High Resolution Structure of the N-terminal Domain of Tissue Inhibitor of Metalloproteinases-2 and Characterization of Its Interaction Site with Matrix Metalloproteinase-3*

The high resolution structure of the N-terminal domain of tissue inhibitor of metalloproteinases-2 (N-TIMP-2) in solution has been determined using multidimensional heteronuclear NMR spectroscopy, with the structural calculations based on an extensive set of constraints, including 3132 nuclear Overhauser effect-based distance constraints, 56 hydrogen bond constraints, and 220 torsion angle constraints (an average of 26.9 constraints/residue). The core of the protein consists of a five-stranded β-barrel that is homologous to the β-barrel found in the oligosaccharide/oligonucleotide binding protein fold. The binding site for the catalytic domain of matrix metalloproteinases-3 (N-MMP-3) on N-TIMP-2 has been mapped by determining the changes in chemical shifts on complex formation for signals from the protein backbone (15N,13C, and 1H). This approach identified a discrete N-MMP-3 binding site on N-TIMP-2 composed of the N terminus of the protein and the loops between β-strands AB, CD, and EF. The β-hairpin formed from strands A and B in N-TIMP-2 is significantly longer than the equivalent structure in TIMP-1, allowing it to make more extensive binding interactions with the MMP catalytic domain. A detailed comparison of the N-TIMP-2 structure with that of TIMP-1 bound to N-MMP-3 (Gomis-Ruth, F.-X., Maskos, K., Betz, M., Bergner, A., Huber, R., Suzuki, K., Yoshida, N., Nagase, H., Brew, K., Bourne, G. P., Bartunik, H. & Bode, W. (1997) Nature 389, 77–80) revealed that the core β-barrels are very similar in topology but that the loop connecting β-strands CD (P67-C72) would need to undergo a large conformational change for TIMP-2 to bind in a similar manner to TIMP-1.

Remodeling of connective tissue is an important event in many normal and pathological processes such as growth, wound healing, tumor invasion, and rheumatoid and osteoarthritis. The matrix metalloproteinases (MMPs) 1 are a group of enzymes thought to be primarily responsible for this catabolism (1,2). As expected, MMP activity is highly regulated and includes inhibition by a family of specific protein inhibitors known as tissue inhibitors of metalloproteinases (TIMPs). Four TIMPs have now been identified (TIMP-1 to TIMP-4), which are about 190 amino acids in length and share 40 -50% sequence identity (3)(4)(5)(6). The TIMPs inhibit activated MMPs by forming tight (K d Ͻ 1 nM), 1:1, noncovalent complexes and do not undergo cleavage by the enzymes (7). The TIMPs show relatively little selectivity in binding to the MMPs, but substantial differences have been reported for their binding to members of the recently discovered membrane-bound MMP family (8).
The TIMPs appear to consist of two domains, although only the N-terminal one has been shown to form a stable, autonomously folded unit (N-TIMP). This domain, encompassing about two-thirds of the molecule and including three of the six disulfide bonds, retains full inhibitory activity and is able to form stable tight complexes with MMP catalytic domains (9). Previously, we were able to determine a low resolution solution structure for N-TIMP-2 (127 residues) using only two-and three-dimensional 1 H NMR data (10); this revealed the backbone topology of the molecule and surprisingly identified it as a member of the oligosaccharide/oligonucleotide binding protein fold family (11). These studies were carried out using recombinant N-TIMP-2 produced in mammalian cells. To obtain a high resolution solution structure for N-TIMP-2, it was necessary to produce uniformly 15 N-and 15 N/ 13 C-labeled N-TIMP-2. This was recently achieved by expression of the protein in Escherichia coli grown on labeled minimal medium followed by refolding of the labeled N-TIMP-2 from inclusion bodies (12).
