Molecular map of the desmosomal plaque

Desmosomes are symmetrical, disc-shaped intercellular junctions found primarily in epithelial tissues. They perform dual roles in mediating adhesion between cells and in linking the intermediate filaments (IF) of one cell to those of its neighbour, thereby establishing an integrated scaffold across the entire epithelium. The importance of this scaffold is reflected by the existence of severe skin blistering diseases, such as the autoimmune pemphigus disorders that target the desmosomal cadherins (Amagai, 1995), and by recently described inherited disorders caused by mutations in desmosomal genes (Armstrong et al., 1999; McGrath et al., 1997). The ultrastructural organisation of desmosomes has been described in detail (Burdett, 1998; McNutt and Weinstein, 1973; Staehelin, 1974) and their principal molecular components have been well characterised (Buxton and Magee, 1992; Garrod, 1993; Green and Jones, 1996; Koch and Franke,

1994), enabling detailed investigation of the molecular interactions that lead to IF attachment and adhesion. Biochemical analysis and in vitro binding or transfection studies employing mutant constructs have revealed much about interactions between specific domains of the principal desmosomal components (see Discussion and references therein). However, it is also crucial to determine the precise locations of the different components within the native structure, in order to relate structure and function. This study set out to provide such complementary data, and thereby to test specific predictions that have emerged from molecular studies, by performing ultrastructural localisation using domain-specific antibodies against all of the principal desmosomal components.

Desmosomes are readily identified in transmission electron micrographs of conventional thin sections by their characteristic ultrastructural appearance (see Fig. 1 for a schematic diagram). They consist of two principal domains: (1) the extracellular core domain (ECD) or ‘desmoglea’, ∽30 nm wide and bisected by an electron-dense mid-line region; (2) symmetrical dense cytoplasmic plaques lying parallel to the plasma membrane and separated from it by a less dense zone. Each plaque is commonly described as consisting of two regions, the outer dense plaque (ODP), 15-20 nm thick, and separated by an 8 nm electron-lucent zone from a slightly less dense inner plaque (inner dense plaque; IDP), into which the IF are seen to insert. The region between the inner face of the ODP and the IF domain has also been referred to as the ‘satellite region’ (Garrod et al., 1990; Miller et al., 1987). Differing values have been reported regarding the width of these domains, particularly the IDP (Garrod et al., 1990; Nilles et al., 1991; Steinberg et al., 1987). The orientation of IF at desmosomes is also unclear, although in classical thin sections IF were reported to converge towards the plaque and then loop away from it at a distance of 20-40 nm from the cell membrane (Fawcett, 1961; Kelly, 1966). Further ultrastructural details have been revealed by alternative methods of tissue preparation. In the ECD, lanthanum infiltration exposed staggered quadratic arrays of side-arms linking the dense mid-line to the plasma membrane (Rayns et al., 1969). In the cytoplasmic plaque, a population of 4-5 nm wide ‘traversing filaments’ (TF) which intervene between the IF loops and the desmosomal membranes were visualised by freeze-fracture EM (Kelly and Kuda, 1981; Leloup et al., 1979; McNutt and Weinstein, 1973).

