Domain-swapped Dimerization of the Second PDZ Domain of ZO2 May Provide a Structural Basis for the Polymerization of Claudins*

Zonula occludens proteins (ZOs), including ZO1/2/3, are tight junction-associated proteins. Each of them contains three PDZ domains. It has been demonstrated that ZO1 can form either homodimers or heterodimers with ZO2 or ZO3 through the second PDZ domain. However, the underlying structural basis is not well understood. In this study, the solution structure of the second PDZ domain of ZO2 (ZO2-PDZ2) was determined using NMR spectroscopy. The results revealed a novel dimerization mode for PDZ domains via three-dimensional domain swapping, which can be generalized to homodimers of ZO1-PDZ2 or ZO3-PDZ2 and heterodimers of ZO1-PDZ2/ZO2-PDZ2 or ZO1-PDZ2/ZO3-PDZ2 due to high conservation between PDZ2 domains in ZO proteins. Furthermore, GST pulldown experiments and immunoprecipitation studies demonstrated that interactions between ZO1-PDZ2 and ZO2-PDZ2 and their self-associations indeed exist both in vitro and in vivo. Chemical cross-linking and dynamic laser light scattering experiments revealed that both ZO1-PDZ2 and ZO2-PDZ2 can form oligomers in solution. This PDZ domain-mediated oligomerization of ZOs may provide a structural basis for the polymerization of claudins, namely the formation of tight junctions.

Epithelial and endothelial cells can form selective barriers between tissues and different body compartments. They polarize and adhere to each other through intercellular complexes, including tight junctions (TJs), 3 adherens junctions, and desmosomes (1). TJs, the most apical component of the junction complex, separate the apical from the lateral plasma membrane through forming a continuous belt-like attachment at the outer end of the intercellular space between cells. TJs play important roles in regulating the passage of ions and molecule through the paracellular pathway (2)(3)(4)(5). TJs are also crucial for correct function of blood-brain barrier. Disruption of the TJs in blood-brain barrier is a hallmark of many central nervous system pathologies, including stroke, human immunodeficiency virus encephalitis, Alzheimer disease, multiple sclerosis, and bacterial meningitis. Systemic-derived inflammation has been shown to cause disruption of TJs and increased paracellular permeability (6).
Recent studies have provided accumulated evidences to unveil domains that are responsible for the interactions between ZO proteins. The association between ZO1 and ZO2/ ZO3 is through their second PDZ domains (20,21). Utep-bergenov's group showed significant amount of ZO1 homodimers in MDCK cells and demonstrated that the second PDZ domain is both necessary and sufficient for the dimerization (22). The SH3/GK domain was also reported to be important in ZO1 dimerization, similarly as of in other MAGUK proteins, such as PSD-95 and Dlg/SAP90/SAP102 (23)(24)(25)(26)(27)(28)(29).
Dimerization or oligomerization of ZO proteins plays a pivotal role in determining the activity of their binding partners. Claudins are thought to constitute the backbone of TJ strands (30). They can form homo-and hetero-oligomers during their engagement in the formation of paracellular channels or pores (31)(32)(33)(34)(35). ZOs bind to claudin YV C termini through their first PDZ domain (36). The dimerization of ZO1 (probably also ZO2) not only initiates the polymerization of claudins, namely the formation of TJ strands, but also directs the correct localization of TJ strands (23).
To understand the molecular basis of the PDZ domain-mediated dimerization/oligomerization of ZO family proteins, we have determined the solution structure of ZO2-PDZ2 using NMR spectroscopy. Unexpectedly, ZO2-PDZ2 forms very tight 2-fold symmetric homodimers via a three-dimensional domain swapping assembly mode (37). To the best of our knowledge, this ZO2-PDZ2 dimer is the first example for dimerization of PDZ domain via three-dimensional domain swapping. Further chemical cross-linking and dynamic laser light scattering experiments demonstrated that both ZO1-PDZ2 and ZO2-PDZ2 can form oligomers in solution. The ZO2-PDZ2 dimer structure is ideally suited for the tight assembly of ZOs in TJs.
Expression and Purification of Fusion Proteins-The recombinant plasmids harboring the respective target genes were transformed into Escherichia coli BL21(DE3) host cells for large scale protein production. The purified recombinant protein His-ZO2-PDZ2 contains an N-terminal Met and a C-terminal His tag (-LEHHHHHH), whereas the purified recombinant His-ZO1-PDZ2 protein contains a C-terminal His 6 tag, with the N-terminal Met being cleaved during expression (data from unpublished mass spectrum).
