Characterization of ADAMTS-9 and ADAMTS-20 as a distinct ADAMTS subfamily related to Caenorhabditis elegans GON-1.

We demonstrate that in humans, two metalloproteases, ADAMTS-9 (1935 amino acids) and ADAMTS-20 (1911 amino acids) are orthologs of GON-1, an ADAMTS protease required for gonadal morphogenesis in Caenorhabditis elegans. ADAMTS-9 and ADAMTS-20 have an identical modular structure, are distinct in possessing 15 TSRs and a unique C-terminal domain, and have a similar gene structure, suggesting that they comprise a new subfamily of human ADAMTS proteases. ADAMTS20 is very sparingly expressed, although it is detectable in epithelial cells of the breast and lung. However, ADAMTS9 is highly expressed in embryonic and adult tissues, and therefore we characterized the ADAMTS-9 protein further. Although the ADAMTS-9 zymogen has many proprotein convertase processing sites, pulse-chase analysis, site-directed mutagenesis, and amino acid sequencing demonstrated that maturation to the active form occurs by selective proprotein convertase (e.g. furin) cleavage of the Arg(287)-Phe(288) bond. Although lacking a transmembrane sequence, ADAMTS-9 is retained near the cell surface as well as in the ECM of transiently transfected COS-1 and 293 cells. COS-1 cells transfected with ADAMTS9 (but not vector-transfected cells) proteolytically cleaved bovine versican and aggrecan core proteins at the Glu(441)-Ala(442) bond of versican V1 and the Glu(1771)-Ala(1772) bond of aggrecan, respectively. In contrast, the ADAMTS-9 catalytic domain alone was neither localized to the cell surface nor able to confer these proteolytic activities on cells, demonstrating that the ancillary domains of ADAMTS-9, including the TSRs, are required both for specific extracellular localization and for its versicanase and aggrecanase activities.

The ADAMTS (A disintegrin-like and metalloprotease (reprolysin type) with thrombospondin type I motif) family consists of secreted zinc metalloproteases with a precisely ordered modular organization that includes at least one thrombospondin type I repeat (TSR) 1 (1,2). Important functions have been established for several members of the family. AD-AMTS-4, ADAMTS-5, and (less efficiently) ADAMTS-1 degrade the cartilage proteoglycan aggrecan and are referred to as aggrecanases (3)(4)(5). They play a major role in aggrecan loss in arthritis (6,7). ADAMTS-1 and ADAMTS-4 participate in the turnover of the aggrecan-related proteoglycans versican and brevican in blood vessels (8) and the nervous system, respectively (9). ADAMTS2 mutations cause dermatosparaxis, a recessively inherited disorder characterized by severe skin fragility that results from incomplete proteolytic removal of the procollagen I amino propeptide (N-propeptide) (10). AD-AMTS-3 and ADAMTS-14 are procollagen N-propeptidases with probable roles in procollagen II processing in cartilage or procollagen I processing in tissues other than skin, respectively (11,12). ADAMTS13 mutations lead to inherited thrombocytopenic purpura, a coagulation disorder caused by deficient proteolytic processing of von Willebrand factor (13). Adamts1-null mice have abnormal adipogenesis, defective angiogenesis in the adrenal gland, and a defect of ureteric ECM turnover, leading to hydronephrosis (14). Adamts2-null mice have fragile skin, and males are infertile (15). Many other ADAMTS enzymes have been discovered through molecular cloning, and their functions are presently unknown. Altogether, 19 human AD-AMTS symbols identifying 18 distinct genes and their products have been assigned (note that ADAMTS5 (1) and ADAMTS11 (4) designate the same gene). 2 ADAMTS are also present in invertebrates, which contain fewer ADAMTS genes than mammalian genomes. A Caenorhabditis elegans ADAMTS gene, gon-1, has an essential role in reproduction (16). The protease (GON-1) encoded by gon-1 is required for migration of distal tip cells during gonadal morphogenesis. It may have a role in degradation of basement membrane or for processing of extracellular cues required for cell migration (16). GON-1 is the largest of all ADAMTS en-zymes described to date and contains 18 TSRs (16). In addition, it has a presumed globular domain at the C terminus without similarity to known proteins.
Human ADAMTS-9, as previously described (17) contains four TSRs. Despite being a much smaller enzyme than GON-1, it had greater sequence similarity to it than to any other human ADAMTS (17). Here, we characterize a considerably longer form of ADAMTS-9 (designated ADAMTS-9B, but referred to subsequently in this paper as ADAMTS-9) that we propose is the authentic full-length product of ADAMTS9. In addition, we have discovered a novel enzyme, ADAMTS-20, and determined its complete primary sequence. ADAMTS-9 and ADAMTS-20 have an identical domain organization and exon structure and a very similar primary sequence, showing that they comprise a distinct subfamily of GON-1-related ADAMTS proteases in the mammalian genome. We have characterized the zymogen maturation and cellular localization of the more highly expressed of these two proteins, ADAMTS-9, and have investigated its role in proteolysis of the large aggregating proteoglycans versican and aggrecan. Our data demonstrate the critical requirement of the ancillary domains for the proteolytic function and localization of ADAMTS-9.

EXPERIMENTAL PROCEDURES
cDNA Cloning and Sequence Analysis of ADAMTS9 and AD-AMTS20 -BLAST (Basic Local Alignment Search Tool) programs from the National Center for Biotechnology Information were used to search the data base of expressed sequence tags (dBEST), using the protein sequences of ADAMTS proteases previously discovered by us (1,18). To extend the initially identified ADAMTS9 cDNA to the 5Ј-end, human chondrocyte, muscle, heart, or fetal brain mRNA (Marathon cDNA, Clontech, Palo Alto, CA) was used as the template for rapid amplification of cDNA ends as previously described (1). To confirm that the overlapping cDNA clones obtained represented a contiguous mRNA, the complete ORF was amplified by PCR. The oligonucleotide primers 5Ј-AAGCGGCCGCACCATGCAGTTTGTATCC-3Ј (NotI site underlined and start codon italicized) and 5Ј-CTCGAGAATAAAACTCGCACCTC-CAGGC-3Ј (XhoI site underlined and modified stop codon italicized) were used for PCR with human fetal skeletal muscle cDNA as template and Advantage 2 polymerase (Clontech, Palo Alto, CA). The 5.8-kb PCR product was cloned into pGEM-T Easy (Promega, Madison, WI) and sequenced completely. cDNA cloning of Adamts9 will be reported elsewhere. 3 To ask whether there existed additional ADAMTS proteases with a domain organization similar to GON-1 and ADAMTS-9, the human genome sequence (Celera, Rockville, MD) was searched using the amino acid sequence of the unique C-terminal domain of ADAMTS-9. GEN-SCAN (available on the World Wide Web at genes.mit.edu/GEN-SCAN.html) analysis of genomic DNA upstream and downstream of the initially identified ADAMTS20 sequence was used to identify putative ADAMTS20 exons. Oligonucleotide primers based on the sequences of these putative ADAMTS20 exons were used for PCR spanning adjacent exons using cDNA derived from the human K562 (erythroleukemia) and A549 (lung cancer) cell lines.
