Structural basis of collagen fiber degradation by cathepsin K

Significance

Fibrillar collagens constitute 90% of the organic bone matrix and are subjected either to physiological remodeling or excessive degradation during diseases such as osteoporosis. Cathepsin K is the critical collagenase in bone and represents a major antiresorptive drug target. Despite its critical role in bone remodeling, its mechanism of collagen degradation remained elusive. Here, we demonstrate that the degradation of fibrillar collagen requires the presence of a cathepsin K dimer bound at the surface of collagen fibers via glycosaminoglycans. Structural modifications of the protease dimerization site or the removal of collagen fiber-associated glycosaminoglycans specifically block fibrillar collagen degradation. The provided structure allows the development of a strategy to inhibit this highly relevant drug target in a substrate-specific manner.

Abstract

Cathepsin K is the major collagenolytic protease in bone that facilitates physiological as well as pathological bone degradation. Despite its key role in bone remodeling and for being a highly sought-after drug target for the treatment of osteoporosis, the mechanism of collagen fiber degradation by cathepsin K remained elusive. Here, we report the structure of a collagenolytically active cathepsin K protein dimer. Cathepsin K is organized into elongated C-shaped protease dimers that reveal a putative collagen-binding interface aided by glycosaminoglycans. Molecular modeling of collagen binding to the dimer indicates the participation of nonactive site amino acid residues, Q21 and Q92, in collagen unfolding. Mutations at these sites as well as perturbation of the dimer protein–protein interface completely inhibit cathepsin-K–mediated fiber degradation without affecting the hydrolysis of gelatin or synthetic peptide. Using scanning electron microscopy, we demonstrate the specific binding of cathepsin K at the edge of the fibrillar gap region of collagen fibers, which suggest initial cleavage events at the N- and C-terminal ends of tropocollagen molecules. Edman degradation analysis of collagen fiber degradation products revealed those initial cleavage sites. We propose that one cathepsin K molecule binds to collagen-bound glycosaminoglycans at the gap region and recruits a second protease molecule that provides an unfolding and cleavage mechanism for triple helical collagen. Removal of collagen-associated glycosaminoglycans prevents cathepsin K binding and subsequently fiber hydrolysis. Cathepsin K dimer and glycosaminoglycan binding sites represent novel targeting sites for the development of nonactive site-directed second-generation inhibitors of this important drug target.

Cathepsin K (CatK), a papain-like cysteine protease, is predominantly expressed in osteoclasts and held responsible for the degradation of bone collagen (1, 2). Among the 11 known human cysteine cathepsins, only CatK exerts a triple helical collagen hydrolase activity (3). It is capable of disintegrating compact collagen fibers into fragments and to further solubilize them into soluble peptides (4). Excessive CatK activity in humans is associated with osteoporosis, arthritis, and certain bone cancers (5–7). Knock-out studies in mice reveal an impairment of bone resorption reflected by an osteopetrotic bone phenotype (8). Similarly, human CatK deficiency leads to pycnodysostosis, a skeletal dysplasia characterized by dwarfism, generalized osteosclerosis, dysmorphic appearance, and pathologic fractures (9). The crucial role of this protease in collagen fiber degradation is underlined by the finding that osteoclasts as well as fibroblasts from pycnodysostosis specimens accumulate undigested collagen fibrils in their endosomal–lysosomal compartments (10). These findings identified CatK as a well sought-after antiresorptive drug target for the treatment of osteoporosis (11). Presently, several active site-directed small molecular weight inhibitors are in development, with odanacatib having advanced through phase III clinical trials (12).

Although the pathophysiological role of CatK has been thoroughly characterized, little is known about the mechanism of collagen fiber degradation by this protease. In contrast with collagenases of the matrix metalloprotease (MMP) family (13, 14), CatK lacks any protein domains implicated in collagen unwinding or specific binding. Structurally, CatK is divided into the left (L) and right (R) domains linked at its center by an interdomain 2-strands beta sheet, creating a V-shaped cleft carrying the active site cysteine (C25) and histidine (H159) residues (15). These structural features are widely conserved among all papain-like cysteine proteases (16) and do not provide mechanistic insights into the collagenolytic activity of CatK. The entrance into the active site of CatK and other cysteine cathepsins is about 5 Å wide and thus too small to accommodate 15-Å-wide triple helical collagen molecules.