The ECD of the desmosome is largely composed of the

extracellular domains of desmocollins (Dsc) and desmogleins (Dsg), two families of transmembrane glycoproteins belonging to the cadherin superfamily of cell-cell adhesion molecules (Buxton and Magee, 1992; Koch and Franke, 1994). Both Dsc and Dsg occur as three related proteins, products of distinct genes, that show graded and overlapping distributions across the different cell layers of epidermis (Arnemann et al., 1993; Legan et al., 1994; North et al., 1996; Shimizu et al., 1995). Each Dsc is additionally subject to alternative splicing, resulting in the ‘a’ and ‘b’ forms, which differ in the length of their cytoplasmic tail (Collins et al., 1991; Parker et al., 1991). The cytoplasmic dense plaque is composed of the cytoplasmic tails of the glycoproteins and a number of cytoplasmic components, including desmoplakins (DP) I and II, plakoglobin (PG) and plakophilins (PP) (reviewed by Cowin and Burke, 1996). DPI and DPII are also produced by alternative splicing: both proteins consist of an α-helical coiled-coil rod domain between globular amino (N)- and carboxy (C)-termini, DPII lacking most of the rod domain (Green et al., 1990). Desmoplakins are constitutive desmosomal components, although DPII is absent from cardiac muscle tissue (Angst et al., 1990), and their critical role in epidermal integrity is demonstrated by the striate subtype of palmoplantar keratoderma caused by DP haploinsufficiency (Armstrong et al., 1999). DPI is believed to occur as homodimers or higher order filamentous structures in desmosomes, but whether by parallel or antiparallel aggregation remains unclear (Bornslaeger et al., 1994). PG and PP belong to the armadillo gene family of signalling proteins (Peifer et al., 1992), with distinct N- and C-termini flanking a number of central arm repeats. PG contains 13 arm repeats, while PP comprise a separate subclass of armadillo proteins, each including 9 arm repeats preceded by an N-terminal head domain of variable length (Hatzfeld et al., 1994; Heid et al., 1994; Riggleman et al., 1989). PG is notable as the only common component of desmosomes and adherens junctions (Cowin et al., 1986). PP1 is localised in the desmosomal plaques of certain stratified and complex epithelia (Heid et al., 1994; Kapprell et al., 1988, 1990), PP2 in the desmosomes of a wide range of cell types (Mertens et al., 1996) and PP3 in the desmosomes of most simple and almost all stratified epithelia as well as cell lines derived from these tissues (Bonné et al., 1999; Schmidt et al., 1999). Additional minor components, such as IFAP 300 (Skalli et al., 1994), pinin (Ouyang and Sugrue, 1996; but see also Brandner et al., 1997), desmocalmin (Tsukita and Tsukita, 1985), plectin (Eger et al., 1997), envoplakin and periplakin (Ruhrberg and Watt, 1997) may also contribute to plaque structure.

Immunoelectron microscopy has been used previously to assign the major components to extracellular or intracellular desmosomal domains (Miller et al., 1987; Steinberg et al., 1987). These studies revealed important information, but largely employed polyclonal antisera raised against whole molecules or monoclonal antibodies of unknown epitope specificity. Many further immunoEM studies have indicated the approximate locations of components within the desmosomal ultrastructure, but precise localisations could not be achieved without a quantitative approach. Here we use antibodies of known specificity directed against the ends of the major desmosomal components, in conjunction with quantitative immunogold EM, to construct a molecular map of the cytoplasmic plaque of desmosomes in bovine nasal epidermis. Important information concerning the relative locations and the structural organisations of the component proteins has been gleaned. The relevance of previous in vitro or transfection data to desmosome organisation has in some cases been confirmed and in others brought into question. This map will aid the interpretation of new results and allow future studies to be focussed on potentially important interactions.

Where possible, antibodies against the extreme termini of each protein were obtained. However the two anti-plakophilin antibodies were specific only to one half of the molecule, rather than to the actual protein termini (see Discussion). Desmoplakin N terminus (NW161): rabbit antiserum against bovine DP, residues 1-189, expressed as a His-tag fusion protein (Bornslaeger et al., 1996) (used at 1:250 on sections and 1:2500 on blots). Desmoplakin C terminus (11-5F): mouse monoclonal IgG against bovine DP (Parrish et al., 1987). The epitope was mapped using deletion constructs to between residues 2019-2803 within the C terminus (T. Stappenbeck and K. Green, unpublished data) (used at 1:20 on sections and 1:500 on blots). Desmocollin ‘a’ form C terminus: rabbit antiserum against mouse Dsc2a C-terminal residues 838-904, expressed as a GST fusion protein (J. Hyam and D. R. Garrod, unpublished) (affinity purified IgG used at 650 μg/ml on sections and 40 μg/ml on blots). Desmocollin ‘b’ form C terminus: rabbit antiserum against a peptide comprising the last 11 residues of human Dsc2b ESIRGHTLIKN (KLH conjugate) (A. I. Magee and B. Trinnaman, unpublished) (used at 1:40 on sections and 1:500 on blots). Desmoglein C terminus (#10): rabbit antiserum against human Dsg3-specific C-terminal peptide residues LCTEDPCSRLI (KLH conjugate) (A. I. Magee and B. Trinnaman, unpublished) (used at 1:100 on sections and 1:1000 on blots). Plakoglobin N terminus (#2008): rabbit antiserum against Xenopus PG residues 1-106, expressed as a GST fusion protein (H. C. Cordingley and A. I. Magee; Cordingley, 1996) (used at 1:100 on sections and 1:1000 on blots). Plakoglobin C terminus (#1): rabbit antiserum against Xenopus PG residues 666-738, expressed as a fusion protein (H. C. Cordingley and A. I. Magee; Cordingley, 1996) (used at 1:100 on sections and 1:1000 on blots). Plakophilin N-region: rabbit antiserum against human PP1, residues 1-285, expressed as a His-tag fusion protein (Kowalczyk et al., 1999; M. Hatzfeld, manuscript in preparation) (used at 1:50 on sections and 1:500 on blots). Plakophilin C-region: rabbit antiserum against human PP1, residues 286 – 726, expressed as a His-tag fusion protein (Kowalczyk et al., 1999; M. Hatzfeld, manuscript in preparation) (used at 1:50 on sections and 1:500 on blots). Both PP antibodies are known to be specific for the PP1 isoform (M. Hatzfeld, unpublished data).

