The Cross-Rho'ds of Cell-Cell Adhesion*

In vivo, cells live, proliferate, move, and die in a densely populated neighborhood surrounded by a lush forest of tissue and stromal components. Few, if any, cellular processes are not affected by a cell’s adhesive interactions with its neighbors and with the surrounding matrix. Consequently, changes in a cell’s adhesive properties accompany the misregulations in other cell functions that cause and/or result from various pathological conditions such as cancer progression. During the transition from a benign to a metastatic state, there is a breakdown of adhesive bonds between neighboring cells. At the same time, the renegade metastatic cells assume the ability to interact with, migrate over, and invade foreign tissues. Most human tumors are of epithelial origin; however, as carcinoma cells become invasive, they undergo a switch from their differentiated epithelial state to a fibroblastic mesenchymal state (epithelial-mesenchymal transition, EMT) (1). This dedifferentiation and consequent increased motility and invasiveness is accompanied by, and perhaps even caused by, a breakdown in the integrity of intercellular junctions. Although junctional integrity can be compromised in a number of ways, misregulation of E-cadherinmediated adhesion is a key factor in most epithelial cancers (2, 3).

From the Fox Chase Cancer Center, Philadelphia, Pennsylvania 19111 In vivo, cells live, proliferate, move, and die in a densely populated neighborhood surrounded by a lush forest of tissue and stromal components. Few, if any, cellular processes are not affected by a cell's adhesive interactions with its neighbors and with the surrounding matrix. Consequently, changes in a cell's adhesive properties accompany the misregulations in other cell functions that cause and/or result from various pathological conditions such as cancer progression. During the transition from a benign to a metastatic state, there is a breakdown of adhesive bonds between neighboring cells. At the same time, the renegade metastatic cells assume the ability to interact with, migrate over, and invade foreign tissues.
Most human tumors are of epithelial origin; however, as carcinoma cells become invasive, they undergo a switch from their differentiated epithelial state to a fibroblastic mesenchymal state (epithelial-mesenchymal transition, EMT 1 ) (1). This dedifferentiation and consequent increased motility and invasiveness is accompanied by, and perhaps even caused by, a breakdown in the integrity of intercellular junctions. Although junctional integrity can be compromised in a number of ways, misregulation of E-cadherinmediated adhesion is a key factor in most epithelial cancers (2,3).

Cell-Cell Adhesion and RhoGTPases
Cadherin-mediated adherens junctions (AJs) are the most thoroughly studied cell-cell adhesion structures. Classical cadherins include E-cadherin, which is expressed in epithelial cells, N-cadherin, which is predominantly expressed in neural tissues and fibroblasts, and VE-cadherin, which is expressed in endothelial cells (4). These cadherins are single span transmembrane proteins that establish Ca 2ϩ -dependent cell-cell contacts. Homophilic interactions between the extracellular domains of cadherins on neighboring cells are integral to the establishment of AJs. The intracellular regions of cadherins interact with various partners, most notably, the catenins, p120catenin (p120) and ␤-catenin, which then interacts with ␣-catenin, linking the cadherin complex to the actin cytoskeleton and strengthening the adhesion.
Several signaling pathways regulate cadherin adhesion, and in turn, the adhesive complexes influence a number of pathways. We limit our discussion to the effects of Rho signaling on cell adhesion and its relevance to EMT. Although we focus on cadherin-mediated adhesion, processes regulating cell-cell adhesion are intimately coupled to those regulating adhesion to the extracellular matrix.
The Rho proteins, Cdc42, Rac1, and RhoA, are best known as master regulators of the actin cytoskeleton and promote the formation of filopodia, lamellipodia, and stress fibers, respectively. However, it has become clear that the functions of these proteins extend far beyond remodeling actin, including regulating motility, polarity, microtubule dynamics, trafficking, and cell adhesion (5).
