Manner of Interaction of Heterogeneous Claudin Species within and between Tight Junction Strands

Rabbit anti–mouse claudin-2 pAb and rabbit anti–mouse claudin-3 pAb were raised and characterized previously ( Kubota et al. 1999; Morita et al. 1999a). Guinea pig anti–mouse claudin-1 pAb was raised against the synthetic polypeptides CPRKTTSYPTPRPYPKPTPSSGKD, which corresponds to the COOH-terminal cytoplasmic domains of mouse claudin-1 (amino acid 186–209). These pAbs were affinity-purified on nitrocellulose membranes with GST fusion proteins with the COOH-terminal cytoplasmic domain of respective claudins. GST fusion proteins were expressed in Escherichia coli (strain DH5α) and purified with the glutathione-Sepharose 4B beads (Amersham Pharmacia Biotech).

Rat anti–mouse claudin-1 mAb and rat anti–mouse claudin-2 mAb were produced as described previously ( Itoh et al. 1991). Wistar rats were immunized with GST fusion proteins with the COOH-terminal cytoplasmic domain of respective claudins, and the lymphocytes were fused with P3 myeloma cells to obtain hybridoma cells. Fusion plates were screened by immunofluorescence staining of L transfectants expressing claudin-1 or -2 or by immunoblotting of GST fusion proteins. Mouse anti-FLAG mAb (M2) was purchased from Eastman Kodak Co.

To construct plasmids for the expression of GST/claudin-1 and GST/claudin-2 fusion proteins in E. coli, cDNAs encoding the COOH-terminal cytoplasmic region of claudin-1 (amino acid 188–211) and that of claudin-2 (amino acid 189–230) were amplified by PCR. Each cDNA fragment was subcloned into pGEX4T-1 (Amersham Pharmacia Biotech) to produce pGEX-IN (GST/claudin-1) and pGEX-IU (GST/claudin-2). The plasmid for the expression of GST/claudin-3 was constructed as described previously ( Morita et al. 1999a).

The mammalian expression vectors for claudin-1, -2, and -3 were constructed as follows. The cDNA fragment containing the whole open reading frame of claudin-1 or -2 was excised from pGTCL-1 or pGTCL-2 ( Furuse et al. 1998b) by SacI or EcoRI digestion, respectively. Each DNA fragment was subcloned into pCAGGSneodelEcoRI ( Niwa et al. 1991) to produce the expression vectors pCCL-1 (claudin-1) and pCCL-2 (claudin-2). Similarly the whole open reading frame of claudin-3 cDNA was subcloned into pCAGGSneodelEcoRI to construct the expression vector pCCL-3 ( Morita et al. 1999a). To construct the expression vector for the claudin-1 mutant lacking its COOH-terminal cytoplasmic domain, the cDNA fragment of claudin-1 corresponding to amino acid 148–188 was obtained by PCR using primers GGCGACATTAGTGGCCACAGCATG (sense) and CGCGGATCCTTTCCGGGGACAGGAGCA (antisense), followed by EcoRI and BamHI digestion. The cDNA fragment of claudin-1 encoding amino acid 148–211 was removed from the plasmid pSKCL-1F, which contained FLAG-tagged claudin-1 cDNA ( Furuse et al. 1998a), by EcoRI and BamHI digestion and was replaced by this PCR-amplified truncated cDNA fragment. Then the FLAG-tagged COOH-terminal cytoplasmic domain-deleted claudin-1 cDNA was subcloned into the expression vector pCAGGSneodelEcoRI to make pCCL-1ΔCF.

