Characterization of the Interaction between Protein 4.1R and ZO-2

Multiple isoforms of the red cell protein 4.1R are expressed in nonerythroid cells, including novel 135-kDa isoforms. Using a yeast two-hybrid system, immunocolocalization, immunoprecipitation, and in vitro binding studies, we found that two 4.1R isoforms of 135 and 150 kDa specifically interact with the protein ZO-2 (zonulaoccludens-2). 4.1R is colocalized with ZO-2 and occludin at Madin-Darby canine kidney (MDCK) cell tight junctions. Both isoforms of 4.1R coprecipitated with proteins that organize tight junctions such as ZO-2, ZO-1, and occludin. Western blot analysis also revealed the presence of actin and α-spectrin in these immunoprecipitates. Association of 4.1R isoforms with these tight junction and cytoskeletal proteins was found to be specific for the tight junction and was not seen in nonconfluent MDCK cells. The amino acid residues that sustain the interaction between 4.1R and ZO-2 reside within the amino acids encoded by exons 19–21 of 4.1R and residues 1054–1118 of ZO-2. Exogenously expressed 4.1R containing the spectrin/actin- and ZO-2-binding domains was recruited to tight junctions in confluent MDCK cells. Taken together, our results suggest that 4.1R might play an important role in organization and function of the tight junction by establishing a link between the tight junction and the actin cytoskeleton.

Multiple isoforms of the red cell protein 4.1R are expressed in nonerythroid cells, including novel 135-kDa isoforms. Using a yeast two-hybrid system, immunocolocalization, immunoprecipitation, and in vitro binding studies, we found that two 4.1R isoforms of 135 and 150 kDa specifically interact with the protein ZO-2 (zonula occludens-2). 4.1R is colocalized with ZO-2 and occludin at Madin-Darby canine kidney (MDCK) cell tight junctions. Both isoforms of 4.1R coprecipitated with proteins that organize tight junctions such as ZO-2, ZO-1, and occludin. Western blot analysis also revealed the presence of actin and ␣-spectrin in these immunoprecipitates. Association of 4.1R isoforms with these tight junction and cytoskeletal proteins was found to be specific for the tight junction and was not seen in nonconfluent MDCK cells. The amino acid residues that sustain the interaction between 4.1R and ZO-2 reside within the amino acids encoded by exons 19 -21 of 4.1R and residues 1054 -1118 of ZO-2. Exogenously expressed 4.1R containing the spectrin/actin-and ZO-2-binding domains was recruited to tight junctions in confluent MDCK cells. Taken together, our results suggest that 4.1R might play an important role in organization and function of the tight junction by establishing a link between the tight junction and the actin cytoskeleton.
Erythrocyte protein 4.1R is an 80-kDa cytoskeletal protein critical in circulating red cells for the dynamic organization and maintenance of the spectrin/actin cytoskeleton and for the attachment of the cytoskeleton to the cell membrane through interactions with integral membrane proteins such as glycophorin C and band 3 (1). In nonerythroid cells, an additional and prevalent 135-kDa 4.1R isoform class has been detected. It contains a 209-amino acid extension at its N terminus end (2).
Multiple isoforms of 4.1R exist. They arise through complex alternative pre-mRNA splicing pathways (2,3), post-translational modifications (4), and use of at least two translation initiation sites (5,6). In nucleated erythroid and nonerythroid cells, a spectrum of proteins ranging from 30 to 210 kDa has been detected that contains epitopes for 4.1R (7,8). Unlike the strict peripheral distribution of 4.1R in mature erythrocytes, 4.1R isoforms in nonerythroid cells are localized in the nucleus and nuclear matrix (9 -13), at points of cell-cell or cell-matrix contact (14,15), and in centrosomal and Golgi structures (16,17). In addition, 4.1R may also interact with microtubules and stress fibers (14,18). However, the precise identity of 4.1R isoforms in these subcellular structures, their binding partners, and their biological significance in nonerythroid cells are not well understood.
Like 4.1R, many members of the protein 4.1 superfamily, including ezrin, radixin, moesin, merlin, and myosin, are known to associate with the plasma membrane (reviewed in Ref. 31). A Drosophila homolog of protein 4.1, Coracle, is localized at septate junctions in ectodermally derived epithelial cells (32). Drosophila Neurexin and DLG (Discs Large), homologs of proteins that are known to bind to 4.1, glycophorin C, and human DLG, respectively, are also localized at the septate junctions (33,34). Collectively, these studies suggest the localization of 4.1 to the septate junction, an invertebrate specific junction with molecular components analogous to the vertebrate tight junction (TJ) (35). In addition to membrane binding activity, members of this family participate in important cell signaling events. For example, merlin is involved in growth regulation (36), and ezrin, radixin, and moesin are involved in Rho-dependent signaling pathways (37,38).
To explore the function of 4.1R isoform(s) in nonerythroid cells, we searched for their binding partners. In this study, using a combination of molecular/genetic, biochemical, and immunofluorescence studies, we establish that two nonerythroid 4.1R isoforms bind to the C-terminal proline-rich domain of X-104, a human homolog of the canine protein ZO-2 (henceforth referred to as hZO-2), through its C-terminal domain. ZO-2 belongs to the membrane-associated guanylate kinase (MAGUK) family of proteins and is known to organize the TJ in association with other TJ-associated proteins (39 -41). We report here that these isoforms of 4.1R colocalize with ZO-2 and occludin and also form an intracellular complex with ZO-2, ZO-1, occludin, actin, and ␣-spectrin at MDCK cell TJs. Our results suggest that in nonerythroid cells, 4.1R may have a role in the organization and function of the TJ by establishing a direct link between the TJ and the actin cytoskeleton.
