Characterization of ZO-2 as a MAGUK Family Member Associated with Tight as well as Adherens Junctions with a Binding Affinity to Occludin and α Catenin*

ZO-2, a member of the MAGUK family, was thought to be specific for tight junctions (TJs) in contrast to ZO-1, another MAGUK family member, which is localized at TJs and adherens junctions (AJs) in epithelial and nonepithelial cells, respectively. Mouse ZO-2 cDNA was isolated, and a specific polyclonal antibody was generated using corresponding synthetic peptides as antigens. Immunofluorescence microscopy with this polyclonal antibody revealed that, similarly to ZO-1, in addition to TJs in epithelial cells, ZO-2 was also concentrated at AJs in nonepithelial cells such as fibroblasts and cardiac muscle cells lacking TJs. When NH2-terminal dlg-like and COOH-terminal non-dlg-like domains of ZO-2 (N-ZO-2 and C-ZO-2, respectively) were separately introduced into cultured cells, N-ZO-2 was colocalized with endogenous ZO-1/ZO-2, i.e. at TJs in epithelial cells and at AJs in non-epithelial cells, whereas C-ZO-2 was distributed along actin filaments. Consistently, occludin as well as α catenin directly bound to N-ZO-2 as well as the NH2-terminal dlg-like portion of ZO-1 (N-ZO-1) in vitro. Furthermore, immunoprecipitation experiments revealed that the second PDZ domain of ZO-2 was directly associated with N-ZO-1. These findings indicated that ZO-2 forms a complex with ZO-1/occludin or ZO-1/α catenin to establish TJ or AJ domains, respectively.

Generation and maintenance of specialized membrane domains are required for cells to exert their physiological functions, and the underlying molecular mechanisms of these processes are attracting increasing interest from cell biologists. As components of the machinery responsible for membrane specialization, a new gene family was identified which is now called the MAGUK family (membrane-associated guanylate kinase homologues) (for reviews, see Refs. [1][2][3]. MAGUKs are multidomain proteins that consist of PDZ, SH3, and GUK (guanylate kinase-like) domains, and through their direct association with cytoplasmic domains of integral membrane proteins, they are thought to be directly involved in clustering of integral membrane proteins to create specialized membrane domains (4 -7).
Simple epithelial cells contain three specialized membrane domains at the most apical part of lateral membranes for intercellular adhesion, tight junctions (TJs), 1 adherens junctions (AJs) and desmosomes (8). In these domains, occludin/ claudin (9 -11), cadherin (12)(13)(14), and desmoglein/desmocollin (15,16) were identified as major integral membrane proteins (adhesion molecules), respectively, but our knowledge regarding how these integral membrane proteins are sorted into three distinct junctional membrane domains is still fragmentary. To date, three MAGUKs have been shown to be associated with these intercellular junctions, and these molecules are now called ZO-1, ZO-2, and ZO-3.
ZO-1 was first identified as a peripheral membrane protein with a molecular mass of 220 kDa and was concentrated at TJs in epithelial cells (17,18). However, in nonepithelial cells lacking TJs, such as cardiac muscle cells and fibroblasts, it was precisely colocalized with cadherins (19,20). ZO-1 molecule is roughly divided into two functional portions: the NH 2 -terminal half, which shows similarity to Drosophila lethal (1) discs large-1 (dlg) consisting of three PDZ, one SH3, and one GUK domain; and the COOH-terminal half with no sequence similarity to dlg (20 -22). Consistent with the subcellular distribution of ZO-1 in epithelial and nonepithelial cells, its NH 2terminal dlg-like half bound directly to the cytoplasmic domain of occludin (23,24) as well as ␣ catenin (24) that associates with the cytoplasmic domain of cadherin via ␤ catenin (25)(26)(27)(28)(29)(30)(31)(32). However, how ZO-1 is excluded out from AJs in epithelial cells where ␣ catenin is highly concentrated remains unclear. In addition, the COOH-terminal non-dlg-like half of ZO-1 was shown to be directly associated with actin filaments in vitro as well as in vivo (24,44).
