An EMMPRIN–γ-catenin–Nm23 complex drives ATP production and actomyosin contractility at endothelial junctions

A fundamental issue in vascular biology is the understanding of the mechanisms that govern vessel stabilization, because this process is required for development and growth of the organism and because changes in vessel permeability contribute to several pathologies. Vascular integrity depends on the assembly of multicomponent endothelial junctions, the most extensively studied being homophilic VE-cadherin (also known as cadherin-5) interactions at adherens junctions (Dejana, 2004; Giannotta et al., 2013); however, the complete molecular map of the assembly and disassembly of these junctions remains to be defined. Endothelial dynamics largely depend on the spatiotemporal reorganization of the actin cytoskeleton during adherens junction maturation (Smutny et al., 2010) and on the tension that actin filaments, in conjunction with non-muscle myosin II (NM-II), exert over the membrane. This tension actively participates in the expansion from single unstable intercellular cadherin interactions to stable cell contacts involving multiple cadherin pairings (Baum and Georgiou, 2011; Mège et al., 2006; Smutny et al., 2010). Although it is known that non-muscle myosin II (NM-II) contributes to generate the tension required for proper maintenance of linear adherens junctions through several mechanisms (Mège et al., 2006), the spatiotemporal regulation of NM-II-driven tension at cell–cell junctions is just beginning to be elucidated, especially in endothelial monolayers, which experience stronger shear and flow forces than epithelial barriers (Abraham et al., 2009; Ando et al., 2013; Baum and Georgiou, 2011; Conway et al., 2013; Huveneers et al., 2012; Wimmer et al., 2012).

EMMPRIN (extracellular matrix metalloproteinase inducer, also called basigin and CD147) is a glycosylated type-I transmembrane protein of the Ig superfamily (Muramatsu and Miyauchi, 2003), and EMMPRIN molecules are therefore able to engage in homophilic interactions with other monomers and in heterophilic interactions with other partners, such as monocarboxylate transporters, integrins, caveolin-1, CD98, cyclophilins and Shrew1 (Berditchevski et al., 1997; Deora et al., 2005; Schreiner et al., 2007; Tang and Hemler, 2004; Xu and Hemler, 2005). EMMPRIN function has mostly been characterized in relation to tumorigenesis and inflammation (Muramatsu and Miyauchi, 2003). EMMPRIN can also contribute to angiogenesis, the formation of new vessels, in several ways: (1) by endothelial cell activation when present on tumor-derived microvesicles (Millimaggi et al., 2007), (2) by regulation of the PI3K–Akt signaling pathway in activated endothelial cells (Chen et al., 2009) and (3) by modulating the secretion of specific VEGF isoforms (Bougatef et al., 2009). Whether EMMPRIN plays additional roles in the quiescent vasculature has not been investigated.

Here, we report that EMMPRIN orchestrates junction stability in vitro and in vivo by forming a novel molecular complex at endothelial junctions with γ-catenin (also known as junction plakoglobin) and the nucleoside diphosphate kinase Nm23 (also known as NME1), which provides ATP for the local actomyosin contractility required for proper junction maturation.

Analysis of mouse tissue sections revealed colocalization of EMMPRIN with the endothelial cell junction marker PECAM-1 in the continuous non-fenestrated endothelium of heart and lung vessels and in the specialized tight endothelium of the blood-brain barrier, but not in the continuous fenestrated endothelium of kidney glomeruli and spleen – a pattern indicating a correlation between EMMPRIN expression and junction strength (supplementary material Fig. S1A). In whole-mount specimens, EMMPRIN was also detected at the junctions of the microvasculature of the ear and at the endothelium of large vessels (aorta) (supplementary material Fig. S1B).

The enrichment of EMMPRIN at endothelial cell–cell junctions of different vascular beds suggests a function at these sites. We therefore analyzed the vasculature of EMMPRIN-deficient mice. EMMPRIN-null mice show no apparent vascular defect, and skin and organ coloration is similar to that of wild-type animals, indicating no hemorrhage or gross alterations to circulation. However, ex vivo whole-mount staining for the junction marker PECAM-1 revealed altered intercellular contacts in several vascular territories. In the microvasculature of the trachea and the ear, the cell-boundary PECAM-1 staining in EMMPRIN-deficient mice was diffuse and discontinuous compared with the sharp pattern observed in wild-type mice, especially at branch points, where the vessels also appeared to be dilated (Fig. 1A). A similar PECAM-1 phenotype was observed in the tracheal microvasculature of EMMPRIN-heterozygous mice, indicating that a wild-type amount of EMMPRIN is required for proper junction formation (data not shown). The pattern of PECAM-1 staining was also subtly irregular in straight large vessels of the ear and in aortas from EMMPRIN-null mice (Fig. 1A).