For domain mapping, plasmids carrying inserts fused to Gal4-BD or Gal4-AD were cotransformed into Y190 and assayed for ␤-galactosidase activity on nitrocellulose filters. In some cases, liquid ␤-galactosidase assays using o-nitrophenyl ␤-D-galactopyranoside or chlorophenol red-␤-D-galactopyranoside were used as described in the CLONTECH manual.
Cell Culture and Transient Transfection-MDCK cells (American Type Culture Collection CRL 1772) were used in this study because they are known to express TJ proteins and to form TJs at confluent density (47). We have described the culture conditions for MDCK cells and have documented the expression of 4.1R in this cell line at both the protein and RNA levels (13). Transfection was performed using Lipo-fectAMINE reagent (Life Technologies, Inc.) according to the manufacturer's protocol. Cells were plated on poly-D-lysine-coated coverslips in a six-well tissue culture plate 1 day prior to transfection. Cells at 50 -70% confluency were transfected with 2 g of plasmid DNA. After 36 -48 h, cells on coverslips were removed, fixed, and processed for immunofluorescence staining with anti-ZO-1 Ab as described below.
Rabbit preimmune serum equivalent to 50 g of IgG was added to 2 mg of proteins and incubated for 2 h at 4°C, followed by the addition of 200 l of a 50% suspension of protein A-Sepharose CL-4B (Amersham Pharmacia Biotech) in immunoprecipitation buffer for 2 h at 4°C. The supernatant was collected by centrifugation at 3000 ϫ g for 5 min at 4°C and split equally into seven tubes. Preimmune serum, affinitypurified anti-HP Ab, anti-ZO-1 Ab, anti-ZO-2 Ab, anti-occludin Ab, rabbit IgG, or anti-p53 Ab (an irrelevant antibody as a control) containing 6 g of IgG was added to different tubes and incubated for 2 h at 4°C. Protein A-Sepharose CL-4B (100 l of a 50% suspension) was then added and incubated overnight at 4°C on a rocking platform. Following incubation with protein A-Sepharose, the samples were centrifuged at 3000 ϫ g for 10 s at 4°C and washed 10 times with 1 ml of immunoprecipitation buffer. The samples were resuspended in 80 l of SDS sample buffer (62.5 mM Tris-HCl, 10% glycerol, 2% (w/v) SDS, 5% 2-mercaptoethanol, and 10 g/ml bromphenol blue, pH 6.8) and boiled in a water bath for 5 min. The samples were centrifuged at 10,000 ϫ g for 15 min at 4°C, and the supernatants were fractionated on 6 -12% SDS-polyacrylamide gels.
Transfer of proteins to nitrocellulose or polyvinylidene difluoride membranes and detection of 4.1R, ZO-1, ZO-2, occludin, actin, and spectrin were carried out by immunoblotting using an ECL detection kit (Amersham Pharmacia Biotech) as described earlier (13). Quantitation of proteins from chemiluminograms was done with NIH Image software for the Apple Macintosh computer. To visualize proteins, the gels were stained with GelCode ® SilverSNAP TM from Pierce.
In Vitro Binding Assay-Recombinant 4.1R/GST proteins were expressed and affinity-purified by coupling to glutathione-Sepharose beads as described (13). The recombinant proteins were quantitated by Coomassie Brilliant Blue staining using bovine serum albumin as a standard. Desired [ 35 S]methionine-labeled segments of hZO-2, hZO-1, and human occludin were in vitro translated using a TNT ® SP6 Quick Coupled transcription/translation system (reticulocyte lysate system; Promega) according to the manufacturer's recommendations. The translation products were quantitated based on percent of methionine incorporation and 35 S-specific activity. For in vitro binding, equal amounts of labeled proteins were incubated with 20 g of 4.1R/GST fusion proteins coupled to glutathione-Sepharose beads for 2 h at 4°C in 300 l of binding buffer (50 mM Tris, pH 7.4, 50 mM NaCl, 10 mM sodium pyrophosphate, 1.5 mM sodium orthovanadate, 4 g/ml aprotinin, 4 g/ml leupeptin, 4 g/ml antipain, 12.5 g/ml chymostatin, 12 g/ml pepstatin, 130 g/ml ⑀-aminocaproic acid, 200 g/ml paminobenzamidine, and 1 mM phenylmethylsulfonyl fluoride). After incubation, the beads were washed 10 times with 1 ml of binding buffer containing 0.5% Triton X-100 each time. The beads were then resuspended in 30 l of SDS sample buffer, boiled for 5 min, and analyzed by SDS-polyacrylamide gel electrophoresis (12% gel; National Diagnostics, Inc., Atlanta). The gels were treated with Enlightning TM (NEN Life Sciences Products), and binding of 35 S-labeled proteins was detected by fluorography.
To determine the dissociation constants between 4.1R and ZO-2, ZO-1, or occludin, 5 g of 24-kDa/GST or GST (as a control) coupled to glutathione-Sepharose beads was incubated with 0.0025-0.015 g of 35 S-labeled translated products as described above. After extensive washing, the bound proteins were eluted with 50 mM Tris-HCl containing 20 mM glutathione, pH 8.0. The amount of bound and free labeled protein was determined from 35 S-specific activity and adjusted for binding to the GST-conjugated Sepharose beads. A Scatchard plot of the data was generated. Each experiment was repeated three times for estimation of K d .
