The F-BAR protein pacsin2 inhibits asymmetric VE-cadherin internalization from tensile adherens junctions

Vascular homoeostasis, development and disease critically depend on the regulation of endothelial cell–cell junctions. Here we uncover a new role for the F-BAR protein pacsin2 in the control of VE-cadherin-based endothelial adhesion. Pacsin2 concentrates at focal adherens junctions (FAJs) that are experiencing unbalanced actomyosin-based pulling. FAJs move in response to differences in local cytoskeletal geometry and pacsin2 is recruited consistently to the trailing end of fast-moving FAJs via a mechanism that requires an intact F-BAR domain. Photoconversion, photobleaching, immunofluorescence and super-resolution microscopy reveal polarized dynamics, and organization of junctional proteins between the front of FAJs and their trailing ends. Interestingly, pacsin2 recruitment inhibits internalization of the VE-cadherin complex from FAJ trailing ends and is important for endothelial monolayer integrity. Together, these findings reveal a novel junction protective mechanism during polarized trafficking of VE-cadherin, which supports barrier maintenance within dynamic endothelial tissue.

T he endothelium is a cell monolayer that covers the luminal side of blood vessels and maintains vascular barrier function. Regulation of endothelial monolayer integrity in homoeostasis, as well as its remodelling during angiogenesis, and transendothelial trafficking of immune cells, occurs at the VEcadherin adhesion complex, the central component of endothelial cell-cell junctions 1,2 . Perturbation of the tight balance between endothelial cell-cell junction stabilization and remodelling leads to developmental defects and vascular diseases, such as chronic inflammation, oedema and atherosclerosis [3][4][5][6] .
To keep barrier integrity in vascular homoeostasis, perturbations at endothelial cell-cell junctions, induced by the highly dynamic mechanical changes that blood vessels experience 7 , have to be compensated. Maintaining endothelial cell-cell adhesion relies on the capacity of junctions to adapt to changes in local forces they experience 8 . Interactions between cells in tissue are often collectively regulated and occur via polarized cell-cell junction dynamics [9][10][11] . For instance cells that lead processes of collective migration, such as during sprouting angiogenesis, generate actomyosin pulling forces and pull along follower cells [12][13][14][15] . Importantly, cadherin-based complexes are the adhesive entities that integrate these mechanical cues from leader cells to guide follower cells 16 . It is still unclear how endothelial cells maintain vascular integrity during local differences in actomyosin-mediated pulling.
The VE-cadherin complex, which comprises the core catenins that connect the transmembrane cadherins to the actin cytoskeleton 17,18 , is regulated by mechanical forces that are either external or induced by the actomyosin cytoskeleton [19][20][21][22][23] . We have previously shown that actomyosin pulling forces induce the formation of a distinct type of cell-cell junction: the focal adherens junction (FAJ) 24 . FAJs are connected to radial pulling F-actin bundles and contain the mechanotransduction protein vinculin. Force-dependent remodelling FAJs are prominent in monolayers of endothelial tissue cultures 24 , and in the endothelium of (remodelling) blood vessels 14,15,25 . Actomyosin-derived forces not only regulate the formation of FAJs and the interaction of the VE-cadherin complex with the actin cytoskeleton, but may also alter the plasma membrane by inducing local membrane curvatures 26 . Curved membranes are recognized and regulated by proteins that contain Bin-Amphiphysin-Rvs (BAR) domains [27][28][29] . In the current study we identify, using various imaging approaches, a completely novel asymmetric signalling event at adherens junctions. We uncover that unbalanced actomyosin activity near FAJs provides an asymmetric signal for recruitment of the Fer-CIP4 homology-BAR (F-BAR) protein pacsin2, an inhibitor of endocytosis 30 , in only one of the cells taking part in the junction. Polarized internalization of VE-cadherin occurs from the junctional side, where pacsin2 is recruited. We further demonstrate that pacsin2 recruitment is important for maintenance of cell-cell adhesion by stabilizing the VE-cadherin complex within FAJs and inhibition of its internalization.

