The Hemidesmosomal Protein Bullous Pemphigoid Antigen 1 and the Integrin (cid:1) 4 Subunit Bind to ERBIN MOLECULAR CLONING OF MULTIPLE ALTERNATIVE SPLICE VARIANTS OF ERBIN AND ANALYSIS OF THEIR TISSUE EXPRESSION*

The bullous pemphigoid antigen 1 (eBPAG1) is a con-stituent of hemidesmosomes (HDs), cell-substrate adhesion complexes in stratified epithelia. Although its COOH terminus interacts with intermediate filaments, its NH 2 terminus is important for its recruitment into HDs. To identify proteins that interact with the NH 2 terminus of human eBPAG1, we performed a yeast two-hybrid screen, which uncovered a protein belonging to the LAP/LERP (for L RR a nd P DZ domain) protein family with 16 NH 2 -terminal leucine-rich repeats and a COOH-terminal PDZ domain. The gene for this LAP/ LERP protein comprises at least 26 exons located on the long arm of chromosome 5. In most human tissues, several transcripts were detected differing in the coding region situated upstream of or within the PDZ domain. One of the encoded variants was found to correspond to the recently described protein ERBIN. In yeast and in vitro binding experiments, ERBIN was shown to interact not only with eBPAG1 but also with the COOH-ter-minal region of the cytoplasmic domain of the integrin (cid:1)

Bullous pemphigoid antigen 1 (epithelial BPAG1 or eBPAG1) 1 is a component of hemidesmosomes (HDs), multiprotein adhesion complexes promoting cell-substrate adhesion in stratified and complex epithelia. Ultrastructurally, these complexes appear as electron-dense structures in close contact with the basal cell membrane associated with the cytoskeleton (1,2). eBPAG1 was identified originally as an autoantigen in the autoimmune subepidermal blistering disorder of the skin called bullous pemphigoid (3,4). This protein is a member of a protein family involved in cytoskeletal organization, the plakins that also comprise desmoplakin, plectin, envoplakin, and periplakin (reviewed in Ref. 5). The structural organization of the plakins is similar, with a central coiled-coil domain responsible for dimerization flanked by two large globular end domains. Although the NH 2 termini of plakins seems to mediate their recruitment to distinct plasma membrane sites (6 -9), their COOH termini are implicated in their attachment to the intermediate filament cytoskeleton (7,10,11). Specifically, cell transfection and yeast two-hybrid analyses have demonstrated that the last 768 residues of human eBPAG1 contain sequences important for their interaction with intermediate filaments (7,12). Consistent with this result, BPAG1-knockout mice show discrete signs of blistering, most likely as a result of an impaired attachment of keratin-intermediate filaments to HDs leading to the mechanical fragility of basal keratinocytes (13). These mice also develop neurological degeneration with dystonia and ataxia, which results from the concomitant inactivation of the neuronal isoforms of eB-PAG1, nBPAG1, or dystonin, which differ from eBPAG1 exclusively at their NH 2 termini (13,14). These variants, which contain an actin or a microtubule-binding domain (7,15), link the actin, the intermediate filament, and the microtubule networks to each other. They are essential for the maintenance of the cytoarchitecture of neurons (7,15). The NH 2terminal domain of eBPAG1 is implicated in its recruitment to HDs. This region associates with the cytoplasmic domain of two transmembrane components of HDs, the bullous pemphigoid antigen 2 (BPAG2, also termed BP180) and the integrin ␤4 subunit (9).
BPAG2 is a type II transmembrane collagenous molecule (16). Transfection and yeast two-hybrid experiments have shown that its cytoplasmic domain associates with the distal COOH-terminal half of the cytoplasmic domain of the integrin ␤4 subunit (17,18) and with eBPAG1 2 (9). The large collagenous extracellular domain serves as a cell surface receptor for extracellular matrix proteins. The ␣6␤4 integrin plays a key role in HD assembly and in cell adhesion (reviewed in Ref. 2). The cytoplasmic domain of the integrin ␤4 subunit consists of over 1000 amino acids and contains two pairs of type III fibronectin (FN-III) repeats separated by a connecting segment. It interacts with both BPAG2 and plectin, another cytoplasmic component of HDs (17)(18)(19)(20)(21). The ␣6␤4 integrin is a cell surface receptor that binds to various isoforms of laminin including laminin 5. Furthermore, ␣6␤4 is implicated in signaling pathways that control cell growth and cell cycle progression (reviewed in Ref. 22). Activation of the ␣6␤4 integrin in response to clustering or cell adhesion results in the sequential phosphorylation of ␤4 and recruitment of the adaptor proteins Shc and Grb2. In turn, Grb2 associates with exchange factor mSOS, leading to the activation of the Ras-Erk pathways. In addition, tyrosine phosphorylation of ␤4 also occurs as a result of treatment of epithelial cells with epidermal growth factor, which has been associated with the disassembly of HDs and reduction of adhesion but is not associated with Shc recruitment (23). These findings suggest that the phosphorylation state of ␤4 controls the interaction of the various components of HDs with each other and the cytoskeleton.
To identify proteins that interact with the NH 2 -terminal domain of eBPAG1 that might be implicated in the assembly of HDs and their regulation, we have screened, using the yeast two-hybrid system, a human keratinocyte cDNA library. We found that in addition to the two hemidesmosomal components BPAG2 and the integrin ␤4 subunit, eBPAG1 binds in yeast and in vitro binding experiments to a protein belonging to a recently recognized family of proteins containing leucine-rich repeats (LRRs) and PDZ domain (LAP/LERP) 3 (reviewed in Ref. 24). We here describe the cloning of the full-length cDNA coding for this protein and the identification of several variants at the RNA level as well as their tissue expression. One of the splice variants of this novel protein corresponds to the recently described protein ERBIN, which has been found to associate via its PDZ domain with the COOH terminus of the tyrosine kinase transmembrane receptor Erb-B2 (25). Because ERBIN was shown previously to interact with the transmembrane tyrosine kinase receptor Erb-B2, which in turn associates with the integrin ␤4 subunit (26), we suggest that ERBIN provides a link between HD assembly and Erb-B2 receptor signaling.
