Migration of epithelial cells occurs in a variety of important biological processes including tissue morphogenesis, wound healing, and the metastasis of epithelial tumors. In some instances, the cells remain attached to each other and migrate together as a sheet, maintaining epithelial integrity. In others (e.g., metastasis), junctional complexes are disrupted and cells migrate individually. In both cases, motility involves the extension of membranous protrusions (filopodia and lamellipodia) in the direction of movement and the transient assembly and disassembly of integrin-mediated adhesions with the extracellular matrix. The driving force for these events is provided by regulated changes in the organization of the actin cytoskeleton, which are thought to be coordinated with alterations in intracellular membrane traffic. In this themes article, I review current hypotheses about how these processes are integrated and attempt to identify fruitful areas for future research.
- ADP-ribosylation factor 6
RHO FAMILY GTPASES IN CELL MOTILITY
much of the actin remodeling that occurs during cell migration is driven by the coordinated action of small Ras-like GTPases of the Rho family. Like all GTPases, members of the Rho family function as molecular switches, being “on” in the GTP-bound state and “off” when GDP is bound. In the GTP-bound form, these proteins bind to and activate a variety of downstream effector proteins including kinases, actin-binding proteins, and lipid-modifying enzymes (for review see Ref. 32). The prototype Rho family members, RhoA, Rac1, and Cdc42, are best known for their distinct effects on the actin cytoskeleton. In early experiments performed in Swiss 3T3 fibroblasts, mutational activation of Cdc42 induced formation of filopodia, Rac1 activation resulted in membrane ruffling and the formation of lamellipodia, and RhoA activation stimulated the formation of stress fibers as well as the stabilization of focal adhesions. Conversely, inhibition of Rac function by expression of dominant-negative mutants has been found to prevent formation of lamellipodia and to inhibit migration. Interestingly, Cdc42 function does not appear to be required for cell motility per se but is essential for vectorial migration in chemotaxis or wound healing. Finally, whereas Rho function is necessary for cell adhesion during motility, focal adhesions and stress fibers are not. In fact, expression of activated Rho mutants inhibits migration, apparently by preventing the disassembly of focal adhesions that occurs as cells move forward.
As for other Ras-like GTPases, interconversion between GTP- and GDP-bound states is facilitated by two classes of accessory proteins:1) guanine nucleotide exchange factors (GEF) that catalyze the exchange of GTP for GDP (leading to activation) and 2) GTPase- activating proteins (GAP) that stimulate the hydrolysis of bound GTP (leading to inactivation). More than 30 GEFs have been identified with specificity for one or more Rho family members (for review see Ref. 33). Although divergent in overall sequence, these proteins share a common motif consisting of a catalytic domain (referred to as the Dbl homology or DH domain) invariantly followed by a phosphoinositide-binding pleckstrin homology domain. Similarly, at least 16 Rho-GAPs have been identified that exhibit a similar degree of divergence outside a conserved catalytic domain (33). It is widely believed that the regulated recruitment of these GEFs and GAPs and the balance between their activities determines the level of local activation of individual Rho family members in response to extracellular cues.
REGULATION OF EPITHELIAL CELL MOTILITY
Whereas the roles of RhoA, Rac, and Cdc42 in cell motility appear to be similar in most cell types examined, their function in epithelial cells is more complex. In this context, they appear to be necessary for both cell-cell adhesion and for motility. For example, the function of both Rac and Rho is required for establishment of new E-cadherin-mediated junctions, as well as for the maintenance of preexisting ones (for example see Ref. 14). In Madin-Darby canine kidney (MDCK) cells, hyperactivation of Rac1 (by expression of activated mutants or of Rac GEFs) leads to an accumulation of both E-cadherin and cortical actin at lateral membrane borders and prevents cell migration induced by either hepatocyte growth factor (HGF) or oncogenic Ras (14). Cdc42 may also play a role in the assembly of adherens junctions, although its precise function in this context is currently unclear.
