Putting VE-cadherin into JAIL for junction remodeling

Both the vascular endothelial cells (ECs) and the epithelium exhibit tissue-specific phenotypes, with significant differences in differentiation and cell shape, as well as cell junction organization and dynamics. The vascular endothelium is composed from ECs that are connected to each other via cell junctions; they form a highly dynamic and plastic interface between the blood stream and the interstitial tissue, controlling exchange of water and solutes. EC junctions also play a critical role in regeneration, vasculogenesis and angiogenesis, inflammation and tumor progression (for reviews, see Vestweber, 2012; Fromel and Fleming, 2015; Kazmi et al., 2015; Tesfamariam, 2016; Zou et al., 2016; Filipowska et al., 2017; Godo and Shimokawa, 2017; Szymborska and Gerhardt, 2017; Choi and Moon, 2018; Hida et al., 2018). ECs and EC junctions undergo a specific and different differentiation program in organs and vascular segments (Simionescu et al., 1975, 1976a,b, 1983; Corada et al., 1999; Dejana et al., 2008; Frye et al., 2015; Augustin and Koh, 2017); thus a heterogeneous regulation and dynamics can be assumed (Trepat and Fredberg, 2011). EC adherens junctions are complex structures (Lampugnani, 2010; Dejana and Orsenigo, 2013; Komarova et al., 2017), in which the Ca²⁺-dependent vascular endothelial (VE)-cadherin (also known as CDH5) is a key protein. It mediates physical adhesion between cells and is involved in signaling and transcription control. VE-cadherin is critical in vascular and angiogenesis, and the control of barrier function in health and disease (Wallez et al., 2006; Harris and Nelson, 2010; Dejana and Vestweber, 2013; Giannotta et al., 2013; Bentley et al., 2014; Lagendijk and Hogan, 2015; Vestweber, 2015; Li et al., 2018). Recent work further indicates that VE-cadherin acts as a mechanosensor (Tzima et al., 2005; Grashoff et al., 2010; Coon et al., 2015; Lagendijk et al., 2017), underpinning its central importance in the vasculature. VE-cadherin consists of an extracellular domain, a short helical transmembrane domain and a C-terminal domain that provides binding sequences for p120-catenin (p120^ctn), β-catenin and γ-catenin [also known as CTNND1, CTNNB1 and plakoglobin (JUP), respectively] (Suzuki et al., 1991; Lampugnani et al., 1995). A protein chain consisting, in order, of VE-cadherin, β-catenin and one of the three α-catenin proteins (CTNNA1, CTNNA2 or CTNNA3) binds to the actin filament, while γ-catenin (Cowin et al., 1986) is linked to vimentin intermediate filaments (Valiron et al., 1996; Kowalczyk et al., 1998). α-Catenin can directly bind to actin filaments in a tension-dependent manner (catch bound association), as has been shown for α-E-catenin (CTNNA1) in the epithelium (Yonemura et al., 2010; Buckley et al., 2014; Ladoux et al., 2015; Kang et al., 2017), or might use a further linker protein such as the epithelial protein lost in neoplasm (EPLIN; also known as LIMA1) (Maul and Chang, 1999). EPLIN is an actin-binding protein with possible mechanosensitive functions (Maul et al., 2003; Abe and Takeichi, 2008; Sanders et al., 2010; Taguchi et al., 2011) that localizes at EC junctions and has been suggested to modulate angiogenesis (Chervin-Pétinot et al., 2012). Therefore, EPLIN could be involved in the control of EC junction dynamics. Vinculin is an α-catenin homolog that can act as a force sensor and transmitter (Leerberg and Yap, 2013; Yao et al., 2014; Goldmann, 2016; Bays and DeMali, 2017). In endothelium, vinculin is supposed to link the interrupted VE-cadherin–catenin complex to stress fibers, particularly after thrombin stimulation (Huveneers et al., 2012) (illustrated in Fig. 1). However, there is evidence that force-dependent stretched α-catenin is stabilized by vinculin (Yonemura et al., 2010; Yao et al., 2014), indicating that the effect of vinculin on force transmission could be indirect.