As compared with ZO-1, our knowledge of ZO-2 is still limited. ZO-2 with a molecular mass of 160 kDa was first identified as a ZO-1-binding protein by immunoprecipitation with anti-ZO-1 mAb (33). Cloning and sequencing of dog ZO-2 cDNA revealed that it also contained a dlg-like domain containing three PDZ, one SH3, and one GUK domain at its NH 2 -terminal region, followed by a short COOH-terminal non-dlg-like domain (34,35). In contrast to ZO-1, ZO-2 was reported to be absent from intercalated discs (N-cadherin-based AJs) of cardiac muscle cells and exclusively concentrated at TJs in epithelial cells using dog ZO-2-specific pAb (34). ZO-3 was also identified in ZO-1 immunoprecipitate as a phosphorylated 130-kDa peptide (36), and its cDNA was recently cloned (37).
At the initial phase of junction formation of epithelial cells, ZO-1 was precisely colocalized with cadherins in primordial spot-like AJs, where TJs were not yet assembled, and then at the later stage, ZO-1 was transferred from AJs to TJs (38). Although these findings suggest the direct involvement of ZO-1 in the establishment of two distinct membrane domains, AJs and TJs, in epithelial cells, lack of information concerning ZO-2 has hampered more direct assessment of the junction sorting mechanism at the molecular level. In this study, we isolated a mouse ZO-2 cDNA and characterized its product in detail.

EXPERIMENTAL PROCEDURES
Cloning of Mouse ZO-2 cDNA-Based on the human cDNA clone X104, which was later recognized as human ZO-2 cDNA (39), a partial human ZO-2 cDNA fragment (2897-3241) was obtained by RT-PCR using mRNA from human T84 cells. This fragment was used as a probe to screen a mouse lung ZAP cDNA library. Eleven positive clones were isolated, one of which, clone 10, contained the entire open reading frame of mouse ZO-2.
Constructs and Transfection-For expression of NH 2 -terminally HAtagged proteins in mammalian cells, an oligonucleotide encoding an HA epitope was subcloned into the eukaryotic expression vector pME18S producing pME18S-HA. The cDNA fragment containing the entire open reading frame of mouse ZO-2 was produced by PCR. The amplified product was digested with StuI and SalI, then subcloned into EcoRV/ SalI-cleaved pME18S-HA. The cDNA fragments encoding the NH 2terminal dlg-like portion (1-938), COOH-terminal non-dlg-like portion (939 -1167) (see Fig. 1), and other deletion mutants of the NH 2 -terminal dlg-like portion of mouse ZO-2 (see Fig. 6) were generated by PCR using appropriate primers and subcloned into pME18S-HA. MDCKII and 3Y1 cells were transfected with each expression vector and selected as described previously (24).
Immunofluorescence Microscopy-Cells plated on glass coverslips were rinsed in PBS and fixed with 1% formaldehyde in PBS for 15 min at room temperature. The fixed cells were processed for immunofluorescence microscopy as described previously (24).
Production of the NH 2 -and COOH-terminal Portions of ZO-2 by Recombinant Baculovirus Infection-The cDNA fragment encoding the NH 2 -terminal portion (1-938) or COOH-terminal portion of ZO-2 (939 -1167) was generated by PCR as described above. These fragments were subcloned into the pBlueBac vector (Invitrogen) and then integrated into the baculovirus genome. The recombinant virus carrying each cDNA was isolated and condensed using a MAXBAC kit (Invitrogen). Insect Sf9 cells were infected with recombinant viruses, and the total cell lysate was prepared as described previously (24).