We have previously shown that the unique and potent collagenase activity of CatK requires the presence of glycosaminoglycans (GAGs) with which it forms high molecular complexes. In the absence of GAGs, CatK only exerts a gelatinase and telopeptide cleaving activity similar to that of other cathepsins (17). This has fueled speculations about a possible role of GAG–CatK complexes in promoting triple helical collagen unfolding by CatK similar to the role the hemopexin domain plays in the hydrolysis of collagens by MMPs (13, 14). Our recently published structure of a complex between E64-inhibited CatK and chondroitin 4-sulfate (C4-S) revealed a “beads-on-a-strand”–like binding of multiple CatK molecules to a single GAG chain (18). However, analysis of the collagenase activity of CatK at the stoichiometric concentrations required to form a beads on a strand–like structure, did not display a significant collagenase activity (17), suggesting that different cathepsin K–GAG complexes must be responsible for collagen degradation.

By using X-ray crystallography, mutagenesis, molecular modeling, and electron microscopy, we identified a unique CatK dimer with previously unidentified GAG binding sites and a specific binding site on collagen fibers that reveal a mechanism for the unique activity of this protease.

Results

Structures of CatK–GAG Complexes.

Chondroitin 4-sulfate (C4-S) and dermatan sulfate (DS) are the most common GAGs associated with collagen fibers in bone (19). Here, we crystallized CatK bound individually to DS and C4-S in the absence of an active site inhibitor. The structures have been refined to 2.62 Å (DS) and 2.02 Å (C4-S) with data collection and refinement statistics summarized in Table S1. The overall structure of CatK in both complexes is highly similar to previously published structures (18, 20). Superposing these structures on CatK-E64 C4-S structure [Protein Data Bank (PDB) ID code 3C9E] shows a CatK rmsd of (0.373) for the DS and (0.311) for the C4-S complex. Fig. 1A displays the presence of three major protein–GAG interaction sites. The first GAG chain, G1, binding site is located at the bottom of the R domain (opposing the active site area) corresponding to a surface rich in positively charged residues including K119, K122, R123, and R127 (Fig. 1 B and C). A twofold axis passes adjacent to and parallel to the GAG chain thereby placing a second GAG chain, G2, parallel to the first at a distance of 7.5 Å, creating symmetry-related GAG dimers (Fig. 1A). This symmetry-related GAG chain presents a second unique CatK–GAG contact through its interaction with K40, K41, R108, R111, and K214 (Fig. 1 B and C). Interestingly, R108, R111, K122, R127, and K214 are unique to CatK among cysteine cathepsins (21). A third binding site is seen via the interaction of the G2 chain with Q92 of the adjacent molecule thereby identifying a CatK dimer (Fig. 1A). This dimer (molecules A and B in Fig. 1A) has the l domain of molecule A interacting via G2 with the R domain of an inversely oriented molecule B. The dimer is further stabilized via protein–protein contacts comprising residues G174, N175, K176, N199, and K200 from molecule A and residues P88, D85, N99, and T101 from molecule B (Fig. 2). GAG conformations and their interacting residues along with their contact distances are shown in Fig. 2 B and C.

Fig. 1.

Interactions of CatK with GAG chains. (A) Cartoon representation of CatK dimer (molecules A and B in yellow) with the GAG chain G1 and symmetry-related G2 (green stick). Molecules C and D (magenta) show the relative packing of CatK dimers in the crystal. The star (black) represents the location of a twofold axis in our crystal packing. (B) Surface representation of the highlighted region in A. Location of G1 and G2 and their interaction sites on CatK (colored orange, G1 and gray, G2) compared with previously published C4-S binding site (18) (blue surface color and C4-S chain in magenta). (C) A 90° rotation of (B).

Fig. 2.

CatK dimer interface and GAG binding sites. (A) Transparent electrostatic charge with underlying line representation of the dimer interface is shown with interacting residues from molecules A and B in stick representation (in yellow). Residues mutated in IFM1 are shown in magenta. The GAG chain is shown as green sticks. (B) Stick representation of DS interacting residues. Residues of CatK are shown in yellow and symmetry-related residues are cyan. DS chain shown in green sticks adopts a linear conformation. ASG6 is the only DS residue not binding to CatK. (C) Stick representation of CatK interactions with C4-S (light gray). All interacting residues are labeled and within 3.6 Å of the GAG chain. GCU6 shows no binding to CatK. (D) Stick representation of GAG chains (colors as in B and C) showing a slightly more curved C4-S compared with DS.

GAG Structures.

