Hyper-adhesion in desmosomes: its regulation in wound healing and possible relationship to cadherin crystal structure

Tissues such as cardiac muscle and epidermis are extremely resistant to shearing forces and physical stress. A vital contributing factor in stress resistance is strong intercellular adhesion mediated by cell-cell junctions. Among these is the desmosome, which is particularly abundant in tissues that are subject to stress. Analysis of human autoimmune and genetic disease, and targeted deletions of desmosomal genes in mice shows that abnormality of desmosomes leads to tissue disruption (reviewed by Chidgey, 2002; Garrod et al., 2002a; Garrod et al., 2002b; Getsios et al., 2004; Huber, 2003; McMillan and Shimizu, 2001; Payne et al., 2004; South, 2004).

Formation of intercellular adhesion appears to be initiated by adherens junctions and subsequently reinforced by desmosomes (Vasioukhin et al., 2000). The essential nature of desmosomal reinforcement is demonstrated by the loss of epidermal integrity, which occurs following conditional knockout of the desmosomal plaque protein desmoplakin from the epidermis (Vasioukhin et al., 2001). Thus it appears that desmosomes are of prime importance for maintaining tissue integrity. Apart from their abundance in tissues such as epidermis, which is well documented (Skerrow et al., 1989), no clear explanation for the ability of desmosomes to mediate strong intercellular adhesion has been advanced.

The principal adhesion molecules of both desmosomes and adherens junctions are cadherins, members of a large family of molecules that mediate Ca²⁺-dependent cell-cell adhesion (Nollet et al., 2000). In adherens junctions these are classical cadherins such as E-cadherin, and in desmosomes they are the desmosomal cadherins, desmocollin and desmoglein. In accordance with the Ca²⁺-dependent nature of these molecules, the formation and disruption of desmosomal adhesion have both been shown to be Ca²⁺ dependent (Hennings and Holbrook, 1983; Kartenbeck et al., 1982; Mattey and Garrod, 1986a; Mattey and Garrod, 1986b; Watt et al., 1984). However, Ca²⁺ dependence is only a temporary property of desmosomal adhesion; after initial desmosome assembly desmosomal adhesion becomes Ca²⁺ independent, which is resistant to disruption by reduction of extracellular Ca²⁺ concentration or Ca²⁺ chelation (Mattey and Garrod, 1986b; Watt et al., 1984).

Factors regulating the Ca²⁺ independence of desmosomes of Madin-Darby canine kidney (MDCK) cells in tissue culture have been investigated in some detail (Wallis et al., 2000). We showed that development of Ca²⁺ independence requires culture confluence. The desmosomes of cells maintained at subconfluent density do not progress beyond Ca²⁺ dependence. Moreover Ca²⁺-independent desmosomes in confluent cell sheets revert to Ca²⁺ dependence when confluence is destroyed by wounding the cell sheet. Having been initiated at the wound edge, reversion to Ca²⁺ dependence is propagated through the cell monolayer. Reversion to Ca²⁺ dependence occurs in the presence of cycloheximide and thus does not require protein synthesis. Rather, it appears to be regulated by protein kinase C (PKC) signalling because PKC activation promotes Ca²⁺ dependence and inhibition promotes Ca²⁺ independence. The change between Ca²⁺ dependence and Ca²⁺ independence caused by PKC activators and inhibitors is rapid (∼15 minutes) and much less than the time thought to be required for desmosome assembly. Immunolocalisation and antisense depletion experiments showed that the PKCα isoform is involved in this regulation. The acquisition of Ca²⁺ independence is specific to desmosomes because adherens and tight junctions do not become Ca²⁺ independent (Wallis et al., 2000). We suggested that the modulation of desmosomal adhesion on disruption of epithelial confluence somehow facilitates downregulation of desmosomal adhesion and the promotion of cell migration in wound healing (Wallis et al., 2000). It was also shown that desmosomes in several mouse epithelial tissues are Ca²⁺ independent (Wallis et al., 2000). Previously desmosomes resistant to disruption by Ca²⁺ chelation have been reported in frog tissues (Borysenko and Revel, 1973).

