Type XIII Collagen and Some Other Transmembrane Collagens Contain Two Separate Coiled-coil Motifs, Which May Function as Independent Oligomerization Domains*

Type XIII collagen is a homotrimeric transmembrane collagen composed of a short intracellular domain, a single membrane-spanning region, and an extracellular ectodomain with three collagenous domains (COL1–3) separated by short non-collagenous domains (NC1–4). Several collagenous transmembrane proteins have been found to harbor a conserved sequence next to their membrane-spanning regions, and in the case of type XIII collagen this sequence has been demonstrated to be important for chain association. We show here that this 21-residue sequence is necessary but not sufficient for NC1 association. Furthermore, the NC1 association region was predicted to form an α-helical coiled-coil structure, which may already begin at the membrane-spanning region, as is also predicted for the related collagen types XXIII and XXV. Interestingly, a second coiled-coil structure is predicted to be located in the NC3 domain of type XIII collagen and in the corresponding domains of types XXIII and XXV. It is found experimentally that the absence of the NC1 coiled-coil domain leads to a lack of disulfide-bonded trimers and misfolding of the membrane-proximal collagenous domain COL1, whereas the COL2 and COL3 domains are correctly folded. We suggest that the NC1 coiled-coil domain is important for association of the N-terminal part of the type XIII collagen α chains, whereas the NC3 coiled-coil domain is implicated in the association of the C-terminal part of the molecule. All in all, we propose that two widely separated coiled-coil domains of type XIII and related collagens function as independent oligomerization domains participating in the folding of distinct areas of the molecule.

Type XIII collagen is a type II transmembrane protein that is expressed in many tissues throughout development and adult life (1). It is located in focal adhesions of cultured fibroblasts and other cells and in adhesive structures of tissues such as the myotendinous junctions in muscle, intercalated discs in heart, and the cell basement membrane interphases (1,2). The type XIII collagen ectodomain can bind to fibronectin, heparin, the basement membrane components nidogen-2 and perlecan, and the ␣ 1 -subunit of integrin (3)(4)(5). Due to its location at the tissue and cell level and its binding properties, it has been postulated that type XIII collagen is involved in cellular adhesion and migration.
Type XIII collagen ␣ chains produced in a recombinant insect cell culture system have been shown to form homotrimers (6). The primary structure is composed of three collagenous domains (COL1 1 to COL3), which are flanked and interrupted by non-collagenous domains (NC1 to NC4). The short cytosolic domain and the transmembrane domain encompass about half of the NC1 domain, whereas the rest of the molecule forms the ectodomain, which is a rod of about 150 nm with two flexible hinges coinciding with the NC2 and NC3 domains (5). The primary structures of COL1, NC2, COL3, and NC4 can vary, on account of complex alternative splicing (7)(8)(9)(10). It has been shown recently that extracellular sequences adjacent to the transmembrane domain are important for the association of type XIII collagen into trimeric molecules (4). It appears that triple helix formation proceeds in the opposite orientation than for the fibrillar collagens, i.e. from the N terminus to the C terminus. Because type XIII collagen does not contain a signal sequence, its translocation to the endoplasmic reticulum has been thought to be mediated by the transmembrane domain sequences, residues 37-59 in the mouse and 39 -61 in man (11), as is known to occur with other type II transmembrane proteins (12).
By using homologous gene targeting, we have previously generated a mouse line, Col13a1 N/N , expressing modified type XIII collagen that lacks the extreme 96 N-terminal residues, including the cytosolic, transmembrane, and association domains, which are replaced by unique sequences not found in any other protein (13). Analysis of tissues and cultured cells derived from homozygous Col13a1 N/N mice has shown that the altered type XIII collagen molecules are transported to roughly the correct location despite the lack of a transmembrane domain. Expression of the N-terminally altered type XIII collagen molecules results in changes in muscle integrity in the genetargeted mice, including abnormalities in the sarcolemmabasement membrane interphase. Immunoelectron microscopy has indicated that the mutant molecules are situated in the adjacent extracellular space, whereas wild-type type XIII collagen molecules are adherent to the plasma membrane. Moreover, cells extracted from the mutant mice showed decreased adhesiveness to the basement membrane component type IV collagen.
Studies with the Col13a1 N/N mice have revealed a role for the N terminus of type XIII collagen in anchoring muscle cells to the basement membrane. Nevertheless, the mutant molecules were secreted, and they may retain intact some aspects of their functional properties. This prompted us to study the molecular properties of the N-terminally altered type XIII collagen. The amount of type XIII collagen in tissues is very low, and thus insect cell expression was used to obtain sufficient protein for these studies. A series of N-terminal variants was tested for their ability to form stable disulfide-bonded type XIII collagen molecules. The data led us to search for other regions in addition to the NC1 domain that may be important for chain association and stability in type XIII collagen and other collagenous transmembrane and non-transmembrane proteins.

Construction of Expression Vectors and Generation of Recombinant
Baculoviruses-Human type XIII collagen variant del1-38 and del1-83 viruses have been described previously (6). E-26 (14), a cDNA clone covering the coding sequence for type XIII collagen except for the beginning of the translation, was used as a template for generating human variant del1-61. Complementary oligonucleotides (nucleotides 660 -692 in human type XIII collagen cDNA bearing a translation start codon in position 678 -680, under GenBank TM data base accession number AJ293624 (4)) were used as primers in a mutagenesis reaction performed using a site-directed mutagenesis kit according to the manufacturer's protocol (Stratagene). The cDNA was transported to the insect cell expression vector pVL1392 (Invitrogen) by EcoRI digestion and ligation.