Many excellent studies on VE-cadherin and actin have contributed to the current concepts for the regulation of cell permeability, wound healing and angiogenesis (for reviews, see Vestweber et al., 2009; Eilken and Adams, 2010; Harris and Nelson, 2010; Dejana and Vestweber, 2013; Gavard, 2014; Schnittler et al., 2014; Szymborska and Gerhardt, 2017). The mechanisms underlying cell junction remodeling involve actin- and myosin-mediated contractility, tyrosine phosphorylation of VE-cadherin, p120^ctn-mediated endocytosis and the formation of actin-driven plasma membrane protrusions (Doggett and Breslin, 2011; Hoelzle and Svitkina, 2012; Martinelli et al., 2013; Adderley et al., 2015). In addition, live-cell imaging is being increasingly used (Day and Davidson, 2009; Frigault et al., 2009; Cranfill et al., 2016; Seebach et al., 2016; Hofer et al., 2018), as it allows the direct observations of junction remodeling with high spatial and temporal resolution, thus providing valuable hints regarding the underlying mechanisms. In this context, our laboratory has focused on the interdependency between actin-driven membrane protrusions and VE-cadherin-mediated adhesion. We documented that actin-driven protrusions at EC junctions directly form new VE-cadherin adhesion sites, thus driving VE-cadherin dynamics and introduced the term ‘junction-associated intermittent lammelipodia’ (JAIL) based on this function and their temporally and spatially limited appearance (Abu Taha et al., 2014). Here, we discuss the molecular control and functional significance of JAIL, taking into account the known mechanisms of cell-contact regulation in the context of permeability, monolayer integrity, cell migration and angiogenesis.

In the endothelium, actin filaments exist within different structures: filament bundles, such as stress fibers and junction-associated actin filaments (JAAFs), filopodia, lamellipodia-like protrusions and the membrane cytoskeleton (for reviews, see Drenckhahn, 1982; Lampugnani, 2010; Marcos-Ramiro et al., 2014; Schnittler et al., 2014; Alon and van Buul, 2017; Schnoor et al., 2017a). Actin-driven membrane protrusions at EC junctions have been suggested to act as important regulators in cell junction dynamics (Hoelzle and Svitkina, 2012; Martinelli et al., 2013; Breslin et al., 2015; Belvitch et al., 2017; Paatero et al., 2018). JAIL develop as semicircular-shaped small (1 to 5 µm long) actin protrusions at cell junctions, which subsequently form new VE-cadherin adhesion sites and thus drive VE-cadherin dynamics (Abu Taha et al., 2014). This complex process occurs according to the following sequence (Fig. 2). A local decreased VE-cadherin concentration/density (see below) is the prerequisite for the formation of JAIL in the very same place. JAIL arise from branched actin filament formation in one cell in an Arp2/3-dependent manner (see below) and protrude across the apical membrane of the adjacent cell. Exclusively in the overlapping area, trans-interactions between VE-cadherin form, which are visible as so-called VE-cadherin plaques (Fig. 2). VE-cadherin trans-interactions are likely being formed by freely diffusing VE-cadherin molecules in the plasma membrane, as has been described for E-cadherin (also known as CDH1) in MDCK cells and upon ectopic expression of E-cadherin in fibroblastoid L-cells (Bacallao et al., 1989; Sako et al., 1998; Troyanovsky et al., 2006). Expansion of JAIL is terminated by the dissociation of the branched actin network and a loss of Arp2/3 complexes from the actin filaments at the JAIL front (Fig. 2). This initiates clustering of the plaque-forming VE-cadherin molecules, which thereafter move into the junctions, thereby forming new VE-cadherin adhesions (Fig. 2). This mechanism drives VE-cadherin dynamics and contributes to junction remodeling (Fig. 2; also see the animation in Movie 1).

Protrusions, such as filopodia and lamellipodia, are also basic structures that form at epithelial cell junctions (Yamada and Nelson, 2007; Baum and Georgiou, 2011; Troyanovsky, 2012) and are controlled by small GTPases (Garcia-Cattaneo and Braga, 2013; Erasmus et al., 2015). JAIL-mediated VE-cadherin plaque formation in endothelium is in line with the formation of E-cadherin plaques during cell-contact formation in MDCK and L-cell colonies, as was observed by the Nelson group as early as 1998 (Adams et al., 1998). Because the Arp2/3 complex is also important in E-cadherin-mediated cell adhesion (Verma et al., 2004), protrusion-dependent cadherin dynamics might be a shared process in both epithelium and endothelium. However, it can be assumed that junction dynamics is tissue specific and depends on cell differentiation. For example, ablation of VE-cadherin and application of VE-cadherin antibodies in mice leads to tissue-specific changes in endothelial barrier function. A decrease barrier function was found in lung and heart, while the barrier remained largely intact in the skin and the brain (Corada et al., 1999; Frye et al., 2015). These intriguing reports open many questions regarding EC junction regulation and dynamics. Further paradigms are needed to explain differentiation-dependent and tissue-specific junction regulation. Whether JAIL play a role in tissue-specific regulation will need to be explored in more detail.

Under physiological conditions in vivo and in highly confluent EC cultures, VE-cadherin distributes along the junctions in a linear arrangement of individual VE-cadherin clusters, such as those we observed in cell culture models by super resolution microscopy (Seebach et al., 2007; Cao et al., 2017). Upon stimulation with VEGF or under wound healing conditions, cells start to elongate, which is accompanied by VE-cadherin remodeling, forming an interrupted pattern at the cell poles and a faint linear pattern at the lateral junctions (also see below, including for references). VE-cadherin remodeling is also observed after the application of proinflammatory mediators such as thrombin, which also causes such a heterogeneous VE-cadherin pattern. Even between two adjacent cells, both interrupted and linear VE-cadherin patterns can appear (Seebach et al., 2015). These observations indicate that there is a variation in junction regulation at the subcellular level, as indeed has been quantitatively demonstrated (Seebach et al., 2015). Another impressive example for a change in VE-cadherin pattern is during cell growth. Subconfluent growing EC cultures preferentially exhibit an interrupted pattern, which is progressively replaced by an linear pattern with increasing confluence (see below), as first described by Elisabetta Dejana's group in 1995 (Lampugnani et al., 1995).