The GST and GST fusion proteins were purified through the glutathione-Sepharose 4B beads (Amersham Biosciences). The soluble His 6 -tagged fusion proteins were purified using HiTrap Chelating columns (Amersham Biosciences) according to the procedures specified by manufacturer. The purity of recombinant proteins were confirmed by Tricine-SDS-PAGE (15%, w/v), and the concentration was measured using BCA kits (Pierce). Uniformly 15 N-and 15 N/ 13 C-labeled His 6 -tagged proteins were prepared through growing bacteria in SV40 medium using 15 NH 4 Cl (0.5g/liter) and 13 C 6 -glucose (2.5g/liter) as stable isotope sources.
In Vitro Binding Assay using GST Fusion Proteins-For in vitro binding studies, the bacterially purified His-ZO1-PDZ2 was equally divided into three parts. Then they were mixed with GST alone, GST-ZO2-PDZ2 or GST-ZO1-PDZ2 fusion protein coupled to the glutathione beads individually. His-ZO2-PDZ2 protein was also equally divided and mixed with GST, GST-ZO1-PDZ2, or GST-ZO2-PDZ2 fusion protein coupled to the glutathione beads individually. The GST-conjugated Glutathione-Sepharose beads were used as a negative control. Samples were then centrifuged at 4000 rpm at 4°C for 4 min and washed ten times using 1ϫ phosphate-buffered saline (PBS). Proteins were eluted from beads by 5-min boiling in sample buffer (Bio-Rad) and separated on 15% denatured SDS-PAGE gel, followed by Coomassie Brilliant Blue staining.
Western Blot Analysis and Co-immunoprecipitation-H1299 cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum at 37°C in 5% CO 2 . Transfection of cells with mammalian expression vectors (see "Plasmids Construction") was performed using Lipofectamine 2000 (Invitrogen) according to the manufacturer's specifications.
For Western blot analysis, cells were washed with 1ϫPBS and resuspended using 5 volumes of cold lysis buffer (50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1% Triton X-100, 1% SDS, 1% sodium deoxycholate) supplemented with protease inhibitor mixture (Roche Applied Science). The cell lysates were incubated on ice for 30 min and were then centrifuged for 10 min at 4°C. Equal amounts of total cellular proteins were separated by SDS-PAGE, and the resolved proteins were transferred to nitrocellulose membrane. After blocking with 5% nonfat milk in TBST (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.1% Tween 20) for overnight at 4°C, the blot was incubated with primary antibody for 2 h at room temperature, and followed by incubating with a alkaline phosphatase-conjugated secondary antibody for 1 h. After washing three times with TBST, the blots were developed using the chemiluminescence detection system (Amersham Biosciences) according to the manufacturer's protocol.
Sedimentation Velocity Analysis-Sedimentation velocity studies were carried out with a Beckman Optima XL-A analytical ultracentrifuge with an An60Ti rotor at 20°C and 56,000 rpm. The partial specific volumes (0.7389 ml/g for ZO1-PDZ2 and 0.7448 ml/g for ZO2-PDZ2), and molecular masses of the monomers (10010 Da for ZO1-PDZ2 and 10081Da for ZO2-PDZ2) were calculated based on their amino acid composition using the program SEDNTERP. Protein samples (at approximate concentrations of 5 mg/ml and 0.5 mg/ml) were prepared in sedimentation buffer (PBS buffer including 1 mM EDTA, pH 7.2). 380-l samples and 400-l reference volumes were loaded into cells. Absorbance scans at 280 or 235 nm were collected at a time interval of 2 min. Sedimentation profiles were analyzed with the software SEDFIT (version 8.9) using the continuous distribution c(s) Lamm equation model (38,39).