To establish L transfectants singly expressing claudin-1, -2, or -3 (C1L, C2L, and C3L, respectively), mouse L cells cultured in 35-mm dishes were transfected with 1 μg of pCCL-1, pCCL-2, or pCCL-3 in 1 ml of Opti-MEM using lipofectamine plus (GIBCO BRL). After 14–16-d selection in DME containing 500 μg/ml of G418, resistant colonies were picked up. Isolated clones were screened by immunofluorescence microscopy. C1C2L cells coexpressing claudin-1 and -2 were established by transfection of C1L cells with a mixture of 1 μg of pCCL-2 and 0.1 μg of pPGKpuro, and then selected in DME containing 8 μg/ml of puromycin. Similarly, C1C3L and C2C3L cells were produced by transfecting C3L cells with pCCL-1 and pCCL-2, respectively, together with pGKpuro. C1FC2L cells coexpressing claudin-1 with a FLAG tag at its COOH terminus and claudin-2 were established by transfecting C1FL cells expressing FLAG-claudin-1 ( Furuse et al. 1998b) with 1 μg of pCCL-2 and 0.1 μg of pSV2hph ( Santerre et al. 1984), followed by selection in 200 μg/ml of hygromycin B. C1ΔCFL cells expressing COOH-terminal cytoplasmic domain-truncated claudin-1 with FLAG-tag was produced by transfecting L cells with pCCL-1ΔCF, followed by G418 selection. Several stable clones were isolated for each transfection experiment. Among these, clone 16 of C1L, clone 12 of C2L, clone 17 of C3L, clone 1 of C1C2L, clone 12 of C1C3L, clone 15 of C2C3L, clone 11 of C1FC2L, and clone 16 of C1ΔCFL were recloned and used for this study, since they expressed relatively large amounts of introduced proteins. Mouse L cells and transfectants were cultured in DME supplemented with 10% FCS.

SDS-PAGE was performed according to the method of Laemmli 1970, and proteins were electrophoretically transferred from gels onto polyvinylene difluoride membranes. The membranes were soaked in 5% skimmed milk and incubated with the primary antibodies. After washing, the membranes were incubated with the second antibodies for rat, rabbit (Amersham Pharmacia Biotech), or guinea pig (Chemicon International, Inc.) IgG, followed by incubation with streptavidin-conjugated alkaline phosphatase (Amersham Pharmacia Biotech). Nitroblue tetrazolium and bromochloroindolyl phosphate were used as substrates to visualize the enzyme reaction.

L transfectants (1.5 × 10⁶ cells) were plated on 60-mm culture dishes with coverslips. For coculture, 7.5 × 10⁵ cells of each transfectant were mixed and plated. After a 48-h culture, cells on coverslips were fixed with 1% formaldehyde in PBS for 10 min at room temperature, washed with PBS, and treated with 0.2% Triton X-100 in PBS for 10 min. Cells were then washed with PBS, soaked in 1% BSA in PBS, and incubated with primary antibodies for 30 min in a moist chamber. After washing with PBS three times, cells were incubated with the fluorescently labeled second antibodies for 30 min. FITC-conjugated goat anti–rat IgG (BioSource International), Cy3-conjugated goat anti–rabbit IgG (Amersham Pharmacia Biotech), Cy3-conjugated goat anti–mouse IgG (Amersham Pharmacia Biotech), rhodamine-conjugated goat anti–guinea pig IgG, and FITC-conjugated goat anti–rabbit IgG (Chemicon International, Inc.) were used as secondary antibodies. Cells were washed three times with PBS and then mounted in 90% glycerol-PBS containing para-phenylenediamine and 1% n-propylgalate. Frozen sections of mouse liver were stained immunofluorescently, as described previously ( Furuse et al. 1993). Specimens were observed using a fluorescence Zeiss Axiophot photomicroscope, and the images were recorded with a SensysTM cooled CCD camera system (Photometrics).

For conventional freeze-fracture EM, L transfectants cultured on 60-mm dishes were fixed with 2% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.3) overnight at 4°C, washed with 0.1 M sodium cacodylate buffer three times, immersed in 30% glycerol in 0.1 M sodium cacodylate buffer for 2 h, and then frozen in liquid nitrogen. Frozen samples were fractured at −100°C and platinum-shadowed unidirectionally at an angle of 45° in Balzers Freeze Etching System (BAF060, BAL-TEC). Samples were then immersed in household bleach, the replicas floating off the samples were picked up on formvar-filmed grids and examined with a JEOL 1200 EX electron microscope at an acceleration voltage of 100 kV.