Work from several laboratories indicates that cadherin ligation activates Cdc42 and Rac1 and inhibits RhoA, although the mechanisms involved are not yet clear (6 -8 and references therein). The exchange factor, Tiam1, appears to be one important mediator. In NIH3T3 cells, Tiam1-induced Rac1 activation coupled with RhoA inhibition results in increased cadherin adhesion and an epithelial phenotype. Conversely, in Ras-transformed MDCK cells, Tiam1 is transcriptionally down-regulated, Rac1 activity decreases, RhoA activity increases, cadherin adhesion decreases, and EMT is favored. In VE-cadherin-null endothelial cells, transfection of VE-cadherin cDNA resulted in increased Tiam1 expression, its localization to cell-cell junctions, and a concomitant increase in Rac1 activity and decrease in RhoA activity (9). Interestingly, the molecular events triggered by VE-cadherin expression were accompanied by a morphological transition from a fibroblastoid to an endothelial phenotype, reminiscent of the behavior of MDCK and NIH3T3 cells. These examples suggest that the balance between Rac1 and RhoA activities is crucial. This balance is subject to antagonistic signaling between Rac1 and RhoA pathways. Rac1 signaling, via the production of reactive oxygen species, inhibits low molecular weight tyrosine phosphatase, the target of which is the RhoA inhibitor p190RhoGAP (10). Consequently, phosphorylation and activation of p190RhoGAP is favored, and RhoA is inhibited. Phosphorylation of p190RhoGAP is mediated by Src family kinases, which are activated following cadherin ligation (11). Thus, during adhesion signaling, RhoA inhibition may be mediated by two complementary steps: phosphorylation of p190RhoGAP and Rac1-mediated inhibition of low molecular weight tyrosine phosphatase. As p190RhoGAP has been shown to localize to protrusive structures, it has been suggested that the inhibition of RhoA may be required to decrease tension and contractility, thereby facilitating AJ formation (11).
Other work in the literature points to the importance of Rho-GTPases in facilitating the establishment of cadherin adhesion. In MDCK cells, Rac1 and lamellipodia localized only transiently to newly forming cell-cell contact sites and were removed as contacts stabilized and became older (12). Because E-cadherin remained present even at older contact sites, the mere presence of E-cadherin was not sufficient for Rac1 localization to AJs. Thus, the authors suggested that lamellipodia formation is the initial driver for cellcell adhesion and Rac1 functions in the initiation of cell-cell contacts, not in their stabilization (12).
Cdc42-driven filopodia are thought to precede the formation of lamellipodia. In endothelial cells, even non-junctional VE-cadherin was found to activate Cdc42 (13). Because VE-cadherin is nonjunctional in wounded cells, filopodia resulting from Cdc42 activation may be involved in the initiation of cell-cell contacts. Indeed, in keratinocytes, filopodia extension appears to be the initial trigger for AJ formation (14). Interestingly, ligation of the tetraspanin CD151 resulted in Cdc42-mediated filopodia formation in different epithelial cell lines, and overexpression of CD151 promoted the formation of E-cadherin puncta (15). The filopodia extension occurred even in E-cadherin-deficient cells, strengthening the view that filopodia formation is the first step in AJ formation.
Whether Cdc42-driven filopodia or Rac1-driven lamellipodia is the initial step in AJ formation might depend on experimental variables. Alternatively, it is possible that cells extend filopodia independent of cadherin ligation. Once a contact is formed, cadherins activate Rac1, which then facilitates recruitment of other complex components and polymerization of actin to strengthen nascent contacts leading to maturation of AJs.
RhoA also participates in AJ formation, most likely via its effectors Dia and Rho kinase (ROCK) (16). Dia is known to promote actin polymerization and microtubule organization, whereas ROCK functions to enhance contractile force. Inhibition of RhoA by C3 exoenzyme disrupted AJs in different epithelial cell lines in a Dia-dependent manner. On the other hand, overexpression of activated RhoA also disrupted AJs and caused (non-apoptotic) membrane blebbing, which further destabilized adhesions. These activ-ities were dependent on ROCK. In addition, in SW620 cells (colon carcinoma cells overexpressing RhoA), inhibition of ROCK activity led to enhanced AJ integrity. The authors suggested that during AJ formation a low amount of RhoA activity is required for Dia-mediated stabilization; however, once AJs are established, RhoA is inhibited to prevent ROCK-mediated disruption (16).