NMR Experiments-The purified proteins were dissolved to a final concentration of 0.8 mM in 500 l of 20 mM phosphate buffer (pH 6.5), containing 200 mM NaCl and 1 mM EDTA. Unless otherwise specified, all NMR spectra were acquired at 310K on a Bruker DMX 600 spectrometer. 1 H-15 N HSQC spectra were acquired on uniformly 15 N-labeled or 15 N/ 13 C-labeled samples. The latter sample was also used to collect triple resonance spectra. Backbone resonance assignments of ZO2-PDZ2 were carried out by inspection of three-dimensional HNCO, HN(CA)CO, CBCANH, and CBCA(CO)NH. Side-chain resonance assignments of ZO2-PDZ2 were obtained from HBHA(CBCACO)NH, 15  To determine whether domain swapping has occurred, 13 C F1-filtered, F2-edited two-dimensional NOESY spectra were recorded separately for samples of ZO1-PDZ2 and ZO2-PDZ2 (40). These samples were made from a 1:1 mixture of uniformly FIGURE 1. The interactions between ZO1-PDZ2 and ZO2-PDZ2 and their self-associations. A, in vitro binding assay. The interactions in vitro were examined by GST pulldown assay; Coomassie Brilliant Blue staining was used to ensure the protein loads. GST alone was used as a negative control. His-ZO1-PDZ2 was retained by GST-ZO1-PDZ2 or GST-ZO2-PDZ2 (lanes 2 and 3), but not by GST alone (lane 1). His-ZO2-PDZ2 was captured by GST-ZO2-PDZ2 or GST-ZO1-PDZ2 but not by GST alone (lanes 4 -6). Samples from GST, GST-ZO1-PDZ2, or GST-ZO2-PDZ2, incubating with no His-tagged proteins, were also shown as a control (lanes 7-9). The mobility of molecular mass markers (in kilodaltons) is indicated on the right of the gels. B, in vivo binding assay. The immunoprecipitates were analyzed by Western blot using anti-GFP antibody (upper panel). The presence of GFP-ZO1-PDZ2 or GFP-ZO2-PDZ2 in each immunoprecipitate precipitated was confirmed by Western blot analysis (middle panel) with an anti-GFP antibody. The expression of respective transfected FLAG-tagged constructs was also detected by anti-FLAG antibody (bottom panel). GFP-ZO1-PDZ2 interacted with both FLAG-ZO1-PDZ2 and FLAG-ZO2-PDZ2 (lanes 2 and 6) but not by FLAG alone (lane 1). GFP-ZO2-PDZ2 interacted with both FLAG-ZO2-PDZ2 and FLAG-ZO1-PDZ2 (lanes 3 and 4) but not by FLAG alone (lane 5). The mobility of molecular mass markers (in kilodaltons) is indicated on the left of the gels. 15 N/ 13 C-labeled and unlabeled proteins by unfolding and refolding with 8 M urea, which therefore contained 25% labeledlabeled, 50% labeled-unlabeled, and 25% unlabeled-unlabeled dimers. The appropriate processing of such spectra yields a sub-spectrum where predominantly only NOESY cross-peaks resulting from intermolecular interactions are observed. All of NMR spectra were processed with NMRPipe (41) and analyzed using Sparky3 software. 4 Structure Calculations of ZO2-PDZ2 Domain-NMR distance restraints were collected from two different NOESY spectra: three-dimensional 15 N-separated NOESY in H 2 O for amide protons and three-dimensional 13 C-separated NOESY in D 2 O for aliphatic protons. NOEs were grouped into four distance ranges according to their relative intensity: strong, 1.8 -3.0 Å; medium, 1.8 -4.0 Å; weak, 1.8 -5.0 Å; very weak, 1.8 -7.0 Å. The 1.8-Å lower limits were imposed only implicitly by the van der Waals repulsion force. The NOEs were specified as intermolecular and intramolecular according to domain swapped topology inferred from the 13 C F1-filtered, F2-edited two-dimensional NOESY spectrum. Dihedral restraints ( and ) were determined using the 1 H ␣ , 15 N, 13 C ␣ , 13 C ␤ , and 13 CO chemical shifts and the program TALOS (43). Hydrogen bond (H-bond) restraints were generated by a combination of the H/D exchange experiment and the standard secondary structure of the protein based on NOE patterns. Structures were calculated using the program CNS version 1.1 (44), employing a simulated annealing protocol for torsion angle dynamics. The calculated structures were analyzed by the programs PROCHECK (45) and MOLMOL (46).