EMMPRIN-null mice do not display macroscopic edema, and skin-tone recovery after pressure application was comparable to that of wild-type animals; moreover, histological skin sections showed no signs of edema. Nonetheless, given the altered endothelial junction structure in EMMPRIN-null mice, we directly explored vessel integrity in vivo by injecting animals with Evans Blue. We observed greater accumulation of Evans Blue in the inflamed ears (after application of mustard oil) of EMMPRIN-null mice than in their wild-type counterparts (Fig. 1B); Evans Blue leakage in the inflamed ear was also significantly higher than normal in EMMPRIN heterozygous mice (data not shown). As a complementary approach, we monitored the tissue leakage of fluorescently labeled low- or high-molecular-mass dextran (4 kDa or 70 kDa) by confocal microscopy. Leakage of both dextrans was higher in the ears of EMMPRIN-null mice when compared with that of wild-type ears, both under basal conditions and upon inflammation induced by aural application of mustard oil (Fig. 1C).

Confocal analysis confirmed that EMMPRIN is also located at junctions formed by human umbilical vein endothelial cells (HUVECs), where it partially colocalizes with the junction components VE-cadherin (48% colocalization), PECAM-1 (95%) and JAM-A (also known as F11R; 81%) (data not shown). We next investigated how adherens junctions were affected by EMMPRIN deficiency in human endothelial cells. siRNA-mediated knockdown of EMMPRIN expression with two independent oligonucleotides efficiently decreased protein expression for up to 72 h in culture (Fig. 2A). Immunostaining for endothelial junction components in EMMPRIN-silenced HUVECs revealed a marked alteration in the VE-cadherin pattern. In non-transfected (control) cells or cells transfected with a non-specific siRNA (Neg), VE-cadherin displayed a well-defined pattern of sharp contours at endothelial cell–cell contacts; by contrast, a broader, more irregular pattern was detected at most contacts between EMMPRIN-silenced cells (Fig. 2B). Image analysis confirmed a significantly wider peak of VE-cadherin intensity at the borders of EMMPRIN-knockdown cells (Fig. 2C). A similarly broadened distribution was observed for the VE-cadherin partner β-catenin (data not shown). These structural alterations resulted in impaired barrier function, as shown by the significantly higher permeability of EMMPRIN-knockdown HUVEC monolayers to labeled dextran in transwell assays (Fig. 2D). The function of EMMPRIN in endothelial cell–cell junctions was independently demonstrated in EMMPRIN-deficient primary mouse lung endothelial cells (MLECs), which also displayed a VE-cadherin distribution of reduced intensity, lower definition and an overall broadened pattern at cell–cell junctions (Fig. 2E; supplementary material Fig. S2E,F).

VE-cadherin stability of adherens junctions is related to the tension exerted by circumferential actin bundles together with NM-II at these sites (Hoelzle and Svitkina, 2012; Liu et al., 2010). We first observed that EMMPRIN colocalized with F-actin (as visualized by phalloidin staining) at filopodia and lamellipodia in both early exploratory immature cell–cell contacts (1 h) and mature junctions established over 48 h (Fig. 3A). The previously reported pattern of the actin network at immature endothelial junctions (Hoelzle and Svitkina, 2012) was detected at junctions formed by control endothelial cells after 1 h, whereas junctions in EMMPRIN-knockdown cells lacked this actin pattern at this early stage and were characterized by the persistence of radial actin filaments (Fig. 3B). To visualize the impact of EMMPRIN deficiency on live actin polymerization during both the formation and maturation of cell–cell contacts, we first knocked down EMMPRIN in human endothelial cells, transfected them with LifeAct and then stained them with a directly labeled anti-EMMPRIN antibody. Time-lapse video microscopy revealed colocalization of polymerized actin and EMMPRIN clusters at membrane contacts established by non-knockdown endothelial cells (Fig. 3C; supplementary material Movie 1). In these cells, actin transited from a network with few filaments perpendicular to the junction (see also Fig. 3B) to a pattern of parallel filaments as described previously (Mège et al., 2006). By contrast, in EMMPRIN-knockdown cells, the actin filaments mostly remained perpendicular to the junction (see also Fig. 3B) until the end of the sequence (8:20 h of time-lapse video microscopy, at 24 h after seeding (Fig. 3C; supplementary material Movie 2).

To further define actomyosin activity at endothelial junctions, we quantified active NM-II (by staining the phosphorylated regulatory myosin light chain; pMLC) and polymerized F-actin (by phalloidin staining) at cellular regions of interest defined around the junctions (ROIs; see Fig. 4A,B and Materials and Methods for details). EMMPRIN-knockdown human endothelial cells displayed reduced actomyosin activity at the junctions, as shown by the significantly decreased staining of pMLC and F-actin, contrasting with the punctate staining of pMLC and the enrichment of F-actin observed at the junctions of control cells (Fig. 4A,C). This reduction was further demonstrated by the significant decrease in the pMLC∶MLC ratio observed by western blotting in membrane-enriched fractions from EMMPRIN-knockdown human endothelial cells compared with that of controls (Fig. 4D). Moreover, a similar reduction in pMLC and F-actin and, therefore, actomyosin activity was also found at the aortic endothelial junctions of EMMPRIN-null mice analyzed ex vivo; this reduction was better visualized in a three-dimensional-rendering reconstruction of pMLC and F-actin images (Fig. 4E,F). We next quantified the linearity index (defined as the ratio of junction length to the distance between vertices) introduced by Takeichi's group in epithelial cells as a read-out of actomyosin contractility (Otani et al., 2006); as shown in Fig. 4G, VE-cadherin-based linearity index was significantly increased in EMMPRIN-knockdown human endothelial cells compared with that of control cells, demonstrating reduced actomyosin contractility in the absence of EMMPRIN. To directly quantify the forces exerted by actomyosin at the endothelial junctions, we plated human endothelial cells on an array of fluorescently labeled elastomeric microneedles, which allowed quantification of intercellular tugging forces in pairs of cells (Cohen et al., 2013; Liu et al., 2010). In accordance with image analysis and western blot data (Fig. 4A–D) and to increased linearity index (Fig. 4G), the intensity of tugging forces (the force across the cell junction) was significantly lower in EMMPRIN-knockdown cells (Fig. 4H).