Results
F-BAR protein recruitment to force-dependent junctions. The F-BAR protein pacsin2 (also known as syndapin2) is known to regulate the actin cytoskeleton [31][32][33] and endocytosis 30,34 . While studying the role of pacsin2 in cytoskeletal remodelling during spreading of primary human umbilical vein endothelial cells (HUVECs) 33 , we observed that a subset of pacsin2 protein localizes near cell-cell junctions. This suggested that this F-BAR protein is involved in junction regulation. To investigate if pacsin2 associates with the VE-cadherin complex, we performed immunofluorescence (IF) on HUVEC monolayers and stained for endogenous pacsin2, VE-cadherin and F-actin. These IF stainings show that pacsin2 is specifically recruited to VE-cadherin-based FAJs, which appear as perpendicularly oriented junction fragments between cells 24 , but not to stable linear adherens junctions (Fig. 1a). IF stainings of pacsin2, VE-cadherin and F-actin in human dermal microvascular endothelial cells (HMEC-1 s) and human bone marrow endothelial cells (BMEC-28 (ref. 35) corroborate these findings (Fig. 1b), indicating that the association of pacsin2 with FAJs is conserved in endothelial cells from different tissue origins. In high-magnification IF images, we observed two intriguing aspects of the pacsin2-FAJ association: first, pacsin2 is present only at a subset of FAJs (Fig. 1c); and second, pacsin2 does not precisely colocalize with VE-cadherin, but concentrates asymmetrically at one side of FAJs (Fig. 1a,b,d; Supplementary Movie 1). Moreover, pacsin2 is recruited to FAJs of mesenchymal stromal and epithelial cells, respectively (Fig. 1e), extending this finding to other classical cadherin-based adhesions. To study if pacsin2 is recruited to endothelial junctions in human vasculature, we applied IF on vessels that were isolated from healthy mesentery sections from colon resections 25 . En face confocal microscopy shows that pacsin2 is recruited to endothelial adherens junctions in human arteries (Fig. 1f). Subsequently, we investigated which other pacsin isoforms might be expressed in endothelial cells. Western blot analysis shows that small amounts of pacsin3, but not of pacsin1, are present in endothelial cell lysates ( Supplementary Fig. 1a). IF experiments further indicate that endogenous endothelial pacsin3 is not detectable at FAJs ( Supplementary Fig. 1b), suggesting that pacsin2 is the dominant pacsin isoform present at endothelial junctions. To investigate if the junctional recruitment of pacsin2 is specific for this F-BAR protein, we next analysed the distribution of nostrin, another F-BAR protein expressed in the endothelium 36 , and found that just like pacsin2, nostrin is recruited specifically to FAJs (Fig. 1g). In contrast to the recruitment of pacsin2 to a subset of FAJs, nostrin is recruited to almost all FAJs in HUVECs (95.0% ± 2.5 s.e.m.; Fig. 1g).
FAJs are not only phenotypically distinct from stable linear adherens junctions, but are also distinguished by the presence of vinculin, which interacts with the VE-cadherin-a-catenin complex upon force-induced tension 23,24 . To investigate the association of pacsin2 and vinculin with VE-cadherin at FAJs, we performed triple IF stainings. These experiments clearly show that pacsin2 and vinculin are both specifically present at FAJs (Fig. 1h). However, we observed that pacsin2 was recruited to the VE-cadherin-vinculin complex only at a fraction of FAJs (Fig. 1c,h), whereas vinculin is present at all FAJs 24 . Moreover, vinculin distribution at FAJs compares to that of VE-cadherin, whereas pacsin2 is asymmetrically present at one side of FAJs (Fig. 1d). These data argue against a direct force-induced molecular interaction of pacsin2 with the VE-cadherin-acatenin complex itself, but indicate that pacsin2 is recruited to FAJs via a different mechanism. Because the F-BAR protein nostrin is recruited to FAJs as well, we propose that specific membrane curvature at FAJs recruits pacsin2. Together, these data reveal that F-BAR proteins are recruited to force-dependent remodelling junctions in vascular endothelium, as well as in other cell types that form cell-cell adhesions via classical cadherins.
Unbalanced actomyosin activity at FAJs recruits pacsin2. To understand how pacsin2 is recruited to cell-cell junctions, we first investigated the formation of FAJs by increasing cytoskeletalderived tension on junctions, following thrombin treatments as described previously 24 . Strikingly, during the formation of FAJs by thrombin, we observed a decrease in the fraction of pacsin2-positive FAJs compared with control monolayers (Fig. 2; Supplementary Movie 2). Inhibition of cytoskeletal tension by using the Rho-kinase inhibitor Y-27632 also resulted in significantly fewer pacsin2-positive FAJs ( Fig. 2; Supplementary Movie 3). These results suggest that the universal increases or decreases in tension on junctions in monolayers do not trigger pacsin2 recruitment. Next, we investigated pacsin2 recruitment to FAJs during polarized cell migration in scratch wound assays. IF  pMLC-Ser19) showed that the near junctional distribution ratio of pMLC-Ser19 is unbalanced between cells sharing pacsin2positive FAJs (Fig. 3b). The average ratio between pMLC-Ser19 intensity at the pacsin2-negative side and the intensity at the opposite side is 2.46±0.34 s.e.m. Thus, pacsin2 is recruited to the FAJ side opposite from the highest myosin-II-based pulling activity (Fig. 3b). These results indicate that differences in the geometry of the near junctional actomyosin network between cells relates to pacsin2 recruitment. The local organization of actomyosin at junctions is tightly controlled by the activation of small GTPases [37][38][39] . To study if forcing the actin cytoskeleton into an asymmetric geometry near FAJs directly regulates pacsin2 recruitment, we ectopically expressed an activated mutant of the GTPase Rac1 (Rac1[Q61L]) in HUVECs, and analysed their FAJs formed with untransfected cells in mosaic cultures. We find that pacsin2 is preferentially recruited to the FAJ side of cells expressing activated Rac1 (Fig. 3c), whereas pacsin2 recruitment in cells ectopically expressing wild-type Rac1 does not localize at a preferred FAJ side (Fig. 3c). Together, these data show that local asymmetric organization of the actomyosin cytoskeleton induces pacsin2 recruitment to FAJs. We next addressed the mechanism that underlies the asymmetric presence of pacsin2 at junctions. A curve of the double membrane at AJs is intracellularly perceived both as convex and concave shapes, depending from which cellular side of the junction this is sensed. To investigate if the recruitment of pacsin2, known to interact with convex membranes 33,40 , to FAJs is derived from one or both cells forming the junction, we monitored dynamics of mosaic endothelial monolayers, in which half the cell population expresses lentivirally transduced pacsin2-green fluorescent protein (GFP) and the other half pacsin2-mCherry. These studies show that the presence of pacsin2 at FAJs is either GFP or mCherry fluorescent ( Fig. 3d; Supplementary Movie 4), indicating that pacsin2 recruitment to junctions occurs from one side of the cell-cell junction.
