Apical junctions and cell signalling in epithelia

Epithelial cell layers protect the organism from its environment, separate different compartments of the organism, transport materials from one compartment to another, and produce both liquid- and solid-phase extracellular components of the body. These functions depend upon the physical integrity and impermeability of the epithelial sheet, and on the apical-basal polarity of its constituent cells. The specialized junctions that are formed between adjacent epithelial cells are critical to epithelial properties: they provide a strong physical attachment between cells, they provide a tight barrier to transepithelial movement of molecules and ions, and they restrict mobile membrane proteins to either the apical or the basal-lateral membrane domain, thus contributing to apical-basal cell polarity (Rodriguez-Boulan and Nelson, 1989). These functions of specialized junctions have been revealed by molecular, physiological and cell biological studies. However, evidence from genetic analysis is now suggesting an additional and critically important role of apical cell junctions in the development of epithelial cell populations. These structures appear to mediate the interactions between cells and their neighbors that control both cell proliferation and the formation of spatial patterns of differentiation.

Intercellular junctions of vertebrate epithelial cells include, in apical-basal order (Fig. 1): tight junctions (also called occluding junctions or zonulae occludens), adherens junctions (also called belt desmosomes or zonulae adherens), both of which form belts around the apical end of the cell, as well as gap junctions and desmosomes scattered along the lateral membrane basal to the adherens junctions (Rodriguez-Boulan and Nelson, 1989). In arthropods the tight junction is missing, but an additional junction, the septate junction, is found basal to the adherens junction (Figs 1 and 2).

Interactions between epithelial cells and their neighbors control cell proliferation, differentiation and morphogenesis, and these interactions must take place via the unspecialized lateral membrane or via the specialized junctions listed above. At the present state of knowledge any or all of the specialized junctions connecting epithelial cells could be involved in these cell interactions. Most investigations of cell communication have been focused on gap junctions, at least in part because simple physiological tests for their function are available. However, an increasing volume of evidence points to the apical junctions as the sites of developmentally significant cell interactions. Much of this evidence comes from immunolocalization of proteins identified by genetic analysis in Drosophila, and a summary of the locations of these proteins is presented in Table 1.

The adherens junction is a thick density that forms a continuous belt around the apical end of each epithelial cell and is associated with actin microfilaments (Poodry and Schneiderman, 1970; Rodriguez-Boulan and Nelson, 1989). One of the main functions of adherens junctions appears to be in cell adhesion (Geiger et al., 1987) and at least one group of cell adhesion molecules, the cadherins, and their associated anchoring molecules, the catenins, have been localized at adherens junctions (Geiger and Ayalon, 1992).

Recent studies of human tumor cells show that loss of expression of cadherins and related molecules is associated with loss of cell proliferation control as well as adhesion (Schipper et al., 1991; Shimoyama and Hirohashi, 1991; Field, 1992). In fact, the E-cadherin locus at chromosome band 16q22.1 (Natt et al., 1989) may correspond to a candidate tumor suppressor gene identified by loss of heterozygosity in hepatocellular, breast and prostate carcinomas (Carter et al., 1990; Sato et al., 1990; Tsuda et al., 1990; Zhang et al., 1990). Furthermore, treatment with anti-E-cadherin antibodies can lead to invasive behavior of otherwise non-invasive epithelial and carcinoma cell lines (Behrens et al., 1989; Frixen et al., 1991) and transfection with E-cadherin cDNA can reverse the invasiveness of invasive carcinoma cell lines (Frixen et al., 1991). On the other hand some transformed cells show loss of adhesiveness while maintaining cadherin expression. For example, the lung cancer cell line PC-9 shows reduced adhesiveness in spite of apparently normal cadherin expression. This is apparently because a-catenin is not expressed in these cells, and lack of this molecule prevents cadherin function (Shi-moyama et al., 1992).

