Interaction of plakophilins with desmoplakin and intermediate filament proteins: an in vitro analysis

Among the diverse forms of the plaquebearing adhering junctions (for a review, see Schmidt et al., 1994) the desmosomes are characterized by their molecular composition and by their specific anchorage of bundles of intermediate filaments (IFs; Schwarz et al., 1990; Kowalczyk et al., 1999a). They represent clusters of isoforms of two types of transmembrane glycoproteins, the desmogleins (Dsg1-3) and desmocollins (Dsc1-3), both members of the larger family of cadherins. In their carboxyterminal, cytoplasmic domains, these desmosomal cadherins assemble the common plaque protein, plakoglobin (Cowin et al., 1986; Franke et al., 1987a,b, 1989; Fouquet et al., 1992), and a distinct set of desmosomal plaque proteins. These include general desmosomal proteins such as desmoplakin I (DPI) and cell typespecific proteins such as desmoplakin II (DPII; Franke et al., 1982; Mueller and Franke, 1983; Cowin et al., 1985) as well as members of the plakophilin (PKP) subfamily of arm-repeat proteins (PKP1: Kapprell et al., 1988, 1990; Hatzfeld et al., 1994; Heid et al., 1994; Schmidt et al., 1997; PKP2: Mertens et al., 1996, 1999; PKP3: Bonné et al., 1999; Schmidt et al., 1999; for a review see Hatzfeld, 1999). In addition, some other lesswellstudied proteins are found within the desmosomal plaque (e.g. Tsukita and Tsukita, 1985; Schwarz et al., 1990; Hatzfeld and Nachtsheim, 1996; Kowalczyk et al., 1999a).

The basic protein PKP1a (‘band 6 protein’, Kapprell et al., 1988), known to bind cytokeratins (CKs) in vitro (Kapprell et al., 1988; Hatzfeld et al., 1994; Smith and Fuchs, 1998), has been identified on the basis of its amino acid (aa) sequence, together with its splice variant PKP1b, as a member of a large family of proteins characterized by variable numbers of socalled arm-repeats comprising a motif of a mean number of 42 aa (Schäfer et al., 1993; Hatzfeld et al., 1994; Heid et al., 1994; Schmidt et al., 1994). This arm-repeat motif, first identified in the developmentally defined gene armadillo of Drosophila (Peifer and Wieschaus, 1990; Peifer et al., 1994), has been found in more than a dozen other junctional plaque and nuclear proteins, including plakoglobin (Franke et al., 1989) and β-catenin (McCrea et al., 1991).

The plakophilins (PKP1-PKP3) are remarkable as they occur constitutively in the nucleoplasm of cells normally forming desmosomes, cells induced to form desmosomes and cells devoid of desmosomes (e.g. Mertens et al., 1996; Schmidt et al., 1997; Bonné et al., 1999). In certain states of differentiation, plakophilins are recruited to the plasma membrane in a cell typespecific manner, targeted to very specific ensembles of plaque proteins where specific IF proteins are inserted. PKP1a, originally isolated from bovine muzzle epithelium as ‘band-6-protein’ under harsh extractive conditions, has been found primarily in desmosomes of stratified and complex epithelia (Franke et al., 1983; Kapprell et al., 1988, 1990; Hatzfeld et al., 1994; Heid et al., 1994; Schmidt et al., 1997). By contrast, PKP2, also present in two splice forms designated a and b, is characteristic of desmosomes of onelayered (‘simple’) epithelia and certain nonepithelial, desmosomepossessing tissues such as myocardium, but has also been localized to certain complex and stratified epithelia where it colocalizes with PKP1a and/or PKP3 (e.g. Mertens et al., 1996, 1999). PKP3 has been found in desmosomes of both simple and stratified epithelia but not in hepatocytes and in myocardium (Bonné et al., 1999; Schmidt et al., 1999).

Desmoplakins DPI and DPII, the latter being a splice variant of DPI (Green et al., 1990; Virata et al., 1992), together with envoplakin, periplakin, plectin and bullous pemphigoid antigen 1 (BPAG1), have been grouped into a distinct protein family, the plakins (Ruhrberg and Watt, 1997). All these proteins have been localized, in one or the other cell type, to certain IFs and their plasma membrane anchorage sites and share a common domain structure, i.e. a central rod domain flanked by globular aminoand carboxyterminal domains. Given their expression pattern, abundance, and the absence of a homologous protein in actinanchoring cell junctions, DPI and DPII have always been prime candidates for a role in the specific connections between desmosomes and IFs. Correspondingly, transient transfection studies using cDNA constructs encoding truncated DPI have suggested that the carboxyterminal desmoplakin domain interacts with IFs (Stappenbeck and Green, 1992; Stappenbeck et al., 1993; Bornslaeger et al., 1996). This concept has been confirmed by in vitro studies using CKs and the recombinant carboxyterminal desmoplakin domain (Kouklis et al., 1994), as well as by observations made in yeast twohybrid experiments (Meng et al., 1997).