The interrupted VE-cadherin pattern is most often investigated, as it accompanies changes in barrier function. It is characterized by the presence of ∼2 to 5 µm long VE-cadherin clusters that are typically aligned either perpendicularly or diagonally to the longitudinal axis of the EC junctions (Table 1). These clusters alternate with small VE-cadherin-free regions (nanometers up to a few micrometers), which most likely are responsible for the outflow of solutes and water through the endothelium. The interrupted VE-cadherin pattern also develops in response to thrombin exposure (Rabiet et al., 1996; van der Heijden et al., 2011; Kronstein et al., 2012), or histamine (Andriopoulou et al., 1999; Guo et al., 2008) or TNFα treatment (Wahl-Jensen et al., 2005; Marcos-Ramiro et al., 2014). This interrupted VE-cadherin pattern has recently also been termed focal adhesion junctions (Huveneers et al., 2012) or cadherin fingers (Hayer et al., 2016). However, at these sites contractile stress fibers can terminate, involving vinculin (Millan et al., 2010; Huveneers et al., 2012; Hayer et al., 2016). The contractile features of the stress fibers are suggested to contribute to both the formation of the interrupted VE-cadherin pattern as well as to junction formation and collective cell migration (Schnittler et al., 1990; Wysolmerski and Lagunoff, 1990; Millan et al., 2010; Hoelzle and Svitkina, 2012; Huveneers et al., 2012; Hayer et al., 2016).

We and others have identified three additional types of supra-molecular VE-cadherin organization at junctions: reticular patterns, invaginations and VE-cadherin plaques (Geyer et al., 1999; Abu Taha et al., 2014; Fraccaroli et al., 2015; Seebach et al., 2015; Cao et al., 2017; Neto et al., 2018) (Table 1). The reticular VE-cadherin pattern appears to contribute to the maintenance of barrier function (Fernandez-Martin et al., 2012; Daniel et al., 2015), whereas VE-cadherin plaques result from the JAIL-mediated VE-cadherin trans-interaction as described above. VE-cadherin-positive invaginations (Table 1), however, have been described upon VEGF stimulation and after inhibition of ROCK proteins (Dorland et al., 2016; Cao et al., 2017). These two stimuli, which lead to desphosphorylation of myosin light chain and in turn to a loss of actin-myosin-mediated tension at the junctions. Invaginations closely localize to the cell junctions, but it is unclear whether they are still bound or disconnected from the plasma membrane. Invaginations could be plasma membrane curvatures, which ‘dilute’ VE-cadherin and thus forces JAIL formation (Cao et al., 2017), or might be a result of an endocytotic-like process that, however, is inhibited from further processing into Rab-positive endocytotic vesicles by pacsin 2 (Dorland et al., 2016). In addition, an increased Triton-X 100 solubility of VE-cadherin and β-catenin upon VEGF stimulation (Wright et al., 2002) indicate a disconnection of the VE-cadherin–β-catenin complex from the cytoskeleton in these conditions, and invaginations might contribute to this process, in turn increasing junction dynamics. Removal of VE-cadherin from the junctions by endocytosis and its recycling, mediated by p120^ctn (Xiao et al., 2003a,b; Gavard and Gutkind, 2006), is another possible way by which VE-cadherin-mediated cell adhesion and dynamics could be modulated, and this has been shown to also take place in the case of E-cadherin (Ulrich and Heisenberg, 2008; Troyanovsky, 2009; de Beco et al., 2012; Kowalczyk and Nanes, 2012). It remains to be clarified whether and to what extent endocytosis and invaginations contribute to the increase of cell contact dynamics. The quantitative analysis of such processes would be extremely helpful.

VE-cadherin remodeling occurs in response to the formation of membrane protrusions, such as JAIL and lamellipodia (Abu Taha et al., 2014; Breslin et al., 2015; Neto et al., 2018). This process is rapid (<5 min), and the different VE-cadherin patterns merge into each other (Huveneers et al., 2012; Abu Taha et al., 2014; Breslin et al., 2015; Neto et al., 2018). This has also been shown using a photoswitchable α-catenin–Dendra2 construct, expressed in ECs, which generated an interrupted VE-cadherin pattern from pre-existing VE-cadherin, but not from VE-cadherin at de novo adhesion sites formed after thrombin application (Huveneers et al., 2012). Furthermore, the JAIL size and frequency directly depends on the VE-cadherin concentration at the junctions, which in turn depends on cell shape and cell junction length, as discussed in the next sections.