Epidermis was dissected from bovine nose, frozen in liquid nitrogen and fractured into powder using a pestle and mortar while still frozen. The powder was thawed into Laemmli sample buffer (Bio-Rad), homogenized, boiled for 5 minutes, and spun for 5 minutes at 10,000 g in a bench centrifuge. SDS-PAGE was performed as described (Laemmli, 1970), using an 8% gel for subsequent immunoblotting with PG and PP antibodies and a 6% gel for DP, Dsc and Dsg antibodies. Proteins were transferred onto a nitrocellulose membrane, which was then blocked with 5% skimmed milk powder before primary antibody incubation. Bound antibodies were detected with an appropriate peroxidase-conjugated secondary antibody and an ECL detection system (Amersham).

Preparation of conventional and polyvinyl alcohol (PVA)-embedded samples for ultrastructural studies[Freshly obtained bovine nasal epidermis was dissected into small pieces (approximately 1 х 1 х 0.5 mm) and fixed by immersion in 2% formaldehyde (FA) (freshly made from paraformaldehyde powder) plus 2% glutaraldehyde (GA) in 0.1 M sodium cacodylate buffer, Ph 7.3. Tissue was fixed for 2 hours at room temperature (RT), then taken through 4 washes of cacodylate buffer. Samples for resin embedding were post-fixed for 2 hours with 1% osmium tetroxide, dehydrated through an ethanol series and embedded in Spurr’s resin. Samples to be embedded in PVA were immersed in 20% aqueous PVA, M. Wt. 10,000 (Air Products and Chemicals Inc., Allentown, Pennsylvania) and hardened by drying overnight at 60°C. Sectioning of PVA blocks was performed as described (Small et al., 1986).

Resin sections were contrasted using uranyl acetate and lead citrate. PVA sections were incubated on aqueous buffer to extract the PVA from the section, rinsed with 40 μg/ml aqueous bacitracin (Sigma), and contrasted by negative staining using 2% aqueous uranyl acetate.

Preparation of ultrathin cryosections for immunoEM

Bovine nasal epidermis was dissected as above and fixed in 2% FA in 200 mM Hepes buffer, pH 7.4 (Hepes buffer) for 1 hour at RT. Samples were washed in 4 changes of Hepes buffer, 10 minutes in each, and then infiltrated with a sucrose/polyvinyl pyrrolidone mixture (Tokuyasu, 1989) for a minimum of 2 hours at RT. Infiltrated specimens were plunge-frozen and stored in liquid nitrogen.

Cryosectioning was performed using a Leica Ultracut S/FCS cryoultramicrotome following the method of Tokuyasu (1980). Ultrathin sections were cut using tungsten-coated glass knives (Roberts, 1975) at temperatures around _100°C. Sections were retrieved on a droplet of 2 M sucrose plus 0.75% gelatin (in Hepes buffer) and transferred to Formvar-coated grids. Grids were inverted onto 2% gelatin/PBS (solidified) and stored overnight at 4°C.

Immediately prior to immunolabelling, cryosections on 2% gelatin plates were incubated at 37°C to fluidify the gelatin and then for a further 10 minutes to block non-specific labelling. All other steps of the labelling procedure were carried out at RT. Grids were transferred across 3 droplets of 0.02 M glycine/PBS (10 minutes total), then blocked for 15 minutes with 5% normal goat serum (NGS) plus 1% BSA. Sections were transferred to primary antibodies (diluted into 1% BSA/PBS to give optimal staining) for 1-2 hours, then washed 5 times in 0.1% BSA/PBS (total 10 minutes). Goat anti-mouse or anti-rabbit IgG gold conjugates (BioCell Research Laboratories, Cardiff) were diluted 1 in 15 in 10% NGS, 1% BSA/PBS, and incubated for 30 minutes on ice before use. Gold labelling was performed for 30 minutes, the sections washed 5 times in PBS (total 20 minutes), and sections were then post-fixed using 2.5% GA/PBS for 10 minutes. After 4 washes in ddH₂O (1 minute each), sections were contrasted for 5 minutes on 2% uranyl acetate oxalate (Tokuyasu, 1980), washed briefly 3 times in ddH₂O, then incubated on 2% aqueous PVA plus 0.2% uranyl acetate (Tokuyasu, 1989). Grids were looped out and drawn across filter paper to remove excess PVA, air-dried, and examined under a Phillips 400 transmission electron microscope. All micrographs were taken at a magnification of 46,000.