Pacsin2, like many other F-BAR proteins, contains an F-BAR domain that is crucial for its binding to phospholipids in curved membranes and an SH3 domain, which interacts with effector proteins 27 . To further characterize its junctional recruitment mechanism, we expressed pacsin2 variants in HUVECs. We first tested if the pacsin2 F-BAR domain is involved in its asymmetric recruitment to FAJs. We expressed a GFP-tagged pacsin2-R50D mutant, which can no longer recognize convex membrane curvature, and a double mutant pacsin2[M124,125E], which cannot induce membrane tubulation 33,40 , in HUVECs. In particular the pacsin2[R50D] mutant is no longer concentrated at FAJs (Fig. 4a). Next, we expressed a truncated variant of pacsin2 (pacsin2-D306), which contains the complete F-BAR domain, but lacks Asn-Pro-Phe (NPF) motifs and the SH3 domain. Also pacsin2-D306 protein is significantly perturbed in its capacity to concentrate at FAJs. Finally, expression of an SH3-domain-mutated pacsin2[P478L] did not affect concentration of pacsin2 at FAJs (Fig. 4a). These results indicate that the pacsin2 F-BAR domain is crucial, but not sufficient, for its recruitment to FAJs.
To assess pacsin2 recruitment in relation to FAJ remodelling, we performed live imaging of HUVECs with lentiviral expression of pacsin2-GFP and a-catenin-mCherry (as marker for adherens junctions). Imaging of these transduced cells indicates that when pulling activity between cells is unbalanced, FAJs themselves displace in the direction of the cell that pulls away. This provides FAJs with a front (that is, the junction side of increased pulling) and a trailing end that follows the movement of the FAJ. Pacsin2 is consistently recruited to FAJ trailing ends ( Fig. 4b; Supplementary Movies 4 and 5). We confirmed these findings in live cell imaging experiments of HUVECs expressing VE-cadherin-GFP and pacsin2-mCherry, or pacsin2-GFP and p120-catenin-mCherry (Supplementary Movies 6 and 7). These experiments also show that pacsin2 recruitment is highly dynamic; its presence at junctions is not constitutive, but relates to the movement of FAJs during unbalanced remodelling. We find that the increased speed of FAJ movement significantly correlates with a higher concentration of pacsin2 at FAJs (Fig. 4b). Moreover, live imaging of cells expressing pacsin2[R50D]-GFP and a-catenin-mCherry confirms that a functional F-BAR domain is crucial for the recruitment of pacsin2 to fast-moving FAJs (Supplementary Movie 8).
Taken together, these data show that asymmetric remodelling of FAJs, mediated by local actomyosin changes, induces pacsin2 recruitment to the trailing ends of FAJs in an F-BAR domain-dependent manner.
Association of pacsin2 with VE-cadherin at super-resolution. Because (i) recruitment of pacsin2 to FAJs occurs via a different mechanism than recruitment of vinculin, and (ii) pacsin2 recruitment depends on a functional F-BAR domain, which is known to interact with the plasma membrane, we next used stimulated emission depletion (STED) super-resolution microscopy to further dissect the molecular organization of FAJs. First, we analysed the colocalization of endogenous VE-cadherin with the core catenin proteins in STED images and, as expected, found a high Pearson's correlation coefficient (R) of these proteins with VE-cadherin at FAJs: a-catenin (R ¼ 0.72), b-catenin (R ¼ 0.77) and p120-catenin (R ¼ 0.64; Fig. 5a,b). Localization of vinculin also correlates well with that of VE-cadherin at this resolution (R ¼ 0.53). By contrast, pacsin2 molecules show a significantly lower correlation with VE-cadherin (R ¼ 0.16) and the super-resolution imaging reveals that pacsin2 is decorating the invaginating ends of FAJs (Fig. 5a,b). This detailed structure is not discernible in lower optic resolution imaging such as by conventional widefield (for example, Fig. 1a) or confocal imaging (Fig. 5a). Thus, the super-resolution imaging revealed that pacsin2 is not part of the VE-cadherin complex, but decorates around the VE-cadherin-based adhesion structure instead, suggesting this may be the location of curved membranes at asymmetric remodelling FAJs.
