The 1.8-Å Crystal Structure of a Matrix Metalloproteinase 8-Barbiturate Inhibitor Complex Reveals a Previously Unobserved Mechanism for Collagenase Substrate Recognition*

, The individual zinc endoproteinases of the tissue de-grading matrix metalloproteinase (MMP) family share a common catalytic architecture but are differentiated with respect to substrate specificity, localization, and activation. Variation in domain structure and more subtle structural differences control their characteristic specificity profiles for substrates from among four distinct classes (Nagase, H., and Woessner, J. F. J. (1999) J. Biol. Chem. 274, 21491–21494). Exploitation of these differences may be decisive for the design of anticancer or other drugs, which should be highly selective for their particular MMP targets. Based on the 1.8-Å crystal structure of human neutrophil collagenase (MMP-8) in complex with an active site-directed inhibitor (RO200-1770), we identify and describe new structural determinants for substrate and inhibitor recognition in addition to the primary substrate recognition sites. RO200-1770 induces a major rearrangement at a position relevant to substrate recognition near the MMP-8 active site (Ala 206 –Asn 218 ). In stromelysin (MMP-3), competing stabilizing interactions at the analogous segment hinder a similar rearrangement,

The matrix metalloproteinases (MMPs), 1 one of the five families that form the metzincin group of zinc proteinases (3), function to degrade the extracellular matrix during embryonic development, reproduction, and tissue remodeling (1) but are disregulated in arthritis, cancer, and other diseases. The "minimal" MMPs matrilysin and endometase (MMP-7 and MMP-26, respectively), have a Zn 2ϩ and Ca 2ϩ binding catalytic domain, and an N-terminal pro-domain. All other known MMPs possess additionally a hemopexin-like domain near the C terminus, and further domain insertions differentiate MMP subfamilies. Gelatinases A and B (MMP-2 and MMP-9) possess three fibronectin type II-like repeats inserted at a loop in the catalytic domain; these form an independent folding domain adjacent to the catalytic domain. Membrane-type MMPs possess an anchoring transmembrane helix C-terminal to the hemopexinlike domain (4). Hierarchical regulation of MMP activity occurs on many levels, including gene expression control (1,5), proteolytic activation of MMP zymogens (6), inhibition by endogenous tissue inhibitors of metalloproteinases (7), and both positive and negative proteolytic feedback loops (8,9). Crystal structures of several MMPs have been determined (for a review, see, e.g., Ref. 10), revealing overall domain structures, catalytic mechanisms, and many aspects of MMP regulation mechanisms; these include collagenase 1 (MMP-1) (11,12) and collagenase 2 (MMP-8). Structures of the latter are represented by two forms of the catalytic domain, resulting from activation cleavage alternately at two cleavage sites, leaving either Met 80 (13,14) or Phe 79 as the N-terminal residue (15). The latter form is "superactivated," as Phe 79 forms a salt bridge with Asp 232 and thereby prevents the N-terminal sequence from transient or other interference with the active site. The result is a 3-fold increase in activity compared with activation cleavage at Met 80 (16).
As their early nomenclature implies, collagenases I, -II, and -III (17) (MMP-1, -8, and -13, respectively) degrade mainly fibrillar collagens (18 -20), although the structural origin of this specificity is not well understood (4). Disruption of MMP tissue remodeling function causes a variety of disorders, including cancer (tumor growth, invasion and metastasis), rheumatoid arthritis and osteoarthritis, and a variety of diseases involving neovascularization. The resulting clinical need has fostered an enormous interest in the development of inhibitors against MMPs. As part of these efforts, crystal structures of MMPs with a variety of synthetic inhibitors have been determined. For MMP-8, complexes reported include peptide mimet-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
We dedicate this work to Prof. H. Tschesche on the occasion of his 65th birthday.
