Phage Display of Tissue Inhibitor of Metalloproteinases-2 (TIMP-2)

Tissue inhibitor of metalloproteinases-2 (TIMP-2) is a broad spectrum inhibitor of the matrix metalloproteinases (MMPs), which function in extracellular matrix catabolism. Here, phage display was used to identify variants of human TIMP-2 that are selective inhibitors of human MMP-1, a collagenase whose unregulated action is linked to cancer, arthritis, and fibrosis. Using hard randomization of residues 2, 4, 5, and 6 (L1) and soft randomization of residues 34–40 (L2) and 67–70 (L3), a library was generated containing 2 × 1010 variants of TIMP-2. Five clones were isolated after five rounds of selection with MMP-1, using MMP-3 as a competitor. The enriched phages selectively bound MMP-1 relative to MMP-3 and contained mutations only in L1. The most selective variant (TM8) was used to generate a second library in which residues Cys1–Gln9 were soft-randomized. Four additional clones, selected from this library, showed a similar affinity for MMP-1 as wild-type TIMP-2 but reduced affinity for MMP-3. Variants of the N-terminal domain of TIMP-2 (N-TIMP-2) with the sequences of the most selective clones were expressed and characterized for inhibitory activity against eight MMPs. All were effective inhibitors of MMP-1 with nanomolar Ki values, but TM8, containing Ser2 to Asp and Ser4 to Ala substitutions, was the most selective having a nanomolar Ki value for MMP-1 but no detectable inhibitory activity toward MMP-3 and MMP-14 up to 10 μm. This study suggests that phage display and selection with other MMPs may be an effective method for discovering tissue inhibitor of metalloproteinase variants that discriminate between specified MMPs as targets.

Normal biological processes, including embryo implantation, developmental remodeling of the extracellular matrix, and wound healing, require active MMPs. 5 Excess active MMPs can have pathological effects and are tightly regulated at the levels of transcription, zymogen activation, and inhibition by endogenous high affinity protein inhibitors, the TIMPs. Disruption of the balance between active MMPs and their inhibitors results in diseases linked to unregulated matrix turnover, including arthritis, cancer, cardiovascular diseases, nephritis, neurological disorders, tissue ulceration, and fibrosis (1)(2)(3)(4). The mammalian TIMPs are a family of four two-domain proteins (TIMP-1-4), with an N-terminal domain of ϳ125 amino acids and a C-terminal domain of ϳ65 amino acids; each domain is stabilized by three disulfide bonds. They show 41-52% identity in pairwise sequence comparisons.
Engineered TIMPs that are specific inhibitors of individual or restricted groups of metalloproteinases have potential applications in the treatment of diseases associated with excess metalloproteinase activities, including arthritis and cancer (6,7). The N-terminal domains of TIMPs (N-TIMPs) can be expressed separately, are fully active as metalloproteinase inhibitors, and have been widely used in studies of TIMP/MMP interactions (6). In the crystallographic structures of TIMP or N-TIMP complexes with MMPs (8 -12), the core of the TIMP interaction site is a surface ridge formed by the N-terminal five residues, Cys 1 -Ser-Cys-Ser-Pro 5 in TIMP-2, and the loop connecting ␤-strands C and D (CD loop), residues Ser 68 -Ser-Ala-Val-Cys 72 , that are covalently joined by the disulfide bond between Cys 1 and Cys 72 . This ridge interacts with the MMP active site and is oriented so that the conserved N-terminal Cys 1 of the TIMP coordinates the metal ion through the ␣amino group and carbonyl group. Residue 2, serine or threonine in mammalian TIMPs, interacts with the S1Ј specificity pocket of the MMP, whereas residue 4 interacts with the S3Ј subsite, and Ala 70 and Val 71 interact with the S2 and S3 subsites, respectively. The loops connecting ␤-strands A and B and strands E and F and the C-terminal end of ␤-strand D make variable interactions with the MMP in different complexes (6). Previous studies have shown that specific substitutions for residues in the TIMP interaction site can strongly affect its relative affinity for different MMPs and ADAMs; TIMPs with restricted specificity for groups of MMPs have been developed by rationally combining mutations that enhance selectivity (1,(13)(14)(15). However, rational design has limited value because mutations at different sites do not necessarily have additive effects on the free energy of binding (1). TIMPs have relatively large interaction sites for MMPs so that systematic multisite mutagenesis is an impractical approach because of the enormous potential for sequence variation., e.g. saturation mutagenesis of five sites can generate 3.2 ϫ 10 6 sequence variants. To overcome this, we have used phage display to attempt to identify and isolate mutants of TIMP-2 that are specific for MMP-1cd. Large combinatorial phage libraries carrying mutants of human TIMP-2 were panned using positive selection with MMP-1cd combined with negative selection using MMP-3cd to identify MMP-1selective TIMP-2 variants. These two MMPs were used for selection because they have been identified as a cancer target and anti-target, respectively (16). Several studies support the choice of MMP-1 as a target for tumor metastasis inhibition. For example, gene profiling studies identified MMP-1 as important gene rendering the metastatic potential of breast cancer (17, 18); up-regulation of MMP-1 was associated with poor prognosis in cancer patients (19). MMP-3 is considered to be an anti-target because studies indicated that it might have protective action during tumorigenesis (20,21). N-TIMP-2 variants corresponding to several of these positive clones were expressed in Escherichia coli as inclusion bodies, folded in vitro, and characterized for their inhibitory activity toward an array of full-length and truncated MMPs. A double mutant S2D/S4A has been identified as a highly potent inhibitor of MMP-1 that is essentially inactive against MMP-3. The structural basis of MMP selectivity in the mutant was investigated by modeling docked complexes of MMPs with N-TIMP-2 variants.
Oligonucleotides-Equimolar DNA degeneracies are represented using the IUB code (K ϭ G/T, N ϭ A/C/G/T, V ϭ A/C/G, and W ϭ A/T) ( Table 1). Doping codons are represented as follows: 5 ϭ A; 6 ϭ G; 7 ϭ C; 8 ϭ T, with 70% of WT and 10% of each of the other bases at mutated sites.
Display of TIMP-2 on M13 Phage-Full-length human TIMP-2 was displayed on the surface of M13 bacteriophage by modifying a previously described phagemid pS2202d (22). The fragment of pS2202d encoding for the gD tag plus Erbin PDZ domain was replaced with a DNA fragment encoding for fulllength human TIMP-2, using standard molecular biology techniques. The resulting phagemid (p3TIMP2Cys) contained an open reading frame that encoded the maltose-binding protein secretion signal, followed by TIMP-2 and ending with the M13 minor coat protein p3. E. coli cells harboring p3TIMP2Cys were coinfected with M13-KO7 helper phage, and cultures were grown in 30 ml of 2YT medium supplemented with 50 g/ml carbenicillin and 25 g/ml kanamycin at 30°C overnight. The propagated phage was purified using the standard protocol (23) and resuspended in 1 ml of TBT buffer (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5% BSA, and 0.1% Tween 20) to produce phage particles that encapsulated p3TIMP2Cys DNA and displayed full-length TIMP-2. The display level was analyzed using phage ELISA.
Phage ELISA-The purified phage was diluted in a 1:3 series with TBT buffer and was applied to 96-well MaxiSorp immunoplates coated with the target proteins (anti-TIMP-2, MMP-1cd, or MMP-3cd) and blocked with 0.5% BSA (Fig. 1). The plate was washed, and the bound phage was detected with anti-M13-HRP followed by tetramethylbenzidine substrate. In these assays, phage binding to BSA was tested in parallel as the control. The clones, whose binding signal to target is more than five times higher than that of BSA, are considered to be positive clones.
Library Construction and Sorting-Libraries of TIMP-2 mutants were constructed using the Kunkel mutagenesis method (24). The following three regions on TIMP-2 were mutated: Cys 1 -Val 6 (L1), Asp 34 -Ile 40 (L2), and Pro 67 -Gly 73 (L3). The stop template is the single strand DNA of p3TIMP2Cys containing three stop codons in the L1 region and was used to construct four libraries, L1, L1L2, L1L3, and L1L2L3, by mutating regions of L1, L1 plus L2, L1 plus L3, and all three regions. The combined four libraries contained ϳ2 ϫ 10 10 unique members, and they were pooled for sorting.
