The Effect of Matrix Metalloproteinase Complex Formation on the Conformational Mobility of Tissue Inhibitor of Metalloproteinases-2 (TIMP-2)*

The backbone mobility of the N-terminal domain of tissue inhibitor of metalloproteinases-2 (N-TIMP-2) was determined both for the free protein and when bound to the catalytic domain of matrix metalloproteinase-3 (N-MMP-3). Regions of the protein with internal motion were identified by comparison of the T 1and T 2 relaxation times and1H-15N nuclear Overhauser effect values for the backbone amide 15N signals for each residue in the sequence. This analysis revealed rapid internal motion on the picosecond to nanosecond time scale for several regions of free N-TIMP-2, including the extended β-hairpin between β-strands A and B, which forms part of the MMP binding site. Evidence of relatively slow motion indicative of exchange between two or more local conformations on a microsecond to millisecond time scale was also found in the free protein, including two other regions of the MMP binding site (the CD and EF loops). On formation of a tight N-TIMP-2·N-MMP-3 complex, the rapid internal motion of the AB β-hairpin was largely abolished, a change consistent with tight binding of this region to the MMP-3 catalytic domain. The extended AB β-hairpin is not a feature of all members of the TIMP family; therefore, the binding of this highly mobile region to a site distant from the catalytic cleft of the MMPs suggests a key role in TIMP-2 binding specificity.

Breakdown of the extracellular matrix is an important event in many normal and pathological processes, such as growth, wound repair, tumor metastasis, and arthritis (1)(2)(3). A large family of zinc-dependent proteinases, the matrix metalloproteinases (MMPs), 1 are thought to be primarily responsible for this matrix catabolism. The activities of the MMP family in the extracellular matrix are highly regulated by transcriptional control, zymogen activation, and inhibition by a family of specific protein inhibitors, the tissue inhibitors of metalloproteinases or TIMPs (4). The TIMPs bind tightly to the active proteinases to form an inactive TIMP⅐MMP complex (5). Four mammalian TIMP proteins have now been identified (TIMP-1 to -4) (6 -9), and their high degree of sequence similarity and conservation of 12 Cys residues suggests that each consists of the same basic fold but with some variations in loop structures and glycosylation. The location of the MMP inhibitory site on the TIMP molecule has been shown to reside predominantly in the N-terminal two-thirds of the protein, defined by the first three disulfide bonds. This domain (N-TIMP) can be expressed independently to generate a fully folded, stable, and active inhibitor (10,11).
High resolution three-dimensional structures are now available for both the active N-terminal domain of TIMP-2 (12) and for the full-length inhibitor (13). Crystal structures have also been published for full-length TIMP-1 and TIMP-2 in complexes with the catalytic domains of MMP-3 (N-MMP-3) and MT1-MMP, respectively (14,15). The structures of the TIMP⅐MMP complexes, together with NMR data on chemical shift perturbation seen for N-TIMP-2 on complex formation with N-MMP-3 (16), have identified the key features of the TIMP inhibitory binding site. It is now clear that the N terminus of TIMP (residues 1-5), together with the two loops with which it is disulfide-bonded, form a "wedge"-like structure that interacts with the active site cleft of the proteinase. A further region in the N-terminal domain of TIMP that interacts with the proteinase is the loop between ␤-strands A and B. In TIMP-2, this region is extended by 7 residues compared with TIMP-1 and is therefore capable of making more extensive interactions with the proteinase. The involvement of the extended AB ␤-hairpin in TIMP-2/MMP interactions was first proposed on the basis of chemical shift changes observed on binding of N-TIMP-2 to N-MMP-3 (16), and further confirmed in the crystal structure for the TIMP-2⅐MT1-MMP complex (15).
NMR studies of N-TIMP-2 provided the first three-dimensional structure for the inhibitor and allowed it to be identified as a member of the OB (oligonucleotide/oligosaccharide)-fold protein family (17). This work was later extended to provide a high resolution structure of N-TIMP-2 using heteronuclear NMR-based methods (12). These studies provided some insights into the dynamics of N-TIMP-2, particularly for the AB ␤-hairpin, which we proposed to be a highly flexible structure due to the relatively narrow line widths seen for signals in this region and the lack of long range NOEs. Furthermore, we suggested that another region of the N-TIMP-2 binding site (Ser 68 -Cys 72 ) was in relatively slow exchange between multiple conformations interconverting on a millisecond time scale, as the backbone amide resonances for these residues were missing from NMR spectra (16).
