Zonula occludens (ZO)‐1 is a multi‐domain scaffold protein known to have critical roles in the establishment of cell–cell adhesions and the maintenance of stable tissue structures through the targeting, anchoring, and clustering of transmembrane adhesion molecules and cytoskeletal proteins. Here, we report that ZO‐1 directly binds to MRCKβ, a Cdc42 effector kinase that modulates cell protrusion and migration, at the leading edge of migrating cells. Structural studies reveal that the binding of a β hairpin from GRINL1A converts ZO‐1 ZU5 into a complete ZU5‐fold. A similar interaction mode is likely to occur between ZO‐1 ZU5 and MRCKβ. The interaction between ZO‐1 and MRCKβ requires the kinase to be primed by Cdc42 due to the closed conformation of the kinase. Formation of the ZO‐1/MRCKβ complex enriches the kinase at the lamellae of migrating cells. Disruption of the ZO‐1/MRCKβ complex inhibits MRCKβ‐mediated cell migration. These results demonstrate that ZO‐1, a classical scaffold protein with accepted roles in maintaining cell–cell adhesions in stable tissues, also has an active role in cell migration during processes such as tissue development and remodelling.
Tight junctions (TJs), localized to the apical ends of the basolateral domains of plasma membranes, have key roles in regulating paracellular permeability and maintaining the apico‐basal polarity of epithelial cells (Balda and Matter, 1998; Matter and Balda, 2003; Anderson et al, 2004). In addition to their prime function as barriers and fences, TJs also contain proteins involved in the regulation of numerous cellular functions. To date, >40 different TJ proteins have been discovered (D'Atri and Citi, 2002; Gonzalez‐Mariscal et al, 2003; Schneeberger and Lynch, 2004). They can be divided into two groups. The first group are the integral membrane TJ proteins, including occludins, claudins, and JAMs (junctional adhesion molecules), which bridge apical intercellular spaces and form physical barriers (Furuse et al, 1993, 1998; Martin‐Padura et al, 1998). The second group are the plaque proteins (e.g. zonula occludens (ZO)‐1, ZO‐2, ZO‐3, cingulin, MAGI, Crumbs/PALS/PATJ, Par‐3/Par‐6/aPKC complex, etc), which serve as scaffolds to assemble integral membrane proteins, actin cytoskeletons, and cytosolic proteins in TJs (Citi et al, 1988; Ide et al, 1999; Gonzalez‐Mariscal et al, 2000; Roh et al, 2002; Hurd et al, 2003). Extensive studies have generated a wealth of information pertaining to the functions of these proteins in the formation of cell–cell junctions, but the functions of these junctional proteins during tissue development and remodelling (i.e. before the formation of stable cell–cell junctions) are much less clear.
ZO‐1 was the first TJ protein identified (Stevenson et al, 1986). Together with ZO‐2 (Jesaitis and Goodenough, 1994) and ZO‐3 (Haskins et al, 1998) they all belong to the membrane‐associated guanylate kinase (MAGUK) family of scaffold proteins, and these three proteins are now collectively called TJ MAGUKs (Woods and Bryant, 1993; Anderson et al, 1995; Kim, 1995). Each of these three ZO proteins contains three PDZ (PSD‐95/Discs large/ZO‐1) domains, one SH3 (Src‐homology 3) domain, and one GK (guanylate kinase) domain (Figure 1A) (Itoh et al, 1993; Willott et al, 1993; Jesaitis and Goodenough, 1994; Haskins et al, 1998). In contrast to other MAGUK proteins, ZO proteins (ZO‐1 and ZO‐2 in particular) contain a long proline‐rich carboxyl terminal region that may be responsible for their unique properties in TJs (Jesaitis and Goodenough, 1994; Itoh et al, 1997; Fanning et al, 1998; Gonzalez‐Mariscal et al, 1999; Ryeom et al, 2000). Interestingly, ZO‐1 also contains an additional ZU5 domain (initially found in ZO‐1 and UNC5; Ackerman et al, 1997; Leonardo et al, 1997) at its C‐terminus (Figure 1A). ZO proteins have long been assumed to have essential roles in TJ establishment and epithelial polarity maintenance, as their PDZ domains directly bind to the carboxyl tails of claudins and their GK domains interact with occludins (Fanning et al, 1998; Haskins et al, 1998; Itoh et al, 1999). Indeed, the simultaneous deletion of ZO‐1 and ZO‐2 leads to the complete disappearance of TJs in epithelial cell cultures, indicating that ZO‐1 and ZO‐2 are required for the formation of TJs (Umeda et al, 2006).