The C4-S and DS GAGs in our structures both reveal a linear orientation of the sugar chain in contrast with the previously published cosine-curved shape of C4-S (18, 20). We speculate that the conformation of the GAG chain is determined by adopting an optimal fit into the binding site it docks into, as has been previously proposed (22). Superposing the DS structure on the C4-S shows the same spatial orientation with slight differences arising from the bend on the C4-S chain compared with the DS chain (Fig. 2D). There were protein–sugar chain interactions in all but the last N-acetyl-galactosamine (ASG) of the DS and the last glucoronic acid (GCU) of the C4-S chains, respectively.

CatK Dimer Formation, Collagen Docking, and Collagen Unwinding.

The interactions of the GAGs with CatK presented in these structures reveal a dimeric organization of the enzyme held together by protein–protein as well as protein–GAG contacts. To identify the mechanism of collagen digestion, we first built a CatK dimer based on unique contacts with a single GAG chain in the crystal. This dimer revealed an elongated C-shaped assembly with the GAG chain at its center perpendicular to its longitudinal axis. Its dimensions of 9 nm × 4 nm × 3 nm reveal a groove running parallel to the GAG chain on the back of the first CatK protomer and adjacent to the active site of the second molecule (Fig. 3A). We docked a tropocollagen into this groove and analyzed the potential interactions with the GAG–CatK dimer assembly. We observed that individual chains of the triple helical collagen interact with distinctly different residues in the complex: one chain with Q21, another with Q92, and the third chain binds the GAG sugar chain. The GAG chain runs alongside and slightly rotated at a 15° angle to the long axis of collagen. This rotation indicates that an extended GAG chain would run down the length of the tropocollagen, either twirling around it or crossing over to neighboring triple helices in the fibril. Both GAG binding modes have been demonstrated via biochemical means and by electron microscopy (23–25). The CatK dimer–GAG complex interaction with the collagen sheds light on a potential mechanism of collagen uncoiling and cleaving. It is tempting to speculate that Q21 and Q92 in concert with the interacting GAG molecule participate in dividing the interacting collagen chains and thus allowing their proteolytic cleavage by CatK (Fig. 3 A and B).

Fig. 3.

CatK dimer–tropocollagen model. (A) Cartoon representation of CatK dimer (yellow) with bound GAG chain G1 (green stick). Residues Q21 and Q92 are shown in red and C25 in black stick. A docked tropocollagen chain (cyan, purple, and orange) is shown in surface representation sandwiched between the GAG, Q21, and Q92. (B) Schematic representation of the proposed mechanism of collagen degradation by CatK dimers. CatK dimer binds to a collagen–GAG complex which is unfolded by Q21 and Q92 before being cleaved by C25. Color codings as in A.

Fibrillar Collagen Degradation by CatK.

To determine the possible binding mode of CatK to a collagen fiber, we superimposed the tropocollagen in our CatK–GAG–tropocollagen docking experiment onto the collagen fiber diffraction structure (PDB ID code 3HQV). The only position where a CatK dimer would be sterically able to assemble is at the edges of the gap regions of the fibril. We superimposed the CatK dimer with a GAG chain bound to the G1 binding surface on each protomer into this gap area. This displayed a perfect fit where the dimer had one tropocollagen located at its active site and another positioned right next to the G1 chain of the second CatK protomer (Fig. 4A). Using scanning electron microscopy (SEM), we confirmed this binding site of CatK to fibrillar collagen. SEM images revealed a very specific distribution of immunolabeled CatK molecules on intact collagen fibrils (Fig. 4B). The majority of labels were found at sites where the N- and C termini of tropocollagen molecules point into the gap region (Fig. 4C). To verify that the binding of CatK to the collagen fiber is mediated via GAG molecules, fibers were pretreated with chondroitinase ABC, which led to over 90% reduction of GAG and CatK immunolabeling and an abrogation of the hydrolysis of the collagen fiber (Figs. 4 B and C and 5C). Primary cleavage sites in both α1- and α2 chains of triple helical collagen were identified by the Edman degradation method. For this purpose we used both insoluble and soluble collagen subjected to time-dependent degradation. These results indicate that initial cleavage sites are located in the vicinity of two intertropocollagen cross-linking sites at the N-terminal region of the helix (Fig. 4D). Together with previously reported C-terminally located CatK cleavage site (26), this would explain the release of the 95–120-kDa tropocollagen fragments as seen by SDS/PAGE analysis (Fig. 5A).

Fig. 4.

