Type XIII collagen is identified as a plasma membrane protein.

The complete primary structure of the mouse type XIII collagen chain was determined by cDNA cloning. Comparison of the mouse amino acid sequences with the previously determined human sequences revealed a high identity of 90%. Surprisingly, the mouse cDNAs extended further in the 5' direction than the previously identified human clones. The 5' sequences contained a new in-frame ATG codon for translation initiation which resulted in elongation of the N-terminal noncollagenous domain by 81 residues. These N-terminal sequences lack a typical signal sequence but include a highly hydrophobic segment that clearly fulfills the criteria for a transmembrane domain. The sequence data thus unexpectedly suggested that type XIII collagen may be located on the plasma membrane, with a short cytosolic N-terminal portion and a long collagenous extracellular portion. These sequence data prompted us to generate antipeptide antibodies against type XIII collagen in order to study the protein and its subcellular location. Western blotting of human tumor HT-1080 cell extract revealed bands of over 180 kDa. These appeared to represent disulfide-bonded multimeric polypeptide forms that resolved upon reduction into 85-95-kDa bands that are likely to represent a mixture of splice forms of monomeric type XIII collagen chains. These chains were shown to contain the predicted N-terminal extension and thus also the putative transmembrane segment. Immunoprecipitation of biotinylated type XIII collagen from surface-labeled HT-1080 cells, subcellular fractionation, and immunofluorescence staining were used to demonstrate that type XIII collagen molecules are indeed located in the plasma membranes of these cells.

The collagen family of proteins presently includes 19 types of collagen, and several additional proteins have collagen-like domains (1,2). The collagens can be divided into two subgroups in terms of their structural and functional characteristics, the fibril-forming and the nonfibril-forming collagens. Members of the former group, i.e. types I-III, V, and XI, aggregate into prominent fibrillar structures in many collagen-containing tissues. These molecules are structurally homologous and char-acterized by a long, uninterrupted collagen triple helix. The other collagens are unable to form fibrils, and they show considerable diversity in structure, macromolecular organization, tissue distribution, and function. One common feature is that they all have one or more interruptions in the collagenous sequence. Several subfamilies can be distinguished among the nonfibril-forming collagens as follows: the network-forming collagens (types IV, VIII and X), fibril-associated collagens with interrupted triple helices (which include types IX, XII, XIV, XVI and XIX), a beaded filament-forming collagen (type VI), the family of types XV and XVIII collagens, and a collagen with a transmembrane domain (type XVII). The last mentioned collagen is distinct from the other family members, because it is not secreted into the extracellular matrix.
Type XIII collagen is a nonfibrillar collagen that has so far been characterized via human cDNA and genomic clones (2)(3)(4)(5)(6), but its function is still unknown. The predicted ␣1(XIII) collagen polypeptide consists of short N-and C-terminal noncollagenous domains, termed NC1 1 and NC4, respectively, and three collagenous domains, COL1-3, separated by the noncollagenous domains NC2 and NC3. A striking feature of type XIII collagen is that sequences corresponding to nine exons of the human gene undergo complex alternative splicing during the processing of primary transcripts, which can be predicted to affect the structures of the COL1, NC2, COL3, and NC4 domains (2)(3)(4)(5)(6)(7)(8)(9). The length of the human ␣1(XIII) collagen chains has been estimated to vary between 614 and 526 amino acid residues, depending on the composition of alternatively spliced exons involved (7)(8)(9). The functional significance of this complex alternative splicing is not known, however. In situ hybridization experiments with human tissues indicate that type XIII collagen mRNAs are found at least in fetal bone, cartilage, intestine, skin, striated muscle, and in the placenta (9,10). In fact, they have been found in all the tissues examined so far and appear to be expressed in low amounts in virtually all connective tissue-producing cells. This suggests that type XIII collagen may serve a general function in connective tissue.
We have cloned and characterized the primary structure of mouse type XIII collagen. Comparison of mouse and human amino acid sequences indicated 90% identity. Surprisingly, characterization of the mouse clones, which extended further in the 5Ј direction than in previously isolated human clones, suggested that type XIII collagen is a plasma membrane protein. This was supported by cell fractionation analyses and immunofluorescence staining of human HT-10180 cells known to express type XIII collagen.

