Occludin was the only integral membrane protein known to be localized at TJ strands (Furuse et al., 1993
; Ando-Akatsuka et al., 1996
; Saitou et al., 1997
). However, occludin-deficient epithelial cells, differentiated from occludin gene double knockout ES cells, bore well-developed TJ (Saitou et al., 1998
). To interpret this unexpected finding, we searched for as yet unidentified TJ integral membrane proteins. Initially, we expected the occurrence of isotypes of occludin, and performed RT-PCR using various combinations of primers and similarity searches of databases. However, this approach failed to find sequences that showed similarity to occludin. To identify the occludin-binding integral membrane proteins, yeast two-hybrid analysis as well as immunoprecipitation with anti-occludin antibodies were carried out, but again no good candidates were identified.
We then re-examined the isolated junction fraction from chick liver, from which occludin was first identified (Tsukita and Tsukita, 1989
; Furuse et al., 1993
). Among the numerous bands observed on SDS-PAGE of this fraction, we screened for bands that were co-partitioned with the occludin band through guanidine extraction and sonication, followed by sucrose density gradient centrifugation; we identified a broad 22-kD band as a candidate. The broadness of this band was partially explained by the occurrence of multiple (at least two) similar but distinct ~22-kD polypeptides. Two peptides were obtained from the chicken 22-kD band, and cDNAs encoding their mouse homologues were cloned. These two proteins were structurally related (38% identical at the amino acid sequence level), but showed no sequence similarity to occludin. We did not succeed in obtaining mAbs or pAbs specific for these proteins, probably because of the small size and/or low antigenicity of extramembrane portions of these molecules. Instead, we introduced cDNAs encoding FLAG- or GFP-tagged proteins into MDCK cells, and found that they exclusively targeted to TJ strands at both light and electron microscopic levels. We then designated these newly identified, related 22-kD proteins as “claudin-1” and “claudin-2” from the Latin word “claudere” (to close). This nomenclature is similar to that of TJ-associated MAGUK family members (Woods and Bryant, 1993
; Kim, 1995
; Anderson et al., 1995
) such as ZO-1, ZO-2 and ZO-3 (Haskins et al., 1998
No sequences similar to occludin were found in databases except for the occludin sequence itself. In contrast, several sequences similar to claudin-1 and -2 have been previously reported as ~22-kD four transmembrane domain proteins (Briehl and Miesfeld, 1991
; Katahira et al., 1997
) and also found in databases, although their physiological functions have not been elucidated. These findings point to the existence of a new multiple gene family which could be called the “claudin family” (Morita, K., M. Furuse, and Sh. Tsukita, manuscript in preparation). As reported previously, the levels of occludin expression in various types of cells are generally correlated well with the number of TJ strands (Furuse et al., 1993
; Saitou et al., 1997
). In contrast, as shown on Northern blots (see Fig. ), no such correlations were found for the expression of claudin-1 or -2. For example, in the lung which shows well- developed TJ, occludin was abundant, whereas the levels of expression of claudin-1 and -2 were rather low. As occludin-deficient mice were born with normal TJ in various tissues (Saitou, M., K. Fujimoto, Y. Doi, M. Itoh, T. Fujimoto, M. Furuse, H. Takano, Sh. Tsukita, and T. Noda, manuscript in preparation), it is likely that members of the claudin family other than claudin-1 and -2 are also involved in the formation of TJ strands in various tissues.
Hydrophilicity analysis predicted that both claudin-1 and -2 bore four transmembrane domains (see Fig. b
). Since the putative first extracellular loop is rather hydrophobic in the four transmembrane domain folding model of claudins, the possibility cannot be excluded that these molecules contain five transmembrane domains. However, considering that claudin-1 and -2 show sequence similarity to several proteins that were reported to have four transmembrane domains as discussed above (Briehl and Miesfeld, 1991
; Katahira et al., 1997
), and that other newly identified members of claudin family were also predicted to bear four transmembrane domains (Morita, K., M. Furuse, and Sh. Tsukita, manuscript in preparation), it appears to be reasonable to conclude that claudin-1 and -2 belong to the so called “four transmembrane domain proteins.” If this interpretation is the case, the hydrophobic character of the first extracellular loop of claudin-1 and -2 may be implicated in the barrier function of tight junctions.
In chicken liver, at least three proteins, occludin, claudin-1, and claudin-2, comprise TJ strands. Since among the guanidine-insoluble bands in isolated junction fraction, only the band ~22 kD behaved similarly to the occludin band, integral membrane proteins other than claudin family members or occludin do not appear to be major components of TJ strands. As epithelial cells derived from occludin-deficient ES cells still bore well-developed TJ strands (Saitou et al., 1998
), we speculated that under specialized conditions the TJ strand itself can be formed solely from claudin family members without occludin. On the other hand, overexpression of occludin in insect Sf9 cells (Furuse et al., 1996
) suggested that occludin alone can form TJ strand-like structures, although these were very short. These observations are similar to the polymerization of intermediate-sized filaments (for review see Fuchs and Weber, 1994
; Heins and Aebi, 1994
). The constituents of the different types of intermediate-sized filaments differ in amino acid sequence and molecular mass, but they all contain a structurally homologous central rod domain that forms an extended coiled-coil structure when hetero- or homodimers are formed, which allows them to be polymerized into filamentous structures collectively categorized as an intermediate-sized filament. Similarly, the integral membrane proteins with four transmembrane domains such as claudin family members and occludin may be polymerized into TJ strands as both hetero- and homo-polymers, although the efficiency of polymerization may differ depending on their combination. At present, it is still premature to discuss in what manner these integral membrane proteins with four transmembrane domains are integrated into TJ strands. Any model, however, must explain the thickness of TJ strands (~10 nm), which is similar to the diameter of the gap junction channel (connexon) consisting of six connexin molecules that also bear four transmembrane domains (for reviews see Staehelin, 1973
; Jiang and Goodenough, 1996
; Kumar and Gilula, 1996
ZO-1 has been thought to be associated with TJ strands through its direct interaction with occludin (Furuse et al., 1994
). However, in epithelial cells differentiated from occludin-deficient ES cells, ZO-1 was still exclusively localized at TJ (Saitou et al., 1998
). As ZO-1 is a multidomain protein containing one SH3, one GUK, and three PDZ domains (Itoh et al., 1993
; Tsukita et al., 1993
; Willott et al., 1993
), it is possible that ZO-1 is associated not only with occludin but also with claudin family members directly or indirectly. Further analyses of the interactions of the cytoplasmic domains of claudin family members with TJ-associated peripheral membrane proteins are required to determine the molecular architecture of TJ in detail.
This study revealed that the molecular architecture of TJ strands is more complex than expected. We concluded that multiple integral membrane proteins with four transmembrane domains, occludin, and claudins, constitute TJ strands. Of course, the most significant limitation of this study is that it has not been definitely proven that claudins (alone or in combination with occludin) actually form TJ strands, but we recently found that the introduction of claudin-1 (and also -2) into cultured L fibroblasts resulted in the formation of a well-developed network of TJ strands at cell–cell borders (Furuse, M., H. Sasaki, K. Fujimoto, and Sh. Tsukita, manuscript in preparation). Molecular manipulation of occludin and claudin functions including their overexpression and targeted gene disruption will lead to a better understanding of how TJ strands are formed as a multi-component complex in vivo.