Cross-linking of His Fusion Proteins-Purified His fusion proteins were diluted into indicated concentrations and centrifuged at 13,200 rpm for 5 min. The supernatants were then collected and treated with Me 2 SO alone or disuccinimidyl suberate (DSS) in Me 2 SO from a 10-fold stock solution to a final concentration of 5 mM. After incubating at 37°C for 30 min, the cross-linker was quenched by the adding of 1 M Tris-HCl (pH7.5) to a final concentration of 20 mM at room temperature for 15 min. Samples were then solubilized in sample buffer (Bio-Rad), boiled, and centrifuged at 13,200 rpm for 5 min. His fusion proteins were separated on 15% SDS-PAGE gel, followed by Coomassie Brilliant Blue staining. Dynamic Laser Light Scattering Measurements-DLS measurements (47) were conducted on an ALV/DLS/SLS-5022F spectrometer with a multidigital time correlation (ALV5000) and a cylindrical 22-milliwatt UNIPHASE He-Ne laser ( 0 ϭ 632 nm) as the light source. The intensity-intensity time correlation function G (2) (t,q) was measured to determine the linewidth distribution G(⌫). For diffusive relaxation, ⌫ is related to the translational diffusion coefficient (D) of the scattering object (macromolecules or colloid particles) in dilute solution or dispersion by D ϭ (͗⌫͘/q 2 ) C30,q30 and further to hydrodynamic radius (R h ) from the Stokes-Einstein equation: R h ϭ k B T/ (6D), where , k B , and T are the solvent viscosity, the Boltzmann constant, and the absolute temperature, respectively. Hydrodynamic radius distribution f(R h ) was calculated from the Laplace inversion of a corresponding measured G (2) (t,q) using the CONTIN program.
All DLS measurements were conducted at a scattering angle () of 45°at 37.0°C. Protein samples (at approximate concentrations of 3.0 ϫ 10 Ϫ3 g/ml) were prepared in physiological buffer conditions (PBS buffer including 1 mM EDTA, pH 7.2).

RESULTS
The Interactions between ZO1-PDZ2 and ZO2-PDZ2 and Their Self-associations-It has been demonstrated that ZO1 interacts with both ZO2 and ZO3. Deletion of any PDZ2 domain of these proteins can disrupt the complex formation (20,21). The striking homology (67.5%) between ZO1-PDZ2 and ZO2-PDZ2 suggests a great possibility of their self-associations. To test this possibility, we transformed E. coli BL21(DE3) with the plasmids encoding GST-ZO1-PDZ2, GST-ZO2-PDZ2, or GST alone. The lysate from each transformant was incubated with glutathione-Sepharose 4B beads and mixed with bacterially purified His-ZO1-PDZ2 or His-ZO2-PDZ2 proteins in pulldown buffer. As shown in Fig. 1A, His-ZO1-PDZ2 was retained by both GST-ZO2-PDZ2 and GST-ZO1-PDZ2; His-ZO2-PDZ2 was retained by both GST-ZO1-PDZ2 and GST-ZO2-PDZ2, indicating that ZO1-PDZ2 and ZO2-PDZ2 were able to bind not only to each other, but also to themselves. Neither His-ZO1-PDZ2 nor His-ZO2-PDZ2 was remained by GST alone. To further confirm the physical interactions between PDZ2 domains of ZO1 and ZO2 in vivo, co-    (42). Residues identical in all sequences are colored white in the red column; conserved residues are colored red, whereas the others are in black. Secondary structure elements for ZO2-PDZ2 and Shank1PDZ are shown above or below the sequences, respectively, with helices as squiggles, strands as arrows, and turns as TT. DECEMBER 7, 2007 • VOLUME 282 • NUMBER 49 immunoprecipitation was performed as described under "Materials and Methods." As expected, GFP-ZO1-PDZ2 could be co-precipitated by both FLAG-ZO2-PDZ2 and FLAG-ZO1-PDZ2, and GFP-ZO2-PDZ2 could be co-precipitated by both FLAG-ZO1-PDZ2 and FLAG-ZO2-PDZ2. Neither GFP-ZO1-PDZ2 nor GFP-ZO2-PDZ2 was co-precipitated by FLAG only (Fig. 1B). The above results strongly suggest that interactions between ZO1-PDZ2 and ZO2-PDZ2 as well as their self-associations indeed exist in mammalian cells.