EXPERIMENTAL PROCEDURES
Cloning of Mouse Type XIII Collagen cDNA Sequences-The mouse sequences corresponding to exon 21 of the human type XIII collagen gene (6) were amplified from mouse genomic DNA using primers derived from the human type XIII collagen sequence. Additional cDNA clones were generated by performing reverse transcriptase-polymerase chain reaction using nested primers and poly(A ϩ ) RNAs extracted from the gut of 2-7-day-old newborn mice (strain B6) with guanidine thiocyanate (11) and oligo(dT)-cellulose chromatography (12). 3Ј-Rapid amplification of cDNA ends-polymerase chain reaction (13) was employed to isolate cDNA sequences extending beyond the translational stop codon. To isolate the extreme 5Ј end of the mouse type XIII collagen mRNA, a cDNA library in gt10 vector (Stratagene) was prepared from newborn mouse gut RNA using random hexamers as primers and the You-Prime-cDNA synthesis kit (Amersham Pharmacia Biotech), according to the manufacturer's protocol, and screened with the previously identified clones under stringent conditions (14). The obtained cDNA clones were sequenced in both directions by the dideoxynucleotide method (15) using Sequenase enzyme (U. S. Biochemical Corp.) or T7 polymerase (Amersham Pharmacia Biotech). Nucleotide and amino acid homology comparisons were carried out against the GenBank TM , EMBL, PIR, and Swiss-Prot data bases at NCBI (National Institutes of Health) using the BLAST network service (16). The search for functional patterns of amino acid sequences was carried out using the PROSITE data base (17).
Preparation and Affinity Purification of Antipeptide Antibodies-Synthetic peptides corresponding to residues 21-34 (GAPGTVAL-VAARAE) in the NC1 domain and residues 451-472 (EMVDYN-GNINEALQEIRTLALM) in the NC3 domain of human type XIII collagen ( Fig. 1) were synthesized with an automated Applied Biosystems 433A peptide synthesizer (Department of Biochemistry, University of Oulu, Finland). The sequence of the reversed-phase high pressure liquid chromatography purified peptides was confirmed by peptide sequencing (Applied Biosystems 477A). Five mg of the purified peptides were coupled to keyhole limpet hemocyanin (Sigma) by a standard procedure using glutaraldehyde (18). For immunization, the coupled peptide solutions were injected subcutaneously into four rabbits with complete Freund's adjuvant followed by booster injections with incomplete Freund's adjuvant at intervals of 14 days. The sera were analyzed by enzyme-linked immunosorbent assay (Vectastain, Vector Labs) using the uncoupled peptide as an antigen. Positive antisera were subsequently analyzed by immunoblotting and immunoprecipitation of recombinant human type XIII collagen expressed in insect cell lysates (lysates were a kind gift of A. Snellman and H. Tu, Department of Medical Biochemistry, University of Oulu). The expression of prolyl 4-hydroxylase in insect cells has been previously described (19), and its co-expression with type XIII collagen followed the protocol described for the production of type III collagen in the same expression system (20). Thereafter two sera named anti-XIII/NC1-1 and anti-XIII/NC3-1, one specific for each peptide, were selected for use in the experiments. The controls included the use of preimmune serum and antigen competition controls. In the latter the antibody was incubated with 10 times molar excess of the peptide for 2 h at 4°C before use in the experiments. Affinity purified antibodies were used in all experiments, except where otherwise stated.
The antibodies were purified by affinity chromatography by coupling the peptide antigens to epoxy-activated Sepharose 6B according to the manufacturer (Amersham Pharmacia Biotech). The antisera were diluted 1:5 with 20 mM K 2 HPO 4 , 0.1 M NaCl, pH 7.0, and applied to the columns, which were subsequently washed with 20 mM K 2 HPO 4 , 0.5 M NaCl, pH 7.0, and eluted with 30 mM glycine HCl, pH 2.9, and thereafter with 100 mM triethylamine, pH 11.0. The protein-containing fractions were detected by absorbance at 280 nm, immediately neutralized with 0.2 volumes of 2 M Tris-HCl, pH 7.5, pooled, and concentrated to 0.5 mg/ml (Microsep 30, Filtron Technology Corp.).
SDS-PAGE and Immunoblotting-SDS-PAGE and immunoblotting were performed as described (18). Briefly, the cells were lysed in Triton lysis buffer as described below for immunoprecipitations, and culture media were precipitated with ammonium sulfate as described previously (3). The samples were then boiled with SDS-PAGE sample buffer (with or without reduction with 100 mM 2-mercaptoethanol) followed by electrophoresis and transfer onto nitrocellulose membranes. The antitype XIII collagen antisera and mouse anti-human ␤1-integrin antibodies (Serotec) were applied at a dilution of 1:1000 to the filters, and the affinity purified antibodies were used at a concentration of 5 g/ml. The filters were washed thoroughly after 1 h of primary antibody incubation at room temperature and then incubated with horseradish peroxidase-conjugated goat anti-rabbit or goat anti-mouse secondary antibody (Bio-Rad) at a dilution of 1:5000 -1:10,000. The immunosignal was detected after washings using the enhanced chemiluminescence system and films (Amersham Pharmacia Biotech).