The exon-intron structures of ADAMTS9 and ADAMTS20 were deduced by comparison of the respective cDNAs with human genome sequences using BLAST searches of private (Celera) and public (Gen-Bank TM ) databases.
Northern Blot and Quantitative RT-PCR of ADAMTS9 and AD-AMTS20 and ADAMTS20 RNA in Situ Hybridization Analysis-Multiple tissue northern blots containing 1 g/lane poly(A ϩ ) RNA from mouse embryos and individual adult mouse and human tissues (Clontech, Palo Alto, CA) were hybridized to [␣-32 P]dCTP-labeled ADAMTS9, ADAMTS20, or Adamts9 probes, followed by autoradiographic exposure for 3-7 days. cDNA panels derived from human adult and fetal organs normalized with respect to GAPDH mRNA levels were purchased from Clontech. Real time PCR of these cDNA templates was performed in an ABI Prism 7700 sequence detector using SYBR Green PCR Core Reagents (Applied Biosystems, Foster City, CA), as previously described (12). PCR amplifications were performed in triplicate for all templates, along with parallel measurements of GAPDH cDNA for normalization.
The GAPDH-normalized quantitative data for ADAMTS9 and AD-AMTS20 were used to determine the ADAMTS9/ADAMTS20 transcript ratio in all templates examined. The following primers were used for amplification at a concentration of 300 nM each: ADAMTS9 forward, 5Ј-GGACAAGCGAAGGACATCC-3Ј; ADAMTS9 reverse, 5Ј-ATCCATC-CATAATGGCTTCC-3Ј; ADAMTS20 forward, 5-GGTGGCATGTTATT-GGCAAAA-3Ј; ADAMTS20 reverse, 5Ј-CACAGTTACCATGGCATAGT-TCTTG-3Ј; GAPDH primers were described previously (12). RT-PCR performed in the absence of template was negative with all primer pairs. RNA in situ hybridization was performed essentially as previously described (19), using 35 S-labeled antisense and sense cRNA probes transcribed from a 600-nt cDNA template encoding the unique domain of ADAMTS-20. Normal human breast and lung samples as well as samples of squamous cell carcinoma of breast and adenocarcinoma of lung were obtained under a Cleveland Clinic Foundation Institutional Review Board-approved protocol and fixed in formalin (tissue samples were provided by the Cooperative Human Tissue Network). 5-m-thick paraffin sections were hybridized to the probes prior to dipping in photographic emulsion (Eastman Kodak Co.) and followed by autoradiographic exposure for 7 days. Nuclei were stained with 4Ј,6-diamidino-2-phenylindole.
The insert of the KIAA0688 gene (20) encoding ADAMTS-4 (3) in pBluescript SK (Stratagene) was excised with EcoRI and XhoI and inserted into the corresponding sites of pcDNA3.1MYC/HIS A-(Invitrogen) to generate a mammalian expression vector producing untagged ADAMTS-4. The ADAMTS4 and ADAMTS5 ORFs from the convertaseprocessing site to the stop codon were PCR-amplified and cloned into p3XFLAG-CMV-9 (Sigma) for expression in frame with a preprotrypsin leader sequence and three tandem FLAG tags present just downstream of the signal peptidase cleavage site. These proteins are therefore secreted with N-terminal FLAG tag ( 3ϫFLAG ADAMTS-4 or 3ϫFLAG ADA-MTS-5). All expression plasmids and site-directed mutations were verified by DNA sequencing.
To release ADAMTS-9 from the cell surface, transfected 293 cells and ECM were harvested by scraping and resuspended in phosphate-buffered saline (10 mM phosphate buffer, pH 7.4, 2.7 mM KCl, 137 mM NaCl). Cells and ECM were gently agitated by end-over end rotation in PBS alone or in PBS plus 100 mM or 200 mM NaCl at 4°C for 30 min.
ADAMTS-9 1-508MYC/HIS Purification and Analysis-To obtain stably transfected 293 cells expressing ADAMTS-9 1-508MYC/HIS , selection with G418 (750 g/ml) was applied after transfection, and selected clones were maintained in culture medium containing 5% serum and 250 g/ml G418. Conditioned medium was dialyzed into binding buffer (20 mM sodium phosphate, 500 mM NaCl, pH 7.8, containing 0.03% Brij-35 (Sigma)) prior to binding on a 5-ml Ni 2ϩ -Sepharose column (ProBond; 3 K. A. Jungers and S. S. Apte, unpublished data. Invitrogen). The column was washed with 3 column volumes of binding buffer. A gradient of 0 -42.5 mM imidazole in binding buffer was used to remove nonspecifically bound molecules from the column. Stepwise elution was done using one-column volume batches of 0 -250 mM imidazole in binding buffer. Elution was monitored by Western blotting using antibody 9E10. The majority of protein was determined to elute at 50 mM imidazole. ADAMTS-9 1-508MYC/HIS was electrophoresed on 10% SDS-PAGE, electrotransferred to polyvinylidene difluoride membrane, and lightly stained with modified Coomassie Blue (Simply Blue Safe Stain; Invitrogen). The 28-kDa band was excised and subjected to Edman degradation on an Applied Biosystems Procise 492 sequencer in the Molecular Biotechnology Core Facility of the Lerner Research Institute.
Deglycosylation of lysate and conditioned medium from ADAMTS-9 1-508FLAG -transfected cells was done using 10 units of PNGase F (Roche Molecular Biosciences) for 3 h at 37°C in 150 mM sodium phosphate, pH 7.4, 50 mM EDTA, 0.1% SDS, 1% 2-mercaptoethanol, 0.5% Triton X-100, followed by immunoprecipitation with anti-FLAG M2 as described below. Stably transfected cells were cultured with and without tunicamycin A homolog (Sigma) as previously described (21), followed by Western blotting of conditioned medium and cell lysates.