Cell shape and the extent of cell spreading determine the extent of the cell perimeter, which, in the case of endothelial cell monolayers, is largely identical to the length of cell junctions. The cell perimeter usually changes during cell growth, wound healing and angiogenesis. These processes are accompanied by changes in cell migration, which is obvious in growing EC cultures (Movie 2), and relates to cell junction, in particular VE-cadherin, dynamics. The cell shape is responsible for the distribution of VE-cadherin along cell junctions, whereas JAIL-mediated junction dynamics and cell migration depend on the distribution and concentration and/or density of VE-cadherin. A low VE-cadherin concentration forces the development of JAIL, while a high concentration blocks it.

Importantly, VE-cadherin expression remains constant during cell growth (Lampugnani et al., 1995; Abu Taha et al., 2014) and after treatment with the proangiogenic growth factor VEGF or the γ-secretase inhibitor DAPT, which also induces angiogenesis (Cao et al., 2017). These treatments all cause a change in cell shape from polygonal to elongated cells (Tsuji-Tamura and Ogawa, 2016; Cao et al., 2017). This results in an increased length of cell junctions, which dilutes the given amount of VE-cadherin along the junctions accordingly. Since initiation of JAIL depends on the local VE-cadherin concentration/density, the shape change induces JAIL formation and in turn affects VE-cadherin dynamics. For example, confluent EC cultures (10⁵ cells/cm²) exhibit a cobble-stone pattern with polygonal small cells and a short perimeter or cell junction length; accordingly, the local VE-cadherin concentration is high. In contrast, in subconfluent cells, which have a high degree of spreading, with long perimeter and cell junctions, the VE-cadherin is diluted in an irregular pattern. At sites where there is low VE cadherin occupancy, JAIL-mediated VE-cadherin dynamics is increased (Abu Taha et al., 2014; Cao et al., 2017). Thus, provided the VE-cadherin expression remains unchanged, the cell perimeter (or the length of the cell junction) determines JAIL-mediated VE-cadherin dynamics and thus that of cell junctions (Fig. 3; Movie 2). In summary, there are a number of parameters, which interdependently modulate VE-cadherin dynamics. To correlate VE-cadherin distribution and JAIL formation with functional parameters, such as permeability or cell migration activity, even at the subcellular level, it is helpful to combine interdependent parameters into a single parameter. One such useful parameter is the ratio between VE-cadherin concentration and the corresponding cell junction length in a region of interest or even for cell collectives, which we term the relative VE-cadherin concentration (Rel-VEcad-C) (Fig. 3). Thus, the Rel-VEcad-C can be used to correlate the dynamics of short regions of interest at the junction with, for instance, permeability or cell migration, and it can also be used as a mean value when determined from a collective of cells. In human umbilical vein endothelial cell (HUVEC) cultures that undergo wound healing and after VEGF treatment, the Rel-VEcad-C decreased by ∼20 to 40% owing to cell elongation and this was accompanied by increased JAIL-mediated junction dynamics and cell migration (Cao et al., 2017)

Quantitative determination of cell junction features can be performed using our recently described software, the CellBorderTracker (CBT) (Seebach et al., 2015). Using either immunofluorescence labeling or fluorescently tagged molecules, many cell junction parameters, such as local or mean VE-cadherin concentration/density, cell border length, cell shape and VE-cadherin remodeling can be analyzed (CBT is freely available upon request). Determination of the VE-cadherin concentration (density) in ECs by analyses of image is possible because the three-dimensional spatial expansions of EC contacts (z-axis) are usually thinner than 1 µm (Simionescu, 1977). Above all, changing the cell border length, either through spreading (growing cell cultures) or elongation, modulates the Rel-VEcad-C and thus JAIL-mediated VE-cadherin distribution, dynamics and migration activity (Fig. 3).