In this paper we report both the high resolution solution structure of N-TIMP-2 and the mapping of its binding site for the catalytic domain of human stromelysin-1 (N-MMP-3), both of which were achieved using double and triple-resonance heteronuclear NMR-based methods. Recently, a crystal structure for the complex formed between TIMP-1 and N-MMP-3 has been reported (13). The coordinates of this model have been made available by the authors and a comparison with the high resolution structure reported here for N-TIMP-2 reveals several notable differences between the structures for bound TIMP-1 and free N-TIMP-2. For example, the extended ␤-hairpin formed from strands A and B in N-TIMP-2 allows this region to make extensive interactions with N-MMP-3, which are simply not possible for TIMP-1 and may form the basis for the selectivity reported for membrane bound MMPs (8). In addition, the well defined loop connecting ␤-strands C and D in N-TIMP-2 would need to undergo a substantial conformational change on binding to N-MMP-3 if TIMP-2 is to interact with the catalytic domain in a manner analogous to TIMP-1.

EXPERIMENTAL PROCEDURES
Sample Preparation-The following protein samples were prepared as described previously: nonisotopically enriched and uniformly 15 N-or 15 N/ 13 C-labeled N-TIMP-2, N-MMP-3, and complexes formed between N-TIMP-2 and N-MMP-3 (10,12,14,16). The NMR experiments were carried out on 0.6-ml samples of 1.5-3.0 mM nonisotopically enriched, 1.7 mM 15 N-labeled and 1.4 mM 15 N/ 13 C-labeled N-TIMP-2 dissolved in 25 mM sodium phosphate and 100 mM potassium chloride at pH 6.7. In the case of the 15 N/ 13 C-labeled N-TIMP-2⅐unlabeled N-MMP-3 complex, spectra were recorded from 0.9 ml samples of 0.6 mM complex in 5 mM deuterated imidazole, 100 mM sodium chloride, and 5 mM calcium chloride at pH 6.7 using 8-mm Shigemi tubes. The NMR samples were prepared in either 100% D 2 O or 90% H 2 O, 10% D 2 O as appropriate.
NMR Spectroscopy-All NMR experiments were performed on Varian Unity and UnityPlus spectrometers operating at 500 and 600 MHz, with all the data collected in phase-sensitive mode using the method of States (17) and at a temperature of 35 or 40°C. Detailed descriptions of the two-and three-dimensional 1 H, 15 N/ 1 H, 13 C/ 1 H, and 15 N/ 13 C/ 1 H experiments used to obtain nearly complete sequence-specific 1 H, 15 N, and 13 C resonance assignments for N-TIMP-2 have been reported previously (10,14). In addition, three-dimensional 15 N/ 1 H NOESY-HSQC (18) and 13 C/ 1 H HMQC-NOESY (19,20) spectra were acquired with an NOE mixing period of 125 ms and three-dimensional 15 N/ 1 H HNHA (21) and HNHB (22) spectra recorded. A series of 10 two-dimensional 15 N/ 1 H HSQC spectra (23)(24)(25) were also acquired over 4 h from a sample of 15 N-labeled N-TIMP-2 freshly dissolved in D 2 O to identify slowly exchanging amide protons in the protein. In the case of the N-TIMP-2⅐N-MMP-3 complex, experiments carried out were two-dimensional 15 N/ 1 H HSQC, two-dimensional 13 C/ 1 H HMQC, three-dimensional 15 N/ 13 C/ 1 H HNCA and CBCA(CO)NH (25)(26)(27).
The three-dimensional spectra were recorded over about 65 h with acquisition times in the indirect dimensions (F 1 of F 2 ) of 9.4 -15.5 ms for 15 N, 6 -11.3 ms for 13 C, and 18.7-24.6 ms for 1 H, as appropriate, and in the real time domain of 48 -128 ms. Water suppression in the experiments was achieved by using the pulsed-field gradient-based WATER-GATE method (28).
Two-and three-dimensional NMR data were processed essentially as described previously (10,14,29) using the NMRPipe software package on Silicon Graphics workstations (30). All the spectra were analyzed on-screen using the program XEASY (31).