Biotinylation of HT-1080 Cells, Immunoprecipitations, and Collagenase Digestions-Surface labeling of subconfluent HT-1080 cells with biotin and immunoprecipitations were performed essentially as described (21). Briefly, cells on 78-cm 2 plates were incubated with 1 mg/ml of the water-soluble biotin derivative Sulfo-NHS-LC-Biotin (Pierce) for 90 min on ice, followed by rinsing with PBS, and inactivation of the remaining biotin reagent with 50 mM glycine. For immunoprecipitations the cells were scraped in Nonidet P-40 lysis buffer (0.1 M Tris-HCl, pH 7.5, containing 1% Nonidet P-40, 2 mM phenylmethylsulfonyl fluoride, 10 g/ml leupeptin, 1 mM EDTA, and 20 g/ml of aprotinin) or Triton lysis buffer (1ϫ PBS, pH 7.5, 1% Triton X-100, 1 mM phenylmethylsulfonyl fluoride, 1 mM N-ethylmaleimide, and 1 mM EDTA), lysed by repeated pipetting on ice, and centrifuged 10,000 g ϫ min to pellet the nuclei. For immunoprecipitation from subcellular membranes, aliquots of the membrane preparation were drawn, pelleted by centrifugation, and suspended to Triton lysis buffer. The lysates were precleared with protein A-Sepharose CL-4B (Amersham Pharmacia Biotech) to reduce unspecific binding. Immunoprecipitations were carried out on 1 mg of precleared lysate protein.
Protein samples were incubated with the type XIII collagen-specific antiserum, antigen-adsorbed control antiserum, or the corresponding preimmune serum at a 1:200 dilution or with 5-10 g of affinity purified antibodies in 600 l at 4°C for 16 h. The resulting immunocomplexes were collected on protein A-Sepharose beads during a 4-h incubation at 4°C followed by sequential washes. Collagenase digestions were performed to washed samples by suspending the washed beads into 400 l of collagenase buffer (50 mM Tris-HCl, pH 7.5, 5 mM CaCl 2 and 0.1% bovine serum albumin (fraction V, Sigma)) and incubating them with 50 units of bacterial collagenase (Worthington, grade CLSPA) for 4 h at 37°C. Parallel controls without added enzyme were always included. The immunoprecipitated proteins were visualized using streptavidin-conjugated horseradish peroxidase and ECL or immunostained with anti-XIII/NC3-1 or a universal anti-pancollagen monoclonal antibody (known to recognize at least collagen types I, II, III, IV, and IX, although the epitope is not known), 2 and the procedure was as above except that the cells were not labeled with biotin, and the filters were incubated with the primary antibody followed by detection with horseradish peroxidase-conjugated anti-rabbit or anti-mouse IgG antibodies and ECL.
Subcellular Fractionation of HT-1080 Cells-Fifteen 78-cm 2 cell culture plates of HT-1080 cells were grown to subconfluency as above. The fractionation was performed on ice using prechilled instruments and solutions. After the medium had been removed, the cells on 13 plates were washed twice with 0.25 M saccharose, 1 mM EGTA, and 5 mM Hepes, pH 7.4 (HES), scraped off into 0.5 ml of HES for each plate, and pooled. Two of the plates were surface-biotinylated as described above, washed with HES, and pooled with the rest of the cells. The cells were pelleted at 20,000 ϫ g ϫ min and resuspended in 3 v/w HES. Homogenization was achieved using a motor-driven Potter-Elvehjem apparatus at 1500 rpm until a microscopically satisfactory homogenate was achieved. The homogenate was diluted to 12.5 ml with HES and centrifuged at 6,000 ϫ g ϫ min to prepare the postnuclear supernatant. Total membranes were pelleted from the postnuclear supernatant at 1.5 ϫ 10 6 ϫ g ϫ min and resuspended in 12.5 ml of HES. The clear supernatant was saved as the soluble protein fraction. Aliquots for immunoblotting and immunoprecipitation were drawn. The membranes were then subjected to self-formed Percoll (Amersham Pharmacia Biotech) density gradient centrifugation as described (22). Briefly, 10.4 ml of the membrane preparation was mixed with 1.4 ml of Percoll stock solution and centrifuged 5.5 ϫ 10 6 g ϫ min in a Beckman SW41 Ti rotor using the slow deceleration option of the centrifuge. The gradient was unloaded by puncturing through the wall of the tube, and the fractions obtained in this way were washed by pelleting and resuspension in HES as above.
Extraction of HT-1080 Cell Membranes-HT-1080 cell membranes were prepared by centrifugation of the postnuclear supernatant at 1.5 ϫ 10 6 g/min as described above for density gradient centrifugation, except that neither of the two plates were biotinylated. The membranes were washed with HES and suspended in 1.5 ml of 0.1 M sodium carbonate, pH 11.5, or 1.5 ml of 1 M NaCl. After a 45 min incubation on ice, the membranes were centrifuged as above and were solubilized in the supernatant volume of HES supplemented with 1% Triton X-100. The NaCl-extracted membranes and the corresponding supernatant were diluted with 1 volume of HES to reduce ionic strength. The supernatant and membrane samples were then boiled with SDS-PAGE sample buffer and analyzed by immunoblotting.