Fluorescence and Confocal Microscopy-MDCK cells were grown on poly-D-lysine-coated coverslips to confluence. The coverslips were washed three times with PBS and fixed in freshly prepared 2% paraformaldehyde in PBS for 25 min at room temperature. Cells were washed with PBS and permeabilized with a solution of 0.2% Triton X-100 and 10% normal goat serum in PBS for 15 min. The cells were blocked with 10% normal goat serum and 50 mM NH 4 Cl for 1.5 h at room temperature and washed three times with PBS containing 0.3% bovine serum albumin. Cells were incubated with primary antibodies (1:100 dilution of anti-occludin Ab, 1:100 dilution of anti-ZO-2 Ab, 1:200 dilution of anti-ZO-1 Ab, 1:50 dilution of anti-10-kDa Ab, 1:50 dilution of anti-16-kDa Ab, and 1:25 dilution of anti-HP Ab) for 1 h at room temperature and washed three times as described above. They were then incubated with a 1:100 dilution of Texas Red-conjugated goat anti-rabbit IgG (Jackson ImmunoResearch Laboratories, Inc.) or a 1:50 dilution of fluorescein isothiocyanate-conjugated goat anti-rabbit IgG (Jackson ImmunoResearch Laboratories, Inc.) at room temperature for 1 h and washed three times again as described above. When double staining was desired, after the first set of staining, cells were blocked with unconjugated goat anti-rabbit Fab fragment (10 g/ml in PBS; Jackson ImmunoResearch Laboratories, Inc.) for 1 h at room temperature and washed three times as described above. Incubation with the second set of antibodies was the same as with the first set of antibodies. Mounting of the coverslips and processing of images using a Noran confocal laser scanning image system and Intervision software (Noran Instruments Inc., Middleton, WI) were the same as described previously (13).
Immunoelectron Microscopy-Confluent MDCK cells were washed with PBS and fixed in 4% paraformaldehyde and 0.1% glutaraldehyde in PBS overnight at 4°C. The cells were then infiltrated with 15% polyvinylpyrrolidone and 1.95 M sucrose in PBS (48), and ultrathin cryosections were cut at Ϫ100°C with a Reichert FCS cryoultramicrotome (Leica Inc.). Cryosections were stretched using 2.3 M sucrose and mounted on Formvar/carbon-coated grids. Sections were blocked with 10% normal goat serum and 20 mM glycine in PBS for 30 min. Sections were sequentially incubated for 60 min with anti-ZO-2 (1:20 dilution) or anti-HP (1:5 dilution) antibody and AuroProbe EM goat anti-rabbit IgG (1:50 dilution; Amersham Pharmacia Biotech). The sections were postfixed in 2% glutaraldehyde in PBS. The immunogold-labeled sections were examined in an electron microscope (EM 10, Carl Zeiss, Inc.).
The X-104 (hZO-2) gene was identified in relation to the FIG. 1. Schematic representation of 4.1R and X-104 (hZO-2) peptides found to interact in the yeast twohybrid screening of a human brain cDNA library. A, the organization of protein 4.1R (P 4.1R) and its 22-24-kDa domain was used as the bait in the yeast two-hybrid assays. Different exons of the 4.1R gene and corresponding chymotryptic domains are shown. The asterisks represent the alternatively spliced exons. B, shown is a schematic representation of the structural organization of the hZO-2 protein and its segments encompassed by the positive clones obtained from the yeast two-hybrid screen. GUK, guanylate kinase; ϩ, basic domain; Ϫ, acidic domain; ␣, alternatively spliced.
Colocalization of 4.1R and ZO-2 in MDCK Cell Tight Junctions-To verify the interaction of 4.1R with ZO-2 and to locate the site(s) of their intracellular interaction, we examined the subcellular distribution of 4.1R and ZO-2 in confluent MDCK cells by double-label immunofluorescence microscopy. As shown in Fig. 2 (A and B), ZO-2 (red) and 4.1R (green) localized at the cell-cell contacts and displayed honeycomb-like staining patterns. Both ZO-2 and 4.1R also showed some diffuse cyto- Occludin has been shown to exclusively localize at TJs of epithelial and endothelial cells (52,53) in association with ZO-2 and ZO-1. To verify the localization of 4.1R at tight junctions, we performed double immunofluorescent staining of confluent MDCK cells with 4.1R and occludin. As shown in Fig. 2, occludin (red; panel D) and 4.1R (green; panel E) localized at apical cell borders and, to some extent, were diffuse in the cytoplasm. A transverse view of the cells taken at the position of the dotted line (in Fig. 2, D and E) showed that both occludin (Fig. 2DЈ) and 4.1R (Fig. 2EЈ) localized at the TJ along the apical side of the cells. As shown in Fig. 2F, 4.1R precisely colocalized with occludin at the MDCK cell TJs. Colocalization of 4.1R with ZO-2 and occludin at MDCK cell TJs was also observed when cells were stained with anti-10-kDa Ab or anti-16-kDa Ab (data not shown). The colocalization of 4.1R with occludin and ZO-2 indicated that 4.1R localizes specifically to TJs of confluent MDCK cells.
To study the subcellular localization of ZO-2 and 4.1R in more detail, we performed high resolution immunogold electron microscopy using fixed confluent MDCK cells. Anti-ZO-2 Ab was used as a control. As shown in Fig. 2 (G and H), labeling for ZO-2 (10-nm gold particles) and 4.1R (5-nm gold particles) was found concentrated at the tight junctions as clusters (arrows). This further supports our contention that 4.1R localizes at tight junctions.
