RAF dimers control vascular permeability and cytoskeletal rearrangements at endothelial cell‐cell junctions

The endothelium functions as a semipermeable barrier regulating fluid homeostasis, nutrient, and gas supply to the tissue. Endothelial permeability is increased in several pathological conditions including inflammation and tumors; despite its clinical relevance, however, there are no specific therapies preventing vascular leakage. Here, we show that endothelial cell‐restricted ablation of BRAF, a kinase frequently activated in cancer, prevents vascular leaking as well metastatic spread. BRAF regulates endothelial permeability by promoting the cytoskeletal rearrangements necessary for the remodeling of VE‐Cadherin‐containing endothelial cell–cell junctions and the formation of intercellular gaps. BRAF kinase activity and the ability to form complexes with RAS/RAP1 and dimers with its paralog RAF1 are required for proper permeability control, achieved mechanistically by modulating the interaction between RAF1 and the RHO effector ROKα. Thus, RAF dimerization impinges on RHO pathways to regulate cytoskeletal rearrangements, junctional plasticity, and endothelial permeability. The data advocate the development of RAF dimerization inhibitors, which would combine tumor cell autonomous effect with stabilization of the vasculature and antimetastatic spread.


Introduction
A functioning vascular barrier is vital for many physiological processes, such as tissue-fluid homeostasis, vascular tone, or angiogenesis [1]. Endothelial cellcell junctions are the gatekeepers of the vascular barrier, and their tight regulation is crucial for vascular function in both physiological and pathological conditions [2]. Permeability-inducing factors secreted during inflammation or tumorigenesis not only cause the efflux of protein-rich fluid (edema) characteristic of inflammation but also the extravasation of leukocytes tasked with combating an infection [3] or of tumor cells on their way to form distant metastasis [4]. These processes take place at the level of the microvasculature, where the permeability-inducing factors locally weaken the junctions between endothelial cells by coordinated regulation of cell-cell adhesion molecules, such as VE-Cadherin, and cytoskeletal rearrangement [5] through pathways including Src, RHO-GTPase, or calcium signaling [1]. As an example, vascular endothelial growth factor (VEGF), which plays a central role in both tumor angiogenesis and vessel permeability [6,7], induces endothelial permeability through PLC-dependent calcium release [8], by Src kinasedependent phosphorylation and internalization of VE-Cadherin [2,9] and by AKT/eNOS/p190RHO-GAP (GTPase Activating Protein)-dependent RHOA GTPase activation [10]. RHO signaling also plays a key role in the induction of vascular permeability by histamine, a crucial event in allergic reactions [11] and by thrombin, which causes prolonged hyperpermeability during inflammation [12,13]. Activation of the RHO pathway by these stimuli affects F-actin quantity and actomyosin contractility, leading to the formation of radial stress fibers (RSF) associated with junctional plasticity and intercellular gap formation. In contrast, circumferential actin bundles (CABs) strengthen cellular junctions [5,14,15] and must dissolve to allow their remodeling. RAP1, activated via the cAMP-inducible GEF (Guanine nucleotide Exchange Factor) EPAC (Exchange Protein directly Activated by cAMP), prevents CAB disruption; permeability-inducing agents such as thrombin reduce cAMP levels [16], promoting CAB weakening. Thus, induction of permeability requires fine-tuning of RAP1 and RHO pathways, both of which must be dimmed at the junctions to allow gap formation. Simultaneously, RHO activity must increase along the RSF, at least partially through RHO GEF relocalization.
Downstream of growth factors, the RAS/RAF/ MEK/ERK pathway regulates cell proliferation, migration, and survival [17]. While homo-and heterodimerization of RAF proteins is crucial for the activation of the MEK/ERK module, RAF1 is capable of modulating parallel signaling pathways by binding and inhibiting other serine/threonine kinases, including ASK1 and ROKa. RAF1 promotes endothelial cell (EC) survival, mainly through ASK1 [18][19][20] and regulates adherens junction (AJ) dynamics, through RAP1dependent binding to ROKa [21]. However, the role of RAFs in endothelial permeability has not been investigated.