Biosynthesis of ADAMTS-9 -Unless specified, reagents were purchased from Sigma. QBI 293A cells (Quantum Biotechnologies, Montreal, Canada) were maintained in complete DMEM containing 10% heat-inactivated fetal bovine serum, 2 mM L-glutamine, 50 units/ml penicillin, and 50 g/ml streptomycin. Cells were transiently transfected with ADAMTS-9 1-508FLAG or ADAMTS-9 1-508MYC/HIS using Fugene 6 (Roche Molecular Biosciences). 24 h following transfection, cells were washed twice with warm phosphate-buffered saline and incubated in Met/Cys-free medium (MEM SelectAmine kit; Invitrogen), supplemented with 10% dialyzed FCS, 1 mM glutamine, and a [ 35 S]methionine/cysteine mixture (EXPRE 35 S 35 S; PerkinElmer Life Sciences). A 15-or 30-min labeling (pulse) was followed by incubation in complete nonradioactive medium (chase) for the indicated times. The cell layer was washed with PBS, and cells were lysed with 1 ml of radioimmune precipitation buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 1% Nonidet P-40, 0.5% deoxycholic acid, and 4 mM EDTA) containing protease inhibitors (1 M aprotinin, 10 M pepstatin, 10 M leupeptin, and 1 M phenylmethylsulfonyl fluoride). Samples were centrifuged to remove insoluble material. FLAG M2 antibody or penta-His antibody (Qiagen, Mississauga, Canada) was added to cell lysate and medium followed by overnight incubation at 4°C. Protein A/G Plus-agarose beads were added and incubated with samples for 1 h at 4°C. Beads were washed three times with 1 ml of radioimmune precipitation buffer, and labeled proteins were resolved by reducing SDS-PAGE. Gels were treated with ENTENSIFY reagent (PerkinElmer Life Sciences), dried, and exposed for fluorography.
Analysis Of Proprotein Convertase Processing-CHO.RPE 40 cells (22) were maintained as previously described (32). They were transfected with ADAMTS-9 1-508MYC/HIS alone or in combination with furin. QBI 293A cells were transiently transfected either with ADAMTS-9 1-508MYC/HIS , or its derivatives obtained by site-directed mutagenesis. Cells were metabolically labeled as above for 3 h, and immunoprecipitation and fluorography were done as above.
Proteolytic Processing of the Versican and Aggrecan Core Proteins-Versican monomer (a mixture of the V1 and V0 forms) was isolated from bovine aorta, as previously described (8). COS-1 cells were transfected with ADAMTS-9 FLAG , ADAMTS-9 1-508FLAG , or ADAMTS-4 expression plasmids or with empty vector pFLAG-CMV-5a (as negative control) in six-well plates in DMEM plus 10% fetal bovine serum. Cells from a single well were used for each experiment. 48 h after transfection, cells were scraped off and suspended in serum-free DMEM followed by six washes in fresh serum-free DMEM. Cells were resuspended in a final volume of 75 l of serum-free DMEM. Versican (5 g in a volume of 25 l) was added, and the reaction was incubated at 37°C for 18 h. The reaction was centrifuged briefly, and the cell pellets were retained for Western blotting of ADAMTS-9 FLAG using the anti-FLAG M2 monoclonal antibody (Sigma). An equal volume of 2ϫ glycosaminoglycan digestion buffer (200 mM Tris, pH 6.5, 100 mM sodium acetate) containing 0.5 units of chondroitinase ABC (Seikagaku) was added to the supernatant, followed by incubation for 16 -18 h at 37°C. Protein was precipitated with 5 volumes of acetone at Ϫ20°C for 15 min, dissolved in 50 l of Laemmli sample buffer, and electrophoresed on 7% SDS-PAGE prior to blotting to nitrocellulose. A rabbit polyclonal antiserum to the versican Asp-Pro-Glu-Ala-Ala-Glu (DPEAAE) neoepitope (8) (provided by Dr. John Sandy) was used at 1:1000 dilution for Western blotting, followed by enhanced chemiluminescence detection of antibody binding. Anti-DPEAAE recognizes the new C terminus resulting from cleavage of versican at the Glu 441 -Ala 442 bond (this enumeration describes the site in the V1 isoform); the corresponding peptide bond is Glu 1428 -Ala 1429 in the V0 isoform, since this form contains an additional GAG-bearing region, GAG-␣, as a result of alternative splicing (8).
Aggrecan monomer was isolated from bovine articular cartilage as previously described (23). Aggrecan (20 g) was incubated with transfected cells as described above. Neoepitope Western blot analysis was as performed for versican (above), except that the proteolytic cleavage at the Glu 1771 -Ala 1772 bond of aggrecan was detected using anti-Ala 1772 -Gly-Glu-Gly (AGEG) antiserum (24) (provided by Micky Tortorella).

RESULTS
Cloning of ADAMTS9 and ADAMTS20 cDNAs-Our search for novel ADAMTS proteases identified a human expressed sequence tag (GenBank TM accession number AA205581 encoded by IMAGE clone 646675) from neuroepithelium-derived NT2 cells treated with retinoic acid. The ORF of this expressed sequence tag was homologous to ADAMTS proteases and encoded four TSRs followed by a C-terminal domain containing 10 cysteines that was similar to the C terminus of a polypeptide predicted by the C. elegans F25H8.3 cosmid (C. elegans protein data base Wormpep, www.sanger.ac.uk/Projects/C_elegans/ wormpep) and subsequently identified as GON-1. The novel human ORF was designated ADAMTS-9. Completion of the full-length protein coding sequence to the putative start codon required several rounds of rapid amplification of cDNA ends. Together, the cloned cDNA sequences represent an mRNA of 8 kb (Fig. 1a). The 3Ј-untranslated region in IMAGE clone 646675 contained a consensus polyadenylation signal (AATAAA) 15 nucleotides upstream of the poly(A) tail. The most 5Ј clone obtained (TS9-B10) contained 32 bp of the 5Јuntranslated region. The putative signal peptide coding sequence was preceded by a methionine codon within a satisfactory Kozak consensus sequence (A at Ϫ3 relative to ATG), but there was no upstream, in-frame stop codon.
