Deletion of epidermal growth factor-like domains converts mammalian tolloid into a chordinase and effective procollagen c-proteinase

Bone morphogenetic protein (BMP)-1 and mammalian tolloid (mTld) are Ca 2 (cid:1) -dependent metalloproteinases that result from alternative splicing of the bmp1 gene. They have different proteinase activities, e.g. BMP-1 ef-fectively cleaves procollagen (an extracellular matrix protein) and chordin (a BMP antagonist), whereas mTld is a poor procollagen proteinase and will not cleave chordin in the absence of twisted gastrulation. This is perplexing because mTld (being the longer variant) might be expected to cleave all substrates cleaved by BMP-1. Studies have shown that the minimal structure for procollagen proteinase activity is proteinase-CUB1-CUB2 (BMP-1 (cid:2) EC3) and therefore lacking the epidermal growth factor (EGF)-like domain thought to account for the Ca 2 (cid:1) dependence of BMP-1. In this study we generated three deletion mutants of mTld that lacked either one or both EGF-like domains (referred to as “mTld- (cid:2) EGF”). The mutated proteins were poorly but sufficiently secreted from 293-EBNA cells for in vitro assays of procollagen and chordin cleavage. Most sur-prisingly, the

Bone morphogenetic protein (BMP)-1, 1 which is also known as procollagen C-proteinase-1 (PCP-1), was first isolated from osteogenic extracts of bone (1,2). BMP-1 belongs to the tolloid group of astacin-like metalloproteinases that are fundamental to tissue patterning and extracellular matrix assembly in animals. The proteinase is synthesized as a proprotein (from the N terminus) consisting of a signal peptide, a prodomain (which is cleaved by dibasic furin-like proprotein convertases (3)), a metalloproteinase domain that is homologous to astacin (4), two CUB domains (a protein domain first found in the complement components C1r/C1s, the sea urchin protein Uegf, and BMP-1), an EGF-like domain, a further CUB domain, and a short specific sequence at the C terminus. BMP-1 cleaves precursors of fibrillar collagens (5)(6)(7)(8), and other extracellular matrix proteins including biglycan (9), type VII procollagen (10), prolysyl oxidase (11,12), and laminin chains (13,14). Dentin matrix protein-1, a protein involved in initializing mineralization of bones and teeth, has also been identified recently as a new substrate for BMP-1/tolloid-like proteinases (15), as well as myostatin (16), a transforming growth factor-␤ family member that is essential for proper regulation of skeletal muscle growth (17). The importance of BMP-1 in tissue assembly and development is illustrated in bmp1 homozygous null mice, which are perinatal lethal and have defects in ventral body wall closure and collagen fibrillogenesis (18). mTld is the longer splice variant of the bmp1 gene. The mTld protein is identical to BMP-1 except that it contains one additional EGF-like domain and two additional CUB domains at its C terminus. The most C-terminal end of the protein contains a sequence specific to mTld. Thus, mTld contains five CUB domains and two EGF-like domains, which in other proteins are involved in protein-protein interactions and Ca 2ϩ binding, respectively. The EGF-like domains consist of 40 amino acids, in which there are six highly conserved cysteine residues that form three disulfide bridges. Two types of 6-cysteine EGF-like domains have been observed, one of which (cbEGF) has a high affinity for Ca 2ϩ . Calcium ion binding occurs in domains containing the consensus sequence (D/N)X(D/N)(E/Q)(X) m (D/ N)*(X) n (Y/F), where X is any amino acid; m and n are integers; and * indicates possible ␤-hydroxylation (19). In a previous study (20), we showed that the EGF-like and CUB3 domains of BMP-1 are not required for cleavage of type I procollagen. Thus the minimal domain structure for procollagen C-proteinase activity is as follows: metalloproteinase-CUB1-CUB2-specific domain (BMP-1⌬EC3). The efficiency with which the minimal C-proteinase cleaves procollagen is in contrast to the poor C-proteinase activity of mTld. We hypothesized that the poor C-proteinase activity of mTld was the result of conformational differences between BMP-1 and mTld, which are the result of the second EGF-like domain or the CUB4 and/or CUB5 domains. These proteinases have proved difficult to study by high resolution structure determination techniques such as x-ray crystallography because they are multidomain glycoproteins that are refractory to crystallization. Thus, we used the approach to engineer site-directed mutations into the mTld molecule to determine the structural basis for the poor C-proteinase activity of mTld.