It is likely that amounts of active Rac1 and Cdc42 are also kept in check. For example, in keratinocytes, sustained activation of Rac1 caused the internalization of E-cadherin and dissociation of cell-cell contacts (17). Because Cdc42 and Rac1 promote cell migration, it is possible that the concentrations of active GTPases determine whether cells adopt an epithelial or mesenchymal phenotype. Alternatively, the signaling partners involved may influence the decision. These possibilities are not mutually exclusive, as shown in the case for RhoA above. Most likely, the degree of activity, localization, and effectors all affect the outcome of RhoGTPase signaling. RhoGTPases have many effectors with varied functions. Here we discuss a few signaling partners that are particularly relevant in pathways regulating cell-cell adhesion.

Rho Effectors in Adhesion Regulation
IQGAP1-IQGAP1 localizes to sites of cell-cell contact and affects E-cadherin-mediated adhesion differently, depending on its binding partners. IQGAP1 binds to ␤-catenin, displacing ␣-catenin from the cadherin complex (18). Consequently, the AJ components are dissociated from the actin cytoskeleton, and adhesion is compromised. Because active Rac1 and Cdc42 associate with IQGAP1, they prevent its interaction with ␤-catenin (19) and prevent AJ weakening.
IQGAP1 interacts with numerous proteins and its interaction with calmodulin directly affects its role linking RhoGTPases to cadherin (20). When intracellular Ca 2ϩ is low, IQGAP1 binds to active Cdc42 and actin. Consequently, Cdc42 is brought to the cytoskeleton, and the actin cross-linking ability of IQGAP1 is enhanced. Furthermore, it has recently been shown that Rac1 requires IQGAP1 for actin accumulation at E-cadherin contact sites (21). Thus, the interaction of IQGAP1 with Cdc42 or Rac1 enhances E-cadherin cell adhesion on two levels: by preventing dissociation of ␣-catenin and by promoting actin polymerization. When the intracellular Ca 2ϩ concentration increases, IQGAP1 preferentially binds calmodulin and dissociates from both Cdc42 and actin, dissolving the connection of Cdc42 with the cytoskeleton. Because calmodulin binding also prevents IQGAP1 from binding to ␤-catenin, there is probably a very fine-tuned regulation of the binding preferences of IQGAP1, depending on intracellular Ca 2ϩ concentration and active Cdc42 and Rac1, and consequent effects on intercellular adhesion.
In an in vitro model for EMT and cancer cell dispersal, hepatocyte growth factor (HGF) induces cell-cell dissociation and scattering. HGF treatment results in loss of ␣-catenin from cell-cell contact sites, followed by loss of E-cadherin, prior to cell dispersal (22). Expression of activated Cdc42 or Rac1 in MDCK2 cells prevents both the HGF-induced disappearance of ␣-catenin from contact sites and the subsequent cell dissociation and scattering. HGF treatment decreases active Rac1, reducing the IQGAP1-Rac1 association. Freed IQGAP1 then interacts with ␤-catenin, displacing ␣-catenin, with subsequent cell-cell dissociation and dispersal (22) (Fig. 1A). In some gastric tumors, a correlation was found between increased membrane localization of IQGAP1, decreased membrane localization of ␣-catenin and E-cadherin, and dysfunction of Ecadherin-mediated adhesion (23). The correlation was greatest in dedifferentiated, diffuse, invasive tumors rather than those that were differentiated, demonstrating the physiological relevance of this signaling module during EMT of tumor progression.