Lethal mutations at the fat locus in Drosophila, which encodes an enormous cadherin-like transmembrane protein, cause hyperplastic overgrowth of the imaginal discs (Mahoney et al., 1991). In the mutant imaginal discs, some of the excess cells are shed as small closed vesicles, which are lost from the epithelium, suggesting a failure of cell adhesion (Bryant et al., 1988). The phenotypes of these mutants suggest an important role for this cadherin-related molecule in the control of Drosophila cell growth. However, it is not yet known whether the protein is localized in cell junctions.

Adherens junctions appear to participate in cell-cell interactions not only through the cadherin/catenin complex but also through their association with a protein tyrosine kinase (PTK)-mediated signalling pathway. Tyrosine phosphorylation is very rare in normal vertebrate cells, accounting for only 0.02-0.05% of all protein phosphorylation events (Sefton et al., 1980). However, in epithelial cells of both vertebrates (Maher et al., 1985; Takata and Singer, 1988; Tsukita et al., 1991; Volberg et al., 1991) and Drosophila (J.-W. Wu, D. F. Woods and P. J. Bryant, unpublished data) most of the small amount of phosphotyrosine is highly localized at adherens junctions.

Tyrosine phosphorylation at adherens junctions is at least partly a function of non-receptor PTKs. Two of these enzymes, pp62^c-yes and pp60^e-src, are highly enriched in the adherens junctions of hepatocytes, kidney epithelial cells and kératinocytes (Tsukita et al., 1991). Expression of the oncogenic pp60^v-src in epithelial cells causes abnormally high levels of tyrosine phosphorylation, breakdown of adherens junctions and loss of cell-cell adhesion (Warren and Nelson, 1987; Volberg et al., 1991). Inhibiting the activity of phosphatases with vanadate and thereby increasing the level of tyrosine phosphorylation at adherens junctions can also cause breakdown of these junctions and changes in cell properties that suggest loss of adhesiveness (Matsuyoshi et al., 1992; Volberg et al., 1992).

An important substrate of pp60^v-src at adherens junctions appears to be the cadherin/catenin complex. Thus in fibroblasts or epithelial cells transformed with v-src, cadherins are expressed and localized at the cell surface but are apparently unable to function properly in cell adhesion or metastasis suppression. These transformed cells show a high level of pp60^v-src-mediated tyrosine phosphorylation on catenins, and a lower level on cadherins (Matsuyoshi et al., 1992; Behrens et al., 1993; Hamaguchi et al., 1993). This raises the intriguing possibility that the adhesive functions of cadherins depend critically on their association with catenins, and that this association is (or can be) functionally modulated by tyrosine phosphorylation mediated by PTKs.

Adherens junctions are clearly important for cell adhesion, but genetic studies show that they are also sites of developmentally significant signalling between epithelial cells. Studies of the Drosophila embryo have provided highly suggestive evidence for a role of adherens junctions in cell-cell interactions controlling the spatial pattern of cell differentiation. Prior to cellularization, the spatial patterns of gene expression that lead to anterior-posterior patterning of the embryo are controlled by the distribution of maternal products and interactions between nuclei in the syncytium. Soon after cellularization, however, the patterns of gene expression come to depend on interactions between the newly formed cells (for review see Woods and Bryant, 1992). This is seen most clearly in genes in the segment polarity class, such as wingless (wg) and engrailed (en). Lethal mutations in these genes cause loss of structures normally found in the posterior parts of body segments and their replacement by anterior structures. The expression of en in the posterior part of each embryonic segment depends on the expression of the wingless gene in anteriorly adjacent cells (DiNardo et al., 1988) as shown by the premature loss of en expression in wg embryos (Heemskerk et al., 1991) and by the expanded domain of en expression produced by ectopic expression of wg (Noordermeer et al., 1992). The sequence of the wg gene shows that it encodes a protein with a signal sequence but no transmembrane domain, indicating that it is secreted (Cabrera et al., 1987; Rijsewijk et al., 1987). This prediction is confirmed by immunocytochemical studies, which show that the Wg protein is actually transfered from the expressing cell into the responding cell across the boundary separating the anterior from the posterior compartment of the body segment. The transfer occurs in the apical part of the cell just basal to the adherens junctions (González et al., 1991).