The functional differences between BMP-1 and mTld are not restricted to the ability to cleave the C-propeptides of procollagen. A major substrate for BMP-1 is chordin, which, significantly, is not cleaved by mTld, at least in the absence of twisted gastrulation (21). Chordin (and the fly homologue short gastrulation (sog) (22,23)) are potent antagonists of BMPs, which are important regulators of early vertebrate and invertebrate dorsal-ventral development (24 -27). The activities of BMPs in vertebrates and dpp (decapentaplegic) in Drosophila are modulated by several secreted factors including chordin/sog (24, 28 -30). Chordin forms latent complexes with BMP2/4 and BMP4/7, thereby preventing the BMPs from binding to cognate receptors (31). BMP1 Xenopus xolloid cleaves chordin (32), producing ventralization and anti-neural activities. In Drosophila, a similar process occurs in which tolloid, the Drosophila homologue of xolloid, cleaves sog (28,33) and enhances the activity of Dpp.
In this study, we investigated the roles of EGF-like domains on the secretion, enzymic activities, and Ca 2ϩ dependence of BMP-1/mTld, using human type I procollagen and human chordin as substrates in vitro. We generated a series of mTld mutants lacking one or both EGF-like domains, expressed and purified the variant proteins. We found that the ⌬EGF mutants were expressed at similar levels to wild-type mTld but were not efficiently secreted from the cells. Wild-type BMP-1 and the mutated BMP-1 lacking its EGF and CUB3 domains (BMP-1⌬EC3) were also included in our study. We showed that mTld-⌬EGF1, mTld-⌬EGF2, mTld-⌬EGF1 ϩ 2, and BMP-1⌬EC3 remained calcium ion-dependent. The mTld-⌬EGF mutants were better C-proteinases than mTld in the presence of Ca 2ϩ and acquired chordinase activity. Furthermore, the chordinase activity was retained by BMP-1⌬EC3. Although wild-type mTld undergoes little conformational change upon Ca 2ϩ binding, the mTld-⌬EGF mutants were compact in the absence of Ca 2ϩ . We conclude that the EGF-like domains do not account for Ca 2ϩ dependence and that they are structure stiffeners. We hypothesize that they may function to hold the CUB4 and -5 domains of mTld in locations that exclude substrates cleavable by BMP-1.

EXPERIMENTAL PROCEDURES
Source of Materials-Full-length BMP-1 cDNA (GenBank TM accession number M22488) was cloned from a human placental cDNA library. A His 6 tag amino acid sequence was introduced into the BMP-1 sequence (BMP-1-His) immediately 5Ј of the stop codon. The cDNA encoding His-tagged BMP-1 was subcloned into the episomal expression vector pCEP4 (Invitrogen). Both wild-type BMP-1 and BMP-1 minimal structure for C-proteinase activity (i.e. BMP-1⌬EC3, lacking EGF and CUB3 (20)) were His 6 -tagged. Mammalian tolloid (mTld) (GenBank TM accession number U50330) was obtained by synthesis of the sequence from position 1883 to the stop codon (Genscript). A V5-His 6 tag amino acid sequence (GKPIPNPLLGLDST-(His) 6 ) was introduced into the mTld sequence immediately 5Ј of the stop codon. V5 epitope is recognized by a mouse monoclonal anti-V5 antibody (Invitrogen). An XhoI restriction site was introduced after the stop codon for cloning purposes. The synthesized cDNA in pUC18 (Invitrogen) was digested by AviII (position 1954) and XhoI (after the stop codon) and ligated into the BMP-1 cDNA in pCEP4, hence forming the mTld-His cDNA clone. The rabbit polyclonal neoepitope antibody 1210 was raised