Here we describe the 1.8-Å crystal structure of MMP-8 inhibited by a barbituric acid derivative. Conformational rearrangements accompanying the inhibitor binding lead to a new and highly ordered crystal packing arrangement. The high resolution structural data enables a thorough analysis of determinants of MMP selectivity toward both low molecular weight substances as well as substrate classes. A previously unreported cis-peptide bond (Asn 188 -Tyr 189 ) could be unambiguously identified. The conservation patterns of the sequence at the cis-peptide bond position support the hypothesis that this cis-peptide plays a critical role in substrate recognition mechanisms specific to the collagenases I and II (MMP-1 and MMP-8).
Inhibition Assay-All measurements were performed at 25°C using a buffering solution of 50 mM Tris, pH 7.6, 100 mM NaCl, 10 mM CaCl 2 . Based on multiple measurements, all data are precise to within 5%. Depending on activity, enzymes were used at 5-50 nM concentration range with a substrate concentration of 2.55 M. The enzyme was briefly pre-incubated with the inhibitor at a resultant Me 2 SO concentration of 1%. The reaction was started with the addition of the substrate M-1855. Substrate was excited at 280 nm and the substrate fluorescence was monitored at 346 nm using the FuoroMax-3 fluorometer (SPEX, Horiba Group, Grasbrunn/Munich, Germany).
Crystallization, Data Collection, and Structure Refinement-MMP-8 was concentrated to 8 mg/ml and then mixed with 3-fold molar excess of an aqueous solution of RO200-1770 for a final MMP-8 concentration of 6 mg/ml. 3 l of protein-inhibitor complex was mixed with 2 l precipitant solution containing 100 mM cacodylate pH 5.5-6.5, 10 mM CaCl 2 , 100 mM NaCl, and 10% polyethylene glycol 6000. The hanging drop was equilibrated by vapor diffusion at room temperature against a reservoir containing 1.0 -1.5 M phosphate buffer. Data were collected on a multiwire detector (X1000, Bruker AXS) to 1.8-Å resolution and processed using SAINT data reduction software (29), yielding an agreement of redundant measurements of R merge ϭ 9.3% over all data and 41% in the outer resolution shell (completeness 98% and 87%, respectively). The space group of the crystal was determined as I 222 with unit cell dimensions a ϭ 61.02 Å, b ϭ 69.24 Å, c ϭ 88.47 Å. The orientation and translation of the molecule within the crystallographic unit cell was determined with Patterson search techniques (30 -32) using the program AMoRe (33). Electron density calculation and model building proceeded using the program MAIN (34). The structure has been refined by using the program X-PLOR (35) to a crystallographic R-value of 21.1% (R free ϭ 29.6%) with bond deviations of 0.009 Å and angle deviations of 1.7°from ideality (36). The molecular structure was analyzed and compared using appropriate tools within the program MAIN (34).

RESULTS
Inhibitor Conformation-The inhibitor RO200-1770 is a barbituric acid derivative, doubly substituted with phenyl and 4-ethanolpiperidyl rings as depicted in Fig. 1. The barbiturate ring chelates the zinc and rigidly orients the two cyclic substituents into the S1Ј and S2Ј substrate binding sites. Neither substituent ring system appears strained by the protein environment, although their relative orientations may be induced by protein binding. The phenyl moiety occupies the MMP-8 binding site and is perfectly planar to within the 1.8-Å resolu-tion. The electron density observed for the piperidine ring allows an interpretation whereby two chair conformations related by a 180°rotation along the C 5 -pN 1 bond might be superimposed; either conformation would allow favorable hydrophobic contacts in the S2Ј-site. Adopting an all-trans conformation, the alcohol group points toward the solvent.