Although both mutations are embryonically lethal, mice lacking ZO‐1 and ZO‐2 display distinct phenotypes, indicating that the two scaffold proteins have non‐redundant roles in animal development (Katsuno et al, 2008; Xu et al, 2008). ZO‐2−/− mice died shortly after implantation because of an arrest in early gastrulation with decreased proliferation at E6.5 (Xu et al, 2008). In contrast, ZO‐1−/− mice embryos are indistinguishable from ZO‐1+/+ mice at E9.5 or earlier, and embryonic development defects of ZO‐1−/− mice begin appearing at E10.5 (Katsuno et al, 2008). Interestingly, ZO‐1−/− mice embryos also display severely inhibited angiogenesis in their yolk sacs without defects in vasculogenesis. In view of the well‐accepted roles of ZO‐1 in the formation of TJs, this inhibited angiogenesis is thought to be the result of defects in cell–cell adhesions (Katsuno et al, 2008). An alternative explanation is that it is caused by vessel endothelial cell migration defects. It has recently been shown that α5‐integrin and ZO‐1 physically interact with each other at the leading edge of migrating cells (Tuomi et al, 2009), suggesting that ZO‐1 has a role in cell migration in addition to its well‐accepted roles in mediating cell adhesion. Consistent with the observed α5‐integrin and ZO‐1 interaction, α5‐integrin null mice share a number of phenotypes with ZO‐1−/− mice (Goh et al, 1997). In cell culture models, ZO‐1 and ZO‐2 can partially replace each other both in TJ formation and in polarity establishment in cultured epithelia. However, ZO‐1−/− cells need a significantly longer time to form TJs and to polarize fully in spite of the upregulated expression of ZO‐2 (Umeda et al, 2006), implying that the C‐terminal ZU5 domain of ZO‐1 is important for the establishment of cell polarity and possibly cell migration (ZO‐2 lacks a ZU5 domain). It has been shown that the deletion of the ZU5 domain leads to a partial dissociation of ZO‐1 from TJs to the cytoplasm in MDCK cells (McNeil et al, 2006), although the underlying molecular mechanism is unclear.
Here, we show that ZO‐1 physically interacts with MRCKβ, a Cdc42 effector kinase involved in the membrane protrusions of motile cells (Leung et al, 1998; Gomes et al, 2005; Wilkinson et al, 2005; Tan et al, 2008), via its ZU5 domain. The biochemical basis of the ZO‐1 ZU5 domain‐mediated MRCKβ interaction is characterized in detail. We further show that a fraction of ZO‐1 specifically localizes at and targets MRCKβ to the leading edge of migrating cells. We discover that the formation of the ZO‐1/MRCKβ complex requires the binding of Cdc42 to the kinase. Finally, we demonstrate that the ZO‐1‐mediated positioning of MRCKβ at the leading edge of cells is necessary for their migrations. Our studies reveal that, in addition to its well‐accepted roles in mediating cell–cell adhesion of organized tissues, ZO‐1 also has an active role in cell migrations, and thus is likely to be linked to diverse cellular processes ranging from tissue remodelling in the development to cancer cell migrations.
ZO‐1 binds to MRCKβ via its unique ZU5 domain
We hypothesized that the unique ZU5 domain of ZO‐1 is at least partly responsible for the differences between the functions of ZO‐1 and ZO‐2. To test this hypothesis, we searched for potential ZO‐1 ZU5‐binding proteins using yeast two‐hybrid screening (Y2H) with the wild‐type (WT) ZU5 domain as the bait. Among 10 positive clones isolated under the highly stringent screening condition, seven encode partial fragments of MRCKβ sharing an overlapping region consisting of 330 amino‐acid residues (residues 745–1074; Figure 1A and B). Y2H‐based assays and in vitro biochemical binding experiments were further used to map out the minimal ZO‐1 ZU5‐binding fragment of MRCKβ (Figure 1C and D), which consists of a 153‐residue fragment (referred to as ZBD for ZO‐1‐binding domain), including the C1 domain and a connecting sequence between the CC3 (the coiled‐coil three region) and the C1 domain of MRCKβ (Figure 1A, C, and D). Interestingly, neither Y2H nor in vitro binding assays could detect any interaction between ZO‐1 ZU5 and MRCKα (Figure 1D3), indicating that ZO‐1 ZU5 specifically binds to the β isoform of MRCK (see Figure 5 for more details).
To our surprise, the endogenous ZO‐1 and MRCKβ do not seem to interact with each other in COS‐7 cells cultured to near confluent states, as MRCKβ is largely diffused in cytosol and nuclei with enrichment in cell periphery (cell protrusions in particular) (Leung et al, 1998), whereas ZO‐1 is diffused throughout the cells (Figure 1E1, top panels). Scratch wounding of COS‐7 cells led to drastic changes in the localizations of both ZO‐1 and MRCKβ in cells facing the wounding front. In these wounded cells, a large portion of ZO‐1 becomes to localize at the leading edge of the wounded cells (Figure 1E1; Supplementary Figure S1A). Concomitantly, a significant proportion of MRCKβ is also localized at the leading edge of the wounded cells together with ZO‐1 (Figure 1E1). A co‐immunoprecipitation assay showed that endogenous ZO‐1 and MRCKβ physically associate with each other only in wounded COS‐7 cells (Figure 1E2). The above results indicate that the interaction between ZO‐1 and MRCKβ only occurs in motile cells (see Figure 6 for more details). We further showed that the leading edge co‐localization of ZO‐1 and MRCKβ was also detected in two migrating human cancer cell lines (Supplementary Figure S2), pointing to a potential general role of the ZO‐1/MRCKβ complex in cell migrations.
ZO‐1 anchors MRCKβ at the leading edge of migrating cells
We next asked whether ZO‐1 determines MRCKβ localization at the leading edge of migration cells or vice versa by transiently transfecting COS‐7 cells with RFP‐tagged MRCKβ and GFP‐tagged ZO‐1. Co‐transfection of the WT MRCKβ and ZO‐1 led to prominent co‐localization of the two proteins at the leading edge of transfected cells (Figure 2A3 and D). Deletion of ZBD from MRCKβ dramatically reduced its leading edge localization, but the leading edge localization pattern of ZO‐1 did not change (Figure 2A4 and D). Conversely, although the deletion of the ZU5 domain did not change the leading edge localization of ZO‐1, essentially no MRCKβ was found at the leading edge of the co‐transfected cells (Figure 2A5 and D). Thus, the binding of ZO‐1 through its ZU5 domain to MRCKβ ZBD serves to target the enzyme to the leading edge of migrating cells, and regions outside of the ZU5 domain in ZO‐1 determine its own leading edge localization in migrating cells. The PDZ domains of ZO‐1 might be responsible for its leading edge localization by binding to α5‐integrin (Tuomi et al, 2009).
