Cortical Actin Organization: Lessons from ERM (Ezrin/Radixin/Moesin) Proteins*

In recent years to clarify the molecular mechanism of dynamic organization of the cortical actin filaments, which is important not only for the determination of cell-surface structures but also for the functions of integral membrane proteins themselves, various types of submembrane proteins involved in cortical actin filament/ plasma membrane interaction have been intensively studied. In this minireview, we focus on ezrin/radixin/moesin (ERM) proteins, which are general cross-linkers between cortical actin filaments and plasma membranes and are involved in the formation of microvilli, cell adhesion sites, ruffling membranes, and cleavage furrows. ERM proteins have attracted a great deal of interest because their functions have been shown to be regulated by the Rho signaling pathway (for recent reviews, see Refs. 1–7).

the C-terminal region of which binds to the N-terminal half of ezrin (36). Sequence analyses revealed that EBP-50 is identical to a Na ϩ /H ϩ exchanger regulatory factor (NHE-RF). This NHE-RF and its isoform, E3KARP, bear two PDZ domains, which were shown to directly bind to the C terminus of NHE3 (36,37). Thus, NHE-RF and E3KARP can function as adapters between NHE3 and ezrin (38). Interestingly, because NHE-RF regulates the NHE3 function in a protein kinase A (PKA)-dependent manner and because ezrin specifically binds to the RII subunit of PKA (38), ezrin appears to play an important role in recruiting PKA to the NHE3⅐NHE-RF complex to regulate the function of NHE3. Furthermore, NHE-RF and E3KARP were found to also be associated with other integral membrane proteins such as the ␤ 2 -adrenergic receptor and the cystic fibrosis transmembrane conductance regulator (39,40), which are not always co-localized with ERM proteins (41). The physiological relevance of the existence of two mechanisms of binding of ERM proteins to integral membrane proteins, direct and indirect, is an interesting subject for future study.

Intramolecular Interdomain Interaction between N-and C-terminal Halves of ERM Proteins: Molecular Mechanism for Inactivation of ERM Proteins
Given that the cortical actin filaments are dynamically organized in response to various signals, the cross-linking activity of ERM proteins between actin filaments and plasma membranes is expected to be dynamically regulated. Indeed, the pioneering work by Bretscher (42) or Hanzel et al. (43) showed that EGF treatment of A431 cells or the secretion-stimulation of parietal cells rapidly recruited substantial amounts of ERM proteins to the cortical actin layer with concomitant phosphorylation of ERM proteins. When conventionally cultured cells were homogenized and centrifuged in physiological saline, ERM proteins were partitioned almost equally into the soluble and insoluble fractions (44). These findings suggested that there are active (insoluble) and inactive (soluble) forms of ERM proteins in terms of their cross-linking activity inside cells.
Evidence has accumulated in vitro and in vivo that the N-and C-terminal halves of ERM proteins mutually interact intramolecularly and suppress their actin filament and membrane binding activities, respectively (45)(46)(47). Recently, it was shown that when two amino acid residues were deleted from the C-terminal end of ezrin, which do not correspond to the EBP-50-binding domain, the interdomain interaction was affected, allowing ezrin to directly interact with EBP-50 (37). These findings indicated that conformational masking by intramolecular interdomain interaction is the molecular mechanism behind the inactivation of ERM proteins.
The region involved in the interdomain interaction was narrowed down to residues 1-297 and 480 -586 (1-296 and 479 -585 when methionine 1 is posttranslationally removed) as exemplified in ezrin, and these regions are called N-and C-ERMADs (ERMassociation domains), respectively (46). Initially, these N-and C-ERMADs were thought to be responsible for oligomerization of ERM proteins, i.e. intermolecular interaction of ERM proteins, but it has been suggested that they are also important for intramolecular interaction (6,7). Furthermore, yeast two-hybrid analyses identified several middle regions between N-and C-ERMADs that interact with N-and C-ERMADs (48). Although the molecular mechanism and physiological relevance of dimerization and oligomerization of ERM proteins remain elusive, it is now accepted that the intramolecular mutual suppression mechanism keeps ERM proteins in an inactive state and that some activation signal may release this suppression to activate ERM proteins inside cells.