30 desmosomes, which had been sectioned in a plane normal to the plasma membrane, were selected for each antibody. In the case of the Dsg3 C-terminal antibody, desmosomes were selected from the basal

layers and lower spinous layers, as the labelling for this Dsg is strongest in this region. Desmosomes from the mid-spinous region were selected for all other antibodies. Electron micrographs were printed at high magnifications and the perpendicular distance of each gold particle from the cytoplasmic surface of the plasma membrane was measured to the nearest millimetre on print. For this purpose, desmosomes were divided into two symmetrical halves and the distance of each gold particle measured from the nearer membrane. Particles lying over the extracellular region were assigned a negative value. To minimize variation and bias, all measurements were performed blind by the same author and the print magnification for each set was calculated only after all measurements had been completed. The number of gold particles was plotted against distance (in whole mm on print) using the statistical package SIMFIT (Bardsley et al., 1995a,b). For all epitopes the distribution of gold particles approximated to a normal distribution, and a best fit curve was fitted using SIMFIT. Distances were converted back to nanometres (nm) on section (from total magnification) and the mean distance from the plasma membrane was obtained from the curve parameters.

Immunogold labelling was typically performed as double labelling, using the antibody of choice together with 11-5F (anti-DP C terminus). This allowed the distance calculated for the DP C terminus to be compared between data sets, thereby assessing the reproducibility of the results. Three of the many data sets obtained using antibody 11-5F are presented below.

Immunoblots (Fig. 2) demonstrated that the antibodies used in this study detected proteins of the appropriate molecular mass and antibodies against the N- and C-termini of the same protein recognized bands of the same size. The anti-Dsg 3 antibody stained a single band of around 135 kDa, confirming that it did not react with the larger Dsg 1 or 2. The slight doublets obtained with the Dsc antibodies probably reflected weak cross-reactivity with other Dsc isoforms, but there was clearly no cross-reactivity between the two splice variants.

It should be noted that throughout the Results and Figures, DP refers to DP I and II, PP to PP1, and Dsg to Dsg 3.

The ultrastructural appearance of desmosomes was dependent on the chosen method of tissue preparation and section contrasting, yet all preparations revealed a highly ordered arrangement of components. Fig. 3A shows a typical desmosome from the spinous layers of bovine nasal epidermis prepared by a conventional method (see legend). The inner face of the ODP was located 15 to 20 nm from the plasma membrane, and was separated from the IDP by an electron-lucent zone less than 10 nm wide. The width of the more diffuse IDP was more difficult to determine but was in the region of 15 to 20 nm. Thus the total width of the desmosomal plaque was up to 50 nm. In classical images from tissues such as newt epidermis, the ODP, IDP and regions of less electron density on either side are clearly revealed, as are individual tonofilaments converging upon the plaque region at a distance of 40-70 nm from the plasma membrane (Kelly, 1966). However, in bovine nasal epidermis a space between the plasma membrane and the plaque was hardly discernible and clearly defined IF could not be distinguished close to the plaque, consistent with previous studies (Leloup et al., 1979).

More detail was revealed within the negatively-stained cytoplasmic plaque of PVA-embedded desmosomes. The highly ordered arrangement of desmosomal components was observed as lines of differential staining both parallel and perpendicular to the membrane (Fig. 3B). At high magnifications perpendicular fine filaments could be distinguished across the ODP, in some regions appearing as two parallel arrays (Fig. 3C). These filaments may correspond to the 4-5 nm TF previously reported to extend from the intermediate filaments to the desmosomal membrane (Kelly and Kuda, 1981; Kelly and Shienvold, 1976; Leloup et al., 1979; McNutt and Weinstein, 1973).

