Transforming Growth Factor- (cid:1) Induces Secretion of Activated ADAMTS-2 A PROCOLLAGEN III N-PROTEINASE*

The metalloproteinase ADAMTS-2 has procollagen I N-proteinase activity capable of cleaving procollagens I and II N-propeptides in vitro , whereas mutations in the ADAMTS-2 gene in dermatosparaxis and Ehlers-Danlos syndrome VIIC show this enzyme to be responsible in vivo for most biosynthetic processing of procollagen I N-propeptides in skin. Yet despite its important role in the regulation of collagen deposition, information re-garding regulation and substrate specificity of AD-AMTS-2 has remained sparse. Here we demonstrate that ADAMTS-2 can, like the procollagen C-proteinases, be regulated by transforming growth factor- (cid:1) 1 (TGF- (cid:1) 1), with implications for mechanisms whereby this growth factor effects net increases in formation of extracellular matrix. TGF- (cid:1) 1 induced ADAMTS-2 mRNA (cid:1) 8-fold in MG-63 osteosarcoma cells in a dose- and time-depend-ent, cycloheximide-inhibitable manner, which appeared to operate at the transcriptional level. Secreted AD-AMTS-2 protein induced by TGF- (cid:1) 1 was 132 kDa gels were transferred, as above, to Sequi-Blot polyvinylidene difluoride membranes (Bio-Rad). were excised, N-terminal amino acid sequences were determined by automated Edman degrada- tion at the Microchemistry Facility. Enzymatic Deglycosylation Assays— Enzymatic deglycosylation was performed using the Glycopro deglycosylation kit (ProZyme). Briefly, purified protein in 50 m M sodium phosphate buffer, pH 7.0, was dena- tured in 0.08% SDS, 40 m M 2-mercaptoethanol for 5 min at 100 °C. Triton X-100 was added to the cooled sample to chelate SDS, which was then incubated 3 h at 37 °C with various combinations of glycosidases, as and of 0.1 unit/ (cid:3) l PNGase F, 0.1 unit/ (cid:3) l sialidase A, 0.025 milliunit/ (cid:3) l endo- O -glycosidase, 0.06 milliunit/ (cid:3) l (cid:1) -1–4-galactosidase, and 0.8 milliunit/ (cid:3) l glucosaminidase. Reactions were stopped with 10 (cid:5) SDS-PAGE loading buffer and boiling for 5 min.

lagen precursors with N-and C-propeptides that are proteolytically removed to produce mature monomers capable of forming fibrils (1)(2)(3). A wealth of biochemical evidence has indicated the propeptides of these procollagens to be processed by a number of highly specific Ca 2ϩ -dependent metalloproteinases, with neutral pH optima. Cleavage of the C-propeptides of procollagens I-III is via a procollagen C-proteinase (pCP) activity (4 -6), now known to be provided by at least three different proteins, BMP-1, mTLD, and mTLL-1 2 (7-10), all of which are in the astacin family of metalloproteinases (11). Cleavage of the N-propeptides of procollagens I and II is via a procollagen I N-proteinase (pNP) activity (12)(13)(14)(15), which reportedly does not process procollagen III (14 -16), whereas cleavage of procollagen III has been reported to occur via a different procollagen III N-proteinase (pNPIII) activity, incapable of processing procollagen I (17)(18)(19). Partial amino acid sequencing of pNP (20) has led to cloning of full-length cDNA sequences for both bovine and human forms of the enzyme (21,22). Analysis of sequences demonstrated that mutations in the pNP gene lead to the recessively inherited human disorder Ehlers-Danlos syndrome type VIIC (EDSVIIC) and the analogous disease dermatosparaxis in cattle (21), both of which are marked by extreme fragility of the skin and accumulation in the skin of processing intermediates in which the C-but not the N-propeptides of type I procollagen have been removed (23)(24)(25). Analysis of sequences also showed pNP to belong to the recently described ADAMTS (A Disintegrin And Metalloproteinase with ThromboSpondin motifs) family of metalloproteinases (26) and led to designation of pNPI as ADAMTS-2. At present there are 19 reported vertebrate ADAMTS family members (27,28), which share a common domain structure. They resemble the ADAMs family of proteases in having pro-, adamalysin/reprolysin-like metalloprotease, disintegrin-like and cysteine-rich protein domains but differ in that they lack epidermal growth factor-like domains, and unlike many ADAMs proteases, they lack transmembrane domains (26,29). Instead, ADAMTS proteases contain variable numbers of thrombospondin type I-like repeats (26,29) which, at least in some family members, appear to be involved in binding to components of the ECM (30).
There are indications that ADAMTS proteases play a broad range of roles in development, reproduction, disease, and homeostasis. Aside from the role of ADAMTS-2 in biosynthetic processing of procollagens I and II, the phenotype of Adamts2null mice suggests a role in male fertility as well (31). AD-AMTS-1, -4, and -5/11 appear to constitute a subset of highly homologous ADAMTS family members with aggrecanase activity, important to homeostasis of cartilage and etiology of the arthritides (32)(33)(34). ADAMTS-1 was originally identified as a gene product induced by inflammation and associated with cachexia (26). The phenotype of Adamts1-null mice also suggests roles for ADAMTS-1 in growth, organogenesis, and female fertility (35), whereas both ADAMTS-1 and ADAMTS-8 have been shown to have anti-angiogenic activity (36). Roles for ADAMTS-like proteins in fertility and organogenesis in a broad spectrum of species are suggested by the finding that the gon-1 gene, which encodes an ADAMTS-like product, is essential for gonadal morphogenesis in Caenorhabditis elegans (37). Mutations in the gene for ADAMTS-13 have been shown to be causal for thrombotic thrombocytopenic purpura, perhaps due to its demonstrated ability to process von Willebrand factor, showing ADAMTS-13 to have an important role in human vascular homeostasis (38).