Enzyme Assays in Cell Fractionations-Protein concentrations in all experiments were measured by Protein Assay (Bio-Rad) with bovine serum albumin (Sigma) as a standard. Relative biotin concentrations in the subcellular fractions were estimated from slot blots prepared by blotting serial dilutions of samples with equal protein concentrations on a nitrocellulose membrane, followed by detection of the bound biotin using avidin-peroxidase (Vectastain ABC, Vector Labs) and ECL. Mean densities of the fractions were determined by weighing triplicate 100-l volumes of unwashed samples in preweighed micropipette tips. Cytochrome c oxidase (mitochondrial marker) was assayed essentially as described (23), measuring the oxidation of reduced cytochrome c at 550 minus 540 nm. NADPH-dependent cytochrome c reduction (endoplasmic reticulum marker) was measured in the presence of 25 M rotenone (24). Acid phosphatase (lysosomal marker) was assayed using 8 mM p-nitrophenyl phosphate as the substrate in 90 mM sodium acetate buffer, pH 5.0. The occurrence of p-nitrophenol was measured at 410 nm using a molar extinction coefficient of 9620 in the calculations (25).
Immunofluorescence Staining-HT-1080 cells were seeded on glass coverslips and grown to the desired density as described above. The cells were fixed for 5 min in precooled methanol at Ϫ20°C and incubated in 1% bovine serum albumin/PBS, pH 7.2, for 30 min to reduce nonspecific staining. The anti-XIII/NC1-1 antibody and a monoclonal antibody to the ␤ 3 -integrin subunit (Chemicon) were applied at their appropriate dilutions and incubated for an hour at room temperature, followed by extensive washing with PBS. Rhodamine-conjugated swine anti-rabbit and fluorescein-conjugated goat anti-mouse secondary antibodies were diluted according to the manufacturer's (Dako) instructions and were allowed to bind to the specimens for an hour at room temperature. After extensive washing with PBS, the coverslips were mounted on microscope slides using Glycergel aqueous mounting medium (Dako) and viewed and photographed using a Leitz confocal laser scanning microscope. The specificity of the stainings was confirmed by peptide competition controls and by omitting primary antibodies from the stain-ings. All control stainings resulted in a faint and uniform background staining only.

RESULTS
Amino Acid Sequences of the Mouse ␣1(XIII) Collagen Chain Predict a Type II Transmembrane Protein-The overlapping mouse cDNA clones covered 2925 nt (Fig. 1, GenBank TM accession number U30292), and surprisingly, they extended 594 nt further in the 5Ј direction than the previously isolated human cDNA clones (5). The new 5Ј sequences contained an in-frame ATG codon for methionine (residue 1, in Fig. 1) 240 nt in the 5Ј direction from the previously reported ATG codon (residue 84 in the human sequence in Fig. 1), which had been previously thought to represent the initiation of translation for human type XIII collagen (5). Use of the upstream ATG codon predicted an N-terminal noncollagenous domain 81 residues longer than that described on the basis of the human data.
The polypeptide encoded by the clones consisted of three collagenous domains, COL1-3, and four noncollagenous domains, NC1-4 (Fig. 1). The longest possible coding region of the mouse type XIII collagen chain depicted by the overlapping clones contained 739 residues, but the length of the polypeptide may vary markedly depending on the complex alternative splicing of RNA molecules. The mouse type XIII collagen is characterized by complex alternative splicing of its transcripts (Fig.  1), as previously shown for its human counterpart (5,7,9,10). An extensive study of the splicing pattern of the mouse gene was recently reported (26). If we take into account the shortest and the longest possible splice variant combinations of isolated clones, the calculated molecular mass of mouse type XIII collagen can range between 58.3 and 69.7 kDa. The NC3 domain FIG. 1. Comparison of amino acid sequences between the mouse and human ␣1(XIII) collagen chains. Residues 84 -726 of the human sequences of the ␣1(XIII) collagen chain are derived from previously isolated cDNA clones (3,5), and the N-terminal 83 residues of the new sequences are derived from the nucleotide sequences of the human gene (6) and an EST clone. Noncollagenous domains are boxed; short noncollagenous interruptions are in boldface italic; cysteine residues are in boldface, and the potential transmembrane segments are shaded. The human type XIII collagen-derived peptides used in the immunizations are underlined, and a potential protein kinase C phosphorylation site is shown with an arrow. Identical residues and conservatively changed residues in the two polypeptides are connected with bars and colons, respectively. Amino acid sequences corresponding to the alternatively spliced regions of the RNAs are shown in brackets and numbered according to the exon numbering of the human gene (6). In addition to the exons that were previously identified as alternatively spliced in mouse, namely exons 4A, 4B, 5, 6, 8, 12, 13, 28B, and 33 (26), exon 32 was also found to be alternatively spliced in the cDNAs analyzed here. The sequences corresponding to the segment 28B, which have not been observed in human cDNAs, are indicated with asterisks.
contains one potential site for N-linked glycosylation (residues 451-453, in Fig. 1), which is not conserved in the human chain.