The formation of plasma membrane protrusions is a complex molecular process, and depends on signals and mechanisms that involve the cytoskeleton and cytoskeleton-associated factors, as well as integral and peripheral membrane proteins (for reviews, see Bisi et al., 2013; Krause and Gautreau, 2014; Alblazi and Siar, 2015; Mooren et al., 2015; Grikscheit and Grosse, 2016; Tyler et al., 2016; Pizarro-Cerda et al., 2017; Schnoor et al., 2017b; Innocenti, 2018). Actin filament branching is a key process in protrusion formation and requires the actin-related protein complex Arp2/3, which needs to be activated by members of the WAVE/WASP protein family (Pollard et al., 2000; Ti et al., 2011; Welch and Way, 2013; Pollard, 2016; Steffen et al., 2017). This is also the case during JAIL formation, as inhibitors of the Arp2/3 complex, or expression of the (V)CA domain of the nucleation-promoting factor N-WASP (also known as WASL), which acts as a dominant-negative mutant (Pollard, 2007; Rottner et al., 2010), blocked JAIL formation in HUVEC cultures (Abu Taha et al., 2014). Consistent with this, the Arp2/3 complex and N-WASP control EC barrier function (Rajput et al., 2013). However, Arp2/3-mediated formation of JAIL and of classical lamellipodia might be controlled by different subunit compositions of Arp2/3. Indeed, the recent intriguing discovery that isoforms of Arp2/3 subunits display different properties, with the isoforms ARPC1B and ARPC5L showing a greater ability to induce actin assembly than ARPC1A and ARPC5 (Abella et al., 2016; Pizarro-Cerda et al., 2017), could point to a differential control of specific protrusions. It would be very interesting to investigate whether differently composed Arp2/3 complexes modulate the initiation and dynamics of classical lamellipodia and JAIL. Furthermore, the GTP-bound form of the GTPase Rac binds to the WAVE/WASP complex, which is essential for Arp2/3 activation (Eden et al., 2002) and, thus, JAIL formation. This was shown by the use of the Rac inhibitor EHT1864, as well as the expression of dominant-negative Rac1, both of which blocked JAIL formation (Breslin et al., 2015; Cao et al., 2017). Recently, a novel regulatory non-canonical Notch-dependent adherens junction complex, consisting of VE-cadherin, the tyrosine phosphatase LAR (also known as PTPRF) and the guanidine-exchange factor TRIO, was found to activate Rac1 and thus contribute to the assembly and establishment of the barrier function in ECs exposed to shear stress (Polacheck et al., 2017). Since the shear-stress-induced increase in barrier function and EC alignment in the direction of flow are Rac1 dependent and accompany increased VE-cadherin-clustering and protrusion formation (Wojciak-Stothard and Ridley, 2003; Seebach et al., 2007), JAIL might also be targeted by non-canonical Notch signaling.

In addition, other Arp2/3 regulatory factors could also influence JAIL dynamics. The cortical-actin-binding protein cortactin and α-parvin are such candidates. Tyrosine-phosphorylated activated cortactin activates and binds to the Arp2/3 complex, which promotes actin nucleation at adherens junctions, as demonstrated in the epithelium (Han et al., 2014). Cortactin depletion in mice and in cell culture models is accompanied by an increase in vascular permeability (Schnoor et al., 2011). Furthermore, cortactin is enriched at the front of protrusions that close gaps after T-cell diapedesis, which in this system are called ventral lamellipodia (VL) (Martinelli et al., 2013). Expression of a dominant-negative cortactin decreased VL velocity and prolonged gap-closure time, resulting in a delayed reappearance of VE-cadherin to this site (Martinelli et al., 2013). Provided that VL directly lead to new VE-cadherin-mediated adhesion sites (which still needs to be verified), such a process would be consistent with JAIL formation, but this needs to be clarified. Another factor that has been shown to be colocalized with and control JAIL formation is α-parvin, an actin-binding protein and regulator of focal adhesions. Depletion of α-parvin impairs VE-cadherin stability and reduces JAIL formation, which significantly alters angiogenesis (Fraccaroli et al., 2015).

Signaling mechanisms that control JAIL formation are less investigated, but there are some candidates. For instance, sphingosine-1-phosphate [S1P activates the S1P receptor, which in turn activates Rac], increases the formation of junction protrusions (both JAIL and lamellipodia) and improves barrier function (Mehta et al., 2005; Tauseef et al., 2008; Breslin et al., 2015; Cao et al., 2017). Furthermore, signaling through YAP/TAZ proteins (Totaro et al., 2018), the transcriptional co-activators of the Hippo signaling pathway involved in growth control and in cardiovascular development and diseases (Huang et al., 2005; He et al., 2018; Park and Kwon, 2018), has been demonstrated to be important in this process, as YAP/TAZ depletion reduced the formation of JAIL (Giampietro et al., 2015; Neto et al., 2018). It is reasonable to assume that signals that modulate the organization and dynamics of EC junctions, such as the vascular phosphotyrosine phosphatase (VE-PTP; also known as PTPRB), which binds to VE-cadherin and thus might be able to control VE-cadherin cluster formation through phosphorylation (Nawroth et al., 2002; Hayashi et al., 2013), VEGFR1 (Cao, 2009) and Angiopoetin–Tie signaling, which can be activated at the cell junctions and interacts with VE-PTP (Saharinen et al., 2008; Mochizuki, 2009), are also directly involved in JAIL formation.