Phosphorylation and PIP 2 Binding as Activation Signals of ERM Proteins
To date, two molecular events have been shown to generate and/or maintain the active form of ERM proteins in vitro: phosphorylation of their C-terminal threonine residue and PIP 2 binding to their N-terminal domains. The phosphorylation of ERM proteins, especially ezrin, has been examined in detail. EGF stimulation induced tyrosine and serine phosphorylation of ezrin in A431 cells with concomitant translocation from the cytoplasm to the cortical actin layer (42). The tyrosine residues that were phosphorylated in A431 cells were shown to be Tyr-146 and Tyr-354 (Tyr-145 and Tyr-353 when methionine 1 is posttranslationally removed) in ezrin, the former of which was conserved in radixin and moesin (49). ERM proteins were also reported to be heavily tyrosine-phosphorylated by v-Src and hepatocyte growth factor (50,51). However, the substitution of Tyr-146 with phenylalanine in ezrin showed some effects on cell motility but did not affect the cortical localization of ezrin induced by hepatocyte growth factor. Thus, the direct activation of ERM proteins by tyrosine phosphorylation is unlikely.
Secretion/stimulation in gastric parietal cells was reported to induce serine/threonine phosphorylation of ezrin (43). In platelets thrombin activation induced the phosphorylation of moesin at a specific C-terminal threonine residue (Thr-558) (52), causing filopodia formation. This site was effectively phosphorylated in vitro by ROK␣/ROCK-II/Rho-kinase in radixin (Thr-564) and by PKC- (53,54). However, it is likely that ROCK kinases do not phosphorylate ERM proteins in vivo (55). In vitro functional analyses suggested that the C-terminal threonine phosphorylation maintains ERM proteins in the active state by suppressing the intramolecular interaction (53). Immunofluorescence microscopy with monoclonal antibodies specific for C-terminal threonine-phosphorylated ERM . Proteins as well as PIP 2 that bind to ERM proteins are listed at the top (for details, see "Structure of ERM Proteins"). In some papers on ERM proteins the amino acid residues are numbered from proline 1, because methionine 1 is posttranslationally removed in human ezrin and moesin (10,15). In this review, however, they are numbered from methionine 1. The percentage sequence identity with ezrin in each domain is indicated at the amino acid sequence level.

Minireview: Actin/Membrane Interaction by ERM Proteins 34508
proteins revealed that ERM proteins localized beneath plasma membranes were actually phosphorylated at the C-terminal threonine in vivo (53,56). Taken all together, it seems that the threonine phosphorylation just maintains the activated ERM proteins.
Another candidate for the activation signal for ERM proteins is PIP 2 , which has been shown to directly bind to the N-terminal halves of ERM proteins in vitro (28,57). Recently, it has been shown that PIP 2 is a key factor for the activation of ERM proteins in vivo (55).

ERM Proteins Downstream as Well as Upstream of Rho
The question has thus arisen as to the identities of the upstream factors required for activation of ERM proteins. Rho, one of the small GTP-binding proteins, is now considered to be a general regulator of actin-based cytoskeletal organization. To date, in vitro as well as in vivo analyses have suggested an intimate relationship between the Rho signaling pathway and activation of ERM proteins. First, the binding ability of ERM proteins to the cytoplasmic domain of CD44 in crude cell homogenate was reported to be enhanced by activation of Rho (28). In semi-permeabilized Swiss 3T3 cells, at least one of the ERM proteins was shown to be required for Rho-dependent formation of stress fibers and focal contacts (58). Furthermore, transfection of the constitutively active mutant of RhoA (V14RhoA), but not that of Rac1 or Cdc42, induced microvillus formation to which ERM proteins were recruited (55,59). Thus, it is now accepted that when Rho is activated in vivo, ERM proteins in the cytoplasm are activated and recruited to plasma membranes to form microvilli. Although ERM proteins are also suggested to be located downstream of Rac (58), the activation of ERM proteins to form microvilli specifically depends on Rho but not on Rac.
Rho has been reported to activate several serine/threonine kinases such as ROK␣/ROCK-II/Rho-kinase, ROK␤/ROCK-I, citron kinase, protein kinase N, and protein kinase C1 (58). Immunoblotting with a monoclonal antibody specific for C-terminal threonine-phosphorylated ERM proteins revealed that in serum-starved Swiss 3T3 cells Rho activation by lysophosphatidic acid stimulation increased the levels of the C-terminal threonine phosphorylation of ERM proteins (53). Rho-kinase, which effectively phosphorylates the C-terminal threonines of ERM proteins in vitro, is not responsible for this Rhodependent threonine phosphorylation of ERM proteins in vivo. PIP 2 -producing phosphatidylinositol 4-phosphate 5-kinase (PI4P5K) has also been reported to be a direct Rho effector (for a review, see Ref. 60). Because activation of dormant ERM proteins was induced by PIP 2 in vivo as well as in vitro (28,55), one possible pathway for the activation of ERM proteins is as follows. Rho may activate PI4P5K, which in turn increases the amount of PIP 2 . PIP 2 then activates ERM proteins by inhibiting their interdomain interaction, which allows phosphorylation of their C-terminal threonine residue by unidentified kinases. The C-terminally threonine-phosphorylated ERM proteins are stabilized as activated forms, which function as actin filament/plasma membrane cross-linkers to form microvilli (Fig. 3).
On the other hand, immunoprecipitation experiments identified Rho-GDI (GDP dissociation inhibitor) in the CD44-ERM protein complex (28). An in vitro binding study then revealed that active but not inactive forms of ERM proteins directly bound to Rho-GDI at their N-terminal halves (61). Interestingly, this binding of ERM proteins suppressed GDI activity of Rho-GDI, i.e. GDP-Rho was released from Rho-GDI, followed by activation as GTP-Rho possibly through Dbl (GDP/GTP exchange protein) (61,62). These findings suggested that ERM proteins, once activated, can activate Rho, which again activates ERM proteins as a positive feedback system. In this sense, ERM proteins are located not only downstream but also upstream of Rho.