Libraries were cycled through rounds of binding selection using a standard protocol (23), except that 20 mM Tris-HCl, pH 7.5, containing 150 mM NaCl was used instead of phosphatebuffered saline, with MMP-1cd coated on 96-well MaxiSorp

MMP-1 Selective Variants of TIMP-2
immunoplates as the capture target and, starting from round 3, with 1.7 M MMP-3cd as a competitor in solution. Phages were propagated in E. coli XL1-blue with M13-KO7 helper phage at 30°C as described previously (23). After five rounds of binding selection, individual phage clones were picked, propagated, and purified as described above and were analyzed with phage ELISA. Positive clones were subjected to DNA sequence analysis and competitive phage ELISA.
Competitive Phage ELISA-Phages corresponding to the positive clones from phage ELISA analysis were subjected to competitive phage ELISA. Serial dilutions of free soluble enzymes (MMP-1cd or MMP-3cd) within a concentration range of 5-5000 nM were preincubated for 1 h with phage particles at a fixed concentration within the linear range of the corresponding binding curve (obtained from phage ELISA). The solutions were transferred to 96-well MaxiSorp immunoplates coated with MMP-1cd followed by BSA blocking and incubation for 15 min. The plate was washed, and the bound phages were detected with anti-M13-HRP followed by tetramethylbenzidine substrate. The IC 50 values are the mean concentrations of free enzyme that blocked 50% of phage binding to the immobilized MMP-1cd.
Construction of N-TIMP-2 Mutants-N-TIMP-2 mutants ( Table 2) were generated by PCR, using pET-3a plasmid (Novagen) containing the N-TIMP-2 cDNA as template. All PCRs were carried out using a hot start for 4 min at 96°C followed by 35 cycles of 96°C for 1 min, 50°C for 1 min, and 72°C for 1 min and a final extension for 10 min at 72°C. The PCR products were cloned back into the pET-42b vector (Novagen) using NdeI and BamHI restriction sites. All plasmid constructs were confirmed by automated DNA sequencing carried out at Davis Sequencing, LLC (Davis, CA).
Expression, Purification, and Folding of N-TIMP-2 and Variants-Wild-type (WT) N-TIMP-2 and mutants were expressed in E. coli BL21 (DE3) gold cells (Stratagene) and were purified and folded in vitro as described previously (1). Inclusion bodies containing N-TIMP-2 or a mutant were dissolved in 8 M urea containing 20 mM Tris-HCl, pH 8.0, and 10 mM DTT. Before separating insoluble cell debris from the lysate, the solution was treated with 20 g/liter LRA (calcium silicate hydrate) for 1 h at room temperature to remove endotoxin and DNA (25). The cleared lysate was centrifuged using a Beckman-Coulter ultracentrifuge at 30,000 rpm to remove insoluble material, and the soluble fraction was applied at a flow rate of 1 ml/min to a column of Q-Sepharose (GE Healthcare, 1.5 ϫ 10 cm) pre-equilibrated with 20 mM Tris HCl, pH 8.0, containing 8 M urea. The column was washed until the absorbance at 280 nm was reduced to zero and was then eluted with 60 ml of salt gradient of 0 -0.5 M NaCl at a flow rate of 1 ml/min. Fractions containing N-TIMP-2 were identified by SDS-PAGE, pooled, and folded in vitro as described for N-TIMP-1 except that the pH of the folding buffer was 8.0. The folded protein was applied to a column of DEAE-Sepharose (GE Healthcare, 1.5 ϫ 10 cm) pre-equilibrated with 20 mM Tris-HCl buffer, pH 8.0. The column was washed with the same buffer and eluted with a 60-ml gradient of 0 -0.5 M NaCl at a flow rate of 1.5 ml/min. The eluate was collected in 4.5-ml fractions, and those containing N-TIMP-2 were identified by SDS-PAGE and by assaying for inhibitory activity against MMP-1cd. The most active fractions were pooled and concentrated. The expression, purification, and folding of MMP-1cd, MMP-3cd, MMP-9cd, and MMP-14cd were performed as described previously (1,15).