against the first 10 amino acids of the metalloproteinase domain of BMP-1 (amino acids 121-130) (Sigma-Genosys) (34). The full-length human chordin cDNA (GenBank TM accession number AF209928) (35), cloned into the SpeI and EcoRI sites of a pGEM-T Easy vector (Promega), was first deleted of its 5Ј-noncoding region. For this, an SpeI-SfiI fragment was replaced by a small double-stranded oligonucleotide SpeI-SfiI DNA fragment containing the CAAA sequence as translation start site, flanked by a SpeI site and the ATG initiation codon. A c-Myc tag (EQKLISEEDL) (36), recognized by antibody 9E10 (Roche Applied Science), was introduced in full-length human chordin between the putative signal peptidase cleavage site and CR1, 14 residues after the cleavage site (after amino acid 40). The SpeI-EcoRI c-Myc-tagged fulllength chordin cDNA was then ligated to the NheI and EcoRI sites of pcDNA3 mammalian expression vector (Invitrogen). Stably transfected Chinese hamster ovary cells with this construct were shown to secrete the c-Myc-tagged chordin, which was demonstrated to antagonize bone morphogenetic protein (BMP)-2 in a cell differentiation assay. 2 The c-Myc-tagged chordin cDNA was then subcloned into the episomal expression vector pCEP4 (Invitrogen) and used to transfect 293-EBNA cells as described below.
Site-directed Mutagenesis-Deletion of the EGF-like domain 1 was as described (20). The ⌬EGF1-BMP-1 clone was used to generate the His-tagged mTld-⌬EGF1 mutant, by excising and re-cloning the fragment comprising the deletion by BamHI (position 1390) and AviII (position 1954). For the deletion of the EGF-like domain 2 of mTld, NotI restriction enzyme sites were inserted at the borders of the EGF2 domain by PCR, using Pfx polymerase (Invitrogen) as described (20). A two-step strategy was undertaken using the mTld tail in pUC18 as a template and the primers M13pUC18Forward (5Ј-CCCAGTCACGAC-GTTGTAAAACG) and E2NotaReverse (5Ј-ACTCGTCCTTGTCAGCGG-CCGCTGAGAAGAAGT) on the one hand and E2NotbForward (5Ј-GA-AGCCGGCGCGGCCGCTTGTGACCACAA) and pUC18DraIIReverse (5Ј-GGAGTAAAGGTCCTTGGTCTT) on the other hand. PCR products were purified, digested by XbaI (multicloning site of pUC18)/NotI, and NotI/MluNI (position 2494), respectively, and re-inserted into the mTld tail in pUC18. DNA sequencing (Applied Biosystems) was used to verify the deletion and to ensure that the cDNA clones were error-free. The deleted EGF2 domain fragment in pUC18 was then digested by AviII (position 1954) and XhoI (after the stop codon) and inserted into mTld-His/pCEP4, thus generating mTld-⌬EGF2 mutant. The double mutant lacking EGF-like domains 1 and 2 (mTld-⌬EGF1 ϩ 2 mutant) was obtained by cloning of the deleted EGF2 domain fragment into the mTld-⌬EGF1 clone.
Protein Expression-293-EBNA (ECACC 85120602) cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (Invitrogen) and 0.25 mg/ml geneticin (G418, Invitrogen) in a 37°C incubator with 5% CO 2 . Four g of wild-type or mutant plasmid/T25 were incubated with Lipofectin (Invitrogen) and added to 293-EBNA cells in serum-free Opti-MEM (Invitrogen), according to the manufacturer's instructions. Twenty four hours after transfection, media were removed and replaced by Dulbecco's modified Eagle's medium containing 10% serum for a further 24 h. The cells were trypsinized (Invitrogen) and diluted 1:5 for selection. After 48 h, selection was initiated by addition of 0.25 mg/ml hygromycin B (Invitrogen).