In contrast, inhibition of Rac1 function has been shown to block MDCK cell migration in a wound-healing assay (8) and in response to HGF (27), suggesting that Rac also plays a positive role in regulating epithelial migration as it does in fibroblasts and other nonepithelial cells. Rac1, RhoA, and Cdc42 are also required for morphogenetic movements of epithelia duringDrosophila embryogenesis. The best example of this is dorsal closure, where ventrally located epidermal cells migrate dorsally along both sides of the embryo until they meet along the dorsal midline (12). Expression of dominant-negative Rac1, RhoA, or Cdc42 constructs in epidermal cells significantly inhibited both the rate and extent of dorsal closure with slightly different phenotypes depending on which GTPase was mutant (12).
How can Rac and Cdc42 support both cell-cell adhesion and cell migration? One hypothesis is that the choice between adhesion and migration may depend on the intracellular site at which these proteins become activated. In this model, preferential recruitment of nucleotide exchange factors to sites of cell-matrix adhesion (or growth factor binding) would lead to enhanced motility, whereas recruitment to adherens junctions would reinforce cell-cell interactions. Given the number and diversity of GEFs and GAPs, it seems likely that distinct sets of regulatory proteins could be recruited to the leading edge or junctional complexes in response to separate signals and that the balance of their activities would determine whether cell-cell adhesion or motility predominates. Similarly, individual GTPases may encounter distinct sets of effector proteins depending on their site of activation, and the stimulation of each may determine cell behavior. A significant goal for the future will be to localize as many of the GEFs and GAPs as possible in epithelial cells under polarized conditions and in the motile state and to determine their individual roles in cell adhesion and/or motility. In addition, recent advances in imaging technology allow the simultaneous visualization and quantitation of Rac activation in living cells (4), and it should now be possible to apply such techniques to epithelial cell systems under a variety of experimental conditions.
COUPLING OF VESICULAR TRANSPORT TO CELL MOTILITY
Vectorial cell migration involves the extension of filopodia and/or lamellipodia in the direction of movement. It has been suggested that the elaboration of these protrusive structures requires the recruitment of membrane constituents from other parts of the cell and their insertion into the leading edge. This could be accomplished by several mechanisms. First, on induction of migration there is a dramatic reorientation of the secretory apparatus, such that the Golgi is positioned between the leading edge of the cell and the nucleus, along the axis of migration. Secretion along this axis is highly polarized, as demonstrated by the observation that the newly synthesized vesicular stomatitis virus is delivered selectively to the leading edge of migrating cells (2), whereas the influenza virus hemagglutinin is transported to the trailing edge (19). This selectivity is highly reminiscent of biosynthetic transport in polarized epithelial cells, where vesicular stomatitis virus buds exclusively from the basolateral pole of the cell and influenza virus buds apically (see below). Second, endocytosed membrane constituents can be recycled preferentially to the leading edge, as has been shown for internalized transferrin (13). Recent evidence indicates that the endocytic recycling compartment, like the Golgi, becomes localized immediately behind the leading edge in migrating neutrophils (24). It has been widely assumed that integrins would be recycled from the trailing edge to the leading edge via this pathway. Recent evidence suggests that this may be true for some but not all integrins (24, 28). In addition, the quantitative contribution of these pathways to the total lamellar membrane or to the total surface integrin pool remains to be determined.
As mentioned above, the secretory polarity observed in migrating cells is similar to that observed in polarized epithelial cells. Remarkably, polarized secretion also occurs in the budding yeastSaccharomyces cerevisiae and parallels can be drawn with epithelial cells. In yeast, secretion is polarized toward the emerging bud. The docking and fusion of secretory vesicles at the bud site is mediated by a complex of eight proteins referred to as the exocyst (for review see Ref. 21). Six of the eight yeast subunits have mammalian orthologs, and the mammalian complex appears to contain at least one additional subunit. Analysis of the mammalian complex in MDCK cells has shown that it functions in basolateral, but not apical, secretion. Moreover, in fully polarized cells, the exocyst components sec6 and sec8 localize primarily to the apex of the lateral membrane, suggesting that basolateral secretion is even more spatially restricted than had been originally appreciated (for review see Ref.15).