To avoid possible interference from endogenous proteins in the transient transfection experiments, we removed them individually via the RNAi approach and repeated the above experiments. Knockdown of ZO‐1 eliminated the leading edge localization of the endogenous MRCKβ but not MRCKα (Figure 2B2, C, and D; Supplementary Figure S1B), indicating that ZO‐1 is specifically required for the leading edge localization of MRCKβ. As expected, transfection of the RNAi‐resistant WT ZO‐1 rescued the leading edge localization of MRCKβ, but the ZO‐1 mutant without its ZU5 domain did not rescue the WT ZO‐1 knockdown even though the mutant itself was well localized to the leading edge (Figure 2B3, B4, and D). Consistent with the results from the overexpression experiments, knockdown of MRCKβ did not change the leading edge localization of ZO‐1 (Figure 2B6, C, and D). The RNAi‐resistant MRCKβ expressed was also found to co‐localize with the endogenous ZO‐1 at the leading edge of the transfected cells (Figure 2B7 and D). Taken together, the above data reveal that ZO‐1 acts upstream of MRCKβ in anchoring the ZO‐1/MRCKβ complex at the leading front of migrating cells.
ZO‐1 ZU5 adopts a partial ZU5‐fold
To elucidate the mechanistic basis of ZO‐1 ZU5‐mediated MRCKβ binding, we tried to determine the three‐dimensional structure of the domain using NMR spectroscopy. Due to poor sample behaviour caused by unfavourable conformational exchanges and non‐specific aggregations (Supplementary Figure S3, and see below), the WT ZO‐1 ZU5 (residues 1631–1748) is not amendable for NMR‐based structure determination. Aided by extensive amino‐acid sequence analysis and structural predictions, we tested numerous point as well as deletion mutations (>30) of the domain by inspecting the 1H–15N HSQC spectrum of each mutant. One of such mutants, with Met1699 and Cys1700 substituted with Ala (referred to as ZU5_MC/AA) displayed an excellent sample condition suitable for NMR‐based structure determination of the domain (Supplementary Figure S3A).
The solution structure of ZU5_MC/AA was determined to a high resolution using NMR spectroscopy (Figure 3; Supplementary Table 1). Except for two loops (the β4/β5 and β7/β8 loops) that are intrinsically flexible (Figure 3A; Supplementary Figure S3C), all the other regions of the domain are well defined. The ZU5 domain adopts a β‐barrel fold formed by two anti‐parallel β sheets packed with each other in parallel. The first β sheet contains four strands (β1, β4–6), and the second one contains another four strands (β2, 3, 7, and 8) (Figure 3B). A comparison of the HSQC spectrum of the WT ZU5 domain at low concentration (i.e. not heavily aggregated form) with that of the MC/AA mutant revealed that most of the peaks from the WT ZU5 overlapped well with or only experienced relatively small shifts with respect to those from the MC/AA mutant; the exceptions are the residues from β5, β6, and part of β4, indicating that the overall conformation of the protein is not altered by the mutation (Supplementary Figure S3A and B). The residues of the WT ZU5 domain that experience significant mutation‐induced shift changes display either doublets or very broad peaks in the HSQC spectrum (Supplementary Figure S3A and B), indicating that these residues (β5 and the β4/β5 loop in particular) undergo slow‐to‐intermediate conformational exchanges. The most likely interpretation of these data is that the WT ZO‐1 ZU5 adopts two distinct conformations: one seen in the structure of the MC/AA mutant with β5 pairing with β6, and the other with β5 swinging out to form an ill‐defined loop connecting β4 and β6 (Supplementary Figure S3D). The structure seen in the MC/AA mutant represents the closed conformation, as the mutant cannot bind to MRCKβ (Figure 4A1 and B). The conformation with β5 not paired with β6 represents an open state capable of binding to MRCKβ (Supplementary Figure S3D and see below for details).
Next, we explored the structural requirements on ZO‐1 ZU5 for its binding to MRCKβ. The residues forming the β8 strand (‘VSVLI’) at the extreme C‐terminal end of ZO‐1, which is predicted not to be a part of ZU5 domain, make extensive hydrophobic contact with the rest of the ZU5 domain (Figure 3D). All the residues involved in the hydrophobic interactions are evolutionarily conserved (Figure 3E). Additionally, the bulky aromatic ring of the last residue Phe1748 of ZO‐1 is inserted into a hydrophobic pocket (Figure 3D), further stabilizing the interaction between the ZU5 domain and the C‐terminal tail. Deletion of the entire C‐terminal fragment or even only Phe1748 both lead to the complete disruption of the interaction between ZO‐1 and MRCKβ, assayed through in vitro biochemical binding experiments and using an imaging‐based assay for the ZO‐1‐mediated leading edge localization of MRCKβ (Figure 4A2, A3, C2, C3, and D). As a control, the addition of an Ala residue to the C‐terminus of ZO‐1 has no effect on the ZO‐1‐mediated localization of MRCKβ (Figure 4A4 and C4). Thus, the structural integrity of the ZU5 domain is required for ZO‐1 to interact with MRCKβ.