Structural Calculations-Structurally significant intra-and interresidue NOEs were identified in three-dimensional 15 N/ 1 H NOESY-HSQC, three-dimensional 13 C/ 1 H HMQC-NOESY, and two-dimensional NOESY spectra of N-TIMP-2 recorded with mixing times of 125 ms. The relationship between NOE intensity and interproton distance was calibrated using NOEs corresponding to known distances in regular ␤-sheet and ␣-helical regions of N-TIMP-2, and on this basis the NOEs were converted to upper distance constraints using the program CALIBA (32) with the maximum upper distance limit set to 6 Å. Where appropriate, standard distance corrections were applied to constraints involving methyl and aromatic ring protons (33,34). In addition, where spectral overlap prevented reliable determination of volumes for NOE cross-peaks the distance constraints were set to the upper limit of 6 Å.
The ratios of diagonal to cross-peak volumes in HNHA spectrum allowed reliable 3 J HN-H␣ coupling constants (Ϯ1 Hz) to be determined for 100 residues of N-TIMP-2. In addition, the relative intensities of the HN to H␤ cross-peaks in the HNHB spectrum allowed 37 3 J N-H␤ coupling constants to be estimated as either large (Ϫ5 Ϯ 0.5 Hz) or small (Ϫ1 Ϯ 0.5 Hz). This coupling constant data together with intraresidue and sequential NOEs were used as input for the program HABAS (32), which produced 89 , 89 , and 42 1 torsion angle constraints.
The high resolution solution structure of N-TIMP-2 was calculated with the program DYANA (35), which uses simulated annealing combined with torsion angle dynamics. The simulated annealing protocol used consisted of a high temperature conformational search stage of 2000 steps, followed by slow cooling over 8000 steps, with conjugate gradient minimization at the end. To maximize the number of converged structures obtained from an initial set of 100 random starting coordinates, each stage of the calculations also included 5 cycles of the REDAC procedure. The set of NMR-derived structural constraints used for the final N-TIMP-2 calculations was the result of several cycles of an interactive process that involved using converged N-TIMP-2 structures from the previous generation of calculations to assign NOEs that were ambiguous with chemical shift information alone. In addition, these preliminary N-TIMP-2 structures, together with 3 J N-H␤b coupling constant and rotating frame Overhauser effect data, were used to determine stereospecific assignments and 1 angles for more residues. At each stage in the calculation cycle new generations of N-TIMP-2 structures were calculated from 100 random starting coordinates, with the additional information obtained now included. In the final stages of the N-TIMP-2 structure refinement, additional distance constraints were included in the calculations corresponding to NMR determined hydrogen bonds between backbone amide and carbonyl groups. Hydrogen bonds were included only for residues whose amide proton was detectable after at least 1 h in D 2 O and where the distance between the hydrogen bond acceptor and donor atoms was less than 2.5 Å and the NH-O bond angle greater than 135°. For each hydrogen bond identified in N-TIMP-2, appropriate lower and upper distance limits were used to constrain NH to O to 1.8 -2.3 Å and N to O to 2.4 -3.3 Å.
After the final round of DYANA calculations, 49 satisfactorily converged N-TIMP-2 structures were obtained from 100 random starting conformations. The converged structures contain no distance constraint or van der Waals violations greater than 0.5 Å and no dihedral angle violations greater than 5°. The sums of the violations for upper distance limits, lower distance limits, van der Waals contacts, and torsion angle constraints were 64.9 Ϯ 1.5 Å, 0.8 Ϯ 0.2 Å, 34.6 Ϯ 1.1 Å, and 40.7 Ϯ 4.9 o , respectively. Similarly, maximum violations for the converged structures were 0.42 Ϯ 0.04 Å, 0.19 Ϯ 0.05 Å, 0.35 Ϯ 0.04 Å, and 3.8 Ϯ 0.6 o , respectively. The solution structure of N-TIMP-2 is determined to very high precision, which is reflected in very low root mean squared deviation values for both backbone and all heavy atom coordinates of 0.38 Ϯ 0.06 and 0.74 Ϯ 0.08 Å, respectively, for the family of 49 converged structures.