Most surprisingly, the N-terminal sequences of the predicted polypeptide lack a typical signal peptide, whereas a highly hydrophobic region of 23 residues is located at residues 37-59. This clearly fulfills the criteria for a transmembrane region. According to the parameters of Engelman et al. (27), a potential transmembrane segment of 20 residues has a free energy change of 20 kcal/mol or more. The calculated equivalent value for residues 40 -59 in the mouse type XIII collagen chain is 37.8 kcal/mol, suggesting that the sequence represents a potential transmembrane region.
The lack of a signal sequence is typical of those transmembrane proteins that have an intracellular N-terminal portion and an extracellular C-terminal portion, also known as class II transmembrane proteins (28). Thus, the sequence data suggest that type XIII collagen may be located in the plasma membranes of cells, with the N-terminal 36 residues being cytosolic and the primarily collagenous portion of the molecule following the putative transmembrane segment being located outside the cell. In addition, the presumed cytosolic part of the NC1 domain includes a putative protein kinase C phosphorylation site (residues 5-8) (29).
Comparison of the Mouse and Human Type XIII Collagen Sequences-The upstream ATG codon and the next 40 residues were not included in the previously described human cDNA sequences (3,5). All attempts to identify human cDNA clones extending further in the 5Ј direction failed to produce additional sequences, and therefore we presumed in our earlier study that the ATG codon coding for methionine 84 (numbering based on the present study) represents the translation initiation site. Characterization of the human gene revealed the presence of the upstream ATG, but the S1 mapping and primer extension experiments and the finding of a putative TATA box located only 25 nt in the 5Ј direction supported the view of the downstream ATG as the translation initiation codon (6). The finding of the mouse clones extending markedly further in the 5Ј direction prompted us to renewed efforts to generate the corresponding human cDNA sequences from various RNA templates with primers selected from the human genomic sequences. Since no new cDNA clones were obtained, and the sequences could be amplified from genomic subclones only with difficulty, we concluded that the upstream sequence must be a very difficult template in the reverse transcriptase reaction. The discovery of a human cDNA clone covering the upstream ATG during the EST projects study (GenBank TM accession number R25685) made further attempts to find such a clone redundant. The mouse and human polypeptide sequences are compared in Fig. 1, where the 83 extreme N-terminal residues of the human type XIII collagen chain are derived from the isolated human genomic clones (6) and the reported EST sequence. The overall identity between the human and mouse polypeptides is 90% and their similarity 94%.
Preparation of Antipeptide Antisera against Type XIII Collagen and Analysis of Their Specificity-We produced polyclonal antibodies against peptides selected from the noncollagenous NC1 and NC3 domains of human type XIII collagen, thus avoiding possible cross-reactions to other collagens. The region covered by the NC1 peptide shows very low homology to mouse type XIII collagen, whereas the sequence of the other peptide that covers the NC3 domain is almost completely conserved between these species (Fig. 1). Both peptide sequences are present in all type XIII collagen isoforms, and both are also unique in the protein sequence data bases. Recombinant human type XIII collagen produced in insect cells was used in the preliminary analysis of the antibodies. Insect cells infected with baculoviruses expressing both human type XIII collagen and human prolyl 4-hydroxylase and negative control cells expressing human prolyl 4-hydroxylase only were lysed and analyzed by immunoblotting and immunoprecipitation. Recombinant type XIII collagen in the lysates was stained by the NC3 antibodies (anti-XIII/NC3-1), but no staining was seen in the lysates of the cells that were producing only prolyl 4-hydroxylase ( Fig. 2A). The specificity of the NC1 antibodies (anti-XIII/ NC1-1) was studied directly by immunoblot analysis of HT-1080 membrane preparations (Fig. 4A, see below), which revealed the same bacterial collagenase-sensitive polypeptides as immunoblotting with the NC3 antibodies. Further demonstration of the specificity of the NC1 antibodies to type XIII collagen was obtained by immunoprecipitating recombinant human type XIII collagen from the insect cell lysates (Fig. 2B). The specificity of the antibodies was further confirmed by using affinity purified antibodies and by preimmmune serum and peptide competition assay controls in the experiments.
Immunoblotting of Type XIII Collagen from HT-1080 Cell Lysates-Since it was known from our previous studies on type XIII collagen that human HT-1080 cells synthesize the highest amounts of type XIII collagen mRNAs of all the cell types studied, as well as the widest selection of different alternatively spliced variants known so far (3, 5, 7), we decided to use these as a model to study the type XIII collagen protein. Cell lysates were prepared and analyzed using the anti-XIII/NC3-1 antibodies. Immunoblots of the lysates revealed proteins in the 85-95-kDa size range in the blot of the reduced samples (Fig.  3A), whereas no specific staining was seen in ammonium sulfate-precipitated culture medium of HT-1080 cells (not shown). The migration of the detected proteins in SDS-PAGE was in the same range as that of the recombinant type XIII collagen produced in insect cells (compare Figs. 2 and 3).