Endothelial BRAF controls transendothelial resistance and paracellular permeability
We ablated Braf in endothelial cells by combining the VE-Cadherin-Cre (VEC-Cre) transgene [22] with a homozygous Braf F/F allele [23]. Complete conversion of Braf flox to Braf D was confirmed by PCR (Fig. 1A). Braf D/D mice (deleted in ECs) were born at Mendelian objective. Whole mounts were stained with CD31 antibody to visualize endothelial cells. Scale bar represents 1 mm. The graphs show the distance of the angiogenic front from the central optical nerve head (left) and the distance between arteries and the capillary bed (right) in (n = 7) F/F and (n = 8) BRAF D/D retinas. The P value was calculated according to Student's t-test. (E) BRAF ablation does not influence the vascularization of subcutaneous Matrigel plugs containing FGF-2 and VEGF (1 lg each). Whole-mount plugs isolated from F/F (n = 5) and BRAF D/D (n = 4) mice were stained with CD31 antibody. CD31-positive areas were quantified and are plotted in the graph. (F) BRAF ablation does not impact in vitro sprouting in 3D fibrin gels. pMECs were allowed to adhere to microcarriers and embedded in fibrin gels containing FGF-2 and VEGF (200 ngÁmL À1 each). Each pMEC sample consists of a pool of three animals. The number of sprouts/beads and the length of sprouts were microscopically assessed after 3 days in culture. The bar graphs represent means AE SD of biological replicas (E) or technical replicates (F; n equals the number of microcarriers and sprouts evaluated). Scale bars represent 50 lm (E) or 100 lm (F). The P values were calculated according to Student's t-test. ratios (Fig. 1B), were fertile, and had a normal life span. We did not detect any anomalies in tissue architecture of Braf D/D kidneys, lungs, and livers (Fig. 1C).
Retinal angiogenesis proceeded slightly faster in the Braf D/D retinas than in controls; moreover, the distance between arteries and the capillary bed was comparable in Braf D/D and control retinas (Fig. 1D). Thus, Braf ablation did not cause developmental defects or affect endothelial homeostasis. Adult angiogenesis, assessed as the ability to vascularize VEGF-and FGF-containing Matrigel plugs, was similarly unaffected (Fig. 1E); and Braf D/D and F/F cells performed equally well in a sprouting angiogenesis assay in 3D cultures (Fig. 1F). We next determined how BRAF ablation affected the functional properties of 2D monolayers of primary microvessel-derived mouse endothelial cells (pMECs). VEGF-, thrombin-, and histamine-induced paracellular permeability, measured by FITC-dextran leakage [24], was significantly reduced in BRAF-deficient pMEC monolayers ( Fig. 2A).
To monitor the transient disruption of the endothelial barrier by VEGF in real time, we recorded the dynamic changes in electrical impedance (transendothelial resistance, TER) of pMEC monolayers. Figure 2B,C shows typical traces. BRAF-deficient pMEC monolayers monitored for 9 h after plating showed a slightly higher baseline cell index than F/F cultures (Fig. 2B) BRAF-deficient monolayers (Fig. 2C,D), indicating increased endothelial barrier function in good agreement with the results of the paracellular permeability assay ( Fig. 2A).