The structure of desmosomes in ultrathin cryosections of PFA-fixed bovine nasal epidermis has been described previously (Miller et al., 1987). Consistent with this, few structural features could be discerned within the low contrast images (Fig. 3D). The two plasma membranes were clearly visible as well as a cytoplasmic lamina parallel to each. Its inner face, situated around 20 nm from the membrane, was presumed to mark the extent of the ODP. Regular periodic striations across this lamina (Miller et al., 1987) were occasionally seen in optimally contrasted desmosomes (not shown).

On ultrathin cryosections all antibodies labelled desmosomes strongly, with negligible labelling across the rest of the tissue (Fig. 3F). Control sections labelled with gold conjugates alone showed minimal background labelling (Fig. 3D and E). Fig. 4 shows representative desmosomes labelled with each antibody. Gold particles were largely confined to a band of label, which varied in distance from the plasma membrane for different antibodies. To assess the reproducibility of the measurements most sections were double-labelled with the antibody under investigation together with antibody 11-5F against the DP C terminus (Fig. 4A-F). The size of secondary gold conjugate used did not affect the particle distributions (not shown). Although all of the detected epitopes were cytoplasmic, gold particles could also be seen in the extracellular domain using certain antibodies, in particular those directed against PP1 (see Discussion).

The distribution of gold particles obtained using each antibody approximated a normal distribution curve. Fig. 5 depicts the data for each antibody plotted together with a best fit curve. The mean gold particle distance from the plasma membrane was estimated for each antibody from the best fit curve (Table 1), and the results were used to construct a preliminary map of the desmosomal plaque (Fig. 6).

The standard error of the mean and standard deviation were also estimated using the best fit curve (Table 1). The standard deviations, reflecting the spread of gold particles within an individual data set (Table 1), were similar for all antibodies. The ‘tail’ which was apparent on the right hand side of each graph was probably contributed by non-specific labelling. A symmetrical tail on the left side of each graph would not have been detected because the labelling at this position would have

been attributed to the other half of the desmosome and thus measured as a positive value from the other membrane leaflet

In repeat experiments the same pattern of results (that is, order of increasing distance from the membrane) was obtained. However, some variability in the mean distances was found between different tissue specimens. Therefore the results presented here were all obtained from the same specimen. To test for variability between experiments, several sets of data were obtained for the DP C terminus (Fig. 4A-C). The mean values and 99% confidence limits of the three data sets presented here were estimated to be 52.0±1.28 nm, 51.4±2.84 nm and 49.5±1.60 nm. Thus no statistically significant differences were found between them at the 99% confidencelevel, since these confidence limits overlap.

The length of DP was calculated from the distance between the C terminus (52 nm; first data set) and the N terminus (10.3 nm; measured on the same desmosomes as the C terminus using double labelling). This gave a length of ∽42 nm, measured perpendicular to the plasma membrane.

The N terminus of each of the desmosomal glycoproteins is situated in the extracellular core domain: therefore only the C terminus was considered here.

Localisation of the terminal domains of all major desmosomal components has enabled construction of a rudimentary molecular map of the cytoplasmic plaque of desmosomes in bovine nasal epidermis (Fig. 6). The results have important implications for future investigations into molecular interactions between desmosomal components. A number of specific predictions based on previous biochemical and molecular studies have been confirmed or brought into question. Before expanding on these major advances in our understanding of structure-function relationships in desmosomes, several issues, which might impact the interpretation of our results, should be considered.

First, in localising the terminal domains of the molecules, we have shown that no major desmosomal component is aligned parallel to the plasma membrane along its entire length. This is in keeping with ultrastructural features such as the TF (Kelly and Kuda, 1981; Leloup et al., 1979; McNutt and Weinstein, 1973), which have been visualised perpendicular to the membrane, and with the perpendicular filamentous structures seen here in PVA-embedded tissue (Fig. 3C). However this study could not provide information concerning the position of internal regions of each protein. Thus other domains of certain components may be positioned closer to or further from the membrane than the N- and C-termini.

Second, this study has demonstrated the benefit of the immunogold technique in resolving the position of epitopes located only a few nm apart, perpendicular to the membrane. However, higher resolution methods will be required to investigate the lateral organisation of the plaque.

Third, the calculated distances varied between different tissue specimens, and therefore cannot be taken as absolute. This variability might be caused by different degrees of fixation-induced tissue

shrinkage brought about by slight changes in the duration or temperature of fixation. However, the important points to emphasise are that the differences between repeated experiments on one specimen were not statistically significant, and that the relative order of the different protein domains and thus the potential overlap in spatial distribution of particular components remained constant.