Preparation of Media and Cell Lysates-Confluent cells were rinsed three times with phosphate-buffered saline and incubated in Dulbecco's modified Eagle's medium lacking fetal calf serum for 24 h. Medium was centrifuged for 5 min at 1,600 ϫ g to remove cell debris, and the pH was stabilized by the addition of a 1:10 volume of 1 M Tris-HCl buffer, pH 7.4. For preparation of cell lysates, cells were rinsed in phosphatebuffered saline and incubated on ice with 500 l of RIPA buffer (150 mM NaCl, 1% sodium deoxycholate, 0.1% SDS, 10 mM Tris-HCl, pH 7.4 containing 10 mM EDTA, and protease inhibitor mixture (Roche Applied Science)) for 15 min with occasional shaking. The cells were scraped on ice and sonicated. The lysates were clarified by centrifugation at 13,000 ϫ g for 5 min at 4°C.
Purification of the His 6 -tagged Enzymes-The cell medium (10 ml) with 0.01% Tween 20 was loaded onto a HIS-Select HC nickel affinity gel (Sigma). The His-tagged proteins were eluted in 0.01 M imidazole in a buffer containing 0.1 M Tris-HCl, pH 7.4, 0.01% Tween 20, 0.1 M NaCl, and 0.01% NaN 3 . Eluted fractions were analyzed by discontinuous SDS-PAGE (10% separating gel) and analyzed using the GelCode E-Zinc Reversible Stain kit (Pierce) (for mTld) or stained with silver nitrate (for BMP-1). Proteins were excised and analyzed by mass spectrometry to confirm their identities. Where needed, the eluted proteins were concentrated using Microcon 50 (Amicon, Inc.) ultrafiltration devices.
Electrophoresis and Western Blotting-Supernatants, cell lysates, or purified proteins (see above) were resolved by electrophoresis on a 10% (w/v) SDS-Prosieve gel (BioWhittaker Molecular Applications) under reducing conditions and subjected to Western immunoblotting. The following antibodies were used where appropriate: the mouse monoclonal anti-V5 linked to horseradish peroxidase (HRP) antibody (Invitrogen); the mouse monoclonal 9E10 directed against the c-Myc tag followed by the secondary antibody (anti-mouse HRP conjugated IgG (Sigma)); and the rabbit polyclonal antibody 1210, directed against the active form of BMP-1/mTld, followed by the secondary antibody (anti-rabbit HRP conjugated IgG (Sigma)). The signal was detected by the enhanced chemiluminescence method (SuperSignal West Dura extended duration, Pierce) in the case of the 1210 and c-Myc antibody, or by regular ECL (Amersham Biosciences) in the case of the V5 antibody. In experiments in which mTld and deletion mutants of mTld were assayed for proteolytic activity, we quantified the levels of the bands corresponding to the active form of the enzymes by laser densitometry of enhanced chemiluminescence fluorograms exposed to preflashed films.
Native Gel Electrophoresis-The purified proteins were incubated in the presence or absence of 5 mM CaCl 2 for 30 min prior to electrophoresis. To the samples was added loading buffer (100 mM Tris-HCl, 10% glycerol, 0.0025% bromophenol blue, pH 8.6) in the presence or absence of 5 mM CaCl 2 . The proteins were resolved on NOVEX 8% Tris-glycine pre-cast gels (Invitrogen) pre-run beforehand for 30 min in the electrophoresis buffer (25 mM Tris, 192 mM glycine, pH 8.3) containing or not 5 mM CaCl 2 . Proteins were transferred to nitrocellulose, which was allowed to dry prior to immunoblotting as described above.