In migrating MCF-7 cells, IQGAP1 localizes to the leading edge. Overexpression of IQGAP1 in these and other epithelial cells increased their motility in a Cdc42-and Rac1-dependent manner (24). IQGAP1 expression also increased the invasiveness of MCF7 cells. siRNA-mediated reduction in IQGAP1 expression resulted in decreased motility and invasiveness in MCF7 cells. Furthermore, knockdown of IQGAP1 decreased Cdc42-induced invasiveness of T47D cells (24). Thus, IQGAP1 appears to function both upstream and downstream relative to active Cdc42 and Rac1 and is potentially involved in multiple steps of EMT.
Merlin-The product of the Nf2 tumor and metastasis suppressor gene, Merlin, is closely related to the ezrin/radixin/moesin proteins, which provide links between several membrane proteins and the cytoskeleton (25)(26)(27). Merlin alternates between folded and unfolded conformations, depending on its phosphorylation state. The conformationally closed form of Merlin is the active growth suppressor and is inactivated by phosphorylation at Ser-518 (28 -30). Recently, a number of reports point to reciprocal interactions between Merlin and Rac1 signaling. In its active state, Merlin inhibits Rac1-mediated signaling and prevents activation of the Rac1 effector, p21-activated kinase (Pak) (28,31). It is therefore possible that Merlin's inhibition of Rac1 and Pak is a mechanism for growth suppression (Fig. 2A). However, active Rac1 stimulates Pak, which phosphorylates Merlin at Ser-518 and thus inactivates it (29,30). Because active Merlin inhibits Rac1 signaling, its phosphorylation by Pak would be expected to potentiate Rac1 signaling. The interaction between Merlin and Pak was shown to be dynamic: the association was enhanced in cells at high density (abundant cell-cell contact) and in cells detached from substrate (lack of cellmatrix contact) (31). Interestingly, Pak activation by Rac1 has been found to be compromised in non-adherent cells (32). Thus, Merlin may inhibit growth signaling in cells detached from substrate.
The phosphorylation status and activity of Merlin is regulated by cell confluence (33). At low cell density, Merlin is phosphorylated and growth-permissive, whereas at high cell density, Merlin becomes dephosphorylated and activated as a growth inhibitor. These results are consistent with a role for Merlin as a mediator of contact-dependent growth inhibition. In fact, Merlin-deficient cells are unable to inhibit growth at high density (34). These cells are also defective in establishing stable cadherin AJs although AJ components are expressed and localized to the membrane, but diffusely. Because Merlin localizes to AJs in wild type cells, its presence may be required to stabilize cell-cell adhesion. As mentioned previously, Rac1 has been reported to be recruited to nascent AJs and to be removed as AJs mature (12). It has been suggested that Rac1 is transiently recruited to AJs to inhibit Merlin and facilitate actin remodeling. Once contact is established, Rac1 leaves and Merlin is activated and prevents additional Rac1 activation at the AJs, which could then disrupt the adhesion (34). Because Merlin is a regulator of Rac1 signaling and also influences cell-cell and cellmatrix adhesions, it is possible that it may serve as a focal point mediating cross-talk between adhesion complexes and Rac1.
Arf6 -The ADP-ribosylation factors are small GTPases that regulate vesicular transport. However, roles in modulating the cytoskeleton, morphology, and motility have also been uncovered (35). Arf6 regulates endosomal-plasma membrane trafficking and also affects cell morphology and motility via Rac1 signaling.
In MDCK cells, expression of ARNO, an Arf6GEF, activates Arf6, induces lamellipodia formation, and stimulates migration (36). Interestingly, the effect was only observed in cells at the periphery of a group or at the leading edge of a wound. This indicates that the absence of cell-cell adhesion is required for the ARNO-induced effect. The induction of lamellipodia and motility was found to coincide with ARNO-induced increases in the activities of Rac1 and phospholipase D, both of which were necessary for the increased motility (36). It is not yet clear how Arf6 activates Rac1. But, it is interesting that ArfGAPs (GIT and PKL) localize at focal contact sites via multimolecular complexes including PIX, a Rac1GEF (35). It is thus conceivable that GIT/PKL bring Arf6 in proximity to the PIX complex and thereby facilitate Rac1 activation.
