Discovery and Characterization of a Novel Inhibitor of Matrix Metalloprotease-13 That Reduces Cartilage Damage in Vivo without Joint Fibroplasia Side Effects*

Matrix metalloproteinase-13 (MMP13) is a Zn2+-dependent protease that catalyzes the cleavage of type II collagen, the main structural protein in articular cartilage. Excess MMP13 activity causes cartilage degradation in osteoarthritis, making this protease an attractive therapeutic target. However, clinically tested MMP inhibitors have been associated with a painful, joint-stiffening musculoskeletal side effect that may be due to their lack of selectivity. In our efforts to develop a disease-modifying osteoarthritis drug, we have discovered MMP13 inhibitors that differ greatly from previous MMP inhibitors; they do not bind to the catalytic zinc ion, they are noncompetitive with respect to substrate binding, and they show extreme selectivity for inhibiting MMP13. By structure-based drug design, we generated an orally active MMP13 inhibitor that effectively reduces cartilage damage in vivo and does not induce joint fibroplasias in a rat model of musculoskeletal syndrome side effects. Thus, highly selective inhibition of MMP13 in patients may overcome the major safety and efficacy challenges that have limited previously tested non-selective MMP inhibitors. MMP13 inhibitors such as the ones described here will help further define the role of this protease in arthritis and other diseases and may soon lead to drugs that safely halt cartilage damage in patients.

Matrix metalloproteinase-13 (MMP13) is a Zn 2؉ -dependent protease that catalyzes the cleavage of type II collagen, the main structural protein in articular cartilage. Excess MMP13 activity causes cartilage degradation in osteoarthritis, making this protease an attractive therapeutic target. However, clinically tested MMP inhibitors have been associated with a painful, joint-stiffening musculoskeletal side effect that may be due to their lack of selectivity. In our efforts to develop a disease-modifying osteoarthritis drug, we have discovered MMP13 inhibitors that differ greatly from previous MMP inhibitors; they do not bind to the catalytic zinc ion, they are noncompetitive with respect to substrate binding, and they show extreme selectivity for inhibiting MMP13. By structurebased drug design, we generated an orally active MMP13 inhibitor that effectively reduces cartilage damage in vivo and does not induce joint fibroplasias in a rat model of musculoskeletal syndrome side effects. Thus, highly selective inhibition of MMP13 in patients may overcome the major safety and efficacy challenges that have limited previously tested non-selective MMP inhibitors. MMP13 inhibitors such as the ones described here will help further define the role of this protease in arthritis and other diseases and may soon lead to drugs that safely halt cartilage damage in patients.
The National Institutes of Health has estimated that more than 20 million adults in the United States suffer from osteoarthritis (OA), 3 a debilitating disease in which the protective cushion of cartilage is destroyed, resulting in pain and reduced mobility. A critical step in OA pathology is break-down of the main structural protein of articular cartilage, type II collagen. This triple helical protein is resistant to most proteases but is efficiently recognized and degraded by the Zn 2ϩ -dependent enzyme, collagenase-3, known as matrix metalloproteinase-13 (MMP13) (1)(2)(3). MMP13 catalyzes the hydrolysis of type II collagen at a unique site resulting in 3 ⁄4-and 1 ⁄4-length polypeptide products (2)(3)(4)(5)(6). MMP13 is not found in normal adult tissues but is expressed in the joints and articular cartilage of OA patients (4 -8). In addition, regulated expression of human MMP13 in hyaline and joint cartilages induces OA in genetically modified mice (9). Furthermore, a MMP inhibitor that preferentially inhibits MMP13 has been shown to block the degradation of explanted human osteoarthritic cartilage (5). Based on these findings, it is likely that MMP13 is the direct cause of irreversible cartilage damage in OA.
The clinical development of drugs that inhibit the actions of MMPs has been plagued by the association of a painful, joint-stiffening tendonitis-like side effect, termed "musculoskeletal syndrome" (MSS), with these inhibitors (10,11). Such joint side effects are not unique to humans. Rats dosed with non-selective MMP inhibitors (i.e. compounds that inhibit several or all MMPs) also display MSS-like side effects such as soft tissue fibroplasias, inflammation, and pain (12). Although the human joint side effects are reversible upon withdrawal of drug, MSS has halted clinical trials of many non-selective MMP inhibitors (10). We began a search for MMP13-selective inhibitors with the hypothesis that they would effectively prevent cartilage degradation without causing MSS-like side effects.
104 to Asp-270 (SWISS-PROT P45452) was generated by synthetic gene methods and bacterial expression (14), and was used for both high throughput screening and crystallographic studies. In the MMP IC 50 assays the enzyme was used at a level at least 10-fold lower than the IC 50 . The initial rate of substrate hydrolysis was determined by monitoring the increase in absorbance at 412 nm on a microplate reader thermostatted at 25°C. The % of control activity was plotted against inhibitor concentration and fit to the equation % of control activity ϭ 100/(1 ϩ ([I]/IC 50 ) slope ), where [I] is the inhibitor concentration, IC 50 is the inhibitor concentration where the rate is 50% reduced relative to the control, and slope is the slope of the curve at its inflection point.