In yeast, the location of the bud site is determined by the action of the Rho family protein Rho1p, and polarized secretion requires both Rho1p and Cdc42p (21). Both Rho1p and Cdc42p interact with the NH2 terminus of one of the exocyst subunits, sec3p, and truncation of the protein leads to a more random distribution, suggesting an important role for sec3p in localization of the secretion machinery. Moreover, several cdc42 mutant alleles lead to a loss of exocytic polarity. Importantly, Cdc42 also appears to regulate polarized secretion in epithelial cells. Several laboratories have reported that expression of dominant-negative Cdc42 mutants selectively inhibits basolateral transport of membrane proteins, without affecting apical secretion. As mentioned above, Cdc42 does not appear to be necessary for epithelial cell migration but does affect the directionality and persistence of migration in a wound-healing assay using MDCK cells (8). In such assays, cells typically move vectorially from one edge of the wound toward the opposing edge. In a similar assay using fibroblasts, expression of dominant-negative Cdc42 was found to slow wound healing by randomizing the direction of migration, suggesting that Cdc42 is important in determining polarity even in fibroblasts. It is tempting to speculate that Cdc42 becomes activated at the leading edge in response to chemotactic or other spatial cues and that this leads to polarized secretion by direct interaction with components of the exocyst. Although mammalian sec3 appears to lack the Rho/Cdc42 binding domain of its yeast counterpart, it remains possible that Cdc42 interacts with one or more of the other subunits or that it binds them indirectly through another effector molecule (21). These hypotheses remain to be tested. Additional support for this hypothesis can be found in the observation that rab8, which mediates the docking and fusion of basolateral secretory vesicles in MDCK cells, is the closest mammalian homolog to sec4p, the yeast rab that regulates exocyst function. Ectopic expression of rab8 has been shown to induce the formation of actin-rich membrane protrusions in fibroblasts (23).
CROSS TALK BETWEEN RHO AND ARF GTPASES
Like the Rho GTPases, the ADP-ribosylation factors (ARFs) are also members of the Ras superfamily. However, unlike the Rho family, ARFs appear to function primarily in the regulation of vesicular transport (20). Of the mammalian ARFs, ARF1 is currently the best understood and has been shown to nucleate the assembly of coat protein complexes at sites of vesicle formation in the Golgi and endosomes. In contrast, ARF6 has been localized to the plasma membrane and a subpopulation of endosomes, where it has been implicated in a variety of processes including endocytosis, recycling, and phagocytosis. At least some of the functions of ARF6 may be cell-type specific. For example, in CHO cells, expression of a dominant-negative ARF6 mutant has been shown to inhibit recycling of internalized transferrin receptors (7). However, in HeLa cells, the same ARF6 mutant impairs recycling of major histocompatability complex class I and an IL-2 receptor construct but does not appear to affect recycling of transferrin receptors (25). This discrepancy may be due to the presence in HeLa cells of an endosomal compartment that does not exist in the CHO line (25). In adipocytes, ARF6 appears to regulate the endothelin-stimulated translocation of Glut4 glucose transporters from an endosomal compartment to the plasma membrane, although it has no effect on insulin-stimulated translocation of the same transporter (17). Taken together, these data suggest that ARF6 regulates some but not all aspects of endosomal membrane recycling. However, the mechanisms by which this is accomplished remain poorly understood.
Recently, it has come to be appreciated that ARF6 can also regulate actin cytoskeleton assembly. Activation of ARF6 was found to induce the formation of actin-rich membrane protrusions in both HeLa cells (26) and CHO cells (6). Conversely, dominant-negative ARF6 has been shown to inhibit membrane ruffling induced by EGF, CSF-1, or bombesin, suggesting a functional relationship between ARF6 and Rac. Several groups have demonstrated that at least a fraction of Rac1 is localized to endosomal membranes and that this endosomal pool can be translocated to the plasma membrane on ARF6 activation. Combined with the inhibition of membrane ruffling by dominant-negative ARF6, these findings led to a model in which the ARF6 regulates the activation of Rac1 by controlling its access to the plasma membrane. However, this model fails to take into account the large pool of cytosolic Rac1 that could be recruited to the plasma membrane without invoking a vesicular transport mechanism. Furthermore, dominant-negative ARF6 can inhibit membrane ruffling induced by an activated Rac1 mutant targeted to the plasma membrane by fusion to a heterologous transmembrane protein. This latter finding indicates that ARF can regulate Rac function after its translocation to the plasma membrane. The mechanistic links between ARF6 and Rac are therefore unclear but are under active investigation (see below) and are likely to be a fruitful area for future research.