HT-1080 samples analyzed under nonreducing conditions exhibited an additional population of reduction-sensitive aggregates with molecular masses of over 180 kDa (Fig. 3B, lane  1). These bands were not very strong, probably because of their broad range of migration due to the multiple alternatively spliced variants that are present. These proteins could represent disulfide-bonded multimeric forms of type XIII collagen, although some polypeptides were still seen at 85-95 kDa. This was a constant finding in a number of samples analyzed, and the binding of the antibodies to all detected proteins in these samples was abolished by antigen competition (Fig. 3B, lane 2). Furthermore, both molecular populations were also detected in nonreduced samples of the HT-1080 membrane fractions by affinity purified antibodies to both NC1 and NC3 domains of  A and B) or full-length human type XIII collagen and prolyl 4-hydroxylase (lanes 2 in A and B). A, immunoblotting of lysates with anti-XIII/NC3-1. B, immunoprecipitation of the same lysates with anti-XIII/NC1-1 followed by immunoblotting with anti-XIII/NC3-1. Sizes of the molecular weight markers are shown on the left in M r ϫ 10 3 . Migration of the heavy chain of the precipitating antibody is indicated with an arrowhead on the right of field B. type XIII collagen (Fig. 4A, lanes 5 and 6). These results collectively demonstrate that both polypeptide populations indeed contain type XIII collagen and that a portion of the "steady state" type XIII collagen in HT-1080 cells is not disulfidebonded. This could be explained by defective or slow (relative to synthesis rate) processing of type XIII collagen by these cells.
The NC1 antibodies were produced against a peptide chosen from the region encoded by the new extension to the cDNA sequences and could thus be used to demonstrate the presence of the corresponding region in the type XIII collagen protein.
Cross-detection demonstrated that the epitope of the NC1 domain antibodies and the NC3 domain reside in the same molecule, since immunoprecipitation of the HT-1080 cell lysate and the total membrane fraction with the NC1 antibodies followed by immunoblot detection using the NC3 antibodies resulted in the same 85-95-kDa polypeptides (Figs. 3C and 4A, lanes 7-9).
The collagenous nature of the presumed type XIII collagen polypeptides was demonstrated by bacterial collagenase digestions of material immunoprecipitated with our antibodies from HT-1080 whole cell lysates (Fig. 3C) and membrane fractions (Fig. 4A, lanes 7-9). Further evidence was provided by immunoblotting polypeptides immunoprecipitated from HT-1080 whole cell lysate using the NC3 antibodies with a universal anti-pancollagen monoclonal antibody that recognizes se-quences in the triple helical domains of several collagenous proteins (see "Experimental Procedures"). This approach also resulted in the same 85-95-kDa polypeptides (Fig. 3D).

Subcellular Fractionation, Membrane Extractions, and Immunoprecipitations Demonstrate Transmembrane Anchorage of Type XIII Collagen on the Plasma Membranes of HT-1080
Cells-Our preliminary experiments indicated that a detergent such as Triton X-100 or Nonidet P-40 is needed for the solubilization of the majority of type XIII collagen from cells, whereas the corresponding culture media precipitated with ammonium sulfate showed no staining with the type XIII collagen antibodies. This was very well in concert with the idea of a transmembrane topology and a plasma membrane location, as suggested by the cDNA-derived primary structure. Several approaches were used to verify experimentally the hypothesized location of the type XIII collagen on the plasma membrane.
First, a total membrane fraction was prepared from HT-1080 postnuclear supernatants and studied by immunoblotting. The analysis resulted in the detection of the same 85-95-kDa and over 180-kDa polypeptides by both the NC1 and NC3 antibodies as described above for the whole cell lysates (Fig. 4A, lanes  1-6). The sensitivity of the 85-95-kDa proteins to bacterial collagenase digestion was also demonstrated using immunoprecipitated samples (Fig. 4A, lanes 7-9).
Second, the transmembrane anchorage of the polypeptides in the membranes was demonstrated by their resistance to extraction into soluble phase by 1 M NaCl and 0.1 M sodium carbonate, pH 11.5. The former reagent solubilizes loosely bound peripheral membrane proteins by interfering with the osmotic environment and competing for electrostatic interactions, whereas the latter is known to convert membrane vesicles into open sheets, strip the membranes from virtually all peripheral membrane proteins, and to release the luminal contents of the organelles (30). Type XIII collagen remained bound to the membranes after both extractions, which directly demonstrates its transmembrane nature, as did the ␤ 1 integrin subunit that was used as a transmembrane protein control (Fig. 4B).
The above experiments convinced us that type XIII collagen indeed is a transmembrane protein. The following experiments were designed to gain evidence on its location in the plasma membrane. Cell-surface molecules of HT-1080 cells were labeled by biotinylation with a water-soluble biotin derivative. Immunoprecipitation of the cell lysates with the anti-XIII/ NC3-1 antiserum (Fig. 5A) revealed that type XIII collagen is present on the cell surface, as the signal detection using horseradish peroxidase-conjugated streptavidin relied on the cellsurface label.