The close fit of each gold particle distribution to a single normal distribution curve suggested that each of the detected epitopes was localised in a single discrete band parallel to the plasma membrane. The standard deviations were consistent with the use of an indirect labelling protocol, after which a gold particle could lie up to the length of two antibodies (around 20 nm) in either direction from its bound epitope (see Griffiths, 1993). This localisation of gold particles far from their bound epitope also seems to be the likely explanation for the apparent extracellular signal observed for antigens close to the plasma membrane, such as PP1. Although a direct labelling method would have resulted in tighter distributions, the reduced number of bound gold particles would have rendered statistical analysis very difficult for the weaker antibodies. Moreover it is probable that the mean position of the gold particles would be unaltered.

Fig. 6 shows a rudimentary molecular map of the desmosomal plaque, on which the approximate position of each protein terminal domain has been marked relative to the plasma membrane. This detailed analysis of the desmosomal component locations has led to a greater understanding of plaque ultrastructure.

The electron density of the ODP (Fig. 3A) can be explained by the high concentration of proteins in this region. The positions of the PG N terminus and the Dsg3 C terminus (around 20 nm from the membrane) were coincident with the electron-opaque lamina observed on cryosections that appears to mark the distal limit of the plaque. In contrast, the PP1 C terminus was located near to the membrane proximal face of the ODP. Thus the ODP corresponds to the region where PG and PP1 might interact with the cytoplasmic tails of the desmosomal glycoproteins. DP spans both the IDP and most of the ODP, the region harbouring its C-terminal domain marking the innermost face of the IDP. The TF, previously visualised between the ODP and the IF attachment zone (Kelly and Kuda, 1981; Leloup et al., 1979; McNutt and Weinstein, 1973), are most probably comprised of DP, possibly together with the most C-terminal domains of

Dsg1 and Dsg2, and/or with unravelled keratin protofilaments (see below).

The DP N terminus was localised in the outer plaque around

10 nm from the plasma membrane, in keeping with its proposed role in attachment to the plaque (Bornslaeger et al., 1996; Smith and Fuchs, 1998; Stappenbeck et al., 1993). In contrast, the C terminus lay much further from the membrane, consistent with its function in IF attachment (Bornslaeger et al., 1996; Kouklis et al., 1994; Meng et al., 1997; Stappenbeck et al., 1993; Stappenbeck and Green, 1992). Contrasting reports from previous immunogold labelling studies reported DP to extend from the cytoplasmic face of the plasma membrane to the inner boundary of the satellite region (Steinberg et al., 1987) or to be a component of the satellite region only (Miller et al., 1987). Our results demonstrate that DP extends across most of the outer plaque and the whole of the inner plaque and thus is in a position to interact with most desmosomal components.

A single peak of labelling was obtained with both anti-DP antibodies. Thus all the N-termini of DP molecules are located at the same distance from the plasma membrane, as are all the C-termini. This indicates that if DP self-aggregates to form homodimers, tetramers or higher- ordered filamentous structures, it must do so in a parallel fashion. This result is consistent with the demonstration that DP can self-associate through head-head interactions (Smith and Fuchs, 1998). However, this observation differs from the situation recently described for the related plakin family member plectin, which has been proposed to exist in the form of both parallel and antiparallel oligomers within hemidesmosomes (Rezniczek et al., 1998).

DPI and DPII are predicted to be of very different lengths. The length of the rod domain was predicted to be 130 nm for DPI and 43 nm for DPII, both from sequence data (Green et al., 1990) and from rotary shadowing of isolated molecules (O’Keefe et al., 1989). Both DPI and DPII are present in this tissue in approximately equal amounts, and the antibodies were shown to detect both isoforms (Fig. 2). Therefore a double peak of labelling might be expected using either antibody. Yet we found that the C-termini and N-termini of both DPI and DPII were located in a single region respectively and the total length of DP, calculated perpendicular to the plasma membrane, was around 42 nm. This figure is thus consistent with the predicted length of DPII, but not of DPI, suggesting that DPI is either oriented at an angle to the membrane or is somehow folded or coiled in native tissue. Several possible break points and stutters along the length of the rod domain, together with an extended region of predicted flexibility (Green et al., 1990), may allow the molecule to bend. Fig. 7 shows three of the many possible configurations of DPI that would fit our data and reconcile our results with those of O’Keefe et al. (1989). However we consider model A to be unlikely as it would be inconsistent with the reported perpendicular appearance of the TF.