Polarized dynamics of VE-cadherin at asymmetric FAJs. The previous experiments suggest the existence of a mechanism for polarized regulation of AJs in response to unbalanced cytoskeletal pulling between cells. The recruitment of pacsin2 associates with this polarization. To understand the consequences of asymmetric regulation of VE-cadherin-based junctions, we first expressed and visualized a fusion of a-catenin with the Dendra2 green-to-red photoswitchable fluorescent protein. Next, we photoswitched a-catenin-Dendra2 at the front of asymmetric FAJs. These experiments show that a-catenin molecules photoconverted to Dendra2-red at the front of FAJs move to the trailing ends of FAJs over time ( Fig. 6a; Supplementary Movie 9). To investigate if there is differential mobility of junctional proteins in distinct parts of FAJs, we next performed fluorescence-recovery-  after-photobleaching (FRAP) experiments of a-catenin-mCherry at pacsin2-positive FAJs. We compared a-catenin recovery at the front of pacsin2-positive FAJs, to its recovery at the trailing ends. Whereas fluorescence intensity of a-catenin-mCherry at the front of junctions readily recovers after bleaching, we find a marked reduction in the recovery of a-catenin at the side of FAJs that recruited pacsin2 (Fig. 6b,c; Supplementary Movie 10). To quantify these results, we determined the average fluorescence intensity values (corrected for overall bleaching and normalized to fluorescence intensity before (100%) and after (0%) bleaching) of multiple FRAP experiments. These data indicate that the mean mobile fraction of a-catenin-mCherry at the front of pacsin2-positive FAJs (65.5% ± 4.2 s.e.m.) is significantly higher than at their trailing ends (38.1%± 4.9 s.e.m.; Fig. 6d). In contrast, the average fluorescence recovery of a-catenin-mCherry at the front and trailing ends of pacsin2-negative FAJs is comparable (Fig. 6c,d). These data reveal a correlation between pacsin2 recruitment and local stabilization of a-catenin at the trailing ends of polarized FAJs. This is consistent with an inhibitory role of pacsin2 in endocytosis 30 . We further conclude that pacsin2 demarcates the junction side that is lacking nascent formation of VE-cadherin-based adhesions.
Pacsin2 and asymmetric internalization of VE-cadherin. To understand the dynamics of VE-cadherin at the trailing ends of FAJs, we further investigated endogenous VE-cadherin and pacsin2 proteins in HUVEC, HMEC-1 and BMEC-28 monolayers. Since there is differential mobility of a-catenin, a crucial element of the VE-cadherin complex, between the front and trailing ends of pacsin2-positive FAJs, and because pacsin2 is known to inhibit internalization from the plasma membrane 30 , we next analysed the fluorescence intensities of VE-cadherin and pacsin2 across individual FAJs. Linescan analysis demonstrates that pacsin2 recruitment at the trailing ends of FAJs marks the junctional border, where the level of VE-cadherin protein declines ( Fig. 7a), corresponding with their asymmetric difference in distribution along FAJs (Fig. 1d). To assess whether FAJ trailing ends are sites where VE-cadherin is being internalized, we live-labelled HUVEC monolayers with an Alexa-Fluor-647conjugated non-blocking anti-extracellular VE-cadherin antibody (mouse monoclonal 55/7H1 (ref. 41). Two hours after the labelling, cells were fixed and stained for the surface VE-cadherin pool with another anti-extracellular VE-cadherin antibody (monoclonal 75; distinct fluorescence wavelength). IF images of these cells show that the pre-labelled VE-cadherin pool (representing those molecules that were on the surface 2 h before fixation) is concentrated at FAJ trailing ends (Fig. 7b). Under these conditions, the staining with clone 75 antibody was higher at FAJ fronts, suggesting that a significant portion of the VE-cadherin molecules is internalized from the cell surface at FAJ trailing ends (Fig. 7b). To investigate if pacsin2 recruitment correlates to the side of FAJs, where the majority of pre-labelled VE-cadherin molecules locate, we performed IF experiments in which pre-labelled surface VE-cadherin stainings were followed by permeabilization and stainings for pacsin2. These images show higher intensities of pre-labelled VE-cadherin than the surface pool of VE-cadherin at the pacsin2-positive trailing ends of FAJs (Fig. 7c). Of note, we noticed even more enrichment of pre-labelled VE-cadherin molecules at the trailing ends of pacsin2-negative FAJs (Fig. 7c). This is consistent with a stabilizing effect of pacsin2 on the mobility of a À catenin at the trailing ends of FAJs (Fig. 6d). Live imaging of HUVECs expressing pacsin2-GFP and a-catenin-mCherry further reveals that in some cases pacsin2 is recruited before internalization events from the trailing ends of FAJs. Internalization from those FAJs is promoted once pacsin2 dissociates from the junction ( Fig. 8a; Supplementary Movie 11). Taken together these data indicate that pacsin2 is recruited to the internalization side of FAJs and that the presence of pacsin2 at FAJs correlates with suppressed levels of internalized VE-cadherin compared with pacsin2-negative FAJs.
To investigate the subcellular compartments to which internalized VE-cadherin is trafficked, we analysed the localization of surface-chased VE-cadherin molecules. We find that surface-chased VE-cadherin molecules culminate in Rab5 GTPase-positive endosomes (Fig. 8b,g), in line with other recent studies which show that endocytosis directs VE-cadherin to early endosomes 42,43 . Moreover, we find that a fraction (43% ± 2 s.e.m.) of surface-chased VE-cadherin is contained in Rab4-positive endosomes (Fig. 8c,g), but not in Rab11 recycling endosomes (Fig. 8d,g). The presence of VE-cadherin in Rab4-positive vesicles was confirmed in IF stainings for total endogenous VE-cadherin (Fig. 8e,g). Moreover, live cell imaging of VE-cadherin-red fluorescent protein (RFP) and Rab4-GFP shows that internalized VE-cadherin molecules that derive from the trailing ends of FAJs are directed towards Rab4-positive vesicles (Supplementary Movie 12). Colocalization analysis of surface-chased VE-cadherin molecules with the vascular endothelial growth factor receptor 2, known as cargo of recycling endothelial vesicles 44 , confirmed that VE-cadherin is targeted to an endosomal recycling compartment (Fig. 8f,g).