Binding of CatK dimers near the gap region of collagen fibrils. (A) Schematic representation of CatK dimer (yellow) superposed on a collagen fibril (blue) in the gap region. The interaction is magnified and shown in yellow cartoon (CatK), collagen in blue surface representation, and GAG green stick representation. A 90° rotation showing a perfect fit of the G1 of both CatK protomers on two adjacent tropocollagen molecules in the fibril is also shown. (B) Immunolabeling with polyclonal antibody for CatK and monoclonal antibody for chondroitin sulfate highlights their binding sites on the edges of the gap region on collagen fibrils. Pretreating with chondroitinase ABC drastically inhibits the binding of both C4-S and CatK. Magnification bars, 50 µm. (C) Magnification of CatK immunolabeling on collagen fibrils including the schematic representation of the location of tropocollagen molecules and quantitative analysis of the binding location of cathepsin prior and after chondroitinase ABC treatment. (D) Primary CatK cleavage sites within mouse collagen fibers at the N-terminal helical region of α1 and α2 collagen chains.

Fig. 5.

Effect of CatK variants on solubilization of insoluble collagen fibers. (A) SDS/PAGE analysis of the release of α-chains from insoluble collagen fibers by CatK variants [Q21A, Q92A, Q21A/Q92A, and interface mutant 1 (IFM1)] after 4 h of incubation. All variants show potent inhibition of collagen fiber solubilization compared with wild-type CatK (the protein band below 70 kDa represents a keratin contamination from mouse tails). (B) A bar chart analysis showing inhibition of release of α-chains from insoluble collagen. Q21A/Q92A and its single mutants showed an 80% inhibition whereas IFM1 revealed a >95% inhibition. (C) Electron micrographs showing the effect of IFM1 and Q21A/Q92A variants on collagenolysis after overnight incubation. IFM1 treatment left the collagen fiber intact, whereas there was increased diameter of the fiber following partial unfolding and limited collagen breakdown in the presence of the Q21A/Q92A variant. In comparison, wild-type CatK treatment resulted in complete breakdown of the fiber. Magnification bars, 50 µm. Removal of GAGs by chondroitinase ABC also led to an inhibition of fiber disintegration.

Mutational Analysis of CatK Dimer–GAG–Collagen Interaction Sites.

To verify our dimer collagenase model, we constructed four CatK variants. In the first, we mutated the protein–protein interaction site in the A molecule of the dimer into a collagenolytically inactive cathepsin L analog sequence (interface mutant, IFM1). In the second to fourth mutant proteins, residues Q21 and Q92 were replaced by alanine individually and in combination (Table S2). The analogous IFM1 site in human cathepsin L is 8 amino acids long and thus 4 residues longer than in CatK. All variants were analyzed for collagen fiber degradation and noncollagenolytic activity. Compared with wild-type CatK, SDS/PAGE analysis displayed a >95% and ∼80% loss of the collagenase activity for the protein–protein interface variant (IFM1) and the Q21A, Q92A, and Q21A/Q92A variants, respectively (Fig. 5 A and B). The loss of collagenase activity is defined by the lack or reduced release of tropocollagen-sized fragments from intact fibers. SEM analysis of collagen fibers incubated with wild-type CatK revealed their complete disintegration, whereas the IFM1 variant left the fiber intact (Fig. 5 B and C). The treatment of collagen fibers with the Q21A/Q92A double mutant variants resulted in a partial disintegration of the fibrils (Fig. 5C). We speculate that alanine residues in positions 21 and 92 were effective in the structural modification of the collagen triple helix, but too short to effectively mediate its degradation. As the single mutant variants behave like the double mutant protein, we can conclude that both sites are equally required for the collagenase activity of CatK. Mutant variants caused an increase in the collagen fiber diameter (50–75%) likely due to the removal of proteoglycans (4). To exclude the possibility that mutations have compromised the active site of the variant proteins, the Michaelis–Menten kinetic parameters were determined. Specificity constants k_cat/K_m, as well as the individual kinetic parameters k_cat and K_m, remained similar to those of wild-type CatK, indicating that the catalytic machinery of the active site was not affected by the mutations (Table 1). All mutation sites are distant from the catalytic C/H diad. The residual activities of both variants and the wild-type protease were between 25% and 30% after 4 h of fiber digestion, indicating comparable protease stabilities. Moreover, degradation assays using gelatin (heat-denatured collagen) revealed no inhibition by either variant compared with wild-type CatK, indicating that the mutation specifically affected the degradation of intact collagen (Fig. S1).

Table 1.