The functional significance of the different lengths of DP I and II is unknown. Here we found that the distance of the DP C terminus from the membrane (around 51 nm) was entirely consistent with the total width of the entire desmosomal plaque measured on frozen sections by Miller et al. (1987), that is a 17 nm outer plaque plus a 34 nm satellite zone. These observations are consistent with the idea that the width of the desmosomal plaque is defined by the extent of DP. It is therefore conceivable that DPII determines the extent of the inner plaque, while lateral associations between the longer rod domains of DPI could contribute to the structure of both plaques (model B) or the IDP alone (model C). Since cardiac muscle desmosomes contain only the DPI isoform we were interested to determine whether the C-terminal antibody would be localised further from the membrane in this tissue. However, insufficient gold labelling could be achieved on these considerably smaller desmosomes to permit statistical analysis.

The C terminus of Dsg3 was located further from the membrane than that of Dsca or Dscb, consistent with their molecular sizes and the predictions of Miller et al. (1987). We found the Dsg3 C terminus to be located at the inner face of the ODP, suggesting that the entire Dsg3 cytoplasmic tail may be folded within the ODP. Thus Dsg3 is unlikely to participate significantly in direct IF binding, unless keratins penetrate into the ODP in the form of unravelled protofilaments (see below). However, Dsg3 is the smallest of the three Dsg isoforms: its C III domain comprises only 2 of the unique 29-amino acid repeats, compared to 5 in Dsg1 and 6 in Dsg2, and it lacks the glycine-rich C-terminal cytoplasmic domain (Amagai et al., 1991). Thus it is still possible that the longer cytoplasmic domains of Dsg1 and 2 could span the region between the ODP and IDP and even extend through the IDP to interact with IF, as proposed from sequence analysis (Nilles et al., 1991) and rotary shadowing studies (Rutman et al., 1994). Since no antibodies against the extreme C terminus of Dsg1 or Dsg2 have been raised successfully, we were unable to address this point.

Our results concerning the Dsc ‘a’ and ‘b’ forms are consistent with previous studies using transfected cells and in vitro binding assays. The ‘a’ form of Dsc1 has been shown to bind both PG and DP and to recruit them in transfected cells to form a plaque (Chitaev et al., 1996; Smith and Fuchs, 1998; Troyanovsky et al., 1993, 1994; Witcher et al., 1996).

Consistent with these results we found that Dsca, PG and DP all overlap spatially. Conversely, we found that the ‘b’ form of Dsc overlapped with neither PG nor DP (unless with internal domains of these proteins: see following section), consistent with the demonstrated inability of Dsc1b to nucleate plaque formation in transfected cells (Troyanovsky et al., 1993). At present the function of the ‘b’ form of Dsc remains unclear. PP1, for which strong in vitro association with the ‘a’ form of Dsc1 has been reported (Smith and Fuchs, 1998), was found to overlap also with the ‘b’ form. The possibility that Dscb contains a PP1-binding domain should therefore be investigated. Alternatively the clue to the function of the shorter ‘b’ form may lie in its apparent inability to recruit cytoplasmic plaque proteins. The extremely high concentration of proteins localised within the outer plaque is reflected by the marked electron density of this region under the EM. It is possible that the density of Dsc extracellular domains required for adhesion is greater than the density of Dsc plaque component-binding domains that is optimal for plaque assembly.

The localisation of PG is in keeping with its orientation and interactions predicted from biochemical and transfection studies (reviewed by Cowin and Burke, 1996). Thus the C terminus was localised nearer to the plasma membrane than the N terminus and the protein overlapped with its known binding partners Dsca, Dsg and DP, as well as with PP1. PG is believed to bind to the C-domains of the desmosomal glycoproteins via a binding site localised to the first 3 arm repeats (Chitaev et al., 1998; Troyanovsky et al., 1996; Wahl et al., 1996; Witcher et al., 1996). Our results would not be compatible with these interactions if PG were an elongated molecule orientated perpendicular to the membrane. However, it is more probable that PG is globular (Kapprell et al., 1987; Ruediger et al., 1994), folded such that the internal arm repeats are positioned closer to the plasma membrane than either its C- or N terminus (as depicted by Cowin and Burke, 1996). This orientation is also consistent with the binding of central PG domains to the DP N terminus (Kowalczyk et al., 1997, 1998). Thus we cannot exclude the possibility that the arm repeat domain would also be available for binding to Dscb.