A segment polarity phenotype similar to that of wg is also produced by mutations in the armadillo (arm) gene, which encodes an adherens junction protein (Peifer, 1993; Peifer et al., 1993) showing highly significant homology with the vertebrate adherens junction protein β-catenin (Peifer et al., 1992). β-Catenin is known to interact with α-catenin and the cytoplasmic domain of cadherins in a multi-molecular complex at the adherens junction (Magee and Buxton, 1991). In a similar way, the Arm protein interacts with the Drosophila α-catenin homolog and a glycoprotein similar in size to vertebrate cadherins (Peifer, 1993). In the female germ line, although well-defined adherens junctions are not present, the Arm protein appears to be required for cell adhesion and cytoskeletal integrity (Peifer et al., 1993). It is therefore possible that arm mutations produce a wg-like phenotype in developing embryos and imaginal discs (Peifer et al., 1991) because they disrupt adherens junctions, which are required for the normal functioning of the wg signal.

Somewhat later in Drosophila development, cell-cell interactions are important for regulating the spatial distribution of neuroblasts in the developing central nervous system (Artavanis-Tsakonas et al., 1991). After many small groups of cells (proneural clusters) become neuroblast-competent, lateral inhibition between the cells of each cluster results in one cell acquiring the neuroblast fate, and the remainder becoming epidermis (for review see Woods and Bryant, 1992). The Notch gene product, a transmembrane protein with 36 sequences homologous to epidermal growth factor (EGF) in its extracellular domain, is required for these cell interactions as well as others controlling neurogenesis in the peripheral nervous system (Artavanis-Tsakonas et al., 1991), and for several different cell interactions controlling the fate of cells in the developing retina (Cagan and Ready, 1989b). Embryos lacking Notch function give rise to neoplastic tissue when transplanted into adult female hosts (Gateff and Schneiderman, 1974) suggesting an additional role for the gene product in the cell interactions controlling proliferation. The product of the Delta gene, which like the Notch gene product is required for the cell interactions leading to neuroblast specification, is also a transmembrane protein containing multiple EGF repeats in its extracellular domain (Vassin et al., 1987). Delta and Notch proteins expressed on different cells can mediate adhesion between them (Fehon et al., 1990), and genetic studies provide evidence for direct interactions between these two proteins in vivo (Xu et aL, 1990). Notch protein is localized at apical junctions in various epithelia (Fehon et aL, 1991), and the same seems to be true of Delta in the follicular epithelium of the ovary (Bender et al., 1993) although its localization in other cell types has not been reported. Nevertheless, many lines of evidence at the molecular and genetic levels are consistent with the hypothesis that these two proteins act as a signal-receptor pair for cell-cell interactions operating at apical junctions to determine cell fates. Studies of genetic mosaics in imaginal discs indicate that cells mutant for Notch autonomously adopt the mutant phenotype of excess bristle production, whereas cells mutant for Delta mutations can non-autonomously show the wild-type phenotype if they are adjacent to wildtype cells (Heitzler and Simpson, 1991). These results argue in favor of a model in which the Delta product is the signal and the Notch product is the receptor for the signalling event that specifies bristle sites.

During the development of the pattern of ommatidia in the forming eye imaginal disc, a series of fate-determining interactions occurs between the individual cells of each developing ommatidium. The best understood of these interactions is the induction of the R7 photoreceptor cell by the R8 cell (for review see Woods and Bryant, 1992). Analysis of genetic mosaics indicates that the bride-of-sev-enless (boss) product is required in the R8 precursor (Reinke and Zipursky, 1988) and the sevenless (sev) product is required in the R7 precursor (Tomlinson and Ready, 1987) for a successful inductive event. The predicted boss product has seven transmembrane segments as well as extracellular and cytoplasmic domains (Hart et al., 1990). The predicted sev product is a transmembrane protein with a functional protein tyrosine kinase catalytic domain in the cytoplasmic tail (Hafen et aL, 1987; Basler and Hafen, 1988; Bowtell et aL, 1988; Simon et aL, 1989). Both Sev and Boss proteins are predicted transmembrane proteins, leading to the idea that the boss protein expressed on the surface of the R8 cell directly induces the adjacent cell to develop into R7, by interacting with the Sev protein expressed on the surface of the responding cell. Immunocytochemistry using epitope-specific antibodies shows that the entire Boss protein is transferred into the R7 precursor cell during the interaction, and is contained in a prominent multivesicular body similar to a late endosome (Cagan et aL, 1992). Both Sev (Cagan et aL, 1992) and Boss (Tomlinson et aL, 1987) proteins are localized at apical microvilli and adherens junctions, and the internalization of Boss protein occurs in the apical part of the cell. Furthermore, the Drk protein, which functions immediately downstream of sev in the signalling pathway, is also localized to the apical plasma membrane of epithelial cells in the eye imaginal disc (Olivier et aL, 1993).