In Vivo Association of 4.1R and ZO-2-To examine the interaction of 4.1R with ZO-2 in vivo, we performed immunoprecipitation using confluent MDCK cell extracts. The extracts were subjected to immunoprecipitation with anti-HP Ab or anti-ZO-2 Ab. Preimmune serum (rabbit), purified rabbit IgG, and anti-p53 Ab (an irrelevant Ab) were used as controls. Analysis of immunoprecipitates by immunoblotting using anti-HP Ab showed that anti-ZO-2 Ab coprecipitated two polypeptides of Х135 and Х150 kDa; these comigrated with two polypeptides of similar molecular mass that were immunoprecipitated by anti-HP Ab (Fig. 3A). Similarly, analysis of the immunoprecipitates by immunoblotting using anti-ZO-2 Ab showed that ZO-2 coprecipitated with 4.1R (Fig. 3B). Neither 4.1R nor ZO-2 was coprecipitated with preimmune serum, rabbit IgG, or anti-p53 Ab (Fig. 3, A and B). These results suggest that 4.1R and ZO-2 are associated together in vivo. Analysis of the same immunoprecipitates by immunoblotting using anti-24-kDa Ab also revealed the presence of Х135and Х150-kDa polypeptides (Fig. 3C), confirming that these polypeptides are 4.1R isoforms. However, these isoforms of 4.1R were not detected in anti-HP Ab or anti-ZO-2 Ab supernatants by anti-24-kDa Ab, but were detected by anti-HP Ab. Because anti-24-kDa Ab is raised against exon 19, these data suggest that most of the 135-and 150-kDa 4.1R isoforms that contain alternative exon 19 coprecipitate with ZO-2 and therefore are not detected in the supernatant, whereas those isoforms not containing exon 19 do not coprecipitate with ZO-2 and thus are detected by anti-HP Ab. This is consistent with the results of yeast twohybrid mapping data that exons 19 -21 of 4.1R are required for interaction with ZO-2 (Fig. 4B).
We determined the efficiencies of immunoprecipitation and coprecipitation to discern the fractions of 4.1R and ZO-2 that associate together. Quantitation from three different experiments (representative gels are shown in Fig. 3, A and B) showed that 80 -90% of 4.1R was precipitated by anti-HP Ab, which in turn coprecipitated 35-48% of ZO-2. In addition, Ͼ95% of ZO-2 was precipitated by anti-ZO-2 Ab, which coprecipitated 45-55% of 4.1R. Analysis of the post-extraction pellet (insoluble fraction) revealed that only 5-10% of both 4.1R and ZO-2 were not extracted (data not shown). This is consistent with the notion that both 4.1R and ZO-2 move out of the nucleus at confluent density in MDCK cells (75,76). Thus, ϳ35-45% of total cellular 4.1R and ZO-2 associate together in confluent MDCK cell TJs. Taken together, all of these results strongly suggest that about half of the cytoplasmic 135-kDa nonerythroid isoforms of 4.1R in MDCK cells interact with ZO-2.
The complex splicing pattern of 4.1R generates distinct isoforms with insignificant differences in molecular mass, but with a significant effect on their subcellular targeting. Although the isoform(s) of 4.1R that interact with hZO-2 are also of the same higher molecular class (ϳ135 kDa) that interacts with NuMA in interphase nuclei and during mitosis (13), the same isoform(s) may not interact with NuMA because of their distinct subcellular localization. However, it has been shown that a given 4.1R isoform can adopt several localizations within the cell (11 3. 4.1R and ZO-2 coprecipitate in confluent MDCK cell extracts. MDCK extracts were prepared as described under "Experimental Procedures" and subjected to immunoprecipitation using different antibodies. The immunoprecipitates were analyzed by immunoblotting using anti-HP Ab (A), anti-ZO-2 Ab (B), and anti-24-kDa Ab (C). (In C, anti-ZO-2 Ab appears to coprecipitate more 4.1R than anti-HP Ab because of unequal loading.) One-fourth of the immunoprecipitates of anti-HP Ab, preimmune serum, anti-ZO-2 Ab, rabbit IgG, and anti-p53 Ab and one-eighth of the supernatant fractions of anti-HP Ab (anti-HP Ab sup.) and anti-ZO-2 Ab (anti-ZO-2 Ab sup.) immunoprecipitates were loaded in order. segments of 4.1R and hZO-2 or ZO-2 that interact, we examined the interactions between different domains or their segments of 4.1R and hZO-2 or ZO-2 in the yeast two-hybrid assays. We expressed different domains or segments of 4.1R as fusion products of the Gal4 DNA-binding domain in pAS2-1 and cotransformed them into Y190 with different segments of hZO-2 or ZO-2. The latter were expressed as Gal4-AD fusion products in pACT2, as shown in Fig. 4A. Full-length 4.1R (135 kDa), its 80-kDa isoform, and its 22-24-kDa domain interacted with 1) full-length hZO-2, 2) a segment of the C-terminal proline-rich domain of hZO-2 (aa 980 -1168), and 3) amino acids 187-1174 of ZO-2. They did not interact with the N-terminal part of ZO-2 (aa 1-189) (Fig. 4A). The N-terminal extension (HP) and the 30-, 16-, and 10-kDa domains of 4.1R failed to interact with hZO-2 or ZO-2. None of the domains of 4.1R, when expressed as Gal4-BD fusion proteins in Y190, expressed the reporter gene(s) by themselves or in combination with Gal4-AD in pACT2 alone (data not shown). These data suggest that the C-terminal domains of 4.1R and hZO-2 or ZO-2 are necessary and sufficient for their interaction.