The search for ADAMTS-9-related proteins led to identification of a polypeptide (Celera hCP1629711) predicted by exons on human chromosome 12. The complete 5733-nt-long AD-AMTS-20 ORF was assembled from overlapping cDNA clones (Fig. 1a). The ADAMTS20 mRNA was found in low quantities, routinely requiring 35 cycles of PCR or nested PCR for visualization of the PCR products on a gel. Because of the rarity of ADAMTS20 transcripts as well as the presence of numerous regions that are difficult to PCR-amplify, we have been so far unable to obtain the complete ORF in a single PCR reaction.
Identical Domain Organization and Similar Primary Structure of ADAMTS-9 and ADAMTS-20 -ADAMTS-9 and AD-AMTS-20 are similar in length, containing 1935 and 1911 amino acids, respectively (Figs. 1b and 2). Each contains a C-terminal array of 14 TSRs (15 TSRs/enzyme) that is interrupted by short "linker" peptides located between TSR-6 and -7 and TSR-8 and -9 that do not have similar sequences. AD-AMTS-9 and ADAMTS-20 are very similar to each other, with 48% identity and 64% similarity. The cysteine signatures of individual modules in ADAMTS-9 and ADAMTS-20 are identical to those of most other ADAMTS enzymes, with the exception of the procollagen aminopropeptidases (ADAMTS-2, ADAMTS-3, AD-AMTS-14) and ADAMTS-13, which have distinctive prodomains and catalytic domains (12). Each module in ADAMTS-9 and ADAMTS-20 (with one exception, described below) contains an even number of cysteines, suggesting participation in internal disulfide bonds. There are 126 cysteines in mature ADAMTS-9, predicting 63 intrachain disulfide bonds. ADAMTS-20 has a Cys to Tyr substitution in TSR-13 (Fig. 2). Since the substituted Cys is the fourth of six conserved cysteines in TSRs, TSR-13 in AD-AMTS-20 may contain two intrachain disulfide bonds instead of three and have an unattached cysteine.
The predicted molecular mass of the full-length enzymes is 216 kDa (ADAMTS-9) and 214 kDa (ADAMTS-20). The mass will decrease by ϳ2-3 kDa following signal peptide processing. In addition, both enzymes have a prodomain that is likely to be proteolytically processed prior to or during secretion. AD-AMTS-9 contains five consensus furin cleavage sites in its prodomain, whereas ADAMTS-20 contains three (Figs. 1b and 2). Two sites, those corresponding to Arg 74 and Arg 287 in AD-AMTS-9, are conserved with ADAMTS-20. Following processing at the furin recognition sequence closest to the C terminus, mature ADAMTS-9 is predicted to have a molecular mass of 184,000 and mature ADAMTS-20 a molecular mass of 185,000. Both enzymes contain consensus sites for N-linked glycosylation (Asn-X-Ser/Thr, where X is any amino acid except Pro), 9 in ADAMTS-9 and 15 in ADAMTS-20 (Fig. 1b). Five such sites, including three in the unique C-terminal domain, are conserved in ADAMTS-9 and ADAMTS-20. Because of the high likelihood of utilization of these sites, the molecular mass of ADAMTS-9 and ADAMTS-20 will probably be in excess of that predicted (i.e. Ͼ185,000). Although there are a number of Ser-Gly or Gly-Ser motifs in both ADAMTS-9 and ADAMTS-20, most are within presumed disulfide-bonded domains and lack the expected sequence context for xylosyltransferase recognition (25,26). However, one motif in the middle of the AD-AMTS-9 spacer domain with the sequence Glu-Tyr-Ser 830 -Gly-Ser 832 -Glu-Thr-Ala-Val-Glu lies within a sequence context that is compatible with GAG attachment to Ser 830 or Ser 832 . A similar sequence is present at this location in ADAMTS-20 (Fig. 2). ADAMTS-9 and ADAMTS-20, respectively, contain three and two Cys-Ser-Val-Thr-Cys-Gly (CSVTCG) motifs that are believed to mediate binding to the cell surface molecule CD36 (27, 28) (Fig. 2). In addition, each enzyme contains two BBXB motifs (where B represents basic amino acid and X represents any amino acid) that have been shown to mediate heparin and sulfatide binding (27, 29) (Fig. 2). Neither enzyme contains an Arg-Gly-Asp motif.

ADAMTS-9 and ADAMTS-20 Do Not Have Identical Zincbinding Catalytic Site
Motifs-The ADAMTS-9 catalytic site is identical to that of ADAMTS-1 and ADAMTS-15 and very similar to that of ADAMTS-4 (Fig. 3a). The unique feature of ADAMTS-9, ADAMTS-1, and ADAMTS-15 is the presence of a proline residue preceding the third zinc-coordinating histidine (Fig. 3a). The corresponding amino acid is leucine in AD-AMTS-4, the next most closely related enzyme. The AD-AMTS-20 zinc-binding site is not identical to that of any other ADAMTS but is most closely related to that of ADAMTS-7 and ADAMTS-12 with 4/12 variant amino acids (Fig. 3a). Moreover, all of the substitutions in the ADAMTS-20 active site relative to ADAMTS-7 and ADAMTS-12 are conservative ones. Alignment and clustering of the published ADAMTS proteases confirm the unique place of ADAMTS-9 and ADAMTS-20 in the ADAMTS family (Fig. 3b) and indicate that they constitute a distinct subfamily of proteases.
ADAMTS-9 and ADAMTS-20 Are Related to GON-1-The domain organization and primary sequence of ADAMTS-9 and ADAMTS-20 have a greater similarity to GON-1 than any other mammalian ADAMTS enzyme (Figs. 1b and 2). AD-AMTS-9 and ADAMTS-20 are equally related to GON-1 in paired BLAST comparisons. The percentage identity of AD-AMTS-9 protein to GON-1 is 33% (that of ADAMTS-20 is 32%), and the percentage similarity (including conservative substitutions) is 46% for both ADAMTS-9 and ADAMTS-20 relative to GON-1. The zinc-binding active site sequence of GON-1 resembles ADAMTS-9 more closely than ADAMTS-20, with just 2 of 14 variant amino acids (Fig. 3a). The conserved C-terminalmost convertase-processing site is at an identical location in ADAMTS-9, ADAMTS-20, and GON-1. The unique C-terminal domain varies slightly in length but nevertheless is highly similar in the three enzymes, including an identical cysteine signature (Fig. 2). TSR-1 is well conserved in these ADAMTS enzymes, but there is less similarity between TSRs 2-15 of ADAMTS-9 and ADAMTS-20 and TSR-2 to -18 of GON-1.