The ARNO-induced phenotype was similar to cell scattering induced by HGF (35). However, ARNO was found to elevate Rac1 activity, whereas in other reports, HGF treatment decreased Rac1 activity (see above). This apparent contradiction is resolved in a series of reports by Palacios et al. (37,38,40). This group has shown that HGF treatment activates endogenous Arf6 and that dominantnegative Arf6 blocks HGF-induced internalization of AJ components and cell scattering. In addition, they have shown that direct expression of activated Arf6 in MDCK cells results in plasma membrane ruffling and causes disruption of AJs by inducing internalization of E-cadherin (37). The elevated Arf6 expression also increased migration but after a time lag, possibly because cell-cell contacts first had to be dissolved.
Activated Arf6 recruits the nucleoside-diphosphate kinase, Nm23-H1, to AJs (38). The metastasis-associated protein Nm23-H1 facilitates dynamin-dependent fission of endocytic vesi-cles and was required for Arf6-dependent E-cadherin internalization. Expression of activated Arf6 was also shown to result in a decrease in Rac1-GTP. It has previously been demonstrated that Nm23-H1 decreases Rac1-GTP by sequestering Tiam1 (39). Here, it was shown that Tiam1 coimmunoprecipitates with Arf6 and Nm23-H1. Thus, HGF treatment of MDCK cells initially results in an Arf6-dependent decrease in active Rac1, a step that is suggested to be necessary for dissociation of cell-cell adhesion (Fig. 1B). Subsequently, there is a steady rise in the amounts of active Arf6 and Rac1 (40), probably necessary to induce motility. These results point to Arf6 as a key regulator of cell-cell adhesion and motility, partly through its effects on Rac1 activity.
p120catenin-The catenin p120, via its central armadillo domain (Arm), interacts with the juxtamembrane domain of classical cadherins and may influence the strength of adhesion by regulating cadherin clustering. Recently, a role for p120 in the regulation of RhoGTPases has been uncovered.
The mechanisms by which p120 regulates RhoGTPases are not yet clear. Purified p120 interacts with RhoA in vitro and inhibits its exchange of GDP for GTP, suggesting that p120 may act as a guanine nucleotide dissociation inhibitor for RhoA (41). Vav2 (a GEF for Rho family GTPases) has also been shown to interact with p120 and could mediate its activation of Cdc42 and Rac1 (42). Another group provided evidence that the N terminus of p120 is necessary for its ability to induce motility and influence GTPase activity (44). However, in this report, p120 overexpression was found to activate RhoA without affecting the activities of Cdc42 and Rac1.
Overexpressed p120 is mostly cytoplasmic, and its effects on the RhoGTPases and morphology are inhibited upon its binding to cadherins (41)(42)(43). These results support a model in which p120 shuttles between cadherin-bound and cytoplasmic pools (45). When p120 interacts with cadherins, it relieves RhoA inhibition and, simultaneously, brings RhoA to the membrane, where it can be activated and both p120 and RhoA can strengthen adhesion by promoting cadherin clustering. When p120 is cytoplasmic, it inhibits RhoA and activates Cdc42 and Rac1, promoting motility. Subconfluent MDCK cells contain a substantial pool of cytoplasmic p120. As cells became confluent, the majority of p120 was found at cell-cell junctions associated with cadherins (43). These findings suggest a simple means by which p120 acts as a regulator of contact-dependent inhibition of cell motility (Fig. 2B). Furthermore, because cadherins may inhibit cell motility by sequestering p120, loss of E-cadherin adhesion during EMT may translate into increased cytoplasmic p120 and increased motility. p120 also appears to play a role in stabilizing cell surface cadherin. Sw48 cells, which bear an inactivating mutation in the p120 gene, exhibit impaired cadherin-mediated adhesion and are unable to establish normal epithelial morphology. Phenotypic rescue by restoring p120 expression was associated with an increase in the half-life of E-cadherin, indicating