Assay of Procollagen C-proteinase and Chordinase Activities-Purified BMP-1, BMP-1⌬EC3 mutant, mTld, and its EGF deletion mutants were assayed for procollagen C-proteinase activity using human U-L-14 C-type I procollagen substrate (0.4 g) in 50 mM Tris-HCl buffer, pH 7.4, containing 0.12 M NaCl, and 0.01% (w/v) Brij-35 in the presence or absence of 5 mM CaCl 2 , at 37°C for 16 h. Where indicated, EDTA was added at a final concentration of 10 mM. Analysis of the cleavage products on SDS gels (7% separating; 3.5% stacking) was performed as described (34), and the cleaved products were visualized by exposing dried gels to a PhosphorImager plate (Fuji, type BAS III) in a Phospho-rImager (Fujix BAS 2000). Proteins corresponding to the pro-␣1(I) and pN␣2(I) chains of type I procollagen and type I pNcollagen, respectively, were quantified using AIDA 2.0 software. The percent cleavage was calculated by multiplying the intensity of the pN␣2(I), corrected for molecular mass, by the initial concentration of procollagen. Chordinase activity was determined in vitro by cleavage of c-Myc-tagged chordin in the same buffer as above, for 16 h at 37°C, followed by analysis of the cleavage products on a 10% SDS-Prosieve gel and Western blot using 9E10 antibody as described above. When needed the nitrocellulose was stripped and reprobed with a different antibody.

RESULTS
The mTld-⌬EGF Mutants Are Poorly Secreted from the Cells-To determine the contribution of each EGF-like domain of mTld to the enzymic activities of the molecule, we generated a series of deletion mutants that lacked each or both EGF-like domains, and we expressed the shortened proteinases in 293-EBNA cells (Fig. 1). NotI sites were introduced by PCR at the borders of the EGF domain of the cDNA clone, and restriction enzyme digestion was used to delete individual domains. Previous work showed that the NotI-derived alanine residues had no effect on the procollagen C-proteinase activity of BMP-1 when inserted at the domain junctions (20). The recombinant proteins contained a V5-peptide epitope at the C terminus to facilitate Western blot analysis. The culture media and cell lysates from the V5-tagged wild-type and mutant mTld were analyzed by SDS-PAGE (10%) under reducing conditions, and Western blot analysis was performed using the anti-V5 monoclonal antibody (Fig. 2). Only the mature forms of the wild-type and the mutant proteins (indicated with a black circle) were present in the media, whereas cell lysates showed the presence of the latent forms (indicated with an asterisk). We noted the presence of faster migrating proteins that were immunoreactive to the V5 antibody. These proteins were presumably degradation fragments of mTld because they were absent from the empty vector control (Fig. 2, 1st and last lanes). Therefore, these degradation products were not considered during quantification of the active enzymes. The increased migration of the mutated mTld proteins was expected because of the deletion of ϳ4.9 kDa per EGF-like domain from the polypeptide chain. Whereas wild-type and mutant proteins were equally well ex-pressed (see Fig. 2, cell lysates), the deletion mutants mTld-⌬EGF1, mTld-⌬EGF2, mTld-⌬EGF1 ϩ 2 were secreted less efficiently, representing 41 Ϯ 14, 27 Ϯ 8, and 32 Ϯ 11% (n ϭ 2) of mTld, respectively. The results showed that the EGF-like domains are required for efficient secretion of the molecule.
The mTld EGF Deletion Mutants and Minimal C-proteinase (BMP-1 ⌬EC3) Remain Calcium Ion-dependent-EGF-like domains in other proteins are thought to be involved in calcium ion binding. To evaluate the contribution made by these domains to the calcium ion dependence of mTld and BMP-1, similar amounts of active purified V5His-tagged recombinant proteins were assayed for PCP activity by cleavage of 14 Clabeled type I procollagen in the presence or absence of 5 mM CaCl 2 . mTld, mTld-EGF deletion mutants, BMP-1, and the minimal C-proteinase (BMP-1⌬EC3) had negligible procollagen C-proteinase (PCP) activity in the absence of calcium ions (Fig. 3). As shown previously, BMP-1 and BMP-1⌬EC3 exhibited comparable PCP activity (ϳ75% of type I procollagen cleaved in 16 h at 37°C) in the presence of calcium ions (Fig.  3B) (20). For reasons that were unclear to us, the 1210 antibody was not always effective at detecting mTld in Western blots. Therefore, we could not be confident about comparing levels of mTld and BMP-1 during assays of the proteins. (The use of the anti-V5 antibody was effective in determining the relative concentrations of mTld and mutants of mTld. Likewise, the use of the 1210 antibody was effective in determining the relative concentration of BMP-1 and mutants of BMP-1.) The mTld deletion EGF mutants exhibited PCP activities that were notably better than wild-type mTld, which has been reported to be a weak C-proteinase compared with BMP-1 (21) (Fig. 3A). In conclusion, our results showed that the EGF-like domains do not account for the Ca 2ϩ dependence of tolloids. Furthermore, we were mindful that the EGF deletion mutants of mTld might be less stable than wild-type mTld (even after purification). Nevertheless, the mutants were more effective C-proteinases than wild-type mTld during the 16-h assay period.