Numerous studies have been carried out on the structural basis of adhesion by classical cadherins. A crystallographic study of the entire extracellular domain (EC1-EC5) of Xenopus C-cadherin showed that distal subdomains EC1-EC3 are involved in two main types of molecular interactions (Boggon et al., 2002). One is the `strand dimer', which mediates trans interaction and is formed by the insertion of the side chain of a tryptophan residue (Trp2) near the N-terminus into a hydrophobic pocket of the EC1 subdomain of the partner molecule (Haussinger et al., 2004; Overduin et al., 1995; Shapiro et al., 1995). This trans interaction is interpreted as the adhesive interaction that occurs between two cell surfaces. The other type is a cis interaction. It involves interaction of the face of EC1 opposite Trp2 with the EC2-EC3 linker region of another molecule on the same cell surface. These cis and trans interactions generate a regular molecular 3D array, in which every EC1 domain forms both cis interactions with EC2-EC3 subdomains in parallel chains and trans interactions with the EC1 subdomain of anti-parallel chains (Boggon et al., 2002).

Desmocollins and desmogleins are cadherins whose extracellular domains have many features in common with classical cadherins. The interdomain Ca²⁺-binding sites are fully conserved in the desmocollins and desmoglein 2, and largely conserved in the other desmogleins. Furthermore, there is evidence to suggest that the so-called cell adhesion recognition (CAR) site regions that form part of the hydrophobic pockets involved in classical cadherin strand dimer formation are also involved in adhesion by desmosomal cadherins (Runswick et al., 2001; Tselepis et al., 1998). It is thus reasonable to suppose that the molecular mechanism of adhesive interactions of desmosomal cadherins is similar to that of classical cadherins, although no structural studies of desmosomal cadherins have been published.

We set out to determine whether modulation of desmosomal Ca²⁺ dependence is an in vivo phenomenon, and have made some striking discoveries. Modulation of desmosomal adhesion, identical to that found in tissue culture, takes place on wounding epidermis. Changes in the structure of the desmosomal adhesive material that accompany modulation of desmosomal adhesion lead us to suggest a model for desmosomal Ca²⁺ independence based on the cadherin crystal structure. Furthermore we propose a novel concept, hyper-adhesion, which is unique to desmosomes and explains, for the first time, why they are so fundamentally important in maintaining the strength and integrity of vertebrate tissues.

Two full-thickness wounds were made on the backs of male Balb-C mice aged 8-10 weeks according to published method (Tomlinson and Ferguson, 2003). After the times indicated in the text, figures and tables, the mice were killed and wound-edge material either frozen in liquid nitrogen for cryostat sectioning and immunofluorescence, or fixed and embedded for conventional transmission electron microscopy, or for ultra-thin cryosectioning and immuno-gold labelling. Control, unwounded skin was taken directly from freshly killed mice, without wounding.

Frozen sections (15-20 μm thick) were collected on poly-L-lysine-coated slides and fixed for 5 minutes in acetone at room temperature. After rinsing with phosphate-buffered saline, they were incubated in 5% normal donkey serum and 1% bovine serum albumen for 10 minutes, before rinsing in PBS and incubating with monoclonal antibody to desmoplakin (11-5F) (see Parrish et al., 1987) and rabbit anti-PKCα (Sigma). Secondary antibodies were FITC-conjugated donkey anti-rabbit IgG and Rhodamine Red donkey anti-mouse IgG (Jackson ImmunoResearch Laboratories). Immunofluorescence was examined using a Zeiss LSM510 confocal microscope.

Transmission and immuno-gold EM were carried out as described (North et al., 1999). The primary anti-PKC antibody used for the latter was as for immunofluorescence. The secondary antibody was 10-nm-gold-conjugated goat anti-rabbit IgG (British Biocell International).

Using electron micrographs printed at a total magnification of 46,000×, the distribution of PKCα with respect to the inner leaflet of the desmosomal plasma membrane was determined using two methods: manual measurement with a ruler and image analysis using the Scion Image Program (http://www.scioncorp.com). The same 48 desmosomes that were sectioned transversely were used for each determination. The desmosomes were labelled with the antibodies 11-5F to desmoplakin and anti-PKCα, using 20 nm gold particles for the former and 10 nm for the latter. The perpendicular distance of each gold particle from the inner leaflet of the plasma membrane was measured. Gold particles in the intercellular space were given a negative value. Measurements obtained manually were referred to as PKC1 (corresponding to PKCα) and DP1 (corresponding to desmoplakin), and those obtained by image analysis as PKC2 and DP2. The SIMFIT program (http://www.simfit.man.ac.uk), designed by Bill Bardsley (University of Manchester, UK) was used to analyse the data to determine whether (1) measurements obtained by the two techniques differ significantly; (2) the observations are scattered randomly or if there is evidence for significant clustering; and (3) the distributions can be quantified. For further details and results of this analysis, please refer to the authors.