Steady-state kinetics data were fit by global non-linear least squares regression using GraFit Version 5 software (15) to the equation for noncompetitive inhibition (16) and [S] are the concentrations of inhibitor and substrate, respectively, v is the initial velocity, K m is the Michaelis constant, V max is the maximal rate, and K i is the inhibition constant. To rule out a more elaborate model, we also fit the data to the equation for mixed inhibition (16) where ␣ is the factor by which the K i is changed when the substrate is bound to the enzyme.
Type II Collagen Cleavage Assay-Reaction mixtures to assay for inhibition of MMP13-catalyzed hydrolysis of type II collagen contained 50 mM Tris-HCl buffer (pH 7.4), 10 mM CaCl 2 , 150 mM NaCl, 0.5 mg/ml soluble bovine collagen (Elastin Products Co. # CJ385), 0.005% Brij 35, and inhibitor in Me 2 SO (1% final). Activated MMP13FL (2 nM) was added, and the reactions were incubated at 23°C for 18 h. Reactions were stopped by adding an equal volume of 2ϫ SDS sample loading buffer. Proteins were separated by denaturing electrophoresis (8% Trisglycine) and then visualized with Coomassie R-250. The 3 ⁄ 4length product protein bands were scanned and quantified using a Bio-Rad GS-700 imaging densitometer with Quantity One Version 4.0 software. The IC 50 was defined from a plot of the % of control 3 ⁄ 4-length product density against inhibitor concentration.
Crystallography-MMP13CD protein for crystallography was expressed in E. coli BL21(DE3) cells harboring pLysS and pGEMEX-1/MMP13CD. Cells were grown in a 20-liter fermentor thermostatted to 30°C and sparged with ϳ8 liters of air/ min. The medium (15 liters) contained 260 g of Difco 0127 yeast extract, 260 g of acidicase peptone (BBL 211843), 260 g of Difco 0259 casitone, 260 g of gelysate peptone (BBL 211870), 26 g of KH 2 PO 4 , 26 g K 2 HPO 4 , 26 g of Na 2 HPO 4 . 7H 2 O, and 1.5 g of ampicillin and was titrated to pH 6.8. The culture was stirred with an impeller at 600 rpm and was maintained at pH 6.8 Ϯ 0.2 by the addition of 85% lactic acid, whereas foam was reduced using Antifoam 289. At a culture A 600 ϳ 10, the temperature was raised to 37°C, and protein expression was induced by adding isopropyl-␤-D-thiogalactopyranoside to 3.2 mM. After 3 h the cells were collected by centrifugation. The wet cells (400 g) were suspended in 1.3 liters of 50 mM Tris-HCl buffer (pH 8.0) plus 1% Triton X-100, 10 mM MgCl 2 , and 40 l of Benzonase. The cells were twice passed through a Dyno-Mill (Glenn Mills Type KDL 0.6-liter chamber) with 0.5 liters of 0.25-0.50-mm glass beads at an impeller speed of 4200 rpm and a flow rate of 100 ml/min. Insoluble material was washed by sequentially suspending it in 2 liters each of the following solutions and centrifuging the washed mixtures: 1) 50 mM glycine buffer (pH 10.0), 10 mM EDTA, 1% Triton X-100 (3 times); 2) 50 mM glycine buffer (pH 10.0), 10 mM EDTA; 3) 50 mM Tris⅐HCl buffer (pH 8.0), 10 mM dithiothreitol; 4) deionized water (2 times). The protein pellet was dissolved by suspending it in 50 mM Tris⅐HCl buffer (pH 7.6) plus 6 M guanidine and stirring the solution at room temperature. After centrifugation, the protein in the soluble fraction was brought to ϳ0.15 mg/ml and was refolded at Ϫ15°C by diafiltration into 50 mM Tris⅐HCl buffer (pH 7.6) containing 10 mM CaCl 2 , 0.1 mM ZnCl 2 , and 10% glycerol using an Amicon S3Y10 membrane. The refolded protein was passed through a 0.2-m filter, concentrated to ϳ1 mg/ml, and frozen in aliquots at Ϫ70°C.
For crystallization, inhibitors in Me 2 SO were mixed in a 5:1 molar ratio with MMP13CD (1 mg/ml) (Me 2 SO, Ͻ5% final). Acetohydroxamic acid was added to 0.1 M, and the solutions were incubated 1 h at 4°C. After centrifugation, the soluble protein was concentrated to 7-20 mg/ml, and 2-4-l hanging drops of 1:1 protein:reservoir solution were equilibrated over 0.5 ml of reservoir solution. The reservoir solution for Compound 1 contained 18 -22% polyethylene glycol monomethyl ether 5000 and 0.2 M Li 2 SO 4 in 0.1 M Hepes buffer (pH 7.0), whereas that for Compound 2 contained 2.1 M (NH 4 ) 2 SO 4 in 0.1 M Hepes buffer (pH 7.5). After several weeks, crystals were cryo-preserved in 15% glycerol, 85% well solution and flashfrozen in a stream of nitrogen.