Solution Structure of ZO2-PDZ2 Homodimer
ZO1-PDZ2 and ZO2-PDZ2 Form Both Homodimer and Heterodimer Tightly in Solution-As shown in Fig. 2A, ZO1-PDZ2 was eluted out at a molecular mass of 28 kDa, and ZO2-PDZ2 was eluted out as a 25-kDa molecule, suggesting that they both exist as dimers. Their association states were further examined by analytical ultracentrifugation. Only one major peak was detected for each of these two proteins, indicating that they are structurally homogeneous in solution. The results of data analysis with Sedfit version 8.9d showed that ZO1-PDZ2 and ZO2-PDZ2 sediment at 1.95 S and 1.83 S, corresponding to a molecular masses of 21.2 and 20.6 kDa, respectively (Fig. 2B). These results confirmed that ZO1-PDZ2 and ZO2-PDZ2 both exist as dimers, in agreement with the results of gel-filtration chromatography. The two dimers are very tight, because no monomeric fraction was detected by sedimentation velocity analysis.
To characterize the dimeric states of ZO1-PDZ2 and ZO2-PDZ2 by NMR spectroscopy, equivalent amounts of 15 N and 13 C labeled and unlabeled proteins were mixed in a buffer (containing 20 mM PBS, 200 mM NaCl, and pH 6.5) to reach a heterolabeled dimer. The mixture of proteins was treated with 8 M urea to make the protein unfolded and then refolded through dialysis. Two-dimensional 13 C F1-filtered, F2-edited NOESY experiments performed on the newly prepared samples of ZO1-PDZ2 and ZO2-PDZ2 both allowed the identification of a set of intermonomer NOE connectivities, supporting our conclusion that PDZ2 domains of ZO1/2 form dimers in solution. A twodimensional-NOESY experiment with unlabeled sample showed the H ␣ , H ␣ connectivities characteristic of antiparallel ␤-sheets for ZO2-PDZ2. A total of twelve such NOEs, both intramolecular and intermolecular, were observed (Fig. 3A). In the isotope-filtered experiment of ZO2-PDZ2, strong intermolecular NOEs between the backbone H ␣ protons were found in Gly 3 /Val 76 , Leu 5 /Gln 74 , Met 7 /Lys 72 , Leu 17 /Met 28 , Leu 19 /Val 25 , and Ser 21 /Ile 23 (Fig. 3, B and C). Moreover, the N-terminal residues (1-22), including ␤1 (2-7) and half of ␤2 (17-22) of one monomer demonstrate other NOEs with residues beyond Gln 22 of another monomer, suggesting that the dimeric ZO2-PDZ2 adopted a domain-swapping structure. The control experiment with fully labeled dimer was performed, and the result confirmed that the filtering is fully efficient (data not shown). The 1 H-15 N HSQC and 1 H-13 C HSQC spectra showed no significant change between treated and untreated samples (supplemental Figs. S1 and S2), indicating that the refolded PDZ2 dimers retain their native structure (data for ZO1-PDZ2 not shown).
To further verify if ZO1-PDZ2 and ZO2-PDZ2 can form heterodimer in solution, we mixed 15 N-labeled ZO2-PDZ2 with unlabeled ZO1-PDZ2. The mixture was treated with 8 M urea, and subsequently dialyzed against refolding buffer. The 1 H-15 N HSQC spectrum of the refolded mixture showed two set of resonances (Fig. 4, blue), one completely overlaps with the ZO2-PDZ2 homodimer (Fig. 4, magenta), and the other has its distinct pattern presumably representing ZO1-PDZ2/ZO2-PDZ2 heterodimer. Taken together, the biochemical and NMR spectroscopic data conclusively demonstrate that ZO1-PDZ2 and ZO2-PDZ2 can form not only homodimers, but also heterodimers in solution.