The residue by residue distribution of the NOEs identified for N-TIMP-2 is shown in Fig. 1. The observed NOEs are distributed relatively evenly throughout the protein, with only residues near the N (Cys 1 -Ser 2 ) and C termini (His 120 -Tyr 122 and Gly 125 -Glu 127 ) and in three short stretches corresponding to Ser 68 -Ala 70 , Gly 79 -Gly 80 , and Gly 92 -Gly 94 showing significantly fewer than average NOEs. The backbone amide signals for Cys 1 , Ser 2 , and Ser 68 -Val 71 were not detectable in the NMR spectra, which may in part account for the lack of NOEs involving these residues.
The topology of the protein backbone in the high resolution solution structure of N-TIMP-2 is shown in Fig. 2, A and C. The locations of the elements of regular secondary structure are identical to those described for our low resolution structure of N-TIMP-2 determined previously (10), with the exception that strand C is extended to include Ala 66 . The most striking feature of the protein remains the closed, five-stranded ␤-barrel, which is formed from ␤-strands A (Val 17 -Asp 34 ), B (Asn 38 -Phe 53 ), C (Phe 62 -Ala 66 ), D (Tyr 84 -Glu 91 ), and E (Lys 95 -Ile 98 ). The ␤-hairpin involving strands A and B, together with strands C and D, forms a classic Greek Key motif, with the ␤-barrel closed by strand E, which runs parallel to strand C and antiparallel to strand D. The short strand F (Val 105 -Trp 107 ) can really be considered a continuation of strand E with the disulfide-bonded loop formed by residues Thr 99 -Ile 104 separating the two parts. The two ␣-helices present in N-TIMP-2 (Pro 8 -Asn 14 and Thr 112 -Leu 118 ) run essentially parallel to each other and pack side by side against the face of the ␤-barrel formed by the N-terminal residues of strands A and D, the C-terminal residues of strand B, and strand F.
The majority of the backbone dihedral angles in the high resolution N-TIMP-2 structures are very well defined with order parameters of greater than 0.9 (37), however, the first two N-terminal residues (Cys 1 -Ser 2 ) and a few others (Glu 61 , Ala 90 , Asp 102 ) are significantly less ordered. Many side chain dihedral angles ( 1 , 2 , etc.) are also well defined (Fig. 2B), not only for residues located in the core of the protein, but also for some surface residues such as Val 6 , Glu 28 , and Gln 123 . Examination of the Ramachandran plot obtained for the family of N-TIMP-2 structures (Fig. 3) reveals that almost all residues adopt favorable values of and , although 6 residues (Lys 41 , Gln 49 , Val 78 , Ala 90 , Trp 107 , and Ser 111 ) adopt nonpreferred backbone dihedral angles.
Analysis of the high resolution N-TIMP-2 structures revealed that the side chain of Cys 101 adopts two distinct conformational groups with 1 angles centered around Ϫ130 o and Ϫ30 o (Fig. 2, A and B). This local conformational variability in the N-TIMP-2 structures may be a reflection of insufficient NMR data for this residue, or alternatively it may indicate real conformational heterogeneity in the protein.
Comparison of the high resolution structures reported here with the family of low resolution structures determined previously for N-TIMP-2 (10) reveals that the ␤-barrels have a very similar topology with an average root mean squared deviation of 2.06 Ϯ 0.02 Å for the backbone atoms. However, the orientation of helix-2 (Thr 112 -Leu 118 ) with respect to the rest of the protein is substantially different in the high resolution N-TIMP-2 structures with helix-2 running almost parallel to and packing against helix-1 (Pro 8 -Asn 14 ; Fig. 2, A and C). The improved definition of the helix-2 orientation can be attributed to a very significant increase in the number of NOE-derived distance constraints obtained for residues in helix-1 and -2, in particular, from the 13 C/ 1 H HMQC-NOESY spectra. Apart from the orientation of helix-2, the low resolution structure reported previously for N-TIMP-2 has proved to be essentially correct and has already been very successful in guiding and interpreting site-directed mutagenesis work (38). The inability to detect these resonances for N-TIMP-2 in the complex probably arises because they are significantly exchange broadened compared with signals from other backbone groups in the protein. Interestingly, many of these signals are also not detected in free N-TIMP-2, in particular the two stretches around the C1-C72 disulfide bond (Cys 1 -Ser 4 and Ser 68 -Cys 72 ). We previously proposed that these signals could be broadened due to intermediate exchange between different local conformations (10,14).