We also observed a significant increase in pMLC (but not F-actin) in the interior of EMMPRIN-deficient HUVECs and aortic endothelial cells (supplementary material Fig. S2A). In an attempt to rescue the phenotype through the reestablishment of tension at the junctions, we next analyzed the effect of relaxing the inner non-muscle myosin activity with the inhibitor of Rho kinase Y27632, which only reduces pMLC inside cells, leaving intact punctate pMLC staining at junctions (Smutny et al., 2010; Totsukawa et al., 2000). Y27632 restored the linear VE-cadherin pattern and restored permeability in EMMPRIN-knockdown HUVECs to a level similar to that of control cells (supplementary material Fig. S2B–D); the inhibitor also restored the junction pattern and integrity in EMMPRIN-null MLECs (supplementary material Fig. S2E,F).

The above results indicate that EMMPRIN contributes to pMLC regulation and actomyosin contractility at endothelial cell junctions either through direct interaction with the actomyosin machinery or through an adapter protein. To explore this, we used the EMMPRIN cytosolic tail as bait to search for interacting proteins in epithelial cell lysates. Mass spectrometry of pulled-down proteins identified several candidates, including γ-catenin, a known component of endothelial cell junctions. The association of EMMPRIN with γ-catenin in HUVECs was confirmed by co-immunoprecipitation (Fig. 5A). The immunoprecipitates did not contain VE-cadherin, and γ-catenin colocalized with EMMPRIN in VE-cadherin-null mouse endothelial cells, pointing to the VE-cadherin independence of this complex (Fig. 5B). EMMPRIN colocalized with γ-catenin at both early exploratory and stable junctions formed by human endothelial cells (Fig. 5C; data not shown). Moreover, γ-catenin was barely detected by immunostaining in EMMPRIN-knockdown endothelial cells, particularly at endothelial junctions, despite a lack of significant difference in γ-catenin protein levels as determined by western blotting, suggesting that EMMPRIN is necessary for the proper recruitment of γ-catenin to these sites (Fig. 5C,D). Levels of γ-catenin were also low at cell–cell junctions of the ear vessels from EMMPRIN-null mice, indicating that EMMPRIN contributes to γ-catenin recruitment to endothelial junctions in vivo (Fig. 5E). Microscopy analysis showed that siRNA-mediated knockdown of γ-catenin in human endothelial cells significantly decreased the amount of pMLC and F-actin at junctions, in a manner similar to that of EMMPRIN-knockdown cells, confirming the contribution of this complex to pMLC–actomyosin contractility at junctions (Fig. 5F,G).

Given that actomyosin contractility requires ATP, we next investigated the possible role at the junction of the reported γ-catenin partner nucleoside diphosphate kinase Nm23 (Aktary et al., 2010), which is able to generate ATP by exchanging phosphate groups from NTP to ADP. The presence of Nm23 in the EMMPRIN molecular complex was confirmed by co-immunoprecipitation from endothelial cell lysates (Fig. 6A). In addition, although Nm23 staining was mostly diffuse within endothelial cells, Nm23 colocalized with EMMPRIN and γ-catenin at specific points in some junctions of confluent human endothelial cells (Fig. 6B); this was better visualized after methanol fixation or saponin treatment to remove the soluble cytosolic pool of Nm23 (supplementary material Fig. S3A). The interaction of EMMPRIN with γ-catenin and Nm23 was further demonstrated by using the proximity ligation assay (PLA), which showed positive signal for EMMPRIN–γ-catenin and for γ-catenin–Nm23 at the endothelial cell junctions of confluent HUVEC monolayers (Fig. 6B, lower panels). Moreover, Nm23 localization at endothelial cell junctions was significantly reduced in both EMMPRIN-knockdown and γ-catenin-knockdown HUVECs (Fig. 6C,D), pointing to an essential requirement for EMMPRIN–γ-catenin in Nm23 recruitment to the junctions. To better address this point, we transfected EMMPRIN-knockdown HUVECs with a myristoylated EMMPRIN cytosolic tail; we first confirmed that the myristoylated EMMPRIN cytosolic tail was recruited to endothelial cell junctions, likely due to cytosolic residues involved in EMMPRIN trafficking (Fig. 7A,B; Deora et al., 2004). The myristoylated EMMPRIN cytosolic tail rescued the presence of γ-catenin and Nm23 as well as pMLC and F-actin staining at the junctions of EMMPRIN-knockdown cells (Fig. 7A,B), indicating that the sole cytosolic EMMPRIN tail anchored to the membrane is able to recruit γ-catenin and Nm23, and that this interaction is essential for proper actomyosin contractility at junctions.