Using blot overlay studies and reconstitution experiments in vitro, we have systematically examined the direct interaction between PKP1a and IF proteins and have also identified, by immunoprecipitation of solubilized cellular proteins, desmoplakin as a PKP1a complex partner.

For expression of complete human PKP1a (accession number Z34974) in Escherichia coli the BsiWI/BamHI 2400 bp fragment was subcloned into the corresponding sites of a derivative of pET-21d, containing a 33-mer oligonucleotide pair coding for the first eleven authentic PKP1a amino acids and introducing a BsiWI site, to yield pET-21d(PKP1a). The constructs of various subdomains, as shown in Fig. 1, were generated from pET-21d(PKP1a) using endogenous restriction sites and appropriate oligonucleotide pairs to introduce stop codons or start codons, respectively.

cDNAs coding for the complete CK5 and CK14, respectively, were isolated from a human epidermis cDNA library (Clontech, Palo Alto, CA, USA; 5′-Stretch Plus, HL 1112b) using ³²P-labelled partial clones of human CK5 and 14 (provided by L. Langbein; Langbein et al., 1993) employing standard procedures (Sambrook et al., 1989). Both clones were mutagenized to include their start codons as part of unique NdeI sites, and were subsequently cloned into the prokaryotic expression vector pET-21b (Novagen, Madison, WI, USA).

Total human vimentin (Herrmann et al., 1993), the human vimentin rod domain (Rogers et al., 1995), mouse desmin (Rogers et al., 1995), total CK8 and CK18 (Hofmann and Franke, 1997), the CK8 rod and the CK18 rod (Bader et al., 1991), all subcloned into the pDS5 plasmid, were introduced into E. coli, strain TG1. Human vimentin, mouse desmin (Li et al., 1994) and the mouse desmin rod domain were purified according to Hofmann et al. (1991), and the human vimentin rod domain according to Herrmann et al. (1996). CK5, CK14, CK8, CK18, CK8 rod and CK18 rod were prepared as described (Coulombe and Fuchs, 1990; Hofmann and Franke, 1997).

E. coli strain BL21 was transformed with plasmids containing the cDNA of human PKP1a, various subdomains derived from PKP1a, neurofilament protein NF-L (Heins et al., 1993) and lamin LIII (Stick, 1988), subcloned into the pET-vector system. For the generation of recombinant proteins, transformed bacteria were grown overnight in 400 ml TB medium at 37°C under rigorous shaking. PKP1a and truncated versions of PKP1a were enriched in inclusion body fractions and purified as described (Hofmann et al., 1991). The inclusion body fraction was dissolved in 40 ml of column buffer I (8 M urea, 5 mM Tris-HCl, 1 mM EDTA, 0.1 mM EGTA, 1 mM DTT, pH 7.5) and centrifuged for 90 minutes in a Beckman Ti 50 rotor at 35,000 rpm (Beckman Instruments, München, Germany). The supernatant was directly applied to a 30 ml DEAE-sepharose column in column buffer I. The proteins PKP1a, PKP1a-N268, PKP1a-N353 and PKP1a-N331his were recovered in the flowthrough fraction and were directly applied to a 15 ml CM-sepharose column equilibrated in column buffer I. Bound protein was eluted in a 50 ml gradient of NaCl (0-0.3 M) with column buffer I. Peak fractions were monitored by SDS-PAGE and purified protein was pooled and stored at −80°C until use. NF-L bound to the DEAE-sepharose column was eluted in a 50 ml gradient of NaCl (0-0.3 M) in column buffer I. It did not, however, bind to CM-sepharose in column buffer I. Therefore fractions eluted from the DEAE-sepharose column were dialzyed into column buffer II (8 M urea, 30 mM sodium formate, 1 mM EDTA, 1 mM EGTA, 1 mM DTT, pH 4.0) and subsequently applied to a CM-sepharose column equilibrated in column buffer II. PKP1a-324C and the myosin rod were not deposited in bacterial inclusion bodies and were therefore enriched and purified according to Herrmann et al. (1996) with minor modifications. The S-100 gel filtration column followed by concentration on a DEAE sepharose column was omitted and the protein was directly applied to a CM-sepharose column equilibrated in column buffer II in the case of PKP1a-324C or in column buffer I for the myosin rod.