Yeast two-hybrid analysis with two defined constructs in pAS2-1 or pACT2 (CLONTECH) was performed in two steps. After yeast transformation, double transformants were selected on agar synthetic complete (SC) medium lacking leucine and tryptophane. For each transformation eight colonies were arrayed in 96-well microtiter plates and then transferred onto agar SC medium lacking leucine and tryptophane (positive control), SC medium lacking leucine and tryptophane without adenine, and SC medium lacking leucine and tryptophane without histidine and supplemented with 2 mM 3-amino 1,2,4-triazole. Growth was estimated after a 5-day incubation at 30°C. Transactivation controls were systematically performed for each construct with the opposite vector without insert. Selected positive clones were further confirmed by colony-lift filter assay for ␤-galactosidase activity in the yeast strain Y190 as recommended by the supplier (CLONTECH).
Cloning of the Entire Coding Sequence of the Newly Identified Protein-The insert of the prey plasmid 67, isolated in the yeast two-hybrid screen of a human keratinocyte cDNA library with GAL4-BD-eBPAG1 1-1593 as a bait, was completely sequenced and compared with those in the GenBank TM data base. The 3Ј end of EST N40602 was identical to the 5Ј end of the insert of the prey plasmid 67, and its sequence was used to design the following primers: 1) 5Ј-CCTAAGT-CAACAGAACAA; 2) 5Ј-GTCATGTGGTTCAGTCCTATTAG; and 3) 5Ј-ACTGGGAAGTTTCCTCAGAATGTC for a 5Ј-RACE with the kit from Life Technologies, Inc. (2.0 version) according to manufacturer protocol. Total RNA was purified from NHK cells grown in low Ca 2ϩ (0.07 mM) medium with the SV total RNA isolation system (Promega). First strand cDNA synthesis was performed with 3 g of total RNA/reaction. The amplicons were cloned and sequenced. Then, the same strategy as described above was applied for a second round of 5Ј-RACE. The 3Ј end of EST AA427786 was completely identical to the 5Ј end of the longest 5Ј-RACE amplicon. The following three primers were designed based on the EST sequence and used in a second 5Ј-RACE: 1) 5Ј-TTCAACGAACCAATAGTCTCA; 2) 5Ј-TAGGAGGTCTTGAAGGTTTT-CACAGCTGCTGAAGTGAATTGCTTGA; and 3) 5Ј-TAGGAGGTCTTG-AAGGTTTTCAC. Sequencing of the cloned amplicons of the second nested PCR and data base analysis indicated that the translation initiation codon of the protein partially encoded by the insert of the prey plasmid 67 had been reached (see Fig. 1, A and B).
To clone the entire cDNA encoding the novel protein, RT-PCR (Superscript one-step RT-PCR system, Life Technologies, Inc.) was performed with total RNA (0.2-g/25-l reaction mixture) isolated from NHK cells with three sets of forward and reverse primers, 5Ј-TATGAC-TACAAAACGAAGTTTGTTTGTG and 5Ј-GGTTTGAACAGTTTTGAC-CATA TTC, 5Ј-GCGGGCTCAAGTTGCATTTG and 5Ј-GGATTTTGTT-TGGAGTGCCACA, and 5Ј-GCACTTGGAAAATGGAAACAAGTG and 5Ј-CCCCGCTTTTTTTGTCCACAGT, to amplify three overlapping cDNA fragments corresponding to the 5Ј end, the middle part, and the 3Ј end of the entire coding sequence, respectively (see Fig. 1, B and C). The RT-PCR amplicons were cloned into the mammalian expression vector pcDNA3 (Invitrogen) digested with EcoRV and completely sequenced. The three cDNA pieces were sequentially fused in pcDNA3 to obtain a full-length ORF. The cDNA sequence has been given the GenBank TM accession number AF276423. The isolated cDNA was later found to be a variant of ERBIN.
To attach a HA tag either at the NH 2 or COOH terminus of the identified protein, an adaptor coding for the HA tag and introducing an HpaI and a BsaBI site as well as an initiation and stop codon was inserted between the BamHI and NotI sites of pcDNA3, resulting in a vector called pcDNA3-HA1. The sequence of the forward primer was GATCCGCCACCATGTTAACCTATCCATATGATGTCCCAGATTATG-CTGATTAACATCGC, with the HpaI and BsaBI sequences underlined and the initiation and stop codons in bold. The reverse primer had a complementary sequence with terminal sequences creating BamHI and NotI overhangs. To tag the novel protein at its NH 2 terminus, the 5Ј end was amplified with the first set of RT-PCR primers (see above) and cloned into pcDNA3-HA1 digested with Bsa-BI. To tag its COOH terminus, the 3Ј end of its ORF was amplified with the forward primer of the third set of RT-PCR primers (see above) and the reverse primer 5Ј-GAGGAAACTTCTCGTACAATGA-TGAG and cloned into pcDNA3-HA1 digested with HpaI. Then the rest of the ORF was subcloned into these plasmids to obtain a fulllength ORF.