Sequences coding for the altered N termini of mouse type XIII collagen were obtained by reverse transcription followed by amplification with PCR. Total RNA was isolated from the skeletal muscle of a mouse expressing N-terminally altered type XIII collagen and transcribed into single-stranded DNA using a type XIII collagen-specific reverse oligonucleotide primer complementary to nucleotides 1022-1039 (under GenBank TM data base accession number NM_007731 (11)) as described previously (13). The long altered N-terminal sequences were amplified using a sense oligonucleotide primer corresponding to nucleotides in the loxP sequence (5Ј-CGGGGTACCGAATTCTGTATGCTATACGAAGTT-ATTAG-3Ј, see Ref. 13 for further explanation) and the short sequences using a primer corresponding to nucleotides 5818 -5840 in the first intron (under GenBank TM data base accession number AF063666 (17)). KpnI and EcoRI restriction enzyme recognition sequences were included in the 5Ј-end of both sense primers, and the antisense primer used in both amplifications was complementary to nucleotides 984 -1003 in mouse type XIII collagen cDNA (under GenBank TM data base accession number NM_007731 (11)). The mouse type XIII collagen cDNA moXIII (689), lacking exons 15, 31, and 36 (4), was digested with KpnI thereby cutting off the wild-type 5Ј-sequences from the rest of the clone. PCR products were also digested with KpnI and linked to replace the wild-type 5Ј-sequences. Recombinant expression constructs XIII N-long and XIII N-short were generated by releasing the inserts from the Bluescript vectors by EcoRI digestion and linking them to the pVL1393 expression vector.
Recombinant baculoviruses were generated by transfecting the construct DNAs together with modified Autographa californica nuclear polyhedrosis virus DNA into Spodoptera frugiperda Sf9 insect cells using the BaculoGold transfection kit (Pharmingen). The recombinant viruses were plaque-purified and amplified as described previously (18).

Analysis of Recombinant Proteins Produced in Insect Cells by SDS-PAGE and Immunoblotting-High
Five insect cells were cultured as monolayers in TNM-FH (Sigma) insect cell medium supplemented with 10% fetal bovine serum (Bioclear) and, when infected, in serum-free Express Five medium (Invitrogen) at 27°C. Viruses coding for the various type XIII collagen variants were used at m.o.i. 5 together with the virus coding for both subunits of human prolyl 4-hydroxylase (4PH␣␤) (19) at m.o.i. 1. For infection, insect cells were seeded on plates at a density of 1 ϫ 10 5 cells/cm 2 . As a control, cells were infected with the prolyl 4-hydroxylase virus alone. Fresh ascorbate phosphate (80 g/ml, Wako Pure Chemical Industries Ltd.) was added to the infected cells daily.
The cells were detached 48 h post-infection by gently pipeting with the medium. The cells were collected by centrifuging at 340 ϫ g for 10 min at room temperature, and the medium was supplemented with 2 mM EDTA or Complete Protease Inhibitor Mixture (Roche Applied Science) to prevent proteolytic cleavage. The cells were washed with PBS and collected as described previously. They were then homogenized in 67 mM Tris-HCl, pH 7.5, 267 mM NaCl, 0.2% Triton X-100 supplemented with Complete Protease Inhibitor Mixture and incubated for 30 min on ice. The cell lysate was centrifuged at 8000 ϫ g for 10 min at 4°C; the supernatant was recovered, and the precipitate was dissolved in 1% SDS. Samples of the different fractions were analyzed by denaturing SDS-PAGE under reducing or non-reducing conditions. The samples were subjected to Western blot analysis with anti-human type XIII polyclonal antibody XIII/NC3-1 (11), detected with enhanced chemiluminescence. The percentage of secreted protein was estimated by scanning densitometry using the Quantity One Quantitation software (Bio-Rad).
Recombinant Protein Purification, N-terminal Sequencing, and Pepsin Digestions-High Five cells in suspension were cultured in Express Five medium in a Certomat BS 4 shaker (B. Braun Biotech) with 130 rpm horizontal agitation at 27°C. The High Five cells were co-infected at a density of 1 ϫ 10 6 /ml with the virus encoding mouse XIII N-short at m.o.i. 5 and with the virus 4PH␣␤ at m.o.i. 1. Once they were infected, ascorbate phosphate (80 g/ml) was added to the culture medium daily. 500 ml of cell culture medium was separated from the cells 48 h post-infection by centrifuging at 340 ϫ g for 10 min at room temperature, and 2 mM EDTA was added, after which the medium was further centrifuged at 40,000 ϫ g for 45 min at 4°C to remove the debris and viruses. The medium was then applied to a Resource S 6-ml column (Amersham Biosciences) and eluted using a gradient program on Ä KTA explorer 10 (Amersham Biosciences). The fractions were analyzed by Western blotting using the antibody XIII/NC3-1, and those containing XIII N-short protein were concentrated to 1 ml and purified on a Sephacryl S-500 column (1.6 ϫ 100 cm, Amersham Biosciences) using 20 mM HEPES, pH 7.0, 0.15 M NaCl. The relevant fractions were then concentrated to 0.5 g/l (total protein), and about 10 g of total protein was separated on 10% SDS-PAGE gel under reducing conditions and electroblotted onto a ProBlott TM membrane (Applied Biosystems), which was stained with Serva Blue R (Serva). The authenticity of the XIII N-short protein was confirmed by N-terminal protein sequencing using a 492 Procise TM protein sequencer (Applied Biosystems). 40 g of partly purified XIII N-short protein was digested with 0.8 g of pepsin (50:1 w/w) for 2 min at room temperature. The digestion products were separated by SDS-PAGE and electroblotted onto a ProBlott TM membrane for N-terminal protein sequencing analysis.