An important point to emerge from this study is the localisation of PP1. This plaque component (Kapprell et al., 1988, 1990) was reported to bind keratin IF in vitro (Hatzfeld et al., 1994; Heid et al., 1994; Kapprell et al., 1988), and henceforth has commonly been assumed to be involved in mediating attachment of IF to the desmosome. Moreover, PP2 has been localised near the cytoplasmic face of the desmosomal plaque in bovine nasal epidermis (Mertens et al., 1996), consistent with such a role, although its localisation in cardiac desmosomes appeared closer to the membrane and even enriched over the desmoglea. We have demonstrated PP1 to lie closer to the plasma membrane than either of the other major cytoplasmic plaque components, PG or DP. This may reflect a true functional difference between PP1 and PP2, as the latter has a significantly longer N-terminal domain that could extend further out from the ODP. A recent study suggests that PP3, like PP1, may be localised deep within the dense plaque of desmosomes in cultured epithelial cells, near the plasma membrane (Schmidt et al., 1999).

It is important to reiterate that the epitopes of the two PP1 antibodies are known with less precision than are those of the other antibodies used in this study. The ‘N-region’ antibody was raised against the whole PP1 head domain and the ‘C-region’ antibody against the whole PP1 repeat domain, and reactivity with deletion constructs suggests that both PP1 antibodies recognise numerous epitopes localised along the length of the fragments (M. Hatzfeld, unpublished data). Therefore broader gold particle distribution curves might have been expected using these two antibodies. In fact, the standard deviations of the distributions were similar to those of the other antibodies. The calculated distance between the mean position of the N- and C-terminal regions of PP1 was very similar to the distance between the termini of PG, which is of a similar size and closely related to PP1. It is thus unlikely that the actual termini of PP1 were located more than a few nm from the positions measured above.

Our results were inconsistent with a function for PP1 in direct binding of full diameter 10 nm IF in native desmosomes. However, the association of PP1 with keratins observed in vitro (Hatzfeld et al., 1994; Heid et al., 1994; Kapprell et al., 1988; Smith and Fuchs, 1998) could also occur in vivo if other oligomeric forms of keratins penetrate further into the desmosome than full diameter IF. It has been postulated previously that keratins may penetrate into the plaque in the form of unravelled protofilaments (Leloup et al., 1979). Moreover, filamentous structures resembling the 4-5 nm TF have been formed from aggregates of DP construct proteins containing both the rod and C-terminal domains together with IF proteins. This meshwork was not seen with DP alone, suggesting that keratin IF may be present in the plaque as an anastomosing network of fine protofilaments associated with DP (Stappenbeck and Green, 1992).

The finding that PP1 resides close to the plasma membrane is also interesting given the binding of its close relative, p120^ctn, to the juxtamembrane region of the E-cadherin cytoplasmic tail (Yap et al., 1998). Moreover our results are in line with data showing that PP1 binds to the desmosomal glycoproteins Dsc1a and Dsg1 (Mathur et al., 1994; Smith and Fuchs, 1998; but see also Kowalczyk et al., 1999). However our results are equally consistent with the reported in vitro binding of PP1 to the DP head domain (Smith and Fuchs, 1998) and in particular with more recent data demonstrating that the DP binding site in PP1 resides in its N-terminal non-armadillo head domain (Kowalczyk et al., 1999). Importantly, desmosomes of a patient with no PP1, due to mutations in the PP1 gene, displayed alterations both in adhesion and in DP and keratin IF organization (McGrath et al., 1997), suggesting a key linking role for PP1 in desmosome structure. Recent results demonstrating that PP1 enhances recruitment of DP to cell-cell borders and promotes lateral interactions among plaque components suggest that its loss in patients could compromise epidermal integrity by decreasing the number of binding sites for IF at the desmosome (Kowalczyk et al., 1999).

Our results demonstrate the value of the immunogold labelling technique for low resolution mapping of multimolecular structures. Such data provide crucial in situ information against which to interpret the results of molecular studies. This technique, when used in conjunction with additional domain-specific antibodies, has the potential to yield further valuable information on structure-function relationships within desmosomes and other cellular structures.

A.J.N. and D.R.G. are grateful for the generous support of the Wellcome Trust, the University of Manchester and the Cancer Research Campaign during these studies. The production of antibodies was also supported by the Medical Research Council (A.I.M), the NIH (grants RO1 AR43380 and AR41836 to K.J.G.) and the DFG (grant Ha 1791/3-1 to M.H.). We also thank members of the Garrod and Green laboratories for insightful discussions, the staff of Newton Heath Abattoir for providing bovine nose tissue and the staff of the Biological Sciences E.M. Unit in Manchester for their support.