Despite the seeming importance of ADAMTS proteases in many biological processes, there has been little characterization of the regulation of their expression and activity. In a previous study of the regulation of procollagen C-proteinase activity and expression of BMP-1 and mTLD, by transforming growth factor-␤1 (TGF-␤1), we noted elevation of both pCP and pNPI activities in MG-63 osteosarcoma cells treated with TGF-␤1 (39). In the present study, we show that TGF-␤1 induces expression of ADAMTS-2 mRNA, apparently at the transcriptional level, resulting in expression of fully processed and active 132-kDa ADAMTS-2 protein from MG-63 cells. Processing of the ADAMTS-2 pro-domain is shown to be via a two-step process involving sequential cleavage by furin-like proprotein convertases at two consensus sites. Surprisingly, purified AD-AMTS-2 protein is shown to have pNP activity not only against procollagens I and II but against procollagen III as well, whereas processing of the procollagen III N-propeptide is shown to be decreased in EDSVIIC fibroblast cultures. These latter two findings indicate that the distinction that has been made between pNPI and pNPIII is a false one. Implications of the various data are discussed.

EXPERIMENTAL PROCEDURES
RNA Analysis-To generate probes for analysis of human AD-AMTS-2 RNA, ADAMTS-2 sequences were PCR-amplified from human dermal fibroblast cDNA (Clontech) using oligonucleotide primers 5Ј-TTTGGCCGAGACCTGCACCTGC-3Ј (forward) and 5Ј-TGACAGGAG-CATAGCCTTGCATGC-3Ј (reverse), corresponding to nucleotides 346 -367 and 1147 to 1124, respectively, of the published ADAMTS-2 sequence (GenBank TM accession number AJ003125, see Ref. 21). The resulting PCR product was used to screen a gt10 human placenta cDNA library (Clontech), yielding one positive clone with a 1231-bp insert extending from nucleotides 400 to 1630 that was subcloned into the EcoRI site of pBluescript II KSϩ (Stratagene). This insert was employed as a Northern blot probe. Restriction of the described construct with PstI, which cleaves after ADAMTS-2 nucleotide 786 and within the pBluescript polylinker, followed by religation of the plasmid, yielded a template that, upon linearization with EcoRI and transcription with T7 polymerase, yielded a 387-base riboprobe used for RNase protection assays.
Poly(A ϩ ) RNA was prepared from cultured MG-63 human osteosarcoma cells as described previously (39). Unless otherwise noted in the text, TGF-␤1 treatment of cells harvested for RNA was at 2 ng/ml for 24 h. To test stability of ADAMTS-2 mRNA, just confluent MG-63 cells, untreated or treated with TGF-␤1 as described above, were then treated with 10 g/ml actinomycin D for varying times in Dulbecco's modified Eagle's medium (DMEM) containing 0.1% fetal bovine serum (FBS) containing or lacking 2 ng/ml TGF-␤. Stability was determined by RNase protection assays. RNase protection and Northern blot assays were performed as described by Lee et al. (39).
Production of Recombinant Protein-For expression of recombinant ADAMTS-2 in 293-EBNA cells, a human ADAMTS-2 cDNA insert extending from nts 112 to 3633 of the published sequence (21), corresponding to the full-length protein minus signal peptide sequences and with PCR addition of sequences encoding DYKDDDDK-Stop to the 3Ј-end of the insert, was ligated between the NheI and BamHI sites of episomal expression vector pCEP-Pu/BM40s (40). Coding was thus for full-length human ADAMTS-2 differing from the native protein only in replacement of the native signal peptide by BM40 signal peptide sequences, to enhance secretion, and by addition of a C-terminal FLAG tag. Experiments in Fig. 7 were performed with recombinant AD-AMTS-2 from a clonal line of transfected 293 cells constitutively expressing ADAMTS-2 and prepared toward the end of the study. To prepare this clonal line, in-frame BM40s/ADAMTS-2/FLAG coding sequences were excised from pCEP-Pu with AflII and BamHI and inserted between corresponding sites of expression vector pcDNA3.1 (Invitrogen), which was transfected into 293 cells. 293 cells were grown in growth medium consisting of DMEM supplemented with 1 mM L-glutamine and 10% FBS, whereas 293-EBNA cells were grown in the same growth medium supplemented with 250 g/ml G418 (Invitrogen). Transfection of both types of cells was at 90% confluence with 10 g of expression vector per 100-mm culture dish, using LipofectAMINE (Invitrogen). After 36 h, transfected 293-EBNA cells were selected, and surviving cells were allowed to grow to confluent mass cultures in growth medium containing 5 g/ml puromycin (Sigma) and 250 g/ml G418. After 72 h, transfected 293 cells were selected in growth medium containing 500 g/ml G418 (Invitrogen), and surviving colonies of cells were ring cloned. Culture media of 36 clones were analyzed by immunoblot, and the clone expressing the highest levels of ADAMTS-2 was used for the studies of Fig. 7.