IF proteins were dialyzed by stepwise lowering of the urea concentration into low salt buffers (for type III IF proteins: 5 mM Tris-HCl, pH 8.4; see Hofmann et al., 1991; for CKs: 2 mM Tris-HCl, pH 9.0; cf. Hofmann and Franke, 1997; Hofmann, 1998). IF assembly was started by adding the corresponding tenfold concentrated buffer (‘filament buffer’; final concentrations, for type III IF proteins: 50 mM NaCl, 25 mM Tris-HCl, pH 7.5; for CKs: 10 mM Tris-HCl, pH 7.5). For the assembly of NF-L, LIII and myosin rods, the conditions employed with type III IF proteins were used. PKP1a or mutated forms of PKP1a were dialyzed in parallel by stepwise lowering of the urea concentration into 10 mM Tris-HCl, pH 7.5, 1 mM DTT, and were then added to soluble forms of IF proteins, either simultaneously with filament buffer or 30 minutes after IF assembly start. After addition of PKP1a the mixtures were further incubated for 60 minutes at room temperature or, in the case of human vimentin and mouse desmin, at 37°C. Molar ratios (PKP1a:IF protein) ranging from 1:130 to 8:1 were examined, at total protein concentrations ranging from 0.1 to 0.2 mg/ml. CK5/14 was dialyzed by stepwise lowering of the urea concentration into assembly buffer (10 mM Tris-HCl, 1 mM MgCl2, 1 mM DTT, pH 7.25). PKP1a was added either in urea or after dialysis into assembly buffer to preformed CK5/14 IFs and incubated for 60 minutes at room temperature.

Procedures for negative staining as well as for pelleting assembled material and further processing for electron microscopy have been described (Hofmann and Franke, 1997). Specimens were examined using a Zeiss electron microscope model 10 or 900 (Carl Zeiss, Oberkochen, Germany).

For centrifuation studies, assembly of IF proteins (0.2 mg/ml) was started in Airfuge polyallomer centrifugation tubes (Beckman Instruments). The molar ratio of PKP1a or mutants thereof to IF protein ranged from 1:80 to 8:1, as specified in Results. PKP1a or its truncated forms were added to soluble complexes of IF proteins, either simultaneously with filament buffer or 30 minutes after initiation of IF assembly, and incubated for 60 minutes. In the case of CK5/14, PKP1a was either added to ureasolubilized CKs and carried with them through the various dialysis steps, or dialyzed into low salt buffers (see above) and mixed with CK5/14 samples containing preformed IFs. Assembled material was pelleted by centrifugation for 15 minutes at 10 psi in the Airfuge, the supernatant was withdrawn, and the pellet dissolved in an equal volume of SDS sample buffer containing 6 M urea followed by heating at 95°C for 3 minutes. Protein samples of pellet material and supernatant fractions were analyzed by SDS-PAGE.

Protein samples separated by SDS-PAGE were transferred to nitrocellulose. Coupled in vitro transcriptiontranslation of cDNA clones was performed at 30°C for 90 minutes in the presence of [³⁵S]methionine, using the reticulocyte lysate system and T3 RNA polymerase (Promega, Mannheim, Germany). Filters were incubated either with ³⁵S-labelled CKs or with ³⁵S-labelled fulllength PKP1a, fulllength PKP2a or the aminoterminal head domain (aa 1-367) of PKP2a. Binding assays with [³⁵S]methioninelabelled PKP2 were performed according to Merdes et al. (1991). Blots were incubated three times for 10 minutes with washing buffer (20 mM Tris-HCl, pH 7.3, 154 mM NaCl, 0.1% Tween-20). Incubation was performed at 4°C overnight in ‘gelatin buffer’ (20 mM Tris-HCl, pH 7.3, 154 mM NaCl, 1 mM MgCl2, 1 mM DTT, 0.1% Tween-20, 0.2% (w/v) gelatin (Merck, Darmstadt, Germany)). After blotting filters were washed four times for 30 minutes each in gelatin buffer, airdried and processed for autoradiography.

Affinities of IF proteins to immobilized PKP1a were determined by surface plasmon resonance detection using a Bialite apparatus (Biacore AB, Freiburg i. Br., Germany; cf. Hofmann and Franke, 1997; Hofmann, 1998). Samples of vimentin or desmin rod domains and equimolar mixtures of CK8 rod and CK18 rod were dialyzed into 2 mM Tris-HCl, pH 9.0, diluted to a concentration of 0.1 mg/ml in ‘running buffer’ (10 mM Hepes, 150 mM NaCl, 1 mM DTT, 0.005% surfactant P20, pH 7.4) and subsequently injected into the apparatus. Protein association was monitored at a flow rate of 5 μl/minutes at 25°C over a 7 minute period. The surface of the sensor chip was regenerated between each sample injection by washing with 50 mM NaOH. In parallel, a ‘naked’ surface without immobilized PKP1a was used to measure nonspecific binding to the dextran matrix. The PKP1a binding curves obtained were corrected for nonspecific binding.