As illustrated in Fig. 4B, exons 17 and 18 were not required for interaction of hZO-2 with 4.1R. Fusion proteins consisting of exons 19 -21 of 4.1R (exons 19 -21/pGBT9) bound to hZO-2, but none of these exons alone (exon 19/pGBT9, exon 20/pGBT9, or exon 21/pGBT9) was sufficient for binding. Amino acids 980 -1168 were encoded by the hZO-2 clones obtained from the two-hybrid screening. As shown in Fig. 4C, the C-terminal truncation of the fusion protein from amino acids 1119 to 1168 did not impede binding, but further truncation from the Cterminal end (aa 1074 -1168) abolished the binding of hZO-2 to 4.1R. From the N-terminal end, truncation of the hZO-2 fusion protein from its N terminus up to amino acid 1053 did not affect and their segments were expressed as Gal4-BD or Gal4-AD fusion proteins, respectively, in yeast strain Y190 by cotransformation and were assayed for the expression of the reporter genes (see "Experimental Procedures" for details). Plus signs indicate the expression of the reporter genes lacZ and HIS3 (and thus the interaction between the peptides), and minus signs indicate non-expression of the reporter genes (and thus no interaction between the peptides). B, the C-terminal 96 amino acids of 4.1R are sufficient for its interaction with hZO-2. Schematic diagrams of the various exons within the 22-24-kDa domain of 4.1R (white rectangles) fused to the DNA-binding domain of Gal4 (shaded rectangles) used in the two-hybrid assay are shown. The names of the plasmids that encoded each construct are given to the right of each schematic diagram. Plasmid pAS2-1 expresses the Gal4 DNA-binding domain alone and was used as a negative control. These plasmids were cotransformed with hZO-2/pACT2 expressing amino acids 980 -1168 of hZO-2 fused to the activation domain of Gal4. C, amino acids 1054 -1118 of hZO-2 are sufficient for its interaction with 4.1R. Schematic diagrams of the various hZO-2 polypeptides fused to the activation domain of Gal4 (hatched) used in the two-hybrid assay are shown. Plasmid pACT2, which expresses the Gal4 activation domain alone, was used as a negative control (data not shown). These plasmids were cotransformed with pGBT9 or pAS2-1 expressing the C-terminal domain of 4.1R or its exons 19 -21 fused to Gal4-BD. ␤-gal, ␤-galactosidase; GUK, guanylate kinase; Alt. Spl., alternatively spliced. its binding to 4.1R. The fusion protein encoding amino acids 1054 -1118 of hZO-2 bound to 4.1R. The binding strength was comparable to the fusion protein encoding amino acids 980 -1168 of hZO-2, as assessed by expression of the reporter gene lacZ (data not shown). Therefore, it appears that amino acids 1054 -1118 of hZO-2 are sufficient for its interaction with 4.1R. These results suggest that the amino acids through which 4.1R and hZO-2 interact reside within amino acids encoded by exons 19 -21 of 4.1R and amino acids 1054 -1118 of hZO-2.
Protein 4.1R Associates with Tight Junction Proteins ZO-1, ZO-2, and Occludin in Confluent MDCK Cells-To examine the association of 4.1R with the TJ protein complex, we asked if it occurred in association with other TJ components such as ZO-1 and occludin (54 -56). Therefore, we performed immunoprecipitation using confluent MDCK cell extracts and looked for 4.1R in the immunoprecipitates of ZO-1 and occludin and vice versa. Rabbit preimmune serum, purified rabbit IgG, and anti-p53 Ab (an irrelevant Ab) were used as negative controls to rule out nonspecific aggregation. ZO-2, which is known to coprecipitate with and bind directly to ZO-1 (57,58) and occludin (41,58), was used as a positive control.
As stated earlier and shown in Fig. 5A, two 4.1R isoforms (Х135 and Х150 kDa) were immunoprecipitated by anti-HP Ab. These isoforms of 4.1R were also found to coprecipitate with ZO-2, ZO-1, and occludin when immunoprecipitation was carried out with anti-ZO-2 Ab, anti-ZO-1 Ab, and anti-occludin Ab, respectively, but not with preimmune serum, rabbit IgG, or anti-p53. Analysis of the same immunoprecipitates by immunoblotting using antibodies specific for the 16-kDa domain of 4.1R (anti-16-kDa Ab) also revealed the ϳ135and ϳ150-kDa 4.1R isoforms in the immunoprecipitates of anti-HP Ab, anti-ZO-2 Ab, anti-ZO-1 Ab, and anti-occludin Ab. No other 4.1R isoforms were seen in the immunoprecipitates of ZO-2, ZO-1, and occludin, suggesting that the isoforms of 4.1R that associate with these TJ proteins are of the 135-kDa molecular mass class. This study also confirmed the Х150-kDa protein as a 4.1R isoform.
To examine the involvement of other 4.1R-binding cytoskeletal proteins such as actin and spectrin in this complex, we analyzed the anti-HP Ab, anti-ZO-2 Ab, anti-ZO-1 Ab, and anti-occludin Ab immunoprecipitates by immunoblotting using the relevant antibodies. As shown in Fig. 5F, actin was found to coprecipitate with 4.1R, ZO-2, ZO-1, and occludin, but not with p53, preimmune serum, or rabbit IgG immunoprecipitates. Anti-spectrin Ab detected both the ␣and ␤-isoforms of spectrin in MDCK cell lysates (Fig. 5G). Interestingly, only ␣-spectrin was detected in the anti-HP Ab, anti-ZO-2 Ab, anti-ZO-1 Ab, and anti-occludin Ab immunoprecipitates. The absence of actin and spectrin (despite their abundance in the cell) in immunoprecipitates of preimmune serum, rabbit IgG, and anti-p53 Ab indicated that their presence in the immunoprecipitates of anti-HP Ab, anti-ZO-2 Ab, anti-ZO-1 Ab, and anti-occludin Ab was due to a specific intracellular association with these proteins. The coprecipitation of actin and ␣-spectrin with the TJ proteins was not surprising because the C-terminal half of ZO-1 has been shown to cosediment with actin filaments (58,59), and ␣-spectrin has been shown to coprecipitate with ZO-1 (60, 61).