To provide a more detailed picture of the backbone mobility of N-TIMP-2 and to assess how MMP complex formation affects these motions, the values of T 1 , T 2 , and heteronuclear 1 H-15 N NOE have now been determined for the backbone amide 15 N signals of N-TIMP-2 when free in solution and when bound to N-MMP-3. These studies show that several regions of N-TIMP-2 have significant local mobility on time scales both slower or faster than the overall tumbling time of the protein (estimated to be 10 ns). In particular, rapid motion was seen for the AB ␤-hairpin of free N-TIMP-2, and this mobility was lost on complex formation with N-MMP-3, suggesting that the extended AB loop is a general and important feature of the TIMP-2 binding site.
NMR Spectroscopy-The NMR experiments were carried out at 35°C on a Varian INOVA spectrometer operating at 1 H frequency of 600 MHz. The spectra used to determine T 1 , T 2 , and 1 H-15 N NOE values for the backbone amide 15 N signals of both free and N-MMP-3 bound N-TIMP-2 were recorded with acquisition times of 16.8 ms in F 1 and 142 or 148 ms in F 2 using the sensitivity enhanced pulse sequences described by Farrow et al. (20). The 1 H carrier was centered at the water frequency for the T 1 and T 2 measurements and at the center of the amide region for the heteronuclear 1 H-15 N NOE experiments. The 15 N carrier was placed in the center of the amide region.
In the case of free N-TIMP-2, spectra used to determine T 1 values were acquired with 16 transients per increment and a relaxation delay Data Processing and Analysis-The NMR spectra were processed on a Silicon Graphics Indigo 2 workstation using the program NMRPipe (21). To increase the resolution in the final spectra the number of data points in F 1 was extended 2-fold by linear prediction, and both F 1 and F 2 were zero-filled once. The spectra were examined and peak heights determined using the program XEASY (22).
The T 1 and T 2 values for backbone amide 15 N signals were calculated by non-linear, least-squares fitting of the observed changes in peak heights to appropriate single exponential functions (23, 24), using the program SigmaPlot. The values of steady-state 1 H-15 N NOEs were determined from the peak heights measured in spectra recorded either with (I s ) or without (I o ) presaturation of backbone amide 1 H resonances during the relaxation delay, according to the formula NOE ϭ (I s Ϫ I o )/I o (25).

RESULTS
Relaxation Measurements-This study provides a detailed picture of the backbone dynamics of the inhibitory domain of TIMP-2 (N-TIMP-2), and identifies those regions of the protein that undergo substantial changes in backbone mobility on formation of a stable complex with the MMP-3 catalytic domain (N-MMP-3). T 1 , T 2 , and heteronuclear 1 H-15 N NOE measurements were obtained for the majority of the backbone amide 15 N signals of free N-TIMP-2 (102/121) and for N-TIMP-2 in the complex (98/121). The residues for which no relaxation data could be obtained were Cys 1 -Cys 3 , Val 6 , Cys 13 , Asp 16 (12,16). 15 N relaxation data for the remaining residues could not be determined due to peak overlap in the spectra or very broad signals that prevented accurate measurement of peak heights. For residues Lys 27 , Val 71 , and Thr 112 of free N-TIMP-2, and Tyr 36 , Gln 49 , Glu 83 , and Tyr 122 of N-TIMP-2 in the complex, it was possible to determine the size of the 1 H-15 N NOE but reliable estimates of T 1 and T 2 could not be obtained.
The T 2 values for backbone amide 15 N signals of proteins are sensitive to both fast and slow local motions of the polypeptide backbone. NMR signals with T 2 values significantly longer than the mean are indicative of rapid local motion on a time scale (picosecond to nanosecond) significantly shorter than the overall tumbling time ( m ) of the protein, and have relatively narrow line widths. In contrast, signals with T 2 values significantly shorter than the mean have relatively broad line widths and arise from exchange between two or more states (conformations) with different local environments. To significantly broaden NMR signals and decrease T 2 values, this exchange must be on a time scale (microsecond to millisecond) similar to the chemical shift difference between the two states (so-called intermediate exchange). If the chemical shift difference between the two states is considerable (Ͼ100 Hz), then intermediate exchange will broaden the NMR signals to such an extent that they can no longer be detected. Conformational exchange on time scales significantly longer that the chemical shift difference will result in the different states being detected as separate signals, while conformational exchange on time scales significantly shorter than the chemical shift difference will result in sharp signals at an intermediate frequency.