The relative orientations of the rings of the inhibitor may be described by considering the ring planes and the bonds linking the substituent rings to the C 5 atom of the barbiturate ring. The C 5 -fC 1 bond linking the phenyl ring is nearly perpendicular to the plane of the barbiturate ring (excluding C 5 ). This arrangement necessarily orients the plane of the phenyl ring likewise perpendicular to the barbiturate plane. The dihedral angle C 6 -C 5 -fC 1 -fC 2 fixes the ring orientation with an eclipsed geometry (at 1.2°, while the C 4 -C 5 -C 1 -fC 2 dihedral is staggered at 60.8°). In contrast, the C 5 -pN 1 bond lies nearly in the plane of the barbiturate, extending the P2Ј-piperidyl ring away from the barbiturate; all dihedrals across the C 5 -pN 1 bond have staggered orientations. This results in an arrangement where all three rings are mutually perpendicular, as follows: the angle between (the normal vectors) of the barbiturate and phenyl rings is 91°, between the barbiturate and the piperidyl rings is 103°, and between the phenyl and piperidyl rings is 111°. highlights the inhibitor binding at the "primed" substrate recognition sites and at the Zn 2ϩ ion. The Zn 2ϩ is coordinated by atoms N 3 and O 2 of the barbiturate ring. The Zn 2ϩ -N 3 coordination has a favorable distance of 2.09 Å and highly symmetric Zn 2ϩ -N 3 -C 2 and Zn 2ϩ -N 3 -C 4 angles of 119°and 117°, respectively. Positioned where the catalytic water is expected for peptidic substrates, a partial negative charge at the hydroxyl O 2 is stabilized by the adjacent Glu 198 , thereby strengthening its binding to Zn 2ϩ (Figs. 2 and 3, Table II). The enol form of the barbiturate is thus favored by the protein matrix over the tautomeric keto form, which dominates in solution (37). In addition, the polar H 1 -N 1 -C 6 ϭO 6 atoms of the barbiturate (Fig.  3B) mimic the P1Ј-S1Ј antiparallel main chain interactions of a substrate (Fig. 3A). The amide N 1 -H 1 thereby is hydrogen bonded to the carbonyl of Ala 161 , and the ketone C 6 ϭO 6 is stabilized by the amides of Leu 160 and Ala 161 . This latter interaction is reminiscent of the oxyanion hole binding of serine proteinases, although here there is no evidence of oxygen anion stabilization. The C 4 ϭO 4 ketone seems unlikely to contribute to the binding energy for two reasons. First, unfavorable geometry (Table I) precludes a role as a third ligand in the Zn 2ϩ chelation. More importantly, the C 4 ϭO 4 ketone would collide with the carbonyl oxygen position of Pro 217 at the "southern" rim of the active site as defined by other MMP-8 structures (2,14,26). Instead, the inhibitor induces a reorientation at the Pro 217 position at an energy cost we discuss below.
The pentacoordinated Zn 2ϩ binding geometry resembles a highly distorted trigonal bipyramidal structure with O 2 , N ⑀2 (His 197 ), and N ⑀2 (His 207 ) approximately in plane with the Zn 2ϩ ion, with N ⑀2 (His 201 ) and N 3 lying above and below the basal plane, respectively (Table I). Alternatively, the coordination can be described as a distorted square pyramid where O 2 , N 3 , N ⑀2 (His 201 ), and N ⑀2 (His 207 ) form the basal ligands (dihedral deviation from planarity 15°, Table I). The metal ion lies outside of the basal plane but within 0.5 Å, and the fifth ligand N ⑀2 (His 197 ) forms the apex of the pyramid Table I. (If considering O 4 to be a sixth ligand, the geometry may be described as a pentagonal pyramid with O 4 as basal ligand in addition to O 2 , N 3 , N ⑀2 (His 201 ), and N ⑀2 (His 207 )).