In order to gain further mechanistic insights, we next analyzed the impact of decreasing Nm23 levels on endothelial junction stability. Endothelial cell junctions were disrupted in Nm23-knockdown human endothelial cells, as indicated by decreased levels of pMLC and altered F-actin at junctions, resulting in a significantly increased linearity index together with reduced VE-cadherin staining and enhanced endothelial permeability (Fig. 8A–C; supplementary material Fig. S3B,C). Because Nm23 can directly regulate cytoskeleton contractility in Dictyostelium by providing ATP to the myosin molecule (Aguado-Velasco et al., 1996), we explored the possible involvement of this mechanism in mammalian endothelial cells by quantifying ATP production at endothelial junctions. We first used the Perceval plasmid to visualize and quantify ATP∶ADP ratios (Berg et al., 2009) at endothelial junctions in the presence of inhibitors of the main sources of ATP production in endothelial cells (glycolysis and mitochondria; see Materials and Methods for details). We visualized and quantified ATP∶ADP ratios before and after inhibitor treatment and observed that the ATP∶ADP ratio at the endothelial cell periphery (junctional area) was decreased to a greater extent in EMMPRIN-knockdown compared with control knockdown endothelial cells upon inhibitor addition, pointing to lower production of ATP in the absence of EMMPRIN at the membrane (Fig. 8D). To assess this, we directly quantified ATP production in membrane-enriched fractions in the presence of inhibitors of ATPase synthase and adenylate kinase and in the presence of GTP and ADP as substrates for Nm23. ATP production in these conditions was lower in membranes from EMMPRIN-knockdown cells than in membranes from control knockdown cells (Fig. 8E). Moreover, an inhibitory anti-Nm23 antibody significantly decreased ATP production by membranes from control human endothelial cells but had only a minor effect in EMMPRIN-knockdown cells, indicating that Nm23 makes an essential contribution to ATP production at the membrane in the presence of EMMPRIN (Fig. 8E). Overall these findings point to decreased pMLC and actomyosin contractility and impaired junctions in EMMPRIN-deficient endothelial cells as a result of defective local ATP production.

Our results show that EMMPRIN is essential for endothelial junction stability and vascular integrity. EMMPRIN, through a newly identified association with γ-catenin and Nm23 during junction formation, acts as a key intermediary between the plasma membrane and actomyosin contractility by providing ATP, thus coordinating the forces required for junction initiation, stabilization and integrity in vitro and in vivo (supplementary material Fig. S4).

Endothelial cell–cell adhesions are essential for vascular integrity, and their alteration results in severe disorders (Dejana et al., 2009). Despite a lack of previous vascular phenotype described in EMMPRIN-null mice (Muramatsu and Miyauchi, 2003), we observe that they show above-normal vascular leakage under basal and inflammatory conditions, demonstrating an essential contribution of EMMPRIN to endothelial barrier functioning in vivo. This permeability phenotype is unlike that of mice deficient for other Ig family members present at endothelial junctions, such as PECAM-1 or JAM-C (also known as JAM-3) (Graesser et al., 2002; Orlova et al., 2006), suggesting that each junction component has selective and specific functions in the regulation of vessel integrity. Because permeability differences in arteries, capillaries and venules correlate with the heterogeneous actin organization and function in these vascular beds (Prasain and Stevens, 2009), it is also possible that EMMPRIN contributes to this heterogeneity through its preferential expression at the junctions of certain vascular beds, in particular at those with a tighter endothelial barrier.

Forces generated at cell–cell junctions through dynamic interactions between transmembrane proteins and the actomyosin cytoskeleton are especially important for the maintenance of endothelial barrier fitness, but the underlying molecular mechanisms were not fully understood. Here, we identify EMMPRIN as an essential coordinator of these forces because, in its absence, actomyosin-driven contractility is reduced and VE-cadherin contacts are not properly established; in this regard, we have extended the utility and complementarity (to the quantification of junctional pMLC and F-actin by image analysis) of the linearity index as a read-out of actomyosin contractility not only in epithelial but also in endothelial junctions (Otani et al., 2006) Direct quantification by elastomer sensors (Liu et al., 2010) revealed that forces across cell–cell junctions (tugging forces) are significantly diminished in EMMPRIN-knockdown endothelial cells. Because this method only allows force quantification in endothelial cell pairs it will be of interest to use alternative approaches, such as the recently established fluorescence resonance energy transfer (FRET)-based tension sensors to quantify junctional forces in established endothelial monolayers (Conway et al., 2013). The relevance of balanced forces for endothelial junction integrity is further supported by the restoration of the wild-type junction phenotype in EMMPRIN-deficient cells upon relaxation of inward tension with a Rho kinase inhibitor.