Diffraction data were collected at 100 K using the MARCCD-165 detector installed on the Industrial Macromolecular Crystallographer Associations beamline 17-ID at the Advanced Photon Source in Argonne, IL. Data were integrated with HKL2000 and reduced using SCALEPACK (17). During structure solution and refinement, initial phases were obtained by molecular replacement with AMoRe (18) using the crystal structure coordinates of human MMP3 (PDB 1CIZ) as a search model (19), modified for sequence differences. After several rounds of model refinement using REFMAC5 (20) and manual correction, the inhibitor, acetohydroxamate, and water molecules were placed using QUANTA-2000 (Accelrys, San Diego, CA).
Cytokine-stimulated Cartilage Explant Degradation Assay-Bovine nasal cartilage explants (1 ϫ 3 mm, 9 per group) were cultured for 2 weeks with interleukin-1␣ (50 ng/ml) and oncostatin M (50 ng/ml) in the presence of vehicle or up to 10 M Compound 2. The culture medium was refreshed twice weekly, and the conditioned medium was pooled and collected for quantification of hydroxyproline (21). At the end of the study, the remaining cartilage was digested with papain, and the hydroxyproline in this sample was measured.
In Vivo Cartilage Degradation Induced by Exogenous MMP13-To assess the ability of orally dosed Compound 2 to reach the knee joint target tissue and inhibit MMP13-induced degradation of cartilage, rats (12 per group) were dosed orally with vehicle or various doses of Compound 2. After 3 h, the knee joints were injected with 12 g of activated MMP13FL enzyme. The animals were sacrificed 2 h later, the knee joints were opened, and the synovial space was lavaged twice with 80 l of saline. The type II collagen neoepitope biomarker in the lavage fluid was quantified by an enzyme-linked immunosorbent assay (22).
Surgical Model of OA-To assess the ability of an MMP13 inhibitor to prevent cartilage damage in vivo in a chronic disease model, Compound 2 was tested in a surgically induced model of OA in rabbits. The Institutional Animal Care and Use Committee reviewed and approved the protocol for this rabbit study. Animals were male specific-pathogen-free New Zealand White rabbits weighing 2.5-3.5 kg. The rabbits (12 per group) were individually housed in stainless steel caging in an Association for Assessment and Accreditation of Laboratory Animal Care accredited facility. Rooms were environmentally controlled to provide a temperature of 61-71°F, a relative humidity of 35-65%, 100% fresh air at a rate of 10 -12 room exchanges per h, and a light:dark cycle of 12:12 h. The rabbits were fed a designated amount of rabbit chow (LabDiet 5321, PMI Nutrition, Brentwood, MO) and provided with a continuous supply of deionized drinking water. The anterior cruciate ligament of one knee joint was transected, and ϳ70% of the medial meniscus was excised. The joints were closed, and the animals were allowed to recover. Beginning 1 day post-operation, the animals were dosed orally twice daily with 30 mg/kg of Compound 2 or vehicle. Five weeks later, the animals were sacrificed, and the knee joints were removed and stained with India ink to evaluate the presence and severity of cartilage lesions as follows. After joint removal and separation of the femoral condyle from the tibial plateau, the joint was wrapped in a paper towel moistened with saline and kept on ice. Each bone was unwrapped and dipped into waterproof drawing India ink (Sanford no. 4418) to immerse the entire articular surface. The bone was then briefly rinsed in a beaker of normal saline then blotted with a clean paper towel to remove excess ink. The remaining ink reveals damage, fibrillation, and fissuring of the cartilage. Each articular surface of the joint was then imaged using three different angles for the femoral condyle and one for the tibial plateau. Images of the lesions were captured using a Nikon SMZ800 dissecting microscope (1ϫ setting) with a SPOT Insight color CCD camera (Diagnostics Instruments Model 320) and a Cadmet MS2000 light source. Images were saved as bitmap files (1600 ϫ 1200, 24 bits per pixel) using SPOT Version 4 software using auto-exposure setting. Images were burned onto compact discs, and lesions were demarcated in blinded fash-ion by two independent, trained investigators. Calibration was performed using images of standards taken on the same microscope at the same time.