We note that the 3D structure of ZO‐1 ZU5 is significantly different from the recently determined structures of the ZU5 domains from Unc5H2 (Wang et al, 2009) and ankyrin‐R (Ipsaro et al, 2009). The β barrel of ZO‐1 ZU5 contains 8 β strands, compared with 12 in UNC5 ZU5 and 11 in ankyrin‐R ZU5 (Supplementary Figure S4). Amino‐acid sequence alignment analysis, together with the structures of these three ZU5 domains, reveals that ZO‐1 ZU5 lacks the sequences corresponding to the β8, β9, β10, and β12 strands of UNC5 ZU5 and the β8, β9, and β11 strands of ankyrin ZU5 (Supplementary Figures S4 and S7), indicating that the ZO‐1 ZU5 domain adopts a partial ZU5‐fold. Interestingly, the solvent‐exposed surface of ZO‐1 ZU5 corresponding to the missing β strands side is highly hydrophobic (Figure 3C), and this hydrophobic surface is responsible for the binding between the partial ZU5 domain and MRCKβ (see below for details).
Structural basis of the ZO‐1/MRCKβ complex formation
The optimal experiment to elucidate the molecular basis governing the ZO‐1 and MRCKβ interaction would be to determine the structure of the ZO‐1 ZU5/MRCKβ complex. However, due to the poor behaviour of numerous complex samples, we were not able to achieve this goal either by NMR spectroscopy or by X‐ray crystallography. As an alternative approach, we investigated the structural basis of the interaction between the WT ZO‐1 ZU5 (not the MC_AA mutant as it does not bind to MRCKβ) and GRINL1A (glutamate receptor, ionotropic, N‐methyl‐d‐aspartate‐like 1A combined protein; Roginski et al, 2004), another binder identified from our Y2H screening. To probe the structural relevance between the ZO‐1 ZU5/GRINL1A complex and the ZO‐1 ZU5/MRCKβ complex, we used the same set of ZO‐1 ZU5 mutants for the disruption of the ZO‐1 ZU5/MRCKβ complex (Figure 4C and D) to test their binding to GRINL1A. Neither ‘ZO‐1 ZU5‐ΔCT’ nor ‘ZO‐1 ZU5‐ΔF’ was capable of binding to the myc‐tagged full‐length GRINL1A (Supplementary Figure S5A). Additionally, the full‐length GRINL1A was found to target ZO‐1 ZU5 to junctional membranes in polarized MDCK cells, but failed to do so to the ‘ZO‐1 ZU5‐ΔCT’ and ‘ZO‐1 ZU5‐ΔF’ mutants (Supplementary Figure S5B). The above results suggest that GRINL1A and MRCKβ likely bind to the same exposed hydrophobic pocket in ZO‐1 ZU5 (see Figure 5 for details). Next, a 22‐residue peptide fragment of GRINL1A was mapped as the minimal ZO‐1 ZU5‐binding region (Supplementary Figure S6), and circular dichroism analysis indicated that this 22‐residue peptide binds to ZO‐1 ZU5 in a β‐strand conformation (Supplementary Figure S6C). Finally, we determined the structure of the WT ZO‐1 ZU5 in complex with the GRINL1A peptide using NMR spectroscopy (see Supplementary Table 1 and Supplementary Figures S6–S8 for the details of the complex structure).
All of the ZU5 β strands except for β5 adopt similar conformations upon binding to the GRINL1A peptide (Figure 5A). The GRINL1A peptide forms two additional β strands (β9 and β10 hairpin) that pair in parallel with β5 and part of β6 from ZU5. In apo‐ZU5_MC/AA, β5 pairs with β6 to form part of the β sheet, thus preventing the target peptide (β9 specifically) from forming anti‐parallel β sheet with β5 and β6 (i.e. apo‐ZU5_MC/AA adopts a closed conformation; see Figure 3 and Supplementary Figure S8A). In the ZU5/GRINL1A peptide complex, β5 dissociates from β6 and swings outwards to one end of the β barrel. Together with β6, it forms a continuous receiving β strand for pairing with β9 of the GRINL1A peptide (Supplementary Figure S8B). The most prominent conformational changes of ZO‐1 ZU5 induced by the GRINL1A peptide binding are in the β5 strand (as discussed above) and the β4/β5 loop. While the β4/β5 loop is flexible in apo‐ZU5_MC/AA (Figure 3A), it becomes highly ordered in the complex due to its extensive interactions with residues from β5/β6 loop and the β9 strand (Figure 5A; Supplementary Figure S8). The structure of the ZO‐1 ZU5/GRINL1A peptide complex, together with the biochemical studies presented in this work, reveals a unique target‐binding mechanism. Upon binding to ZO‐1 ZU5, the GRINL1A peptide forms a β hairpin that augments the existing partial β barrel, analogous to the β8 and β12 in UNC5 ZU5 and β8 and β11 in ankyrin‐R ZU5 (Figure 5B; Supplementary Figures S4 and S7), and the resulting ZO‐1 ZU5/GRINL1A complex assumes a compact, complete ZU5‐fold stable in solution. Therefore, the target‐binding mode seen in ZO‐1 ZU5 can be described as a target‐induced domain complementation: the β strands of ZU5 and its target combine to form a complete ZU5‐fold.