Other Drosophila PTKs and growth factor homologs may be associated with adherens junctions. The non-receptor PTK Dabl is enriched at adherens junctions (Bennett and Hoffmann, 1992) and the 66 kDa src-related protein encoded by the D.src28C locus is localized to the cell periphery (Vincent et al., 1989), possibly at adherens junctions. DER, the Drosophila homolog of the epidermal growth factor receptor, has been described as being present at the apical microvillar surface of imaginal discs cells (Zak and Shilo, 1992), but the data are also consistent with localization at the adherens junction. Finally, the product of the crumbs gene, a transmembrane protein with EGF-like repeats in its extracellular domain (Tepass et aL, 1990) that is required to maintain epithelial integrity in the early embryo, is localized apically (Tepass and Knust, 1990), possibly at adherens junctions, in embryonic epithelia.

In the epithelia of vertebrates, tight junctions form a belt around the end of the cell, apical to the adherens junctions. They provide a transepithelial diffusion barrier to movements of ions and small molecules, and restrict mobile membrane proteins to either the apical or the basal-lateral membrane domains, thus maintaining apical-basal cell polarity (Rodriguez-Boulan and Nelson, 1989). Breakdown of tight junctions by treatment of cells with proteases or chelating agents leads to loss of apical-basal polarity and intermixing of apical and basal-lateral membrane proteins (Pisam and Ripoche, 1976; Ziomek et aL, 1980; Herzlinger and Ojakian, 1984). Some protein components of tight junctions have been identified, including ZO-1 (Anderson et al., 1989), ZO-2 (Gumbiner et al., 1991), cingulin (Citi et al., 1991), and the protein identified by antibody 7H6 (Zhong et al., 1993). However, critical tests of the developmental functions of tight junctions have not yet been reported.

The most obvious difference between arthropod and vertebrate epithelia is that, in general, arthropods have septate junctions but no tight junctions, whereas vertebrates have tight junctions but no septate junctions. Septate junctions (Fig. 2) are characterized by electron-dense septa between cells and are associated with both actin filaments and microtubules (Lane, 1991). Although claims have been made for the existence of tight junctions at the blood/brain barrier in insects (Lane, 1992), other studies of this region show the presence of typical septate junctions (Juang and Carlson, 1992). Because they replace each other and show some structural and functional similarities (Green and Bergquist, 1982; Wood, 1990), it has been suggested that tight and septate junctions have similar roles (Noirot-Timothee and Noirot, 1980), even though the septate junctions are basal to the adherens junctions whereas tight junctions are apical (Figs 1 and 2). Our results support the idea that the septate junctions also maintain apical-basal polarity, and they further suggest that these junctions play an important role in the cell-cell interactions controlling proliferation.