Inactivation Signals for ERM Proteins
Because native monomers of ERM proteins have a tendency to inactivate themselves by the interdomain interaction, the downregulation of activation signals may inactivate ERM proteins inside cells. For example, at the initial phase of apoptosis, ERM proteins were dephosphorylated and inactivated, resulting in their cytoplasmic translocation with concomitant microvillar breakdown (44). In this process, the C-terminal threonine was confirmed to be dephosphorylated by the monoclonal antibody specific for C-termi-nal threonine-phosphorylated ERM proteins. 2 Recently, myosin light chain phosphatase was shown to bind to moesin through its myosin-binding subunit in vitro, although its physiological relevance remains unclear (63). On the other hand, proteolysis could be another method of inactivation of ERM proteins because ezrin is a good substrate for calpain in vitro (6,7). Although evidence is still lacking, on and off activation and/or inactivation signals and their balance may regulate the function of ERM proteins.

ERM Proteins and Merlin
Some, but not all, of the characteristics of ERM proteins appear to be shared by merlin. The subcellular localization of ERM proteins was similar to that of merlin in fibroblasts (microvilli and ruffling membranes) but different in epithelial cells; merlin, but not ERM proteins, was concentrated at lateral membranes together with Ecadherin (64). The N-terminal half of merlin bound to the cytoplasmic domains of CD44 (65) and NHE-RF (66) as well as Rho-GDI in vitro (64). Merlin has two major alternatively spliced isoforms, I and II, which differ at their C-terminal ends (67). Although neither of these C-terminal ends shows any similarity to those of ERM proteins, the major actin-binding domains, merlin has been reported to bind to actin filaments at its middle region (68). Interdomain interaction between the N-and C-terminal halves has been suggested in isoform-I, but the interaction does not appear to affect its binding affinity to actin filaments or to NHE-RF (66,68). No interdomain interaction has been detected in isoform-II (69).
At present, there are three distinct possible explanations for the sequence similarity between ERM proteins and merlin. First, it is possible that merlin functions in cells by competing for shared binding partners such as CD44, NHE-RF, and Rho-GDI. However, 2 Sa. Tsukita, unpublished data.
FIG. 3. Activation pathway of ERM proteins. Rho activates PI4P5K, which increases the amount of PIP 2 . PIP 2 then activates ERM proteins by inhibiting their interdomain interaction, which allows phosphorylation of their C-terminal threonine residue by some kinases. The C-terminally threonine-phosphorylated ERM proteins are stabilized at the activated forms, which function as actin filament/plasma membrane cross-linkers to form microvilli. Activated ERM proteins are associated directly with the adhesion molecules such as CD44 and ICAM-1, -2, and -3 and indirectly with other integral membrane proteins such as NHE3 through EBP-50/NHE-RF. Activated ERM proteins also bind to Rho-GDI at their N-terminal halves, suppressing GDI activity of Rho-GDI to release GDP-Rho, which is activated to GTP-Rho. This GTP-Rho can be used to activate ERM proteins just beneath the plasma membranes, providing a positive feedback pathway.

Minireview: Actin/Membrane Interaction by ERM Proteins 34509
because the molar ratio of endogenous merlin/ERM was calculated to be 0.05-0.15 in cultured fibroblasts and epithelial cells, the ERM-merlin competition is not likely in vivo. Second, merlin would be functionally redundant for ERM proteins. Data from the suppression experiments of ERM proteins by antisense oligonucleotides discussed above are inconsistent with this explanation. Furthermore, this explanation was not supported by the observation that merlin-deficient mice showed an embryonic lethal phenotype (70). Third, it is also possible that merlin shows sequence similarity to ERM proteins because both are involved in Rho signaling pathways; for example, both bind to Rho-GDI. If this is the case, loss of function mutations of merlin may induce tumorigenesis through some disturbance in the Rho signaling pathway. In this connection, it should be pointed out that neurofibromin, which is responsible for neurofibromatosis type I, shows GTPase-activating protein activity for Ras (71). It is reasonable to speculate that both neurofibromatosis types I and II are caused by some dysregulation of small GTP-binding protein-dependent signaling pathways.

Perspective
In the past decade, it has been established that ERM proteins function as general cross-linkers in the cortical layer, coupled with signal transduction pathways such as Rho signaling. Because ERM proteins are expressed almost ubiquitously, this cross-linking system would be involved in various cellular events in various types of cells. Thus, in the coming decade, ERM proteins will attract increasing interest in many fields from not only biological but also medical researchers. For example, in the immune system ERM proteins are thought to play an important role in cell recognition of T lymphocytes by producing uropods (17,29).
It will also be interesting to study the interactions between ERM proteins and microtubules. Radixin was characterized as a marginal microtubule band-associated protein in nucleated erythrocytes (72), and in activated T-lymphocytes tubulin was co-concentrated at the uropods together with ERM proteins (17). Furthermore, it was also shown that ERM proteins have some homology with Tea 1, which is localized on the ends of microtubules and is critical for polarization in yeast (73). Further studies of the ERM proteinsmicrotubule interaction will provide new insight into the physiological functions of ERM proteins as well as the cortical actin layer.