ADAMTS9 and ADAMTS20 Are Located on Different Chromosomes but Have a Highly Conserved Gene Structure-Exons corresponding to the overlapping ADAMTS9 and ADAMTS20 cDNA clones were found arranged sequentially on human chromosomes 3p14 (as previously mapped (18)) and 12q11, respectively (GenBank TM and Celera Genomics, Rockville, MD). AD-AMTS9 and ADAMTS20 are large, being 137 and 200 kb in size, respectively. The ADAMTS9 and ADAMTS20 ORFs are each encoded by 39 exons (Fig. 2). Notably, all of the splice boundaries between exons are identical in ADAMTS9 and AD-AMTS20 mRNAs (Fig. 2). The exons vary in size, with the largest, coding exon 2, encoding half the prodomain and the smallest, exon 7, encoding just 14 amino acids within the catalytic domain. Like other ADAMTS proteases, the active site sequence is split by an intron. The majority of the TSRs are encoded by single exons whose 5Ј end is just upstream of the first cysteine codon, but there are three exceptions to this rule: TSR-1, TSR-2, and TSR-3. TSR-1 and TSR-2 are each split by an exon junction and are thus encoded by two separate exons, although the exon junctions are not at the same location in each TSR. The 5Ј exon junction in TSR-3 is 8 amino acids upstream of the first cysteine residue. Linker 2 is encoded by the exon encoding TSR-8, whereas a separate exon encodes linker 1. Interestingly, the unique C-terminal domain is encoded by six exons, so that with the exception of 13 of 15 TSRs, none of the presumed intrachain disulfide-bonded domains are encoded in their entirety by single exons.
ADAMTS9 and ADAMTS20 Are Differently Regulated-In contrast to ADAMTS9, expression of ADAMTS20 was not detectable in Northern blots of human adult mRNA. In mouse embryos, a single ADAMTS9 mRNA of ϳ8.5 kb was detected (Fig. 4a). Expression was highest in 7-and 17-day-old embryos and lower in 11-and 15-day-old embryos. A number of adult human tissues expressed ADAMTS9, with highest expression in heart, placenta, and skeletal muscle (Fig. 4b). All these tissues contained an 8.0-kb mRNA, but kidney and ovary contained additional mRNAs of 4.5 kb, and kidney contained a hybridizing mRNA of 3.0 kb. Spleen, thymus, prostate, testis, small intestine, and peripheral blood leukocytes had low to undetectable levels of ADAMTS9 on Northern blots.
Quantitative RT-PCR was undertaken to determine which tissues, if any, expressed ADAMTS20 and to measure relative levels of ADAMTS9 and ADAMTS20 mRNAs. In all human fetal and adult tissues examined (other than peripheral blood leukocytes), ADAMTS20 mRNA levels were 1-3 orders of magnitude lower than ADAMTS9 (Fig. 4, c and d). ADAMTS9 was expressed at higher levels than ADAMTS20 in ovary and testis, the two tissues relevant to GON-1 function (Fig. 4d). By Northern blot analysis, expression of ADAMTS9 was substantially higher in ovary than in testis (Fig. 4b).
Since ADAMTS20 expression levels were so low, we asked whether it was detectable at the single cell level using the highly sensitive RNA in situ hybridization approach. The data demonstrate that in normal breast (Fig. 4f) and lung (Fig. 4g), as well as in breast cancer (Fig. 4i) and lung cancer (Fig. 4l), ADAMTS20 mRNA was detectable at low levels in epithelial cells but was not expressed in stromal cells (Fig. 4, f, g, i, and  l). Sense probe showed no hybridization (Fig. 4, e, h, and k).

ADAMTS-9 Is Located Near the Cell Surface but Not in
Conditioned Medium-ADAMTS-9 FLAG was detected in lysates of transiently transfected COS-1 and 293 cells as two major anti-FLAG reactive bands migrating at ϳ180 and Ͼ250 kDa under reducing conditions (Fig. 5a), although the 250-kDa band was inconsistently seen. In addition, a number of smaller FLAG-tagged bands, presumably derived from the full-length ADAMTS-9 were also seen (Fig. 5a, upper panel). Treatment of ADAMTS-9-expressing cells with an increasing concentration of NaCl demonstrated a concentration-dependent release of ADAMTS-9 from the cells (Fig. 5a, lower panel). Due to the unfavorable effects of supraphysiological salt concentrations on cell viability, concentrations higher than 340 mM were not tested.
To identify the cellular or extracellular location of AD-AMTS-9 and contrast it with ADAMTS-4, ADAMTS-5, and the ADAMTS-like protein, punctin (21), transiently transfected COS-1 and 293 cells were immunostained with anti-FLAG M2 antibody with or without permeabilization. In permeabilized COS-1 cells, there was cytoplasmic staining characteristic of localization to endoplasmic reticulum and Golgi apparatus as well as cell surface staining (data not shown). In nonpermeabilized cells, ADAMTS-9 was localized to the cell surface of transiently transfected cells and/or to their substratum (Fig. 5, b  and c). Negative controls did not show immunostaining with the FLAG M2 antibody. ADAMTS-4 immunolocalization resembled ADAMTS-9, with granular staining present in the cell substratum and on the cell surface (Fig. 5d), whereas AD-AMTS-5 was exclusively present in the substratum (Fig. 5e). The substratum staining of ADAMTS-4 or ADAMTS-9 and ADAMTS-5 were qualitatively different; ADAMTS-5 staining was of fine granularity and was spread under and around the transfected cell in a nebulous appearance (Fig. 5e), whereas that of ADAMTS-9 and ADAMTS-4 was coarsely granular and limited to cell boundaries.