The COOH terminus of the identified protein, residues 932-1412, was fused in frame with the COOH terminus of enhanced GFP (EGFP) by transferring the SalI fragment from the yeast two-hybrid prey plasmid 67 into pEGFP-C3 (CLONTECH) digested with XhoI.
Other cDNA Constructs-cDNAs encoding various portions of the cytoplasmic domain of the integrin ␤4 subunit were derived from expression plasmids described in detail elsewhere (19) and subcloned into the eukaryotic expression vector pcDNA3-HA2 (kindly provided by E. Sander, The Netherlands Cancer Institute, Amsterdam) allowing the expression of HA-tagged fusion proteins. The expression plasmid pMT2 encoding a HA-tagged COOH-terminal truncation of eBPAG1 encompassing residues 1-2161, HA-eBPAG1 1-2161 , is described elsewhere. 4 Correctness of all constructs was verified by sequence analysis. Sequencing reactions were performed with Thermo Sequenase dye terminator cycle sequencing premix kit (Amersham Pharmacia Biotech) and analyzed on an ABI 373 (PerkinElmer Life Sciences).
Expression and Purification of Recombinant GST-fusion Proteins-Various cDNA fragments encoding residues 805-1179, 805-1412, and 1307-1412 of the novel protein were amplified by PCR with Pfu polymerase from variant 1 cDNA and cloned in frame with the COOH terminus sequence of glutathione S-transferase (GST) into pGEX-2T (Amersham Pharmacia Biotech) cut with SmaI. Escherichia coli BL21 (Novagen) was transformed with these constructs or the vector pGEX-2T without insert. Expression of GST or GST-fusion proteins was induced by adding 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside in the broth medium for 3 h. Bacteria were collected by centrifugation, resuspended in phosphate-buffered saline (PBS) supplemented with 1% Triton X-100 and 5 mM EDTA, and lysed by sonication. Purification of GST-fusion proteins was performed as described previously (19). The protein concentration was determined with the protein assay reagent from Bio-Rad using bovine serum albumin as a standard. For pull-down assays, glutathione-Sepharose beads (Amersham Pharmacia Biotech) with bound GST or GST-fusion proteins were equilibrated in CHAPS buffer (25 mM Hepes, 150 mM NaCl, 5 mM MgCl 2 , and 0.5% CHAPS (w/v), pH 7.5), and nonspecific binding sites were blocked by incubation with 1% heat-inactivated bovine serum albumin diluted in CHAPS buffer.
Semiquantitative Analysis of ERBIN Transcript(s) in Human Tissues-The level of expression of ERBIN transcripts in 24 human tissues was determined by using a human Rapid-Scan panel (OriGene Technologies). In each well of the 96-well microtiter plate, 25 l of the following PCR reaction mixture was added: 20 mM Tris-HCl, pH 8.4, 50 mM KCl, 1.5 mM MgCl 2 , 0.2 mM (each) dNTP mixture, 0.2 mM each of forward (5Ј-GTAGTGTGTCCTCCACAGCC) and reverse (5Ј-CCCCGC-TTTTTTTGTCCACAGT) primers, and 2.5 units of Taq polymerase (Life Technologies, Inc.). The PCR conditions were: 1 ϫ 94°C for 90 s, 35 ϫ (94°C for 30 s, 53°C for 30 s, and 72°C for 2 min), and 1 ϫ 72°C for 10 min in a GenAmp 9600 cycler (PerkinElmer Life Sciences). At the end of the PCR, two control reactions, without template and with 10 fg of pcDNA3 plasmid containing the 3Ј end of ERBIN cDNA, were performed with the same PCR reaction mixture (kept on ice) and cycling conditions.
Southern and Northern Blotting-The probe was labeled with digoxigenin (DIG) antigen using the PCR DIG labeling mix (Roche) with the forward and reverse primers 5Ј-TCCCCATTGTTCTCCTAGACAA and 5Ј-GAGGAAACTTCTCGTACAATGATGAG, respectively, from a plasmid derived from the yeast two-hybrid prey plasmid 67 and lacking both primer sequences used in the semiquantitative analysis of the transcripts from a human Rapid-Scan panel (see above). Labeling and Southern blotting were carried out exactly as described by the manufacturer. The Southern blot was developed with nitroblue tetrazolium salt and 5-5-bromo-4-chloro-3-indolyl phosphate as substrates for the alkaline phosphatase coupled to the anti-DIG Ab. For Northern blotting, a 3.8-kilobase pair EcoRI fragment from ERBIN cDNA was used as template to synthesize by PCR an antisense DIG-labeled DNA probe with the reverse primer 5Ј-CCCCGCTTTTTTTGTCCACAGT. Northern blotting was carried out exactly as described by the manufacturer (Roche). The DIG-labeled probe was detected on the Northern blot with nitroblue tetrazolium salt and 5-5-bromo-4-chloro-3-indolyl phosphate as substrates for the alkaline phosphatase coupled to the anti-DIG Ab.
In Vitro Transcription/Translation-The different constructs in pcDNA3 or pcDNA3-HA1 were transcribed and translated in vitro with the kit TNT T7 quick-coupled transcription/translation system (Pro-FIG. 1. Cloning of the full-length ORF encoding ERBIN. A, a schematic presentation of human ERBIN cDNA with the position of some restriction enzyme sites on the top. The first 162 nucleotides come from EST AI022650, and the remaining come from our own sequencing, which has been given the GenBank TM accession no AF276423. The ORF corresponds to the gray box. B, the cloning strategy applied to obtain the complete ORF sequence of ERBIN cDNA from the yeast two-hybrid clone. 5Ј-RACE from human keratinocyte total RNA and GenBank TM EST sequences coding for the same cDNA, as deduced by comparison of the overlapping sequences, were used to quickly reach the 5Ј end of the ORF. Then, three RT-PCRs were performed from human keratinocyte total RNA to finally construct a full-length ORF. Sequence heterogeneity is indicated by the [caret]. C, RT-PCRs 1-3 (lanes 1-3) were separated on 0.8% agarose gel and stained with ethidium bromide. RT-PCR3 gave three distinct bands. The size in kilobase pairs of some DNA marker bands (lane m) is indicated. D, a schematic representation of the domain organization in ERBIN. Sixteen LRRs and a PDZ domain at the NH 2 and COOH terminus, respectively, flank the internal portion of the protein, the most of which does not resemble any other protein.

