibody was engineered and purified according to the Abgent Phospho-Antibody Pack, using the following KHL-conjugated peptide sequence as immunogen: CDVSVLN(pS)VRRGG (Abgent, San Diego, CA).
Plasmids, DNA constructs and shRNA lentivectors.
pCEFL–AU5–RacQL, pCEFL–AU5–RacN17, pCEFL–AU1–PAKN, pCEFL–AU1–PAKNL2 were previously described42, as well as pCEFL–c-SrcYF (active from), pCEFL–c-SrcYF/KM (inactive form)19, pEGFP–N-cadherin44, pIRES–HA–VEGFR-2 (ref. 45) and pEGFP–Src46. pCEFL–GFP–-arrestin2 was generated by cloning the coding sequence of -arrestin2 downstream from GFP in the pCEFL expression vector. The 83–149 amino-acid fragment of PAK1 (termed the PAK inhibition domain, PID), which inhibits autophosphorylation and substrate phosphorylation, was cloned in pCEFL–GFP. The coding sequence for human VE-cadherin (hVE-cad, cadherin-5) was cloned in the pCEFL expression vector from EST clone (ID5299873, Invitrogen) and pcDNAI–hVE-cad47. Point mutations of Ser 665 to valine (S665V) or aspartic acid (S665D) were introduced in hVE-cad nucleotide sequence using the Quick-ChangeII XL-site-directed mutagenesis kit (Stratagene, La Jolla, CA). The wild-type (WT), S665D and S665V mutants of the hVE-cad ICD (starting position, nucleotide 1860) were cloned into the pRSETA–His bacterial expression plasmid. MouseVav2 shRNAs were prepared in pENTR-shRNA plasmid derived from pShag–Magic2 using the following nucleotide sequences: Number 1 starting position 586, ctcctcaaagtctgccatgaca; Number 2 starting position 2443, ctcgccatcagcatcaagttca; and control (c) starting position 529, cgacatcaacttccgccg; the latter as an inefficient knockdown sequence. To knockdown mouse -arrestin2, the same strategy was used with the following nucleotide sequences: Number 1 starting position 85, cagctcaccgtgtacttgggca; Number 2 starting position 248, cgggcctgtctttccgcaaaga; and control (c) starting position 527, cgcggcttatcatcagaaaggt. shRNA sequences were transferred from pENTR-shRNA into pLentiGW–EF using the LR clonase kit (Invitrogen) and lentiviruses were collected from HEK-293T supernatants 2.5 days after transfection with shRNA expression vector pLentiGW–EF, the envelope protein encoded by pVSV, and the packaging vector pSPAX2.
Permeability assay.
Permeability was assessed by the passage of FITC-conjugated dextran (relative molecular mass, 40,000 (Mr 40K; 1 mg ml-1; Molecular Probes) as previously described18. Briefly, 1 
105 SVECs were plated onto 6.5 mm Transwell Collagen-coated 3 m pore PTFE membrane inserts (Corning Costar Corp, Acton, MA) and left for 2 days to form mature monolayers (see Supplementary Information, Fig. S1). When required, pools of 1 106 cells were first transfected and then distributed equally in inserts, and cell lysates were prepared in parallel for western blot analysis. Each sample from the bottom chamber was read in triplicate on GENios fluorescent plate reader (Tecan, Durham, NC) or on Victor3 V, 1420 multilabel counter (Perkin-Elmer, Wellesley, MA). All data were from at least three independent experiments and statistical analysis was performed using Prism 4.2 software. Statistical significance was depicted as a single asterisk for P <0.05, a double asterisk for P <0.01 and a triple asterisk for P <0.001. When P >0.05, data were considered not significantly different.
Immunostaining and internalization assay.