The program SegMod (Levitt, 1992) was used to build a homology model for the sequences of Dsg2 and Dsc2 (Q14126, Q02487) using the C-cadherin ectodomain structure (1L3W.pdb) as a template. The initial models obtained were then refined according to the gradient minimisation routine in Xplor v3.8 (Brunger, 1987) to minimise van der Waals interactions. Following the minimisation, a 3D array of molecules was created based on the crystallographic cell and symmetry operations obtained in the 1L3W.pdb structure.

To confirm and extend our previous observation that desmosomes in normal adult mouse epidermis are Ca²⁺ independent (Wallis et al., 2000), small pieces of epidermis (1 mm³ or less) were dissected, placed in Ca²⁺-free tissue-culture medium containing 3 mM ethylene glycol bis(2-aminoethyl ether)-N,N,N′N′-tetraacetic acid (EGTA), and incubated for up to 6 hours at 37°C. Controls were incubated in normal medium. They were then examined by electron microscopy. All desmosomes in normal epidermis remained intact after such treatment (Fig. 1A, Table 1) and were therefore Ca²⁺ independent.

To examine wound-edge epithelial desmosomes, similar experiments were carried out with fragments from a zone from the edges of full-thickness wounds on the backs of mice. A majority of desmosomes (55% by 48 hours and 63% by 72 hours) in wound-edge epithelium became Ca²⁺ dependent; adhesion was lost by separation of the desmosomal halves within 1 hour of exposure of wound-edge epidermis to EGTA (Fig. 1B, Table 1). This change was first detectable close to the wound edge (within 10 cell diameters) by 24 hours post wounding and spread throughout the entire wound-edge zone up to at least 50 cell diameters (∼500 μm) by 48 hours. Thus desmosomes in wound-edge epidermis show similar changes to those previously reported in cultured sheets of MDCK cells (Wallis et al., 2000; Mattey and Garrod, 1986b).

Change to Ca²⁺ dependence was associated with a remarkable change in desmosome structure (Fig. 1C-F; Table 2). Desmosomes in unwounded epidermis viewed in transverse section in electron micrographs possessed a dense midline joined to the plasma membrane by alternating cross-bridges on either side of the midline (Fig. 1A,C, Fig. 5A). By contrast, the majority of desmosomes in wound-edge epithelium showed no evidence of such dense structures (Fig. 1D-F). We also measured the width of the intercellular spaces of 50 desmosomes with and without midlines. Those with midlines were 23.9±0.5 nm in width and those without 21.9±0.5 nm (P=0.05 by Student's t-test). In addition we measured the width of the intercellular space of desmosomes from normal epidermis that had been exposed to 3 mM EGTA for 6 hours. These showed no significant difference from desmosomes exposed to normal culture medium for the same time. These desmosomes also retained the midline structure following EGTA exposure (Fig. 1A). We conclude that desmosomes in wound-edge epidermis show a spontaneous change in the organisation of their adhesive material compared with normal desmosomes. Furthermore, the adhesive material of normal, Ca²⁺-independent desmosomes appears resistant to Ca²⁺ chelation.

We previously showed that PKCα became rapidly localised to the periphery of MDCK cells following wounding of confluent cultured cell sheets (Wallis et al., 2000). Furthermore, in unpublished observations, we showed that cell-peripheral PKCα was colocalised with desmosomal plaques. To determine whether similar changes take place in vivo, normal and wound-edge epidermis were examined by double-labelling immunofluorescence for PKCα and desmoplakin (a marker for desmosomes) with confocal microscopy, and by immuno-gold labelling of utra-thin cryosections.

PKCα was diffusely distributed in the cytoplasm principally of the basal cells of unwounded epidermis (Fig. 2A). On wounding, PKCα became colocalised with the desmosomal plaque, within 5 minutes at the extreme wound edge and by 48 hours 500 μm distant from the wound edge (Fig. 2A-F; Table 3A). Thus, in vivo as in culture, wounding epithelial cell sheets results in relocalisation of PKCα to desmosomal plaques.