To prepare purified recombinant ADAMTS-2, mass cultures of transfected 293-EBNA cells were grown to confluence, washed twice with phosphate-buffered saline (PBS), and switched to serum-free DMEM containing 40 g/ml soybean trypsin inhibitor (Sigma) with or without 5 g/ml soluble heparin (Sigma). Conditioned media were harvested 24 h later, and protease inhibitors were added to final concentrations of 0.4 mM phenylmethylsulfonyl fluoride and 10 g/ml leupeptin. Harvested media were centrifuged to remove debris, and supernatants were stored at Ϫ70°C. FLAG-tagged ADAMTS-2 in media was bound to 1 ml of FLAG M2 matrix, and the matrix was washed with 10 ml of binding buffer (50 mM Tris-HCl, 150 mM NaCl, pH 7.4), using the supplier's protocols (Sigma). ADAMTS-2 was eluted with 1 mg/ml FLAG peptide (Sigma) in binding buffer, and fractions were collected and analyzed by Western blot. Purified full-length recombinant human ADAMTS-1, ADAMTS-4, and ADAMTS-5 were provided by Wyeth Research, Cambridge, MA.
Preparation of Cell Lysate and ECM-For the experiments of Fig. 7, confluent cells were rinsed twice with PBS, detached in PBS containing 5 mM EDTA, pelleted by centrifugation, and lysed in 1ϫ SDS-PAGE loading buffer. ECM remaining on dishes was rinsed twice with PBS and then scraped into 1ϫ SDS-PAGE loading buffer. Treatment of cells in serum-free growth media with 20 M of furin inhibitor decanoyl-RVKR-chloromethyl ketone (Bachem) was as described previously (41).
Induction of Secreted ADAMTS-2 by TGF-␤-MG-63 cells at 70% confluence were trypsinized and replated in low salt, low cysteine, low sulfate DMEM containing 10% dialyzed FBS (Hyclone), 4 mM L-glutamine, and 30 mM sodium chlorate. After 48 h, cells were trypsinized again and replated in the same type of chlorate-containing media. The next day, cells were switched to serum-free, chlorate-containing DMEM with 2 ng/ml TGF-␤1 (R & D Systems) and 40 g/ml soybean trypsin inhibitor, and conditioned media were harvested 48 h later. Cell layers were then treated 2 h with 0.006 IU/ml heparitinase (Seikagaku America) in serum-free DMEM at 37°C. These media were removed and pooled with conditioned media from the previous step, and pooled media were incubated 20 h with heparin-Sepharose beads (Amersham Biosciences) at 4°C. Beads were centrifuged 1 min at 1,000 ϫ g at 4°C and washed three times with PBS, and protein was eluted by boiling beads for 5 min in 30 l of 4ϫ SDS-PAGE loading buffer.
Immunoblots-Samples subjected to SDS-PAGE were transferred to Immobilon-P membranes (Millipore) as described (39). Blots were blocked 1 h with 2% bovine serum albumin in T-PBS (PBS, 0.05% Tween 20) and incubated overnight with primary antibody diluted 1:5000 in the same solution. Blots were washed with T-PBS and blocked again by 2% bovine serum albumin in T-PBS, followed by incubation with 1:5000 diluted secondary antibody. After washing 6 times in T-PBS, blots were incubated 4 min in SuperSignal West Pico substrate (Pierce) and exposed to film. The blot in the right panel of Fig. 9B was washed overnight in T-PBS after transfer, blocked 1 h with 3% bovine serum albumin in T-PBS, incubated overnight with primary antibody diluted 1:2000 in the same solution and, the next day, washed as above. Antibodies 1429, 1435, 1374, and 1442, raised against peptides whose sequences are presented in Fig. 3, were prepared and affinity-purified using techniques described previously (39). Peptide antibodies directed against sequences in the pro-␣1(I) C-telopeptide (LF-67), pro-␣1(III) C-propeptide (LF-69), and N-telopeptide (LF-71) have been described previously (42) and were the kind gifts of Dr. Larry Fisher (NIDCR, National Institutes of Health, Bethesda). Although antibody LF-71 was raised against pro-␣1(III) N-telopeptide sequences, it appears to recognize only forms that retain the N-propeptide (i.e. pro-␣1(III) and pN␣(III) chains) and does not recognize forms from which the N-propeptide has been removed (i.e. pC␣1(III) and mature ␣1(III) chains), perhaps due to conformational changes affecting availability of epitopes 3 (42).
Proteinase Assays-ADAMTS-2, prepared and purified as above, and/or mTLL-1, prepared and purified as described previously (7), were incubated for 20 h with 210 ng of types I-III procollagen in 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM CaCl 2 at 37°C. Reactions were stopped by adding 10ϫ SDS-PAGE loading buffer and boiling 5 min. Procollagen substrates, radiolabeled with [2,3-3 H]proline, were prepared and purified essentially as described previously (7). Samples were subjected to SDS-PAGE on 5% acrylamide gels, which were treated with EN 3 HANCE (DuPont) and exposed to film.
Amino Acid Sequence Analysis-Proteins on 5% acrylamide SDS-PAGE gels were transferred, as above, to Sequi-Blot polyvinylidene difluoride membranes (Bio-Rad). Bands were excised, and N-terminal amino acid sequences were determined by automated Edman degradation at the Harvard University Microchemistry Facility.
Culturing of EDSVIIC and Normal Human Dermal Fibroblasts-Human dermal fibroblasts (kind gifts of Dr. Peter H. Byers, University of Washington, Seattle) were grown in the presence of dextran sulfate to augment efficiency of enzymatic conversion of procollagens to collagens, essentially as indicated in Smith et al. (25). Cells were grown to ϳ80% confluence and treated overnight in growth medium containing 50 g/ml ascorbate. The next day, cells were switched to serum-free medium containing 50 g/ml ascorbate, 40 g/ml soybean trypsin inhibitor, and 0.01% dextran sulfate (Amersham Biosciences). After 20 h, cell layers were rinsed twice with PBS and scraped into hot SDS-PAGE loading buffer.