This function of EMMPRIN in orchestrating spatiotemporal junctional actomyosin (pMLC and F-actin) contractility is particularly relevant, because this network is just beginning to be elucidated in endothelial cells (Ando et al., 2013; Huveneers et al., 2012; Wimmer et al., 2012). EMMPRIN absence resulted in delayed actin maturation at endothelial junctions, in accordance with its role in organizing cortical actin at neuromuscular junctions in Drosophila (Besse et al., 2007). In terms of the mechanism involved, the polybasic juxtamembrane region of EMMPRIN is important for cortical actin organization in Drosophila, but no EMMPRIN cytosolic interactions have been described, except for an association with the microtubule protein Hook1 that was reported recently (Besse et al., 2007; Maldonado-Báez et al., 2013). We have identified by mass spectrometry a novel molecular interaction of the cytosolic tail of EMMPRIN with γ-catenin, with this complex being required for proper actomyosin-dependent forces during the formation of endothelial junctions (data not shown). This is in line with the specific role of γ-catenin in organizing the cortical actin skeleton described in Xenopus (Kofron et al., 2002). Because γ-catenin can also associate with CPI-17 (also known as PPP1R14A), an inhibitor of MLC phosphatase, this might contribute to the overall regulation of MLC phosphorylation at the junction (Kim et al., 2013), supporting broader effects of γ-catenin on junctional F-actin and on adherens junctions.

The identification of the EMMPRIN–γ-catenin complex raises the question of how it regulates actomyosin contractility and junction stability. ATP is required for phosphorylation of MLC by MLCK, and we therefore explored the possible role of Nm23, a member of the nucleoside diphosphate kinase (NDPK) family, which was recently identified as a partner of γ-catenin in tumor cells (Aktary et al., 2010). Nm23 could be found at endothelial junctions together with EMMPRIN and γ-catenin, and knockdown of either of these proteins resulted in a significant decrease in Nm23 localization to these sites. A myristoylated EMMPRIN cytosolic tail rescued γ-catenin and Nm23 presence at junctions, together with pMLC and F-actin organization at these sites, further supporting an essential contribution of the EMMPRIN–γ-catenin–Nm23 complex to the regulation of tension at junctions. A detailed analysis of how this molecular complex is assembled and the protein domains involved in the interactions deserves further investigation.

However, the mechanism by which Nm23 might regulate endothelial intercellular adhesion remained unclear. The primary function of NDPK is to provide phosphates to ADP, thus generating ATP, which could then be used for myosin-dependent cytoskeleton contraction, as reported in Dictyostelium (Gajewski et al., 2003). By using (for the first time in endothelial cells) the recently developed Perceval tool (Berg et al., 2009; Zala et al., 2013), we could visualize and quantify ATP∶ADP ratios at the subcellular level and show that, in inhibitor-treated cells, the absence of EMMPRIN resulted in lower ATP∶ADP ratios at the junctional area compared with those of control cells. The requirement for EMMPRIN at the endothelial membrane in the process of local ATP production was further confirmed in isolated membranes; in addition, the ability of an inhibitory anti-Nm23 antibody to mimic the effect of EMMPRIN depletion, significantly reducing ATP production in membranes, further demonstrates the contribution of Nm23 to ATP production within the EMMPRIN–γ-catenin complex at human endothelial junctions. The EMMPRIN–γ-catenin–Nm23 complex would provide a local source of ATP required for MLCK-dependent MLC phosphorylation and actomyosin-driven forces at junctions; alternatively, local ATP can be crucial for the dynamic regulation of weak/strong actomyosin force cycles required for efficient contractility (Aprodu et al., 2008). This Nm23/NDPK function would be particularly important when ATP levels are low (Aguado-Velasco et al., 1996), as might be the case at maturing junctions, where the packed actin cytoskeleton might restrict free diffusion of ATP from the cytosol. Nm23 knockdown resulted, however, in a more severe junction phenotype than that observed in EMMPRIN-knockdown or γ-catenin-knockdown endothelial cells, because, in addition to decreased junctional pMLC and increased linearity index and to altered VE-cadherin pattern and augmented monolayer permeability, endothelial cells displayed lamellipodia-like protrusions at the cell–cell adhesions, resembling disrupted junctions. This might be related to global cellular effects of Nm23 knockdown and to additional functions attributed to Nm23 in regulating adherens junctions, such as Rab5-mediated endocytosis and inhibition of Rac signaling by sequestering Rac guanine nucleotide exchange factors (GEFs) in Drosophila, and Arf6-dependent E-cadherin endocytosis in canine epithelial cells (Palacios et al., 2002; Woolworth et al., 2009). The EMMPRIN–γ-catenin–Nm23 complex reveals an unexpected link between compartmentalized energy production and endothelial cell junction stability that might be relevant to situations in which changes in endothelial cell metabolism and junction remodeling need to be coordinated, such as during angiogenesis, collective migration and tissue morphogenesis.

EMMPRIN-deficient mice were generated as described previously (Muramatsu and Miyauchi, 2003). EMMPRIN heterozygotes were crossed following the strategy described previously (Chen et al., 2004), and null littermates were used for experiments. Mice were kept in the CNIC Animal Facility under pathogen-free conditions and according to institutional guidelines. All animal studies were approved by the CNIC Animal Care Ethical Committee.