NMR Structure Determination of ZO2-PDZ2-To uncover the molecular basis governing the dimerization of PDZ2 of ZOs, the solution structure of ZO2-PDZ2 was determined by NMR spectroscopy. Only one set of backbone resonances was observed in the 1 H-15 N HSQC spectrum of the protein, indicating that the ZO2-PDZ2 dimer is symmetrical. It should be noted that ZO2-PDZ2 sample at low salt concentration (Յ50 mM NaCl) showed obviously weakened signal intensities after days of experiments in NMR tube, suggesting the protein might aggregate to give too broadened resonance peaks to be observed. In contrast, the protein was comparatively stable in solution of high salt concentration (Ն200 mM NaCl), because the 1 H-13 C HSQC spectrum of the protein showed no significant change after 2 weeks at 37°C. Therefore, unless otherwise specified, all NMR experiments were carried out with samples in 200 mM NaCl solution and at a high temperature 37°C to ensure the protein predominantly existed in dimeric state. Finally the three-dimensional structure of the dimeric ZO2-PDZ2 was determined using a total of 2854 experimental restraints derived from NMR spectroscopy, including 590 intermolecular NOEs and 24 intermolecular H-bonds. Intermolecular NOEs were obtained from the three-dimensional 15 N-edited NOESY-HSQC spectrum and three-dimensional 13 C-edited NOESY-HSQC spectrum and further confirmed by the 13 C F1-filtered, F2-edited two-dimensional NOESY spectrum. There are 20 intermolecular H-bonds and 26 intramolecular H-bonds determined experimentally from H/D exchange experiments, and 4 intermolecular H-bonds and 12 intramolecular H-bonds inferred from the standard secondary structure of the protein based on NOE patterns. Fig. 5A shows an ensemble of 20 lowest energy NMR structures. It was selected from 200 accepted structures by requiring no NOE violation Ͼ0.5 Å and no dihedral angle violations Ͼ5°. A ribbon representation of the energy-minimized average structure of ZO2-PDZ2 domain was shown in Fig. 5B. The coordinates of these 20 NMR structures have been deposited into the Protein Data Bank (code 2OSG).
The structural statistics are listed in Table 1. The root mean square deviation (r.m.s.d.) of the well defined regions (Ile 2 -Lys 8 and Tyr 15 -Leu 78 of both subunits) of the 20 structures to the average structure was 0.485 Å for the backbone and 1.016 Å for the heavy atoms. In contrast, the regions of Ser 9 -Glu 14 of two subunits are disordered because of fewer medium and long range NOEs. PROCHECK analysis showed that Ͼ97% of the residues lied in the most favored and additional allowed regions of the Ramachandran plot.
Three-dimensional Domain-swapping Structure of ZO2-PDZ2 Homodimer-ZO2-PDZ2 homodimer consists of two identical polypeptide chains. Each chain adopts the ␤␤␣␤␤␣␤ topology, with two strands from the other subunit (Fig. 5B). Domain swapping of the homodimer occurs through exchanging their N-terminal ␤1and ␤2-strands. The ␤1-strand (residues 2-7) from one subunit forms an antiparallel ␤-sheet with ␤5Ј (residues 77Ј-72Ј) from the other subunit. Uniquely, a 12-residue fragment (residues 17-28) of ZO2-PDZ2 homodimer, which, corresponding to the ␤2 and ␤3 strands in canonical PDZ domain, forms a single extended ␤-strand spanning two PDZ subunits (Fig. 5, B and C). In canonical PDZ domains, strands 2 and 3 are linked by a loop with various lengths, and the two strands form an intramolecular antiparallel ␤-sheet. In the domain-swapped ZO2-PDZ2 dimer, the cor- There are 6 H-bonds between ␤1 and ␤5Ј strands, 6 H-bonds between ␤1Ј and ␤5 strands, and 12 H-bonds between ␤2 and ␤2Ј strands. C, the residues located at the dimer interface between the two monomeric subunits. Main-chain traces are rendered as green tubes for the ZO2-PDZ2 dimer; side chains of the residues located at the dimer interface are rendered as sticks and in different colors. DECEMBER 7, 2007 • VOLUME 282 • NUMBER 49
Dimer Interface of ZO2-PDZ2-A great part of the dimer interface is involved in extended ␤-strands matching most swapped domains (Fig. 6A). The protruding ␤1, ␤2, and ␤5 strands of one subunit are paired off with ␤5Ј, ␤2Ј, and ␤1Ј strands of the other subunit, respectively, in an antiparallel manner. Intermolecular hydrogen bonds are deduced from the geometry of the structure. 24 H-bonds are found in the three antiparallel ␤-sheets (Fig. 6B). The interactions of some residues in the interface between the two monomeric subunits, such as Phe 24 /Ile 45 Ј, Phe 24 /Glu 55 Ј, and Asn 56 /Leu 59 Ј, steady the hinge region of the dimer (Fig. 6C). The protein complex with abundant domain-swapping structure also shows abundant hydrophobic residues in the dimeric interface, suggesting that the hydrophobic interactions might play an important role in maintaining the stabilization of dimer structure.