We next investigated what portion of internalized VE-cadherin molecules is intrinsically endocytosed, or alternatively, might be exo-endocytosed from neighbouring cells 43,45 . VE-cadherin-GFP was expressed in a sparse fraction of endothelial cells (donor cells) and VE-cadherin internalization in neighbouring adhered non-transduced cells (acceptor cells) was monitored. Subsequently, we surface-chased VE-cadherin for 2 h using Alexa-647-labelled 55/7H1 antibody and determined the number of GFP þ /Alexa647 þ versus GFP -/Alexa647 þ vesicles in acceptor cells. These experiments show that 81%    (Fig. 8h). Taken together, these data demonstrate that the VE-cadherin complex is internalized from asymmetric remodelling FAJs and transported to endosomal compartments.
Pacsin2 controls VE-cadherin and endothelial barrier. To assess the role of pacsin2 in endothelial cell-cell adhesion, we next performed knockdown of pacsin2 protein expression by lentiviral expression of short hairpin RNAs (shRNAs). Two of the five tested shRNA constructs induced robust pacsin2 knockdown in HUVECs (Supplementary Fig. 2a). We detected no changes in total VE-cadherin protein levels in lysates of HUVECs depleted for pacsin2 (Fig. 9a) nor on total surface levels of VE-cadherin as determined by fluorescence-activated cell sorting (FACS; Fig. 9b). IF imaging showed that shPacsin2 knockdown cells still form VE-cadherin-based adherens junctions, including FAJs, as well as tight junctions, and we did not observe major phenotypic changes of endothelial monolayers ( Fig. 9c; Supplementary Fig. 2b).
To address whether pacsin2 controls VE-cadherin internalization, we first compared the fate of surface-chased VE-cadherin between shControl and shPacsin2 cells. These experiments clearly show that in the absence of pacsin2, increased levels of surface-chased VE-cadherin accumulate in internalized vesicles (Fig. 9d). Internalization of VE-cadherin is associated with the appearance of a 100 kDa fragment of VE-cadherin 46 . Western blot analysis of VE-cadherin in lysates of shControl or shPacsin2 transduced HUVECs indicate an increased quantity of the 100 kDa fragment in the absence of pacsin2 (Fig. 9e). Together these experiments evidently demonstrate that pacsin2 inhibits internalization of the VE-cadherin complex.
To investigate whether pacsin2 has a role in the polarized trafficking of a-catenin at FAJs, we next studied the mobility of a-catenin-mCherry at the trailing end of FAJs by FRAP in shControl and shPacsin2 transduced HUVECs. We find that the average mobile fraction of a-catenin-mCherry at the trailing end of FAJs is significantly higher in shPacsin2 HUVECs (51% ± 3.5 s.e.m.) than in shControl HUVECs (35%±1.6 s.e.m. ; Fig. 9f). These experiments indicate that pacsin2 controls the polarized trafficking of the VE-cadherin complex at FAJs.
To functionally investigate the role of pacsin2 in endothelial cell-cell junctions, we analysed barrier function by measuring endothelial resistance using electric cell-substrate impedance sensing (ECIS) across endothelial cells transduced with pacsin2 or control shRNA. These experiments show that shPacsin2 cells have a significantly reduced capacity to form endothelial barriers, which is apparent both during cell-cell junction formation between recently plated cells and within established monolayers (4 and 16 h after seeding, respectively; Fig. 9g). To investigate which domains of pacsin2 are important for its barrier protective function, we knocked down pacsin2 in HUVECs expression by means of an shRNA targeting the 3 0 -untranslated region (UTR) of the endogenous pacsin2 messenger RNA (the shPacsin2 3 0 -UTR plasmid also contains a TagRFP complementary DNA). In parallel, we re-expressed GFP-tagged pacsin2, pacsin2[R50D] or pacsin2[P478L]. After transduction and puromycin selection, cells were sorted by FACS for dual GFP and TagRFP positivity. Western blot analysis shows efficient knockdown of endogenous pacsin2 and expression of the rescue constructs (Fig. 9h). Next, we investigated the capacity of these cell types to form an endothelial barrier using ECIS. These experiments confirm that pacsin2 is needed for efficient junction formation. Rescuing the shPacsin2 3 0 -UTR expressing cells with pacsin2-GFP significantly restores the barrier forming capacity, whereas expressing the pacsin2[R50D] mutant, which fails to concentrate at FAJs (Fig. 4a), does not mediate this effect (Fig. 9i). Expression of the pacsin2[P478L] mutant, which does become recruited to FAJs (Fig. 4a), rescues the barrier formation to a certain extent (Fig. 9i). Taken together, these data indicate that pacsin2 is required for efficient barrier formation and stabilization in an F-BAR domain dependent manner. The data point towards a protective role for F-BAR proteins during asymmetric remodelling of endothelial junctions in vascular barrier maintenance.