TGF-␤ Increases ADAMTS-2 mRNA Levels in MG-63 Osteosarcoma Cells but Not in Keratinocytes-
To determine whether intracellular levels of ADAMTS-2 mRNA were influenced by treatment with TGF-␤, just-confluent MG-63 cultures were treated 24 h with 2 ng/ml TGF-␤1. As can be seen (Fig. 1A), a single major mRNA ADAMTS-2 species (ϳ7-kb), just detectable in untreated MG-63 cells, is increased ϳ8-fold in TGF-␤1treated MG-63 cells. Moreover, ϳ4.5and 2.3-kb bands, not detected in untreated cells, become evident in the TGF-␤treated sample. All three bands are similar in size to Northern blot bands reported previously (21) to correspond to human ADAMTS-2 mRNA. To ensure that the observed induction was for ADAMTS-2 mRNA, and not for mRNA of some other closely related ADAMTS family member, a highly specific RNase protection assay, capable of detecting even single nucleotide differences, was conducted on the same RNA samples. The result was protection of a riboprobe fragment of the size predicted for ADAMTS-2 mRNA, and a level of induction in response to TGF-␤ similar to that detected by Northern blot (Fig. 1, A  and B).
In contrast to MG-63 cells, we had previously been unable to detect pNPI activity in cultured human keratinocytes, even upon treatment with TGF-␤1 (39). We were thus interested in determining whether this lack of activity was due to a lack of expression of ADAMTS-2 in keratinocytes or, as is the case for BMP-1 and mTLD of pCP (39), whether pNPI might be secreted by keratinocytes in an inactive form in response to TGF-␤1. To address this issue, an RNase protection assay was performed on RNA from human keratinocytes cultured in the presence or absence of TGF-␤1. As can be seen (Fig. 1B), no ADAMTS-2 mRNA is detected by the highly sensitive RNase protection assay in cultures of nonfibrogenic keratinocytes, even after treatment with TGF-␤1.
Incubations of MG-63 cell cultures with TGF-␤1 concentrations, ranging from 0.01 to 10 ng/ml, showed induction of AD-AMTS-2 mRNA to be dose-dependent ( Fig. 2A), with maximal induction at ϳ1.0 ng/ml. Induction at higher concentrations was somewhat less, possibly due to toxic effects of TGF-␤1 at higher concentrations (43). Kinetics of induction of ADMTS-2 mRNA were examined by incubating MG-63 cultures in 2 ng/ml TGF-␤1 for varying times (Fig. 2B). As can be seen (Fig. 2B), noticeable induction first occurs at ϳ6 h, with maximal induction at ϳ24 h post-treatment. Thus, kinetics for induction of ADAMTS-2 mRNA overlap those for induction of procollagen C-proteinases BMP-1 and mTLD, in which initial and maximal increases were at 12 and 24 h post-induction, respectively (39), and overlap the kinetics of induction of RNAs for various extracellular matrix proteins, as such induction generally occurs within 3-5 h of TGF-␤ induction of a variety of cell types (44). Treatment of confluent MG-63 cultures with cycloheximide prior to TGF-␤1 treatment inhibited induction of ADAMTS-2 (Fig. 2C). Thus, as with BMP-1 and mTLD mRNAs (39), induction of ADAMTS-2 mRNA by TGF-␤1 is indirect and requires prior protein synthesis. To obtain insights into roles that transcriptional and post-transcriptional processes might have on 3 L. W. Fisher and P. H. Byers, personal communication.

FIG. 1. TGF-␤1 induces ADAMTS-2 mRNA in MG-63 cells. A,
Northern blot previously used to analyze induction of BMP-1 and mTLD mRNA in MG-63 cells, by 24 h treatment with 2 ng/ml TGF-␤1 (39), was stripped and hybridized to a probe specific for human AD-AMTS-2. The result from hybridization of a ␤-actin probe to the same blot as a control for loading has been published previously (39). B, an autoradiogram is shown of an RNase protection assay of 20 g of total RNA from human epidermal keratinocytes (HEK) or MG-63 cells either treated with TGF-␤1 (T) or untreated controls (C). Protected fragments were electrophoresed on 6% sequencing gels.
increases in steady state levels of ADAMTS-2 mRNA in TGF-␤1-treated cells, just-confluent MG-63 cells, untreated with TGF-␤ or pretreated with 2 ng/ml TGF-␤1 for 24 h, were incubated with 10 g/ml of the transcriptional inhibitor actinomycin D for varying times (in media containing or lacking 2 ng/ml TGF-␤1), followed by determination of ADAMTS-2 mRNA levels by RNase protection assays (Fig. 2D). Although levels of ADAMTS-2 RNA were higher at all time points in TGF-␤1treated cells, the rate of decrease in ADAMTS-2 RNA levels over time was approximately the same for treated and untreated cells. This similarity in stabilities of ADAMTS-2 mRNA in TGF-␤1-treated and -untreated cells is consistent with the probability that increased steady state levels of ADAMTS-2 mRNA induced by TGF-␤1 in MG-63 cells are due to transcriptional effects, rather than to post-transcriptional enhancement of RNA stability.