ARF6 also exhibits functional interactions with Rho. Expression of an activated ARF6 mutant (3), of the ARF6 nucleotide exchange factors ARNO (10), or EFA6 (9) induces a loss of actin stress fibers whose assembly is known to be regulated by RhoA. Again, the mechanistic basis of this interaction is not known; however, the recent discovery of a family of ARF-GAPs that also possess Rho-GAP domains suggests that recruitment of these proteins to sites of ARF activation could lead to local downregulation of Rho activity (for review see Ref. 30). The further characterization of these proteins and their regulation (each has 5 pleckstrin homology domains) should prove to be interesting and informative.
INTERNALIZATION OF E-CADHERIN COMPLEXES
Epithelial cells can acquire a more motile, fibroblastic phenotype in a process referred to as epithelial-mesenchymal transition (EMT). Examples of EMT can be found both in normal development (e.g., gastrulation, neural crest cell migration) and pathologically, as in the onset of tumor metastasis. This process can be recapitulated in vitro by treatment of epithelial cell monolayers with growth factors such as EGF, HGF, or FGF or by the expression of oncogenes such as v-Src. An early event in EMT both in vivo and in vitro is the loss of adherens junctions, which correlates with internalization of E-cadherin complexes from the plasma membrane. E-cadherin is the primary adhesion molecule of epithelia. Dysregulation of cadherin expression or function is a common feature of epithelial cancers, and in fact E-cadherin has been identified as a tumor suppressor (for review see Ref.5).
In confluent monolayers of MDCK cells, E-cadherin has been shown to undergo constitutive endocytosis at a low but detectable rate (18); however, internalization is dramatically increased on treatment of cells with HGF, phorbol esters, or reduction in the level of extracellular calcium. Although the E-cadherin cytoplasmic domain becomes phosphorylated on tyrosine after exposure to growth factors, the mechanisms by which it is internalized have been poorly understood. Recently, Birchmeier and colleagues (11) have shown that, like many growth factor receptors, E-cadherin becomes ubiquitinated after its tyrosine phosphorylation and that this modification significantly enhances its rate of endocytosis. These authors have identified a novel ubiquitin ligase, Hakai, which appears to interact directly and specifically with phosphorylated E-cadherin. Overexpression of Hakai was found to stimulate ubiquitination of E-cadherin and to significantly enhance the sensitivity of the cells to HGF. Conversely, a mutant E-cadherin that could not bind Hakai was not internalized after HGF treatment and prevented the scattering of HGF-treated cells. Together these findings indicate that ubiquitin-mediated internalization of E-cadherin is a key event in the onset of epithelial cell motility. Whether ubiquitination promotes the association of E-cadherin with the clathrin endocytic machinery remains to be determined, although this appears likely. Considering that the half-life of E-cadherin is reduced by nearly 50% in cells expressing exogenous Hakai, it also seems likely that ubiquitination targets the protein for degradation in lysosomes. It will be important to determine whether ubiquitination occurs under other experimental regimens that stimulate E-cadherin endocytosis, such as phorbol ester treatment or Ca2+ depletion. Takai and coworkers (16) have shown that E-cadherin is cointernalized with the HGF receptor c-met. Notably, internalization of both proteins was inhibited by expression of constitutively active RhoA or Rac1 or by a dominant-negative mutant of the Rab GTPase Rab5. A similar Rab5 mutant has been shown to inhibit endocytosis of both EGF and transferrin receptors in nonpolarized cells, and it is possible that this is due to inhibition of coated pit assembly (1). Similarly, D'Souza-Schorey and coworkers (22) recently found that internalization of E-cadherin is also regulated by ARF6. Expression of constitutively active ARF6 stimulated adherens junction disassembly and the intracellular accumulation of E-cadherin, whereas a dominant-negative ARF6 mutant inhibited internalization induced by HGF. Inhibition of ARF6 function also inhibited scattering induced by activation of a temperature-sensitive v-Src construct, indicating that ARF6 functions downstream of Src in the regulation of cell motility. It will be interesting to determine whether any or all of these GTPases modulate the ubiquitination of E-cadherin (and c-met) or if internalization is impaired despite ubiquitination at normal levels.