Cultured human keratinocytes of line HaCaT (Boukamp et al., 1988) were grown to confluency in a 10 cm dish. They were rinsed in PBS buffer and incubated with 1 ml extraction buffer (80 mM KCl, 20 mM NaCl, 5 mM EDTA, 1 mM DTT, 250 mM sucrose, 15 mM Hepes, pH 7.5, 0.2% Nonidet-P40) for 5 minutes at 4°C. Cells were then gently scraped off with a rubber policeman, disrupted by pipetting up and down in a narrow bore pipette and centrifuged at 13,000 g for 10 minutes at 4°C. The supernatant fraction was directly loaded on top of a 5%-30% (w/v) or 10%-60% (w/v) linear sucrose gradient buffered with 10 mM Tris-HCl, pH 7.5. Centrifugation was performed in a SW40 rotor (Beckman Instruments) at 35,000 rpm for 19 hours at 4°C. 15 fractions of 0.8 ml each were collected from top to the bottom of the gradient. Marker proteins (bovine serum albumin (BSA), catalase, thyroglobulin; all from Sigma, München, Germany) were separated on parallel gradients.

Protein was extracted from confluent HaCaT cultures in icecold immunoprecipiation (IP)-buffer (140 mM NaCl, 5 mM EDTA, 20 mM Hepes, pH 7.5, 1% Nonidet-P40) followed by centrifugation at 13,000 g for 10 minutes at 4°C. Samples were cleared by addition of protein A sepharose beads (Amersham Pharmacia Biotech, Freiburg i. Br., Germany) for 2 hours on a rotating wheel at 4°C. The beads were then pelleted, and the supernatants transferred to a tube containing protein A sepharose beads preloaded with guinea pig antibodies specific for PKP1a (B6-4; Schmidt et al., 1997). After overnight incubation at 4°C on a rotating wheel the protein A sepharose beads were washed four times in icecold IP-buffer, then boiled in sample buffer, processed by SDS-PAGE and stained either with Coomassie Brilliant Blue or blotted to PVDF membranes. As a control unrelated guinea pig antibodies were processed in parallel. Protein bands were excised and digested in the gel strip for peptide mass fingerprinting by matrixassisted laser desorption/ionization (MALDI) mass spectrometry as described (cf. Schmelz et al., 1998).

Immunoblotting for PKP1a was performed with monoclonal antibodies (mAbs) 2D6, 5C2 or 9E7 (Heid et al., 1994). Desmoplakin was detected with desmoplakinspecific rabbit antibodies (Natutec GmbH, Frankfurt, Germany; Arnemann et al., 1993) or a mixture of mAbs specific for DPI and DPII (2.15, 2.17, 2.20; Cowin et al., 1985). Detection of bound horseradish peroxidasecoupled secondary antibodies was performed using the ECL-system (Amersham Pharmacia Biotech).

The fulllength human PKP1a polypeptide and various subdomains (Fig. 1) were synthesized in E. coli, isolated and purified to homogeneity. In order to test if the recombinant PKP1a was able to bind to CKs, as previously shown for native PKP1a (Kapprell et al., 1988; Hatzfeld et al., 1994), blot binding assays were performed (Fig. 2). The immobilized proteins (immunoblots are shown in Fig. 2A) were overlaid with the ³⁵S-labelled epidermal cytokeratins CK5 or CK15. Both cytokeratins bound to fulllength PKP1a. Binding was also detected for the aminoterminal half (PKP1a-N331his) whereas the carboxyterminal half (PKP1a-324C) remained undecorated (Fig. 2Ab,c). The same type of binding was demonstrated for the ³⁵S-labelled α-helical rod domains of CK8 and CK18 (data not shown). In the reverse experiment, when cytokeratins were bound to nitrocellulose, both PKP1a and PKP2a bound strongly to CK5, CK14, CK8 and CK18 and to some extent to vimentin (Fig. 2Bb,c). In addition, the aminoterminal, non-arm-repeat domain of PKP2 (aa 1-367), bound to CK8, CK18 and vimentin in a very similar way to that seen with the complete molecule (data not shown). Hence, both PKP1a and PKP2a interact strongly with various IF proteins, the major binding activity probably residing in the aminoterminal non-arm-repeat domain.

To study interactions of PKP1a and IFs at the ultrastructural level we performed in vitro reconstitution experiments, followed by negative staining of the assembled structures. For IF proteins, especially cytokeratins, pathways for productive refolding from the fully denatured state, as obtained in 9.5 M urea, to assemblycompetent soluble complexes, mainly tetramers, have been established and used extensively (see Hofmann and Franke, 1997, and references therein; for a review see Herrmann and Aebi, 1999). We therefore employed the same type of dialysis regimen as previously used for cytokeratins with PKP1a. When dialyzed from 8 M urea into 10 mM Tris-HCl, pH 7.5, 1 mM DTT, PKP1a was still present as a monomer, as revealed by sedimentation equilibrium experiments with the analytical ultracentrifuge (I. Hofmann, N. Mücke, J. Reed, H. Herrmann and J. Langowski, unpublished). Moreover, according to circular dichroism data as well as results obtained by endoproteinase AspN digestion, the recombinant PKP1a evidently folded into a distinct conformation (data not shown).