The first protein to be identified at septate junctions was the product of the discs-large (dig) tumor suppressor gene in Drosophila (Woods and Bryant, 1991). Mutations in this gene cause the imaginal discs, which are normally single-layered, to overgrow and become disorganized masses in which the cells have almost completely lost apical-basal polarity (Fig. 3). The discs also lose the ability to develop into adult parts after transplantation into normal hosts. As shown using antibodies against recombinant Dig protein, the gene product is expressed in most epithelial tissues throughout development, and it also shows expression in other tissues including the nervous system (Woods and Bryant, 1991). In the carboxy-termi-nal 179 amino acids, the predicted Dig protein shows 35.5% identity to yeast guanylate kinase (Stehle and Schulz, 1992), an enzyme that catalyzes the conversion of GMP to GDP, using ATP as the phosphate donor. However, although the GMP-binding features of yeast GUK are highly conserved in Dig, the putative ATP-binding site has a three-amino acid deficiency (Koonin et al., 1992), suggesting that the protein may have lost its kinase function during evolution and may now have a different function dependent on GMP binding.

The predicted Dig protein, as well as its two mammalian homologs (Bryant and Woods, 1992; Cho et al., 1992), also includes a copy of the 71-amino acid SH3 domain (Musac-chio et al., 1992a,b), which is found in many membrane-associated signal transduction proteins and is thought to mediate binding to other proteins. Recently a protein that binds with high affinity to the SH3 domain of the c-abl protein was identified and found to show homology to the GTPase activating protein associated with the ras-related protein rho (Cicchetti et al., 1992). The rho protein is involved in actin bundling and regulates the assembly of focal adhesions (Ridley and Hall, 1992). Other binding targets for SH3 include the GTP-hydrolyzing motor protein dynamin (Booker et al., 1993), and guanine nucleotide exchange factors (Olivier et al., 1993; Rozakis-Adcock et al., 1993). The Dig protein also contains an OPA repeat (Wharton et al., 1985) and a PEST motif (Rogers et al., 1986), both of which are thought to control protein synthesis and turnover rates. The N-terminal half of the molecule contains three copies of a ∼91-amino-acid motif called GLGF (Cho et al., 1992) or DHR (Bryant et al., 1993), the function of which is unknown. The results obtained with the dig gene suggest that cell interactions important for growth control occur at septate junctions, that the Dig protein is important in morphogenesis of the septate junction, and that it may also regulate the production, at these junctions, of guanine nucleotides that act as messenger molecules within the cell.

In epithelial cells the Dig protein is localized at the apical end of the lateral cell membrane, just basal to the adherens junctions (Fig. 4B). This is exactly the location of the septate junction (Fristrom, 1982). Furthermore, our genetic studies (unpublished) show that the dig product is required for the formation of septate junctions. The effects of dig mutations on cell polarity, adhesion and proliferation may therefore be a result of disruption of the septate junction. A cell adhesion molecule, fasciclin III, is localized at septate junctions (Table 1), so some of the effects of dig mutations, including the loss of cell adhesion, could be mediated by their effects on the localization or function of this molecule.

If cell communication occurs at apical junctions, the signals generated must be transmitted to the nucleus where they can be translated into effects on patterns of gene expression and/or replication. This might suggest the need for a signal transduction mechanism to relay the signal through the cytoplasm to the nucleus, and many molecules that may participate in this signal transduction have recently been identified (Pelech and Sanghera, 1992). However, a special aspect of the biology of epithelial cells provides a natural mechanism for direct interaction between the nucleus and apical cell junctions.

Epithelial cells undergo a characteristic set of morphological changes during the course of a cell cycle (Figs 4, 5 and 6; Fujita, 1962). The nucleus is basal during S phase, but it migrates apically during G2, and goes through mitosis at the extreme apical end of the cell. At this time the cell itself rounds up (Figs 4 and 5), but at least in Drosophila imaginal epithelium the cell retains a thin connection to the basal lamina (Mathi and Larsen, 1988). Following mitosis the cell regains its columnar shape, and the nucleus migrates basally to begin the next S phase. This apical-basal ‘elevator movement’ (Jinguji and Ishikawa, 1992) accounts for the original erroneous interpretation of simple epithelia as stratified, with a layer of ‘germinal cells’ at the apical surface. Thymidine incorporation studies (Fujita, 1962) clearly reveal that cells seen at different apical-basal levels are simply cells at different stages of the cell cycle (Fig. 6). In vertebrate cells the tight junctions remain intact during the entire cell cycle, preserving the integrity of the epithelium as a barrier to trans-epithelial diffusion (Jinguji and Ishikawa, 1992). In Drosophila the adherens and septate junctions are also maintained in cells through mitosis (Figs 4 and 5).