BRAF ablation impacts signaling to the cytoskeleton
To gain more insight into the mechanism by which BRAF regulates paracellular permeability, we monitored morphological changes in monolayers of pMEC continuously growing, starved, or exposed to VEGF. VEGF induced RSF formation, elongation of VE-Cadherin-containing AJs (indicative of radial tension), and intercellular gap development in F/F pMECs, but were severely impaired in BRAFdeficient pMECs (Fig. 3A). These qualitative results are consistent with, and complement, the quantitative measurement of barrier function ( Fig. 2A,C,D). Notably, the reduction in RSF and prominent CAB were stimulus-independent and could also be observed in unstimulated or continuously growing BRAF-deficient pMECs (compare Fig. 3A,B), where they also correlated with reduced F-actin content (Fig. 3C). The morphology of BRAF-deficient pMECs was similar to that of F/F cells treated with the EPAC activator 007 (Fig. 3D), which decreases permeability of endothelial monolayers through a RAP1-dependent tightening of VE-Cadherin-containing AJ [14,15,25]. Treatment with 007 significantly increased TER in both F/F and BRAF-deficient MECs; however, there was no significant difference between 007-treated F/F pMECs and untreated BRAF-deficient pMECs, indicating that 007 treatment and BRAF ablation have a similar impact on AJ tightening (Fig. 3E). In good correlation with the reduction in RSF and Factin and with the prominent CAB observed in continuously growing, unstimulated or VEGF-treated BRAFdeficient pMECs, we observed a decrease in the phosphorylation of the ROKa (RHO-dependent kinase a) effector LIMK (LIM Kinase) and of its target, the actin-severing protein COFILIN, used as a readout for ROK signaling, under both basal and VEGF-induced conditions (Fig. 3F, left panel). VEGF signaling upstream of ROK was unaltered or slightly increased in BRAF-deficient pMECs compared with F/F cells (Fig. 3F, right panel), suggesting a roadblock in RHOA signaling at the level of ROKa. Reduced COFILIN phosphorylation, RSF formation, and F-actin content have also been observed in BRAF knockout fibroblasts, where they correlated with ERK-dependent reduction in ROKa expression [26]. ROKa expression, however, was indistinguishable in BRAF-proficient and -deficient pMECs (Fig. 3F), indicating that a distinct mechanism impacts ROKa signaling in the latter cell type. BRAF could also promote actomyosin formation, cell contractility [27], and endothelial permeability [28] through its effectors MEK/ERK, which activate MLCK (myosin light chain kinase) [29]. VEGF-induced ERK activation was reduced in BRAF-deficient pMEC monolayers (Fig. 3F). However, the MEK inhibitor trametinib, which completely blunted ERK activation in F/F pMECs, had no impact on VEGF-induced loss of TER (Fig. 3G), indicating that the reduced MEK/ERK activation in BRAF-deficient pMEC is not the cause of decreased permeability.

BRAF ablation increases RAF1 interaction with ROKa at VE-Cadherin-containing AJs
The VE-Cadherin-containing junctions are crucial for the regulation of vessel permeability. Association of VE-Cadherin with VEGFR2 induces its endocytosis, destabilizing the junctions; in contrast, the association with the cytoskeleton and particularly with CAB increases AJ stability [30]. Consistent with the decreased sensitivity to permeabilizing agents and with the prominent CAB observed in BRAF-deficient pMECs, VE-Cadherin association with VEGFR2 and with the cytoskeleton (measured by binding to a, b, and p120 catenins; Fig. 4A) was increased in these cells. Low amounts of BRAF could be detected in F/F VE-Cadherin immunoprecipitates; importantly, however, the association of VE-Cadherin with ROKa was increased (2.8-fold) in BRAF-deficient pMECs ( Fig. 4A; see also Fig. 4D). ROKa binding to recombinant RHOA-GTPcS was not decreased in BRAFdeficient lysates, indicating that this is not the activation step inhibited by BRAF ablation. The ROKa interactor RAF1, but not BRAF, could be recovered in the RHOA-GTPcS pull downs (Fig. 4B). Similar amounts of ROKa were recovered in RHOA-GTPcS pull downs from control and RAF1-deficient lysates; thus, RAF1 is dispensable for the binding of ROKa to active RHOA. BRAF ablation slightly increased the amount of RAF1 present in the RHOA-GTPcS pull downs; consistently, more ROKa was present in RAF1 immunoprecipitates from BRAF-deficient than from F/F pMECs (Fig. 4C, threefold increase). The amount of RAF1 and ROKa interacting with VE-Cadherin was also increased to a similar extent (2.8fold) in BRAF-deficient pMECs, as shown by VE-Cadherin immunoprecipitation ( Fig. 4D; see also