Immunoelectron microscopy for examining freeze-fracture replicas was performed as described by Fujimoto 1995. Mouse liver was cut into small pieces and quickly frozen in high pressure liquid nitrogen with an HPM010 high pressure freezing machine (BAL-TEC). L transfectants were fixed with 1% paraformaldehyde in 0.1 M phosphate buffer (pH 7.3) for 5 min at room temperature, washed three times in 0.1 M phosphate buffer, immersed in 30% glycerol in 0.1 M phosphate buffer for 3 h, and then frozen in liquid nitrogen. Frozen samples were then fractured and shadowed as described. Samples were immersed and stirred in lysis buffer containing 2.5% SDS, 10 mM Tris-HCl, and 0.6 M sucrose (pH 8.2) for 12 h at room temperature, then replicas floating off the samples were washed with PBS containing 5% BSA. The replicas were incubated with primary antibodies for 2 h, washed with PBS-BSA, incubated with colloidal gold-conjugated secondary antibodies, washed with PBS-BSA, and then picked up on formvar-filmed grids. Goat anti–rabbit IgG coupled with 5-nm gold, goat anti–mouse IgG coupled with 15-nm gold (Amersham Pharmacia Biotech), and goat anti–guinea pig IgG coupled with 15-nm gold (British BioCell International) were used as secondary antibodies.

We raised monoclonal and polyclonal antibodies against the COOH-terminal cytoplasmic domains of claudin-1, -2, and -3. Immunoblotting of GST fusion proteins with the COOH-terminal tails of claudin-1, -2, and -3 confirmed the specificity of these antibodies ( Fig. 2 A). Then, cDNAs encoding claudin-1, -2, and -3 were transfected into L fibroblasts and stable transfectants were obtained (C1L, C2L, and C3L cells, respectively). When the total cell lysates of these transfectants were immunoblotted with anticlaudin-1, -2, and -3 pAbs, bands of respective claudins with the expected molecular masses were clearly detected ( Fig. 2 B).

Previously, we obtained L transfectants stably expressing claudin-1 or -2 that were tagged with a FLAG sequence at their COOH termini (C1FL and C2FL cells, respectively) and found that introduced FLAG-claudin-1 and -2 were concentrated at cell–cell contact planes in an elaborate network pattern ( Furuse et al. 1998b). Immunofluorescence microscopy of C1L, C2L, and C3L cells with anticlaudin-1, -2, or -3 antibodies revealed that introduced claudin-1, -2, and -3 without an epitope tag were also highly concentrated at cell–cell borders to form elaborate networks ( Fig. 2 C). Similar to C1FL and C2FL cells, well-developed networks of rTJ strands/grooves were induced at these cell–cell contact planes of C1L, C2L, and C3L cells, as revealed by conventional freeze-fracture EM (see Fig. 5, g–i and Fig. 8, a and b). Judging from the morphological characteristics, as well as the size of the networks detected by immunofluorescence microscopy at cell–cell contact planes ( Fig. 2 C, c, f, i, l, and o), each line in these networks can be regarded to represent individual rTJ strands. These immunofluorescence observations further confirmed the specificity of anticlaudin-1, -2, and -3 pAbs and mAbs used in this study.

Northern blotting revealed that several distinct species of claudins are coexpressed in individual organs, such as the lung, liver, and kidney ( Furuse et al. 1998a; Morita et al. 1999a). Since claudin-1, -2, and -3 appeared to be expressed in the liver, we first double stained frozen sections of the mouse liver with guinea pig anticlaudin-1 pAb/rabbit anticlaudin-2 pAb or guinea pig anticlaudin-1 pAb/rabbit anticlaudin-3 pAb. As shown in Fig. 3, a–d, claudin-1, -2, and -3 were colocalized at the junctional complex regions along bile canaliculi. These findings indicated that claudin-1, -2, and -3 were evenly mixed in these cells, at least at the level of immunofluorescence microscopy. Furthermore, the freeze-fracture replicas obtained from mouse liver were double labeled with anticlaudin-1 pAb and anticlaudin-3 pAb ( Fig. 3 e). Although, with unknown reason, the labeling ability of guinea pig anticlaudin-1 pAb for freeze-fracture replicas was very low, the 15-nm and 5-nm gold particles representing the localization of claudin-1 and -3, respectively, appeared to be admixed along individual TJ strands. Therefore, at the level of TJ strands, it was likely that claudin-1, -2, and -3 were copolymerized into individual TJ strands as heteropolymers of claudins (see Fig. 1 B, Model C and D).