Rat Fibroplasia Model of Musculoskeletal Joint Side Effects-To determine the propensity of an MMP13 inhibitor to induce joint side effects, Compound 2 was tested in a rat model of MSS that has been used previously to demonstrate joint toxicities of the non-selective MMP inhibitor, marimastat (12). We orally dosed Sprague-Dawley rats (6/sex) with vehicle or 2000 mg/kg of Compound 2 per day (i.e. the maximum feasible dose) for a period of 2 weeks. A third group of rats given a single daily oral dose of 2000 mg/kg of a non-selective hydroxamic acid MMP inhibitor, Compound 4, (R)-2-(4Ј-bromobiphenyl-4-sulfonylamino)-N-hydroxy-3-methylbutyramide (23), served as the positive control.
The calcaneon tendon and femorotibial, tibiotarsal, and humeroradial joints were prepared bilaterally for microscopic assessment of potential joint-related lesions. After 14 days of oral dosing to rats of either vehicle, MMP13 inhibitor Compound 2 or non-selective MMP inhibitor Compound 4 representative samples of femorotibial joints were collected at necropsy, fixed in 10% neutral-buffered formalin, decalcified, and processed with hematoxylin and eosin for light microcopy to assess any compound-induced changes in joint histology.
Statistical Analyses-For the MMP13 inhibitor kinetic mechanism of inhibition study, the initial rate data were fit to the various models for inhibition using GraFit software (15). This program globally fits the rate data at the various inhibitor concentrations and provides a reduced X 2 value representing the goodness of fit of the data to the kinetic model. For the statistical analysis of animal model data, SigmaStat software (Jandel Scientific, San Rafael, CA) was used. One-way analysis of variance was carried out to see if differences existed among the experimental groups, and Dunnett's test was used to see if the mean responses of the experimental groups were statistically significantly different from control group. Statistical tests were two-tailed and carried out at the p ϭ 0.05 level of statistical significance. The rabbit OA study consisted of at least two groups, and the normality and equal variance tests used for data analysis indicated that the samples satisfied the standard statistical assumptions required.

Discovery and Biochemical Characterization of MMP13
Inhibitors-We screened our compound library for inhibitors of the hydrolytic activity of human MMP13CD and found Compound 1 (6-benzyl-5,7-dioxo-6,7-dihydro-5H-thiazolo[3,2-c]pyrimidine-2-carboxylic acid benzyl ester) to be a potent inhibitor with an IC 50 ϭ 30 Ϯ 2 nM (Fig. 1A). Remarkably, when we tested Compound 1 at up to 100 M against nine other MMPs, we found that this novel inhibitor chemotype did not inhibit these closely related proteases (Table 1). Also highlighted in Table 1 is the improved potency (IC 50 ϭ 0.67 nM) and extremely selective inhibition of MMP13 by an advanced derivative, Compound 2 (4-[1-methyl-2,4-dioxo-6-(3-phenyl-prop-1-ynyl)-1,4-dihydro-2H-quinazolin-3-ylmethyl]-benzoic acid). The high selectivity of Compounds 1 and 2 for MMP13 is in stark contrast to the lack of selectivity of Compound 3, a hydroxamic acid-containing broad-spectrum MMP inhibitor known as GM-6001, or Galardin. In addition, when tested against isolated human ADAMTS4 and -5 in in vitro assays where cleavage of native aggrecan substrate was quantified with the BC-3 neoepitope antibody, Compound 2 showed IC 50 Ͼ 25 M with both of these aggrecanases. 4 In some of the MMP selectivity assays, which are carried out in pH 7 Hepes buffer, Compound 2 showed visible insolubility at 100 M, as noted in Table 1. In routine solubility tests carried out in phosphate buffer at pH 7.5 and 6.5, however, Compound 2 showed solubility of less than 3 M, whereas in pH 6 phosphate buffer it was soluble to 20 M. Compound 1, on the other hand, showed no sign of insolubility up to 100 M, the highest concentration tested in the MMP selectivity assays. Thus, the solubility of Compound 2 is sensitive to the buffer system and pH used.
In steady-state kinetics experiments, Compound 1 is a linear noncompetitive inhibitor of MMP13FL (Figs. 1, B and C) with and a goodness-of-fit reduced X 2 value of 0.0262. In noncompetitive inhibition the inhibitor and the substrate bind independently and simultaneously to the enzyme at non-overlapping sites, and the inhibitor binds with equal affinity to the free enzyme and to the enzyme⅐substrate complex. The presence of bound substrate does not alter the binding affinity of the inhibitor and vice versa. Attempts to fit the data to the rate equations for mixed, competitive, and uncompetitive inhibition models resulted in less satisfactory reduced X 2 values of 0.0270, 0.2947, and 0.1439, respectively. In addition, we determined that Compound 2 is also a linear noncompetitive inhibitor of MMP13. 5 During MMP-catalyzed peptide bond hydrolysis, it is thought that the carbonyl oxygen atom or oxyanion of the scissile amide group binds to, and is polarized by the catalytic zinc ion. Nearly all MMP inhibitors described to date possess a hydroxamic acid, carboxylic acid, or other metal-chelating group that binds to this catalytic zinc ion and competes with substrate binding (24). In contrast, the noncompetitive mechanism of inhibition of MMP13 displayed by Compounds 1 and 2 indicates that these inhibitors do not bind to the catalytic zinc ion or to the peptide binding active site cleft. We confirmed that inhibition of MMP13 by Compound 2 is rapidly and fully reversible using enzyme-inhibitor preincubation-dilution activity studies. 5 Furthermore, when tested in an assay using a more physiologically relevant macromolecular substrate such as type II collagen, Compound 1 retains its ability to potently inhibit MMP13-catalyzed proteolysis with an IC 50 ϭ 51 Ϯ 6 nM (Fig. 2).