RAF1 ablation rescues the permeability defects of BRAF-deficient pMECs
We next investigated whether increased RAF1/ROKa interaction and recruitment to VE-Cadherin observed in BRAF-deficient pMECs was causally linked to the decrease in COFILIN phosphorylation, filamentous actin, RSF, and TER. In BRAF D/D /RAF1 D/D pMECs, ERK phosphorylation was decreased to a level comparable to that of BRAF D/D pMECs (Fig. 5A). The residual ERK phosphorylation in BRAF D/D /RAF1 D/D pMECs does not correlate with ARAF upregulation (data not shown). A similar phenotype has been observed in primary keratinocytes [31][32][33], and may be due to the activity of alternative MEK kinases, such as TPL2 or MOS, due to reduced DUSP expression, or due to the attenuation of ERK-dependent negative feedback mechanisms.

RAF dimerization regulates VEGF-induced permeability and cytoskeletal rearrangements
To gain insight on the mechanism by which BRAF impacts the binding of RAF1 to ROKa, we transfected pMECs either with empty vector (eV) or with constructs encoding wild-type (WT) or kinase-dead (K483M) BRAF proteins [34]. Wild-type BRAF, but not the K483M mutant, efficiently rescued permeability and increased both COFILIN and ERK phosphorylation (Fig. 6A). These results were confirmed using a second kinase-dead mutant (D594A; Fig. 6B) [34]. Additionally, a BRAF mutant which cannot bind to RAS or RAP1 (R188L) [35] failed to rescue both the biological and the biochemical phenotypes of BRAFdeficient pMECs (Fig. 6B). Thus, both RAS/RAP1 binding and BRAF kinase activity are required for the control of pMEC permeability by BRAF. We analyzed the significance of RAF dimerization in the control of pMEC permeability by BRAF using mutants with either reduced (R509H) or increased (E586K) affinity for RAF1 (Fig. 6C) [36]. R509H BRAF failed to rescue the TER phenotype and led to a marginal increase in pCOFILIN and pERK. Conversely, E586K significantly increased VEGFinduced permeability as well as COFILIN and ERK phosphorylation (Fig. 6C). As confirmed in   cotransfected COS7 cells, the ability of BRAF mutants to dimerize with RAF1 correlated with their ability to rescue the endothelial cell phenotype (Fig. 6D,E). Collectively, the data indicate that the role of BRAF in permeability is kinase dependent and that it requires RAS/RAP1 binding and dimerization with RAF1.

Endothelial BRAF controls vessel permeability in vivo
To determine whether BRAF was required for the control of endothelial permeability in vivo, we next injected VEGF, histamine, or thrombin, all of which act through the RHO/ROK signaling pathway [10- Input (μg) 13], intradermally in Braf D/D , and control littermates. BRAF-deficient vessels were more resistant to all three permeability-inducing stimuli; however, intradermal injection of VEGF, histamine, or thrombin induced similar levels of permeability in RAF1 D/D , BRAF D/D /RAF1 D/D , and control mice (Fig. 7A). To assess whether the permeability phenotype impacts tumor growth, we used two different allograft models that depend on tumor vascularization, namely Lewis lung carcinoma (LLC-1) and B16F10 melanoma grafts, which depend on VEGF for growth [37,38]. Braf D/D and F/F littermate supported the     growth of LLC-1 and B16F10 grafts at indistinguishable levels (Fig. 7B,C). However, colonization of the lung by B16F10 melanoma cells injected in the tail vein, a widely used model for tumor cell extravasation in the lung vasculature [39], was less efficient in Braf D/D than in control littermates (Fig. 7D). Consistently, VEGF, histamine, and thrombin also promoted the migration of B16F10 melanoma cells through a monolayer of F/F, but not BRAF knockout endothelial cells, and this phenotype was rescued in BRAF/RAF1 knockout monolayers (Fig. 7E). Taken together, the results show that endothelial BRAF ablation reduces the paracellular permeability of endothelial monolayers in culture and vessel permeability in vivo irrespectively of the inducer, and that these phenotypes depend on the presence of RAF1 and on the formation of BRAF/RAF1 dimers (Fig. 8).