In contrast to the polar interactions of the barbiturate ring, the interactions mediated by the phenyl and piperidyl rings are predominantly hydrophobic and involve the S1Ј and S2Ј pockets, respectively. The most prominent interaction in the S1Ј pocket is the ideally parallel planar stacking of the phenyl ring and His 197 at a distance of 3.6 Å (Fig. 4). The conserved Leu 160 contributes to ligand binding also with its side chain in the S1Ј site. The phenyl ring does not by itself fill the S1Ј site, but leaves space filled by a network of three ordered water mole-cules. The first of these (Sol 595 ) is probably incompletely occupied and forms hydrogen bonds with the inhibitor, with MMP-8, and with a second water molecule. The proximity of the inhibitor phenyl fC 4 atom to Sol 595 (3.1 Å) indicates a O . . . H-C interaction (38). The carbonyl group of Leu 193 forms a 2.9-Å hydrogen bond with Sol 595 with, however, an unfavorable CϭO 193 -O 595 angle of 113°. The second water molecule, Sol 602 , is positioned deeper inside the S1Ј pocket at a hydrogen bonding distance of 2.7 Å from Sol 595 . Sol 602 in turn is hydrogenbonded (2.8 Å) with the third solvent molecule in the S1Ј pocket, Sol 592 . Sol 592 is in a channel bounded by Arg 222 , which forms a link between the three water network in S1Ј and, via Sol 667 (2.9 Å from Sol 602 ), water in the adjacent cavity. Muta- tion of Arg 222 would connect the two cavities, opening a "back door" to S1Ј for solvent access. The guanidinium group atoms N 2 and N 1 of Arg 222 are fixed by hydrogen bonds to the carbonyl oxygen of Ala 213 (3.3 Å) and the O of Tyr 227 (3.3 Å), respectively. Since most MMPs lack an equivalently stabilized Arg, MMP-8 has a comparatively restricted S1Ј site.
The hydrophobic interactions of the piperidine ring are mediated by aliphatic surfaces from Pro 217 -Asn 218 -Tyr 219 at the "southern" rim of S2Ј and by the main chain Gly 158 -Ile 159 -Leu 160 at the "northern" rim. The latter residue (Leu 160 ) separates the S1Ј and S2Ј pockets. No ordered water molecule can be detected in the vicinity of the hydroxyl group pOH 9 , although the position of this solvent exposed ethanol group is well defined by electron density (Fig. 2) and is thus presumably hydrated by disordered water.
Protein Conformational Changes-Significant differences are apparent in the protein structure compared with previously determined MMP-8 structures (2,14,21). The catalytic Zn 2ϩ ion of the three reference structures occupies the same position to within 0.2 Å; it is, however, shifted from that average position by 0.6 Å in the RO200-1770 complex structure. Corresponding shifts of the Zn 2ϩ protein ligand positions are also apparent, with the respective N ⑀2 and C ␥ values measured as follows: His 197 (0.4 Å, 0.3 Å), His 201 (0.3 Å, 0.3 Å), and His 207 (0.6 Å, 0.7 Å). Consistent with this overall shift, the side chain of the catalytic Glu 198 is translated by 0.2 Å. This displacement of the catalytic Zn 2ϩ and its protein ligands is evidently induced by inhibitor binding, as the net effect of the optimization of barbiturate-Zn chelation geometry and the inhibitor orienting forces arising from the other inhibitor-protein interactions.
Of the two partial sequences harboring the Zn 2ϩ binding histidine residues, the loop Ala 206 -His 207 -Asn 218 is more exposed to the solvent and anchored by fewer protein contacts than the internal helix L 191 -H 197 EXXH 201 -L 203 . The conformation of this loop is altered by several effects associated with the binding of RO200-1770. First, the greater inherent plasticity of this loop leads to greater compensation by the Zn ligand His207 for shear stresses induced at the catalytic site. Second, the Pro 217 -Asn 218 peptide bond is rotated by ϳ100°, evidently to prevent a repulsive interaction between the barbiturate C 4 ϭO 4 keto group with the Pro 217 carbonyl. Third, residues Ser 209 , Tyr 216 , Pro 217 , and Asn 218 form crystal contacts. These effects in combination lead to a translation along the entire loop from Ala 206 to Pro 217 , which, however, is relatively rigid, leaving most dihedral angles similar to those in the reference structures. In the "north" rim of the active site, the largest change compared with the inhibitor free MMP-8 structure is a 0.98 Å displacement and disorder of the Ile 159 side chain; the electron density shows a branched but symmetric side chain interpretable as two equally populated rotamers, which "swap" C ␥1 and C ␥2 positions.