Interestingly, activation of ARF6 appears to be sufficient to induce motility of epithelial cells. We have found that expression of the ARF nucleotide exchange factor ARNO in MDCK cells leads to a dramatic morphological change in which the cells separate from each other and develop broad, fan-shaped lamellipodia. This morphological transition correlates with a significantly enhanced rate of migration, even in the absence of motogens such as HGF (29). The onset of this migratory phenotype requires two downstream events initiated by ARF6: the activation of Rac1 and the stimulation of a known ARF effector protein, phospholipase D (PLD). Enhanced Rac1 activation is consistent with the observed formation of lamellipodia, and we found that expression of a dominant-negative Rac1 mutant inhibited both lamellipodium formation and migration. More surprisingly, we found that inhibition of PLD activity also blocked the formation of lamellipodia. The product of PLD is phosphatidic acid, a charged lipid with a number of biological properties. Among these is its function as a cofactor in the synthesis of l-α-phosphatidylinositol 4,5-bisphosphate (PIP2) by phosphoinositide 4P 5-kinase-α, an enzyme that is also allosterically activated by ARF. Importantly, we found that activation of Rac1 was not dependent on PLD activity, nor was PLD activity dependent on Rac1, suggesting that each represents a distinct signaling pathway downstream of ARF6 whose combined activities are required for motility. Considering that PIP2 modulates the activities of a number of actin-binding proteins, we have proposed a model in which the products of PLD are necessary for the reorganization of the actin cytoskeleton orchestrated by Rac1.
The mechanism by which ARF6 activates Rac1 remains unclear; however, emerging evidence suggests that ARF6 and Rac can interact indirectly through several recently characterized ARF GAPs. These proteins, which include GIT1, GIT2/Pkl, and p95APP, each contain an NH2-terminal ARF-GAP domain, a central domain that binds the Rac nucleotide exchange factor β-PIX, and a COOH-terminal domain that interacts with the focal adhesion component paxillin (for review see Ref. 31). It is therefore tempting to speculate that activation of ARF6 leads to Rac1 activation through either recruitment or enhanced local activation of β-PIX at sites of ARF activation. This hypothesis is currently being tested.
Taken together, these data suggest that ARF6 regulates at least two distinct aspects of epithelial cell motility: adherens junction disassembly and the formation of lamellipodia. However, as always, each observation generates multiple new questions: How does ARF6 activation lead to junction disassembly? Does it affect the interaction of E-cadherin with the cytoskeleton, or induce endocytosis directly? Does it integrate with Rho family proteins in this context, or does it operate independently? Is PLD activity necessary for cadherin internalization? Does ARF6 activity modulate ubiquitination of E-cadherin by Hakai and, if so, how? Is tight junction disassembly regulated coordinately? Does ARF6 regulate the flow of membrane into lamellipodia? If so, how is this flow targeted to the leading edge? Is more than one Rac GEF activated by ARF6 and, if so, which ones?
In summary, although there has been an explosion of information in the fields of membrane traffic and cell motility, surprisingly little is known about how these two aspects of cell function are integrated. It is clear that the actin cytoskeleton plays a central role and that cross talk among a large number of regulatory signaling pathways is necessary to coordinate intracellular transport with cell movement. Hopefully, work in the near future will target this interface more directly.
Address for reprint requests and other correspondence: J. E. Casanova, Dept. of Cell Biology, University of Virginia Health System, Box 800732, 1300 Jefferson Park Ave., Charlottesville, VA 22908-0732 (E-mail:).
- Copyright © 2002 the American Physiological Society