Discussion
Vascular permeability defects are common to many pathological conditions. Weakening of the endothelial barrier causes vascular leakage and edema in cardiovascular and inflammatory diseases. In cancer, the leaky tumor-associated vasculature facilitates metastatic spreading and hampers drug delivery. In both instances, normalization of the vasculature would be desirable; the search for therapeutic approaches based on the molecular understanding of the endothelial barrier function is ongoing.

A BRAF/RAF-RAF1/ROKa rheostat regulates paracellular permeability in endothelial monolayers
The regulation of AJ and cytoskeletal remodeling by RHO GTPases play a crucial role in endothelial permeability. Specifically, RAP1 and RHO have opposite functions, the former stabilizing CAB and AJs, the latter driving RSF formation, contractility, and AJ remodeling [2,30]. Permeability-perturbing agents cause activation and relocalization of RAP1 and RHO through their activators (GEFs) or inhibitors (GAPs). While this part of the signaling pathways leading to paracellular permeability is rather well studied [40], what happens downstream is less clear.
Our data now show that BRAF controls endothelial permeability by reducing both the binding of RAF1 to ROKa and the recruitment of this complex to VE-Cadherin-containing AJs. All players are found in complex with VE-Cadherin (Fig. 4). In BRAF-deficient pMECs, increased ROKa signaling at the AJs favors the formation of CAB over RSF and reduces overall F-actin content. These morphological and biochemical phenotypes are evident in unstimulated pMECs. In contrast, the physiological phenotype is revealed both in vivo and in vitro by stimulation with permeability-inducing agents. In BRAF-deficient cells and vessels, the efficacy of these agents is reduced due to the stabilization of AJs and the increased strength of the tonic permeability barrier induced by increased RAF1/ROKa signaling. This conclusion is backed by the fact that only BRAF proteins able to bind to RAF1 are able to rescue the permeability phenotype in pMEC monolayers; equally importantly, the phenotypes of BRAF-deficient cells are rescued by the concomitant ablation of RAF1. By demonstrating that BRAF, RAF1, and ROKa receive and integrate signals from permeability stimuli, and that BRAF/RAF1 and RAF1/ROKa heterodimers act as a rheostat fine-tuning endothelial barrier function, our results advance our understanding of the mechanisms modulating AJ dynamics and cytoskeletal remodeling.