We have previously mapped the locations of desmosomal component molecules within the desmosomal plaque by immuno-gold electron microscopy (North et al., 1999). To determine the distribution of PKCα in relation to other desmosomal plaque components, similar mapping was carried out using localisation of the C-terminus of desmoplakin with monoclonal antibody 11-5F as an internal control to check that no undue shrinkage had occurred and that our measurements were comparable with those we made previously (North et al., 1999).

Statistical analysis of the distribution of gold particles associated with anti-PKCα in the desmosomal plaque showed that they were not normally distributed because many particles were located very close to the plasma membrane. Rather, the analysis is consistent with the presence of a sharp peak of PKCα distribution ∼1 nm from the inner face of the plasma membrane, and another, more diffuse peak centred around 23 nm, that is within the outer dense plaque (Fig. 2G,H; Table 3B). Comparison of this distribution with our map of the desmosomal plaque suggests that PKCα located very close to the plasma membrane may have the cytoplasmic domains of the desmosomal cadherins or plakophilin as its phosphorylation targets. PKCα located within the outer dense plaque may have a wider group of possible targets including the cadherin cytoplasmic domains, plakoglobin, plakophilin and the N-terminus of desmoplakin.

To determine whether splitting of desmosomes occurred in wound-edge epithelium as part of the normal downregulatory process, the wound edge was examined by electron microscopy. By 48 hours after wounding all leading-wound-edge keratinocytes examined by electron microscopy showed substantial reduction in the numbers of desmosomes at their surfaces. No evidence for splitting of desmosomes as seen following EGTA treatment was found. Instead, 94 cells of 100 examined within ten cell diameters of the edges of 48-hour wounds showed whole intra-cytoplasmic desmosomes. Since `cells' in this case means cell profiles seen in ultra-thin sections, it is probable that all cells in this region of the wound edge contained internalised desmosomes. Intra-cytoplasmic desmosomes were also occasionally noticed further from the edge; of 100 cells within the region of 40-50 cells from the edge examined in 48-hour wounds only 4 showed internalised desmosomes. Some internalised desmosomes appeared to be attached to membrane fragments (Fig. 3C-F), or to vesicles in the cytoplasm (not shown), giving the characteristic `tennis-racket' shape reported by others (Caputo and Prandi, 1972). However, other internalised desmosomes appeared to have lost any associated membranous material, instead appearing free in the cytoplasm (Fig. 3B,F). We conclude that internalisation of entire desmosomes is the normal method of downregulation of desmosomal adhesion at the wound edge.

Several structural studies of classical cadherin fragments have shown that three Ca²⁺ ions bind at the linker regions between individual extracellular subdomains (Overduin et al., 1995; Shapiro et al., 1995). It is generally accepted that Ca²⁺ binding provides rigidity to the otherwise flexible linker region loops and accounts for the Ca²⁺ dependence of cadherin-mediated adhesion (Koch et al., 1997; Nagar et al., 1996; Pokutta et al., 1994). These studies indicate regular arrangements based on intramolecular and intermolecular interactions observed in the crystallographic cell, suggesting homophilic dimerisation of the cadherin extracellular domains in both cis (i.e. molecules on the same cell surface) and trans (i.e. molecules on opposite cell surfaces) (Boggon et al., 2002; Overduin et al., 1995; Shapiro et al., 1995). Trans interaction is clearly essential for adhesion and there is evidence to suggest that adhesion also requires cadherin clustering in cis (Brieher et al., 1996; Troyanovsky et al., 1999; Troyanovsky et al., 2003; Yap et al., 1997). Interestingly such regular structures, previously termed `adhesion zippers', are more readily comparable to the ultrastructure of desmosomes than adherens junctions (Lasky, 1995).

Homology models for Dsc2 and Dsg2 were generated using the C-cadherin ectodomain as a template. Modelling of the desmosomal cadherins (Dsc2 and Dsg2) in the crystal packing observed for C-cadherin shows that it is feasible to produce a similar array, despite the modest sequence identity between them (∼30%). However, some differences in detail between the intermolecular and intramolecular interactions in the desmosomal cadherins maybe expected. The final quality of the models for Dsc2 and Dsg2 was compared with the C-cadherin structure. The root-mean-square deviation for all equivalent carbons was 1.03 Å for Dsc2 and 1.04 Å for Dsg2.