Before conducting the assays, different concentrations of inhibitor were incubated with the appropriate concentration of MMP at 37°C for 3 h before addition of the NFF-3 substrate (7-methoxycoumarin-4-yl)acetyl-Arg-Pro-Lys-Pro-Val-Glu-Nva-Trp-Arg-Lys(dinitrophenyl)-NH 2 ) to a final concentration of 3 M (for MMP-3) and Knight substrate to a concentration of 1.5 M for all other MMPs. The preincubation and assays with MMP-9 were conducted at 20°C because of stability problems at 37°C. Reaction velocities were measured as the slope of the linear portion of the fluorescence curve. The percentage of residual MMP activity was calculated by dividing the velocities measured with inhibitor by the velocities measured without inhibitor (v/v 0 ). K i(app) values for tight binding inhibition (Ͻ100 nM) were calculated by fitting inhibition data to Equation 1, For weak binding inhibition (K i Ͼ100 nM), the data were fitted to Equation 2, where v is the experimentally determined reaction velocity; v 0 is the activity in the absence of inhibitor; E is enzyme concentration; I is inhibitor concentration, and K is the apparent inhibi-  ). As discussed previously (1), K i(app) is essentially identical to the true K i value because the substrate concentration is very low relative to the K m . For stoichiometric titration of N-TIMP-2 and mutants, increasing concentrations of the inhibitor were incubated with MMP-3cd (300 nM) for 4 h at 37°C, diluted 300-fold with TNC buffer, and immediately assayed with 1.5 M NFF-3 substrate, as described above. Residual MMP activity (%) was calculated as described above and plotted against a molar ratio of TIMP/ MMP (0 -4 in this case). The stoichiometry was determined by linear regression analysis of the appropriate data points. For weak binding mutants, it was assumed that the fraction of active inhibitor was 45%.
Modeling of Complexes and Analyses-Swiss Model and Patchdock were used to generate models of MMP⅐TIMP complexes. Inhibitor-free MMP structures were used as receptor PDB files and protease-free TIMP structures as inhibitor PDB files. PDB files of 966c, 1qib, 1slm, 1mmp, 1jan, 1gkd, 2ow9, 1bqq, 1uea, 2e2d, and 2j0t were used for docking MMPs (MMP-1-3, -7-9, -13, and -14) with N-TIMP-2 and variants. Inhibitorfree MMPs and protease-free TIMPs structures were extracted from PDB files. These analyses provided a set of 50 PDB files of docked structures. The structure with a good fit between the MMP and TIMP was usually the one with the highest score, but some models with high scores were eliminated based on structural criteria. The docked PDB file was further subjected to 100 -200 energy minimization cycles to refine the molecular contacts between the two molecules. The criteria used for the best docked structure were as follows: (i) Zn 2ϩ -Cys 1 coordination distance between 2.0 and 3.5 Å; (ii) positioning of N-terminal binding ridge of TIMP in close contact with the variable S1Ј wall forming residues; (iii) arrangement of loops A-B and C-D connector, and (iv) P1Ј residue-catalytic Glu coordination. van der Waals (VDW) contacts or overlaps were identified using the Findclashes/contacts plugin of Chimera beta version 1 build 2429 (26). Previously, Arumugam et al. (27) and Iyer et al. (28) have analyzed van der Waals contacts of TIMP⅐MMP VDW complexes. Mutations were introduced in N-TIMPs using program COOT (29), SWISSMODEL (30), and swapaa command in Chimera. Energy minimization calculations were performed using the GROMOS96 plugin in the DEEPVIEW (sPDBv) (31) and the minimize structure command in Chimera to refine molecular contacts between the protease and the inhibitor. According to Chimera, the overlap between two atoms is defined as the sum of their VDW radii minus the distance between them and minus an allowance for potentially hydrogen-bonded pairs: overlap ij ϭ rVDW i ϩ rVDW j Ϫ d ij Ϫ allowance ij . An allowance of Ͼ0 reflects the observation that atoms sharing a hydrogen bond can come closer to each other than would be expected from their VDW radii. It is only subtracted for pairs composed of a possible hydrogen bond donor atom (or donor-borne hydrogen) and possible acceptor atom.
The default criteria were used for clash detection. The Helix Systems web server (molbio.info.nih.gov) was used to identify H-bonding pairs in the modeled complexes.