The size of the 1 H-15 N NOE for backbone amide signals is also sensitive to local mobility of the peptide chain on a picosecond to nanosecond time scale (i.e. significantly faster than the overall tumbling rate of the protein). Regions of the protein backbone with rapid local motion are characterized by large negative 1 H-15 N NOEs.
Free N-TIMP-2-The T 2 relaxation times for the backbone amide 15 N signals of free N-TIMP-2 are shown in Fig. 1A. The mean and standard deviation (indicated on the chart) were calculated after omitting the data for 3 residues which were judged to have unusually long T 2 values (Gly 32 , Gly 92 , and Glu 127 ). Fig. 1A clearly shows five regions where the T 2 values are significantly longer than the mean. The N terminus (residues Ser 4 , His 7 , and Gln 10 ), the end of ␤-strand A, the AB loop and the beginning of strand B (the AB ␤-hairpin, residues Ser 31 -Ile 35 , Gly 37 , and Lys 41 ), the region between the Cys 1 -Cys 72 disulfide bond and strand D (residues Asp 77 , Gly 79 -Gly 80 ), the loop between strands D and E (residues Gly 92 -Asp 93 ) and the C terminus of the protein (residues Gly 125 -Glu 127 ). Residue Lys 58 (in the loop between strands B and C) is also raised. These elevated T 2 values suggest that these residues experience rapid internal motion on a picosecond to nanosecond time scale. The 1 H-15 N NOE results (Fig. 1B) identified very similar regions of N-TIMP-2 as having rapid local mobility. Large negative NOEs were recorded for residues near the N and C termini (Ser 4 , His 7 , and Gly 125 -Glu 127 ), in the AB ␤-hairpin (Ser 31 -Lys 41 ), and in the loops between strands B and C (Lys 58 ), the Cys 1 -Cys 72 disulfide bond and strand D (Asp 77 and Lys 82 ) and strands D and E (Gly 92 -Asp 93 ). The T 2 and 1 H-15 N NOE data for free N-TIMP-2 are summarized on the loop diagram in Fig. 2A.
The T 2 data for free N-TIMP-2 also identified several regions of the molecule with significantly shorter T 2 relaxation times than the mean, suggesting exchange between two or more conformations on the microsecond to millisecond time scale (Fig. 1A). Residues Gln 49 -Lys 51 near the end of strand B, residue Ile 60 in the loop between strands B and C, and residues His 97 -Cys 101 in strand E and the EF loop all have T 2 values below a threshold of 1 standard deviation from the mean (Figs. 1A and 2A).
The T 1 values for the backbone amide 15 N signals of free N-TIMP-2 showed very little variation with sequence (data not shown). The average T 1 was found to be 0.807 with a standard deviation of 0.088. The T 1 /T 2 ratios for N-TIMP-2 were used to estimate the overall rotational correlation time ( m ) of the molecule (26). A value of 10 ns (at 35°C) was obtained, which is comparable to the m values reported for several other proteins on the basis of T 1 and T 2 data (26 -28).
N-TIMP-2⅐N-MMP-3 Complex-The T 2 data for N-TIMP-2 bound to N-MMP-3 are shown in Fig. 1C. Significantly elevated T 2 values were found for Cys 13 in helix 1, Lys 22 in strand A, Glu 57 -Lys 58 in the BC loop, Gly 79 -Gly 80 in the loop between the Cys 1 -Cys 72 disulfide bond and strand D, Gly 92 in the DE loop, Ser 117 in helix 2, and Gly 123 -Glu 127 at the C terminus. In addition, the T 2 values for Gly 32 -Asn 33 in the AB ␤-hairpin were also marginally above the threshold (mean plus 1 standard deviation) considered to be significant (Fig. 1C). As observed for free N-TIMP-2, these regions of rapid internal motion were similarly identified in the 1 H-15 N NOE experiment (Fig. 1D). Large negative 1 H- 15  bound to N-MMP-3 (Fig. 1C) were Ile 19 and Thr 21 in strand A, Ile 50 , Lys 51 , and Phe 53 at the C-terminal end of strand B, and His 97 , Ile 98 , and Leu 100 in strand E and the EF loop. The T 2 data suggest that these residues exist in two or more conformations with interconversion on the microsecond to millisecond time scale. The location of these residues are shown in Fig. 2B.