In Vitro Binding Assay Using GST Fusion Proteins-GST fusion proteins with full-length ␣ catenin (GST-␣ catenin) or the cytoplasmic domain of occludin (GST-Oc358) were prepared as described previously (24). Fusion proteins were expressed in E. coli and purified using glutathione-Sepharose 4B beads (Pharmacia LKB Biotechnology, Uppsala, Sweden). Aliquots of 200 l of glutathione-Sepharose beads slurry containing GST fusion proteins were washed with 100 volumes of PBS containing 0.1% Triton X-100, 2 mM phenylmethylsulfonyl fluoride, and 4 g/ml leupeptin by brief centrifugation, and then 2 ml of the lysate of Sf9 cells expressing N-ZO-1 or N-ZO-2 was added, followed by incubation for 3 h at 4°C. The beads were then washed with 40 volumes of the same solution, and bound proteins were eluted with 1 ml of 50 mM Tris-HCl buffer (pH 8.0) containing 10 mM glutathione. The amounts of GST fusion proteins in each eluate were determined by SDS-PAGE. An appropriate amount of each eluate was again applied to SDS-PAGE to contain the same amount of GST fusion proteins.
For biotinylation of the Sf9 cell lysate, Sulfo-NHS-biotin (Pierce) was added to the lysate at a final concentration of 0.5 mg/ml and incubated for 10 min at room temperature. The reaction was stopped by adding Tris-HCl (pH 8.0) up to 50 mM. To detect biotinylated proteins bound to GST-␣ catenin, each eluate was separated by SDS-PAGE, transferred onto nitrocellulose membranes, followed by incubation with alkaline phosphatase-conjugated streptavidin.
Immunoprecipitation-Confluent monolayers of cultured cells on 9-cm dishes were washed three times with ice-cold PBS, and then cells were lysed in 2 ml of extraction buffer (PBS containing 0.5% Nonidet P-40, 2 mM phenylmethylsulfonyl fluoride, 2 mg/ml leupeptin). Cell lysates were clarified by centrifugation at 100,000 ϫ g for 30 min and incubated with 100 l of protein G-Sepharose bead slurry (Zymed Laboratories Inc., San Francisco, CA) coupled with mouse anti-HA mAb for 3 h at 4°C. As a control, nonspecific mouse IgG was coupled to the beads. After five washes with the extraction buffer, immunoprecipitates were eluted with SDS-PAGE sample buffer. For metabolic labeling of transfectants, cells were washed once with methionine-free medium supplemented with 2% fetal calf serum and then incubated with 3 ml of the same medium containing 0.2 mCi [ 35 S]methionine (Amersham Pharmacia Biotech, Buckinghamshire, UK) for 3 h before lysis.
Gel Electrophoresis and Immunoblotting-One-dimensional SDS-PAGE (10 -12.5% gel) was performed based on the method of Laemmli (41), and immunoblotting was performed as described previously (21).

Isolation of Mouse ZO-2 cDNA and Generation of Anti-ZO-2
pAb-When we began this study, only a partial dog ZO-2 cDNA had been isolated (34). Similarity searches in data bases identified a human cDNA (X104) which showed marked similarity to dog ZO-2 cDNA (39). Based on this sequence, we isolated a part of human ZO-2 cDNA by PCR using the first strand cDNA generated from total RNA of cultured human epithelial cells (T84). Using this cDNA fragment as a probe, we screened a ZAP cDNA library of mouse lung, and obtained a full-length cDNA encoding mouse ZO-2 (data are available from GenBank/ EBI/DDBJ under accession number AF113005). Its open reading frame encoded a protein of 1,167 amino acids with a calculated molecular mass of 132 kDa. During the course of this study, however, a full-length cDNA encoding dog ZO-2 was reported (35). Mouse ZO-2 was 85% identical to dog ZO-2 at the amino acid sequence level, and consisted of three PDZ, one SH3, and one GUK domain from the NH 2 -terminal end (Fig. 1).
Based on the deduced amino acid sequence of mouse ZO-2, we synthesized two polypeptides corresponding to the middle and COOH-terminal end portions of ZO-2 that showed no sequence similarity to ZO-1 and raised pAbs (pAb59 and pAb62, respectively) in rabbits using them as antigens. Because both pAbs showed the same properties on immunoblotting as well as immunofluorescence microscopy, we will report here only the data obtained with pAb59. As shown in Fig. 2a, the affinitypurified pAb59 specifically recognized recombinant ZO-2 but not recombinant ZO-1 produced in Sf9 cells by baculovirus infection.