To examine the presence of other 4.1R-binding and/or TJassociated proteins in this protein complex, anti-HP Ab immunoprecipitates were examined by SDS-polyacrylamide gel electrophoresis and silver staining (Fig. 5H). About 17 protein bands were visualized in the anti-HP Ab immunoprecipitates, but not in the preimmune serum immunoprecipitates. The molecular mass of some of these proteins was identical to that of some of the TJ-associated proteins. A duplicate of the gel in Fig. 5H was transferred to polyvinylidene difluoride membrane and analyzed by immunoblotting for the identification of 4.1R, ZO-1, ZO-2, occludin, actin, and spectrin using the relevant antibodies (data not shown). The protein bands on the silverstained gel that were identified by immunoblotting are labeled (Fig. 5H). These results suggest that 4.1R isoform(s) and TJ proteins ZO-1, ZO-2, and occludin (and possibly others) along with actin and ␣-spectrin associate together in vivo.
Protein 4.1R Does Not Associate with Tight Junction Proteins ZO-1, ZO-2, and Occludin in Nonconfluent MDCK Cells-To examine whether or not the association of 4.1R isoforms is TJ-specific, we repeated the immunoprecipitation experiments in nonconfluent (Ͻ50% confluent) MDCK cell lysates (TJ not organized). As shown in Fig. 6A, in contrast to confluent cell lysates, only one isoform of 4.1R (Х135 kDa) was precipitated by anti-HP Ab. Because 135-kDa 4.1R was expressed in confluent as well as nonconfluent cells, but ϳ150-kDa 4.1R was detected in confluent cells only, it appears that ϳ150-kDa 4.1 might play a critical role in tight junction organization. However, our initial efforts to delineate the proper exon combination of the ϳ150-kDa isoform have not yielded precise identity of this isoform. The characterization of the ϳ150-kDa isoform is in progress. Anti-ZO-1 Ab, anti-ZO-2 Ab, and anti-occludin Ab failed to co-immunoprecipitate 4.1R in the nonconfluent state (Fig. 6A), even though ZO-1 was coprecipitated with ZO-2 and vice versa (Fig. 6, B and C). The fractions of ZO-2 and ZO-1 that coprecipitated were also less compared with those of the confluent cell lysates. Occludin also did not coprecipitate with 4.1R, ZO-1, ZO-2, or any of the controls (Fig. 6D). Analysis of the immunoprecipitates by immunoblotting showed that neither actin (Fig. 6E) nor spectrin (Fig. 6F) was coprecipitated with 4.1R, ZO-1, ZO-2, or occludin. The difference in results observed in contrast to confluent cells was not due to nonexpression of the above proteins because all these proteins were detected in whole cell lysate (Fig. 6, A-F). The results from Figs. 5 and 6 suggest that the association of 4.1R, ZO-1, ZO-2, occludin, actin, and ␣-spectrin together in MDCK cells is dependent on TJ formation, which is organized only when these cells become confluent.
Protein 4.1R Binds to ZO-2, ZO-1, and Occludin in Vitro-Because the results from immunoprecipitation experiments suggested that 4.1R associates with TJ proteins ZO-2, ZO-1, and occludin in vivo, we asked if there were additional binary interactions between 4.1R and ZO-1 or occludin. We constructed yeast expression vectors expressing full-length ZO-1, occludin, or their different segments in frame with Gal4-AD and performed yeast two-hybrid assays with 4.1R and its subdomains. The results of filter lift ␤-galactosidase activity assays were inconclusive. Therefore, chlorophenol red-␤-D-galactopyranoside-based liquid ␤-galactosidase activity assays were performed. As shown in Fig. 7A, the N-terminal cytoplasmic domain of occludin (aa 1-57) did not show substantial ␤-galactosidase activity over the control. However, full-length occludin (data not shown) as well as its C-terminal cytoplasmic domain (aa 250 -504) showed weak interactions with the 135-and 80-kDa isoforms and C-terminal domain of 4.1R. A weak interaction was also observed between the N-terminal cytoplasmic domain of occludin (aa 1-57) and HP of 4.1R. In addition, weak interactions were detected between the 135-and 80-kDa isoforms and the 24-kDa domain of 4.1R and the C-terminal proline-rich domain (aa 1045-1737) (Fig. 7B), but not the Nterminal part (aa 1-1044) of ZO-1 (data not shown).
To confirm direct binding of 4.1R to ZO-2, ZO-1, or occludin, we performed in vitro binding assays. Different isoforms and domains of 4.1R were expressed as GST fusion proteins and purified by coupling to GST-Sepharose beads as described (13). revealed, respectively. These results suggest direct binding of ZO-2, ZO-1, and occludin to 4.1R and are consistent with results from the yeast two-hybrid assays.