Kinetic values for the peptidase activity of CatK and its variants

Enzyme	kcat, s−1*	Km, µM*	kcat/Km, 106 s−1⋅M−1*
Wild-type CatK	19.5 ± 1.0	7.5 ± 1.0	2.6
Gln21Ala	23.4 ± 0.6	8.2 ± 0.5	2.9
Gln92Ala	16.7 ± 0.9	7.7 ± 1.0	2.2
Gln21Ala/Gln92Ala	16.7 ± 0.8	8.3 ± 1.0	2.0
IFM1	20.4 ± 3.1	5.5 ± 2.4	3.7

^*The constants K_m, k_cat, and K_cat/K_m were determined according to Michaelis–Menten kinetics with procedures described in SI Materials and Methods. Varying concentrations of Z-FR-MCA (carbobenzoxy-phenylalanine-arginine-methyl coumarin amide) as fluorogenic substrate were used.

Discussion

CatK, the predominant collagen-degrading cysteine protease in bone-degrading osteoclasts, has been the subject of extensive biochemical and pharmaceutical studies. Presently, active site inhibitors are in clinical testing where odanacatib has recently passed a large-scale phase III osteoporosis trial. The 3D structure of CatK is rather simple, having an R- and l domain with a central cleft accommodating the active site (16). This structure is highly conserved within the cysteine cathepsin family where, however, CatK is the only member exerting a collagenase activity (3). Furthermore, it lacks other domains, which have been identified as collagen binding sites and domains that have been implicated in a triple helical unfolding mechanism (13). Therefore, how CatK cleaves collagen fibers remained enigmatic.

We have previously shown that the collagenase activity of CatK is dependent on the presence of GAGs (17, 18, 27, 28), but how these polysaccharides facilitate collagen degradation remained elusive. However, a previously determined structure of a CatK–C4-S complex with a beads-on-a-strand–like organization provided little insight into a mechanism as maximal collagenolytic activity was obtained at rather equimolar CatK–GAG ratios instead of at molar excesses of CatK as predicted by the structure (17). A reduction of the GAG content significantly below the protease concentration led to a complete inhibition of the collagenase activity (17). Using ∼8.5-kDa C4-S and DS preparations, we obtained two very similar CatK–GAG complex structures, which differed from the previously published complex structure (18). The earlier reported GAG-binding residues K9, I171, and Q172 are all buried in the protein–protein interfaces of the present structures and residues N190, K191, and L195 are distant from the novel GAG binding sites (18). Two long-stretched GAG binding sites were identified, which include for the G1 GAG molecule the basic amino acid residues K119, K122, R123, and R127 and for the G2 molecule, residues K40, K41, R108, R111, R127, and K214. An analysis of species conserved CatK specific surface arginine and lysine residues demonstrated that R108, R111, K122, and R127 were important for the collagenolytic activity of CatK (21). Interestingly, a study combining modeling, binding kinetics, and cross-linking experiments has also suggested additional GAG binding sites. One site contained cross-links to K10, K39, and K77, whereas the other site was modeled into a positively charged area containing residues K40, K41, R108, R111, R127, and K214 (29). The latter putative binding site resembles the binding of the G2 chain in our X-ray structures.

Of particular interest is the GAG molecule G1 that interacts with two neighboring CatK molecules. Both protease molecules have close protein–protein contacts suggesting the presence of a functional dimer. This dimer is C-shaped with the GAG chain at its center and apposed to the active site cleft of CatK molecule B. Here, two CatK molecules bind to each other in a head-to-tail organization, whereas in the previously solved CatK–GAG complex a dimer formation revealed a head-to-head organization (20). To postulate a mechanism of collagen digestion, we built a CatK dimer based on unique contacts with a single GAG chain and docked a tropocollagen into its central groove (Fig. 3A). Space constraints in a fibril arising from highly intertwined collagen molecules along with an extensive intermolecular and GAG cross-linking will generally prevent the assembly of a CatK dimer within it. To access the tightly packed tropocollagens in microfibrils, it is generally expected that collagenases must bind and initiate digestion from the surface of the fibril (30). This is also supported by our earlier findings that type II collagen fibers in cartilage specimens are getting thinner in the presence of CatK (17). We have demonstrated that this CatK–GAG assembly can only take place adjacent to the gap areas on a collagen fibril (Fig. 4A). Consequently, digestion of collagen would start at the N- or C terminus of a tropocollagen molecule. This has been demonstrated using both electron microscopy (Fig. 4C) and N-terminal sequencing in this study (Fig. 4D) and documented for C-terminal cleavage sites previously (26, 31). The binding in the gap region must be facilitated by collagen-associated GAG molecules. Their enzymatic removal prevents CatK binding to the collagen fibers as well as their degradation (Figs. 4 B and C and 5C). We propose that one CatK protomer binds via its GAG binding site to a GAG-bound collagen molecule on the surface of the fibril. This CatK–GAG complex serves as an anchor allowing the docking of a second protease molecule via Q92 to the GAG chain and stabilized by the dimer–interface interactions. The triple helical collagen molecule then gets trapped between Q21 and Q92, potentially forcing the unwinding of the triple helix before cleavage (Fig. 3B). The functionality of these residues was supported by mutagenesis analysis, where we showed an 80% reduction in collagen fiber degradation activity consistent with our model. Both residues are equally important as the mutations at both sites resulted in the same loss of collagenase activity. This was further corroborated by the lack of fiber disintegration in the presence of this variant in contrast with the complete destruction of the fiber by the wild-type enzyme (Fig. 5C). Further evidence in support of this model is the proteolytically active but collagenolytically inactive variant IFM1 that potentially disrupts enzyme dimerization.