Subcellular Distribution of ZO-2 in Comparison with ZO-1 in Cultured Cells and Tissues-Using the affinity-purified anti-ZO-2 pAb, we first examined the expression levels of ZO-2 in various cultured cells by immunoblotting and compared them with those of ZO-1 (Fig. 2b). As previously reported, ZO-1 was detected in all the cell types examined including epithelial, fibroblastic, and myeloma cells. ZO-2 was also detected in cultured epithelial cells as well as in fibroblasts such as 3Y1, NIH 3T3, and Swiss 3T3 cells, but not in L fibroblasts or P3 myeloma cells.
We next examined the subcellular localization of ZO-2 in cultured MDCK cells and 3Y1 cells using affinity-purified anti-ZO-2 pAb. In MDCK cells, as expected from previous studies, ZO-2 was precisely colocalized with ZO-1 in a linear pattern at cell-cell borders, and computer-generated cross-sectional views confirmed that ZO-2 was colocalized with occludin at TJs and concentrated more apically than E-cadherin (data not shown). Interestingly, ZO-2 was also precisely colocalized with ZO-1 in a punctate or serrated pattern at cell-cell borders of 3Y1 fibroblasts lacking TJs (Fig. 3, a and b). This characteristic distribution of ZO-2 was identical to P-cadherin (Fig. 3, c and d). This finding prompted us to re-examine the expression and subcellular distribution of ZO-2 in cardiac muscle cells. Frozen sections were doubly stained with anti-ZO-1 mAb and affinitypurified anti-ZO-2 pAb, and intense signals of ZO-2 as well as ZO-1 were detected from intercalated discs (Fig. 3, e and f). This ZO-2 signal was specific because it was not observed on staining with pAb preabsorbed with GST-ZO-2 fusion protein (data not shown).
Characterization of Dlg-like and Non-dlg-like Domains of ZO-2 in Vivo and in Vitro-Previously, we divided ZO-1 into the NH 2 -terminal half dlg-like (N-ZO-1) and COOH-terminal half non-dlg-like (C-ZO-1) portions and characterized them both in vivo and in vitro (24). In this study, the full-length ZO-2 (F-ZO-2) was also divided into the NH 2 -terminal dlg-like (N-ZO-2) and COOH-terminal non-dlg-like (C-ZO-2) domains (see Fig. 1). First, F-ZO-2, N-ZO-2, and C-ZO-2 were tagged with HA peptide at their NH 2 ends and introduced into cultured MDCK cells as well as 3Y1 cells (Fig. 4). Under transient expression conditions, immunofluorescence microscopy with anti-HA mAb revealed that in both MDCK and 3Y1 cells, F-ZO-2 as well as N-ZO-2 were recruited to the cell-cell borders where endogenous ZO-2 was concentrated, i.e. TJs in MDCK cells and P-cadherin-based spot AJs in 3Y1 cells (Fig. 4, a-d).
In marked contrast, C-ZO-2 appeared to be colocalized with actin filaments. In 3Y1 fibroblasts, C-ZO-2 was clearly distributed along stress fibers (Fig. 4f), although in MDCK cells it was distributed diffusely with some concentration at cell-cell borders (along circumferential actin bundles) and plasma membranes (along microvilli) (Fig. 4e).
We have previously reported that N-ZO-1 directly binds to the cytoplasmic domain of occludin as well as ␣ catenin in vitro (24). The binding of N-ZO-2 to occludin and/or ␣ catenin was compared with that of N-ZO-1 (Fig. 5). N-ZO-1 and N-ZO-2 were produced in Sf9 cells by baculovirus infection, and the cell lysate of Sf9 cells containing almost the same amounts of N-ZO-1 or N-ZO-2 was incubated with a GST fusion protein with the cytoplasmic domain of occludin (Fig. 5a). CBB staining of the eluates from the GST/occludin fusion protein beads revealed that the cytoplasmic domain of occludin directly bound to N-ZO-2 as well as N-ZO-1. Next, we examined the binding of N-ZO-2 to ␣ catenin, but in this case the electrophoretic mobilities of recombinant N-ZO-2 (and also that of N-ZO-1) were almost the same as that of GST-␣ catenin fusion protein, making it difficult to estimate the amount of bound N-ZO-2 (and of bound N-ZO-1) by CBB staining (Fig. 5b). Thus, the total proteins of Sf9 cells containing almost the same amounts of N-ZO-1 or N-ZO-2 were biotinylated, then the in vitro binding assay with GST-␣ catenin fusion protein was performed. The amounts of bound N-ZO-1 and N-ZO-2 were estimated by detection with alkaline phosphatase-avidin. As shown in Fig. 5b, GST-␣ catenin fusion proteins bound directly to N-ZO-2 as well as N-ZO-1.