Levels of Secreted ADAMTS-2 Protein Are Up-regulated by TGF-␤1 Treatment of MG-63 Cells-To determine whether TGF-␤1 induction of higher steady state levels of ADAMTS-2 mRNA is paralleled by increased secretion of ADAMTS-2 protein, we attempted to detect endogenous ADAMTS-2 protein in media of TGF-␤1-treated or -untreated MG-63 cells. Toward this end, we produced four polyclonal antibodies directed against four different peptides corresponding to different regions of ADAMTS-2 (Fig. 3). Only one of the four antibodies, antibody 1435, directed against sequences within the protease domain, was able to detect endogenous ADAMTS-2 produced by MG-63 cell cultures. Moreover, as seen in Fig. 4A, 1st and  2nd lanes, antibody 1435 detected endogenous ADAMTS-2 protein only in media of cells treated with TGF-␤1. The size of the induced band was ϳ132-kDa. This is larger than the molecular weight of 107 previously estimated for active ADAMTS-2 isolated from fetal calf skin, based on mobility on SDS-PAGE gels (20), but is similar to the 131,697 molecular weight predicted by the human cDNA sequence (21) for full-length ADAMTS-2 in which only the signal peptide has been removed. Thus, this result left open the possibility that the major form of AD-AMTS-2 secreted by MG-63 cells was the inactive precursor form. It should be noted that the ϳ132-kDa band recognized by antibody 1435 is unlikely to represent ADAMTS-3 or ADAMTS-14, as these proteins, although closely related to ADAMTS-2, are considerably diverged in sequence from AD-AMTS-2 in the region corresponding to the peptide against which antibody 1435 was raised (Fig. 3).
The Nature of Secreted ADAMTS-2-To obtain further insights into the nature of secreted ADAMTS-2, recombinant ADAMTS-2 with a C-terminal FLAG tag was produced in 293-EBNA human embryonic kidney cells that had been transfected with human ADAMTS-2 cDNA in an expression vector. As can be seen (Fig. 4B), anti-FLAG antibodies recognize a doublet of ϳ132and ϳ118-kDa bands in media of transfected, but not untransfected, cells. As reported for other ADAMTS proteins (30), these bands were observed only in media of transfected cells cultured in the presence of heparin, which is thought to compete for ADAMTS proteins bound to sulfated glycosaminoglycans in the cell layer. As can be seen in Fig. 4C, antibodies 1435 and 1374, directed against sequences in the ADAMTS-2 protease domain, and antibody 1442, directed against the C-terminal domain (see also Fig. 3), all recognize the 132-kDa band. In contrast, antibody 1429, directed against putative proregion sequences, does not, suggesting the 132-kDa form to represent ADAMTS-2 in which at least a portion of the proregion has been removed. Interestingly, antibodies 1374 and 1442 also recognize the 118-kDa band (Fig. 4C), whereas antibody 1435 does not. Because antibody 1435 recognizes sequences in the more N-terminal portion of the protease domain than does antibody 1374, this suggests that the 118-kDa band represents ADAMTS-2 in which some sequences from the Nterminal portion of the protease domain have been removed. Antibody 1435 recognizes an ϳ95-kDa band not recognized by anti-FLAG antibodies (e.g. compare Fig. 4, B and C) and which, therefore, is missing ADAMTS-2 C-terminal sequences. It is not clear whether this band is not recognized by antibodies 1374 and 1442 or whether it is not apparent on Western blots using these antibodies because they produce a lower signal than antibody 1435 and because the 95-kDa form is a relatively minor species. The minor 95-kDa form was not characterized further.
Importantly, when recombinant ADAMTS-2 and samples from media of MG-63 cells treated with TGF-␤1 are run in adjacent SDS-PAGE lanes and analyzed by Western blot with antibody 1435, the 132-kDa forms from the two samples comigrate exactly (Fig. 4A). Thus, the same 132-kDa form likely represents the predominant form of secreted ADAMTS-2 in both the recombinant system and in TGF-␤1-treated MG-63 cells.
To determine whether the ADAMTS-2 produced in the recombinant system was an active procollagen N-proteinase, pu- rified recombinant ADAMTS-2 was incubated with human type II procollagen. As can be seen (Fig. 5), the recombinant AD-AMTS-2 cleaves procollagen II to produce pC␣1(II) chains and together with the procollagen C-proteinase mTLL-1 cleaves procollagen II to produce mature ␣1(II) chains, with apparent absence of nonspecific cleavages. Thus, the ADAMTS-2 produced in the recombinant system is an active procollagen Nproteinase, although it was not clear from these data whether this activity might be furnished by the 132-, 118-, or 95-kDa form of the protein.
Human ADAMTS-2 has eight potential sites for Asn-linked glycosylation (Ref. 21 and Fig. 3). Thus, a significant proportion of the apparent molecular masses estimated for various AD-AMTS-2 forms by SDS-PAGE may be due to carbohydrate side chains. To determine the degree and nature of glycosylation of ADAMTS-2, the recombinant protein was treated with the enzymes peptide N-glycosidase F (PNGase F), endo-O-glycosidase, sialidase A, ␤-1-4-galactosidase, and glucosaminidase. Treatment with a mixture of all five enzymes increased the mobilities of the ϳ132and ϳ118-kDa forms to ϳ110 and ϳ90-kDa, respectively, upon SDS-PAGE (Fig. 6, lane 5). The 110-kDa form is thus similar in size to the 107,450 molecular weight predicted by the cDNA coding sequence for ADAMTS-2 proteolytically processed at the more C-terminal of two potential recognition sequences for cleavage by furin-like proprotein convertases (see Fig. 3).