Discussion
Many recent studies underscore the importance of the activity and organization of the actomyosin cytoskeleton for endothelial cell-cell junction dynamics (reviewed in ref. 47). The regulation of junctions is crucial for development and immune surveillance, and its involvement in vascular disease is unmistakable. This study provides evidence for a novel mechanism of protection against asymmetric internalization of the VE-cadherin complex that follows unbalanced force-induced junction remodelling and is important for endothelial barrier function. Our data show that upon imbalanced pulling activity between cells (that is, when one cell pulls stronger than the other), the cell-cell junctions adopt an asymmetric geometry with a junction front where new VEcadherin complex adhesions are formed, and a trailing end from which the VE-cadherin complex is internalized. We demonstrate that pacsin2, which is recruited to the trailing ends of a subset of FAJs, is important to maintain cell-cell adhesion during their rapid asymmetric remodelling (Fig. 10).
We identify the F-BAR protein pacsin2 as a novel regulator of endothelial cell-cell adhesion. Its recruitment to, and function at, FAJs involves its specific ability to recognize convex membrane curvature 40 . A potential alternative mechanism for its recruitment to junctions is via specific conformations of the F-actin cytoskeleton that is distinctive at FAJs 48 , and might enable a direct interaction of F-actin with the pacsin2 F-BAR domain 31 . Our super-resolution studies revealed that pacsin2 is covering around the trailing ends of FAJs and is not precisely colocalizing with the F-actin-binding proteins a-catenin and vinculin, which favors a model in which plasma membrane curves recruit pacsin2 to FAJs.
We find that pacsin2 is recruited to a subset of polarized junctions, where internalization takes place. The comparison of VE-cadherin internalization in shControl and shPacsin2 transduced cells clearly show that once pacsin2 is functionally perturbed, there are increased levels of overall internalized VE-cadherin. This increase in internalization of VE-cadherin is associated with a decreased endothelial barrier capacity, which corresponds with the notion that VE-cadherin-based adherens junctions are crucial for endothelial barrier function. The FRAP experiments comparing pacsin2-positive and pacsin2-negative FAJs indicate that once recruited to the trailing ends of FAJs, pacsin2 locally stabilizes a-catenin. This may explain the protective role of pacsin2 during junction internalization and endothelial barrier function. Others previously reported that the presence of pacsin2 at the plasma membrane inversely correlates with the recruitment of dynamin 2, a protein that mediates membrane scission 49 . Possibly, a similar competing mechanism with dynamin is responsible for the inhibitory role pacsin2 has during VE-cadherin internalization from FAJs.
Our data shows that pacsin2 is not essential for the polarized movement of VE-cadherin molecules along FAJs or their internalization. This is visible, for example, in Supplementary Movie 11, which captured the internalization process from FAJs.
Pacsin2 is not continuously covering the FAJ and internalization from that FAJ is promoted once pacsin2 dissociates from the junction trailing end. In fact, in cells where pacsin2 is functionally perturbed by shRNAs, VE-cadherin internalization is strongly enhanced. Importantly, when we observe junctional recruitment of pacsin2 we consistently find it at the trailing end, where the internalization of VE-cadherin molecules occurs, and not at the FAJ front where new cell-cell adhesion complexes are being formed. Because we find a correlation between increased pacsin2 recruitment during the moments of faster polarized FAJ movements, we speculate that the pacsin2 negative, but polarized junctions, might have been pacsin2 positive earlier in their lifetime. Taken together, the data suggest that pacsin2 inhibits VE-cadherin internalization from those faster moving FAJs.
Because pacsin2 is ubiquitously expressed 30,34 , and we detect it also at E-cadherin-and N-cadherin-based-FAJs, we surmise that pacsin2 serves similar functions in other cell types that form classical cadherin-based cell-cell junctions. Recently, in Caenorhabditis elegans, the F-BAR protein SRGP-1 (an srGAP orthologue) was shown to regulate membrane dynamics during cell-cell junction formation, and its function is needed for efficient cadherin-based cell-cell adhesion 50 . Two other F-BAR proteins have been implicated in regulating cell-cell adhesion in Drosophila; the F-BAR proteins CIP4 and nostrin also appear to function as cadherin internalization suppressors 51  of the VE-cadherin complex. Of note, it is not yet clear whether these other F-BAR proteins are specifically recruited to certain cell-cell junction conformations, or whether they are controlled by actomyosin activity in these model systems. In epithelial cell types, depletion of junctional CIP4 impaired E-cadherin trafficking, and inhibited EGF-induced cell scattering and invasion of epithelial cancer cells 52 . Moreover, increased CIP4 protein expression levels were associated with the increased tensile adherens junctions and invasiveness of human breast cancer 52 . Together with our experiments, these data point towards tailored mechanisms regulating endothelial and epithelial cell-cell junctions by F-BAR proteins. Intriguingly, we find that nostrin is present at almost all endothelial FAJs, whereas pacsin2 is present at a subset of FAJs. There could be various reasons for this difference: there might be distinct expression levels of nostrin and pacsin2, the presence of pacsin2 at FAJs might be more transient than nostrin's or another possibility is that the degree of curvature of the F-BAR dimers is slightly different and there are more FAJs that fit the structure of nostrin. Even so, both F-BAR proteins are preferentially recruited to FAJs, and not to linear adherens junctions, which suggests that the membrane at FAJs adopts specific conformations that are recognized by these F-BAR proteins.