Interaction of ZO-2 with ZO-1-Because ZO-2 was first identified in the ZO-1 immunoprecipitate (33), ZO-1 is thought to be directly associated with ZO-2. Recently, the ZO-2 binding domain on ZO-1 was narrowed down to its PDZ2 domain (44). We attempted to identify the ZO-1 binding domain on ZO-2. First, we examined the in vitro binding of GST fusion proteins with N-ZO-1 or C-ZO-1 with recombinant N-ZO-2 or C-ZO-2 produced in Sf9 cells, but we detected no significant binding. These findings suggested that some modifications on ZO-1 or ZO-2 molecules within cells are required for the ZO-1/ZO-2 interaction. Therefore, we introduced HA-tagged F-ZO-2 cDNA into EL cells (L cells transfected with E-cadherin) expressing myctagged N-ZO-1 (NZ-EL cells) or myc-tagged C-ZO-1 (CZ-EL cells), and F-ZO-2 in the total cell lysate was immunoprecipi- tated with anti-HA mAb. As shown in Fig. 6a by immunoblotting with anti-myc mAb, N-ZO-1 but not C-ZO-1 was co-immunoprecipitated with F-ZO-2, indicating that the NH 2 -terminal dlg-like domain of ZO-1 bound to F-ZO-2. When HA-tagged C-ZO-2 cDNA was introduced into NZ-EL cells or CZ-EL cells, neither N-ZO-1 or C-ZO-1 was co-immunoprecipitated with C-ZO-2, indicating that the NH 2 -terminal dlg-like domain of ZO-2 is responsible for ZO-1 binding (data not shown). Then, to further narrow down the domain responsible for the ZO-1 bind-ing, various deletion constructs of N-ZO-2 were introduced into NZ-EL cells (Fig. 6b). Each construct was tagged with HA at its NH 2 -end. Stable transfectants were metabolically labeled with [ 35 S]methionine and solubilized, and then mutant ZO-2 produced from these constructs was immunoprecipitated with anti-HA mAb. As shown in Fig. 6b, N-ZO-1 was co-immunoprecipitated only when the introduced N-ZO-2 constructs contained the PDZ2 domain, indicating that the PDZ2 domain of ZO-2 is responsible for ZO-1/ZO-2 interaction. DISCUSSION ZO-1 is expressed not only in epithelial/endothelial cells but also in nonepithelial/endothelial cells such as cardiac muscle cells, fibroblasts, and astrocytes (19,20,42) and that, in these nonepithelial/endothelial cells, ZO-1 is precisely colocalized with cadherins (19,20). In contrast to the ubiquitous expression and the peculiar subcellular distribution of ZO-1, ZO-2, another MAGUK family member, was reported to be specific for TJs, and to be absent from AJs in cardiac muscle cells (34). However, in the present study using anti-ZO-2 pAb, we found that ZO-2 was precisely co-concentrated with ZO-1 at intercalated discs (AJs). At present, the reason for this discrepancy remains unclear, but the following observations favored the notion that ZO-2 is very similar to ZO-1 in terms of AJ-association in nonepithelial/endothelial cells. First, a human cDNA called X104, which was later recognized to be human ZO-2 cDNA, was ubiquitously detected in various tissues and was abundant in heart (39). Second, our anti-ZO-2 pAb also detected concentration of ZO-2 at the P-cadherin-based spot-like AJs in cultured fibroblasts, which had not been examined in the previous report (34). Third, when HA-tagged F-ZO-2 and N-ZO-2 were introduced into cultured fibroblasts, it was correctly targeted to the P-cadherin-based spot-like AJs.