Potential mechanisms of BRAF/RAF1 and RAF1/ ROKa heterodimerization
We have recently shown that the RAF1 phosphospecies able to bind ROKa is generated in the context of RAF dimers formed during ERK activation. However, in the context of the RAF dimer, BRAF promotes RAF1 autophosphorylation on a 14-3-3 residue which stabilizes RAF dimers, thereby favoring BRAF/RAF dimerization over RAF1-ROKa complex formation and efficiently controlling their levels [41]. How exactly BRAF/RAF1 dimerization is modulated by permeability-promoting signals in pMECs is unclear. RAS activation, which regulates different aspects of endothelial cell biology [42][43][44], occurs upon stimulation with VEGF but also with thrombin [45] and, at least in HEK293T cells, with histamine [46]. Alternatively, RAP1, which has been shown to regulate both RAF1/ ROKa heterodimerization and their association with VE-Cadherin at AJs [21], may also control RAF dimerization. In favor of this, RAP1 activates ERK via BRAF [47,48], activates BRAF in cell-free extracts [49] and binds to both RAF molecules with different affinities, determined by their divergent CRD domains [50].
In this scenario, both BRAF/RAF1 and RAF1/ ROKa heterodimers would be stimulated by the activation of the same GTPase, RAP1 (Fig. 8).
But if this is the case, how do RAF1/ROKa heterodimers form in BRAF-deficient cells?
It is important to point out here that low levels of basal and growth factor-induced ERK phosphorylation are still detectable in BRAF-deficient pMEC (Fig. 4), fibroblasts, and keratinocytes [51], indicating that this function of BRAF is at least partially redundant. It is thus likely that other RAF1 dimerization partners (such as RAF1 itself, ARAF, or KSR) can both maintain ERK activation and prime RAF1 for ROKa complex formation in BRAF-deficient cells. Over time, the interaction with these less efficient dimerization partners/activators would generate an increased number of ROKabinding RAF1 molecules, leading to the cytoskeletal phenotypes observed in BRAF-deficient pMECs. In favor of this hypothesis, increased RAF1/ROKa complex formation has also been observed in BRAF-deficient keratinocytes [31].
Whatever the precise mechanism underlying their yin-yang behavior in pMECs, the BRAF/RAF1 -RAF1/ROKa module impacts permeability induced by agents responsible for vessel leakage not only in tumors but also in other conditions, including cardiovascular and inflammatory diseases. Our results thus suggest that inhibitors preventing RAF dimerization would be beneficial in a broad range of disorders associated with permeability defects. In the specific context of cancer, RAF dimerization inhibitors combine a beneficial cell autonomous effect on tumor proliferation, by reducing the activity of the ERK pathway, with the normalization of vascular permeability, allowing for better drug delivery.

Methods
Generation of BRAF D/D mice BRAF F/F mice were mated to VEC-Cre [22] (Charles River Laboratories, Sulzfeld, Germany) mice to obtain BRAF D/D animals. BRAF ablation was determined by allele-specific PCR analysis as previously described [21]. Compound deletion of BRAF and RAF1 in endothelial cells was obtained by mating BRAF F/F /RAF1 F/F mice with VEC-Cre-expressing animals. Animal experiments were authorized by the Austrian Ministry of Science and Communications, following the approval by the national Ethical Committee for Animal Experimentation.

Retinal angiogenesis
Whole-mount retinas derived from 6-day-old animals were stained with CD31 antibody (BD Pharmingen, BD Biosciences, Franklin Lakes, NJ, USA; cat. No. 550274) to visualize the vascular plexus [21] and quantify the distance between central optical nerve head and angiogenic front and between capillaries and arteries.