Importantly, the ZO‐1 ZU5‐binding sequence of GRINL1A aligns very well, at both amino‐acid sequence and secondary structure levels, with a fragment of MRCKβ N‐terminal to its C1 domain (Figure 5B). Deletion of parts or the whole of this presumed ZO‐1 ZU5‐binding sequence disrupted the interaction between MRCKβ and ZO‐1 (Figure 5D). Mutations of two hydrophobic residues in either β9 (Phe952 and Tyr954 to Gln) or β10 (Leu964 and Phe966 to Gln) also abolished the MRCKβ/ZO‐1 complex formation (Figure 5D). It is further noted that the predicted ZO‐1 ZU5‐binding sequence of MRCKβ is highly conserved among different species (Figure 5B). However, the amino‐acid sequence of the corresponding peptide fragment (the predicted β10 sequence in particular) in MRCKα is significantly different from that of MRCKβ (Figure 5B), explaining the observation made earlier that MRCKα does not interact with ZO‐1 (Figure 1D). We directly tested the above structure‐based analysis by replacing the β10 cassette of MRCKα (‘DQFERS’) with the corresponding segment from MRCKβ (‘DLTFRT’), and found that the resulting MRCKα chimera was capable of binding to ZO‐1 (Figure 5E). Taken together, the above structural and biochemical experiments strongly indicate that the peptide fragment preceding the C1 domain of MRCKβ binds to ZO‐1 in a β‐hairpin conformation, forming a complete ZU5‐fold in the ZO‐1/MRCKβ complex, although the final proof would have to come from the structure of the ZO‐1 ZU5/MRCKβ ZBD complex.
Finally, we note that the structure of the ZO‐1 ZU5/GRINL1A peptide complex (and the corresponding MRCKβ β‐hairpin peptide complex) still lacks one or two β strands (corresponding to β9 in ankyrin‐R ZU5 or β9 and β10 in UNC5 ZU5; Supplementary Figure S7) compared with the complete ZU5‐folds. We showed that the C1 domain is also required for binding to ZO‐1 ZU5 (Figure 1). It is likely that the MRCKβ C1 domain has an additional structural role in the formation of the MRCKβ/ZO‐1 complex (Figure 5C). Supporting this hypothesis, we found that the C1 domain and its preceding β‐hairpin sequence together are required for the formation of the stable ZO‐1 ZU5/MRCKβ ZBD complex (Supplementary Figure S9). This analysis also provides an explanation for the weak interaction observed between ZO‐1 and the GRINL1A peptide (Kd∼20 μM) (Supplementary Figure S6D). It is possible that GRINL1A is not a physiological ZO‐1 partner, but the protein acted as a useful tool here for us to elucidate the ZO‐1 ZU5/MRCKβ interaction mechanism.
ZO‐1 binds to MRCKβ in a Cdc42‐dependent manner
In our initial attempts to verify the direct binding between ZO‐1 ZU5 and the full‐length MRCKβ in vitro, we failed to detect any interaction (Figure 6A1). A Y2H‐based assay also showed that the full‐length MRCKβ does not bind to ZO‐1 ZU5 (Figure 6A2). We further found that the inclusion of the PH‐CH domain tandem C‐terminal to the C1 domain sufficiently prevented MRCKβ from binding to ZO‐1 ZU5, indicating that the MRCKβ C‐terminal tail adopts a closed conformation incapable of binding to ZO‐1 (Figure 6A2). The closed conformation of the MRCKβ tail was further confirmed by the observation of a direct interaction between the CC3–C1 domains and the PH‐CH‐CRIB domains of the enzyme (Supplementary Figure S10A). Interestingly, the removal of the Cdc42‐binding CRIB domain did not alter the closed conformation of MRCKβ, implying that the PH‐CH domain tandem is largely responsible for the closed conformation of MRCKβ (Figure 6A2).
To investigate the potential opening mechanism of the MRCKβ's closed conformation, we tested the effect of Cdc42 on the binding between MRCKβ and ZO‐1 ZU5, as the Cdc42‐binding CRIB domain immediately follows the PH‐CH tandem. Interestingly, we found that Cdc42‐loaded MRCKβ can bind to ZO‐1 ZU5, and removal of the CRIB domain abolished the Cdc42‐dependent binding of MRCKβ to ZO‐1 ZU5 (Figure 6B). Consistent with the above biochemical data, MRCKβ lacking the Cdc42‐binding CRIB domain cannot be anchored at the leading edge of migrating COS‐7 cells (Figure 6C). Overexpression of a dominant negative form of Cdc42 (‘Cdc42N17’; Sudhaharan et al, 2009) also eliminated the leading edge localization of MRCKβ (Figure 6D). As a control, MRCKβ was found to localize at the leading edge of cells expressing a constitutively active form of Cdc42 (‘Cdc42V12’) (Figure 6D). Finally, as phorbol esters are known to bind to the C1 domain of MRCKβ (Choi et al, 2008), we tested their potential effects on the opening of the MRCKβ's closed conformation. We found that 12‐O‐tetradecanoylphorbol‐13‐acetate, a commonly used phorbol ester, did not potentiate the MRCKβ/ZO‐1 ZU5 complex formation, indicating that phorbol esters cannot by‐pass Cdc42 to release the closed conformation of MRCKβ (Supplementary Figure S10B). Taken together, the data shown in Figure 6 uncover that MRCKβ adopts a closed conformation incapable of binding to ZO‐1 ZU5. The binding of Cdc42 to its C‐terminal CRIB domain induces conformational changes to the neighbouring PH‐CH domain tandem, thereby exposing MRCKβ ZBD for binding to ZO‐1 (Figure 6E).