We started our analysis with CKs 8 and 18, since for this cytokeratin pair the in vitro assembly is well characterized (Hofmann and Franke, 1997); furthermore, PKP1a and CK8/18 are bona fide interaction partners in vivo, being colocalized in desmosomes in HaCaT cells and cells of other cultured lines (Boukamp et al., 1988; Schmidt et al., 1997). PKP1a was added either to preformed CK8/18 IFs (Fig. 3A) or to precursors of CK8/18 IFs simultaneously with the assembly initiation by rapid mixing with filament buffer (Fig. 3B). In both cases, thick fibrils of laterally aggregated IFs, up to 120 nm in diameter, were observed, side by side with normallooking 8-14 nm IFs. The thick fibrils often folded back on themselves, thereby forming loops (Fig. 3C). Even at rather low molar ratios of PKP1a to CK8/18, such as 1:30, thick fibrils were observed and by increasing the proportion of PKP1a added to CK8/18 IFs the frequency of individual IFs was drastically reduced. In the absence of PKP1a thick fibrils of this kind were not found in CK8/18 IF assembly experiments (data not shown). In an attempt to narrow down the IF binding domain of PKP1a, we employed the aminoterminal, non-arm- repeat domain of PKP1a (aa 1-268) as well as slightly longer fragments (see Fig. 1) in analogous assembly experiments, demonstrating that the first third of PKP1a was as effective in bundling CK8/18 IFs as the whole molecule (Fig. 3D and data not shown). In sharp contrast, a truncated version containing most of the arm-repeat domain, PKP1a-324C (aa 324-727), had only little effect on the higher order interaction of CK8/18 IFs (Fig. 3E).

Since simple epithelial CKs such as CK8 and CK18 differ quite significantly from CKs found in epidermis, such as CK5, CK14 and CK15, and since PKP1a colocalizes with CK5/14 in desmosomes of the basal cell layer of epidermis (Heid et al., 1994; Moll et al., 1997), we also investigated the in vitro interaction of PKP1a with reconstituted CK5/14 IFs (for assembly of CK5/14, see Coulombe and Fuchs, 1990). In both principal types of coassembly experiments, addition of PKP1a to assembled IFs or presence of PKP1a during dialysis into assembly buffer, PKP1a induced the formation of thick fibers containing numerous individual IFs (Fig. 3F).

The apparently similar type of interaction of two different CK pairs with PKP1a prompted us to investigate if representatives of the other assembly groups (see Herrmann and Aebi, 2000) would also interact with PKP1a. The solidphase binding experiments had shown that vimentin bound both PKP1a as well as PKP2a to some extent (see Fig. 2B). In conventional assembly experiments, as introduced above (Fig. 3), both PKP1a (data not shown) and the aminoterminal domain PKP1a-N331his induced the formation of highly ordered fibrils, bundling numerous individual vimentin IFs into structures persisting for more than 4 μm in length (Fig. 4A), eventually fusing with another fiber that partially folded back on itself, thereby forming a loop (thick arrow). Enlargements of different types of structures formed by PKP1a-N331his and vimentin are shown in Fig. 4B-E. In Fig. 4B a bundle terminates in few individual IFs; Fig. 4C depicts the transition from a tightly packed bundle into a zone of loose association back into a more condensed bundle; Fig. 4D shows the appearance of a ‘normal’ bundle, with the contours of individual IFs being visible in the upper middle part among the ‘crossbridging’ PKP1-derived material. The type of organization of PKP1a molecules on individual IFs is shown in Fig. 4E (arrowheads). Amorphous material in the form of knobs with a diameter of 30 nm is frequently seen lying on and connecting several filaments to loose, parallel running fiber arrays. For desmin the addition of PKP1 mainly gave the same result (data not shown). Further insight into the thick, densely packed fibrils was obtained with ultrathin sections of pelleted material (Fig. 4F). Often, normallooking IFs protruded out of these fibers and, on occasion, longitudinally sectioned bundles revealed individual IFs making a connection to the next bundle (arrows). Moreover, the distinct minimal inter-IF spacing, typical for vimentin filaments (e.g. Hofmann et al., 1991), was lost and these bundles appeared to be packed with maximal density (Fig. 4G). Thus, a bundle of 100 nm diameter may, theoretically, contain up to 100 IFs. The small crosssectioned bundle in the upper left apparently consists of eleven IFs, making this estimate indeed realistic. Fibers formed by CK8/18 IFs and PKP1a were processed in parallel and gave principally the same results (data not shown).