Enzyme Inhibition Analysis-Utilizing the crystal structure of the MMP8-RO200-1770 complex, several follow-up compounds were synthesized and tested against the panel of metalloenzymes shown in Table II. The lead compound RO200-1770 shows broadly nonspecific micromolar inhibition, excepting only stromelysin 1 (MMP-3) with its ϳ10-fold weaker binding affinity to RO200-1770. To facilitate synthesis, the piperidine of the lead compound RO200-1770 was substituted by an essentially isosteric piperazine, RO204-1924. The almost uniform decrease in binding affinity might be rationalized by higher desolvation penalties for piperazine binding. The theoretical clogP values calculated for 1,4-dimethylpiperidine (1.9) and 1,4-dimethylpiperazine (0.8) support this hypothesis (39). I-COL043 and RO206-0027 represent the results of two orthogonal approaches to optimize P1Ј-S1Ј and P2Ј-S2Ј binding, respectively. For each inhibitor, an ϳ10-fold increase in inhibition toward MMP-8, -2, -9, and -3 was accomplished, while inhibition of MT1-MMP and MMP-1 was weakened or remained relatively unchanged. With its 4-fold weaker inhibition of MMP-1, I-COL043 showed significantly enhanced selectivity potential against the latter enzyme. The P1Ј and P2Ј optimizations of I-COL043 and RO206-0027 are combined in RO206-0032 and the inhibition values demonstrate, to a first approximation, additivity of the effects for MMP-8, -2, -9, and MT1-MMP. The improvement in its binding affinity to stromelysin (MMP-3) is less distinct, while fibroblast collagenase (MMP-1) binding averages rather than sums the effects of the precedent compounds.
Crystal Packing Effects-The MMP8-RO200-1770 complex did not crystallize as previously described (26), but also under the previously reported crystallization conditions formed the crystal packing arrangement described here. Thus, the inhibi-  (4,40,41). The relative domain arrangement of the catalytic and C-terminal domains, as seen for MMP-1 (11) and MMP-2 (42), shows the importance of the primed substrate recognition sites, since these are located at the interface of the two collagenase domains. Intriguingly, Asn 188 -Tyr 189 , located at the corridor connecting the catalytic and the C-terminal domain, adopts a cis-peptide bond (Fig. 5).
Although not yet recognized, this cis-peptide bond is not unique to the present crystal form; re-inspection confirmed its presence also in the alternative crystal form (26). This cis-peptide bond is located on the solvent-exposed loop preceding the "catalytic" ␣-helix L 191 -H 197 EXXH 201 -L 203 . The only restraint apparent for this structural framework is a stabilizing hydrogen bond between carbonyl oxygen of Thr 181 with the amide of Tyr 189 . Sequence comparison of this segment with related MMPs reveals a subdivision within the MMP family. Only collagenases 1 and 2 (MMP-1 and MMP-8) lack a glycine at position 188, a feature otherwise absolutely conserved, including nonhuman species as well. We therefore predict that the Glu 188 -Tyr 189 peptide bond of MMP-1 also adopts a cis-conformation. As exemplified by the crystal structure of stromelysin 1 (MMP-3) (23, 43), Gly 188 exhibits dihedral angles (, ) ϭ (150°, 165°), which correspond to a conformation allowed only for glycine. Therefore, glycine is conserved at position 188 presumably to stabilize the local fold; conservation of a nonglycine residue (MMP-1, MMP-8) suggests a function related to the cis-peptide bond. DISCUSSION Inhibitor Conformation and Its Interaction with the Protein-Although identified as potent collagenase inhibitor by an independent screening program, the barbiturate-based inhibitor family exhibits striking similarities with well characterized classes of inhibitors, namely hydroxamic and malonic acidbased compounds (2,26). Fig. 3C illustrates that the Zn 2ϩ chelation geometry of the hydroxamate, exemplified by batimastat, is mirrored by the barbiturate with its N 3 nitrogen substituting for the keto group of the hydroxamate. Additionally, the interaction of the barbiturate N 1 -H 1 and O 6 with the protein backbone Ala 160 -Ala 161 parallels that of batimastat (Fig. 3). A subtle difference is found at the O 6 interaction of the barbiturate ring, since the additional amide interaction with Ala 161 could stabilize a greater negative charge on O 6 .