For the sequencing of XIII N-long proteins in the cell lysate, infected insect cells from a 100-ml suspension culture were harvested after 48 h of infection by centrifuging at 340 ϫ g for 10 min at room temperature. They were then washed with 10 ml of PBS twice, homogenized with 10 ml of PBS, and incubated for 30 min on ice. The debris and insoluble molecules were removed by centrifugation at 40,000 ϫ g for 45 min at 4°C. The supernatant of the cell lysate was separated out in steps using a HiTrap Q 5-ml column (Amersham Biosciences), by increasing the concentration of NaCl in the PBS buffer. The fractions containing XIII N-long were concentrated to 0.5 ml and subsequently precipitated with 75% cold ethanol for 30 min. The ethanol precipitates were dissolved in SDS-PAGE sample buffer, applied to SDS-PAGE, and electroblotted onto a ProBlott TM membrane for N-terminal protein sequencing analysis.
For the sequencing of XIII N-long proteins in medium, the medium sample described above was centrifuged at 40,000 ϫ g for 45 min at 4°C to remove the virus and small debris and then loaded onto a HiTrap Q 1-ml column (Amersham Biosciences) pre-equilibrated with 20 mM MES, 0.3 M NaCl, 2 mM EDTA, pH 6.0, at 0.5 ml/min, followed by step-by-step elution with the same buffer containing 0.3, 0.5, 0.8, and 1 M NaCl, respectively. All the fractions were analyzed by Western blotting using the antibody XIII/NC3-1. The fractions containing XIII N-long protein were combined and dialyzed against 20 mM Tris, 0.15 M NaCl, 2 mM EDTA, pH 7.4, and loaded onto a Resource Q 1-ml column (Amersham Biosciences) pre-equilibrated with the same buffer at 0.5 ml/min at 4°C. The elution was performed by means of a programmed gradient with Ä KTA Explorer 10 starting from 0.15 to 1.0 M NaCl in the 20 mM Tris, 2 mM EDTA, pH 7.4, buffer for a 20-column volume. The elution fractions containing the XIII N-long protein were concentrated to 1 ml and further separated by Superdex 200 (Amersham Biosciences) in Ä KTA Explorer 10 with 20 mM HEPES, 0.15 M NaCl, pH 7.0. The fractions containing XIII N-long protein were concentrated and applied to the SDS-PAGE followed by electroblotting onto a ProBlott TM membrane for N-terminal protein sequencing analysis.
Amino Acid Sequence Analysis and Secondary Structure Prediction-Sequence alignment of collagen types XIII, XXIII, XXV, and XXVI was performed using the ClustalW method (20) and compiled into a figure using BOXSHADE (ch.EMBnet.org). Coiled-coil predictions for collagen types XIII, XVII, XXIII, XXV, XXVI, MARCO, and EDA were made using the COILS program (version 2.1 (21)). The GenBank TM data base accession numbers for collagen types XIII, XVII, XXV, XXVI, MARCO, and EDA are AJ293624, NM_000494, AF293340, AB085837, NM_006770, and NM_001399, respectively. The positions of the heptad repeats in the NC1 and NC3 domain of type XIII collagen were predicted according to the COILS program.

Secretion of Type XIII Collagen Molecules Expressed in Insect
Cells-The type XIII collagen molecules synthesized by the Col13a1 N/N mice lacked the cytosolic, transmembrane, and association domains (the 96 extreme N-terminal residues encoded by exon 1) but retained the large collagenous ectodomain. Surprisingly, these molecules were located in association with adherence structures apparently secreted into the pericellular matrix, and they were correctly located in focal adhesions in cultured cells derived from these mice (13). To test the effect of the altered N terminus on association and folding of the mutant ␣1(XIII) chains, we prepared recombinant DNA expression constructs encoding identically altered mouse ␣1(XIII) chains ( Fig.  1), the deleted N-terminal sequences in the Col13a1 N/N mice being replaced by either 65 or 11 residues of sequences unique to the mutant ␣1(XIII) chains depending on which of the two potential new translation initiation sites was used. Consequently, two constructs were prepared, namely XIII N-long and XIII N-short , corresponding to full-length ␣1(XIII) chains from residue 97 onwards and preceded by the longer or shorter mutant N termini (Fig. 1A). The del1-38 variant lacks the cytosolic domain but corresponds to full-length human ␣1(XIII) chains in terms of chain association and folding, and this was used as a control on account of its superior expression levels relative to full-length human and mouse ␣ chains ( Fig. 1B (6)). In addition, constructs del1-61 and del1-83 were prepared, encoding human ␣1(XIII) chains lacking either the first 61 or 83 residues, respectively (Fig. 1B). In each case insect cells were infected with viruses encoding ␣1(XIII) and prolyl 4-hydroxylase (4PH␣␤), the latter being necessary in order to obtain hydroxylated recombinant collagen chains when using insect cells as hosts (22).
Proteins were extracted from infected insect cells by homogenizing the cells in a buffer containing Triton X-100, and the remaining precipitates were solubilized in 1% SDS. The volumes of the cell and medium fractions were adjusted to correspond to the same cell number, and the samples were fractionated on denaturing SDS-PAGE gels under reducing or non-reducing conditions and analyzed by Western blotting. Because the SDS fractions contained only minimal amounts of protein, these were excluded from the final pictures, which contain only the Triton X-100 cellular fractions and medium samples for each variant (Fig. 2). It has been shown previously that about half of the del1-38 ␣1(XIII) chains are secreted into the medium through proteolytic cleavage by one or more furintype proteases when cultured in medium supplemented with serum (4). A furin consensus sequence can be found at amino acid residues 105-108 in human type XIII collagen (under GenBank TM data base accession number CAC00688 (4)) and at residues 103-106 in the mouse protein (under GenBank TM data base accession number NP_031757 (11)). Here we used serum-free medium, and only about 10% of the del1-38 protein was secreted (Fig. 2B, lanes 1 and 2), possibly because of a decrease in proprotease activators in the medium. Similar or slightly higher portions of the protein, namely about 50% for XIII N-short , 15% for XIII N-long , 40% for del1-61, and 15% for del1-83, were found to be secreted (Fig. 2B, lanes 3-10).