mega) as indicated by the supplier.
Production and Purification of Polyclonal Antibodies-A peptide corresponding to the 16 COOH-terminal residues of the novel protein (NH 2 -TFQNTVELIIVREVSS-COOH) was synthesized, isolated to 70% purity, and conjugated to Keyhole Limpet hemocyanin with glutaraldehyde by the company Eurogentech (Belgium). Two rabbits were immunized with the prepared antigen and bled 3 weeks after the third immunization at Eurogentech. The antiserum from one rabbit was precipitated with ammonium sulfate to 50% saturation (33), and IgG was isolated from this fraction after dialysis by chromatography on Protein A-Sepharose CL-4B according to manufacturer protocol (Amersham Pharmacia Biotech). Isolated IgG was purified by affinity chromatography (33) on a column prepared by coupling 10 mg of COOHterminal peptide dissolved in 10 mM sodium phosphate buffer, pH 7.5, to N-hydroxy-succinimide-activated Sepharose 4 Fast Flow (Amersham Pharmacia Biotech) at 4°C for 15 h. Eluted anti-COOH-terminal peptide IgG (anti-ERBIN Ab) was finally concentrated by filtration on an Ultrafree-4 centrifugal filter unit (Millipore), aliquoted, and stored frozen until use for Western blotting. When skin sections or cultured keratinocytes were subjected to immunofluorescence or immunoperoxidase staining procedures, no specific signal could be detected with the anti-ERBIN Ab because of significant background.
Transfection Experiments-For transfection, cells were first grown to 50 -80% confluence in six-well tissue culture plates (Falcon, Becton Dickinson). The COS-7 cells were transfected with the DEAE-dextran method, and the other epithelial cells with the cationic lipid Lipo-fectAMINE 2000 (Life Technologies, Inc.) and assayed for gene expression after 36 h of culture as described previously (18).

RESULTS
Yeast Two-hybrid Screen with the NH 2 Half of eBPAG1 Identifies the Hemidesmosomal Proteins BPAG2 and the Integrin ␤4 Subunit as Interactors-The NH 2 terminus of desmoplakin and plectin mediates their recruitment to plasma membrane sites by interacting with transmembrane glycoproteins (6,8,19). It is therefore probable that the same region is responsible for the subcellular distribution of eBPAG1, which belongs to the same protein family. To identify proteins that interact with the NH 2terminal half of eBPAG1 and may thus mediate its localization into HDs, we used the yeast two-hybrid system. Residues 1-1593 of eBPAG1 were fused to GAL4-BD in pAS2-1 to screen a human keratinocyte cDNA library in the PJ69 -4A yeast strain. Upon selection of 3.3 million independent yeast colonies, 59 colonies grew on selective media lacking adenine and histidine. In agreement with previous studies that provided evidence for an interaction of eBPAG1 with the integrin ␤4 subunit and BPAG2 (9, 28), we identified several cDNAs for the cytoplasmic domain of the integrin ␤4 subunit among the positives. All of them encoded the COOH terminus of ␤4, the shortest starting at residue 1315 (GenBank TM accession no. CAA37656), indicating that this region is critical for the interaction with eBPAG1. Furthermore, one cDNA encompassing the entire cytoplasmic domain of BPAG2 was isolated. These findings provide additional support for the assumption that eBPAG1, by means of interactions with the cytoplasmic domain of both BPAG2 and the ␤4 subunit of the ␣6␤4 integrin, connects the intermediate filament network to the basal plasma membrane (2,9).
Isolation of ERBIN as Interactor of eBPAG1-Among the other clones identified in the yeast two-hybrid screen, two of them, clones 67 and 68, coded for an ORF of 481 residues corresponding to the COOH terminus of a novel protein. Because the partial sequence lacked an ATG start codon, we performed two 5Ј-RACE experiments using EST sequences to obtain a full-length ORF sequence (Fig. 1B). To construct a full-length cDNA clone, three fragments spanning each about one third of the ORF and containing a unique restriction enzymes site in their overlapping sequences were amplified by RT-PCR from total RNA isolated from human keratinocytes. Comparison of the sequence of the RT-PCR 1 and 2 fragments with the 5Ј-RACE and EST sequences indicated a perfect match except for a stretch of 15 nucleotides (nt) (from nt 1611 to 1625), which was missing in the first 5Ј-RACE fragment (indicated by [caret] in Fig. 1B). Surprisingly, the third RT-PCR produced three distinct bands (see Fig. 1C). Cloning and sequencing of these bands revealed that the complexity was even greater (see below). The longest ORF, variant 1, which contains the 3Ј-end sequence of the original yeast two-hybrid clone 67, codes for a protein of 1412 residues with a predicted molecular mass of 158,238 Da. An initial Blast search (34) with the complete sequence identified a single protein with significant homology, rat densin-180 (35). The size of the proteins and the organization of their domains were strikingly similar. Protein sequence analysis (36) revealed the presence of two motifs at both extremities of the protein: 16 LRRs, each 23 amino acids long, close to the NH 2 terminus (from residues 23 to 391) and a PDZ domain at the COOH terminus (from residues 1321 to 1409) (Fig. 1D). Nevertheless, two observations did not support the hypothesis that the identified protein was the human homologue of rat densin-180: 1) the lack of homology in the middle part of the proteins (24% identity) in comparison with the LRRs (71% identity) and PDZ domains (68% identity) and 2) their different tissue expression (see below).