These findings present opportunities with challenges. The structural similarity of both inhibitor classes for example enables the application of knowledge of optimization criteria for one class to the other. On the other hand, the similarity might also indicate a limitation in finding specific metalloproteinase inhibitors; the presence of a similar metal chelation topology in independently identified and structurally unrelated lead compounds indicates that the Zn 2ϩ binding follows a rather universal recognition motif, which dominates the binding characteristics. Consequently, many if not most potent active site directed Zn 2ϩ protease inhibitor will exploit such a universal binding motif and are likely to exhibit a low specificity profile, at least prior to optimization.

Structural Basis of Collagenase Substrate Specificity
barbiturate ring. First, its pK a varies dramatically with the presence of ring substituents. Whereas the pK a of unsubstituted barbituric acid is about 4, its 5,5-diethyl substituted analog ("barbital") has a pK a of around 8 (37). Second, the surrounding protein will also strongly affect the protonation of the barbiturate.
To address this issue, we inspected each polar group of the inhibitor for possible hydrogen bonding partners. The 2.0-Å distance of the catalytic Zn 2ϩ to N 3 excludes its protonation, and the O 2 H 2 hydroxyl group is necessary to avoid repulsion of the carboxylate of Glu 198 . N 1 and O 6 are involved in main chain hydrogen bonds. Consequently, their protonation appears well defined as depicted in Fig. 1. O 4 is the only polar group without apparent attractive interactions with the protein. However, the reorientation of the Pro 217 carbonyl described above would seemingly not occur if O 4 is protonated as an alcohol. These arguments summarize the case for the formula depicted in Fig.  1, which carries no net charge. Tunneling of the proton H 2 (Fig.  1), which bridges the carboxylate group of Glu 198 , however, transfers a partial negative charge to O 2 and by resonance also to N 3 (Fig. 1). (A second line of investigation using conformational correlation analysis of the 1.8-Å resolution structure presented here with barbiturate derivatives deposited in the Cambridge small molecule data base was not conclusive.) S1Ј and S2Ј Interaction, and Enzyme Inhibition Profiles-Compared with MMP-8, human fibroblast collagenase (MMP-1) has a more restricted S1Ј site with its Arg instead of Leu at position 193. Its guanidinium group approximately occupies the three S1Ј solvent sites of MMP-8, namely Sol 595 , Sol 602 , and Sol 592 . Conversely, three solvent molecules are found near Thr 222 in MMP-1, where in MMP-8 the guanidinium group of Arg 222 is found. It appears, therefore, possible to enlarge the MMP-1 S1Ј subsite to an MMP-8 size by swapping its Arg side chain and solvent molecules. Although such a swapped conformation has been confirmed (12) for MMP-1, the rehydration is likely to create a considerable kinetic barrier. Consequently, MMP-1 is expected to bind large P1Ј residues with a k on kinetic rate considerably lower than for MMP-8. The S1Ј site of TACE appears rather too large to properly accommodate the large P1Ј residue of ICOL 043 and RO206-0032 (Table II). A unique feature of the TACE active site (44) is the occurrence of Ala at the equivalent position of the strictly conserved Tyr 219 (MMP-8 numbering) of MMPs. This renders the TACE S1Ј site both larger and less hydrophobic than in MMPs by almost completely removing the barrier to the S3Ј site. Consequently, the hydrophobic P1Ј residue of ICOL 043 and RO206-0032 is not optimally anchored in the TACE S1Ј pocket. Further, incomplete dehydration of the voluminous site is likely to disrupt the solvent structure within the TACE S1Ј site (44) without the energy compensation of a good fit.