To evaluate this possibility, we used L transfectants as a model system. As shown previously ( Furuse et al. 1998b) and in Fig. 2 C, claudin-1, -2, and -3 can form homopolymers in L transfectants. We then cotransfected claudin-1 and -2, claudin-1 and -3, or claudin-2 and -3, and obtained stable transfectants (C1C2L, C1C3L, or C2C3L cells, respectively; Fig. 4 A). When confluent cultures of C1C2L cells were double stained with anticlaudin-1 mAb and anticlaudin-2 pAb, both species of claudins showed the same concentration pattern at cell–cell contact planes ( Fig. 4 B, a and b). Also, at higher magnification, almost identical network patterns of staining were observed at cell–cell contact planes by anticlaudin-1 mAb and anticlaudin-2 pAb ( Fig. 4 B, g–i). Similarly, claudin-1 and -3, and claudin-2 and -3 were also colocalized in C1C3L and C2C3L cells, respectively ( Fig. 4 B, c–f).

Next, we observed the morphology of rTJ strands in C1C2L, C1C3L, and C2C3L cells by conventional freeze-fracture EM. As previously shown in C1FL and C2FL cells ( Furuse et al. 1998b), in glutaraldehyde-fixed C1L cells claudin-1–induced strands were largely associated with the protoplasmic (P) -face as mostly continuous structures with vacant grooves at the extracellular (E) -face (P-face–associated TJs; see Fig. 5 g), whereas in C2L cells claudin-2–induced strands were discontinuous at the P-face with complementary grooves at the E-face that were occupied by chains of particles (E-face associated TJs; see Fig. 5 h and 8 a; Tsukita and Furuse 1999). Similar to C1L cells, P-face–associated TJs were induced in C3L cells (see Fig. 5 i and 8 b). Interestingly, C1C3L cells bore typical P-face–associated TJs ( Fig. 5 b), whereas C1C2L and C2C3L cells showed an intermediate morphology of rTJ strands/grooves between P- and E-face–associated TJs ( Fig. 5, a and c); the grooves at E-face of C1C3L cells were completely vacant ( Fig. 5 e), whereas those in C1C2L and C2C3L cells were characterized by evenly scattered particles ( Fig. 5d and Fig. f). These findings suggested that two distinct species of claudins were completely mixed up in individual rTJ strands in these L transfectants.

To confirm this speculation, the freeze-fracture replicas obtained from confluent C1C2L cells were double immunolabeled with guinea pig anticlaudin-1 pAb and rabbit anticlaudin-2 pAb. As shown in Fig. 6 a, the 15- and 5-nm gold particles representing the localization of claudin-1 and -2, respectively, appeared to be admixed along individual rTJ strands. Similar results were also obtained in C1C3L cells ( Fig. 6 b). However, as described above, the labeling ability of anticlaudin-1 pAb for freeze-fracture replicas was fairly low (see Fig. 3 e). Therefore, we obtained L transfectants coexpressing claudin-2 and FLAG-tagged claudin-1 (C1FC2L cells), and the freeze-fracture replicas obtained from confluent C1FC2L cells were double labeled with anti-FLAG mAb (15-nm gold particles) and anticlaudin-2 pAb (5-nm gold particles; Fig. 6c and Fig. d). In these images, numerous 15- and 5-nm gold particles appeared to distribute evenly and specifically along individual rTJ strands. Taken together with Fig. 5, we concluded that in the model system of L transfectants, distinct species of claudins can be copolymerized to form rTJ strands as heteropolymers (see Fig. 1 B, Model C or D).

As described in Fig. 1 A, in TJs each TJ strand associates laterally with another TJ strand in the apposing membrane, which is responsible for intercellular adhesion at TJs. The next question was whether this lateral association, i.e., the cell adhesion at TJs, is attributable to homophilic or heterophilic interactions of the extracellular regions of claudins. Since paired rTJ strands were induced in C1L, C2L, and C3L cells, the homophilic interaction of claudins, as described in Model A ( Fig. 1 B) can be expected in these cells. We next examined the possibility of the heterophilic interaction of claudins between homopolymers ( Fig. 1 B, Model B) by coculturing pairs of C1L, C2L, and C3L cells.