Shortly after we submitted the cDNA sequence of the identified protein in the GenBank TM data base (accession no. AF276423), the same protein was identified in a yeast twohybrid screen of a mouse kidney cDNA library with the tyrosine kinase transmembrane receptor Erb-B2/HER2 as a bait (25  Bank TM accession no. AF263744) that corresponds to our variant 2 (see below) and termed the protein ERBIN. Therefore, we will also use this terminology in the rest of the article. Chromosomal Localization of ERBIN-Based on the partial draft sequence of the human genome, the ERBIN gene is located on the long arm of chromosome 5, reference interval D5S427-D5S647 (69.6 -74.cM), physical position 322.02 cR3000 (GB4 map). ERBIN mRNA comprises at least 26 exons, the borders of which have been deduced by comparison with the consensus intron/exon and acceptor/donor sequences (37) ( Table I).
Characterization of Several Splice Forms of ERBIN that Differ in their Coding Sequence upstream of and within the COOHterminal PDZ Domain-Seven ERBIN variants have been identified thus far, and these variations are probably caused by alternate splicing between exons 21 and 26 (Fig. 2). Variants 2, 3, 6, and 7 differ from the longest isoform, variant 1, in a region situated upstream of the PDZ domain. Interestingly, variants 4 and 5, which lack exons 24 or 25, respectively, have a truncated PDZ domain. Finally, we also found a polymorphic difference at position 3909 A/G, leading to the amino acid substitution K1207E.
ERBIN Is Expressed in Keratinocytes as is eBPAG1-To assess whether ERBIN was expressed in the same keratinocytes that also express eBPAG1, we first performed Northern blot analysis. The ERBIN transcript was expressed as a doublet of ϳ6900 and 7400 nt in human keratinocytes and the HaCat cell line, respectively (Fig. 3). Next, we performed Western blot analysis by using affinity-purified Abs directed against the COOH terminus of ERBIN, which recognizes the GST-ERBIN 1307-1412 fusion protein (Fig. 4A). When tested on NHK cell extracts, the anti-ERBIN Ab identified one major protein with an M r of 200,000 in differentiated cells (Fig. 4B), but only small amounts could be visualized in undifferentiated keratinocytes after long film exposures. The M r is higher than the theoretical molecular mass deduced from the cloned cDNAs, which range from 146,019 to 158,238 Da (Fig. 2). However, when the cDNAs for various variants were translated in vitro and their products were separated on SDS-PAGE, the isoforms migrated with an M r ϳ180,000 -200,000 (Fig. 4C), suggesting that the above lower electrophoretic motility was an intrinsic property of the protein(s).
Tissue Expression of ERBIN Transcripts-To extend the data of the expression of ERBIN in man, we performed a semiquantitative PCR analysis with cDNA synthesized from total RNA isolated from 24 different tissues that had been normalized to obtain the same level of amplification of the ␤-actin transcript. Primers annealing close to the variable region of the ERBIN transcripts were chosen for the PCR to get further insight into the expression of the different variants. Southern blotting using a probe encompassing nt 3809 -4540 ( Fig. 1) was carried out to confirm that the PCR-amplified fragment(s) did indeed correspond to the ERBIN sequence. Based on the results presented in Fig. 5, different ERBIN   FIG. 2. Variants of ERBIN. The sequence varying in the variants 2-7 (of decreasing length) in comparison with that of the variant 1 (the longest) is presented. On the extreme right hand side the exon (Ex) number (see Table I transcripts are expressed in most human tissues; they seem particularly abundant in brain, heart, kidney, muscle, and stomach (Fig. 5, lanes 4, 5, 6, 12, and 13, respectively).
The PDZ Domain of ERBIN Is Not Necessary for the Interaction with eBPAG1-In independent yeast two-hybrid transformation experiments we first confirmed that the region of ERBIN from residues 932 to 1412, GAL4-AD-ERBIN 932-1412 , that was identified in the yeast two-hybrid transformation screen was able to interact with residues 1-1593 of eBPAG1, eBPAG1  . Furthermore, binding also occurred with a construct containing residues 1-887 of eBPAG1, eBPAG1 1-887 , from which the sequences encoding the coiled-coil domain of BPAG1 were deleted (Table II). Next, because PDZ-containing proteins are known to mediate the interaction between proteins (38), we assessed whether the association between ERBIN and eBPAG1 involved the PDZ domain. Two ERBIN cDNA fragments, coding for residues 932-1159 and 1169 -1412, respectively (see Fig. 1), were fused in frame to the GAL4-AD sequence in the vector pACT2. In independent yeast two-hybrid experiments, only the region of ERBIN encompassing residues 932-1159, which is located upstream of the PDZ domain and does not exhibit significant homology (21% identity) to the sequence of densin-180, was found to interact with both eBPAG1 1-1593 and eBPAG1   (Table II). These findings indicate that the PDZ domain of ERBIN and the coiled-coil domain of eBPAG1 are dispensable for their interaction.