Display of TIMP-2 Mutant Libraries on M13
Phage-Fulllength human TIMP-2 was cloned into pS2202d. Phages harboring p3TIMP2Cys were purified from 30 ml of XL1 culture, and the TIMP-2 display level was determined with anti-TIMP-2 (Oncogene Products) and its ability to bind to the catalytic domains of MMP-1 and MMP-3 as determined by phage ELISA. TIMP-2 displayed on M13 phage bound to anti-TIMP-2 strongly but weakly to MMP-1cd and MMP-3cd (Fig. 1). The affinity between phage-displayed TIMP-2 and MMP-1cd or MMP-3cd was similar, and the binding was specific, compared with helper phage M13KO7 binding to the same enzymes. Regions L1 (Cys 1 -Val 6 ), L2 (Asp 34 -Ile 40 ), and L3 Pro 67 -Gly 73 of TIMP-2 contribute to the protein/ protein interface in the complex of TIMP-2 with MT1-MMPcd (10). L1 has a particularly important role among these three regions because it blocks the catalytic site of MT1-MMP and is critical for TIMP inhibitory activity. We designed the library to hard-randomize L1 with NNK codons and soft-randomize regions L2 and L3 with the doping codons; cysteine residues in these regions were not varied because of their structural roles.
Phages were selected using MMP-1cd, with MMP-3cd being added as a competitor in later rounds. Colonies were collected and amplified after 5 rounds of selection in 30 ml of XL1 cultures, and purified phages were tested for binding to both MMP-1cd and MMP-3cd. Clones that bound to both MMPs were further tested for selectivity using phage competition ELISA. The results showed that clones TM1, TM8, TM13, and TM14 encode TIMP-2 variants that show selectivity between MMP-1cd and MMP-3cd, in contrast to WT TIMP-2, which does not show any selectivity in this assay (Table 3). In particular, TM8 appeared to be the most selective because it showed an apparent affinity for MMP-1cd similar to that of WT TIMP-2 and more than 10 times lower affinity for MMP-3cd.
The sequences of these clones (Table 3) indicate that mutations are present only in the L1 region. In an attempt to improve selectivity, the Cys 1 -Gln 9 region of TM8 (excluding cysteines) was soft-randomized, and the phages were subjected to selection as described above. 25 clones were screened after four rounds of sorting. Four clones (TM8 -12, TM8 -19, TM8 -22, and TM8 -23) showed affinities for MMP-1cd similar to WT but reduced affinity toward MMP-3cd. Nevertheless, none of the TM8-derived variants showed higher selectivity than their parental clone, TM8, in a phage competition ELISA. The N-terminal inhibitory domains of the candidates identified above were subsequently quantitatively characterized as inhibitors of MMP-1cd and MMP-3cd and an array of other MMPs.
Characterization of TIMP-2 Variants-We generated the coding sequences for the N-terminal domains of selected TIMP-2 variants identified by phage display, including TM8 and the TM8-derived library variants by site-specific mutagenesis (Tables 2 and 4). TM8 contained substitutions of Asp and Ala for Ser 2 and Ser 4 , respectively, residues that are known to influence specificity (6). The other variants obtained have several combinations of amino acids at these sites. TM8 -12 (S2V/ S4V), TM8 -23 (S2D/S4A), and TM8 -24 (S2G) mutants have additional substitutions for residues 5-9. Also, mutants with single site S2D and S4A substitutions were constructed to investigate the contributions of the individual mutations to MMP-1 specificity. The oligonucleotides used to generate these mutants are listed in Table 2. The N-TIMP-2 variants were expressed in E. coli as inclusion bodies, extracted, and partially purified using anion-exchange chromatography in 8.0 M urea. The denatured protein was treated in vitro to promote native folding and disulfide bond formation under appropriate redox conditions (see "Experimental Procedures") and finally purified to homogeneity using gel filtration. The purified mutants showed single bands on SDS-PAGE with a mobility corresponding to the expected molecular mass.