The nature of the PKP1a-IF interaction was further examined by centrifugation experiments, when PKP1a was added either to preformed CK8/18 IFs or together with the buffer that initiates IF assembly of CK8/18. The experiments were terminated by centrifugation and the resulting soluble and pelletable material was analyzed by SDS-PAGE (Fig. 5). When PKP1a was present in near stochiometric amounts, both the CK8/18 IFs and PKP1a were almost quantitatively pelleted, irrespective of whether PKP1a was added to preformed IFs or added to the IF assembly mixture. In contrast, PKP1a on its own remained entirely in the supernatant fraction (Fig. 5A). Also, the smallest N-terminal PKP1a fragment (PKP1a-N268) bound strongly to CK8/18 and cosedimented whereas most of the C-terminal PKP1a fragment (PKP1a-324C) remained in the supernatant (Fig. 5B). Furthermore, in order to identify the IF protein domains involved in PKP1a binding, we formed all possible pair combinations between wild type (wt) CK8, wt CK18, the CK8 rod and CK18 rod domain, and studied their interaction with PKP1a in solution (Fig. 5C). Notably, the heterodimer consisting of both rod domains was sufficient to generate structures that sedimented quantitatively. Finally, we investigated if one CK partner alone could bind and mediate the cosedimentation of PKP1a. On their own, CK8 as well as CK18, like PKP1a, remained in the supernatant fraction (data not shown), but when mixed with PKP1a, both were recovered in the pellet together with PKP1a (Fig. 5C). Similarly, two IF proteins from sequence class IV and V, i.e. the low molecular weight neurofilament triplet protein NF-L and Xenopus lamin LIII, bound quantitatively to PKP1a yielding structures that sedimented in the Airfuge (data not shown).

Since PKP1a is a basic protein with an isoelectric point of 9.25, we examined the possibility that the observed binding of IF proteins was due merely to an electrostatic interaction. In centrifugation assays, the soluble, acidic protein bovine serum albumin (BSA) did not cosediment with PKP1a but both proteins remained in the supernatant (Fig. 6). Further support for the specificity of the interaction between IF proteins and PKP1a came from experiments employing another coiledcoil forming protein, the myosin rod domain, known to exhibit clusters of negative charge on its surface. Most of the myosin rod sedimented on its own but leaving, however, PKP1a added in a severalfold molar excess nearly entirely in the supernatant (Fig. 6).

After establishing these qualitative data, we attempted to analyze the PKP1a-IF protein interaction quantitatively. In our sedimentation experiments the molar ratio of PKP1a to IF proteins was varied from 1:130 to 8:1. However, the amount of pelleted PKP1a increased only up to a molar ratio of approximately 1:1. When present in excess over IFs, surplus PKP1a remained in the supernatant fraction. This saturation effect was obtained for the epidermal CK pair, CK5/14, and the simple epithelial CK pair, CK8/18, as well as for vimentin (Fig. 6).

In a first attempt to evaluate the relative affinities of PKP1a to different IF proteins, we applied surface plasmon resonance technology with PKP1a immobilized on the sensor surface. We decided to investigate the binding properties of the α-helical rod domains of CK8/18, desmin and vimentin, since we had shown that the CK8/18 rod by themself already interacted strongly with PKP1 (see Fig. 5C). Various amounts of these IF protein fragments were allowed to react and increases in resonance (RU, resonance units) were monitored and compared (Fig. 7). All three proteins bound to the immobilized PKP1a as indicated by the increase of resonance, although with somewhat differing affinities. For the vimentin rod, final RU values of more than 800 were attained. The rod domain of desmin and CK8/18 complexes showed RU values of around 400 and 100, respectively, indicative of a lower relative affinity under these experimental conditions. Binding of either CK8 rod or CK18 rod alone resulted in final RU values comparable to those seen with the mixture of type I and type II CKs (data not shown). In all combinations examined, practially no dissocation was observed (less than 50 RU), and therefore the determination of affinity constants was not possible. Wildtype IF proteins could not be analyzed as they rapidly assembled in the physiological buffer conditions used. Evidently filaments cannot be analyzed in the flowthrough system of the biacore instrument.

Having shown the direct interaction of PKP1a with IF proteins and IFs in vitro, we searched for protein complexes containing PKP1a in vivo. When HaCaT keratinocyte cultures were lysed with Nonidet-P40-containing buffers, and the soluble proteins obtained used for immunoprecipitation with PKP1a antibodies, substantial amounts of PKP1a were recovered by immunoprecipitation (Fig. 8A, open arrowhead). A Coomassie Bluestained band with less electrophoretic mobility than the heavy chain myosin of the gel standard (212 kDa) was also found specifically enriched in these immunoprecipitates (Fig. 8A, filled arrowhead) and was identified by peptide mass fingerprinting and immunoblotting as desmoplakin I (DPI). The amount of DPI clearly exceeded that of DPII as verified in immunoblots using a serum specific for both proteins (Fig. 8B). The Coomassie Bluestained band comigrating with myosin was not further characterized as it also showed up in the control to some extent.