ERBIN Interacts with the Integrin ␤4 Subunit-The cytoplasmic tail of the ␤4 subunit is crucial for its interaction with various components of HDs and their formation (2,39). Therefore, we tested the ability of the COOH terminus of ERBIN (residues 932-1412) to interact with the integrin ␤4 subunit in the yeast two-hybrid system. Growth was supported selectively when yeast was co-transformed with GAL4-AD-ERBIN 932-1412 and GAL4-BD-␤4 constructs, leading to the expression of the COOH terminus of the ␤4 integrin subunit. Specifically, the COOH terminus of ERBIN, GAL4-AD-ERBIN 932-1412 , interacted with ␤4 fragments containing either the third and fourth FN-III repeats and the extreme COOH terminus (residues 1457-1752) (GenBank TM accession no. CAA37656) or the fourth FN-III repeat and the extreme COOH terminus (residues 1570 -1752) of ␤4. No interaction was observed with the first two FN-III repeats and the connecting segment (residues 1115-1457) or with the COOH extremity of ␤4 (residues 1667-1752). Furthermore, as in eBPAG1, residues 932-1159 located upstream of the PDZ were essential for interaction with ␤4 (Table II).
ERBIN Forms a Complex with Either eBPAG1 or the Integrin ␤4 Subunit in Vitro-The ability of ERBIN to associate with ␤4 and eBPAG1 was tested in in vitro binding assays. Two COOHterminal regions of ERBIN encompassing or not the PDZ domain (from residues 805 to 1179 and from residues 805 to 1412, respectively) were expressed as fusion proteins with GST (GST-ERBIN 805-1179 and GST-ERBIN 805-1412 , respectively) and immobilized on glutathione-Sepharose beads. HA-tagged recombinant proteins of eBPAG1 (from residues 1 to 2161, HA-eBPAG1 1-2161 ) and of the ␤4 cytoplasmic domain (from residues 1115 to 1752 and from residues 1115 to 1328, HA-␤4 1115-1752 and HA-␤4 1115-1328 , respectively) were expressed by transfecting COS-7 cells and used as fluid phase ligands. As illustrated in Fig. 6, recombinant HA-eBPAG1 1-2161 encompassing the NH 2 terminus and the coiled-coil region of eBPAG1 and recombinant HA-␤4 1115-1752 representing a large portion of the cytoplasmic domain of ␤4 bound to both GST-ERBIN 805-1179 and GST-ERBIN 805-1412 but hardly to GST alone. Furthermore, recombinant HA-␤4 1115-1328 , which lacks the region containing sequences critical for interaction with ERBIN in yeast, did not associate with any of the GST-fusion proteins tested. Together, consistent with the yeast two-hybrid results, the COOH-terminal region of ERBIN, containing or not the PDZ domain, forms a complex with eBPAG1 and the ␤4 cytoplasmic domain.
Expression of ERBIN in Epithelial Cell Lines-The ORF sequence of the ERBIN identified in the yeast two-hybrid screen encoding residues 932-1412 was fused to the COOH terminus of EGFP to study the localization of the fusion protein in various cell lines: COS-7, 804 G, PtK2, HaCat, and PA-JEB-␤4 keratinocytes. In all transfected cells, the EGFP-ERBIN 932-1412 fusion protein was found either diffusely distributed over the entire cytoplasm (Fig. 7A) or at the lateral cytoplasmic membrane at sites of cell-cell contact, especially in PtK2 cells (Fig. 7E) and PA-JEB-␤4 keratinocytes (Fig. 7C). However, in both 804G cells (data not shown) and PA-JEB-␤4 keratinocytes the expressed protein was not obviously colocalized with either the ␣6␤4 integrin (Fig. 7, A-D) or eBPAG1 (data not shown), the staining pattern of which was characteristic for hemidesmosomal-like structures or stable anchoring contacts (17,18). Finally, a nuclear staining was also occasionally observed. Similar results were obtained with full-length ERBIN (variant 2) HA-tagged at its COOH terminus (data not shown).

FIG. 3. Northern blot analysis of ERBIN transcripts in human keratinocytes.
Total RNA (10 g) isolated from human keratinocytes grown in low calcium (0.07 mM) (lanes 1) and high calcium (1.87 mM) media for 4 days after low to high calcium switch (lanes 2) or from HaCat cells (lanes 3) was size-fractionated on 1% formaldehyde agarose gel and stained with ethidium bromide (left half). The size in kilobases of RNA markers (lane m) is indicated. RNA was then transferred to Hybond N ϩ membrane. Northern blot analysis was performed at high stringency with a mono-stranded DIG-labeled DNA probe encompassing nt 4568 -889 of ERBIN cDNA (see Fig. 1). The hybridized DIGlabeled probe was immunologically detected using nitroblue tetrazolium salt and 5-5-bromo-4-chloro-3-indolyl phosphate as substrate for alkaline phosphatase conjugated to anti-DIG Ab (right half).

DISCUSSION
In a yeast two-hybrid screen of a human keratinocyte cDNA library with the NH 2 -terminal half of eBPAG1, we have identified the COOH terminus of a protein resembling rat densin-180 (35). Cloning of the full-length ORF demonstrated that this protein of 1412 residues with a predicted molecular mass of 158,238 Da has a domain organization identical to that of densin-180, with 16 LRRs toward the NH 2 terminus and a PDZ domain at the COOH terminus. However, both the low homology of their internal sequences and the very different tissue expression of their mRNA (35) indicate that these proteins are not orthologues. During the preparation of the article, one of the isolated variants, variant 2, was found to be identical to a very recently described protein, termed ERBIN. ERBIN was identified in a yeast two-hybrid screen of a kidney cDNA library with the tyrosine kinase transmembrane receptor Erb-B2/HER2 fused to GAL4-BD as a bait (25). Thus, through a yeast two-hybrid screen we have independently isolated ERBIN as interactor of eBPAG1.