To better elucidate the nature of ADAMTS-2 glycosylation, the recombinant protein was separately incubated with each of the various enzymes from the mixture. PNGase F, which removes virtually all N-linked oligosaccharides from glycoproteins, produced a marked shift in mobility for both the 132-and 118-kDa forms, consistent with glycosylation of at least some of the 8 potential sites (Fig. 6, lane 2). Sialidase A (sialidase from Arthrobacter ureafaciens) removes mono-, di-, and trisialyl residues, the most commonly occurring modifications, from Gal␤(1-3)GalNAc O-linked cores. Treatment of ADAMTS-2 with both PNGase F and sialidase A produced a further increase in electrophoretic mobility, compared with samples treated with PNGase F alone (Fig. 6, lanes 2 and 3), thus indicating the presence of sialylated O-linked oligosaccharides. Endo-O-glycosidase (endo-␣-N-acetylgalactosaminidase) removes Gal␤(1-3)GalNAc cores from serine and threonine residues, but only after modifying monosaccharides have first been removed by exoglycosidases. Treatment of ADAMTS-2 with PNGase F, sialidase A, and endo-O-glycosidase did not clearly produce an additional increase in electrophoretic mobility, compared with treatment with PNGase F and sialidase A alone (Fig. 6, lanes 3 and 4), suggesting the absence of appreciable quantities of simple sialylated O-linked Gal␤(1-3)GalNAc Olinked cores. In contrast, ␤-1-4-galactosidase and glucosaminidase, which remove less common modifying ␤-1,4-linked galactose and ␤-1-6-linked N-acetylglucosamine residues, respectively, from Gal␤(1-3)GalNAc O-linked cores, were found to enable endo-O-glycosidase to induce a detectable mobility shift in ADAMTS-2 (Fig. 6, lane 5). Thus, ADAMTS-2 appears to be decorated with O-linked Gal␤(1-3)GalNAc cores, which are in turn decorated with modifying ␤-1,4-linked galactose and ␤-1-6-linked N-acetylglucosamine residues.
As noted above, the ϳ110-kDa form of recombinant AD-AMTS-2 obtained from deglycosylation using all five glycosidases is similar in size to the ϳ107 molecular weight predicted by the sequence for ADAMTS-2 processed at the more C-terminal of two potential furin cleavage sites. To determine whether the 110-kDa form might indeed result from cleavage at such a site, we incubated transfected 293 cells in the presence of the highly specific furin inhibitor decanoyl-RVKR-chloromethyl ketone. As can be seen (Fig. 7A), in the presence of furin inhibitor, cells secrete into medium a ϳ150-kDa form as the predominant ADAMTS-2 species, in addition to the 132and 118-kDa forms.
To characterize further ADAMTS-2 biosynthesis, we compared ADAMTS-2 forms detected in medium with those detected in cell layer and cell layer-associated ECM, using both anti-FLAG antibody and antibody 1435. Both antibodies detected the 132-and 118-kDa forms in medium, cell layers, and ECM (Fig. 7B). Interestingly, although neither antibody de- FIG. 3. Structural features of ADAMTS-2. Boxes in the schematic of ADAMTS-2 represent the signal peptide (S); the pro-(Pro); protease (Protease); disintegrin-like (Dis); 1st, 2nd, 3rd, and 4th thrombospondin type-1 (Tsp1, 2, 3, and 4); cysteine-rich (Cys); spacer (Spacer); and C-terminal domains (C-term). Placement of the signal peptide cleavage site, between proline residues 29 and 30, was determined using the method of Nielsen et al. (55). The dashed region within the protease domain represents the Zn 2ϩ -binding active site (Zn 2ϩ ). The dashed region within the C-terminal domain indicates the site of the relatively conserved PLAC region. Arrows indicate sites of consensus sequences for potential cleavage by furin-like proprotein convertases (56). Closed circles indicate sites of consensus sequences for potential Asn-linked glycosylation. Vertical lines indicate sites corresponding to the N termini of ϳ132 and ϳ118-kDa forms of ADAMTS-2, as experimentally determined by automated Edman degradation. The cleavage site for production of the ϳ132-kDa form was at the more C-terminal of the two furin sites (RARR 259 / 260 HAAD), whereas the cleavage sites for the two ϳ118-kDa forms were at GFSS 402 / 403 AFVV and SYDC 465 / 466 LLDD, respectively. Open circles above the schematic indicate the positions of amino acid residues corresponding to peptides used in the production of antibodies 1429, 1435, 1374, and 1442. Below the schematic the ADAMTS-2 (TS-2) amino acid sequences are presented for the peptides used to raise the various antibodies. Alignments show amino acid differences between the ADAMTS-2 peptides and equivalent regions of ADAMTS-3 (TS-3) and ADAMTS-14 (TS-14).
tected the 150-kDa form in medium, cell layer, or ECM, in the absence of furin inhibitor, both antibodies detected a cell layerassociated 141-kDa form of ADAMTS-2, not found in either ECM or medium (Fig. 7B). Upon treatment with decanoyl-RVKR-chloromethyl ketone, medium, cell layer, and ECM also contained the 150-kDa form (Fig. 7C). Antibody 1429, which is directed against putative proregion sequences (Fig. 3) and which does not recognize the 132-and 118-kDa secreted forms (Figs. 4C and 7C), recognizes the 150-kDa form in medium, cell layer, and ECM of cultures incubated with the furin inhibitor (Fig. 7C). This is consistent with the identity of the 150-kDa species as a precursor form of ADAMTS-2. In addition, antibody 1429 recognizes the 141-kDa form of ADAMTS-2 found in cell layers of cultures either treated or untreated with furin inhibitor (Fig. 7C). As antibody 1429 is directed against the prodomain sequences located between the two potential furin cleavage sites (Fig. 3), these results suggest the 141-kDa form to represent a processing intermediate in which cleavage has occurred at a site N-terminal to the more C-terminal furin site. The presence of the 141-kDa form in cell layers in the absence of furin inhibitor and absence of the 150-kDa form suggest that cleavage at the more N-terminal site is more rapid than that occurring at the more C-terminal site, thus allowing accumulation of the 141-kDa processing intermediate. Persistence of the 141-and 132-kDa forms in the presence of furin inhibitor probably reflects incomplete inhibition of cleavage at the two furin consensus sites, whereas persistence of the 118-kDa form probably reflects cleavage by non-furin-like proteases (see below). The presence of the 141-kDa form in cell layers, but not in ECM or media, indicates that it is likely an exclusively intracellular and/or cell-associated form.