Furthermore, density gradient centrifugation of a total membrane fraction, prepared as above, was applied to study the distribution of type XIII collagen within the subcellular membranes. The cell surfaces were trace-labeled with biotin as above, and the initial homogenization and low speed centrifugation steps were adjusted to produce a membrane fraction without nuclei and with maximal surface label content. After centrifugation, visual inspection of the gradient revealed four layers, which were collected as fractions 1 to 4 (from top to bottom) and analyzed for the presence of protein, type XIII collagen and biotin (plasma membrane marker), and the activities of NADPH-dependent cytochrome c reductase (endoplasmic reticulum marker), cytochrome c oxidase (mitochondrial marker), and acid phosphatase (lysosomal marker). The analysis showed reproducibly that type XIII collagen was exclusively present in fraction 2 (Fig. 5B), as was the case for the plasma membrane label (Fig. 5C) and the plasma membranebound ␤ 1 integrin subunit (Fig. 5B). The mean density of this fraction (1.048 g/ml, Fig. 5D) is in good agreement with studies on fibroblast plasma membranes in Percoll (31). Other major membranous organelles, as judged by the distribution of the specific activities of their respective marker enzymes, were either almost exclusively present in fractions (lysosomes and mitochondria) containing no type XIII collagen or exhibited a wider distribution among the four fractions (endoplasmic reticulum) than type XIII collagen did (Fig. 5D). These findings indicate that the exclusive presence of type XIII collagen in fraction 2 was most probably due to the contribution of the plasma membrane to this fraction.
Immunofluorescence Staining-Immunofluorescence stainings of HT-1080 cells using the anti-XIII/NC1-1 antibody revealed staining in the outer margins of moving or spreading cells in freshly plated cultures. This staining was highly reproducible, and double immunofluorescence stainings using a monoclonal antibody to ␤3-integrin subunit as a marker for a plasma membrane protein revealed an almost impeccable colocalization of the two proteins in these cells (Fig. 6). In addition there was a specific intracellular staining, which is probably due to the previously reported high synthesis rate of type XIII collagen by these cells. DISCUSSION A surprising finding in the analysis of the cloned mouse cDNA sequence predicted that type XIII collagen is a transmembrane protein. The mouse type XIII collagen cDNA clones described here extend further in the 5Ј direction than the human clones and encode a longer N-terminal noncollagenous domain than has been previously reported for the human sequence (5). The translation of the longer domain initiates at an upstream ATG which is in-frame with that previously thought to represent the initiation codon of translation and results in an N-terminal noncollagenous domain that is 81 residues longer and includes a putative transmembrane domain. The corresponding human cDNA sequence was not previously obtained, and it has become evident that this human sequence is difficult to clone. Nevertheless, a human cDNA clone extending as far in the 5Ј direction as the mouse clones has been identified during the EST projects study (GenBank TM accession number R25685). In addition, a mouse EST clone with a 5Ј end upstream of the new ATG codon has been identified (Genbank TM accession number AA14637). It can be predicted from the available human EST and genomic sequences (6) that the upstream ATG, and thus the longer NC1 domain and the highly hydrophobic transmembrane domain, also occur in human type XIII collagen.
The detection of human type XIII collagen using antibodies raised against a peptide sequence selected from the putative N-terminal extension directly demonstrated that this region was present in the type XIII collagen protein. Furthermore, the immunoprecipitation of type XIII collagen labeled in situ on the cell surface and subcellular fractionation and extraction experiments demonstrate biochemically the functional importance of the hydrophobic domain contained in the deduced protein sequence. The immunostaining of HT-1080 cells likewise corroborate the plasma membrane location of this protein. Considering together the new sequence data and the biochemical as well as the immunostaining studies on the location of the type XIII collagen protein, it appears that type XIII collagen is anchored to the plasma membrane via a hydrophobic domain near its N terminus. This surprising finding suggests a cell surface-associated function for type XIII collagen in tissues.
Nevertheless, type XIII collagen is not the only transmembrane protein that has collagenous domains. The first integral membrane proteins found to contain collagenous sequences were the types I and II macrophage scavenger receptors (32,33), which participate in a variety of macrophage-associated functions, as suggested by their broad polyanion-binding ability, including host defense and inflammation (34). Furthermore, the recently identified bacteria-binding plasma membrane protein MARCO is structurally related to the scavenger receptors and contains a collagenous domain (35). Although these receptors are not true collagen molecules, they have nevertheless been included in the superfamily of proteins with collagenous sequences (1).
The 180-kDa bullous pemphigoid antigen BPAG2 is a hemidesmosomal component expressed in stratified squamous epithelia of the skin, oral cavity, and uterine cervix. This protein was originally recognized as an autoantigen in bullous pemphigoid and herpes gestationalis (36 -38). Its primary structure predicted a highly interrupted collagenous domain and a transmembrane segment, and the molecule was subsequently designated as type XVII collagen (39,40). Types XIII and XVII collagens are not homologous in sequence or in the general appearance of these molecules, but since our results indicate that type XIII collagen is also a membrane-associated collagenous protein that is expressed in mesenchymal tissues, we nevertheless suggest that they should be grouped together on the basis of their plasma membrane location to comprise a new subgroup of transmembrane collagens (1,2).