Characterization of the MMP Catalytic Domain Binding
To identify N-TIMP-2 signals perturbed by N-MMP-3 binding, the backbone assignments for bound N-TIMP-2 were compared with those determined for the free protein and the chemical shift change noted for each signal. The shifts in backbone amide 15 N and ␣ carbon 13 C signals were then scaled to take some account for the difference in spectral dispersion compared with backbone amide 1 H signals giving a more equal weighting to each data set ( 15 N range 130 -107 ϭ 23 ppm; 13 C range 67-44 ϭ 23 ppm; 1 H range 10.1-6.9 ϭ 3.2 ppm; giving a correction factor of 0.14 for both 15 N and 13 C signals). Correction in this manner allows an average change in backbone chemical shift to be calculated for each residue in N-TIMP-2, results of which are shown in Fig. 4. Examination of the chemical shift differences clearly identifies 8 residues in N-TIMP-2 whose backbone signals undergo large chemical shift changes on complex formation, specifically Ser 32 , Tyr 36 , Ile 40 , Lys 41 , Ala 70 , Val 71 , Gly 73 , and Cys 101 . These residues fall into three distinct regions of the protein: (i) the end of strand A, the AB loop, and the beginning of strand B (Glu 28 -Lys 41 ); (ii) Ala 70 -Gly 73 in the long loop between strands C and D, which is disulfide-bonded to Cys 1 through Cys 72 ; and (iii) in the loop between strands E and F (His 97 -Ile 104 ), which is also linked to the N terminus by the Cys 3 -Cys 101 disulfide bond. Val 71 showed the largest chemical shift changes, more than twice as large as those of any other residue, whereas signals from about 70% of the residues in N-TIMP-2 showed no appreciable chemical shift changes on complex formation. Fig. 5 shows a space-filled view of N-TIMP-2 in which residues are color-coded according to the changes in average backbone chemical shift observed on N-MMP-3 binding (red Ͼ 0.20 ppm, yellow Ͼ 0.05 ppm, blue Ͻ 0.05 ppm, and white unknown). This clearly shows that the MMP binding site is formed from the region around the Cys 1 -Cys 72 and Cys 3 -Cys 101 disulfide bonds, which connect loops CD and EF to the N methyl groups in N-TIMP-2. Sequence-specific 1 H and 13 C resonance assignments for methyl groups in bound N-TIMP-2 were made from three-dimensional 13 C/ 1 H HMQC-NOESY spectra of the 15 N/ 13 C-labeled N-TIMP-2⅐unlabeled N-MMP-3 complex, using the ␣ and ␤ carbon assignments obtained from CBCA(CO)NH and HNCA spectra as starting points. This approach allowed resonance assignments to be made for all methyl groups in the protein apart from those of Ile 35 , Ile 40 , Ala 66 , and Val 71 . No significant chemical shift changes occurred for any assigned methyl signals on N-MMP-3 binding, except for Ala 70 . These results are entirely consistent with the chemical shift changes seen for the protein backbone resonances of N-TIMP-2. 1 H signals from a pair of methyl groups were each shifted by over 0.5 ppm on complex formation, but unfortunately showed no NOEs to ␣ or ␤ protons and so could not be assigned. These highly shifted methyl resonances must belong to either Ile 35 , Ile 40 , or Val 71 ; all of these residues are also implicated in MMP binding by the changes seen in the chemical shifts for their backbone signals.