One of the key events implicated in inflammation, as well as in mechanical-induced remodelling of cell-cell junctions, is the trafficking of the VE-cadherin complex. Internalization of VEcadherin occurs in response to permeability factors (for example, vascular endothelial growth factor and bradykinin), actomyosin contraction, changes in (hemodynamic) forces or transendothelial migration of immune cells 43,[53][54][55] . Because we find that in the absence of pacsin2 the overall internalization of the VE-cadherin complex is increased, and endothelial barrier function is perturbed, we propose that pacsin2 recruitment is part of a signalling pathway that protects junctions during asymmetric dynamic events. Tensile adherens junctions are essential structures that orchestrate collective behaviour of cells in a monolayer 56 , and polarized internalization and recycling of cadherin-based adhesions underlies this phenomenon 9 . Polarized trafficking at endothelial junctions likely feeds into pathways that determine the ability of endothelial cells to adapt to the rapid changes in adhesion, pulling forces and repulsive cues 57 . Although, we observe that VEcadherin molecules internalized from FAJs are ending up in Rab4positive vesicles, it remains to be addressed whether pacsin2 has also a direct role in trafficking towards recycling compartments that functionally take part in newly forming cell-cell adhesions.

Methods
Cells and tissue. Pooled HUVECs (cultured up to passage six) from different donors (purchased from Lonza), immortalized HMEC-1 (described in ref. 58) and human BMEC-28 (described in ref. 35) were cultured in EBM-2 culture medium supplemented with EGM-2 BulletKit (Lonza) on gelatin-coated tissue flasks. HEK293T, HeLa and epithelial DU145 cells (purchased from American Type Culture Collection (ATCC)) were cultured in Iscove's modified Dulbecco's medium with L-glutamine (Lonza) supplemented with 10% fetal calf serum and antibiotics. Bone marrow-derived human mesenchymal stromal cells, a kind gift from Dr Carlijn Voermans, were isolated and cultured as described previously 59 . Human blood vessels were isolated from tissue that remains after pathology analysis of mesentery from patients that underwent intestinal tumour resection in the Academic Medical Center (Dr Mat Daemen, Amsterdam, The Netherlands) as described previously 25 . All vessels were obtained with informed consent and according to the Dutch guidelines for secondary used patient material.
DNA constructs and lentiviral transductions. For lentiviral transductions, human pacsin2 fused at its C terminus to a GFP tag was amplified by PCR from a pcDNA6.2/C-EmGFP-DEST-pacsin2 vector (provided by Dr Ronen Zaidel-Bar) and cloned into a self-inactivating lentiviral pLV-CMV-ires-puro vector between the SnaBI and NheI restriction sites to generate pLV-CMV-pacsin2-GFP, and the pLV-CMV-pacsin2-D306 variant that expresses only its F-BAR domain: amino acids 1-306. The pLV-CMV-pacsin2-GFP construct was modified by replacing GFP with mCherry DNA sequence to generate pLV-CMV-pacsin2-mCherry. An arginine (CGC) to aspartic acid (GAC) mutation at amino acid position 50 in the pacsin2 F-BAR domain was generated by site-directed mutagenesis to generate pLV-pacsin2-[R50D]-GFP. A membrane localization mutant was generated by mutating methionine at position 124 and 125 (ATG ATG) to glutamic acid (GAG GAG). An SH3 mutant was generated by mutating proline at position 478 (CCG) to leucine (CTG) in pacsin2. Mouse a-catenin-mCherry and human Rab4a-GFP were cloned into the pLV-CMV-ires-puro vector using NdeI and NheI restriction sites. The lentiviral expression constructs pLV-a-catenin-Dendra2, pLV-VE-cadherin-GFP and pLV-p120-catenin-mCherry have been previously described 24 . Mammalian expression vectors for myc-tagged pacsin1, and pacsin3-and mCherry-tagged Rac1 wild type and Q61L, were previously reported 33 . Transient transfection of these constructs was performed with TransIT-LT1 (Mirus) according to the manufacturers' standard protocol. Pacsin2 expression was knocked down using shRNA expression plasmids from The RNAi Consortium (TRC) library 60 . shRNA encoding pLKO.1 lentiviral vectors targeting human pacsin2 were MISSION TRC1 clones 0000037979, 0000037980, 0000037981, 0000037982 and 0000037983 from Sigma-Aldrich. The 0000037980 and 0000037983 constructs induced most efficient knockdowns in HUVECs. Non-targeting shRNA (SHC002; Sigma-Aldrich) was used as a negative control. For rescue experiments a short hairpin directed against the 3 0 -UTR of human pacsin2, shPacsin2-TagRFP, was designed. The shRNA was created by allowing two oligo's containing the shRNA sequence (5 0 -CCGGTGTCATGTCTCAGT GTCTATCTCTCGAGAGATAGACACTGAGACATGACTTTTTG-3 0 ) to selfligate and subsequently insert the shRNA into a modified version of the pLKO.1 backbone between the AgeI and EcoRI restriction sites. Lentiviral particles were isolated from the supernatant of HEK293T cells transiently transfected with thirdgeneration packaging constructs and the lentiviral expression vectors. HUVECs, cultured to 80% confluency, were infected with supernatant containing lentiviral particles overnight. Cells were analysed at least 48 h after transduction. To obtain cells expressing both shPacsin2-TagRFP and pacsin2 rescue constructs, cells were sorted on FACSARIATM III cell sorter (BD) for TagRFP and GFP dual positivity.