In an attempt to model the Ca²⁺-independent desmosome structure we recreated a 3D array of Dsc2 molecules generated according to the crystallographic cell symmetry of C-cadherin. A similar array was generated for Dsg2. In these arrays, the adhesion interfaces are aligned to the x-axis (along the crystallographic x-axis) (Fig. 4 shows the Dsc2 array). In Fig. 4, the cylinder at the centre roughly represents the desmosomal interspace, containing the molecular stack. A front view (Z=0°) (Boggon et al., 2002), shows strand dimers formed in trans (trans interface) between monomers emanating from one cell surface (blue) and others emanating from the opposed cell surface (red). Rotation of the stack by 90° around the x-axis, shows a quadratic array (Fig. 4, top) with a repeating periodicity of 73.85 Å between rows of molecules, and a periodicity of 75.14 Å between layers. This array is remarkably similar to the en face view of the desmosome seen by lanthanum infiltration (Fig. 5), with a reported periodicity of 75 Å (Rayns et al., 1969). This quasi-crystalline arrangement was confirmed by digitisation and Fourier transformation of EM images (Rayns et al., 1969), revealing strong reflections characteristic of regularly ordered structures such as crystals (Fig. 5).

Rotation of the stack around the z-axis generates four different views: strand dimer 1 at z=0°, `boat' view at z=30°, strand dimer 2 at z=60° and `zipper' view at z=120° (Fig. 4, bottom). Each view occurs twice in a complete 360° rotation because of the crystallographic twofold symmetry axis aligned with the x-axis in Fig. 4. The `zipper' view shows both cis and trans interactions. All four views show a concentration of material in the midline, which correlates well with the midline observed by both conventional electron microscopy and freeze etching of desmosomes (Odland, 1958; Staehelin, 1974). Our comparative analysis of the 3D array for the Dsc2 model and the C-cadherin crystal with the EM data from desmosomes, shows evidence of a highly similar arrangement. Therefore, we propose that this 3D array is a good model for the highly ordered, quasi-crystalline desmosome structure observed in ultrastructural studies.

Ca²⁺-independent desmosomes retain midline structure after prolonged exposure to EGTA (Fig. 1). Measurement of the inter-membrane distance of 50 Ca²⁺-independent desmosomes after treatment with 3 mM EGTA for 6 hours revealed no significant difference from controls, indicating that the structure is unaffected. Since both calcium-dependent desmosomes and adherens junctions (also calcium dependent) lack this highly ordered arrangement of the adhesive material, it may be that some feature of the regular array seen in the crystal structure makes an essential contribution to the Ca²⁺ independence of desmosomes. Trans interactions between cadherin molecules are clearly required for adhesion and must be present in both Ca²⁺-dependent and Ca²⁺-independent junctions. The key feature that distinguishes the crystallographic array is the cis-interaction interface involving binding of the EC1 domain from one molecule to the linker region between the EC2-EC3 domains in an adjacent molecule (Boggon et al., 2002). Therefore, these cis interactions may contribute to Ca²⁺ independence. We suggest that the cis interactions may result in the retention of calcium within the regular array according to the following argument.

The linker regions between the EC1 to EC5 subdomains domain of cadherins are involved in binding three Ca²⁺ ions at each interspace. Two of the ions (Ca1, Ca2) are buried in the main core of the protein domain with an octahedral coordination whereas the third Ca²⁺ ion (Ca3) is more exposed to the solvent. In the EC2-EC3 linker region Ca1 is coordinated by seven groups in N215, N217, D246, D248, A254 and N304 (C-cadherin numbering) and Ca3 by six ligands from residues E119, E182, D213, D216 and D248 (Fig. 6B). Ca3 is coordinated by four ligand groups in E119, D180, E182 and D216, where two side chains, from E119 and E182 are shared between Ca2 and Ca3. These Ca²⁺-binding residues are fully conserved across all cadherins including the desmoglein and desmocollin. The different ligand coordination for the three positions correlates with marked differences in binding affinity for the three Ca²⁺ ions previously reported and with the notion that calcium binding restricts flexibility of the inter-domain region (Alattia et al., 1997; Koch et al., 1999; Koch et al., 1997; Pertz et al., 1999).