Mutant XIII N-short and XIII N-long ␣1(XIII) Chains Are Processed at a Furin Site-Western blotting of XIII N-long cell fractions revealed two bands with molecular masses of 90 and 80 kDa ( Fig. 2A, lane 5), whereas the predicted molecular mass for XIII N-long is 64 kDa. This difference is known to reflect the high imino acid content of collagens in addition to post-translational modification of the polypeptides (23). Purification and sequencing of the cell fraction proteins indicated that the 90-kDa band has the N-terminal MLYEVIRSLE predicted for intact XIII N-long (13), whereas the 80-kDa band corresponded to polypeptides with the N-terminal 147 GQPGEKGAPG located at amino acid residue 147 in the mouse protein, and thus the latter represented a degradation product of the full-length XIII N-long ␣ chain. Sequence analysis of the two bands observed in the medium indicated that the upper band with the Nterminal MLYEVIRSLE represented full-length XIII N-long and the lower band had an N terminus of 107 EAPKMSPGCN (residue 107 of the mouse ␣1(XIII) chains). Thus the lower band lacked the 75 extreme N-terminal residues (65 residues of mutant sequences and residues 97-106 of the mouse ␣1(XIII) chains) preceding the predicted furin cleavage site, indicating that the XIII N-long chains were cleaved at the predicted furin site in the manner described previously for the human del1-38 ␣ chains (4). The observed molecular mass for secreted XIII N-short was 80 kDa (Fig. 2A, lane 4) and the calculated mass for full-length was 58 kDa. Only one major band was observed in the medium of cells synthesizing the XIII N-short ␣ chains, and identification of its N-terminal sequence, 107 EAPKMSPGCN, indicated that it lacked the 21 extreme N-terminal residues (11 residues of mutant sequences and residues 97-106 of the mouse ␣1(XIII) chains) and was also cleaved at the furin site. Also a faint band was seen in this medium sample likely representing a degradation product. Thus the N-terminally altered mouse ␣1(XIII) chains were secreted and processed in the same manner as shown previously for human ␣1(XIII) chains (4), and some part of the full-length XIII N-long ␣ chains were secreted without proteolytic processing.
Formation of Disulfide-bonded Trimers of Type XIII Collagen Variants-Full-length human ␣1(XIII) chains have four pairs of cysteine residues located in the NC1, COL1, NC2, and NC4 domains, as depicted in Fig. 1B. We have shown previously that interchain disulfide bonds are found in the NC1 domain of human type XIII collagen and possibly in the COL1 and NC2 domains, whereas the cysteine residues occurring in the NC4 domain form intrachain bonds (6). Full-length mouse ␣1(XIII) chains contain nine cysteine residues, differing from the human equivalents in possessing an extra pair of cysteines in the transmembrane domain and lacking a cysteine residue in the NC2 region (Fig. 1A). Deletion variant del1-38 contains all eight cysteine residues, and it has been shown that this variant forms disulfide-bonded homotrimers in a similar manner to full-length human ␣ chains (4). Thus we compared the ability of the other recombinant ␣1(XIII) chains to form disulfide-bonded trimers with del1-38. Western blot analysis of non-reduced samples of the del1-38 variant revealed disulfide-bonded trimers both in the cell supernatant and in the medium, and it is notable that monomers are not present in the medium (Fig. 2B,  lanes 1 and 2). The slightly faster mobility of the trimer protein in the medium is due to the furin cleavage, as shown previously (4). Full-length mouse ␣1(XIII) chains form disulfide-bonded homotrimers in the same manner (data not shown). Due to the sequence alteration in the N terminus, XIII N-short and XIII N-long ␣1(XIII) chains lack the NC1 association domain and four of the extreme N-terminal cysteine residues, whereas instead XIII N-long contains two new cysteine residues in its unique sequences (Fig. 1A). The XIII N-long samples also contain trace amounts of trimers in the cell extracts, but the ratio of trimers to monomers is extremely low, so that only monomers could be visualized in the medium (Fig. 2B, lanes 5 and 6). In the case of the XIII N-short ␣1(XIII) chains no disulfide-bonded trimers could be detected in the cells or medium (Fig. 2B, lanes 3 and  4). Thus it can be concluded that the N-terminally mutant ␣1(XIII) chains are deficient in terms of their capacity to form disulfide-bonded trimers. It should be noted that in some of the samples a band between the trimers and the monomers was observed representing dimers of ␣1(XIII) chains (Fig. 2B).
The Transmembrane Domain Is Implicated in Efficient Chain Association-To test whether inclusion of the previously described association domain (4) would restore the capacity of the ␣1(XIII) chains to form disulfide-bonded trimers, the del1-61 variant lacking only the cytosolic and transmembrane domains was tested. Only minute amounts of trimers were detected in the cell extract; however, the trimer/monomer ratio was very low compared with del1-38, lacking only the cytosolic domain (Fig. 2B, lanes 7 and 1, respectively), and only monomers were detected in the medium (Fig. 2B, lane 8). Furthermore, the del1-83 ␣ chains, which practically correspond to the XIII N-long and XIII N-short ␣ chains in terms of their type XIII collagen sequences but lack the 11 or 65 residues of mutant N-terminal sequences, did not form disulfide-bonded trimers (Fig. 2B, lanes 9 and 10). Del1-83 contains the pairs of cysteine residues near the end of the NC1, at the end of the COL1, and at the beginning of the NC2 domain, but at least in this variant, these cysteines appear not to be utilized for interchain bonds. All in all, the data suggest that the previously identified transmembrane-proximal chain association region of about 20 residues is not sufficient to ensure proper association and disulfide bonding of the ␣1(XIII) chains, as the transmembrane domain is also required.