Sufficient 132-and 118-kDa ADAMTS-2 was isolated to enable automated Edman degradation and determination of the N-terminal amino acid sequences of each. As expected from the similarity in size between the 110-kDa deglycosylated version of the 132-kDa form and the molecular weight of 107 predicted by the sequence for ADAMTS-2 processed at the more C-terminal furin site, the N-terminal sequence of the 132-kDa form was HAADDDYNIE, showing cleavage occurred between residues Arg 259 and His 260 of the published sequence, at the more C-terminal consensus furin recognition sequence 256 RARR 259 (Fig. 3). N-terminal sequences for the 118-kDa band showed it to be heterogeneous and to represent two protein species, one of which had been processed between residues GFSS 402 and 403 AFVV and the other of which had been processed between residues SYDC 465 and 466 LLDD. Both of these cleavages occurred within the protease domain (Fig. 3), one immediately N-terminal to and the other C-terminal to the Zn 2ϩ -binding active site. Both of these cleavages occur C-terminal to the sequences recognized by antibody 1435 (Fig. 3), explaining the inability of this antibody to detect the 118-kDa form.
ADAMTS-2 Has Both pNPI and pNPIII Activities-In the course of characterizing recombinant ADAMTS-2, it was tested for pNP activity against procollagens I-III. Surprisingly, AD-AMTS-2 was found to have pNP activity not only against procollagens I and II (pNPI activity) but against procollagen III (pNPIII) as well (Fig. 8). Moreover, incubating varying amounts of ADAMTS-2 with fixed amounts of the three procollagen types showed ADAMTS-2 to have similar levels of pNPI and pNPIII activity.
If ADAMTS-2 is truly a pNPIII, then TGF-␤ induction of ADAMTS-2 in MG-63 cells should be accompanied by induction of pNPIII activity. This appears to be the case, as evidenced by appearance of pC␣1(III) chains in media of MG-63 cells treated, but not in media of those untreated, with TGF-␤1 (Fig. 9A). As a final test of the identity of ADAMTS-2 as a pNPIII, processing of procollagen III was compared in control dermal fibroblasts and those derived from an individual with EDSVIIC. It has been shown that the effects of dermatosparaxis/EDSVIIC on type I procollagen processing are best observed in cell layers of fibroblasts cultured in the presence of neutral polymers (25,45). Addition of the polymers, which appear to act via a volume exclusion mechanism, potentiates normal processing of procollagen to collagen and association of processed forms with cell layers, such that defects in processing are accentuated when mutant and control fibroblast cultures are compared (25,45). As can be seen (Fig. 9B, left panel), whereas pro-␣1(I) chains are almost entirely converted to mature ␣1(I) chains in the control fibroblast cell layer, conversion to ␣1(I) chains is impaired in the EDSVIIC culture, with accumulation of large amounts of incompletely processed pN␣1(I) chains. Such results reflect loss of pNPI activity normally provided by AD-AMTS-2, whereas residual pNPI activity is thought to be provided by another ADAMTS family member (46). When the same samples were analyzed on a separate Western blot using antibodies that recognize pro-␣1(III) and pN␣1(III), but not pC␣1(III) or mature ␣1(III) chains, processing of procollagen III was seen to be impaired in the EDSVIIC culture, with accumulation of pN␣1(III) chains in the EDSVIIC cell layer, and no evidence of such chains in the cell layer of the normal control. This result strongly supports an in vivo role for AD-AMTS-2 as a pNPIII. DISCUSSION TGF-␤1 is known to induce net increases in formation of ECM via a combination of effects. These include inhibiting production of degradative matrix metalloproteinases, while inducing production of (i) inhibitors of such proteinases, (ii) structural matrix components such as procollagens I-III, and (iii) lysyl oxidase, an extracellular enzyme necessary to formation of covalent cross-links in collagen and elastic fibers (44,47). Because of the roles of TGF-␤1 in formation of ECM in general and, in particular, because of its role in inducing production of procollagens I-III and lysyl oxidase, all known substrates of the procollagen C-proteinases BMP-1 and mTLD (8,48), we previously examined the ability of TGF-␤1 to regulate expression of BMP-1 and mTLD as well (39). In that study (39), we observed TGF-␤1-induced increases in (i) levels of BMP-1 and mTLD mRNA, (ii) secretion of active forms of BMP-1 and mTLD, and (iii) pCP activity in cultures of fibrogenic cells, including MG-63 cells (39). In addition, we noted the apparent induction of pNPI activity in TGF-␤1-treated MG-63 cultures, evidenced by increased conversion of endogenous pro-␣1(I) chains to pN␣1(I) and mature ␣1(I) forms (39). In the present study we have further explored the latter observation and have shown, by both Northern blot analysis and highly specific RNase protection assay, that TGF-␤1 induces, in a dose-dependent way, an ϳ8-fold increase in ADAMTS-2 mRNA levels in MG-63 cells. This induction appears to be indirect, as it is inhibited by cycloheximide, and to occur at the level of transcription, as ADAMTS-2 mRNA stability is unchanged by the presence or absence of TGF-␤1 in culture media.