As shown in Fig. 7 a–c, when C1L and C3L cells were cocultured and double stained with anticlaudin-1 mAb and anticlaudin-3 pAb, three types of cell–cell contact planes were distinguished in terms of immunofluorescence staining: the claudin-1–positive/claudin-3–negative planes, which would be formed between adjacent C1L cells; the claudin-1–negative/claudin-3–positive planes, which would be formed between adjacent C3L cells; and the claudin-1–positive/claudin-3–positive planes, which would be formed between adjacent C1L and C3L cells. At higher magnification, in the claudin-1–positive/claudin-3–positive planes, both claudin-1 and claudin-3 in C1L and C3L cells, respectively, were concentrated in an elaborate network pattern ( Fig. 7j and Fig. k), and their networks were mostly overlapped ( Fig. 7 l). The same results were obtained in cocultures of C2L and C3L cells ( Fig. 7, d–f and m–o). These findings indicated that claudin-3–based homopolymers (rTJ strands) can associate laterally with claudin-1– and claudin-2–based homopolymers through the heterophilic interactions of claudin-3 with claudin-1 and claudin-2, respectively. In contrast, when C1L and C2L cells were cocultured and double stained with anticlaudin-1 mAb and anticlaudin-2 pAb, the claudin-1–positive/claudin-2–positive planes were not formed ( Fig. 7, g–i). Therefore, it was likely that, at least in L transfectants, claudin-1 cannot interact with claudin-2 in a heterophilic manner. In conclusion, in L transfectants, paired strands can be formed not only by homophilic interactions, but also by heterophilic interactions of claudins, except for some combinations of claudins.

To further confirm this conclusion, confluent cocultures of C2L and C3L cells were fixed with glutaraldehyde and examined by conventional freeze-fracture EM ( Fig. 8). As described above, induced TJs in C2L and C3L cells were E- and P-face–associated, respectively, i.e., the fracture planes at C2L/C2L contact planes showed discontinuous strands on the P-face (P-C2L in Fig. 8 a) and grooves occupied with chains of particles on the E-face (E-C2L in Fig. 8 a), while the fracture planes at the C3L/C3L contact planes showed continuous strands on the P-face (P-C3L in Fig. 8 b) and vacant grooves on the E-face (E-C3L in Fig. 8 b). In the C2L/C3L coculture, in addition to these C2L/C2L and C3L/C3L fracture planes, fracture planes, which would be derived from the contact regions between adjacent C2L and C3L cells, were occasionally identified. In these planes, as expected, E-C2L and P-C3L ( Fig. 8 c) or E-C3L and P-C2L (data not shown) were seen to be paired. Interestingly, as exemplified in Fig. 8 c, when the fracture plane jumped from the E- to the P-face, the continuity of the network pattern of grooves of E-C2L and strands of P-C3L was completely maintained, conclusively indicating that in these planes individual claudin-2 homopolymers in C2L cells always associated laterally with claudin-3 homopolymers to form TJs in adjacent C3L cells.

In this study, we discussed the formation of TJ strands from the viewpoint of claudin–claudin interaction. However, we should consider the possible involvement of peripheral membrane proteins in TJ strand formation. Since the COOH-terminal cytoplasmic domain of claudins was thought to be responsible for their interactions with peripheral membrane proteins, we constructed a claudin-1 mutant lacking its COOH-terminal cytoplasmic domain (claudin-1ΔC), and obtained L transfectants expressing FLAG-tagged claudin-1ΔC (C1ΔCFL cells; Fig. 9 A). These claudin mutants were concentrated at cell–cell borders as shown immunofluorescently in Fig. 9 B. Freeze-fracture replicas obtained from C1ΔCFL cells revealed that this claudin-1 mutant still bore well-developed network of rTJ strands ( Fig. 9 C), and interestingly, these rTJ strands were largely associated with the P-face as mostly continuous structures with vacant grooves at the E-face. These findings suggested that the interaction between claudins and peripheral membrane proteins was not required for the formation of TJ strands, per se, as well as their P-face association.