The COL1 Domains of the N-terminally Truncated ␣1(XIII) Chain Variants Are Pepsin-sensitive-Although the N-terminally truncated ␣ chains are lacking the sequence shown to be important for ␣1(XIII) chain association, these modified ␣ chains may also associate and trimerize through other sequences. Thus we considered it possible that they could form triple-helical trimers that lack interchain disulfide bonds and thus were not detected in the non-reduced Western blot (Fig.  2B). To study this possibility, High Five insect cells were coinfected with the viruses del1-61, del1-83, XIII N-short , or XIII N-long together with 4PH␣␤, and the ability of the modified chains to form stable triple-helical domains was assessed enzymatically. The results were compared with High Five cells infected with the virus del1-38 together with 4PH␣␤, because insect cell expression of del1-38 ␣ chains has been shown previously (6) to result in the formation of homotrimers with three triple-helical collagenous domains.
The Triton X-100-soluble proteins in the cell supernatants were digested with pepsin, which is unable to digest triplehelical collagen sequences, and thus any pepsin-resistant fragments are likely to be derived from triple-helical molecules. Samples of the pepsin-digested cell supernatants were analyzed by SDS-PAGE under reducing conditions, followed by Western blotting with various antibodies. Digestion of the del1-38 sample resulted in four pepsin-resistant fragments when detected using the antibody XIII/NC1-Q610, which reacts with NC1 domain sequences (Fig. 3A, lane 1). Their observed molecular masses were 19, 22, 26, and 29 kDa. On the other hand, faint 19-and 22-kDa bands could be detected during digestion of the del1-61 and del1-83 samples (Fig. 3A, lanes 2  and 3), but once the digestion was complete, no bands representing the COL1 domain could be detected.
Western blotting of the pepsin-digested del1-38 sample with the NC2-specific antibody XIII/NC2-55 detected four pepsinresistant fragments of sizes 26, 29, 30, and 32 kDa (Fig. 3A,  lane 4). The bands recognized by both XIII/NC1-Q610 and XIII/NC2-55 antibodies are considered to represent pepsinresistant fragments originating from the COL1 domain flanked by non-collagenous sequences, whereas the 30-and 32-kDa fragments represent the COL2 domain, as they can also be detected with an antibody XIII/NC3-1 raised against the NC3 domain of type XIII collagen (Fig. 3A, lane 7). When the pepsindigested del1-61 and del1-83 samples were Western-blotted with antibody XIII/NC2-55, three fragments of size 30, 31, and 33 kDa, representing the COL2 domain (see below), could be observed (Fig. 3A, lanes 5 and 6). No bands representing the COL1 domain could be detected with this antibody (see below). Thus both antibodies against the NC1 and NC2 domains demonstrate that the COL1 domains of the del1-61 and del1-83 chains are pepsin-sensitive.
Western blotting of the pepsin-digested del1-61 and del1-83 ␣ chains with the antibody XIII/NC3-1 detected pepsin-resistant fragments of 31 and 33 kDa (Fig. 3A, lanes 8 and 9), which were also detectable with antibody XIII/NC2-55 (Fig. 3A, lanes  5 and 6). In contrast to the COL1 domain, the COL2 domain fragments of the del1-61 and del1-83 chains can be detected at equal intensity to the COL2 domain fragments of the del1-38 chains. It should be noted that the mobility of the COL2 fragments recognized by the antibodies against the NC2 and NC3 domains is slower in the del1-61 and del1-83 samples than in del1-38. It is conceivable that the del1-61 and del1-83 chains are subject to a higher extent of post-translational modifications, due to delays in chain association and triple helix formation, and that this causes the slight difference in their electrophoretic mobility compared with del1-38.
Pepsin digestion of the del1-38, del1-61, and del1-83 ␣ chains and subsequent Western blotting with antibody XIII/ NC4-SO, produced against a synthetic peptide corresponding to the C-terminal end of the COL3 domain of human type XIII collagen and the whole NC4 domain, revealed one 36-kDa pepsin-resistant fragment of equal intensity in each case (Fig.  3A, lanes 10 -12). This fragment thus represents the COL3 and NC4 domains. The pepsin digestion results with regard to the human variants are summarized in Fig. 3B.
The COL1 Domains of the N-terminally Altered Type XIII Collagen ␣ Chain Variants Are Pepsin-sensitive-The del1-38 variant was used as a control for the mouse N-terminally altered XIII N-short and XIII N-long ␣1(XIII) chains as well. Pepsin digestions were performed, and Western blottings were prepared as described for the human variants above. The antibodies XIII/NC1-Q610 and XIII/NC2-55 were not used with mouse type XIII collagen, because the mouse XIII N-short and XIII N-long variants lacked most of the XIII/NC1-Q610 epitope, and the mouse NC2 sequences differed substantially from the human corresponding sequences. Western blotting with the XIII/  lanes 7-9), and XIII/NC4-SO (lanes 10 -12). Lanes 1,4,7, and 10 refer to del1-38 samples; lanes 2, 5, 8, and 11 to del1-61 samples; and lanes 3, 6, 9, and 12 to del1-83 samples, all digested with pepsin. B provides a schematic presentation of the pepsin-resistant fragments of the variants del1-38, del1-61, and del1-83 and the antibodies used, and the location of the cysteine residues (C). C, the pepsin-digested cell supernatant samples XIII N-short (lanes 2 and 5) and XIII N-long (lanes 3 and 6) together with del1-38 (lanes 1 and 4) were subjected to Western blotting with antibodies XIII/NC3-1 (lanes 1-3) and XIII/NC4-SO (lanes 4 -6). The XIII N-short protein partly purified from the conditional medium (lane 7) was digested with pepsin and analyzed on 12% SDS-PAGE gel under reduced conditions followed by transfer to nitrocellulose filter and Coomassie Blue staining. A schematic presentation of the pepsin-resistant fragments of the mouse N-terminally altered variants XIII N-short and XIII N-long is given in D, in which the cysteine residues (C) and antibodies used are also indicated. NC3-1 antibody revealed 30-and 32-kDa bands in the pepsindigested sample of del1-38 and a 32-kDa band in pepsindigested samples of XIII N-short and XIII N-long (Fig. 3C, lanes  1-3). Western blotting with the antibody XIII/NC4-SO identified a single band in all samples, of size 36 kDa in del1-38 (Fig.  3C, lane 4) and 38 kDa in the N-terminally altered samples (Fig. 3C, lanes 5 and 6).