Crystal Structures of Inhibitors Bound to MMP13-The noncompetitive mechanism of inhibition of MMP13 by these new chemotypes indicated that they did not bind to the catalytic zinc ion. Thus, we could not use crystal structure data for MMPs bound to zinc binding inhibitors to construct reliable models to predict where these MMP13 inhibitors bind. We, therefore, determined the mode of binding of these unique MMP13 inhibitors by crystallography using MMP13CD. 6 Dif-4 G. Munie, A. Wittwer, and M. Tortorella, unpublished data. 5 A. R. Johnson, unpublished observations. 6 The atomic coordinates of MMP13 in complex with Compounds 1 and 2 have been deposited in the Protein Data Bank as entries 2OW9 and 2OZR, respectively. fraction data collection and refinement statistics are reported in Table 2 for co-crystals of Compounds 1 and 2 with a complex of MMP13CD plus acetohydroxamic acid, a ligand added to prevent autolysis. The secondary and tertiary structures for MMP13CD (Fig. 3A) in general resemble those previously described for MMP13 crystallized with zinc binding inhibitors (25,26), except in the N-terminal region of the protein and in the S 1 Ј-specificity loop (S 1 Ј nomenclature (27)). In various MMP13 crystals we have prepared, including those described here, the N terminus of the protein can adopt different orientations depending on the inhibitor used, the crystallization conditions, and/or the resultant space group observed. 7 However, it is the unique disposition of the S 1 Ј-specificity loop that is most relevant to our drug discovery efforts. We observed that the non-zinc binding MMP13 inhibitors described here confer an ordered structure to the S 1 Ј-specificity loop that is otherwise flexible and poorly defined. The MMP13-Compound 1-acetohydroxamic acid crystal structure reveals that the hydroxamate is bound in a bidentate fashion to the catalytic zinc ion. Compound 1 does not interact with this zinc ion but instead binds deep within the S 1 Ј-specificity loop of the protein and extends past this pocket out toward solvent (Fig. 3B). The benzyl ester of Compound 1 points toward the substrate binding cleft but overlaps only slightly with the space that would be occupied by a P 1 Ј leucine amino acid side chain in productively bound substrates such as type II collagen or the synthetic thiopeptolide or in non-selective peptidic MMP inhibitors such as GM-6001 (Fig. 4A). This binding mode for Compound 1 is consistent with its noncompetitive mechanism of inhibition and contrasts with the substrate competitive inhibition expected for MMP inhibitors that bind to the catalytic zinc ion (24 -26, 28 -32).

TABLE 1 Contrasting the MMP selectivity of two MMP13 selective inhibitors with that of the non-selective MMP inhibitor, Compound 3 (GM-6001)
In addition to not binding the catalytic zinc ion, Compound 1 does not occupy space within the substrate binding cleft of MMP13. Its inhibitory potency and target specificity can be explained by complementarities of the inhibitor and the accommodating S 1 Ј-specificity loop of MMP13 in which it binds (Fig. 3B). Three carbonyl oxygen atoms in Compound 1 form hydrogen bonds to the backbone nitrogens of Thr-224 (2.85 Å), Thr-226 (2.98 Å), and Met-232 (2.88 Å) within the S 1 Ј-specificity loop. In addition, important interactions exist between the phenyl rings of Compound 1 and amino acid residues His-201, Tyr-223, Tyr-225, and Phe-231. The phenyl group of the benzyl ester of Compound 1 participates in a nearly co-planar stacking interaction with His-201 (Fig. 3B), similar to interactions found in co-crystal structures of MMP3 and MMP8 bound to non-selective MMP inhibitors (31,32). The other hydrophobic interactions of Compound 1 with MMP13 are of the aromatic ring edge-to-face variety. An overlay of the orientations of Compounds 1 and 2 in the MMP13 crystal structures (Fig. 4B) shows that these structurally different compounds may nevertheless share certain crucial interactions with the enzyme.