Recruitment of 4.1R/GFP Fusion Proteins to Tight Junctions in Confluent MDCK Cells-To verify that 4.1R interacts with ZO-2 and/or other tight junction-associated proteins in intact cells, we constructed expressions vectors with segments of 4.1R fused to GFP and transfected them into cultured MDCK cells. Because results from yeast two-hybrid and in vitro binding assays suggest that the 22-24-kDa domain of 4.1R is sufficient for interaction with ZO-2, ZO-1, and occludin, cells were transfected with 24-kDa/GFP. GFP alone and 10-kDa/GFP were used as controls. However, as shown in Fig. 9, GFP alone, 10-kDa/GFP, or 24-kDa/GFP was not recruited to cell-cell junctions (Fig. 9, A-C), but 10ϩ24-kDa/GFP was efficiently recruited to cell-cell junctions (Fig. 9, D and E). The state of confluency and localization of ZO-1 at cell-cell junctions in the same cells were revealed by staining the cells with anti-ZO-1 Ab (Fig. 9, AЈ-EЈ). Confocal microscopy also revealed that the 10ϩ24-kDa/GFP fusion proteins colocalized with ZO-2 and ZO-1 in cultured MDCK cell tight junctions (data not shown). All these fusion proteins were also concentrated in the nucleus. DISCUSSION Several studies suggest that the cortical cytoskeleton is involved in the structural and functional organization of TJs (reviewed in Ref. 56; Ref. 62), but little is known about the molecule(s) that link these two distinct structures. In this study, we demonstrate that two nonerythroid isoforms of 4.1R (ϳ135 and ϳ150 kDa) interact with hZO-2. In addition to ZO-2, these isoforms of 4.1R also associate with other TJ proteins such as ZO-1 and occludin and the cytoskeletal proteins ␣-spectrin and actin in one protein complex. Endogenous components of TJs also efficiently recruit tagged 4.1R segments containing the spectrin/actin-and ZO-2-binding domains to TJs. We thus hypothesize that 4.1R isoforms in epithelial cells may participate in regulation of TJ by mediating a direct link between the TJ proteins with the underlying cytoskeleton.
ZO-2 belongs to the MAGUK protein family. MAGUK proteins such as p55, human DLG, and the human LIN2 homolog are known to associate with the cortical actin cytoskeleton. In erythrocytes, the MAGUK protein p55 links the transmembrane protein glycophorin C to the spectrin/actin cytoskeleton through the 30-kDa domain of protein 4.1R (20). Human DLG and the human LIN2 homolog are also known to bind to 4.1R in epithelial cells (15,29). Similarly, ZO-1, a member of the MAGUK family of proteins, binds to ZAK, a serine-threonine kinase (63), and links the transmembrane protein occludin to cytoskeleton by binding to F-actin (59). However, the interaction between hZO-2 and 4.1R was unexpected because hZO-2 lacks the 4.1R-binding motif used by the MAGUK proteins p55, human DLG, and human LIN2, a conserved lysine-rich sequence motif located between the SH3 and guanylate kinase domains (15,20,29).
Although the C-terminal domain encoded by exons 19 -21 of 4.1R interacts with multiple PDZ domains containing hZO-2, the expected PDZ domain-tail interaction (64) was not ob- Because deletion of exon 19, 20, or 21 abolishes this interaction, it appears that the amino acids of 4.1R that interact with hZO-2 may be distributed in these three exons and that there may be multiple contact sites between these proteins. It is also possible that the presence of these three exons of 4.1R together gives rise to a particular folding of 4.1R required for its interaction with hZO-2 or ZO-2. A recent study shows that these amino acids of 4.1R are highly conserved among other 4.1R-like genes (30), raising the possibility that these gene products may also interact with hZO-2 and ZO-2.
ZO-2 is known to associate with the cytoplasmic surface of the TJ (51,57). It shares a strong homology with ZO-1, especially within the conserved MAGUK domains (39). It is expressed in almost every tissue (44) and associates with ZO-1 through its second PDZ domain at both the adherens junctions and TJs (41,51,58,65). ZO-2 binds directly to the C-terminal 147 amino acids of occludin (66) and coprecipitates with occludin and ␣-catenin (41). It is a component of the TJ along with ZO-1, ZO-3, cingulin, 7H6 antigen, symplekin, several unidentified proteins, and transmembrane proteins such as occludin (reviewed in Ref. 56) and members of the claudin family (67).
ZO-1, ZO-2, and occludin have been shown to localize at TJs of confluent MDCK cells (51-53, 61). Using anti-HP Ab, an antibody specific for nonerythroid isoforms of 4.1R, we found that 4.1R epitopes are also localized at TJs of MDCK cells by confocal microscopy and high resolution electron microscopy (Fig. 2, B and H, respectively). Confocal microscopic analysis of double-labeled confluent MDCK cells showed that these 4.1R epitopes colocalize at TJs with ZO-2 and occludin (Fig. 2, C and  F). In immunoprecipitation studies using the same antibodies, ZO-2 and two nonerythroid 4.1R isoforms were found to coprecipitate together. About 40 -50% of 4.1R and ZO-2 appear to remain associated together under our experimental conditions. Unlike ZO-2 and ZO-1, which also localize at adherens junction in addition to TJs, occludin is exclusively localized at the TJs (52,53). We found that 4.1R, ZO-1, ZO-2, and occludin coprecipitated together. This association of 4.1R with ZO-2, ZO-1, or occludin was not seen in cells that were not confluent (and thus in which the tight junction was not organized) (Fig. 6), strongly suggesting that these proteins associate together at TJs.
Coprecipitation does not reveal whether there is a direct interaction between 4.1R and ZO-1 or occludin. We thus tested this hypothesis in yeast two-hybrid and in vitro binding assays. As shown in Fig. 7, only weak interactions between the Cterminal domain of 4.1R and the proline-rich domain of ZO-1 (Fig. 7B) or the C-terminal cytoplasmic domain of occludin ( Fig. 7A) were observed in yeast two-hybrid assays. These interactions were only 25-30% of the interaction between 4.1R and hZO-2 or ZO-2 as detected in yeast two-hybrid assays. ZO-2, ZO-1, and occludin bound to 4.1R in in vitro binding assays (Fig. 8). Interestingly, a recent study suggested that a third protein might mediate the interaction between ZO-1 and ZO-2 (58). Similarly, a mutant occludin that bound to ZO-1 in vitro failed to localize at TJs (68). Both ZO-1 and ZO-2 also localize to TJs in occludin-deficient TJs (41,69). These findings imply that association of other factors, in addition to ZO-1 and ZO-2, may be required for localization of occludin at TJs.