The T 1 data obtained for the complex showed the same general trends in mobility as identified by the more sensitive T 2 and 1 H-15 N NOE data (data not shown). The clearest trend was a decrease in T 1 toward the C terminus of the protein (Tyr 122 -Glu 127 ) indicative of rapid internal motion on a nanosecond time scale. Depressed values of T 1 were also seen near the N terminus (His 7 and Gln 10 ), and for two glycine residues in the loop between the Cys 1 -Cys 72 disulfide bond and strand D (Gly 79 and Gly 80 ). The average T 1 value for N-TIMP-2 was

FIG. 2. Loop diagram of N-TIMP-2 showing secondary structure elements and the backbone mobility for each residue in free N-TIMP-2 (A) and N-TIMP-2 bound to N-MMP-3 (B).
Residues are colored according to the following scheme: red (rapid picosecond to nanosecond internal motion), heteronuclear 1 H-15 N NOE Ͻ Ϫ0.34; green (slow microsecond to millisecond internal motion), T 2 values Ͻ mean Ϫ S.D.; blue (no significant internal motion) 1 H-15 N NOE Ͼ Ϫ0.34 and T 2 Ͼ mean ϩ S.D.; white (no data), residues for which data could not be collected for relaxation analysis; gray (slow microsecond to millisecond internal motion), residues in the CD loop judged to be in intermediate conformational exchange due to very broad (Val 71 and Cys 72 ) and missing (Ser 68 -Ala 70 ) backbone amide signals (12,16). found to increase from 0.81 s for the free protein to 1.48 s in the complex. This increase is consistent with the change in molecular size of the system on complex formation resulting in a longer overall correlation time ( m ). The variation in T 1 values across the sequence for N-TIMP-2 in the complex was found to be substantially greater than that observed for the free molecule (standard deviations of 0.288 and 0.088, respectively). This greater range of T 1 values is thought to reflect the increased anisotropy of the N-TIMP-2 molecule when bound to the proteinase.
The most dramatic mobility difference seen for N-TIMP-2 on N-MMP-3 binding is the loss of rapid local motion in the AB ␤-hairpin (Fig. 2, compare A and B). Residues Asp 34 , Tyr 36 -Asn 38 , and Lys 41 no longer show elevated T 2 values and large negative 1 H-15 N NOEs.

DISCUSSION
Backbone Mobility in Free N-TIMP-2-In free N-TIMP-2 the regions of the protein backbone that showed greatest internal motion on a rapid picosecond to nanosecond time scale were the AB ␤-hairpin (Ser 31 -Lys 41 ), the tight turn between strands D and E (Gly 92 -Asp 93 ), and the C terminus of the protein, which shows increasing mobility from residue Gly 125 onwards ( Fig.  2A). In addition, several other regions showed a more moderate degree of rapid internal motion including the N-terminal region of the protein (Ser 4 and His 7 ), the loop between strands B and C (Lys 58 ), and the loop between the Cys 1 -Cys 72 disulfide bond and strand D (Asp 79 -Lys 82 ). The core ␤-barrel and the two helices of N-TIMP-2 were all found to be comparatively rigid on a picosecond to nanosecond time scale. This picture of N-TIMP-2 is consistent with that seen for other proteins, where in general the highest flexibility of the protein backbone is found in surface loop regions (29). The rapid motion found for the AB ␤-hairpin confirms our earlier suggestion that this region is flexible and able to move through a relatively large conformational space (17). This region is a very prominent feature on the surface of the protein, where it extends away form the core ␤-barrel into the surrounding solvent (12,13). A clear indication of the extent of the rapid mobility of this and other surface regions of free N-TIMP-2 is given in Fig. 3A, where the width of the ribbon representing the backbone topol-ogy of N-TIMP-2 is scaled according to the size of the 1 H-15 N NOE and therefore illustrates the degree of rapid motion.
Regions of high mobility in proteins are usually those least well defined in NMR structures. The average r.m.s.d values for the backbone atoms (N, C, CЈ) of the solution structures determined for N-TIMP-2 are shown in Fig. 4. A comparison of this graph with that for 1 H-15 N NOE (Fig. 1B) reveals that the most poorly defined regions of N-TIMP-2 (i.e. highest r.m.s.d values) also show the largest negative 1 H-15 N NOEs. In addition, there is a distinct correlation between the number of 1 H-1 H NOEderived distance constraints per residue and the extent of rapid mobility. Regions of N-TIMP-2 with large negative 1 H-15 N NOE values also showed significantly fewer than average medium and long range 1 H-1 H NOEs (12).