Intracellular Maturation of ADAMTS-9 Involves N-Glycosylation of the Prodomain and Furin Processing at the Arg 287 -
Phe 288 Bond-The predicted molecular masses of signal peptide processed and ADAMTS-9 1-508 that is processed at the consensus proprotein convertase sites are shown in Fig. 6a. Transient expression of ADAMTS-9 1-508FLAG in 293 cells followed by pulse-chase analysis, immunoprecipitation using anti-FLAG M2 antibody, and fluorography identified three major immunoreactive bands in cell lysates with molecular masses of ϳ66, 56, and 54 kDa, respectively. The relative intensity of these bands varied with the duration of pulse and chase. After a 15-min pulse and 60-min chase, the amount of the 66-kDa protein seen was significantly greater than that seen after a 15-min chase (Fig. 6b). Conversely, the 54 -56-kDa doublet was more prominent after a 15-min chase (Fig. 6b). The 66-kDa band intensified substantially after a 135-min chase with very little of the 54 -56-kDa doublet being detectable. When cell lysate and culture medium were immunoprecipitated and immunoblotted with anti-FLAG M2 antibody 48 h following transfection of QBI 293A cells, the cells contained the 66-kDa band and essentially no 54 -56-kDa doublet (Fig. 6c). When these cell lysates were treated with PNGase F, this 66-kDa band was reduced to a doublet of ϳ54 -56 kDa (Fig. 6c). Collectively, these observations suggest that the 66-kDa band is derived from a 54 -56-kDa precursor by N-linked glycosylation. N-Glycosylation of ADAMTS-9 1-508HIS was confirmed by culture of stably transfected cells in the presence of the tunicamycin A homolog (data not shown). Under the pulse-chase conditions used, no labeled protein could be immunoprecipitated from the conditioned medium (Fig. 6b), and protein corresponding in size to the active form (28 kDa) was not seen in cell lysate. However, in stably transfected cells (not shown) or immunoprecipitation 48 h after transfection, the mature, tagged protein could be detected in culture medium (Fig. 6c). Deglycosylation did not alter the migration of the secreted mature enzyme (Fig. 6c). N-terminal sequencing of the secreted mature ADAMTS-9 1-508HIS gave the sequence Phe-Ser-Leu-Tyr-Pro-Arg-Phe.
Furin-deficient CHO.RPE 40 cells did not process AD-AMTS-9 (Fig. 7a). Processing was rescued by transfection with furin (Fig. 7a). In QBI 293A cells, the Arg 33 3 Ala, Arg 74 3 Ala, or Arg 280 3 Ala mutants did not affect the appearance of FIG. 3. a, comparison of the active site sequences of mammalian ADAMTS and GON-1 proteases. These are arranged in descending similarity to ADAMTS-9. The overlined residues are conserved in all ADAMTS proteases. b, phylogenetic relationship between human ADAMTS protein sequences. These were obtained using the ClustalW algorithm (Megalign software; DNAStar Inc., Madison, WI). In both a and b, ADAMTS-9 and AD-AMTS-20 are in boldface type, the previously identified aggrecanases are italicized, and the procollagen amino propeptidases are enclosed in a box.
the mature protein in the medium, but abrogation of the most C-terminal processing site (Arg 287 3 Ala) resulted in failure of processing to the mature form (Fig. 7b). Expression of the Arg 74 3 Ala mutant resulted in anomalous bands of ϳ40 and ϳ45 kDa in conditioned medium in addition to the mature protein (Fig. 7b). Instead of the mature 28-kDa form, expression of the Arg 287 3 Ala mutant resulted in the appearance of ϳ37and ϳ42-kDa proteins in culture medium whose identity is not known (Fig. 7b).

Identification of the Full-length Product Of ADAMTS9 -
Although a report of the ADAMTS9 mRNA (GenBank TM accession number AF 261918) and ADAMTS9 chromosomal localization was published (17) while our work was in progress, the novel sequence data we report here extend the predicted C terminus of that protein further to include an additional 10 TSRs and the unique C-terminal domain. Our data suggest that the ADAMTS9 transcript presented here encodes the full- FIG. 4. ADAMTS9 and ADAMTS20 mRNA expression. a, Northern blot of mRNA from mouse embryos hybridized to an Adamts9 cDNA probe. b, Northern blots from human tissues hybridized to an ADAMTS9 cDNA probe. RNA kilobase markers are shown at the left of the autoradiogram, and tissue origin is indicated above each lane. c and d, comparative quantitative RT-PCR analysis of ADAMTS9 and ADAMTS20 mRNA in fetal human tissues (c) and in adult human tissues (d). Values for each tissue were normalized to GAPDH levels and are indicated as -fold increase of ADAMTS9 over ADAMTS20, except in leukocytes, where there was a relative excess of ADAMTS20 over ADAMTS9 (indicated by hatched bar and vertical scale to the right). Tissue source is indicated under each bar. e-m, RNA in situ hybridization of ADAMTS20 mRNA in normal breast, normal lung, and squamous cell carcinoma of breast and lung adenocarcinoma. e, h, and k show a lack of hybridization of normal breast, breast cancer, and lung cancer, respectively, to sense probe. f and g, hybridization of antisense probe to epithelial cells in the normal breast duct epithelium (arrow) and bronchial lining epithelium (arrow) respectively. i and l, hybridization of antisense probe to carcinoma cells in breast cancer (arrow in i) and lung cancer (arrow in l). Note that tumor stroma (asterisk) does not express ADAMTS20, and no expression is seen in necrotic lung tumor (N). e and f, h-j, and k-m, serial sections. j and m, hematoxylin and eosin-stained sections corresponding to i and j, respectively. length, authentic product of this gene for several reasons. First, the previously described ADAMTS9A cDNA diverges from our ADAMTS9 sequence at an unspliced intron (deduced by comparison of the ADAMTS9A and mRNA sequence with the AD-AMTS9 genomic sequence). Another ADAMTS9 product predicted by the sequence of the KIAA1233 gene (GenBank TM accession number AB037733) is incomplete at both the amino and carboxyl termini. Comparison of this sequence with the cDNA sequence reported here and the ADAMTS9 genomic sequence suggests the inclusion of an unspliced intron leading to a premature stop codon. Intron inclusion suggests cloning of partially processed pre-RNA, not authentic mRNA. Second, the ADAMTS-9A transcript does not contain a consensus polyadenylation sequence upstream of the poly(A) tail, in contrast to the ADAMTS9 transcript reported here. Third, Northern analysis demonstrated that probes from the novel sequences we describe here, as well as a probe from the region shared by all transcripts (data not shown), hybridized to the same major 8-kb band on Northern blots, suggesting that the dominant transcript in most tissues encodes the longer form, ADAMTS-9B. Previous studies of ADAMTS9 had shown widespread expression in fetal and adult tissues by RT-PCR (18). Our studies using Northern blots and quantitative RT-PCR are in agreement with this and provide additional information about the mRNA size. The 8-kb ADAMTS9 mRNA is compatible with the long ORF we have cloned. The identity of the smaller mRNAs found in kidney and ovary is presently unclear. These could represent alternative splice forms or unrelated transcripts that cross-hybridize with the probe. Since ADAMTS20 mRNA is undetectable on Northern blots, both its size and the existence of alternative forms is unknown. In fact, ADAMTS20 transcripts are extremely rare in all of the tissues we have examined, and there are only two human ADAMTS20 expressed sequence tags (AU132053 and BG212007) reported in Gen-Bank TM . Nevertheless, a sensitive RNA in situ hybridization approach did demonstrate low levels of expression in epithelial cells of breast and lung origin. The prevalence and biological significance of this low level expression is unknown. Therefore, at the protein level, detailed characterization of the more abundantly expressed enzyme, ADAMTS-9, was subsequently undertaken.