To study the properties of all of the collagenous domains, insect cells in suspension culture were infected with the variant XIII N-short virus, and the corresponding protein was partly purified from the conditional medium, pepsin-digested, and analyzed by SDS-PAGE under reducing conditions, followed by transfer onto a ProBlott TM membrane and Coomassie Blue staining. This revealed three pepsin-resistant fragments of sizes 30, 32, and 38 kDa (Fig. 3C, lane 7). The bands were cut from the membrane and subjected to N-terminal sequencing. The 32-kDa band was found to represent the fragment containing the COL2 domain and a portion of the NC2 domain, as the N-terminal sequence 245 KGEQSQTGIQ (residue 245 of the moXIII (689) sequence) is located in the latter. The 30-kDa band has the same N-terminal sequence as the 32-kDa band, indicating that these protein fragments have been processed by pepsin at the same site in their N terminus and thus vary in their C-terminal sequence. The 38-kDa band corresponded to the COL3 and NC4 domains, with the N-terminal 442 LALMG-PPGLP (residue 442 of the moXIII (689) sequence) located at the end of the NC3 domain. No pepsin-resistant band representing the COL1 domain was detected. These results indicate that folding of COL1 to a pepsin-resistant triple helix is compromised when the NC1 association domain is missing, but surprisingly, the COL2 and COL3 domains are correctly folded. A summary of the pepsin-resistant triple-helical fragments of mouse variants is provided in Fig. 3D.
Potential Coiled-coil Structures in the NC1 Domain of Several Collagenous Transmembrane Proteins-We next examined the association domain sequence of the type XIII collagen NC1 domain more closely in order to obtain a better understanding of its nature and extent. It is known that ␣-helical coiled-coils function as specific oligomerization domains that determine the association state and chain composition of various proteins, whereas collagenous triple helices are characterized by a slow folding speed and poor selection properties (24). We thus set out to analyze whether the type XIII collagen NC1 sequence would qualify as a possible coiled-coil element, using the COILS program (21). Type XIII collagen revealed a high probability of forming coiled-coils just after the transmembrane domain when a sliding window size of 14 residues was used (Fig. 4). On the other hand, if a window size of 21 or 28 residues was adopted, the coiled-coil domain of type XIII collagen already begins within the transmembrane domain, albeit at a low probability (Fig. 4). The fact that lack of the transmembrane domain sequences reduced the capability for forming disulfidebonded trimers in the case of the del1-61 ␣1(XIII) chains confirms that sequences within the transmembrane domain are needed for efficient chain association.
The search was extended to cover the other collagenous transmembrane proteins as well, namely collagen types XVII, XXIII, XXV, MARCO, and EDA, all of human origin (26 -30). Collagen types XVII, XXIII, XXV, and MARCO were also found to have a potential coiled-coil domain next to the transmembrane domain, this already starting within the transmembrane domain (Fig. 4). The same was observed for EDA, albeit with a very low probability (Fig. 4). These coiled-coil regions of collagen types XIII, XVII, XXIII, XXV, MARCO, and EDA were 40, 37, 35, 22, 39, and 24 residues long, respectively.
The NC1 and NC3 Coiled-coil Sequences of Type XIII Colla-gen Are Conserved in Collagen Types XIII, XXIII, and XXV-The pepsin sensitivity of COL1 and pepsin resistance of COL2 and COL3 raised the question of additional association and folding elements to those present in the NC1 of type XIII collagen. Interestingly, the COILS program predicted that type XIII collagen would have a second coiled-coil sequence toward its C-terminal part, namely at residues 437-465 (Fig. 4, 29 residues in length). Similarly, strong coiled-coil predictions were given for type XXIII collagen, at residues 391-410 (Fig. 4, 20 residues), and type XXV collagen, at residues 422-450 (Fig.  4, 29 residues). In all of these collagens the second coiled-coil domain encompasses the NC3 domain. Sequence comparisons indicated that a 21-residue stretch (residues 60 -80) of the 40-residue NC1 coiled-coil domain of type XIII collagen is 48 and 38% identical and 52 and 57% homologous to the corresponding sequences in collagen types XXIII and XXV, respectively (Fig. 5A). In the case of the NC3 coiled-coil domain, 24 residues (residues 437-460) out of the 29-residue type XIII collagen sequence are 21 and 54% identical and 38 and 71% homologous to the corresponding sequences in collagen types XXIII and XXV, respectively (Fig. 5B).
Coiled-coil Regions of Type XIII Collagen-Coiled-coil domains are characterized by a heptad repeat of amino acids (abcdefg) n , with a predominance of hydrophobic residues in the a and d positions and charged residues frequently in the e and g positions. Despite the preference for hydrophobic residues, ϳ20% of the a and d residues are polar or charged (31). The heptad repeats of type XIII collagen, as predicted on the basis of the COILS program, are shown in Fig. 5, C and D, respectively. It is predicted that the first 12 residues of the 40-residue NC1 coiled-coil domain are included in the transmembrane domain (11).