Analysis of the cis-interactions between EC1′ and EC2-EC3 (Fig. 6B) shows that the low-affinity Ca²⁺ ion (Ca3) is covered by the β-helix region of the EC1′ domain. This protection of the Ca3 binding site may limit its exchange with the solvent, and presumably its chelation by EGTA. According to this observation, we proposed the following model for the Ca²⁺-independent hyper-adhesion in desmosomes. If, as we suggest, the extracellular domains of the desmosomal cadherins are able to form a quasi-crystalline array in which cis interactions are maximised, all of the Ca3 molecules at the EC2-EC3 boundaries would be protected and the structure of these interfaces would be maintained throughout upon exposure to EGTA. The regular array of cis interfaces would be matched by an equally regular array of trans interfaces, which are maintained by hydrophobic interactions (Trp2 binding). The combined effect of these multiple interactions involving the EC1-EC3 subdomains would make possible a crystalline array structure that would provide strong, calcium-independent adhesion.

By contrast, in the calcium-dependent desmosome and the adherens junction, where the cadherin extracellular domains are less ordered, we propose that the cis interactions would be either more dynamic or fewer in number. Addition of EGTA would then result in removal of some of the Ca3 ions, and a domino effect would result in collapse of all extracellular domains and loss of adhesion.

The calcium ions at the other EC subdomain interfaces have different coordination; Ca3 in EC3-EC4 has only two ligands and Ca3 at EC4-EC5 has six. The question arises as to whether they would be removed by exposure to EGTA. We envisage two possible scenarios. Either the ordered structure retains at least some of the calcium ions at these interfaces (certainly Ca1 and Ca2, which are strongly coordinated), or that the majority of the other calcium ions are removed, but the combined multiple interactions at the cis and trans interfaces in the mid line are sufficient to maintain adhesion.

Our results show that the vast majority of desmosomes in normal epidermis are Ca²⁺ independent. Previous results, although not quantitative, indicated that this was also true of desmosomes in several other tissues, including trachea, oesophagus, tongue, liver and cardiac muscle (Wallis et al., 2000). We suggest that Ca²⁺ independence is the normal state for desmosomes in adult tissues in vivo.

We refer to the Ca²⁺-independent adhesive state of desmosomes as `hyper-adhesion' because it appears to represent a higher-affinity, more stable state of adhesive binding than that shown by either Ca²⁺-dependent desmosomes or other adhesive junctions such as adherens junctions. On wounding, when greater tissue lability is required to facilitate cell migration and wound repair, desmosomes spontaneously adopt a lower-affinity adhesive state. In doing so they appear to acquire characteristics that mimic those found in adherens junctions; they become Ca²⁺ dependent, their intercellular space narrows and they lose the highly organised structure of their adhesive material. A detailed consideration of the structure of adherens junctions is published elsewhere (Miyaguchi, 2000). Moreover, adherens junctions (and tight junctions) do not acquire Ca²⁺ independence (Wallis et al., 2000). Thus, we believe that hyper-adhesion represents a unique, more strongly adhesive state that can be adopted by desmosomes but not by other junctions. Furthermore we believe that hyper-adhesion can explain why desmosomes are so important in maintaining normal tissue architecture and function.

Clearly desmosomal hyper-adhesion is not the only reason for the strength of epithelial cell layers. Desmosomes and the intermediate filament cytoskeleton constitute a complex that forms a supporting scaffolding throughout an epithelium (Moll and Franke, 1982). When the intermediate filaments are disrupted, as in the human genetic disease epidermolysis bullosa and transgenic mice that mimic it, the epidermis is greatly weakened (Fuchs, 1996; Lane and McLean, 2004). Weakening or disrupting the interaction between desmosomes and intermediate filaments also diminishes the strength of epithelia (Huen et al., 2002; Russell et al., 2004). Scaffolding, like a chain, is only as strong as its weakest link. Thus desmosomal hyper-adhesion is crucial to the strength of the desmosome–intermediate-filament complex. We propose that the ability of desmosomes to adopt hyper-adhesiveness represents a specific evolutionary acquisition to enable them to fulfil their role in the complex.

A remarkable feature of desmosomes that was revealed by early ultrastructural studies is the apparent highly ordered arrangement of the intercellular material, characteristically exemplified by an electron-dense midline half way between the opposed plasma membranes (Odland, 1958). We have shown that this structure is present when desmosomes are in their Ca²⁺-independent, hyper-adhesive state, but absent from wound-edge epithelium when they become Ca²⁺ dependent and lose hyper-adhesiveness. This suggests that an organised structure of the adhesive material may be associated with high-affinity adhesion.