ZO‐1/MRCKβ complex is required for cell migrations
Finally, we tested the roles of the ZO‐1/MRCKβ complex in cell migrations using a wound‐healing assay (Figure 7; Rodriguez et al, 2005). The control cells showed expected migration‐induced wound closure after scratch wounding (Figure 7A). Compared with the scRNA‐mediated control (Figure 7B and F), RNAi‐mediated knockdown of either ZO‐1 or MRCKβ largely eliminated cell movements upon wounding (Figure 7C and G). We checked that the defective cell migrations induced by the ZO‐1 or MRCKβ knockdowns are not the results of inhibited cell proliferations (Supplementary Figure S11). Restoring the expression of ZO‐1 or MRCKβ effectively rescued cell migration defects induced by RNAi‐mediated knockdowns of the two proteins, respectively (Figure 7D and H). In contrast, neither the ZO‐1 (ΔZU5) mutant nor the MRCKβ (ΔZBD) mutant rescued cell migration defects induced by the knockdowns (see Figure 7E, I, and J for quantifications), indicating that the direct interaction between ZO‐1 and MRCKβ is required for cell migration (with the assumption that these domain deletions specifically disrupt the ZO‐1/MRCKβ interaction without other adverse effects). The data in Figure 7, together with the results presented above, demonstrate that ZO‐1 has an active role in regulating cell migration by targeting Cdc42‐activated MRCKβ at the leading edge of migrating cells.
As a founding member of the MAGUK family of scaffold proteins, ZO‐1 is well recognized for its vital roles in the assembly of cell–cell adhesion complexes that are critical for the formation of intercellular connection machineries, including TJs and adherens junctions, in a wide variety of tissues (Mitic and Anderson, 1998; Fanning and Anderson, 2009; Tsukita et al, 2009). In this work, we show that ZO‐1 actively regulates cell migration by binding to the cytoskeletal dynamics regulatory protein kinase MRCKβ, and targeting the enzyme to the leading edge of migrating cells. Our findings reveal that ZO‐1, and perhaps other scaffold proteins involved in cell–cell adhesions, can function either as cell‐to‐cell connectors in maintaining tissue organization or as facilitators in modulating cells migration and tissue remodelling. The ability of ZO‐1 to regulate cell migration is highly consistent with the in vivo developmental functions of ZO‐1 from genetic studies. The severe inhibition of angiogenesis in mice lacking ZO‐1 at early developmental stages could be the result of migration defects of blood vessel endothelial cells (Katsuno et al, 2008). The ZO‐1/MRCKβ complex in the regulation of cell migration is likely to be important for many human pathological conditions, including cancers. It has been shown that cancer cell migration requires Cdc42‐MRCK (Wilkinson et al, 2005; Gaggioli et al, 2007). Additionally, ZO‐1 has been found to be localized at the leading edge of migrating cancer cells together with α5‐integrin (Tuomi et al, 2009), although it is not clear how the ZO‐1/α5‐integrin complex promotes cell migration. Our finding of co‐localization of ZO‐1 and MRCKβ at the leading edge of several migrating human cancer cell lines (Supplementary Figure S2) suggests that the ZO‐1/MRCKβ complex might act as downstream of the ZO‐1/integrin complex and have a critical role in the cancer cell migration.
The direct interaction between ZO‐1 and MRCKβ at the leading edge of migrating cells provides a molecular explanation for ZO‐1's role in regulating cell migration and perhaps tissue remodelling. Interestingly, other scaffold proteins have also been implicated to actively participate in the regulation of cell migration, in addition to their well‐established roles as scaffolds for cell–cell junction organization. For example, Scribble, another multiple PDZ scaffold protein required for the establishment and maintenance of cell polarity in diverse tissues, has been shown to have active roles in positioning both Cdc42 and PAK in migrating cells (Osmani et al, 2006; Nola et al, 2008), although Scribble does not directly bind to PAK.
In this work, we also show that MRCKβ adopts a closed conformation incapable of binding to ZO‐1 ZU5 (Figure 6). The binding of Cdc42 to the MRCKβ CRIB domain exposes the ZO‐1 ZU5‐binding site, which is otherwise covered by the PH‐CH tandem of the enzyme, even though the CRIB domain is not directly involved in covering the ZO‐1 ZU5‐binding site (Figure 6E). Thus, ZO‐1 specifically binds to the activated form of MRCKβ. It is envisioned that the binding of Cdc42 induces some sort of conformational changes in the PH‐CH tandem, thereby allowing the activated MRCKβ to interact with ZO‐1. Further work is under progress in elucidating the nature of the Cdc42‐induced conformational changes of MRCKβ. In the process of converting cultured cells from confluent states (i.e. with stable cell–cell adhesions) to migrating states, a fraction of ZO‐1 moves to the leading edge of cells (Figure 1E), although the molecular basis governing this ZO‐1 localization change is unclear at this stage.