Our previous study suggested that ZO-1 functions as a crosslinker between occludin and actin filaments in epithelial/endothelial cells or between ␣ catenin and actin filaments in nonepithelial/endothelial cells (24). The present study revealed that the cytoplasmic domain of occludin and ␣ catenin also bound to N-ZO-2 in vitro. Exogenously expressed C-ZO-2 was distributed along stress fibers in cultured fibroblasts similarly to exogenously expressed C-ZO-1. From these observations, we concluded that ZO-2 is very similar to ZO-1 also as a crosslinker. At present, it remains unknown why two similar crosslinkers, ZO-1 and ZO-2, exist in intercellular junctions such as AJs and TJs. Furthermore, the recent knockout study of occludin revealed that ZO-1 is still exclusively concentrated at occludin-deficient TJs (43). Preliminary experiments revealed that ZO-2 also remained at occludin-deficient TJs (data not shown). These findings indicate that ZO-1 as well as ZO-2 are recruited to normal TJs through direct or indirect interactions not only with occludin but also with other TJ-specific membrane proteins such as the recently identified claudins (10).
The possible interaction of ZO-1 and ZO-2 would make the relationship of these two similar cross-linker proteins more complex. Recent immunoprecipitation analyses indicated that the PDZ2 domain of ZO-1 was responsible for ZO-1/ZO-2 interaction although it was not clear whether this interaction was direct or indirect (44). L cells and their transfectants gave a good model in which to examine the ZO-1/ZO-2 interaction because they lack endogenous expression of ZO-2. In the present study, immunoprecipitation experiments using metabolically labeled cells demonstrated that N-ZO-1 binds to the PDZ2 domain of ZO-2. Of course, although the possibility cannot be completely excluded that a third protein mediates this binding, these findings favored the notion that ZO-1 and ZO-2 directly form a heterodimer (or oligomer) through PDZ2/PDZ2 interaction. This type of PDZ/PDZ interaction has been reported be-FIG. 6. ZO-1/ZO-2 interaction. a, association of F-ZO-2 with N-ZO-1. F-ZO-2 (with HA-tag) cDNA was introduced into NZ-EL (NZ-EL) or CZ-EL (CZ-EL) cells that expressed E-cadherin and N-ZO-1 or C-ZO-1 (with myc-tag), respectively. When F-ZO-2 was immunoprecipitated with anti-HA mAb, N-ZO-1 (N-ZO-1) but not C-ZO-1 was coimmunoprecipitated with F-ZO-2 (arrowhead), which was confirmed by immunoblotting with anti-myc pAb. Arrows, IgG. b, domain of ZO-2 responsible for ZO-1 interaction. Various HA-tagged deletion N-ZO-2 constructs (as schematically shown above the panel) were introduced into NZ-EL cells. After stable transfectants were metabolically labeled with [ 35 S] methionine, each ZO-2 construct (1-7) was immunoprecipitated with anti-HA mAb and examined by autoradiography. N-ZO-1 (arrowheads) was co-immunoprecipitated only with bands 2, 5 and 6 which contained the PDZ2 domain. tween neuronal nitric oxide synthase and PSD-95 (or PSD-93) (45).
In this study, two intercellular-associated MAGUK family members, ZO-1 and ZO-2, were compared in detail. The elucidation of the molecular mechanism behind the peculiar behavior of ZO-1 and ZO-2, i.e. their respective localization at TJs and AJs in epithelial/endothelial and nonepithelial/endothelial cells, is necessary to better understand the molecular mechanism not only underlying the establishment of intercellular junction domains but also that behind the polarization of epithelial/endothelial cells. Because information concerning the molecular components of intercellular junctions is rapidly accumulating, further analyses of ZO-1 and ZO-2 including knocking-out their genes will lead to a better understanding of the physiological functions of MAGUK family members in general.