When this structure was infiltrated with the electron-dense tracer lanthanum, it appeared as a zigzag extending the full width of the junction and connected to the plasma membrane by alternating cross-bridges 30 Å in thickness (Rayns et al., 1969). In en face view, such infiltrates appeared as regular arrays of alternating white and black dotted lines with a periodicity of 75 Å. These observations suggest that the adhesive material of the desmosome has a quasi-crystalline arrangement. The crystal structures of the extracellular domains of classical cadherins, the adhesion molecules of adherens junctions, revealed such ordered structures that were referred to as `adhesion zippers' and which bore striking resemblance to desmosomes (Lasky, 1995; Shapiro et al., 1995). In fact, as we showed, the spacings calculated from the crystallographic array are strikingly similar to those found in ultrastructural studies of desmosomes. We also built homology models for Dsc2 and Dsg2, which are fully compatible with the 3D array reported for C-cadherin. This evidence prompted us to consider this array as a good model for the hyper-adhesive desmosome. Obviously, the situation in vivo could differ from the observations in the crystal, mainly because the cytoplasmic regions of cadherins and associated proteins undoubtedly play an important role in assembly and organisation of the desmosome.

Such a crystalline array does not seem to comply with previous observations of adherens junctions, which even in their most elaborate form, show limited and discontinuous organisation of their adhesive material (Miyaguchi, 2000). We predict that in this case, the contribution of the cytoplasmic region may have a totally different effect on junction assembly and adhesiveness that compromises a highly ordered organisation. Unfortunately, structural data on the full-length cadherin are not yet available.

Our attempt to model the Ca²⁺-independent desmosome from the crystal structure of C-cadherin produced two intriguing results. First, the periodicity shown in the en face views of the desmosome by electron microscopy and the C-cadherin crystal lattice are identical. Second, an arrangement of extracellular domains such as suggested by the C-cadherin lattice gives rise to a midline at no less than eight positions around the vertical axis. The appearance of the midline generated from such a structure might be expected to differ somewhat when viewed from different directions. It is not clear whether the preparation techniques used and the resolution achieved in conventional electron microscopy are adequate and sufficient to discriminate such differences.

A major consideration in modelling the desmosome was to attempt to seek an explanation for Ca²⁺-independent adhesion by Ca²⁺-dependent molecules. Our modelling was prompted by the consideration that Ca²⁺ independence seemed to be associated with ordered structure and Ca²⁺ dependence with lack of it. We show that Ca²⁺-independent desmosomes retain midline structure after prolonged exposure to EGTA and measurement of the inter-membrane distance of Ca²⁺-independent desmosomes after exposure to 3 mM EGTA for 6 hours revealed no significant difference from controls, indicating that the structure is unaffected. However, we do not know whether Ca²⁺ is removed from the desmosomes by such treatment, or is so tightly bound that it resists removal. Thus our model suggests that Ca²⁺ ions may be sequestered within the highly ordered, quasi-crystalline arrangement of the cadherin extracellular domains. The cis interactions between the molecules appear to be of key significance for this.

In the C-cadherin crystal structure the molecules are curved giving a predicted distance of 24.5 nm between the apposed plasma membranes of adhering cells (Boggon et al., 2002). This figure is very close to the intermembrane distance we measured for Ca²⁺-independent desmosomes. A recent study has examined desmosome structure by cryo-electron microscopy of frozen sections following high-pressure rapid freezing of the tissue (Al-Amoudi et al., 2004). This showed a highly organised arrangement of the desmosomal cadherins, but showed them to be straight and indicated an inter-membrane distance of 34 nm. The structure shown by Al-Amoudi et al. (Al-Amoudi et al., 2004) is in fact remarkably similar to the predicted structure of a desmosome shown by Miyaguchi (Miyaguchi, 2000). It is not clear how cis interactions of the desmosomal cadherin extracellular domains would be accommodated in such a structure. The desmosomal interspace may appear somewhat shrunken by conventional transmission electron microscopy (Al-Amoudi et al., 2004), but its regularity of structure appears to be maintained.