Two regions of N-TIMP-2 appear to exist in several local conformations that interconvert on a microsecond to millisecond time scale resulting in a significant reduction in their backbone amide T 2 values. The regions that show this behavior were the ␤-bulge at the C-terminal end of strand B (Gln 49 -Lys 51 ) and the second half of strand E and the EF loop (His 97 -Cys 101 ). The conformational exchange observed for Gln 49 -Lys 51 suggests that the core ␤-barrel is not rigid but able to flex slightly at this point where strand B coils tightly to form the ␤-barrel structure (17). Interestingly, residues Gln 49 -Lys 51 are immediately adjacent to Thr 21 across the strands of the ␤-barrel, and two variants of TIMP-2 are known with either Ala (considered to be the wild-type sequence) or Thr (the protein used in this study) at this position (30). The Thr 21 variant is known to be less stable than the wild-type protein (11), and this may contribute to the conformational heterogeneity seen at Gln 49 -Lys 51 . A similar picture is seen for N-TIMP-2 in the complex, but in this case the region of slow internal motion was somewhat larger and includes Ile 19 and Thr 21 , suggesting that the flexibility of the ␤-barrel is greater when N-TIMP-2 is bound to the enzyme.
Backbone Mobility of the Binding Site for N-MMP-3-The binding site on TIMP for the catalytic domain of the MMPs has now been mapped by both NMR spectroscopy and x-ray crystallography (12, 14 -16). These studies have identified a binding site on N-TIMP-2 comprising residues Cys 1 -Pro 5 at the N Site 1 (N Terminus)-15 N relaxation data could only be obtained for Ser 4 , which clearly experiences rapid local mobility. The backbone amide signal of Ser 2 has never been detected in NMR spectra, and the amide signal of Cys 3 (although overlapped in these relaxation experiments) has nevertheless been shown in previous experiments to have average line widths suggesting no additional slow or rapid internal motion.
Site 2 (CD Loop)-The backbone amide signals for Ser 68 -Ala 70 have never been detected in NMR spectra, and we have previously argued that this suggests that they are significantly broadened by exchange between several conformational states on a microsecond to millisecond time scale (12,16) (Fig. 2). Furthermore, the backbone amide line widths for Val 71 and Cys 72 are very broad (hence their omission from the relaxation data), a finding consistent with the view that this entire region is in conformational exchange. This flexibility may play an important role in facilitating the large conformational change suggested for this region on MMP binding by a comparison of the structures available for both free and bound TIMP (12,14).
Site 3 (EF Loop)-The T 2 values determined for Leu 100 and Cys 101 (Figs. 1A and 2A) suggest that this region, like site 2, is also in slow exchange between two or more conformational states on a microsecond to millisecond time scale.
Site 4 (AB Loop)-The backbone atoms of this region (Ile 35 -Arg 42 ) were found to be highly mobile on a rapid picosecond to nanosecond time scale.
The NMR data clearly show that the binding site for the catalytic domain of MMP-3 on N-TIMP-2 is substantially more mobile than the protein as a whole. This property has been reported for a number of protein and nucleic acid binding sites on proteins (31,32) and may be an important factor controlling the specificity of the binding interaction. It has been shown by Kay and co-workers (33,34) that the flexible region of the peptide binding site in the phospholipase C-␥1 SH2 domain has a much wider ligand binding specificity than the more rigid phosphotyrosine binding site.
Mobility Changes Seen on Complex Formation with N-MMP-3-The binding of N-TIMP-2 to N-MMP-3 resulted in a dramatic change in the mobility of the AB ␤-hairpin (Ser 31 -Lys 41 ), with the majority of this region losing the rapid motion seen in the free protein. Only residues Ser 31 -Asn 33 still retain significant mobility in the complex; these residues, however, form a ␤-bulge in strand A that may act as a flexible hinge for the AB ␤-hairpin. The highly mobile regions around Lys 58 , Gly 92 -Asp 93 , and the C terminus of N-TIMP-2 are distant from the MMP binding site and, as expected, retain rapid motion in the complex. The loss of rapid local motion for the majority of the AB ␤-hairpin on binding to N-MMP-3 suggests that this region interacts with the enzyme, and this result strongly supports our earlier NMR chemical shift perturbation studies that showed a change in the environment of the backbone amide resonances of Glu 28 -Lys 41 on binding to N-MMP-3 (12,16). There is currently no high resolution structure for the TIMP-2⅐MMP-3 complex, but our NMR results clearly show that the AB ␤-hairpin plays an important role in complex formation.