IF microscopy. For standard IF staining, cells were cultured on coverslips coated with 5 mg ml À 1 fibronectin and subsequently fixed in 4% paraformaldehyde in PBS   (Fig. 7b,c) or indicated endogenous proteins. For the IF stainings in Fig. 9d, HUVECs were pre-labelled with 55/7H1 antibody in EGM-2 medium at 4°C for 30 min and internalization was followed after 2 h of culturing the cells at 37°C. To visualize intracellular proteins, VE-cadherin-labelled cells were permeabilized in 0.5% Triton X100 in PBS for 5 min and stained according to the standard IF protocol. Mesenteric arteries were stored in PBS þ þ at 4°C for no longer than 24 h after surgery until further preparation as described previously 25  Super-resolution microscopy. STED microscopy was performed on a Leica TCS SP8 STED 3X microscope. HUVEC monolayers were stained using the following antibodies: rabbit polyclonal anti-pacsin2 (Abcam), mouse monoclonal anti-vinculin hVIN-1 (Sigma), mouse monoclonal anti-p120-catenin clone 98/pp120 (BD Bioscience), rabbit anti-a-catenin and anti-b-catenin serum (Sigma) combined with anti-VE-cadherin mouse monoclonal F-8 (Santa Cruz) or rabbit polyclonal (Cayman) antibodies. Secondary antibodies used were STAR 580 and STAR 635P (Abberior). Samples were irradiated with a pulsed white light laser at wavelengths 587 and 633 nm. A pulsed STED laser line at a wavelength of 775 nm was used for the depletion of the 587-and 633-nm fluorophore, reaching a lateral resolution of B45 nm to distinguish individual fluorescent signals in these IF stainings. The signal was detected using a gated hybrid detector in counting mode. STED images were acquired using a dedicated 100 Â 1.4 NA oil objective. Finally, deconvolution was performed with Huygens Professional (Scientific Volume Imaging).
Western blot analysis. Cell lysates were prepared in reduced sample buffer and analysed by standard western blotting and enhanced chemiluminescence detection. Uncropped scans of blots are presented in Supplementary Fig. 3.
Flow cytometry. For flow cytometry analysis in Fig. 9b, cells were collected with Accutase cell detachment solution (GE healthcare), blocked in PBS containing 1% BSA, incubated with anti-surface VE-cadherin antibody using Alexa-Fluor-647conjugated mouse monoclonal 55/7H1 for 45 min at 4°C, and subsequently analysed on a FACS CANTO II (Becton Dickinson).
Electric cell-substrate impedance sensing. Endothelial barrier function was determined by measuring the electrical resistance of monolayers by ECIS. Electrode arrays (8W10E; IBIDI) were treated with 10 mM L-cysteine (Sigma) for 15 min at 37°C and subsequently coated with 10 mg ml À 1 fibronectin (Sigma) in 0.9% NaCl for 1 h at 37°C. Cells were seeded and electrical resistance was continuously measured during monolayer formation at 30 kHz using ECIS model 9600 (Applied BioPhysics) and 37°C at 5% CO 2 incubator conditions. Resistane values were corrected for background signal that was measured in ECIS wells without cells (274 Ohm, n ¼ 2). In Fig. 9i, values are normalized to the resistance of shControl cells. The barrier capacity during junction formation was determined at 4 h after cell seeding, and for established monolayers at 16 h after cell seeding.
Image and statistical analysis. For evaluations of fluorescence signals along linescans in IF images (Fig. 1a,d,g and 4a,b), fluorescence integrated intensities was measured along junctional linescans, corrected for the integrated intensity along linescans in the cytoplasm in direct proximity of the junction. A measure of 8 mm linescans were made along random junctions in the original imaging data using ImageJ. The categorization of FAJs and linear adherens junctions was based on phenotypic organization of VE-cadherin signal. FAJs are defined as interrupted and perpendicularly oriented junctions between cells. For symmetric ratio analysis at FAJs (Figs 1d and 6a), the background-corrected integrated fluorescence intensities along linescans were split into two halves of which the values of the trailing end side of junctions were divided by the values of their opposite side. A ratio of 1 represents symmetric distribution along FAJs. The evaluation of the percentage±s.e.m. of FAJs that are pacsin2-or nostrin-positive was calculated manually in widefield images (Figs 1c,g and 2). The orientation angle of pacsin2 at FAJs, in relation to the front of the scratch wound or the bottom of the image in control monolayers (Fig. 3a), was obtained by linescans from the front to the pacsin2-positive trailing end along every pacsin2-positive FAJ within a distance of four cell rows from the scratch front. To determine the average near junctional intensities of activated myosin light chain (Fig. 3b), the local integrated fluorescence instensity of p-MLC Ser19 signal was measured in fixed square areas of 18 mm 2 at the pacsin2-positive versus pacsin2-negative side of FAJs. The values are corrected for background and normalized to maximal intensities in the overall image. To determine the speed of FAJ movement (Fig. 4b), FAJ trailing ends were tracked using the MtrackJ plugin of ImageJ in time-lapse recordings of 150 s. Pearson's split colocalization coefficients (Fig. 5b) were calculated using ImageJ.