that p120 may stabilize E-cadherin (46). Using siRNA, two other studies have also implicated p120 in maintaining cadherin. Knockdown of p120 in multiple cell lines (expressing different cadherin subtypes) resulted in decreased steady-state cadherin amounts and compromised cell-cell adhesion (47). The decrease resulted from cadherin internalization, not alterations in synthesis or delivery to the membrane (47). Likewise, in endothelial cells, loss of p120 led to endocytosis and degradation of VE-cadherin, whereas overexpression of p120 increased surface VE-cadherin by inhibiting its internalization (48). These results point to a role for p120 in the retention of cadherins at the cell surface. p120 is also involved in the transport of cadherin to contact sites. Through its N-terminal region, p120 interacts with kinesin heavy chain (49,50). This interaction resulted in the recruitment of N-cadherin to kinesin and facilitated rapid delivery of N-cadherin to the membrane (49). Kinesin binding also prevented the ability of overexpressed p120 to alter cell morphology (50). A direct interaction between p120 and microtubules has also been reported. The interaction occurred via the p120 Arm domain and prevented association with E-cadherin (50). Although the function of p120 may depend on the cell type, in most settings p120 appears to function as a molecular switch: it may mediate an epithelial morphology by promoting cadherin delivery or retention; alternatively, p120 may influence the transition to a mesenchymal morphology through its effects on morphology and motility via the RhoGTPases.

Concluding Remarks
Clearly, many proteins participate in pathways that regulate cell-cell adhesion and maintain the integrity of the cellular neighborhood. These regulators also participate in pathways that induce the breakdown of the neighborhood. RhoGTPases and their signaling partners lie at the crossroads of many of these pathways, implicated in the formation and disruption of cell-cell adhesion and in the maintenance and destruction of epithelial morphology. The extracellular domains of E-cadherin (purple) on neighboring cells interact, and AJs are formed to establish strong cell-cell adhesion. A, the cytoplasmic domain of E-cadherin interacts with ␤-catenin (green), which interacts with ␣-catenin (red). ␣-Catenin associates with the actin cytoskeleton and fortifies the adhesion complex. Active Cdc42 and Rac1 interact with IQGAP1 and promote actin polymerization at contact sites. Upon HGF stimulation, amounts of active Rac1 and Cdc42 decrease; IQGAP1 interacts with ␤-catenin, displacing the actin link and dissolving the adhesion complex. B, Tiam1 is often recruited to contact sites to activate Rac1, and GDP-Arf6 at AJs prevents E-cadherin internalization. Thus, strong adhesion is achieved. Upon HGF stimulation, Arf6 is activated and recruits Nm23-H1, and E-cadherin is endocytosed. Also, Nm23-H1 interacts with Tiam1, preventing Rac1 activation. As a consequence of both steps, AJs are disrupted.
Hence, the regulation of RhoGTPase activity and signaling specificity is integral to the fate of cell-cell adhesions.

FIG. 2. Models for contact-mediated inhibition of growth and motility.
A, subconfluent cells contain phosphorylated Merlin, which is inactive as a growth suppressor. Phosphorylation of Merlin is mediated by Rac1 signaling and Pak. In this case, Rac1/Pak signaling is active and promotes cell proliferation. As cells reach confluence, Merlin is dephosphorylated, activated, and associates with AJs. Active Merlin inhibits Rac1 signaling and Pak activation and inhibits proliferation. B, subconfluent cells contain cytoplasmic p120, which inhibits RhoA activation by direct interaction and stimulates Cdc42 and Rac1 activation, possibly via Vav2. Active Cdc42 and Rac1 induce membrane protrusions and promote motility. Confluent cells contain many AJs. p120 is sequestered by cadherin binding and no longer promotes Cdc42 and Rac1 activation or motility. Also, RhoA, released at contact sites, may be activated and promote cadherin clustering.