The ERBIN gene is located on chromosome 5 within the reference interval D5S427-D5S647, in a region that has not yet been linked to a genetic disease. Northern blot analysis of the total RNA isolated from human keratinocytes demonstrated the existence of two major ERBIN transcripts, whereas RT-PCR experiments show the presence of several variants in most human tissues. Sequence analysis of seven variants indicates   FIG. 4. Detection and expression of ERBIN and its variants. A, whole extracts (100 g) from bacteria transformed with the GST-ERBIN 1307-1412 construct (lanes 1 and 2) or pGEX-2T vector (lanes 3 and 4) in which protein expression had been induced with isopropyl-1-thio-␤-D-galactopyranoside for 3 h (lanes 1, 3) or not induced (lanes 2 and 4) were size-fractionated on 12.5% SDS-PAGE and blotted to an Immobilon-P membrane. Western blot analysis (development with diaminobenzidine) indicated that anti-ERBIN Ab bound to GST-ERBIN 1307-1412 fusion protein (lane 1, and also 2 because of the leaky promoter), an apparent proteolytic fragment (M r of 32,500) of GST-ERBIN 1307-1412 (lane 1), and a few additional endogenous proteins in all lanes but not GST (lane 3). B, immunoblot analysis (ECL development) with anti-ERBIN Ab of extracts (40 g/lane) from undifferentiated (lanes 1 and 3) or differentiated (lanes 2 and 4) NHK cells resolved on 7.5% SDS-PAGE. Membranes were incubated without (lanes 1 and 2) or with (lanes 3 and 4) 0.1 mM of ERBIN COOH-terminal peptide. Addition of the peptide abolished the specific recognition by the Ab of a protein with an M r of 200,000 (arrow). The same immunoreactive protein could also be detected in extracts from undifferentiated keratinocytes after a long film exposure. The Ab also specifically detected a smaller protein (M r of 67,000, star), which might represent a proteolytic fragment of ERBIN. C, cDNAs for four ERBIN variants cloned in pcDNA3 were translated in vitro and radiolabeled with [ 35 S]methionine. Aliquots (4 l) of the in vitro translation mixture were separated on 5% SDS-PAGE and transferred to Immobilon-P membrane, and proteins were detected by autoradiography. Lanes 1-4 correspond to the variants 7, 3, 2, and 1, respectively (see Fig. 2). The position and size (10 Ϫ3 ) of protein markers are indicated.
FIG. 5. Semiquantitative analysis of the expression of ERBIN transcripts in human tissues. The 3Ј end of the ERBIN ORF was amplified by RT-PCR from total RNA of human keratinocytes with two primers annealing to nt 3538 -3557 and 4568 -4547 (see Fig. 1). Amplified cDNA was size-fractionated on 1% agarose gel and stained with ethidium bromide (lane 1). The size in kilobase pairs of some marker bands (lane m) is indicated on the left. DNA from lanes m and 1 were transferred to a Hybond N ϩ membrane, and Southern blot analysis was performed with a DIG-labeled ERBIN cDNA fragment consisting of nt 3809 -4540 (see Fig. 1). After hybridization at high stringency, the DIG-labeled probe was immunologically detected using nitroblue tetrazolium salt and 5-5-bromo-4-chloro-3-indolyl phosphate as a substrate for alkaline phosphatase conjugated to anti-DIG Ab. that they most likely derive from alternative splicing. These variants differ in a region situated upstream of and within the PDZ domain. Interestingly, the densin-180 gene is subject to similar alternative splicing, and its expression is developmentally regulated (40). However, it remains to be established whether all variants of the ERBIN mRNA are translated, especially those with a truncated PDZ domain. Only one variant appears to be expressed in differentiated human keratinocytes, whereas a doublet around M r 180,000 was systematically detected in several human tissues analyzed by Western blotting with anti-ERBIN Ab (25).
ERBIN belongs to a family of proteins called LAP/LERP that comprises, in addition to ERBIN (25) and densin-180 (rat) (35), Scribble (human and Drosophila) (41) and Let-413 (Caenorhabditis elegans) (42). The domain organization of all of them is strikingly similar: 16 LRRs and one or four PDZ domains at the NH 2 and COOH terminus, respectively (reviewed in Ref. 24). In addition to these two structural features, they all contain a short conserved region of 39 residues next to LRRs, termed the LAP/LERP-specific domain. Putative Src homology 3 and WW binding sites in ERBIN have also been identified (25). Furthermore, there are two potential immunoreceptor tyrosine-based inhibitory motifs (ITIM) (43) before the PDZ domain, 1268 ANYSQI, and 1291 IDYLML, which might attract, when phosphorylated, the tyrosine protein phosphatase SHP-1, even though Ala 1268 is not typical for an immunoreceptor tyrosine-based inhibitory motif (44). Although densin-180 is a putative transmembrane protein that is heavily O-sialoglycosylated (35), in accordance with the analysis of Borg et al. (25), we found no typical transmembrane domain or signal peptide sequence in either ERBIN or densin-180 by using several predictive programs (45)(46)(47).
The PDZ domain of ERBIN has been found recently to interact with the COOH terminus of Erb-B2. This interaction seems to be responsible for the basolateral localization of Erb-B2 in epithelia (25). Interestingly, ERBIN seems to associate with dephosphorylated Erb-B2 and is phosphorylated by Her1/2 (25) upon epidermal growth factor stimulation of COS-1 cells (25). In extension to these findings, our yeast two-hybrid results indicate that ERBIN has the potential to interact with the NH 2 -terminal region of eBPAG1 and the cytoplasmic domain of the integrin ␤4 subunit. Sequences important for these associations reside within the globular end domain of eBPAG1 and probably the fourth FN-III repeat close to the COOH terminus of the ␤4 subunit. A region of ERBIN encompassing residues 932-1159 mediates binding to both proteins. These residues are located upstream of the variable region subjected to alternative splicing and the PDZ domain.