ADAMTS9 and ADAMTS20 Constitute a Distinct Subfamily of ADAMTS Proteases-ADAMTS proteases can be clustered into subfamilies of closely related enzymes on the basis of their domain organization and primary sequences. The procollagen aminopropeptidase subfamily (ADAMTS-2, -3, and -14) represents the most striking example, and other enzymes such as ADAMTS-7 and -12 and ADAMTS-6 and -10 occur in closely related pairs. The ADAMTS-9 and ADAMTS-20 subfamily is particularly interesting, because it is the first such ADAMTS subfamily with a closely related ortholog in invertebrates, indicating, perhaps, a highly conserved physiological role. However, unlike the other ADAMTS subfamilies, ADAMTS-9 and ADAMTS-20 do not have identical zinc-binding active site sequences. Furthermore, their expression patterns are quite different, suggesting they may have nonredundant biological roles.
The genomic organization of ADAMTS9 and ADAMTS20 bears little resemblance to other genes in the family. AD-AMTS-1 is encoded by nine exons, and the prodomain, disintegrin-like domain, and central TSR are each encoded by single exons, whose boundaries coincide with the domain boundaries (30). In ADAMTS-1, a single terminal exon encodes the spacer and two C-terminal TSRs (30). This is clearly not the case with ADAMTS9 and ADAMTS20, where few domains other than the TSRs are encoded by single exons. ADAMTS13 (13) has 29 protein-coding exons whose boundaries are different from AD-AMTS-1, -9, and -20. The procollagen aminopropeptidases share a different genomic organization (12). Therefore, gene structure may be conserved in ADAMTS subfamilies, but there is not a characteristic gene structure that is shared by the entire family.
The Cys to Tyr substitution in TSR-13 is not an artifact of cloning, because we found it both in the genomic DNA (in Celera and GenBank TM databases), in the cloned cDNA, and in a small number of normal human alleles in which the corresponding exon was subjected to PCR-direct sequencing (data not shown). It may represent a non-synonymous single nucleotide (4715A2G) polymorphism, since TSR-13 in mouse ADAMTS-20 4  lence and significance of this amino acid change in humans is not presently known and will be investigated further.

Intracellular Maturation Of ADAMTS-9 Involves Glycosylation of the Prodomain and Processing at a Single Proprotein
Convertase-processing Site-Following removal of the signal peptide and entrance into the secretory pathway, ADAMTS proteases, like ADAMs and some MMPs, are processed further by one or more proprotein convertases to remove the prodomain and undergo additional post-translation modification such as glycosylation. Proprotein convertases (e.g. furin) are serine proteases present in the Golgi apparatus or at the cell surface that typically cleave immediately following a consensus recognition sequence rich in basic residues (31). Our studies showed that processing did not occur in the absence of furin but could be rescued by transfection of furin, demonstrating that proprotein convertases were essential for pro-ADAMTS-9 maturation.
Our studies suggest that there is rapid glycosylation of the ADAMTS-9 prodomain following synthesis that is essentially complete in about 2 h. There is no N-glycosylation of the catalytic domain, consistent with the observation that the prodomain contains three consensus N-glycosylation sites, whereas the catalytic domain has none. Our data indicated processing of the Arg 287 -Phe 288 peptide bond, whereas none of the other furin sites appear to be used for enzyme maturation. We should emphasize that the Arg 280 mutation would abrogate two furin sites, since this residue serves as the P1 Arg for the Arg-Glu-Lys-Arg 287 site as well as the P4 residue for the Arg-Thr-His-Arg 283 site. We could detect the 28-kDa mature form intracellularly in the wild-type and Arg 33 3 Ala mutant. This then accumulates in the medium following secretion through the constitutive secretory pathway. On the other hand, in the Arg 287 3 Ala mutant, the precursor is not processed intracellularly and accumulates in the medium along with other unidentified bands. The N terminus of mature ADAMTS-9 determined by amino acid sequencing was in agreement with the location of the N terminus of mature ADAMTS-1, ADAMTS-4, and ADAMTS-13, suggesting that although more than one processing site may be present, the C-terminal furin-processing site is generally used for production of the mature ADAMTS enzymes.
Western blotting of full-length ADAMTS-9 suggested that it undergoes substantial post-translational modification. In keeping with the number of consensus sites for N-linked glycosylation and the large number of serine and threonine residues, glycosylation of full-length ADAMTS-9 has also been noted (data not shown), as is shown in the prodomain. Expression of full-length ADAMTS-9 demonstrated the existence of a number of smaller FLAG-tagged fragments that were presumably derived from it by proteolysis. Regulated processing has been noted in ADAMTS-1 (32), ADAMTS-4 (33), and ADAMTS-12 6. a, scheme of the protein encoded by ADAMTS9 1-508FLAG . The domains included in the expressed proteins and the locations of N-linked sugar attachment (lollipops) and FLAG tag are shown. Below this are the protein species predicted following signal peptidase cleavage or cleavage at each of five consensus furin cleavage sites. The expected molecular mass of each unmodified protein species is shown at the right. b, pulse-chase analysis of ADAMTS9 1-508FLAG -transfected QBI 293A cells. Cells were pulsed with radiolabeled amino acids and chased for varying times as indicated. Control cells were transfected with empty expression vector. Cell extracts and media were immunoprecipitated with anti-FLAG M2 monoclonal antibody and detected by fluorography. The arrowhead indicates a doublet at 54 -56 kDa, and the arrow indicates a major N-glycosylated band at 66 kDa. C, cell lysates; M, medium. Molecular mass markers are shown at left. c, deglycosylation of ADAMTS9 1-508FLAG by PNGase F. Transiently transfected QBI 293A cell lysates and culture medium were immunoprecipitated with anti-FLAG M2 48 h after transfection. Western blot analysis was done using anti-FLAG-M2. Ig, the immunoglobulin heavy chain. The arrow indicates the mature form in culture medium, and the arrowheads indicate the intracellular zymogen form. (34) and is a potentially intriguing phenomenon because the released ancillary domains could have interesting biological functions or modify the function of ADAMTS-9 (33). Proteolytic fragments of the native enzyme will be sought in tissues and cells once specific high affinity antibodies are available.