The high selectivity of Compounds 1 and 2 for MMP13 can be rationalized by examination of the residues that comprise the S 1 Ј-specificity loops of the MMPs (Figs. 4C and 5). In most reported MMP structures, including those for MMP13 and MMP3, the S 1 Ј-specificity loop is not highly ordered, even in the presence of zinc binding inhibitors (28,29,32). However, we find that this loop is well defined in complexes of MMP13 with Compounds 1 and 2. Residues Leu-197, Tyr-223, Tyr-225, Gly-227, and Phe-231 play significant roles in determining the selectivity of inhibitor binding (Fig. 4C). The sequence alignment of related MMPs (Fig.  5) reveals that the specificity loops in MMP1, 2, and 9 may be too short relative to that in MMP13 to accommodate these inhibitors. A shorter loop makes the S 1 Ј pocket of these enzymes relatively shallow (MMP1) and narrow (MMP2 and -9). On the other hand, the S 1 Ј-specificity loops in MMP3 and -17 are longer than in MMP13, which may increase loop flexibility and may in part explain the lack of detectable inhibition of these isoforms. Perhaps the most critical feature enabling molecular recognition is the presence of the nonconserved Gly-227 residue in MMP13. Because the , angles adopted by Gly-227 residue in MMP13 are in a region of the Ramachandran plot that is unfavorable for the corresponding Glu/Asp residues in MMP8, -12, and -14, the S 1 Јspecificity loop of these enzymes cannot accommodate these MMP13 inhibitors even though they have large S 1 Ј pockets and the same length specificity loop as MMP13. In addition to different S 1 Ј-specificity loop sizes and shapes relative to MMP13, MMPs 1, 3, 7, and 9 lack aromatic residues at either position 225 or 231 (Fig. 5). In MMP13, the side chains of Tyr-225 and Phe-231 make critical hydrophobic stacking or edge-to-face interactions with the terminal phenyl rings of Compounds 1 and 2. Finally, the S 1 Ј pocket in some MMPs contains bulky residues that may prevent these inhibitors from binding (Fig. 5). Leu-197 forms part of the spacious S 1 Ј pocket in MMP13, but MMP1 and -7 contain Arg and Tyr residues, respectively, at this position. Crystal structures of these latter MMPs show that these larger residues may block access to the S 1 Ј pocket (30).
With a detailed knowledge of the MMP13 protein structure surrounding Compound 1, we have rationally designed analogs, such as Compound 2, which are more potent than Compound 1 and retain extremely high selectivity for MMP13 ( Table 1). The ability to modify the central bicyclic ring system and the linker between the bicyclic core and the phenyl group that abuts the active site cleft allows the design of highly varied analogs with improved potency and physical properties. Compound 2, for example, provides a tool to test whether inhibition of MMP13 alone can prevent cartilage damage.
Inhibition of Cartilage Degradation in Vitro and in Vivo-In cartilage explants stimulated with interleukin-1␣ and oncostatin M, Compound 2 significantly inhibited the release of hydroxyproline from bovine nasal cartilage (Fig. 6A) with an IC 50 ϭ 1.3 Ϯ 0.4 M (Fig. 6B). Given that both MMP1 and MMP13 are induced in this type of explant system (33) and MMPs other than collagenases may contribute to the cleavage of hydroxyproline-containing peptides, this level of inhibitory activity for the MMP13 inhibitor Compound 2 is notable. We next assessed the ability of orally dosed Compound 2 to inhibit the degradation of type II collagen in an acute model of cartilage damage induced by the injection of active human MMP13 into rat knee joints. At doses as low as 0.1 mg/kg, Compound 2 significantly inhibited the formation of the type II collagen neoepitope biomarker (Fig. 6C), which appears upon collagen degradation and cartilage damage. Thus, by oral dosing we could deliver to the knee joint an effective amount of an MMP13 inhibitor sufficient to reduce the cartilage damage induced by exogenously added MMP13.
Given the effective inhibition of cartilage damage in these acute models, we next wanted to determine whether an MMP13 inhibitor could prevent cartilage damage in a chronic in vivo model of osteoarthritis. Several such models have been reported. In a rat model after medial collateral ligament transaction and meniscal tear a non-selective MMP inhibitor inhibited cartilage degradation and osteophyte formation by 39% (34). In a rabbit anterior cruciate ligament transection model, both MMP1 and -13 are known to be up-regulated (35). In a rabbit model in which anterior cruciate ligament transection was accompanied by removal of part of the meniscus (partial meniscectomy), a non-selective MMP inhibitor significantly reduced cartilage damage (36).
Using the rabbit anterior cruciate ligament transection/partial meniscectomy model, we found that Compound 2 dosed orally at 30 mg/kg twice daily provided a dramatic level of cartilage protection that can be seen qualitatively in photographs showing the gross morphology of the cartilage surfaces from representative vehicle and Compound 2-dosed rabbits (Fig.  7A). Quantitatively, Compound 2 significantly reduced the cartilage lesion areas on the tibial plateaus and on the femoral condyles by 68 and 51%, respectively, relative to vehicle-dosed animals (Fig. 7B).