Previous studies indicated that claudins are polymerized within plasma membranes to constitute the backbone of TJ strands ( Furuse et al. 1998a, Furuse et al. 1998b). Furthermore, it is accepted that each TJ strand laterally associates with another TJ strand in apposing membranes to form paired strands. Therefore, there would be two methods of interaction between claudins; the side-by-side interaction for polymerization to form TJ strands, and the head-to-head interaction between each of the paired strands for cell adhesion. Since claudins comprise a multigene family ( Furuse et al. 1998a; Morita et al. 1999a; Tsukita and Furuse 1999), each of the above methods of interaction can be further subdivided into homo- and heterophilic interactions ( Fig. 1 B). Therefore, the possible molecular organizations of paired TJ strands can be subclassified into four models ( Fig. 1 B, Model A to D). In this study, we evaluated these models using L transfectants expressing claudin-1, -2, and -3 singly or in combination. First, a single species of claudins was sufficient to reconstitute TJ strands, indicating that homopolymers were formed in these cells (see Fig. 2; Furuse et al. 1998b). Furthermore, these homopolymers were associated laterally, both in a homophilic (see Fig. 2), as well as heterophilic manner (see Fig. 7). Therefore, TJs in L transfectants can adopt both organizations represented in Model A and B ( Fig. 1 B). On the other hand, heteropolymers were also induced in L transfectants (see Fig. 4). Models C and D ( Fig. 1 B) were not distinguished in the L transfectant system, but since both Model A and B were possible, both Model C and D would also be possible ( Fig. 1 B).

The thickness of TJ strands is ∼10-nm in freeze-fracture replica images ( Staehelin 1973, Staehelin 1974). Interestingly, the gap junction channel consisting of six connexins is also ∼10-nm in diameter ( Kumar and Gilula 1996; Jiang and Goodenough 1996). Since connexin also has four transmembrane domains with the same membrane topology as claudins, it is tempting to speculate that oligomers of claudins are also unit structures in TJs, and that these unit structures are arranged linearly to form individual TJ strands. This raises a new question as to whether these unit structures themselves are homomeric or heteromeric, further subdividing the above four models; heteropolymers can be formed by linearly aligned heteromeric unit structures, as well as distinct homomeric unit structures. This type of discussion has been reported in detail for gap junctions, in which the unit structures have been determined ( Kumar and Gilula 1996; Jiang and Goodenough 1996), but in TJ strands the clarification of unit structures is a prerequisite for further discussion.

Northern blotting showed that most tissues expressed more than two species of claudins ( Furuse et al. 1998a; Morita et al. 1999a). Immunofluorescence microscopy revealed that TJs in situ contained claudin-4 and -8 in the kidney ( Morita et al. 1999a) and claudin-1, -2, and -3 in the liver ( Fig. 3). Therefore, taking the results obtained from the cotransfection experiments (see Fig. 5 and Fig. 6), it is likely that in situ most TJ strands are heteropolymers of claudins, although specialized TJ strands in myelin sheaths of oligodendrocytes and in Sertoli cells in the testis appeared to be mainly composed of a single species of claudin, claudin-11 ( Morita et al. 1999b). As shown in this study, not only homophilic, but also heterophilic interactions of claudins between each of the paired TJ strands are allowed. Thus, in the paired TJ strands of heteropolymers in situ homophilic interactions of various claudin species, as well as heterophilic interactions between various combinations of claudin species, are expected to occur, as shown in Model D ( Fig. 1 B). It is reasonable to postulate that the strength of these interactions varies depending on the claudin species involved and their combinations, and that the tightness of each TJ strand is determined as a whole by the number/type of species of claudins and their mixing ratio in the strand. For example, MDCK I cells have a fairly tighter TJ barrier than MDCK II cells, but no difference was detected in the number of TJ strands between these two distinct clones of MDCK cells ( Stevenson et al. 1988). These observations suggested a qualitative difference of the paired TJ strands in their tightness between these two clones, and this difference could be explained as postulated above.