Evidence for Oligomerization of ZO1-PDZ2 and ZO2-PDZ2-The oligomeric state of ZO1-PDZ2 or ZO2-PDZ2 in solution was probed by chemical cross-linking experiments. DSS crosslinking of ZO1-PDZ2 generated a dimeric and some oligomeric species. ZO2-PDZ2 yielded less dimer and more oligomers when compared with ZO1-PDZ2 treated by DSS under the similar cross-linking condition (Fig. 7A, lanes 1-4 versus 5-8). This suggests that ZO2-PDZ2 might have more oligomeric species in solution than ZO1-PDZ2, or alternatively, ZO2-PDZ2 might contain more favorably positioned lysine residues to allow DSS cross-linking than ZO1-PDZ2. Without DSS treatment, both ZO1-PDZ2 and ZO2-PDZ2 appeared as monomers, despite trace amounts of dimers on SDS-PAGE gel (Fig. 7B), which indicate no covalent bond was formed between the two subunits of the dimers.
The oligomeric state of ZO1-PDZ2 and ZO2-PDZ2 was further examined by DLS. The DLS experiments were performed under physiological buffer conditions at 37°C (Fig. 7, C and D). For ZO1-PDZ2 and ZO2-PDZ2, the hydrodynamic radius distributions f(R h ) exhibit single peak at 1-10 nm at 0 h. As time increases, another peak at ϳ210 nm appears, indicating the occurrence of aggregation. Actually, the aggregation should already occur before the DLS measurements. However, the aggregates were filtered out by a 0.2-m filter, namely, no aggregation at the starting point we chose. Note that f(R h ) is z-averaged here. Actually only a small amount of ZO1-PDZ2 or ZO2-PDZ2 molecules form aggregates or oligomers in terms of weight. After a period of time (28 h for ZO1-PDZ2 and 16 h for ZO2-PDZ2), the oligomerization was nearly finished and the system reached an equilibrium. Therefore, either ZO1-PDZ2 or ZO2-PDZ2 mainly consists of their dimers in solution.
To get an insight into the possible molecular mechanism of oligomerization, the electrostatic potential at the solvent-accessible surface for ZO2-PDZ2 dimer was calculated. Apparently, complementary surface electrostatic potential might contribute to oligomerization of ZO2-PDZ2 (Fig. 8). In our experiments, the ion strength of the solution does affect the processes of aggregation, the higher the salt concentration, the narrower the line width of signals in the 1 H-15 N HSQC spectrum of ZO2-PDZ2 or ZO1-PDZ2 (data not shown). Based on these results, we propose that the tight ZO2-PDZ2 dimers are oligomerized by the charge interactions in solution.

DISCUSSION
In this study, we discovered for the first time that a PDZ domain (ZO2-PDZ2 in this case) can form dimer via three-dimensional domain swapping. The unique amino acid sequence within the ␤2a and ␤2b region of the ZO2-PDZ2 is likely responsible for the formation of the domain swapped dimer. Previous studies have discovered several dimeric patterns for PDZ domains. For example, in Shank1PDZ, its dimer interface involves the ␤1 strand and the ␤2/␤3 loop from each monomer (49), as for GRIPPDZ6, it involves the ␤1 strand and the ␣1/␤4 loop (ligand binding state) (50). As revealed by the multiple sequence alignment analysis (Fig. 5D), ZO2-PDZ2 lacks the long loop between ␤2 strand and ␤3 strand that can be found in shank1PDZ and GRIPPDZ6. This means that the linker between ␤2a and ␤2b of ZO2-PDZ2 is not long enough for ␤2b to swing back to form an antiparallel ␤-sheet with ␤2a strand, hence facilitates the swapping structure. There are accumulated evidences supporting that engineering the hinge loop can affect three-dimensional domain swapping (51). With shortened hinge loop, Domain 1 of CD2 (52), staphylococcal nuclease (53), and single chain Fv (54, 55) form domainswapped dimers. On the other hand, extending the hinge loop in the domain-swapped dimer of cro repressor leads to monomer formation (56). Due to the high conservation of PDZ2 domains in ZO proteins, we suppose that domain-swapping structure also applies for the ZO1-PDZ2 and ZO3-PDZ2 homodimers or ZO1-PDZ2/ZO2-PDZ2 and ZO1-PDZ2/ZO3-PDZ2 heterodimers. The tight dimeric structure of ZO2-PDZ2 (and likely ZO1-PDZ2) is due to extensive intersubunit hydrogen bonds and hydrophobic interactions formed via swapping of the N-terminal two ␤-strands of each subunit.