Both the 135-and 80-kDa isoforms interact with ZO-2 in yeast two-hybrid assays, whereas HP alone does not. Thus, our results do not define the precise isoform(s) of 4.1R that interact with ZO-2. Because of the complex splicing pathway that 4.1R undergoes, it is difficult to know exactly which exons are included in the isoform(s) that interact with ZO-2. However, epitopes for anti-HP Ab are observed at TJs; moreover, the two isoforms of 4.1R that coprecipitate with ZO-2 and other TJ proteins are exclusively of the higher molecular mass. Finally, no 80-kDa isoform containing exon 19 (which is required for its interaction with ZO-2) is expressed in MDCK cells (13). We thus believe that isoforms of 4.1R that interact with ZO-2 are of the 135-kDa class or of similar molecular mass. The 80-kDa isoform binds to ZO-2 in yeast two-hybrid and in vitro binding assays, but does not coprecipitate with ZO-2. These data are compatible because the increased concentration and proximity of molecules containing the binding sites in yeast two-hybrid and in vitro binding assays could allow the 80-kDa isoform to interact with ZO-2. In the intact cells, factors such as posttranslational modifications and compartmentalization or presence of modifying cofactors will affect their interaction. In the same context, the amino acids encoded by exons 19 -21 of 4.1R are required and sufficient for interaction of 4.1R with ZO-2 in vitro, but not in vivo. Indeed, the 22-24-kDa domain is not sufficient to target GFP to tight junctions, although per se it binds well to ZO-2. These amino acids are highly conserved among the 4.1-like gene family (30). It can thus be argued that other 4.1R-like gene products interact with ZO-2. However, as discussed above, our results strongly implicate 4.1R isoforms as opposed to 4.1-like gene products because the anti-HP Ab used in our assays does not react with other 4.1-like proteins.
A noteworthy finding is the association of 4.1R with ZO-1, FIG. 9. Localization of transiently expressed 4.1R/GFP fusion proteins in confluent MDCK cells. GFP or the 10-kDa/GFP, 24-kDa/ GFP, or 10ϩ24-kDa/GFP fusion proteins were exogenously and transiently expressed in MDCK cells. The cells were grown to confluence, fixed, and stained with anti-ZO-1 Ab (in red; AЈ-EЈ) for localization of ZO-1. Expression of GFP or GFP fusion proteins was visualized by green fluorescence (A-E). 10ϩ24-kDa/GFP fusion proteins were recruited to cell-cell junctions, but not GFP alone, 10-kDa/GFP, or 24-kDa/GFP. Anti-ZO-1 antibody-stained cells that correspond to cells transfected with GFP or 4.1R/GFP fusion proteins are indicated with arrows in corresponding pictures. Magnification ϫ 40.
ZO-2, occludin, ␣-spectrin, and actin apparently in one protein complex, implying a connection of TJs to the cortical cytoskeleton, as previously suggested (Refs. 62 and 70; reviewed in Ref. 56). Several studies also suggest that actin plays a role in regulation of TJ permeability (71,72). ZO-1 has been shown to cosediment with F-actin (58,59) and to colocalize with actin (60,(72)(73)(74). ZO-1, ZO-2, and occludin cosediment with ␣-spectrin (68). ZO-1 has been suggested to link the actin cytoskeleton to TJ via an "actin/ZO-1/occludin" linkage (58,59). The Cterminal parts of ZO-2 and ZO-3 show poor homology to the C-terminal half of ZO-1, which cosediments with actin. ZO-2 and ZO-3 also lack the lysine-rich "protein 4.1-binding motif" found in other MAGUK proteins. These observations suggest that ZO-1 may be the only MAGUK protein at the TJ that can bind to actin and thereby serve as the structural and signaling component of TJs. In contrast, a recent in vitro study suggests that ZO-1, ZO-2, and occludin can directly interact with F-actin (66).
The following evidence from our study suggests that 4.1R may provide or supplement the linkage between the TJ and the cortical cytoskeleton. First, ZO-2, ZO-1, and occludin were found to interact with 4.1R, which in turn is known to bind to the cytoskeletal proteins actin and spectrin. Second, the association of actin and ␣-spectrin with ZO-1, ZO-2, or occludin was not observed in nonconfluent cells when 4.1R also failed to interact with these proteins. Third, ZO-2, ZO-1, and occludin interacted with 4.1R in in vitro binding and yeast two-hybrid assays. Finally, GFP-tagged segments of 4.1R that contained both the spectrin/actin-and ZO-2-binding domains were recruited to TJs, but the segments that contained only the spectrin/actin-binding domain or the ZO-2 binding domain were not. Thus, as shown in Fig. 10, ZO-2, like other MAGUKs, may be establishing a link between the TJ and the actin-based cytoskeleton via a different binding interaction with 4.1R. This function of 4.1R in nonerythroid cells is not established, but unpublished data from our laboratory suggest that the 135-kDa nonerythroid isoform of 4.1R forms a ternary complex with F-actin and fodrin (nonerythroid spectrin) in vitro. 2 Interaction of 4.1R with members of the MAGUK family, ZO-1, and ZO-2 could assist in recruitment of other proteins to coordinate cell signaling. Other members of the protein 4.1 superfamily such as merlin, ezrin, radixin, and moesin are known to be involved in different signaling pathways (36,37). It is thus tempting to speculate that 4.1R may have a role in signal transduction between the TJ and the cytoskeleton. A similar role of 4.1R has been suggested for the human CASK and 4.1R interaction (29).