In an earlier report (14), we described a preliminary mapping of the N-MMP-3 binding site on N-TIMP-2 using a nearest peak-based approach in which lower limits for chemical shift changes in backbone amide signals on complex formation were determined by noting the 15 N/ 1 H cross-peaks in HMQC spectra of the complex that were nearest to an assigned peak in a comparable HMQC spectrum of free N-TIMP-2. Now that real values (rather than lower limit values) for the chemical shift changes of the backbone nuclei have been determined (Fig. 4), the success of the earlier "nearest-peak" approach can be assessed. In fact, the nearest-peak approach was found to be highly successful in identifying the overall location of the N-MMP-3 binding site on N-TIMP-2, but the availability of actual chemical shift changes on complex formation allows one to focus more clearly on those residues involved in MMP binding as their true degree of change relative to the other residues is not attenuated by the inherent underestimates of chemical shift changes determined by the nearest-peak approach for the most perturbed residues.
Conformational Heterogeneity in the MMP Binding Site on N-TIMP-2-The inability to detect backbone amide resonances from Cys 1 -Ser 4 and Ser 68 -Cys 72 in free N-TIMP-2 indicates that these signals are either significantly broadened compared with other backbone amide signals from the protein or there is intermediate or fast exchange with water protons catalyzed by the local environment (14,15). Localized signal broadening from exchange processes could result if the protein in these regions exists in different conformational states in intermediate exchange on the NMR chemical shift time scale. Although these two regions form a large part of the MMP binding site, the formation of a very tight and stable N-TIMP-2⅐N-MMP-3 complex does not result in the backbone amide signals for residues in this region of N-TIMP-2 becoming detectable in HSQC, CBCA(CO)NH, or HNCA spectra of the complex, except for Val 71 . This implies that complex formation had little effect on the processes responsible for the backbone amide signal line broadening of residues Cys 1 -Ser 4 and Ser 68 -Cys 72 . The binding of N-MMP-3 to N-TIMP-2 almost certainly prevents these backbone amides from being in contact with bulk water, which would argue against rapid exchange via exchange with the water. Thus, the backbone amide line broadening for residues Cys 1 -Ser 4 and Ser 68 -Cys 72 in N-MMP-3-bound N-TIMP-2 would appear to arise from interconversion between multiple local conformations of the protein. The model of the TIMP-1⅐N-MMP-3 complex determined by x-ray diffraction (13) does not suggest a clear mechanism for this conformational heterogeneity. Indeed, the x-ray model shows many of the affected backbone amide protons in well defined conformations participating in hydrogen bonds with backbone carbonyl oxygen atoms of N-MMP-3.
Another interesting observation for bound N-TIMP-2 is the detection of two sets of signals from the side chain methyl groups of Ile 43 and of a pair of substantially shifted methyl resonances arising from either Ile 40 , Ile 35 , or Val 71 . No other signals appeared to be doubled, and the ratio of major:minor forms (80:20) is the same for both sets of signals and did not change with time or between different preparations of the complex. This suggests that the methyl signal doubling arises from the presence of two conformational states of bound N-TIMP-2 in this local region, which are in slow exchange on the NMR time scale. Interestingly, the residues involved are again implicated in N-MMP-3 binding; however, it seems likely that this conformational heterogeneity in N-TIMP-2 is distinct from that affecting Cys 1 -Ser 4 and Ser 68 -Cys 72 discussed previously.
Comparisons between the N-MMP-3 Binding Sites on TIMP-1 and TIMP-2-The N-MMP binding site identified by NMR on N-TIMP-2 is entirely consistent with that seen in the crystal structure of the related TIMP-1⅐N-MMP-3 complex (13), in which the N-MMP-3 binding site on TIMP-1 consists of the N terminus (Cys 1 -Val 4 ), CD loop (Ala 65 -Cys 70 ), AB loop (Val 29 , Thr 33 -Tyr 35 ), and EF loop (Thr 98 -Cys 99 ). The first two sites make three-quarters of all intermolecular contacts, whereas the last two make direct but weaker side chain-mediated hydrophobic interactions. The corresponding residues in TIMP-2 are Cys 1 -Ser 4 , Pro 67 -Cys 72 , the AB loop region around Val 29 -Lys 41 and Leu 100 -Cys 101 , respectively, exactly those residues identified as participating in N-MMP-3 binding by this NMR study. A significant structural difference between TIMP-1 and -2 is the insertion of 7 residues in the AB loop of TIMP-2, which allows it to make more extensive contacts on binding to N-MMP-3. In the case of TIMP-1, Thr 33 and Leu 34 make hydrophobic contacts with N-MMP-3 and are homologous to Ile 40 and Lys 41 in TIMP-2, both of which show substantial chemical shift changes for backbone signals on complex formation. On the other side of the AB ␤-hairpin, Val 29 , which makes hydrophobic interactions with N-MMP-3 in the crystal structure, is conserved in both TIMPs and together with Glu 28 also shows and Gly 32 may also indicate additional interactions with N-MMP-3. However, it is interesting to note that these two residues form part of a ␤-bulge in strand A, which could facilitate movement of the AB loop in N-TIMP-2 to optimize interactions with the MMP catalytic domain.