Coiled-coil Analysis of the Novel Type XXVI Collagen-While extending the search for potential internal coiled-coil domains to encompass non-transmembrane collagens as well, we found the recently described type XXVI collagen (32) to have two potential coiled-coil domains (Fig. 6A), an N-terminal one in the NC1 domain, encompassing residues 129 -159 (31 residues), and a C-terminal one in the NC3 domain, covering residues 348 -388 (41 residues). Comparison of the NC1 and NC3 coiled-coil sequences of type XIII collagen with the whole sequence of type XXVI collagen revealed sequence homology between the respective NC1 coiled-coil domains (Fig. 6B) but not between the NC3 domains. More specifically, a 21-residue portion of the 40-residue type XIII collagen NC1 coiled-coil domain is 48% identical and 57% homologous to the corresponding sequence in type XXVI collagen (Fig. 6B). DISCUSSION Authentic type XIII collagen molecules lack an N-terminal signal sequence, and instead the transmembrane domain is presumed to be important for their endoplasmic reticulum translocation and subsequent transmembrane anchorage. Thus it is surprising that the mouse N-terminal mutant ␣1(XIII) chains lacking the transmembrane and adjacent chain association domain were found to be correctly deposited in adherence structures in tissues and in focal adhesions in cultured cells (13). The data presented here, obtained using recombinant protein expression in insect cells, indicate that the N-terminally altered molecules are translocated into the endoplasmic reticulum and secreted out of the cell in a manner comparable with ␣1(XIII) chains equipped with the transmembrane domain. This is evident from the similar or slightly higher amounts of type XIII collagen found in the insect cell culture media in the case of XIII N-long and XIII N-short ␣ chains by comparison with membrane-anchored ␣1(XIII) chains. Moreover, the N-terminally altered protein molecules found in FIG. 4. ␣-Helical coiled-coil prediction for collagenous transmembrane proteins. The human protein sequences indicated were analyzed for putative coiled-coil domains using the COILS program (version 2.1 (21)). The scoring matrix MTIDK, derived from myosins, paramyosins, tropomyosins, intermediate filaments of types I-V, desmosomal protein, and kinesins, was used, with 2.5-fold weighting of residues in heptad positions a and d. Sliding window sizes of 14, 21, and 28 residues were used. The abscissa shows the scale in amino acids (aa), and the y axis indicates coiled-coil probability. Schematic structures for the corresponding molecules are shown below the graphs. The non-collagenous domains are shown as black boxes, the transmembrane domains as transparent gray boxes, the coiled-coil domains as striped red boxes, the collagenous domains as white boxes, and a tumor necrosis factor motif as a green box. the media were proteolytically processed at a furin cleavage site. Membrane-anchored type XIII collagen molecules have been found previously (4) to be cleaved at the same site, and this cleavage can be inhibited by a furin inhibitor. Furin proproteases are predominantly situated in the trans-Golgi network, but they are also located in the plasma membrane (33). Thus lack of the 96 extreme N-terminal residues of authentic type XIII collagen does not impair proteolytic processing, and this cleavage is indicative of utilization of the same secretory pathway as in authentic molecules.
Human N-terminally truncated ␣1(XIII) chains lacking the first 83 residues have been shown to occur in part in the cytosolic compartment, whereas about half of the protein is translocated to the lumen of endoplasmic reticulum and secreted into the medium (6). In the case of the XIII N-long protein, we observed two bands in the cell fraction, one representing full-length ␣1(XIII) N-long chains and the other lacking 115 amino acids of the altered sequence at the N terminus, the start of localization in mouse protein being at amino acid 147. We consider it likely that this truncated protein is a degradation product of the full-length ␣1(XIII) N-long chains. Moreover, analysis of the ␣1(XIII) N-long chains in the cell culture medium indicated that the more truncated ones were not secreted. Thus it appears that the sequences following residue 147 do not include information on translocation to the lumen of endoplasmic reticulum, but such information may be included in the sequence between residues 96 and 147 in the mouse type XIII collagen sequence.
We have shown previously (6) that interchain disulfide bonds are formed by cysteines located in the NC1 domain or by the cysteines at the junction of the COL1 and NC2 domains, whereas the extreme C-terminal pair of cysteine residues in type XIII collagen molecules form intrachain disulfide bonds. The human deletion variants del1-61 and del1-83 and the mouse variants XIII N-short and XIII N-long lack the cysteine residues of the transmembrane domain but retain the pair of cysteines at the NC1/COL1 junction. Interestingly, none of the N-terminally mutant ␣1(XIII) chains was able to associate efficiently into disulfide-bonded trimers. This is the case even with the del1-61 ␣1(XIII) chains, which contain the entire NC1 association domain, conserved with other membrane-spanning collagenous proteins.
We have recently shown by means of pepsin and trypsin/ chymotrypsin digestions that the del1-38 ␣ chains form correctly aligned disulfide-bonded trimers with all three collagenous domains in triple-helical conformation, and that interchain disulfide bonds were associated with the COL1 domain (6). We considered it possible that the ␣ chains translocated to the lumen of the endoplasmic reticulum and lacking the N-terminal residues could form triple-helical molecules even though they lacked interchain disulfide bonds. Interestingly, the use of pepsin resistance to map triple-helical domains indicated that despite lack of the transmembrane domain and the NC1 association domain, it is possible for the modified ␣ chains to form trimeric molecules, where the COL2 and COL3 domains are in a pepsin-resistant triple-helical conformation. Nevertheless, the COL1 domains of trimeric type XIII collagen molecules composed of modified ␣ chains are sensitive to pepsin. The results imply that the transmembrane and association domain sequences are needed for correct folding of the COL1 domain. These sequences are not included in the proteolytically processed ectodomain, which nevertheless has a stable triplehelical COL1 domain (5). Thus it appears that the triple-helical conformation remains stable once it has been formed.