Mouse monoclonal antibody (mAb) VJ1/9.1 against the EMMPRIN ectodomain was generated in the laboratory of F.S.-M. The anti-VE-cadherin TEA1/31 and anti-PECAM-1 TP1/15 mAbs have been described previously (Yáñez-Mó et al., 1998). The anti-JAM-A BV16 mAb was obtained from Hycult Biotech. Rabbit anti-phospho-myosin light chain (pMLC) antibody was from Cell Signaling, rabbit anti-myosin light chain was from Santa Cruz and phalloidin–Texas-Red was obtained from BD Pharmingen. Whole-mount mouse vessels were stained with anti-PECAM-1 from Millipore, anti-VE-cadherin and anti-γ-catenin from BD Biosciences, and anti-EMMPRIN from eBiosciences. Goat anti-mouse-IgG, anti-armenian-hamster-IgG and anti-rat-IgG secondary antibodies were from Jackson ImmunoResearch. Type-I collagen was purchased from PureCol, Opti-MEM from Gibco, Medium 199 from Lonza, Evans Blue from Sigma and Oligofectamine and dextran–TRITC from Invitrogen. The Rho kinase inhibitor Y27632 was obtained from Sigma. Anti-Nm23 antibodies were purchased from Santa Cruz Biotechnology and GeneTex. LifeAct (Riedl et al., 2008) and myristoylated cytosolic EMMPRIN (Ruiz et al., 2008) constructs were provided by Roland Wedlich-Soldner and Xosé Bustelo, respectively.

Mice were perfused with 1% paraformaldehyde (PFA) in PBS. Postnatal retinas and the trachea, aorta and ears of adult animals were extracted and post-fixed in 1% PFA for 1 h at room temperature. Tissues were then incubated with primary antibodies in 5% goat serum (Jackson ImmunoResearch) overnight at room temperature. Samples were washed for 8 h in PBS containing 0.3% Triton X-100 and labeled with the appropriate secondary antibody overnight at room temperature. The next day, samples were washed again in PBS containing 0.3% Triton X-100 and mounted in Vectashield (Vector Laboratories). Samples were examined with a Leica TCS Sp5 spectral confocal microscope or a Zeiss LSM 700 confocal microscope.

A modified Miles assay was used to examine vessel permeability in 4–8-week-old anesthetized mice (Miles and Miles, 1952). Briefly, the mice were injected intravenously with 100 µl of 5 mg/ml dextran–TRITC (4 KDa or 70 KDa) or 100 µl of Evans Blue to a final concentration of 30 mg/kg and, after 5 min, 5% mustard oil in mineral oil (Sigma) was applied to the right ear. Mineral oil alone was applied to the left ear as a negative control. After 30 min, mice were sacrificed, and ears from dextran-injected mice were extracted and processed for whole-mount staining and ears from Evans-Blue-injected mice were dried and weighed. Evans Blue was extracted from tissues in formamide (Fluka) and measured in a spectrophotometer (BioRad).

Human umbilical vein endothelial cells (HUVECs) were obtained and cultured as described previously (Yáñez-Mó et al., 1998). Cells from passages two to four were used in all assays. For endothelial monolayer formation assay, HUVECs were seeded at subconfluence on collagen-I-coated plates or coverslips. Cells were processed for analysis at 1, 24, 48 and 72 h after seeding. Mouse lung endothelial cells (MLECs) were obtained and cultured as described previously (Oblander et al., 2005). Briefly, lungs from wild-type or EMMPRIN-null mice were excised, disaggregated and digested in 0.1% collagenase (Gibco) for 1 h at 37°C. The cell suspensions were seeded onto plates coated with 10 µg/ml fibronectin (Sigma), 10 µg/ml collagen I (PureCol) and 0.1% gelatin (Sigma). After attachment, cells were negatively selected with anti-CD16/CD32 mAb (BD Biosciences) coupled to magnetic beads (Dynal, Invitrogen), and then positively selected with anti-ICAM-2 (BD Biosciences) coupled to magnetic beads. Cells were seeded onto plates coated with 10 µg/ml collagen I before assays.

Tissue cryosections were fixed with 4% PFA, 2% sucrose in PBS for 15 min and quenched for 1 h at room temperature with 100 mM glycine in PBS. After washing, sections were blocked overnight with 2% bovine serum albumin (BSA) and 5% goat serum in PBS. Sections were then incubated with primary antibodies for 2–3 h at 37°C. After washes, samples were incubated with goat secondary antibodies for 1 h at 37°C, washed again and mounted with Prolong (Invitrogen). Cells were grown on collagen-I-coated coverslips and samples were fixed at different times in 4% PFA, 2% sucrose in PBS for 15 min at room temperature. Samples were blocked with 2% BSA and, when necessary, permeabilized with 0.1% Triton X-100 for 10 min. For colocalization of Nm23 with EMMPRIN and γ-catenin, cells were also fixed in methanol at −20°C for 2 min or treated with 0.01% saponin for 5 min at room temperature before PFA fixation. Coverslips were incubated with the primary antibody for 1 h at 37°C and, after washes, labeled with the appropriate secondary antibody under the same conditions. For double staining of EMMPRIN and other molecules using mouse antibodies, immunofluorescence was performed as described above and coverslips were incubated with anti-EMMPRIN (VJ1/9.1) directly labeled with Alexa Fluor 488 or Alexa Fluor 594 (Zenon, Molecular Probes).