In addition to demonstrating induction of ADAMTS-2 mRNA by TGF-␤1, data presented herein also indicate that TGF-␤1 induces secretion of ADAMTS-2 protein. Moreover, this protein is identical in size to an ϳ132-kDa form of recombinant AD-AMTS-2 that appears to represent the mature active form of the protease, as the prodomain has been removed via cleavage by a furin-like protease at a consensus recognition site. Thus, in addition to increasing formation of ECM by inducing synthesis of matrix components, matrix-metalloproteinase inhibitors (44), lysyl oxidase (47), and pCPs such as BMP-1 and mTLD (39), TGF-␤1 also appears to affect ECM formation by inducing production and secretion of the active form of ADAMTS-2.
ADAMTS-3 and -14 have greater homology to ADAMTS-2 than do any other ADAMTS family members, and the three proteases have been suggested to constitute a subfamily (46,49). In addition, both ADAMTS-3 and -14 have been shown recently (46,49) to have pNPI activity in vitro and have been suggested as possible sources for the residual pNPI activity observable in bone, tendon, cartilage, skin, and other tissues of EDSVIIC patients, dermatosparaxic cattle, and Adamts2Ϫ/Ϫ mice. Nevertheless, correlation in the present study between the level of induction of the ϳ132-kDa band recognized by polyclonal antibody 1435 and the level of induction of AD-AMTS-2 mRNA, the fact that the 132-kDa band is identical in size to mature ADAMTS-2, and the fact that ADAMTS-3 and -14 are considerably diverged in sequence from ADAMTS-2 in the region recognized by antibody 1435, all make it highly likely that the 132-kDa band induced by TGF-␤1 corresponds to ADAMTS-2.
Previously, Colige et al. (46) have reported that levels of ADAMTS-14 RNA in human dermal fibroblasts are unaffected by TGF-␤, and they cite unpublished data as showing the same to be true for ADAMTS-2 and -3 RNA. It is not stated which type of TGF-␤ was employed in that study, although if TGF-␤1 was used, then reasons for the different results obtained for ADAMTS-2 RNA in that study and in the present study are unclear. However, as dermal fibroblasts were used in the previous study and MG-63 cells are used here, these differences may suggest that TGF-␤1 is capable of affecting ADAMTS-2 expression in some cell types but not in others.
A surprise in the present study was the finding that AD-AMTS-2 has pNPIII activity, and that levels of pNPIII and pNPI activity appear similar in this single protease. This finding is contrary to accepted dogma based on various biochemical studies that have reported previously (14 -19) that the protease responsible for pNPI activity is unable to process procollagen III, whereas the protease responsible for pNPIII activity is unable to process procollagen I. In addition, processing of type III procollagen N-propeptides appears to occur more slowly than processing of type I procollagen N-propeptides in tissues and tissue culture systems (50), which has also suggested the operation of two different N-proteinases. Reasons for the difference between previous biochemical findings and those of the current study are unclear but may be related to use here of relatively large amounts of highly purified ADAMTS-2. In contrast, most earlier studies (16 -19) have employed tissue or cell culture isolates of relatively low purity, and such preparations may have contained contaminants (e.g. endogenous inhibitors and/or other proteases) that affected proteolytic activity. Other studies (14,15) that have employed more highly purified preparations of pNPI from tissue or organ culture have nevertheless found this activity to be associated with large 500-kDa complexes containing several polypeptides, some of which might similarly have affected activity. Additional polypeptides, and/or differences in the accessibility of ADAMTS-2 to procollagens I and III, may also result in the different kinetics with which N-propeptides are processed from the two procollagen types in tissues and tissue culture systems.
Importantly, it should be noted that one of the earlier studies (19) cited above described a marked decrease in processing of the procollagen III N-propeptide in skin of dermatosparactic cattle. Thus, as dermatosparaxis is the bovine equivalent of EDSVIIC, this previous report provides genetic evidence that supports the evidence presented here that ADAMTS-2 is both a pNPI and a pNPIII, and that ADAMTS-2 plays an in vivo role as a pNPIII, such that defects in the ADAMTS2 gene lead to a defect in processing of procollagen III in EDSVIIC skin fibroblasts. These results suggest that the phenotypes of dermatosparaxis and EDSVIIC may arise from defects in the processing of both types I and III procollagen N-propeptides. As AD-AMTS-3 and ADAMTS-14 have been postulated to provide residual pNPI activity found in dermatosparactic tissues (46,49), one or both of these two proteinases may be capable of pNPIII activity, perhaps explaining the residual low level processing of procollagen III N-propeptides previously noted in dermatosparactic calf skin (19). It should be noted that pNPIII activity is not intrinsic to all ADAMTS family members, as ADAMTS-1, -4, and -5 were negative for pNPIII activity in our in vitro assays (data not shown).
Interestingly, Apte and colleagues (49) have recently shown ADAMTS-2 mRNA to be at dramatically higher levels in 7 days post-conception murine gastrulas than at later gestational times. This is despite the fact that mRNAs of the major fibrillar collagens are at relatively low levels at 7 days post-conception and are found at abundant levels only markedly later in development 4 (51). This difference in temporal patterns of expression suggests that, in addition to its expanded role as both pNPI and pNPIII, ADAMTS-2 may also be involved in proteolytic processing of substrates other than the major fibrillar collagens. In this regard, it is of interest that the procollagen C-proteinases, each of which processes the C-propeptides of procollagens I-III, have been found to process a wide range of additional substrates involved in formation of the ECM and in other morphogenetic processes, such as BMP signaling (7,44,48,(52)(53)(54). The possibility of such multiple roles for AD-AMTS-2, and perhaps for other ADAMTS family members as well, presents an interesting avenue for future research.