The Minimal C-proteinase (BMP-1⌬EC3) Cleaves Chordin-
It was reported previously that BMP-1, but not mTld, is able to cleave chordin (21). We wanted to determine whether the minimal C-proteinase (BMP-1⌬EC3) exhibited chordinase activity. To compare the ability of BMP-1 and BMP-1⌬EC3 to cleave chordin, similar amounts of purified His-tagged versions of each enzyme were incubated in the presence of 5 mM CaCl 2 with c-Myc-tagged human chordin (Fig. 4). In the conditions of the assay, c-Myc-tagged chordin was stable in the absence of added enzyme (Fig. 4A, lane 1). In contrast, chordin was cleaved by BMP-1 and BMP-1⌬EC3 (Fig. 4A, lanes 2 and 3). Cleavage was abolished by EDTA, which removes the catalytic zinc ion from the active site in the metalloproteinase domain in addition to calcium ions bound by the protein. Cleavage of chordin was assayed by the disappearance of the c-Myc-tagged full-length chordin (Fig. 4C, top band), and the appearance of a partially digested fragment containing the c-Myc tag but lacking the C-terminal end (N-t ϩ Int) and the 15-kDa N-terminaltagged fragment. The results show that the minimal domain structure for PCP activity is sufficient for chordinase activity.
mTld Lacking EGF-like Domains, but Not Wild-type mTld, Undergoes Dramatic Conformational Changes Upon Ca 2ϩ Binding-To investigate the differences between the mTld-EGF deletion mutants and wild-type mTld that could explain the gain of chordinase activity and the better PCP activity of the mutants, we examined the proteins by native gel electrophoresis in the absence and in the presence of 5 mM CaCl 2 , as described under "Experimental Procedures." Each enzyme was purified and equilibrated with 5 mM CaCl 2 prior to electrophoresis. The proteins were separated in a 8% Tris-glycine native gel, transferred to nitrocellulose, and detected by using the anti-V5-HRP antibody. As shown in Fig. 6, no detectable changes could be observed on the conformation of mTld in the presence or absence of Ca 2ϩ (lanes 1 and 5). In contrast, the mutant enzymes exhibited noticeable differences in electrophoretic migration in the presence and absence of Ca 2ϩ . These results indicate that the EGF-like domains have a major influence on the conformation of mTld and resist gross Ca 2ϩ -induced conformational changes of the protein. DISCUSSION In this study we showed that mTld EGF-like domains are important for the secretion, proteolytic activity, and conformation of mTld, but they do not account for its calcium ion dependence. We propose an explanation for why mTld is a poor PCP and chordinase based on a substrate exclusion mechanism in which the EGF-like domains stiffen the molecule in such a way that the CUB4 and/or -5 domains sterically block binding of some substrates.
To test the contribution of each EGF-like domain in mTld function, we used site-directed mutagenesis to generate mTld proteins lacking either or both of the two EGF-like domains and examined the proteins biochemically (Fig. 1). Wild-type and mutant proteins were expressed in 293-EBNA cells. Western blotting showed that the single and double EGF deletion mutants were secreted less well than mTld and that the EGF2 domain makes a major contribution to the stability of the protein. Furthermore, analysis of the cell lysates showed that these mutants are expressed at the same level as the wild-type mTld. Taken together, these results suggest that the EGF-like domains participate in mTld secretion, but without additive contributions. These results are in contrast to those obtained with the ⌬EGF and ⌬EC3 deletion mutants from BMP-1, which showed that deletion of the EGF-like domain had no effect on secretion (20).