Furthermore, the in vitro GST pulldown experiments and in vivo immunoprecipitation study have been utilized to define the interactions between ZO1-PDZ2 and ZO2-PDZ2. We dem- onstrated that ZO1-PDZ2 and ZO2-PDZ2 interact directly both in vitro and in vivo; concomitantly, both PDZ2 domains undergo self-associations. Using chemical cross-linking and the dynamic laser light scattering method, we concluded that both ZO1-PDZ2 and ZO2-PDZ2 exist as a combination of dimers and oligomers in solution. The in vitro self-association of and the interactions between ZO1-PDZ2 and ZO2-PDZ2 should be caused by their homo-and hetero-oligomerization, because both ZO1-PDZ2 and ZO2-PDZ2 are tight dimers. It is likely that the unique surface-charge potential property of the homodimers afford them the ability to aggregate.
The fundamental functions of epithelia and endothelia in multicellular organisms are to separate compositionally distinct compartments and regulate the exchange of small solutes and other substances between them. TJs between adjacent cells constitute the barrier to the passage of ions and molecules through the paracellular pathway. Occludin and claudins are identified as constituents of TJ strands (7,8), and claudins are thought to constitute the backbone of TJ strands (30). ZOs bind the carboxyl termini of claudin YV through their first PDZ domain (36). Umeda et al. have shown that dimerized ZO1 (and probably also ZO2) not only initiates the polymerization of claudins but also direct the correct localization of TJ strands. Utepbergenov et al. found ZO1 forms stable homodimers in solution or higher order oligomers in vivo via its second PDZ domain, whereas PDZ3-acidic (a construct containing PDZ3, SH3, GK, and acidic domain) exists only as monomer. They concluded that it is the PDZ2 of ZO1 instead of SH3/GK domain that functions as a dimerization motif (22). On the contrary, Umeda et al. maintained that the SH3/GK domain, but not PDZ2 domain, is important for ZO1 dimerization, and this is similar to other MAGUK proteins such as PSD-95 and Dlg/ SAP90/SAP102 (23). We found that both ZO1-PDZ2 and ZO2-PDZ2 can form tight homodimers, and the homodimers are formed by domain swapping. It has been suggested that either intramolecular or intermolecular domain swapping between SH3 and GK domains exists in PSD-95, and the intermolecular domain swapping may be promoted by regulatory factors in vivo (25,26). Accordingly, we presume that intramolecular domain swapping kept the PDZ3-acidic of ZO1 in its monomer form in Utepbergenov's study, whereas in Umeda's study, regulatory factors induced intermolecular domain swapping of SH3/GK module in ZO1. For the oligomerization or multimerization of ZO proteins in vivo, both PDZ2 domain and SH3/GK module may be necessary.
Therefore, we propose a model for the roles of PDZ2 domains as well as SH3/GK modules in regulated oligomerization or multimerization of ZOs (Fig. 9). Fig. 9A presents threedimensional domain swapping between PDZ2 domains and between SH3-GK modules of closed dimer; Fig. 9B shows three-dimensional domain swapping between PDZ2 domains and between SH3-GK modules of open oligomer. Both cases may promote homo-, hetero-oligomerization or homo-, hetero-multimerization of ZOs, which in turn initiate the polymerization of claudins and consequently the assembly of TJ strands. This model is consistent with the observation that ZO1 or ZO2 was first recruited to the junction area, which initiated and facilitated the polymerization of claudins (23). The three-dimensional domain swapping of ZOs, perhaps directed by sets of regulatory proteins, could provide combinatorial scaffold diversity, resulting in different protein recruitments. Previous studies have provided evidence for three-dimensional domain swapping as a mechanism for allostery and signal sensing in a macromolecule, and therefore regulating biological functions of proteins (51). ZO proteins' function as scaffolding proteins in the formation of intracellular signaling complexes could be greatly promoted by oligomerization or multimerization via three-dimensional domain swapping.