Paracellular permeability assays
The FITC-Dextran permeability assay was performed by adding FITC-Dextran (AE 200 ngÁmL À1 VEGF, 100 lM histamine or 10 UÁmL À1 thrombin) to pMECs monolayers cultured on fibronectin-coated semipermeable inserts (0.4 lm pore size) and measuring its passage to the lower compartment after 1 h, according to the supplier's protocol (Millipore). Changes in the transendothelial electrical resistance (TER) of pMEC monolayers were measured using xCELLigence system (RTCA-DP version; Roche Diagnostics, Mannheim, Germany), which tracks changes in electrical impedance (expressed as "cell index", proportional to cell attachment and spreading). Permeabilityinducing agents causing the appearance of intercellular gaps result in changes in electrical impedance quantifiable in real time. pMECs were plated (1.5 9 10 5 ) and allowed to grow to confluence overnight on fibronectin-coated 96well E-plates prior to the addition of PBS or permeability-modifying agents [200 ngÁmL À1 VEGF or 100 lM 007 (8-pCPT-2 0 -O-Me-cAMP; Biolog Life Science, Bremen, Germany)] [15]. For MEK inhibition, cells were pretreated with 10 nM trametinib for 1 h before the addition of VEGF. To compare the effect of permeability-inducing stimuli on the different genotypes, the cell index recorded at the time of addition of the permeability-inducing stimuli or their vehicles was set as 0, and the changes in cell index induced by the permeability stimuli were subtracted from those obtained by treating the cells with their vehicles. This normalization is necessary because the cell indexes of unstimulated F/F and BRAF D/D (raw data) differ slightly (see Fig. 2B). Thus, the drop of TER caused by permeability-inducing stimuli appears as a negative value. To further help comparison among experiments and different stimuli, the values representing the maximum drop in TER induced by the stimuli in wild-type pMEC monolayers are normalized to À1 in all plots except Figs 2B and 3E, in which cell index is shown instead, and CCD camera (Hamamatsu ImageEM X2, Hamamatsu, Japan) and a Plan-Apochromat 639/1.4 Oil DIC objective lens, and analyzed with the IMAGEJ software (NIH, National Institute of Health, Bethesda, Maryland, USA). The F-actin/G-actin ratio was determined using an assay kit (Cytoskeleton, Denver, CO, USA) according to the supplier's protocol. Briefly, pMECs were lysed in 500 lL detergent-based lysis buffer and subjected to an ultracentrifugation step which pellets F-actin and leaves Gactin in the supernatant. The amount of actin in supernatant and pellet was determined by immunoblotting.

Immunoprecipitation, pull down, and immunoblotting
For immunoprecipitation, cells lysates prepared in a buffer containing 25 mM HEPES, 150 mM NaCl, 1 mM EGTA, protease inhibitors cocktail, 0.5% NP-40 and 10% glycerol were incubated immunoprecipitated with Protein G Sepharose beads coupled with the relevant antibody at 4°C overnight [31]. Immunoprecipitated proteins were analyzed by immunoblotting. GTP-bound RHOA was determined by the RHO Activation Assay Kit (Millipore) according to the supplier's protocol.

Vascular permeability assays
Vascular permeability was determined using Evans blue dye (Miles assay [54]). Intradermal injections (20 lL) of recombinant VEGF (400 ng) [24], histamine (1 lg) [11], or thrombin (10 U) [55] were performed 10 min after intravenous (i.v.) injection of sterile Evans Blue dye (100 lL, 1% in PBS). After 20 min, the injection sites were excised and incubated in formamide for 5 days, and the extracted dye was determined by spectrophotometric measurement at 620 nm. Values are expressed in fold increase versus the control injection with PBS.

Extravasation assay
B16F10 melanoma cells (1 9 10 6 ) stained with Cell-Tracker TM Orange CMRA Dye (Molecular Probes, Invitrogen Life Technologies) were injected in the tail vein of 8-week-old mice. After 2 h, two mice of each genotype were sacrificed and analyzed to control for similar lodging in the lung microvasculature. Forty-eight hours after injection, images of total lungs were acquired with the stereomicroscope Zeiss SteREO Discovery V.12 and the number of cells in the extravasation area of each of the three lobes of the lungs was quantified using the IM-AGEJ software (NIH) [57]. The numbers in the plots represent the mean AE SEM of the indicated biological replicates.

Transendothelial migration assay
The pMECs were cultured on fibronectin-coated inserts (8lm pore size) for 48 h before B16F10 melanoma cells (2 9 10 5 ) stained with CellTracker TM Orange CMRA Dye were added to the upper chamber and incubated for 6 h with FBS (1.25%) plus VEGF (200 ngÁmL À1 ), thrombin (10 UÁmL À1 ), or histamine (100 lM). Experiments were performed in triplicates and four different areas per well were counted; the integrity of pMECs monolayers was determined by crystal violet staining.

Statistical analysis
Quantitative data are presented as mean AE SD or mean AE SEM as indicated in the figure legend. Pairwise comparisons were performed by Student's t-test (twotailed), respectively.