In this study Compound 2 achieved a total plasma exposure of 200 g⅐h/ml and a free fraction (i.e. not protein bound) AUC 0 -24 h exposure of 6 g⅐h/ml, corresponding to an average plasma free fraction that is 148-fold higher than the in vitro potency of Compound 2 with human MMP13FL (IC 50 ϭ 4 nM). The efficacy demonstrated by oral administration of this MMP13 inhibitor suggests that most of the cartilage damage observed in this model is in fact due to MMP13 activity. Because humans and rabbits can express the same collagenases (MMP1, -8, -13, and -14), the response of a compound in a rabbit model should be predictive of its response in treating patients with OA. Finally, in comparison to the level of cartilage protection that was afforded in this type of model by a nonselective hydroxamic acid-containing MMP inhibitor (36), our results with Compound 2 show that inhibition of MMP13 alone is nearly as effective in reducing the cartilage lesion area as broad-spectrum MMP inhibition.
Lack of Musculoskeletal Joint Side Effects-Achieving efficacy has been difficult for clinically applied MMP inhibitors, but safety has also been a challenge due to the occurrence of MSS joint pathology in subjects dosed with non-selective MMP inhibitors. To assess the potential of an MMP13 inhibitor to induce MSS-like symptoms, we followed a protocol similar to one recently described for evaluation of MMP inhibitor-induced joint toxicity (12). As depicted in the representative photomicrographs of femorotibial joint sections from three treatment groups, no MSS-like fibroplasias were observed in any of the animals dosed orally for 2 weeks with vehicle (Fig. 8A) or MMP13 inhibitor Compound 2 (Fig. 8B). In contrast, all 12 animals treated with the non-selective hydroxamic acid-containing experimental MMP inhibitor, Compound 4, displayed joint fibroplasias characterized by marked expansion of the inner synovial lining of the patellar tendon and adjacent quadriceps femoris muscle with a population of plump, immature spindloid fibroblasts contained in a scanty collagenous matrix (Fig. 8C). The AUC 0 -24 h exposure on day 14 for the combined sex group dosed with Compound 2 was 1010 g⅐h/ml (free fraction exposure of 30 g⅐h/ml), which translates to an average free inhibitor plasma level that is 740-fold higher than the in vitro potency of Compound 2 with MMP13. Compound 4 achieved an exposure of 94 g⅐h/ml, whereas exposure of the carboxylic acid active metabolite of Compound 4 reached 684 g⅐h/ml. Thus, the MMP13 inhibitor Compound 2 did not induce joint side effects in this 2-week study at exposures 5-fold above the level needed to effectively protect articular cartilage in the rabbit model and well above the low yet fibroplasia-inducing exposures achieved by the non-selective MMP inhibitor, Compound 4. . Non-selective MMP and MMP13 selective inhibitor binding modes and specificity determinants. A, secondary structure elements of MMP13 are colored (loops, green; ␣-helices, red; ␤-sheets, blue). The three structural calcium ions (green) as well as the structural and catalytic zinc ions (magenta), respectively, are shown at the top and middle of the figure, the latter were bound by three histidine residues (yellow). The crystal structure orientation of the non-selective MMP inhibitor GM-6001 (orange) is shown here as bound to MMP13CD. The two hydroxamic acid oxygen atoms of GM-6001 bind to the catalytic zinc ion, whereas the rest of this inhibitor binds in the primed side (right) of the substrate-binding cleft, toward the S 1 Ј to S 3 Ј subsites. The binding orientation of Compound 1 (colored by atom type: carbon, gray; sulfur, yellow; nitrogen, blue; oxygen, red) in MMP13 shows that it binds deeply and completely within the S 1 Ј-specificity loop. B, the orientation of Compound 2 (orange) bound to MMP13 is superimposed on the structure of Compound 1, oriented as in A. C, close up view of the MMP13 S 1 Ј-specificity loop that envelops Compound 1, highlighting amino acid residue side chains (orange, labeled) that impart potency and specificity. Ligands to the catalytic zinc ion (magenta) are three histidine residues (orange) and acetohydroxamic acid, AcNHOH (green).

DISCUSSION
By mass screening we have discovered an extremely selective inhibitor of MMP13 that is structurally unique in that it does not contain an obvious metal binding group. When we characterized this initial hit and a rationally designed analog  The joints shown here are from animals that displayed responses representative of the means that are graphically depicted in B. In the lower images, which are duplicates of the upper images, the cartilage lesions have been outlined in green for the vehicle (lower left) and Compound 2-treated (lower right) rabbit joints. B, the average twodimensional lesion surface areas of the two treatment groups (vehicle, solid bars; Compound 2, striped bars) are plotted as the means Ϯ S.E. from one study that has been repeated three independent times with similar quantitative results. The erosion areas for both the tibial plateaus and femoral condyles were significantly reduced with Compound 2 treatment (asterisk, p ϭ 0.002 and Ͻ0.001, respectively). by biochemical, structural, and pharmacological methods, we found that these compounds differ from previously reported MMP inhibitors in many other ways. The noncompetitive mechanism of inhibition by Compounds 1 and 2 is consistent with crystallographic data that confirm that these compounds neither occupy the substrate binding cleft nor bind to the catalytic zinc ion. Furthermore, the MMP13 selectivity of these inhibitors is dramatic. Compound 2 is at least 4 orders of magnitude more selective for inhibiting MMP13CD than closely related MMP isoforms, including the collagenases MMP1, -8, and -14 and the distantly related aggrecanases ADAMTS4 and -5, which are also thought to play a role in osteoarthritis.