We showed previously the minimal BMP-1 structure for PCP activity, BMP-1⌬EC3, was an effective PCP in the presence of calcium ions (20). This was a surprising finding because the BMP-1 EGF-like domain as well as the second EGF-like domain in mTld contain the calcium binding consensus sequence (data not shown) and were therefore believed to account for the calcium ion requirement for cleavage. It was also assumed that calcium dependence of mTld, demonstrated for its chick homologue in early studies (37), was linked to these cbEGF domains. The secretion of the mTld-⌬EGF mutants provided us with the opportunity to assay these molecules for PCP activity and to examine their calcium ion dependence. In the presence of calcium ions, we confirmed our previous result that BMP-1 and the minimal C-proteinase BMP-1⌬EC3 have comparable and efficient PCP activity (20) (Fig. 3B). On the other hand, wild-type mTld was shown to be a poor C-proteinase, with only 26% cleavage compared with 50 and 60% cleavage by the mTld deletion EGF mutants (Fig. 3A). This is consistent with previous reports (5,8,21) showing BMP-1 is a better C-proteinase than its longer splice variant mTld. None of the wild-type or mutant enzymes exhibited detectable PCP activity in the absence of calcium ions. However, the EGF deletion mutants all exhibited PCP activity which showed the following: (i) the EGF-like domains do not account for calcium ion dependence in BMP-1/mTld, and (ii) the calcium ions that are required for  The surprising fact that the deletion EGF mTld mutants were more efficient C-proteinases than wild-type mTld and that BMP-1 is known to be a better C-proteinase than mTld led us to the hypothesis that CUB domains 4 and 5 in mTld might limit or block the access or binding of procollagen and chordin. This "closed conformation" of mTld would be "opened" by deletion of the EGF domain(s) allowing binding of type I procollagen. To test this possibility, we chose human chordin as a substrate because it is efficiently cleaved by BMP-1 but not by mTld. Chordin, a 120-kDa protein essentially composed of four cysteine-rich domains (CR), is secreted by the Spemann's organizer (38) and is an extracellular antagonist of BMP signaling (31). BMP-1 cleaves chordin at two sites, one downstream of CR1 and the other downstream of CR3 (39). As BMP-1 and mTld differ by the replacement of the BMP-1-specific sequence by an EGF-like domain, and two CUB domains (Fig. 1), our data strengthen the assumption that the C-terminal domains of mTld may determine its inability to cleave chordin. Most interestingly, mTld exhibits some chordinase activity in the presence of twisted gastrulation (Tsg) in vitro (40). Nevertheless, it was later demonstrated that in vivo cleavage of chordin in mammals is shared by both bmp-1 and tll-1 gene products (6).
Our first question was to address whether BMP-1⌬EC3 would cleave chordin. BMP-1 and BMP-1⌬EC3 were found to be equally capable of cleaving c-Myc-tagged chordin (Fig. 4), indicating that the EGF-like 1 and CUB3 domains are not required for chordinase activity in vitro. Second, wild-type mTld was shown to be unable to cleave c-Myc-tagged chordin, even with increased amounts of enzyme, thus confirming previous findings (40). Most surprisingly, the deletion EGF mTld mutants that were found to be more efficient C-proteinases than wild-type mTld also acquired the ability to cleave chordin (Fig. 5). To test further our hypothesis that the mTld deletion EGF mutants adopt an opened conformation (thus allowing better accessibility of the substrates to the active site), we performed native gel electrophoresis in the presence and absence of calcium ions (Fig. 6). We observed that whereas wildtype mTld undergoes little conformational change after Ca 2ϩ binding, the mTld-⌬EGF mutants adopted a compact confor-mation in the absence of calcium ions. These observations are in agreement with a number of studies where it was demonstrated that conformational changes in structure can be associated with calcium ion binding, e.g. troponin C (41), fibrillin-1 (42,43), and the clotting factors IX and X, as well as the complement component C1r, where there is a change in orientation or alignment of