When an epithelial cell rounds up and its nucleus enters M phase there is very little, if any, cytoplasm surrounding the nucleus (Fujita, 1962). Thus, when the nuclear envelope breaks down, the nuclear contents are apparently in direct contact with the cell membrane including the apical junctions (Fig. 5). We therefore suggest that signals generated by interactions with adjacent cells at cell junctions are translated directly into effects on transcription and replication factors in the nucleus, without the need for an elaborate cytoplasmic signal transduction mechanism or transport of signals through the cytoplasm. The signals could be mediated by phosphorylation events controlled by junction-associated protein kinases. This hypothesis implies that such signals have their effect on nuclear factors only during M phase, and that the nucleus implements the signals autonomously during the subsequent interphase.

The cell fate decisions that occur during the development of photoreceptor clusters in the Drosophila eye disc provide strong circumstantial evidence for cell-fate decisions associated with nuclear elevator movements (Fig. 7). The elevator movements in this system occur in a precisely defined sequence that is not associated with mitosis but is correlated with specifically defined cell interactions establishing cell fates, some of which can be recognized at the molecular level. The sequence of events is best understood in the determination of the R7 cell (for review see Woods and Bryant, 1992), which requires the expression of the boss gene in the R8 precursor cell at about 3 hours after the furrow passes, boss protein is internalized into the adjacent R7 cell at 6-11 hours, and the expression of the sev protein in R7 is required at 6-7 hours. In the absence of this inductive signal from R8, the R7 cell becomes committed to develop into a cone cell instead of a photoreceptor cell. Committment to the R7 fate occurs after the R7 cell is exposed to the sev signal for about six hours (Mullins and Rubin, 1991), and the nucleus of R7 is apical at exactly this time (Tomlinson and Ready, 1987). Nuclei of the other photoreceptor cells also migrate apically in a sequence that corresponds to the sequence of their subsequent differentiation, and the same is true later of the cone cells. Thus the detailed information available from this system is strikingly consistent with the idea that cell interactions controlling cell fate occur via the apical junctions, where both Boss and Sev protein are localized and where Boss is transferred and internalized, and that this interaction directly programs the nucleus at the time when it is apically positioned.

Molecular genetic studies are providing strong evidence for the idea that apical cell junctions, both adherens and septate, are the main sites of the interactions between epithelial cells that control their developmental fates and proliferative behavior. The well-known elevator movements of epithelial cell nuclei may reflect a mechanism providing for the presentation of signals generated at the apical junctions to the mechanisms that control gene expression and replication in the nucleus. Most of the evidence for this model comes from genes that function in the undifferentiated imaginai disc epithelium or the undifferentiated ectoderm of the Drosophila embryo. In both systems there is a dynamic pattern of spatially organized gene expression that ultimately leads to the intricate spatial pattern of differentiation seen in the cuticle and other derivatives (Akam, 1987; Bryant, 1993). Cell signalling mechanisms operating at the apical junctions appear to be intimately involved in setting up these spatially ordered patterns of gene expression.

We thank Dr Mark Peifer for a gift of antibody against the armadillo protein, and Dr Francis Davis for a gift of the anti-MPM-2 antibody. Our research is supported by grants from the National Institutes of Health and the National Science Foundation.

The hypothesis that tight and septate junctions may represent analogous structures has been strengthened by the recent finding that a major protein component of tight junctions, ZO-1, is homologous to the Dig protein (M. Itoh, A. Nagafuchi, S. Yonemura, T. Kitani-Yasuda and S. Tsukita (1993). J. Cell Biol. 121, 491-502; E. Willot, M. S. Baida, A. S. Fanning, B. Jameson, C. Van Itallie and J. M. Anderson (1993). Proc. Nat. Acad. Sci. USA90, 7834-7838; D. F. Woods and P. J. Bryant (1994). Meeh. Dev. (in press).