One recent report indicated that a weak inhibitor of MMP13 (IC 50 ϳ 10 M) that did not contain a zinc binding group and did not inhibit MMP1 or -9 had been identified, but when a metal binding hydroxamic acid functionality was incorporated into this structure to improve potency, the metal-chelating analog showed lower selectivity and also inhibited MMP9 (37). In contrast, MMP13 inhibitor Compound 2 does not bind to the catalytic zinc ion of MMP13, yet it has high inhibitory potency and selectivity, and it effectively protects cartilage without causing joint fibroplasias. Another recent study described some MMP13 inhibitors that are unrelated in structure to Compounds 1 or 2 but did not indicate whether those compounds exhibit any desirable pharmacological activity (38).
The MMP13 inhibitors Compounds 1 and 2 are unlike traditional non-selective MMP inhibitors that bind to the catalytic zinc ion and occupy the substrate binding cleft and the proximal part of the S 1 Ј-specificity pocket. Compounds 1 and 2 bind deeply in the S 1 Ј pocket where they interact with residues in the MMP13 specificity loop to gain remarkable selectivity. These inhibitors induce order to what is normally a flexible specificity loop that is poorly resolved in crystal structures. Based on the strong electron density we observe for the catalytic zinc ion in the crystal structure data, these MMP13 inhibitors clearly do not inhibit by causing zinc to dissociate from the enzyme. The noncompetitive kinetic mechanism of inhibition indicates that Compounds 1 and 2 do not prevent substrate from binding or alter its affinity. Rather, it is possible that these noncompetitive inhibitors block catalysis simply by inducing rigidity to the S 1 Јspecificity loop. Dynamics and movement of this loop may be essential for turnover. Alternatively, movement during the catalytic cycle of the histidine residues that position the catalytic zinc ion as this metal changes from four-to five-and back to four-coordinate during the catalytic cycle may be required for catalysis. By binding in and immobilizing this loop and/or these critical histidine residues, the MMP13 inhibitors may interfere with subtle structural or electronic rearrangements of the catalytic machinery and/or cause electronic perturbations of critical active site moieties which thereby blocks a step(s) in the catalytic cycle.
MMP inhibitors that have been clinically applied so far have been relatively non-selective, potently inhibiting all or nearly all of the known MMPs and even inhibiting more distantly related metalloproteases such as the aggrecanases and the tumor necrosis factor-␣ converting enzyme. Clinical trials of non-selective MMP inhibitors have been plagued by the occurrence of musculoskeletal syndrome joint side effects in subjects (10). This side effect liability may have prevented clinicians from being able to dose such MMP inhibitors high enough to achieve efficacious exposures. Thus, no MMP inhibitor has yet shown safety and efficacy in the clinic. A, the patellar tendon (PT) and distal femur (F) from a vehicle-treated rat are shown. The patellar tendon is composed of a dense, highly stained tendon (T) and a less dense inner synovial lining (L). As can be seen in A and B, no lesions were observed in vehicle or MMP13 inhibitor Compound 2-treated rat joints, respectively, as denoted by comparable thickness of the patellar tendon in A and B (arrowheads). On the other hand, the non-selective hydroxamic acid-containing MMP inhibitor Compound 4 produced marked joint fibroplasias characterized by a dramatic expansion of the inner synovial lining of the patellar tendon (arrowheads) and adjacent quadriceps femoris muscle (M) by immature fibroblasts (C). The white scale bar in each photomicrograph represents 90 m in length.
The MMP13 inhibitor Compound 2 appears to have overcome these two major hurdles that have slowed the clinical development of MMP inhibitors. Compound 2 effectively prevents cartilage degradation in vivo, and it does not induce joint toxicities in an animal model of MSS. With these promising data in hand, we are optimistic that a major hurdle of future clinical trials, to show that such MMP13 inhibitors are devoid of musculoskeletal syndrome side effects, may now be surmountable. Based on the beneficial pharmacology these compounds have displayed in animal models of arthritis, we are hopeful that MMP13-selective inhibitors will effectively protect cartilage in patients without causing the MSS side effects seen previously with non-selective MMP inhibitors. Finally, insofar as aberrant MMP13 activity contributes to other pathologies such as cancer, heart failure, rheumatoid arthritis, and liver fibrosis (39 -43), MMP13 selective inhibitors will be useful in further characterizing the role of this protease in other diseases.