ADAMTS-9 Is Located near the Cell Surface and Is Involved in Versican and Aggrecan Degradation-Neither ADAMTS-9 or ADAMTS-20 nor any of the other known ADAMTS proteases has a potential transmembrane sequence or a glycophosphatidylinositol signal anchor sequence. Therefore, these are not predicted to be membrane-anchored enzymes. Accordingly, studies with various ADAMTS proteases have shown that they are soluble or associated with the ECM (3,4,35). ADAMTS-9 and ADAMTS-4 are therefore the first ADAMTS proteases shown to localize near the cell surface, as demonstrated by immunofluorescence microscopy, although their precise location relative to the cell membrane or the binding mechanism is presently unknown. In contrast, both the localization and appearance of ADAMTS-5 distribution are different. Furthermore, although restricted to the ECM, ADAMTS-5 presents a different distribution than punctin, an ADAMTS-like protein comprising only ancillary domains (21). Punctin localization to the cell substratum (21) and the failure of ADAMTS-9 1-508 or C-terminally truncated ADAMTS-1 (35) to be located in either the ECM or cell surface strongly validates the role of the ancillary domains in anchoring these enzymes near the cell. ADAMTS-9 has consensus sites for binding to heparin (and therefore to heparan sulfate proteoglycans) and CD36, and these may be candidate cell surface and pericellular ECM ligands. In support of this possibility, ADAMTS-9 was released from cells and ECM by gentle washes with low concentrations of salt.
To identify potential substrates for ADAMTS-9, we relied upon comparison of the ADAMTS active site sequences, the phylogenetic profile of the ADAMTS family, and the previous descriptions of their enzymatic activities. The ADAMTS enzymes (ADAMTS-1, ADAMTS-4, and ADAMTS-5) that process the large aggregating proteoglycans versican, aggrecan, and brevican have very similar (although not identical) active site sequences, but they have different domain structures. Because ADAMTS-9 has an active site sequence identical to that of ADAMTS-1 and similar to that of ADAMTS-4, we considered that it might be a proteoglycan core protein-degrading enzyme. Since ADAMTS-9 was not secreted into the culture medium of cells, we used a cell-based ADAMTS assay. Serum-free culture medium has the appropriate pH and salt concentration for ADAMTS activity and, when supplemented with calcium, provided the reaction conditions necessary for the versicanase and aggrecanase assays.
By analogy with aggrecanase-susceptible sites in aggrecan, Sandy et al. (8) had previously predicted two putative ADAMTS cleavage sites in human versican and had prepared polyclonal antisera recognizing one such predicted neoepitope generated by proteolysis of the V1 Glu 441 -Ala 442 bond (8). Versican V0 and V1 forms differ in the inclusion of the GAG-␣ region that is present in the V0 form but missing in the V1 form. Accordingly, the peptide bond cleaved has a different location in the two forms (8). Consistent with the mixed population of versican made by smooth muscle cell cultures, two bands (70 and ϳ180 kDa corresponding to G 1 versican fragments DPEAAE (V0 form) and DPEAAE (V1 form)) were seen in previous studies of ADAMTS-4 processing of versican (8). Of these, the 70-kDa band was considerably stronger, consistent with there being more of the V1 form in the versican preparation (8). In contrast, neither ADAMTS-4 nor ADAMTS-9 proteolysis gave an anti-DPEAAE reactive band at 180 kDa in our experiments.
A refined comparison of ADAMTS-4 and ADAMTS-9 cannot be done in the cell-based assay, since transfection efficiency, expression levels, secretion, and zymogen processing may be different. Purified ADAMTS-9 is not yet available, and given its complex domain structure and cell surface localization, it may be difficult to obtain. We have purified ADAMTS-9 1-508 , but it does not process versican or aggrecan, demonstrating the essential role of the ancillary domains in substrate recognition and/or binding. Therefore, this form cannot be used in kinetic studies to compare with ADAMTS-4. With these limitations, however, it appears from our studies that ADAMTS-9 may be an efficient versicanase, comparable with ADAMTS-4, but a less efficient aggrecanase. Versican is widely distributed during development and in adult tissues. Like aggrecan, it may have a mechanical role, and since it interacts with fibrillin, fibulin-1, and fibulin-2 through its lectin-like domain, it may have a specific role in matrices enriched in these molecules (36 -38). In addition, versican is believed to provide guidance cues for migrating neural crest cells (39). The presence of ADAMTS-9 near the cell surface and its homology to an enzyme required for cell migration certainly make it an appropriate enzyme for involvement in the migratory process. In addition, the ability to process aggrecan warrants further investigation of its involvement in skeletal development and cartilage destruction in arthritis.
Comparison of Proteoglycan-degrading ADAMTS Enzymes and the Procollagen Amino Propeptidases Suggests Differences in Stringency of Enzyme-Substrate Interaction-The procollagen N-propeptidases have identical domain organizations and identical active site sequences (in fact, ϳ70 amino acids around the zinc-binding site are identical in these enzymes) (12). This identity suggests that the structural requirements for procollagen processing are very stringent, and indeed, the catalytic sites of the procollagen N-propeptidases have a distinctive cysteine signature not found in other ADAMTS. Similarly, AD-AMTS-13, the von Willebrand factor protease, is unlike any other ADAMTS in its domain organization and is clearly the major, if not only, von Willebrand factor-processing enzyme (13), suggesting that the structural requirements for this function are stringent as well. In contrast, the four ADAMTS enzymes that degrade proteoglycan core proteins have neither an identical domain organization nor identical active site sequences. This dissimilarity suggests a relatively relaxed structural requirement for proteoglycan processing and supports the likelihood that additional ADAMTS enzymes may have activity against proteoglycans. In the future, it will be important to define the relative prevalence of each of the proteoglycan-degrading ADAMTS enzymes in different tissues as well as in diseases such as arthritis and to determine their tissue-specific role by targeted inactivation of the corresponding mouse genes. In future studies, it will also be important to ask whether ADAMTS-20 can process versican and aggrecan and to ask whether ADAMTS-9 and ADAMTS-20 have biological roles similar to GON-1.