Last but not least, the pushing of membranes over an adjacent cell during JAIL formation requires a certain amount of energy. The FMNL family of formins modulates actin filament dynamics and has been shown to be required for the generation of pushing forces in classical lamellipodia (Kage et al., 2017), which might be also of relevance in JAIL formation and thus junction dynamics. Furthermore, an elegant recent study addressed the question of what controls pulling and pushing at EC junctions (Efimova and Svitkina, 2018). In this study, the authors used platinum replica electron microscopy (PREM) together with live-cell imaging approaches to provide evidence that Rho-mediated pulling forces stabilize VE-cadherin-mediated cell adhesion, while Rac mediates the pushing forces exerted by VE-cadherin trans-interactions, results that are in full agreement with the observed dynamics of VE-cadherin during JAIL formation.

Termination of JAIL is accompanied by the dissociation of branched actin filaments and a loss of ARP2/3 complex from the actin filaments at the JAIL front (Abu Taha and Schnittler, 2014). Because the concentration of actin and actin-binding factors (Pollard et al., 2000; Falcke, 2016), as well as actin-myosin-mediated contractility (Worthylake and Burridge, 2003; Bergert et al., 2012; Yang et al., 2012) all impact on lamellipodia dynamics, it is feasible that they also have a role in JAIL formation and termination. Assuming the Rel-VEcad-C at cell contacts has increased to a level that inhibits further JAIL formation, the question arises as to what is the mechanism that recruits the contractile actomyosin filaments to the junctions to reinforce stabilization. In endothelium, these actomyosin filaments are called JAAFs (Drenckhahn, 1982), and circumferential actin filament bundles (CAFB) in epithelium (Yonemura et al., 1995). It is generally accepted that contractility is induced by RhoA, which activates ROCK protein and so stimulates phosphorylation of the myosin II light chain (MLC) (Chrzanowska-Wodnicka and Burridge, 1996; Kumper and Marshall, 2011). In epithelial cells, RhoA helps to maintain E-cadherin-mediated adhesion, most likely through myosin II-mediated force generation, which pulls cadherin inward (Sahai and Marshall, 2002; Shewan et al., 2005; Cavey and Lecuit, 2009; Brieher and Yap, 2013). In EC, Rho GTPases have also been shown to significantly contribute to junction regulation (for reviews, see Beckers et al., 2010; Menke and Giehl, 2012; Amado-Azevedo et al., 2014; Marcos-Ramiro et al., 2014; Schlegel and Waschke, 2014; Huveneers et al., 2015; van Buul and Timmerman, 2016). Recent work has shown that RhoB-mediated regulation of Rac at the plasma membrane is responsible for junction recovery during inflammation (Marcos-Ramiro et al., 2016). Thus, a balanced equilibrium between Rho and Rac activity might be critical to JAIL formation and VE-cadherin dynamics, as well as to control acto-myosin formation and contraction. Taken together, these data suggest that Rho-mediated actomyosin contractility may restrict JAIL protrusion. This is further supported by the finding that the Rho GTPase Rnd3, which inhibits Rac activity, increases the activity of RhoA and reduces both lamellipodia and JAIL (Breslin et al., 2016). These findings are also in agreement with the effect of VEGFR2 activation or inhibition of ROCK proteins (Y27632) in cultured ECs, which results in dephosphorylation of MLC at cell junctions, causing a loss of tension that is accompanied by JAAF disassembly and the formation of invaginations that might increase junction dynamics (Cao et al., 2017). Thus, spatiotemporally restricted contractions and relaxations, mediated by RhoA, as well as membrane protrusions, induced by Rac, contribute to both the stabilization and integrity of EC junctions, as well as their dynamic remodeling. Dissecting the spatial and temporal activity of Rho GTPases, together with the identification of proteins and signaling mechanisms that control JAIL dynamics might be helpful to understand dynamic EC junction remodeling and to explain both physiological and pathological processes. In the next paragraph, we discuss how the subcellular JAIL formation can contribute to monolayer integrity.

An EC monolayer consists of individual cells that connect to up to four adjacent cells through cell junctions. However, adjacent cells can have a variable migration direction and velocity, or even move in opposite directions, while nevertheless maintaining overall monolayer integrity. This means that each of the cell contacts between adjacent cells must be regulated individually, resulting in relative cell migration (illustrated in Movie 3, particularly when played slowly). Relative cell migration is a complex process that can only be achieved by a graded, cell-density-dependent establishment of localized cell–cell junctions, as is provided by iterative JAIL formation. Numerous JAIL can be formed at the same time, and their size and frequency depend on the number of sites with low Rel-VEcad-C (i.e. a decrease of 20 to 40%). This is obvious in growing EC cultures. In large spread cells in subconfluent cultures, JAIL frequency and size is high. In contrast, increases in cell density when cells become confluent, and a concomitant graded increase in the Rel-VEcad-C, is followed by a reduction in JAIL frequency and size (Abu Taha et al., 2014). Furthermore, since many JAIL can appear at cell junctions at the same time, it is reasonable to assume an autoregulatory mechanism at each site. Indeed, the direct link between Rel-VEcad-C and JAIL formation (Fig. 4) presents a possibility for autoregulation, as iteration cycles of JAIL formation and disassembly take place until the newly formed cell adhesion sites inhibit further JAIL formation due to the then elevated Rel-VEcad-C, which reaches that found in highly confluent EC cultures (Abu Taha et al., 2014; Cao et al., 2017). Taken together, we propose a subcellular autoregulation process that is based on JAIL-mediated formation of new VE-cadherin adhesion sites. Integrity and barrier function of the entire monolayer is ensured by a gradual modulation of JAIL frequency and size (Figs 4 and 5)