Considering MMP-3, the southern rim of the active site, and in particular Pro 221 (Pro 217 in MMP-8), is rigidified by His 224 (Ala 220 in MMP-8), which hydrogen-bonds via its N ␦1 and N ⑀2 atoms to the backbone carbonyl groups of Leu 222 and Thr 215 (Asn 218 and Pro 211 in MMP-8). The MMP-3 active site is thus incompatible with binding the barbiturate ring, as reflected by the overall lower binding constants. His 224 is unique to MMP-3. The observed progressive increase of binding affinity with enlarging P1Ј or P2Ј residues is likely due to generally increasing hydrophobic surface areas.
Optimization of Binding Affinity and Selectivity-Excepting C4ϭO4, all polar groups of the barbiturate bind within the protein matrix and should, therefore, remain invariant in optimization strategies. The C4ϭO4 keto group, however, presumably weakens the binding because of repulsive interactions with the Pro 217 carbonyl group, as described in the paragraph "Protein Conformational Changes." Consequently, we identify this group as a major variable in optimizing the Zn 2ϩ -chelating moiety. In fact, modeling studies together with small molecule crystallographic data base analyses indicate that analogous five-membered ring systems might be an appropriate substitute Zn 2ϩ -binding group, provided that the Zn 2ϩ chelating properties are maintained. For example, deletion of the C 4 ϭO 4 ketone and retention of the position of the chelating group N 3 -C 2 -OH 2 will allow the new five-membered ring to relax to a chemically reasonable geometry. With appropriate restraints for the relaxation, the result will still possess favorable hydrogen bonding interactions between the HN 1 -C 6 ϭO 6 segment and the protein. To compensate for concomitant displacements of the phenyl ring, it would be necessary to add a spacer atom to restore unstrained occupancies of both the S1Ј and S2Ј pockets.
Additional optimization approaches are suggested by kinetic analyses of earlier x-ray structures of MMPs with peptidic, hydroxamic, and malonic acid-based inhibitors (2,26,27,45), whereby significant improvement in binding affinity is achieved by better filling the respective binding pockets (see Ref. 46 for a comprehensive review of these approaches). Expansion of the P1Ј residue with an appropriate heterocyclic ring should supply both the necessary flexibility to optimally fill the curved S1Ј pocket and the hydrophilicity to adequately replace the binding sites of the three water molecules found in the S1Ј pocket. In summary, the major contribution to specificity can be attributed to the P1Ј-S1Ј interaction, while both substituents similarly contribute to MMP-8 binding (Table II).
Protein Conformational Changes-As described under "Results," the most striking structural changes induced by inhibitor binding occur near the catalytic Zn 2ϩ with a major contribution from repulsive interactions between the C 4 ϭO 4 ketone and the Pro 217 carbonyl group. For the collagenases, considerable flexibility near the active site environment appears physiologically necessary for triple helical peptide processing (47). The flexibility observed in the present structure suggests that the loop Ala 206 -Asn 218 will provide much of the flexibility necessary for collagen substrate recognition (48), along with the additional plasticity seen at the catalytic Zn 2ϩ and its ligating residues. The generally weaker inhibition constants indicate that stromelysin 1 (MMP-3) does not possess the necessary plasticity in this segment for barbiturate binding.
The intermolecular crystal contact Ser 209 -His 207 might provide an unexpected opportunity for synthetic drug design. In particular, this interaction offers the welcome possibility to deviate from peptide-like binding patterns at a highly ordered position in the active site. Intriguingly, since hydroxyprolines and other hydroxylated amino acids are present in the physiological substrates of the extracellular matrix, we speculate that this crystal contact may mimic a hydroxylated substrate interaction. An alternative possibility that also would exploit the common His-Ser motif would be an interaction with His 201 N ␦1 .