Other known desmosomal proteins such as desmoglein and plakoglobin did not coimmunoprecipitate with PKP1a. Neither were CKs found to be enriched in soluble PKP1a complexes (data not shown). This, however, is in accordance with the fact that most of the IF proteins are in an assembled state within cells and therefore are not present in cell extracts that have been cleared by centrifugation. In the reverse experiment, when desmoplakinspecific antibodies were used for immunoprecipitation, PKP1a was coisolated with desmoplakin, whereas desmoglein (see e.g. Pasdar et al., 1991) and plakoglobin were not (data not shown).

To characterize the soluble PKP1a complexes in more detail, sucrose density gradient centrifugations of cell extracts was performed. PKP1a was found in two peaks. The main portion in 5%-30% sucrose gradients was found in the last fraction (Fig. 9A), a minor one fractionated together with soluble desmoplakin (approx. 8 S). In addition, a band of lower molecular mass of approximately 50 kDa was visible, located in fractions 4-7. This band was recognized by the PKP1aspecific serum and probably corresponds to a degradation product of PKP1a. In parallel experiments with recombinant PKP1a, the purified protein sedimented at 4.3 S (data not shown). In order to investigate more closely the composition of the fastsedimenting PKP1acontaining material, we analyzed corresponding cell extracts on 10%-60% sucrose gradients (Fig. 9B,B ′). The main portion of PKP1-containing complexes was found in a distinct peak of around 50 S. This fraction did not contain any of the other known desmosomal proteins. In addition, PKP1a complexes with S-values of approximately 8 cosedimenting with desmoplakin were detected (cf. Duden and Franke, 1988; Pasdar and Nelson, 1988). Further characterization of the high molecular mass complex of PKP1a is in progress.

In the present study we have investigated the interaction of PKP1a with various IF proteins by a series of different in vitro assays. We show that this interaction is specific for IF proteins. Moreover, the PKP1a head domain alone, i.e. the N-terminal non-arm-repeat domain (aa 1-268), is sufficient to mediate bundling of IFs. With respect to the IF proteins, their α-helical rod domain enables the formation of macromolecular structures with PKP1a.

The electron microscopic appearance of PKP1ainduced IF bundles is somewhat similiar to that of CK bundles (‘macrofibrils’) induced by filaggrin, an upper strataspecific protein of squamous stratified epithelia, notably epidermis (Dale et al., 1978, 1988, 1993, 1997; Steinert et al., 1981; Mack et al., 1993; Parry and Steinert, 1995; see also, however, Weidenthaler et al., 1993; Ishida-Yamamoto et al., 1994). Both PKP1a and filaggrin are basic proteins and bind to the IF rod domain in a proteinspecific manner. For filaggrin it has been shown that basic 20-mer peptides consisting of a repeated amino acid motif (SGSR/X, X being a charged residue) interact as efficiently with IFs as does fulllength filaggrin. A similar type of repeat motif does not, however, occur in the N-terminal non-arm-repeat domain of PKP1a. Currently we are examining whether a basic 21-aa stretch in the plakophilin N terminus, found in PKP1a, PKP2 and PKP3 (cf. Schmidt et al., 1999), is involved in IF interaction. The question of how PKP1a induces IFs to bundle cannot be unequivocally answered with our current knowledge. It is clear from sedimentation equilibrium experiments that PKP1a on its own is monomeric in solution (I. Hofmann, N. Mücke, J. Reed, H. Herrmann and J. Langowski, unpublished). This does not necessarily imply that within the PKP1a head domain two binding sites are located. As has been shown for a filaggrinderived peptide, short basic peptides derived from PKP1a may function as an ionic zipper and may be able to bundle IFs. On the other hand, binding of PKP1a to IFs might induce intramolecular, conformational changes such that PKP1a oligomerizes, forming a glue embedded in IFs.

Our finding that the rod domain of type III IFs or CK8/18 is capable for PKP1a interaction and sufficient to mediate bundle formation, seems to be at variance with results reported for the epidermal CK5 by Smith and Fuchs (1998). Using blot overlays they identified a sequence motif in the non-α-helical head domain of CK5 as responsible for PKP1a binding. Clarification of this apparent difference will have to await future experiments. However, we do not exclude the existence of other interaction sites outside the α-helical rod domain.