Human endothelial cells were fixed with methanol at −20°C for 2 min and stained with rabbit antibody against EMMPRIN C-terminus (GTX62657, Genetex) and mouse anti-γ-catenin or with mouse anti-γ-catenin and rabbit anti-Nm23. PLA was performed following the manufacturer's instructions (Duolink In Situ kit, Olink Biosciences). Briefly, samples were blocked with the blocking solution for 30 min at 37°C and incubated overnight with the primary antibodies at 4°C. The following day, the samples were washed and incubated for 1 h at 37°C with PLA probes for rabbit and mouse antibodies. After washing, samples were incubated with the ligase for 30 min at 37°C, washed again and incubated with the polymerase for 100 min at 37°C. Finally, the samples were washed, dried completely and mounted in Duolink In Situ Mounting Medium with DAPI.

HUVECs (3×10⁵) were seeded onto collagen-I-coated six-well plates the day before siRNA transfection. The next morning, two independent EMMPRIN siRNA oligonucleotides or negative siRNA (Ambion) were mixed with 4 µl of Oligofectamine (Invitrogen) in 400 µl of Opti-MEM and added to the cells (final concentration of 100 nM). After 4 h, 600 µl of Opti-MEM containing 30% FBS was added, and the cells were incubated overnight at 37°C. Most experiments were performed with both siRNAs, but for some experiments the results are shown for siRNA#1 only. This siRNA decreases EMMPRIN expression less efficiently than siRNA#2, but in a higher percentage of cells. Transfection of HUVECs with γ-catenin or Nm23 siRNA oligonucleotides (Ambion) followed a similar protocol. For the expression of LifeAct or the myristoylated cytosolic tail of EMMPRIN, cells were first transfected with siRNA as described and, the next day, were transfected with 2 µg of DNA in Lipofectamine 2000 (Invitrogen). Cells were analyzed at 24 h after transfection.

After siRNA-mediated interference, HUVECs were transfected with LifeAct and seeded onto collagen-I-coated glass-bottomed plates. Before imaging, cells were stained with Zenon Alexa-Fluor-488-labeled anti-EMMPRIN antibody (VJ1/9.1; Molecular Probes) for 30 min at room temperature. For time-lapse video microscopy, cells were maintained at 37°C under 5% CO₂, and Z-stacks of eight confocal sections (1.2 µM) were taken every 5 min with a Zeiss LSM 700 microscope for 20 h.

Images were acquired with 40×/1.25 or 63×/1.4 objectives on an Sp5 confocal microscope (Leica Microsystems) or with 40×/1.3 or 63×/1.4 objectives on a LSM700 confocal microscope (Carl Zeiss). Images were converted to TIFF files with LAS AF (Leica Microsystems) or ZEN (Carl Zeiss) software, respectively, and brightness and contrast were adjusted with Photoshop CS3. The colocalization of EMMPRIN with other junction proteins was analyzed with LAS AF software. Data are presented as the percentage of colocalization measured in ROIs surrounding the junction. Junctional ROIs were drawn as an area with a maximum width of 1.5–2 µm surrounding the junction. The VE-cadherin fluorescence staining pattern was quantified with ImageJ, making plot profiles within similar ROIs as in colocalization analysis. In each of three independent experiments, four to six junctions were analyzed per field in an average of five fields. Fluorescence intensities for γ-catenin, Nm23, pMLC and F-actin were also analyzed with ImageJ. Mean intensity per pixel was measured in junctional ROIs and in the rest of the cell (interior). Linearity index, defined as the ratio of junction length to the distance between vertices by Takeichi's group (Otani et al., 2006), was quantified by using ImageJ software based on VE-cadherin staining of endothelial cell junction length.

HUVEC monolayer permeability was measured with the Millipore In Vitro Permeability Assay kit. Briefly, 120,000 cells were seeded onto the membranes of the chambers provided with the kit. After 48 h, dextran–FITC was added to the upper chamber and was recovered at different time-points from the lower chamber and measured in a fluorimeter (Fluoroskan Ascent, Thermo Scientific).

Polydimethylsiloxane (PDMS) microneedle array substrates were fabricated as described previously (Fu et al., 2010). Bowtie patterns had a total area of 1600 µm². To measure forces exerted at the junctions, control HUVECs (Neg) or EMMPRIN-knockdown cells (siRNA#1 and siRNA#2) were plated over elastomer sensors with fibronectin micropatterning, fixed with 4% PFA in PBS for 15 min and stained for EMMPRIN. Tugging forces were measured as described previously (Liu et al., 2010). Briefly, microneedle deflections were calculated using Matlab to determine the displacement of the centroid of the microneedle at the tip. Forces were then computed from the known spring constant of the microneedle (7.2 nN/µm). Cells exist in a quasi-static equilibrium where the net force sums to zero (Balaban et al., 2001; Tan et al., 2003). The intercellular tugging force is therefore equal in magnitude and opposite in direction to the measured net traction force reported by the microneedle array (Cohen et al., 2013; Liu et al., 2010).