The subcellular regulation of cell junctions is also of great importance in leukocyte diapedesis (Vestweber, 2008; Vestweber et al., 2009; Harris and Nelson, 2010; Giannotta et al., 2013; Gavard, 2014) and in controlling barrier function in response to inflammation-promoting factors (Seebach et al., 2015). The possible impact of junction protrusion on leukocyte transmigration has been discussed above with regard to cortactin. In addition, a recent elegant study uncovered a Rho-dependent contraction of an actin ring that appeared at the rim of the gap formed during leukocyte transmigration, which subsequently restored VE-cadherin adhesion (Heemskerk et al., 2016). Because leukocyte transmigration requires just a very small opening of a junction to allow passage, the contraction mechanism appears to be an alternative to protrusion formation for gap closure. Such a mechanism might be at play for gaps that decreases the Rel-VEcad-C to sufficiently low enough to still promote the formation of JAIL. This might be a very nice example for the evolution of graded mechanisms, which also act as simple and effective mechanisms of pulling and pushing force at cell junctions.

With respect to inflammation-promoting factors, a transient thrombin-induced increase in permeability is accompanied by the inhibition of JAIL (Seebach et al., 2015) or local lamellipodia (Breslin et al., 2015). Subsequently, VE-cadherin remodels into an irregular pattern with interrupted and linear VE-cadherin clusters (Seebach et al., 2015). During barrier function recovery, JAIL (also termed ‘local lamellipodia’) preferentially reappear at sites with the interrupted VE-cadherin pattern and re-establish VE-cadherin distribution in a Rho-dependent manner (Breslin et al., 2015; Seebach et al., 2015) (Movie 4). Together, the regulation of junction dynamics at the subcellular level, for example, through JAIL, might provide a fundamental control principle that on one hand allows cell motility and junction remodeling, while on the other hand maintaining the overall monolayer integrity.

A key event in wound healing and angiogenesis is the formation of a leader cell at the wound rim, or a tip cell in the case of angiogenesis, with both cells migrating forward to initiate collective cell migration (Gerhardt et al., 2003; Yang et al., 2016). The cells behind the leader cells are called followers, while the cells behind tip cells in angiogenesis are called stalk cells. Both processes rely on common mechanisms that, in turn, depend on the expression and distribution of VE-cadherin. The effect of VE-cadherin expression in angiogenesis and wound healing has been investigated in detail (Carmeliet et al., 1999; Crosby et al., 2005; Vitorino and Meyer, 2008; Abraham et al., 2009; Mitchell et al., 2010; Wang et al., 2010). Ablation of VE-cadherin or the truncation of its β-catenin-binding domain still allowed primitive vascular plexus formation in mice, but after embryonic day (E)9.25, these vessels regressed and disintegrated (Carmeliet et al., 1999; Crosby et al., 2005). In contrast and surprisingly, a moderate downregulation of VE-cadherin stimulates cell migration and angiogenesis (Abraham et al., 2009). These, at first glance, inconsistent data become more plausible when cell elongation is considered with its effect on relative VE-cadherin occupancy.

Cell elongation increases cell border length and has been proposed to stimulate cell migration (Merks et al., 2006; Qutub and Popel, 2009; Cao et al., 2017), indicating that this effect is crucial. Indeed, cell elongation occurs in angiogenesis in vivo in the mouse retina and in zebrafish, and can also be induced in vitro by pro-angiogenic factors, such as VEGF and the γ-secretase inhibitor DAPT (Sauteur et al., 2014; Tsuji-Tamura and Ogawa, 2016; Cao et al., 2017; Paatero et al., 2018). The signals underlying cell elongation involve the loss of tension and microtubule polarization, as well as Rac, PI3K, Foxo1 and mTOR signaling (Furuyama et al., 2004; Tsuji-Tamura and Ogawa, 2016; Cao et al., 2017). The loss of tension accompanies the formation of invaginations, which might increase the cell junction length; this, in turn, decreases Rel-VEcad-C, inducing JAIL formation and thus affects local cell junction dynamics (see above), which is required for the initiation of cell elongation. This proposed mechanism is supported by overexpression of VE-cadherin in cell cultures, which blocked both VEGF-induced cell elongation, as well as cell migration and JAIL formation (Abu Taha et al., 2014; Cao et al., 2017). In wound healing, in angiogenesis models and in angiogenesis in vivo, cell elongation is accompanied with a polarized distribution of VE-cadherin. An interrupted pattern at the cell poles alternates with JAIL and VE cadherin plaques, while the extended lateral junctions display a faint linear VE-cadherin pattern with small JAIL, allowing cells to migrate relative to each other (Fig. 4A) (Cao et al., 2017). The mechanisms that lead to polarized cell migration in angiogenesis and sheet migration involve protrusive or repulsive and adhesive forces, which are mediated by actomyosin-mediated contraction, as well as formation of plasma membrane protrusions (Ballestrem et al., 2000; Waschke et al., 2004; Noda et al., 2010; Tambe et al., 2011; Yao et al., 2014; Bazellières et al., 2015; Hayer et al., 2016; van Buul and Timmerman, 2016; Paatero et al., 2018).