The ability of ERBIN to form a complex with eBPAG1 or the ␤4 integrin subunit was confirmed in in vitro binding assays. Recombinant HA-␤4 1115-1752 representing a large portion of the ␤4 tail and recombinant HA-eBPAG1 1-2161 containing the NH 2 terminus and coiled-coil domain of eBPAG1 bound to both the GST-fusion proteins, GST-ERBIN 805-1179 and GST-ERBIN 805-1412 , whereas an HA-␤4 1115-1328 recombinant lacking the third and fourth FN-III repeats of the ␤4 tail did not. Consistent with our yeast two-hybrid assays, the in vitro binding experiments indicate that 1) the PDZ domain of ERBIN is not needed for the association with either the ␤4 integrin subunit or eBPAG1, and 2) sequences contained within the COOH-terminal half of the ␤4 integrin subunit tail are crucial for its association with ERBIN.
To assess the potential of ERBIN to become localized in HDs, we performed transient transfection studies. In various epithelial cells transfected with constructs encoding either full-length ERBIN or its COOH-terminal half, the distribution of the expressed proteins in the cytoplasm was diffuse. However, there  Ϫ Ϫ Ϫ a ϩ and Ϫ indicate growth and no growth, respectively, on selective media, tested exactly as described under "Experimental Procedures." Each experiment was independently repeated twice and consisted of the analysis of eight colonies on selective media for each co-transformation.
was also a clear staining of the plasma membrane that was reminiscent of the distribution of endogenous ERBIN in colon epithelial cells (25). These findings indicate that the localization of ERBIN at the plasma membrane does not reflect an intrinsic property of the protein but most likely depends on its association with other proteins. In an immortalized keratinocyte cell line and in 804G cells that form HDs in culture, expressed ERBIN was not colocalized in HD-like structures, suggesting that ERBIN is not a structural component of these complexes. Nevertheless, it seems very unlikely that the interactions detected among eBPAG1, ␤4, and ERBIN are not physiologically relevant because of 1) the involvement of other LAP/ LERP proteins in protein and receptor clustering and scaffolding (24,48), 2) the ability of Erb-B2 to interact with the ␤4 cytoplasmic domain and cooperatively influence cell behavior (49) (see below), and 3) the ability of ERBIN to associate not only with the integrin ␤4 subunit but also with another hemidesmosomal protein, eBPAG1. Therefore, the observed interaction of ERBIN with the integrin ␤4 subunit or eBPAG1 may be transient and thus regulated in mammalian cells and may occur in cytosolic compartments other than HDs.
Scribble and Let-413 proteins seem to have a crucial role in the proper assembly of adherens junctions and epithelial cell polarity in Drosophila and C. elegans, respectively (41, 42). In the Let-413 mutants of C. elegans, epithelial cell polarity is disturbed with abnormal and mislocalized adherens junctions and a disorganized actin cytoskeleton (42). Moreover, in Drosophila Scribble regulates epithelial cell growth in conjunction with Lethal giant larvae (Lgl) and Discs large (Dlg), two proteins with which Scribble is colocalized at the septate junction (the Drosophila homologue of vertebrate tight junction) (41). Like Lgl and Dgl, Scribble may act as a tumor suppressor, linking the maintenance of cell polarity to growth control. Thus, in analogy to the latter, it is conceivable that ERBIN is also implicated in pathways that critically regulate cell polarity and growth in mammalian epithelia. In this regard, the observation of an occasional detection of ERBIN in the nucleus of transfected cells raises the intriguing possibility of an additional nuclear function, in analogy to other proteins that contain PDZ domains such as syntenin (50) and the tight junction protein ZO-1 (51). Both of them lack nuclear localization sequences such as ERBIN but can shuttle between the plasma membrane and the nucleus.
Expression of the integrin ␤4 subunit has been found recently to promote invasiveness of a colon carcinoma cell line (52,53) as well as of NIH3T3 cells transformed by the Erb-B2 oncogene (26). This effect depends on the activation of phosphoinositide-3-OH kinase (49,54). In the transformed NIH3T3 cell line Erb-B2 was found to associate with the ␤4 cytoplasmic domain, and both proteins were required for constitutive phosphoinositide-3-OH kinase activation. Interestingly, the region of the ␤4 cytoplasmic domain important for the interaction with Erb-B2 is located close to the transmembrane domain (49) and is thus distinct from that involved in binding to ERBIN. In view of the properties of other LAP/LERP proteins such as the targeting of secretory vesicles to the basolateral membrane (reviewed in Refs. 24, 55, and 56), the ability of ERBIN to interact with both eBPAG1, and the ␤4 cytoplasmic domain suggests that this protein regulates the assembly of HDs by either bringing various scaffolding proteins together or by affecting the interactions between them. Alternatively, it might be involved in signaling pathways mediated by the Erb-B2/ Her2 receptor and/or the integrin ␤4 subunit.
Future investigations will be necessary to better characterize the ability of ERBIN to interact with various hemidesmosomal components and to define the binding specificity of both the LRRs and LAP/LERP-specific domain. Moreover, the role of the ERBIN variants that result from alternative splicing either upstream of or in the PDZ domain remains unclear. These studies have important implications for our understanding not only of HD assembly and thus cell adhesion but also, perhaps, for mechanisms regulating cell growth, differentiation, and migration.