Cells were grown onto collagen-coated coverslips and treated as previously described44. The donkey anti-mouse and anti-rabbit FITC or TRITC-conjugated immunoglobulins (Ig) were used as secondary antibodies (Jackson ImmunoResearch, West Grove, PA), as well as goat anti-mouse Ig G2a, G1, G2b AlexaFluor 488 or 546-conjugated antibodies (Molecular Probes, Invitrogen). Samples were visualized under conventional microscope (AxioPlan 2, Zeiss Imaging Inc., Thornwood, NJ) for cell counting. Confocal microscopy acquisitions were performed on a TCS/SP2 Leica microscope (NIDCR confocal service, Bethesda, MD). All pictures are representative of 3–5 independent experiments. hVE-cadherin internalization was adapted from a previously described protocol48. A schematic representation of the protocol is shown in the Supplementary Information, Fig. S1. The number of cells exhibiting at least one group of five or more acid-resistant VE-cadherin-positive vesicle-like structures was counted (n >300). Results are expressed as the mean s.e.m. of three independent experiments, and statistical analysis was performed using Prism 4.2 software.
PAK1 kinase assay.
His-tagged ICDs of either wild-type hVE-cad, hVE-cadS665D were expressed in BL21 bacteria and purified in native conditions using His MicroSpin purification kit from Amersham (Piscataway, NJ). Purified protein (20 g) was mixed for 30 min at 30 °C in kinase buffer (50 mM Hepes at pH 7.5, 1 mM EDTA, 1 mM DTT, 10 mM MgCl2, 10 mM NaF, 2 mM MnCl2) with 1 g PAK1 (Cell Signaling, Beverley, MA), 100 M cold ATP, and 10 Ci of adenosine 5'-32P-triphosphate (Amersham). Reactions were stopped by adding prewarmed Laemmli buffer and samples were subjected to SDS–PAGE. Films were exposed onto dry gels for 6 h to overnight at room temperature.
Immunoprecipitations, GST and His pulldown assays.
Immunoprecipitation experiments were performed as previously described19, using 20 L of rabbit anti-Vav2, rabbit anti-VEGFR-2 or goat anti-VE-cadherin antibodies at 4 °C overnight. The activation of Rac and Cdc42 was assessed by affinity precipitation of the GST-bound form of these GTPases using the GST–CRIB domain of PAK1 as previously described19. HeLa cells were lysed in immunoprecipitation buffer, and post-nuclei supernatants were precleared with 100 l of ProBond Resin (Invitrogen) for 45 min before incubation for 2 h with 20 g nickel mock-treated resin or bound to His-tagged wild-type, S665V and S665D hVE-cad ICD, or His-tagged wild-type hVE-cad ICD that was previously subjected to in vitro PAK phosphorylation. After 2 washes in immunoprecipitation buffer and one wash in PNI20 buffer (20 mM phosphate, 0.5 M NaCl at pH 7.4, 20 mM imidazole; Amersham), nickel-conjugated beads were boiled in Laemmli buffer for western blot analysis. All scanned films are representative of at least three independent experiments.
Trypsinization experiments.
To distinguish cell surface and intracellular pools of VE-cadherin, SVECs were rinsed and incubated in trypsin–EDTA at 37 °C to remove cell surface VE-cadherin, as previously described49. Trypsin was inactivated in serum-containing medium. Cells were recovered by centrifugation and pellets were lysed for western blot analysis. For controls, parallel cultures were lysed without trypsinization.
Accession numbers.
Human VE-cadherin, NM 001795; Mouse Vav2, NM 009500; mouse -arrestin2, NM 145429; human E-cadherin, NP_004351; human N-cadherin, NP_001783; simian (chimpanzee; Pan troglodytes) VE-cadherin, XP_523383; bovine (Bos taurus) VE-cadherin, NP_001001601; porcine (Sus scrofa) VE-cadherin, NP_001001649; mouse (Mus musculus) VE-cadherin, NP_033998; rat (Rattus norvegicus) VE-cadherin, XP_226213; and chicken (Gallus gallus) VE-cadherin, NP_989558.
Note: Supplementary Information is available on the Nature Cell Biology website.
Author Contributions
J.G. and J.S.G. planned the experimental design, analysed data and wrote the paper. J.G. conducted the experiments.
(v)