Cells were lysed in RIPA buffer for 5 min at 4°C and scraped. Lysates were cleared by spinning at 20,000 g, 4°C for 15 min, and 5× Laemmli buffer was added to the supernatant. Cell proteins were resolved by SDS-PAGE and transferred to nitrocellulose membranes. Blots were blocked overnight at 4°C in TBS-T containing 5% BSA and were incubated with primary and secondary antibodies in TBS-T for 1 h at room temperature. Signal on washed blots was detected by enhanced chemiluminescence (GE Healthcare Life Sciences). For pMLC and MLC detection, membranes from control-knockdown and EMMPRIN-knockdown human endothelial cells were isolated as detailed for ATP production assays below, and the blots were processed as described above. For co-immunoprecipitation, confluent cells were incubated for 1 h at 4°C with VJ1/9.1 mAb or an IgG1 control (2 µg each per 3×10⁶ cells). Cells were lysed and cleared as before, and the lysates were incubated with Protein-G–Sepharose (GE Healthcare Life Sciences) for 1 h at 4°C. Beads were washed six times with lysis buffer and the proteins were separated and immunodetected as above.

After siRNA-mediated interference, HUVECs were transfected with Perceval plasmid (Addgene) by the calcium phosphate method (Armesilla et al., 1999) and were analyzed in a Zeiss LSM 700 confocal microscope equipped with a 40× oil-immersion objective. For Perceval signal, excitation was set at 405 nm (ADP sensitive) and 488 nm (ATP sensitive) and detection of photomultiplier (PMT) was set at 500–560 nm (Berg et al., 2009; Zala et al., 2013). At 30 min before the experiment, human endothelial cells were stained with the directly labeled (Zenon) anti-EMMPRIN antibody to identify EMMPRIN-knockdown cells. Endothelial cells were equilibrated for 15 min before recording in the following buffer: 15 mM HEPES, 120 mM NaCl, 3 mM KCl, 2 mM CaCl₂, 1 mM MgCl₂, 1.25 mM Na₂HPO₄, 3 mM NaHCO₃, 5 mM glucose and 0.2 mM sodium pyruvate (pH 7.4). As a control for possible effects of pH variations on ATP∶ADP signal, pH was registered by adding 5 µM SNARF 5 probe (Life Technologies) in complete medium for 25 min, 1 h before the experiment. For SNARF 5, excitation was set at 555 nm and detection was set at 575–595 nm (pH sensitive) and 610–647 nm (pH insensitive). pH and Perceval (ATP∶ADP) signals were quantified by the ratio intensity of the two corresponding detection channels. To inhibit ATP production by mitochondria and glycolysis, human endothelial cells were treated with rotenone (1 µM), oligomycin (5 µM) and 2-deoxyglucose (20 mM) (all from Sigma-Aldrich). Cells were visualized, and ATP∶ADP ratios were quantified over a time range spanning from before to after treatment with the combination of inhibitors for which no variations in pH were observed. Analysis of the intensity ratios was performed with ImageJ both at the cell periphery (equivalent to junctional ROI) and over the total cellular area.

Cells were washed and treated with ice-cold hypotonic lysis buffer (10 mM Tris-HCl pH 7.4, 1.5 mM MgCl₂, 5 mM KCl, 1 mM dithiothreitol, 0.2 mM sodium vanadate and protease inhibitors) for 5 min. Cell lysates were homogenized with 30 strokes of a Dounce homogenizer and homogenates were centrifuged at 1000 g for 3 min. The supernatants were then spun at 40,000 g for 30 min at 4°C in a refrigerated centrifuge, and the crude membrane pellet was gently washed with hypotonic lysis buffer and resuspended in RIPA buffer. Membrane fractions were then assayed for total protein. ATP production by membrane-enriched fractions was measured by a kinetic luminescence assay (Vives-Bauza et al., 2007). Briefly, enriched membranes were diluted in buffer A (150 mM KCl, 25 mM Tris-HCl pH 7.4, 2 mM EDTA, 0.1% BSA, 10 mM KPO₄, 0.1 mM MgCl₂) in the presence of 150 µM ADP, 150 µM GTP, 1 µM oligomycin and 150 µM diadenosine pentaphosphate. To inhibit the reaction, anti-Nm23 antibody (Santa Cruz Biotechnology) was added at a final dilution of 1∶100.

Student's t-test was used for statistical comparisons of two groups of data. For three or more groups assuming Gaussian distribution, one-way ANOVA with Dunnet's multiple comparisons post-test was used for comparisons with controls, and Newman-Keuls test was used for comparisons between different groups of data.

We thank Roland Wedlich-Soldner [Institute of Cell Dynamics and Imaging (Max Plank), Münster, Germany] for LifeAct, Xosé Bustelo (Centro de Investigación del Cáncer, Salamanca, Spain) for myr-EMMPRIN cytosolic constructs and Simon Bartlett (Centro Nacional de Investigaciones Cardiovasculares, Madrid, Spain) for English editing.