In addition to PKP1, desmoplakin is so far the only other desmosomal protein reported to bind to IF proteins. In transient transfection studies using simple epithelial cells or fibroblasts, an N-terminally truncated DPI-construct colocalized with IF fibrils and resulted in a gradual disruption of the endogenous IF network (Stappenbeck and Green, 1992). In further analyses, Stappenbeck et al. (1993) pointed to the requirement of the last 68 aa of desmoplakin for its binding to IF proteins, a region known to harbour a repeated amino acid motif (GSRS; cf. Franke et al., 1989; Green et al., 1990) resembling the filaggrin motif mentioned above.

For desmosome assembly not only IF-binding plaque proteins are needed but also the transmembrane cadherins, desmoglein and desmocollin. It has been repeatedly demonstrated that the cytoplasmic portions of both desmosomal cadherins can bind to plakoglobin (Troyanovsky et al., 1993, 1994a,b; Mathur et al., 1994; Chitaev et al., 1996; Wahl et al., 1996; Witcher et al., 1996). Observations made with DP null mouse embryos suggest that DP is important not only in establishing IF cytoskeletal architecture but also for assembly and stabilization of the desmosome (Gallicano et al., 1998). Moreover, Smith and Fuchs (1998) have recently reported, mostly in blot overlay experiments, in vitro interactions of PKP1a with major desmosomal components, i.e. desmoglein, desmocollin and desmoplakin. Given their close amino acid sequence homology, PKP2 and PKP3 would be expected to display similar characteristics of protein interactions but this needs to be further studied.

In search of PKP1a complexes in vivo we have identified, in immunoprecipitation experiments, soluble or solubilized cytoplasmic DPI as a specifically interacting partner. We tend to conclude that this is a genuine cytoplasmic complex, although so far we cannot rule out that part of this complex has been extracted from desmosomes. Both proteins cofractionate during sucrose density centrifugation at approximately 8 S, whereas the purified proteins reveal lower S-values. Recombinant PKP1a sediments at 4.3 S, and purified DPI has been reported at 6.7 S (O’Keefe et al., 1989). Desmoplakin from cultured epithelial cells grown in low calcium medium has been shown to occur in soluble complexes of 7.3-9 S (Duden and Franke, 1988; Pasdar and Nelson, 1988), in fair agreement with our observation. A direct PKP1adesmoplakin interaction is also suggested from results obtained with the yeast twohybrid system and from transient transfection experiments (Kowalczyk et al., 1999b). Obviously, the PKP1adesmoplakin complex shown here is different from the large vesicleassociated structures containing desmoplakin, desmoglein and plakoglobin isolated with vesicle fractionation techniques (Demlehner et al., 1995).

The concept that the interaction of PKP1a with IFs as well as with desmoplakin is important in vivo receives further support from reports that mutations in both alleles of the PKP1a gene result in a form of congenital ectodermal dysplasia (McGrath et al., 1997, 1999). In the epidermis of these patients, the association between IFs and desmosomes, specifically desmoplakin, appears disturbed although plakoglobin, desmogleins and desmocollins are still localized to desmosomelike sites at cell borders.

On the other hand it should not be overlooked that in vivo a decoration of IFs with PKP1a, or any other plakophilin, is not observed (Kapprell et al., 1988; Schmidt et al., 1994, 1997, 1999; Mertens et al., 1996). Only recently the decoration of IFs with fragments of PKP1a has been reported to occur in some transfected cells (Klymkowsky, 1999). This suggests that, in normal living cells, the association of PKP1a with IFs and with DPI is tightly regulated to prevent plakophilins from binding to IFs and PKP1amediated IF bundling. Of course, PKP1a – or other plakophilins – might bind to IF protein configurations (‘roots’) deep in the desmosomal plaque where plakophilins have been localized, at least in certain cell types, relatively close to the membrane (e.g. Kapprell et al., 1990; Mertens et al., 1996; North et al., 1999; Schmidt et al., 1999). Obviously, the regulatory mechanisms governing plakophilin-IF protein interactions in vivo will have to be elucidated in future experiments.

Finally as most, if not all, cells contain plakophilins, usually in the nucleus, we still have to find the regulatory principles for recruiting the specific plakophilins to, or excluding them from, the desmosomes of a given cell type. Sitespecific phosphorylation of desmoplakin, for example, has been shown to regulate its interaction with IFs (Stappenbeck et al., 1994). However, other regulatory principles, including additional posttranslational modifications, should not be excluded a priori as determinants in the observed topological sorting.

We thank Reimer Stick and Lutz Langbein for helpful discussions, Jutta Osterholt for skillful photographic work and Eva Ouis for arranging the manuscript. Moreover, we gratefully acknowledge the technical assistance of Sonja Reidenbach. Alfred Wittinghofer provided the prokaryotic expression plasmid encoding the myosinrod fragment and Ansgar Schmidt helped with the design of the PKP1a constructs. W. W. Franke is thanked for his generous support and critical comments. The work has been supported by the Deutsche Forschungsgemeinschaft (German Research Foundation).