This oligomerization of CatK and GAG molecules represents an evolutionary alternative to MMPs, which are characterized by multidomain structures including a hemopexin-like and tandem C-terminal collagen binding domains (32, 33). These domains have been shown to anchor the collagenases on tropocollagen molecules and provide an unfolding mechanism, whereas an N-terminal catalytic domain cleaves the triple helix (13). Bacterial collagenases appear to use yet another mechanism, although with some analogy to matrix metalloproteases. Collagenase G from Clostridium histolyticum has a multidomain structure consisting of a saddle-shaped two-domain collagenase module and protein recognition motifs (34). The protein recognition domain is thought to bind to the collagen microfibril and the saddle domains are responsible for the unfolding and subsequent cleavage of the triple helices. In the case of CatK, GAGs anchoring to the cysteine protease take over the role of the collagen-binding domain, whereas the C-shaped CatK dimer would represent the functional analogy to the saddle-shaped collagenase domain of the Clostridium protease.

In conclusion, we identified a CatK dimer associated with collagen-bound GAGs, which acts as the active collagenase form and provides a mechanism of collagen fiber degradation by this pharmaceutically important protease. Fig. 6 illustrates the proposed mechanism of collagen fiber degradation in three steps: (i) binding of CatK monomers to GAGs associated with collagen fibrils. (ii) Recruitment of a second CatK molecule to form dimers at the edges of the gap regions of fibrils. (iii) Cleavage at the N- and C termini of surface tropocollagen molecules and their release from the fibril. These tropocollagen fragments are subsequently further degraded by CatK.

Fig. 6.

Schematic representation of collagen fiber degradation by CatK dimers. Collagen fiber degradation occurs in three steps: (i) binding of first CatK molecule to collagen-associated GAG in gap region; (ii) recruitment of second CatK molecule to form collagenolytically active dimer; (iii) cleavage in either the N- or C-terminal region of tropocollagen molecules leading to the release of 95–100-kDa collagen α-chain fragments.

The dimerization as well as GAG interaction sites represent novel inhibitor binding sites, which would selectively block the collagenase activity of CatK without interfering with other proteolytic activities of the protease. We have recently demonstrated this concept by showing that the binding of dihydrotanshinone at exosite 1 of CatK selectively inhibits the collagenase and elastase activity of CatK (35). Exosite I partially overlaps with the protein dimer interaction site described in this report. This approach may exert an advantage over active-site inhibitors of CatK (36), as the therapeutic blocking of the hydrolysis of noncollagen substrates may give rise to adverse side effects as recently suggested (7) and corroborated by the finding that CatK-deficient mice show severe cognitive malfunction (37).

Materials and Methods

Recombinant CatK and variants were expressed in Pichia pastoris and purified as previously described (38). The purification of GAGs followed the procedure previously described in ref. 39. Details about the enzymatic assays are detailed in SI Materials and Methods. Crystals of CaK-GAG complexes were obtained by vapor diffusion method at room temperature after mixing equal volumes of protein (6 mg/mL in 50 mM acetate pH 5.0 and 2.5 mM dithiothreitol) and buffer (24% 2-methyl-2,4-pentanediol, 100 mM sodium acetate pH 4.5, and 100 mM calcium chloride). Further details are provided in SI Materials and Methods.

The CatK-GAG structures were determined via molecular replacement using PDB ID code 3C9E as a search model. The CatK collagen models were built as described earlier using PDB ID codes 3AH9 and 3HQV as docking models. Details are provided in SI Materials and Methods.