Coiled-coil domains are thought to have an important role as oligomerization domains. We show here that it can be predicted that the NC1 association domain will form a coiled-coil structure. Furthermore, the results imply that chains lacking the NC1 coiled-coil region can associate through sequences residing in other parts of the molecule, but the ensuing trimers are incorrectly folded with respect to the COL1 domain. Studies with the various N-terminal variants indicated that type XIII collagen seems to have another domain responsible for association and folding of the molecule, from COL2 to NC4. This is presumably the NC3 domain, in view of the potential coiled-coil structure encompassing this domain and the conservation of this structure among type XIII collagen-like molecules. The importance of the NC1 and NC3 domains is further highlighted by the fact that neither the NC1 nor the NC3 domain of type XIII collagen is affected by alternative splicing, whereas the rest of the domains except for the COL2 domain can be alternatively spliced (7-10, 14, 34).
Mice expressing the N-terminally altered type XIII collagen (Col13a1 N/N ) are viable and fertile but still cannot be considered completely normal. Expression of altered molecules affects mainly skeletal muscle integrity and can be detected as an abnormal plasma membrane-basement membrane interphase (13). Our data obtained with recombinantly produced, identically altered ␣1(XIII) chains indicate that the modified type XIII collagen molecules are correctly folded with respect to the C-terminal two-thirds of their structure. Consequently these mutant molecules may retain some aspects of the function of normal type XIII collagen. The findings recorded with the Col13a1 N/N mouse line and the properties of the mutant molecules also imply that the cytosolic and transmembrane domains of type XIII collagen are of functional significance but not necessarily for all aspects of type XIII collagen function. Moreover, this N-terminally altered type XIII collagen causes a considerably milder phenotype than with another ␣1(XIII) mutation in which a 90-residue in-frame deletion of COL2 domain FIG. 5. Multiple alignment of coiled-coil sequences of collagen types XIII, XXIII, XXV, and heptad repeats in collagen type XIII. Alignment of the amino acid residues in the NC1 (A) and NC3 (B) domains of human type XIII collagen (1st row) with the corresponding residues in human type XXIII collagen (2nd row) and human type XXV collagen (3rd row). Alignment was achieved using a BLOSUM matrix. The amino acid sequences are shown in one-letter codes. Gaps (Ϫ) were introduced for maximal alignment of the polypeptides. Identical amino acids are indicated by black boxes and similar ones by gray boxes. Heptad repeats in the NC1 (residues 50 -89) and NC3 (residues 437-465) domains of human type XIII collagen are shown in C and D, respectively. The letters a-g indicate positions in the heptad coiled-coil repeat. Hydrophobic residues in positions a and d of the heptad are written in boldface italics. Charged and polar residues in positions e and g are underlined.
leads to fetal lethality (35). In this case the mutant molecules remain attached to the plasma membrane, folding of the affected domain is likely to be impaired, and the mutant molecules may produce incorrect signals in the cell.
The family of collagenous transmembrane proteins currently has 9 members, i.e. collagen types XIII, XVII, XXIII, and XXV, the macrophage scavenger receptors (MSRs), a macrophage receptor with collagenous structures (MARCO), a scavenger receptor with C-type lectin (SRCL), the complement component C1q, and ectodysplasin-A (15, 29, 30, 36 -38). Of these, human type XVII collagen, bovine MSRs, mouse MARCO, and human SRCL have been shown previously (15,16,24,25) to possess coiled-coil structures, the coiled-coil region in type XVII collagen and MARCO extending directly from the transmembrane region, whereas that in the MSRs and SRCL is separated from the transmembrane domain by a spacer domain of 32 and 56 amino acid residues, respectively. Furthermore, the coiled-coil region is much shorter in type XVII collagen and MARCO than in MSRs and SRCL, 39 and ϳ28 amino acid residues, respectively, versus 163 and 223. Here coiled-coil structures were also identified in other collagenous transmembrane proteins, namely in collagen types XIII, XXIII, XXV, and EDA. As with collagen type XVII and MARCO, the potential coiled-coils in types XIII, XXIII, XXV, and EDA also start from the transmembrane domain and are relatively short, 40, 35, 22, and 24 residues, respectively.
Interestingly, our results also indicate that the type XIII collagen family of molecules possesses two coiled-coil domains, one located adjacent to the transmembrane domain and partly overlapping with it and an additional one in the NC3 domain region. These function as two independent oligomerization domains. All in all, the collagenous transmembrane proteins can now be characterized as proteins having an N-terminal transmembrane domain and one or more collagenous domains interspersed by one or more coiled-coil domains, each having its own characteristic function: anchoring, ligand binding, and oligomerization, respectively. Furthermore, these family members can be divided into two groups depending on the features of their N-terminal oligomerization domain, i.e. collagen types XIII, XVII, XXIII, XXV, MARCO, and EDA, having a short, conserved oligomerization domain that is linked to the transmembrane domain, as opposed to MRSs and SRCL, having a long, separate oligomerization domain. Moreover, the occurrence of internal coiled-coil domains has also been identified in the recently described type XXVI collagen, which has been shown to have a signal sequence but not a transmembrane sequence. Interestingly, this collagen has a consensus sequence for furin cleavage, RRRR, in its NC1 domain, but this was not shown to be involved in secretion of the protein (32). We detected here two coiled-coil domains for type XXVI collagen, the NC1 coiled-coil being homologous to that of type XIII collagen. Whether this means that the NC1 coiled-coil of type XXVI collagen acts as an oligomerization domain for this collagen remains to be seen.