The scavenger receptors, MARCO and type XVII collagen, all reside on the plasma membrane in an orientation where their N-terminal regions are intracellular, and the collagenous domains are in the extracellular space. Since the hydroxylation of proline residues and disulfide bond formation occurring in the lumen of the endoplasmic reticulum necessitate such a topology, it would be fairly safe to assume that the collagenous portion of type XIII collagen, located toward the C-terminal from the putative plasma membrane, domain is extracellular. This "type II" orientation occurs in only 5% of plasma membrane proteins (28), which suggests that the orientation of collagenous proteins in the membrane may be more than coincidental, particularly with respect to formation of the collage-FIG. 6. Co-localization of type XIII collagen and ␤ 3 integrin subunit on the HT-1080 cell membrane by double immunofluorescence staining. Double immunofluorescence staining of a single spreading HT-1080 cell with antibodies to the NC1 domain of type XIII collagen (red; upper left panel) and ␤3-integrin subunit (green; upper right pane), as seen by confocal laser scanning microscopy. Lower panel shows superimposition of the two stainings, where co-localization is seen as yellow. Co-localization of the signals is seen in the cell periphery. Both antibodies also give a specific intracellular signal, reflecting current synthesis or intracellular stores of adhesion molecules in the early stages of attachment. The type XIII collagen staining was abolished by peptide competition (not shown). nous triple helix. In the case of the fibril-forming collagens the three appropriate procollagen chains associate via their Cpropeptides after the polypeptides have been released into the lumen of the endoplasmic reticulum. Triple helix formation initiates at the C-terminal ends of the collagen domains of the associated pro-␣ chains and proceeds in a zipper-like fashion toward the N terminus (41). Since the N termini of the collagenous transmembrane protein chains are inserted into the rough endoplasmic reticulum membrane before their C termini can associate, formation of the triple helix proceeding from the C terminus toward the N terminus would create torsional tension in their structures. This tension would have to be relieved by rotation of the transmembrane domains of the three polypeptides around each other in the membrane. An alternative hypothesis would be that the macrophage scavenger receptors, MARCO and collagen types XIII and XVII, differ strikingly from the known mode of triple helix formation in that the association of the three chains occurs near the most N-terminal triple helical domain, while the polypeptides are being inserted into the rough endoplasmic reticulum membrane and folding into the triple-helical conformation proceeds in the opposite direction, i.e. from the N terminus toward the C terminus. At the present time there are no experimental data to support either hypothesis.
We do not know the function of type XIII collagen, but our observations on its structural characteristics allow for some new speculations. It is possible that type XIII collagen could play a role in the adhesion of many types of cells to their surrounding extracellular matrix analogous to type XVII collagen, or it could function as a receptor for soluble ligands such as the scavenger receptor isoforms or MARCO.
The predicted large ectodomain is the most conspicuous structural feature of type XIII collagen and is thus most likely involved in its physiological function. This could include interaction with soluble ligands or components of the extracellular matrix, or lateral interaction with other components of the cell surface. Sequence analysis has not revealed any known functional motifs in the ectodomain. However, experiments with scavenger receptors have provided evidence that a collagenous structure can function as a specific binding site for a ligand (42). In this light, the 115 most C-terminal amino acid residues of type XIII collagen that are completely conserved between the man and mouse species form a good candidate for a site of interaction. One feature that could also be of importance is the presence of 10 charged Gly-X-Y triplets in the extreme C terminus of the most protruding collagenous domain COL3, the majority of the charged residues being basic. These could be employed in the binding of mostly negatively charged cellsurface proteins or glycosaminoglycans. The complex alternative splicing of transcripts affecting the structures of the COL1, NC2, and COL3 domains is likely to alter the functional properties of the type XIII collagen molecules. Our results demonstrate that this alternative splicing is conserved between the mouse and human species and affects the structures of the same domains in their ␣1(XIII) collagen chains. It is possible that these regions fulfill certain critical functions. On the other hand, the collagenous domains could simply function as spacer regions that adjust the distances between functional elements. The composition of such collagenous domains would not be critical as long as their lengths remained right. That would explain the finding that only some of the alternatively spliced exons coding for parts of the collagenous domains COL1 and COL3 in the mouse cDNA are subject to alternative splicing in the human chain and vice versa.
The cytosolic domain of type XIII collagen is short and is thus unlikely to have any enzymatic activity, but there is a threo-nine residue at position 6 that could be subject to phosphorylation. Future studies will resolve whether the intracellular domain of type XIII collagen is involved in adhesion or communication between type XIII collagen and the intracellular compartment, or perhaps functions as a signal for correct topological positioning of type XIII collagen in the cell membrane.