Type I Collagen Recognition Exosite-Independent investigations by others on rat MMP-8 have shown the 188-loop to be required for collagenase activity. The single site-directed mutation to N209K, corresponding to N188K in human MMP-8, disrupts collagenolytic activity. 2 In addition, hybrid molecule studies involving stromelysin 1 (N-terminal) and collagenase 1 (C-terminal) underscore the importance of this loop for collagenolytic activity. The segment R 181 WTNNFREY 189 of collagenase 1 is critical for triple-helicase activity (49). In addition to collagenase 1, 2, and 3, the two gelatinases MMP-2 and MMP-9 have tyrosine at position 189, both of which are preceded by a large insertion of three fibronectin II domains (Fig. 6). These domains also are known to be critical for substrate recognition (4,48,50). Therefore, the 188 exosite serves as a collagen substrate recognition site in both collagenases (MMP-1, -8, and -13) and gelatinases (MMP-2 and -9). This proposed substrate recognition site is the position of the cis peptide bond described here for MMP-8 and predicted for MMP-1. As such, it distinguishes collagenases 1 and 2 (MMP-1 and -8) from the other MMPs known to cleave collagen that have a glycine at this position, including MMP-13 (17), MMP-14 (51), and MMP-18 (52). Thus, collagenase 3 (MMP-13) (12, 53), MMP-14 (54), and presumably MMP-18 have a different backbone conformation in this loop segment. This structural relationship is reflected by the biochemical properties of the respective enzymes. MMP-13 is distinct from MMP-1 and -8, as it preferentially hydrolyzes type II collagen, whereas the enzyme was 5 or 6 times less efficient at cleaving type I or III collagen (17). Similarly, MMP-14 is 5-7 times less efficient at hydrolyzing type I collagen than MMP-1, whereas its gelatinolytic activity is 8 times higher than that of MMP-1 (51).
To further investigate the role of the 189 exosite for macromolecular substrate recognition, we docked a collagen triple helix to a full-length collagenase (MMP-1). In addition to optimizing overall contact areas of the substrate-enzyme complex, we were guided by the following localized interactions: (a) the contact of the collagen helix with the primary substrate recognition sites, including the catalytic Zn 2ϩ ; (b) the contact of the collagen helix with the 189 exosite; and (c) the interaction of the collagen hydroxyproline with His 207 . We used published data for modeling the structures of the isolated components (11,55,56). The most reliable and powerful conclusion from these modeling studies is the orientation of the extended collagen peptide relative to the enzyme. Earlier models postulated that the triple helix makes major contacts with the first "blade" of the propeller-like hemopexin domain (48). We conclude, however, that the triple helix will not lie in the MMP active site oriented along the shortest route to the hemopexin-like domain, which would bring it into contact with its first blade. Instead, we propose that the substrate runs through the 188 exosite, leading to major contacts to blade 2 of the C-terminal collagenase domain, consistent with the chimera mutant studies by Nagase and co-workers (49). The extended contact of the collagen substrate with the catalytic domain is consistent with a collagenolytic activity of the catalytic domain alone, as described for MMP-1 (57). On the other hand, the conservation of the substrate exosite within MMP-1 and MMP-8 would suggest that the catalytic domain of MMP-8 should also exhibit a collagenolytic activity that however has not been observed (48). A second important consequence of these modeling studies is that at least one of the collagen strands must be bent or arched by ϳ20 o ; an perfectly straight, rod-like collagen model binding to the active site of MMP-8/1 remains ϳ6.5 Å distant from the collagen exosite (Tyr 189 -Asn 190 ). This bending may be part of the unwinding mechanism of the collagen triple helix necessary for its proteolysis (48). The occurrence of a presumably functional cis-peptide bond at the type I collagen exosite of collagenase 1 and 2 poses the question of its precise mechanism in collagenolysis. The function of the 188 loop should be structurally linked to either (a) the amino acid 188 side chain or (b) the backbone conformation due to the cis-peptide bond. Since the residue at the "nonglycine" position 188 of MMP-1 and MMP-8 ( Fig. 6) shows conserved similarity but not identity, a contribution of option b seems likely. A detailed understanding of the 188-exosite's mechanism in collagen processing, and in particular the role of the amino at 188, awaits further experiments.
We suggest that subscribers photocopy these corrections and insert the photocopies at the appropriate places where the article to be corrected originally appeared. Authors are urged to introduce these corrections into any reprints they distribute. Secondary (abstract) services are urged to carry notice of these corrections as prominently as they carried the original abstracts.