A comparison of the ␤-barrel in the high resolution solution structure of free N-TIMP-2 with that of TIMP-1 in the crystal structure of the complex (13) reveals that there is no major conformational change in the core of the TIMPs on binding to N-MMP-3, which is reflected in an root mean squared deviation of 1.6 Å for backbone atoms of homologous residues in the ␤-barrels. This finding is consistent with our NMR data for the complex, as no significant chemical shift changes were detected for nuclei distant from the N-MMP-3 binding site on complex formation, indicating that there is no overall conformational change in N-TIMP-2 on binding to N-MMP-3. The only major differences in the topologies of the two TIMPs are the extension of the AB ␤-hairpin in N-TIMP-2 (discussed above) and a short C-terminal 3 10 helix present in TIMP-1 but not found in N-TIMP-2. It is possible that this helix is absent from TIMP-2, or destabilized by removal of the C-terminal domain, and therefore only present in the full-length protein.
Comparisons of the conformations of exposed loops in N-TIMP-2 and TIMP-1 highlight several interesting differences between the local conformations of the two proteins, particularly in the TIMP/MMP interaction site. Superposition of the two TIMP structures using the coordinates for the backbone atoms of the ␤-barrel (Fig. 6) places the N-terminal residues of both inhibitors in very similar positions, which would allow the N-terminal amide of N-TIMP-2 to interact with the zinc in the catalytic site of N-MMP-3 in the complex, as observed for TIMP-1 (13). However, the conformation of the CD loop (Pro 67 -Cys 72 ) is very different in bound TIMP-1 and free N-TIMP-2. Thus, with free N-TIMP-2 superimposed on the ␤-barrel of TIMP-1 in the complex, residues in the CD loop of N-TIMP-2 do not occupy the active site cleft of N-MMP-3 at positions P2 and P3 as seen for TIMP-1, but lie to one side, pointing away from the enzyme. Consequently, if TIMP-2 is to bind N-MMP-3 in a similar manner to that for TIMP-1, then this region (Pro 67 -Cys 72 ) must undergo a substantial conformational change on N-MMP-3 binding, to place the backbone and side chains in the correct orientation to form TIMP-1-like hydrogen bonds and hydrophobic interactions with the enzyme. The largest displacement of the backbone atoms between the free and bound forms of TIMP in this region is over 9 Å. In addition to the CD loop, the AB ␤-hairpin of N-TIMP-2 must also change its conformation, probably by a reorientation with respect to the rest of the protein, on binding of N-MMP-3, as when N-TIMP-2 is superimposed on TIMP-1 in the complex, the extended AB ␤-hairpin of free N-TIMP-2 penetrates the catalytic domain making numerous van der Waals clashes (Fig. 6). This extended ␤-hairpin is probably somewhat less coiled in the complex, thereby changing its orientation with respect to the ␤-barrel and allowing favorable interactions with N-MMP-3. This change in conformation could occur about the Ser 31 -Gly 32 ␤-bulge, as suggested earlier. The increased length of the TIMP-2 AB hairpin allows specific interactions by residues at the end of this region (i.e. Ile 35 and Tyr 36 ) with the MMPs that are not possible for the TIMPs with shorter AB loops, for example TIMP-1. It is tempting to speculate that these additional interactions could be important for the specificity differences seen between the TIMPs for the membrane-bound MMPs, where TIMP-2 shows much faster binding than TIMP-1 (8).