The formation of a tight N-TIMP-2⅐N-MMP-3 complex did not have a dramatic effect on the regions of N-TIMP-2 that are in exchange between several conformations on a microsecond to millisecond time scale. T 2 values for the EF loop, which forms part of the interaction site, remained low in the complex, suggesting that this region is not "frozen" into a single conformation by interaction with N-MMP-3. Similarly, the backbone amide signals for Ser 68 -Cys 72 were not detected in spectra of the complex, suggesting that this region is also not locked into a single conformation by its interaction with the enzyme, even though it is thought to undergo a substantial change in conformation on binding (12,13).
Binding Interactions Made by the AB ␤-Hairpin-The role of the AB ␤-hairpin in TIMP-2/MMP binding is of considerable interest. X-ray crystallography revealed that this region makes specific interactions with the catalytic domain of MT1-MMP, particularly the side chain of Tyr 36 , which binds in a hydrophobic groove formed from the MT-loop, a unique feature of the membrane-bound MMPs (15). These findings have been recently confirmed by kinetic analysis of a mutant form of N-TIMP-2, where Tyr 36 was substituted by glycine (35). This mutant shows a 100-fold decrease in the rate of binding (k on ) and a 40-fold increase in the inhibition constant (K i app ) for MT1-MMP, but showed binding kinetics and inhibition constants essentially unchanged to those of the wild-type for several other MMPs (MMP-2, -3, -7, and -13). TIMP-1 lacks the extended AB ␤-hairpin structure present in TIMP-2 and is a comparatively poor inhibitor of MT1-MMP (36). It is clear that the AB ␤-hairpin contributes considerably to the binding of TIMP-2 to MT1-MMP, but otherwise TIMP-1 shows no dramatic difference in MMP specificity and thus far the extended AB ␤-hairpin has not been shown to be critical in other TIMP-2 inhibitory interactions. Interestingly, TIMP-3, which also lacks the AB ␤-hairpin, binds to MT1-MMP with similar affinity to TIMP-2 (37) and must, therefore, make additional binding interactions with the enzyme that in some way compensate for the loss of the extended AB ␤-hairpin structure.
The positioning of the extended AB ␤-hairpin at one end of the inhibitory wedge of TIMP-2 suggests that this structure will interact with all MMP catalytic domains on complex formation. On binding to an MMP catalytic domain, the AB ␤-hairpin must move away from the core ␤-barrel of the molecule to allow the proteinase sufficient access to interact at the TIMP inhibitory site. This movement is quite considerable (up to 8 Å in the case of TIMP-2 binding to MT1-MMP; Ref. 13) and may be facilitated by the ␤-bulge at Ser 31 -Gly 32 in strand A that could act as a hinge. This role is supported by the 15 N relaxation data, which show that this putative hinge region has high mobility in both free and bound forms of the inhibitor.
The nature of the interaction of the AB ␤-hairpin with MMP catalytic domains will have a significant effect on the overall binding energy and kinetics of inhibition. The high mobility observed for this region may be essential for it to adopt complementary conformations with the wide range of MMP catalytic domains. The contribution of this interaction to the overall binding energy will depend on the nature of the interactions formed, but in order to increase the overall binding energy the strength of favorable contacts will first have to overcome the loss of entropy on binding and it is clear that the contribution of the extended AB ␤-hairpin will vary from case to case. For some MMPs, like MT1-MMP, specific interactions mediated by this region (e.g. Tyr 36 ) will help to offset the loss of entropy, and hence overall this region contributes positively to complex formation. In other cases, the nature of the close contacts may simply balance the loss of entropy resulting in little or no effect on the binding affinity, or indeed, may act to reduce the overall binding energy by making non-favorable interactions with the proteinase. The extended ␤-hairpin of TIMP-2 may also have an important role in controlling the rate of TIMP-2/MMP interactions by affecting the way the molecules orientate to one another before contact or by restricting access of the incoming MMP to the TIMP inhibitory site. The precise contribution of the AB ␤-hairpin to the binding kinetics and affinity of TIMP-2/MMP interactions will require careful future studies on engineered forms of TIMP-2 where this region is excised or modified by site-directed mutagenesis.