domains (44,45). Furthermore, the conserved carboxylate residues contained in the EGF domain that are responsible for binding calcium ions with high affinity (D/N)X(D/N)(E/Q)(X) m (D/N)*(X) n (Y/F) (see Ref. 19) are thought to increase the stability of the tertiary structure (46). Of direct relevance to the effects of calcium ion binding to CUB and EGF domains are the recent studies of the x-ray crystal structures of mannan-binding lectin-associated protein 19 (MAp19) (47), mannose-binding protein-associated serine protease (MASP)-2 (48), and C1s complement component CUB1-EGF dimodule (49). These studies show that the CUB-EGF-CUB trimodule is a planar-linear structure when Ca 2ϩ is bound to the centrally located EGF domain (48). Studies of MAp19 show that a CUB-EGF dimodule can bind calcium ions and that the dimodule forms an antiparallel dimer. In the absence of a crystal structure of mTld, it is tempting to speculate the existence of a hairpin conformation of the CUB1 through CUB5 domains of the molecule in which the CUB2-EGF1 and EGF2-CUB4 are in anti-parallel close association. It might be expected that the CUB4-CUB5 dimodule covers totally or partially the CUB1/2 and/or metalloproteinase domains, thereby restricting the binding of mTld to chordin and procollagen. In this scheme, the absence of the EGF domains would disrupt this hairpin conformation and make the molecule more flexible. Consequently, the absence of the EGF2-CUB4-CUB5 domains from BMP-1, as well as the small size of CUB3, would minimize the effects of removing EGF1 from BMP-1 and thereby account for the similar activities of BMP-1 and BMP-1⌬EC3.
Our results show for the first time that calcium ions are bound by other domains of mTld, perhaps the metalloproteinase domain and/or the CUB domains (Fig. 3B). Indeed, calcium ion binding in the metalloproteinase domain has been reported in other metzincins such as snake venom zinc-endopeptidase adamalysin II, which shares a similar overall topology with astacin, and exhibits a virtually identical zinc environment (50). Its refined 2⅐0-Å x-ray crystal structure allowed the local- FIG. 7. Alignment of the CUB domains from BMP-1/mTld. Alignment of CUB1-5 domains of BMP-1/mTld, with the calcium binding CUB module signature (boxed) (49). Conserved residues defining the CUB domain signature (52) are labeled in the one letter code and conserved changes indicated by: h, hydrophobic; n, negatively charged; ϩ, positively charged; p, polar; a, aromatic; and o, presence of a hydroxyl group.
ization of a hepta-coordinated calcium ion. Furthermore, Ca 2ϩ binding to CUB domains has also been documented in the C1s complement component CUB1-EGF module (49), which forms a head-to-tail homodimer. One calcium ion is bound to each EGF domain, and a second calcium ion is bound to the distal end of each CUB1 module, through six ligands contributed by Glu-45, Asp-53, and Asp-98, two water molecules, and Tyr-17 (numbers according to C1s CUB1). Although the first site is involved in the intra-and inter-monomer CUB1-EGF interfaces, the second site provides extensive stabilization of the distal part of the CUB1 module. These acidic residues and the tyrosine are conserved in approximately two-thirds of the CUB repertoire (49), including BMP-1/mTld and procollagen C-proteinase enhancer-1 (51), and define a novel, calcium-binding CUB module subset. Noteworthy, all the CUB domains of mTld exhibit the signature, except the CUB1 domain, which lacks two of the three acidic residues (Fig. 7).
In conclusion, we hypothesize that the EGF domains could confer stiffness to the mTld molecule and might stabilize an anti-parallel arrangement of the CUB domains. Furthermore, we propose that a hairpin conformation of mTld holds the CUB4 and -5 domains in locations that exclude substrates cleavable by BMP-1 or limit their access to the active site. Indeed, binding of Ca 2ϩ by an isolated EGF domain has been shown to result in little effect on its conformation; instead changes appear to have longer range effects involving neighboring domains (44,45). Taken together, our data imply a role of the two C-terminal CUB in regulating the enzymatic activity of mTld by a substrate-exclusion mechanism.