We suggest that conversion to Ca²⁺ dependence involves a conformational rearrangement of the desmosomal cadherin extracellular domains. In the Ca²⁺-dependent configuration they would be less organised and so would not give rise to the appearance of a midline. Adhesive binding would also be more easily disrupted. This may resemble the arrangement of classical cadherins in adherens junctions. A recent model of desmosome structure based on electron tomography of 2-day-old mouse epidermis suggested that the desmosomal cadherins have a disorganised, knotted arrangement in which cis interactions are compromised (He et al., 2003). We suggest that although trans interactions still occur in the Ca²⁺-dependent state, fewer cis-dimerisation interactions may result in a gel-like structure similar to that proposed by He et al. (He et al., 2003), with higher plasticity than the more crystalline 3D mesh observed for Ca²⁺-independent desmosomal structures. It may be that the desmosome chosen for analysis was Ca²⁺ dependent (He et al., 2003), but our unpublished observations show that the great majority of desmosomes in 2-day-old mouse epidermis are Ca²⁺ independent. We do not believe that the disorganised arrangement suggested (He et al., 2003) can account for the biological fact of Ca²⁺ independence.

A crucial component of this hypothesis is that the affinity of desmosomal adhesion is regulated by the phosphorylation of an intracellular component(s) in the desmosomal plaque. We suggest that phosphorylation of one or more desmosomal components by PKCα may generate a transmembrane signal that alters the conformation of the extracellular domains of the desmosomal cadherins, reducing both their affinity for Ca²⁺ and their adhesive binding. This causes a relaxation in the extracellular domain so that the quasi-crystalline structure, exemplified by the midline, is lost. Although the midline disappears, adhesion is still maintained, for we have found no evidence for separation of desmosomal halves in wound-edge epithelium, except following the artificial exposure to EGTA. Recent evidence has shown that plakophilin 1, an armadillo-family protein and component of the desmosomal plaque, plays a role in regulating the Ca²⁺ independence of desmosomes (South et al., 2003). This supports our view that transmembrane effects are involved in regulating desmosomal adhesiveness.

It is well established for other adhesion molecules, especially integrins, that signals generated within the cytoplasm modulate the structure and binding affinity of the extracellular domains (Hynes et al., 2002; Liddington and Ginsberg, 2002). This is referred to as `inside-out' signalling. We postulate that a cytoplasmic signal involving PKCα generates a similar inside-out signal in wound-edge epithelial cells, and that this modulates both the structure and binding affinity of desmosomal cadherins, converting them from Ca²⁺ independent to Ca²⁺ dependent.

Modulation or downregulation of desmosomes in wound-edge keratinocytes has been reported previously (Croft and Tarin, 1970). Our observations suggest that this downregulation is substantial and that it takes place by internalisation of whole desmosomes rather than by splitting into half desmosomes or by desmosome disassembly. To our knowledge this is a novel observation for wound-edge epithelium, although intracytoplasmic desmosomes have been reported in a variety of different skin lesions (Caputo and Prandi, 1972; Klingmuller et al., 1970; Klug and Haustein, 1974; Komura and Watanabe, 1975; Proctor and Sherman, 1975; Schenk, 1975; Schenk, 1980; Takaki et al., 1971; Watanabe et al., 1977) and in the trophoblast (Firth et al., 1980).

Internalisation of whole desmosomes was suggested as a mechanism for downregulation of desmosomes in the epidermis (Allen and Potten, 1975). The desmosomes appeared to be engulfed in a double-membrane vesicle derived from the two adjacent cells that had formed the desmosomes. Some of the internalised desmosomes that we have observed at the wound edge had membrane associated with them, but others seemed to have no membrane. We have no explanation for this apparently common occurrence.

Our suggestion and summary of the sequence of events that accompanies desmosomal downregulation is shown in Fig. 7. What is the role of PKCα in this event? We note that PKC signalling has been shown to be associated with endocytosis and phagocytosis in a number of different systems (Hirai, 2001; Lennartz, 1999; Liu and Anand, 2001; Scaife and Margolis, 1997; Wright et al., 1997) and therefore suggest that PKC signalling in some way primes desmosomes for internalisation by a process akin to phagocytosis. Thus primed, the desmosomes remain weakly adhesive in the wound-edge epithelium until a further signal associated with the onset of cell motility promotes their internalisation. Work in our laboratory currently aims to discover the mechanism of this process. Most importantly, our work has shown how desmosomes ensure a stable network of cells in a physically stressed environment.

We thank the Medical Research Council for financial support and the Egyptian Government for support to M.Y.B. We also thank Jordi Bella, Anita Merritt, Martin Humphries, Charles Streuli, Zhuxiang Nie, Nigel Hodson, Gillian Ashcroft and Diane Silla for valuable help and discussions during the preparation of the manuscript.