To avoid further complexity, we have not discussed occludin. Immunoreplica analyses indicated that occludin was also incorporated into TJ strands in situ in most types of epithelial cells ( Fujimoto 1995; Furuse et al. 1996; Saitou et al. 1997). When occludin was cotransfected into L cells, together with claudin-1, they were coincorporated into rTJ strands (and of course also paired rTJ strands; Furuse et al. 1998b). Occludin singly introduced into L cells was concentrated at cell–cell borders in a punctate manner, indicating that occludins interact with each other in a homophilic manner between adjacent cells ( Furuse et al. 1998b). Furthermore, when L transfectants expressing occludin were cocultured with C1L cells, neither occludin or claudin-1 was concentrated at cell–cell borders, suggesting that occludin does not interact with claudin-1 in a heterophilic manner between adjacent cells (Furuse, M., and S. Tsukita, unpublished data). Therefore, at present it is very difficult to postulate how occludin is incorporated into the paired heteropolymers of claudins (TJ strands) in situ. Occludin may be incorporated into claudin-based TJ strands in the same manner as claudins, and in the paired TJ strands occludin may be positioned opposite to occludin, which allows occludin–occludin interaction ( Furuse et al. 1998b). Alternatively, it may be positioned opposite to a claudin which may result in the formation of small pores. It is also possible that occludin is involved in TJ strand formation differently from claudins.

Finally, we should discuss the possible role of peripheral membrane proteins, such as ZO-1 ( Stevenson et al. 1986), ZO-2 ( Gumbiner et al. 1991), and ZO-3 ( Balda et al. 1993; Haskins et al. 1998) in the formation of TJ strands. ZO-1, but not ZO-2 or ZO-3, was expressed endogenously in L cells, and this endogenous ZO-1 was coconcentrated with claudin-1 and -2 at cell–cell borders as an elaborate network in C1L and C2L cells, respectively (Itoh, M., M. Furuse, K. Morita, and S. Tsukita, unpublished data). In our previous study ( Furuse et al. 1998b), we reported that claudin-1 and -2 with a FLAG tag at their COOH-termini also formed a well-developed network of rTJ strands in L transfectants. Interestingly, however, endogenous ZO-1 was not recruited to these FLAG-claudin-1 or FLAG-claudin-2–based rTJ strands, probably because the FLAG sequence affected the claudin/ZO-1 interaction (Furuse, M., and S. Tsukita, unpublished data). These findings indicated that ZO-1 (and also ZO-2 and ZO-3) was not required for the formation of rTJ strands in L transfectants. Furthermore, Fig. 9 showed that a claudin-1 deletion mutant lacking almost all of its COOH-terminal cytoplasmic domain still formed a well-developed network of rTJ strands in L transfectants, favoring the notion that the peripheral membrane proteins are not involved in the polymerization of claudins within plasma membranes. Therefore, we interpreted the data presented in this study without considering the possible involvement of peripheral membrane proteins.

The model system of L transfectants used in this study was very useful to analyze the properties of each claudin species. To date, 15 members of the claudin family have been identified, but it remains unclear how many claudins will be identified in the future. It is thus very difficult, by the use of epithelial cells, to clarify the nature and function of each claudin species, partly because we cannot determine exactly all the types of claudins expressed in certain epithelial cells. Therefore, to further understand the structure and functions of claudins and TJ strands, the L transfectant system will continue to be used towards the reconstitution of TJs as a complementary technique to the knockout of each claudin gene.

Note Added in Proof. Positional cloning has identified a new member of the claudin family (paracellin-1/claudin-16) in which mutations cause hereditary renal hypomagnesemia in humans (Simon, D.B., Y. Lu, K.A. Choate, H. Velazquez, E. Al-Sabban, M. Praga, G. Casari, A. Bettinelli, G. Colussi, J. Rodriguez-Soriano, et al. 1999. Paracellin-1, a renal tight junction protein required for paracellular Mg²⁺ resorption. Science. 285:103–106). This finding favors the idea that claudins possibly are involved in the formation of aqueous pores within TJ strands.

We thank all the members of our laboratory (Department of Cell Biology, Faculty of Medicine, Kyoto University) for helpful discussions. Our thanks are also due to Ms. K. Yoshida, K. Fukui, and Ms. K. Furuse for their excellent technical assistance, and to Drs. K. Morita and N. Sonoda (Kyoto University) for construction of the claudin-3 expression vector and production of mAbs, respectively.

This study was supported in part by a grant-in-aid for Cancer Research and a grant-in-aid for Scientific Research (A) from the Ministry of Education, Science, and Culture of Japan to S. Tsukita and in part by a grant from the Kazato Research Foundation to M. Furuse.