ZU5 domain, originally identified as an ∼120 amino‐acid residue domain present in ZO‐1 and UNC5, is also found in various ankyrins and an apoptotic machinery scaffold PIDD (p53‐induced protein with a death domain) (Ackerman et al, 1997; Leonardo et al, 1997; Tinel and Tschopp, 2004; Bennett and Healy, 2008). Although these ZU5 domain‐containing proteins are functionally diverse, they are all multi‐domain scaffold proteins with no intrinsic enzyme activities. Interestingly, except for ZO‐1, which contains an orphan ZU5 domain at its extreme C‐terminal end, all other known ZU5 domain proteins contain a UPA domain and a death domain C‐terminal to their ZU5 domain (Ipsaro et al, 2009). Additionally, ankyrins and PIDD contain two ZU5 domains arranged in tandem. Recently, the structures of the ZU5 domains of UNC5 (Wang et al, 2009) and ankyrin‐R (Ipsaro et al, 2009) have been reported (Supplementary Figure S7). The UNC5 ZU5 domain adopts a β‐barrel fold and interacts with its own death domain, and the functions of both the ZU5 and death domains are thereby suppressed by an autoinhibition mechanism (Wang et al, 2009). The ankyrin‐R ZU5 domain specifically binds to a tandem spectrin repeats (Ipsaro et al, 2009). The common theme emerging from these studies, together with this work, is that ZU5 domains function as specific protein–protein interaction modules capable of binding to very different target proteins. A comparison of the structure of ZO‐1 ZU5 with those of UNC5 and ankyrin‐R ZU5 domains reveals that ZO‐1 ZU5 adopts a partial ZU5‐fold lacking 2–3 strands at one edge of the β barrel (Supplementary Figure S7). We show here that target proteins bind to ZO‐1 ZU5 in a β‐sheet conformation. The β strands from the target protein complete the partial ZU5‐fold of ZO‐1 (i.e. ZO‐1 ZU5 binds to its targets via a domain complementation mechanism). The results described in this study not only expand our knowledge of ZU5 domains in general, but also lay a foundation for future investigations of specific cellular functions of ZO‐1 that arise from its unique ZU5 domain, which is not present in ZO‐2 and ZO‐3.
In summary, the identification of the specific interaction between ZO‐1 and MRCKβ and the functional characterization of the ZO‐1/MRCKβ complex in cell migrations presented in this work uncover a previously uncharacterized role that the classical junctional scaffold protein ZO‐1 has in regulating cell migration. The results presented here not only call for further study of this much less explored function of ZO‐1 (and perhaps other scaffold proteins involved in cell–cell adhesions) in cell motility, but also indicate to therapeutic possibilities in cancers based on interfering with cancer cell migrations by targeting proteins such as ZO‐1.
Materials and methods
The ZO‐1 monoclonal antibody (1A12) and the anti‐α‐tubulin antibody were purchased from Invitrogen; anti‐MRCKα, anti‐MRCKβ, and anti‐GFP antibodies were purchased from Santa Cruz; and the anti‐Myc antibody was purchased from Roche.
Protein expression and purification
DNA sequences encoding human ZO‐1 ZU5 domain (residues 1631–1748), various mutants, the fused ZO‐1 ZU5/GRINL1A complex, and Cdc42QL were individually cloned into a modified version of the pET32a vector. The resulting ZU5 domain constructs, the fused ZO‐1 ZU5/GRINL1A complex, and Cdc42QL each contained a His6‐tag in the N‐terminus. Point mutations of ZU5 proteins were created using the standard PCR‐based mutagenesis method and confirmed by DNA sequencing. Recombinant proteins were expressed in Escherichia coli BL21 (DE3) host cells at 16°C. His6‐tagged ZU5 proteins and the fused ZO‐1 ZU5/GRINL1A complex expressed in bacterial cells were purified by Ni2+‐NTA agarose (Qiagen) affinity chromatography followed by size‐exclusion chromatography. Uniformly, 15N or 15N, 13C‐labelled ZU5 proteins and the fused ZO‐1 ZU5/GRINL1A complex were prepared by growing bacteria in M9 minimal medium using 15NH4Cl as the sole nitrogen source or 15NH4Cl and 13C6‐glucose (Cambridge Isotope Laboratories Inc.) as the sole nitrogen and carbon sources, respectively.
NMR structure determination
NMR samples contained 1.0 mM of the ZU5 protein and the fused ZO‐1 ZU5/GRINL1A complex in 50 mM Tris (pH 7.0, with 1 mM DTT, and 1 mM EDTA) in 90% H2O/10% D2O or 99.9% D2O. NMR spectra were acquired at 35°C on Varian Inova 500 or 750 MHz spectrometers. Backbone and side‐chain resonance assignments of ZU5 proteins were achieved by the standard heteronuclear correlation experiments (Bax and Grzesiek, 1993; Kay and Gardner, 1997). The side chains of aromatics were assigned using 1H two‐dimensional total correlation spectroscopy/NOESY experiments. Approximate inter‐proton distance restraints were derived from NOESY experiments (a 1H 2D homonuclear NOESY, a 15N‐separated NOESY, and a 13C‐separated NOESY). Structures were calculated using the program CNS (Brunger et al, 1998).
Yeast two‐hybrid screening
Y2H was performed using the WT ZO‐1 ZU5 domain as the bait. Human ZO‐1 (residues 1631–1748) was subcloned into the pDBLeu vector. The bait plasmid and a random‐primed cDNA library from rat hippocampus in pPC86 vector were transformed into PJ69‐4A yeast cells and grown on quadruple minus plates that lack leucine, tryptophan, adenine, and histidine. Positive clones were scored by both growth and blue assays (Cao et al, 2007). Plasmid cDNAs from positive clones were recovered and back‐transformed with either the bait or the empty pDBLeu vector into both PJ69‐4A and HF7c yeast strains to confirm the interaction. All of the cDNA constructs were confirmed by sequencing.