The impact of contractile forces on cell migration has been demonstrated in both individual epithelial cells (reviewed in Bertet et al., 2004; Lecuit, 2008; Ladoux et al., 2015), and in migration of sheets of epithelial cells (Tambe et al., 2011; Vedula et al., 2012; Das et al., 2015) and of HUVECs (Hayer et al., 2016). Elongated ECs display an interrupted VE-cadherin pattern (called cadherin fingers here) associated with stress fibers at the cell pole, which has been suggested to pull cells forward; this results in the leader cells extending cadherin fingers from their rear end, which are engulfed by the front of the follower cells (Hayer et al., 2016). However, in a scratch-wounding assay, ECs migrate along individual paths and also display variations in velocity, which is a type of relative cell displacement or relative cell migration. Although in these assays, some cells move backwards against the general direction, nevertheless, the wound is closed in the end (Movie 5; Cao et al., 2017). The already complex process of collective cell migration is even more complicated in angiogenesis, as tip and stalk cells can interconvert, changing the leader position (Jakobsson et al., 2010). Therefore, a subcellular junction control is required with a cyclic formation and loosening of cell adhesion sites. Consistent with this, the transient appearance of an interrupted VE-cadherin pattern at the front of migrating cells in a scratch assay correlates with cryptic lamellipodia described in endothelium (Hayer et al., 2016), which we identified as cycling JAIL at the cell pole in migrating cells sheets (Cao et al., 2017). In mice and cell culture, elongated polarized cells display large JAIL that appear at the cell pole and are able to drive directed forward migration, whereas small JAIL at the lateral site allow for a relative displacement of adjacent cells. JAIL-mediated formation of new VE-cadherin adhesion sites has been recently confirmed in zebrafish, termed there ‘junction-based lamellipodia’ (Paatero et al., 2018), indicating that formation of JAIL is indeed a basic mechanism in angiogenesis.

The iterative cycle between an interrupted VE-cadherin pattern and JAIL-mediated formation of new VE-cadherin adhesion sites continues until the Rel-VEcad-C has increased to the threshold (e.g. the level found in highly confluent ECs) where it blocks further JAIL formation (Abu Taha et al., 2014; Cao et al., 2017); thus, it might be critical for the termination of wound healing and angiogenesis. The concept of auto-regulation of JAIL formation in a Rel-VEcad-C-dependent manner provides a plausible mechanism for those complex processes that need to be coordinated with actomyosin-mediated contractions, and their relevance will need to be analyzed in more detail (Fig. 5).

In summary, actin-driven membrane protrusions have an important effect on cell junction dynamics and are crucial for different aspects of endothelial cell biology. The generation of JAIL as actin-driven, highly dynamic and spatiotemporally confined structure that enable the formation of new VE-cadherin adhesion sites, allows for a subcellular control of cell–cell junctions. Furthermore, JAIL and the relative VE-cadherin concentration (Rel-VEcad-C) are regulated interdependently, pointing to an auto-regulatory mechanism. As a number of factors, such as S1P, VEGF, thrombin or histamine, can modulate JAIL formation, it is reasonable to assume that JAIL are involved in several mechanisms that regulate cell junctions. Nevertheless, the signals and molecular mechanisms that control the formation and termination of JAIL will need to be investigated in more detail in order to deepen our understanding of JAIL-mediated regulation of cell junctions. For instance, it will be of great interest whether (and, if so, how) well-established pathways, including actomyosin contraction, and post-translational modifications, such as tyrosine phosphorylation of the VE-cadherin–catenin complex, or kinases, such as PKC kinases and IP₃ kinases, are involved in JAIL formation. Furthermore, it remains to be investigated what mechanism in detail induces JAIL formation in a cell, and it needs to be clarified whether JAIL formation is asymmetric (i.e. occurs only in one of the neighboring cell or in both). JAIL might be more important than is currently appreciated, and further insights into their formation and regulation could help to paint a better picture of cell–cell contact dynamics and EC biology.

We appreciate the detailed comments and suggestions of the reviewers, who have helped to make the topic accessible to a wide audience. We also thank Dietmar Vestweber and Eloi Montanez for discussions. We apologize in advance if papers that relate to the current topic were not cited in this overview.