GST pull‐down assays
Crude extracts of various fragments of N‐terminal Myc‐ or GFP‐tagged MRCKβ expressed in HEK293 cells were first incubated with 50 μl of Glutathione SepharoseTM 4B Fast Flow (GE Healthcare) 50% slurry beads in a lysis buffer (50 mM HEPES, pH 7.6, 150 mM NaCl, 1.5 mM MgCl2, 100 mM NaF, 1 mM EGTA, 1% Triton, 10% glycerol, and various protease inhibitors) for half an hour. The cleared supernatants were transferred to a new batch of 50 μl slurry beads, mixed with GST or GST‐tagged proteins (10 μl from 1 mg/ml stock solutions) for 2 h at 4°C. After washing the beads three times with the lysis buffer, the proteins captured by the beads were eluted by boiling with the SDS–PAGE sample buffer, resolved by SDS–PAGE, and detected either by Coomassie blue staining or by immunoblotting with specific antibodies. Purified His‐tagged Cdc42QL was incubated with 100 μM GTP analogue GMPPNP in a reaction buffer (50 mM Tris, 200 mM (NH4)2SO4, pH 8.5) for 5 h before mixing with GST‐tagged proteins.
For immunoprecipitation of overexpressed proteins, COS‐7 cells grown on 100 mm culture dishes were co‐transfected with specific cDNA constructs. Cells were harvested 4 h after wounding. Each cell lysate was incubated with 50 μl protein G beads (50% slurry) in the lysis buffer for 1 h, and transferred to a tube containing 50 μl protein G beads (50% slurry) and 2.5 μg anti‐ZO‐1 antibody. After incubation for 4 h at 4°C, the beads were washed with the lysis buffer three times. The captured proteins were then eluted using the SDS–PAGE buffer and analysed with specific antibodies.
Cell culture, immunostaining, and imaging
All cDNA constructs used in this study were cloned into mammalian expression vector pEGFP (Invitrogen) and pERFP for imaging studies, and pCMV‐Myc (Invitrogen) and pEGFP (Invitrogen) for pull‐down assays. The pSUPER vector was used to drive the expression of shRNA in COS‐7 cells. The shRNAs used in this study include ZO‐1 (KD1: GATCAAATTCTCAGGGTAA; KD2: CCGAAGAAGTTTTGAGAAT), and MRCKβ (KD1: CCGAAGAGCTCGAGGCTTT; KD2: TCGAGAAGACTTTGAAATA; KD3: GATAAATACGAACGAGAAA). The scramble siRNAs were designed using siRNA Wizard™ v3.1. COS‐7 cells and HEK293 cells were cultured in minimum essential medium supplemented with 10% fetal bovine serum (FBS). Transfection of HEK293 cells with plasmid DNA was performed using Lipofectamine (Invitrogen) according to the manufacturer's instructions. Transfection of COS‐7 cells with plasmid DNA, shRNA or scRNA was performed by electroporation (Amaxa Biosystems). In the knockdown experiments, cells were fixed ∼50 h after transfection. In immunofluorescence studies, cells on cover slips were fixed with 4% paraformaldehyde for 20 min. Cells expressing the transfected constructs were identified by immumostaining of the N‐terminal tag with corresponding primary and secondary antibodies. For the immunostaining of endogenous proteins, cells were incubated with primary antibody at 4°C overnight. Alexa 488‐ and Alexa 594‐conjugated secondary antibodies against goat and mouse immunoglobulin were obtained from Molecular Probes. All images were acquired on a Nikon TE2000E inverted fluorescent microscope.
Wound‐healing assays and cell proliferation analysis
COS‐7 cells at about 90% confluence were scratched with a pipette tip and replaced in fresh 10% FBS‐containing medium to induce cell migrations. Images were processed with the MetaMorph software package. Cell proliferations of COS‐7 cells used for the wound‐healing assays were measured with WST‐1 (Dojindo) according to the manufacturer's protocol.
The atomic coordinates of ZO‐1 ZU5_MC/AA mutant and ZO‐1 ZU5/GRINL1A complex have been deposited in the Protein Data Bank with accession codes 2KXR and 2KXS, respectively.
Supplementary data are available at The EMBO Journal Online (http://www.embojournal.org).
Conflict of Interest
The authors declare that they have no conflict of interest.
We thank Dr Thomas Leung for MRCKβ cDNA, Drs Alan Fanning and Ian Macara for providing the human ZO‐1 cDNA, Dr Raymond Roginski for the GRINL1A cDNA, Miss Ling‐Nga Chan and Jia Chen for technical assists, and Mr Anthony Zhang for editing of the manuscript. This work was supported by grants from the Research Grants Council of Hong Kong to MZ (HKUST663407, 663808, 664009, CA07/08.SC01, SEG_HKUST06, and AoE/B‐15/01‐II), WF (660708 and 660709), and JX (HKUST6/CRF/08 and 663107), the National Major Basic Research Program of China (2009CB918600 and 2011CB910500), the National Natural Science Foundation of China (31070657), and the Knowledge Innovation Program of the Chinese Academy of Sciences (KSCX2‐YW‐R‐154).
Author contributions: LH, WF, WW, RW, and MZ designed the experiments; LH, WW, and RW performed all of the experiments except the initial Y2H screening; CK and JX performed the initial Y2H screening, LH, WW, RW, WF, and MZ analysed the data; LH, WF, and MZ wrote the paper; and MZ coordinated the entire research project.
- Copyright © 2011 European Molecular Biology Organization