Inhibitory Properties of Mutants-All of the mutants identified by phage display were selective inhibitors of MMP-1 over MMP-3 and also poor inhibitors of MMP-14 (Table 5). TM8 (S2D/S4A) did not inhibit either MMP-3cd or MMP-14 at concentrations up to 10 M and is an ϳ1400and 20-fold weaker inhibitor of MMP-2 and MMP-9, respectively, when compared with WT N-TIMP-2. This mutant is also an ϳ70and ϳ150fold weaker inhibitor of MMP-7 and MMP-8, respectively. TM8 -23 carried the same substitutions for residues 2 and 4 as TM8 along with additional substitutions for residues 7 and 9 and had a pattern of inhibitory specificity that is very similar to TM8. These results suggest that N-terminal residues 2 and 4 are critical for generating inhibitory selectivity between MMP-1 and -3. Overall, TM8 was the most selective inhibitor for MMP-1 relative to MMP-3 that was identified in this study (see Fig. 2). To investigate the contributions of the two substitutions present in this variant, the effects of the individual substitutions S2D and S4A on MMP selectivity were investigated. The S2D mutation was found to have a major influence on specificity for MMP-1 when compared with MMP-2, MMP-3, MMP-7, and MMP-14. However, like the T2G mutation in N-TIMP-1, S2D had only a small effect on binding to MMP-9. In contrast, the S4A mutation weakened binding to MMP-9 about 200-fold. This mutant was found to be a partial inhibitor of MMP-1cd. Although it inhibits MMP-1cd effectively at nanomolar concentrations, the level of inhibition did not exceed 80% even at a concentration of 1 M. A similar pattern was previously noted for the inhibition of MT1-MMP by the P5A mutant of N-TIMP-1 (1). The effect of S2D and S4A on the free energy of binding is synergistic in its selectivity toward MMP-1 when compared with MMP-9. Clearly, the S4A substitution did not have additive effects with the S2D mutation on the free energy of binding to any other MMPs studied. This suggests that residue 4 is critical for selectivity toward MMP-9 presumably via its S3Ј subsite interactions.

values for phage-displayed TIMP-2 variants from the screening of libraries that mutate three regions on TIMP-2
The data were obtained by competitive phage ELISA as described under "Experimental Procedures." All mutants contained WT sequences in L2 and L3.   Table 5). The second of these mutants was a particularly effective inhibitor of MMP-9 (K i of 2.4 nM) and MMP-13 (K i of 0.2 nM) relative to other MMPs that were tested. Previous studies on the effects of substitution of Thr 2 of N-TIMP-1 over its affinity for MMP-1-3 have shown that the Val 2 mutant increased selectivity for MMP-1 relative to MMP-2 and MMP-3 (13). It appears from this study that the slightly bulkier and more hydrophobic Val substitution at residues 2 and 4 inter-acts well with the deep S1Ј pockets of MMP-2, MMP-9, and MMP-13 but not with the shallow S1Ј pocket in MMP-7.
Modeling and Docking-Because of the lack of structural information on complexes of TIMP variants with several MMPs, we used molecular modeling in an attempt to understand the structural basis of selectivity in the TM8 and others. Table 6 summarizes the van der Waals contacts of residues of WT N-TIMP-2 in comparison with TM8 in their respective modeled complexes with MMP-1cd and MMP-3cd, which suggest that the substitutions in TM8 substantially reduce the number of interactions. In case of the MMP-1⅐TM8 complex, the effects of the substitutions are projected to be much smaller. The effects of single site mutations emphasize the important role played by P1Ј substitutions on selectivity. This view is supported by an analysis of our modeled structure of the N-TIMP-2⅐MMP-3 complex. Ser 2 of wild-type N-TIMP-2 makes six van der Waals contacts with the base of the S1Ј pocket, whereas in the modeled TM8⅐MMP-3 complex, Asp 2 of the inhibitor makes 10 contacts with the S1Ј wall-forming residues ( Table 6). The large negative effect of the single site Ser 2 to Asp mutation on the affinity for MMP-3 appears to arise, at least in part, from increased separation between the protease and inhibitor in the S1Ј subsite. The distance between O␦2 of Asp 2 of TM8 and O⑀2 of Glu 202 of MMP-3 is about 4.1 Å, whereas the distance between atoms O␥ of Ser 2 and O⑀2 of Glu 202 in the complex with wild-type N-TIMP-2 is 2.4 Å. This increased separation appears to arise from charge repulsion between the side chains of Asp 2 of TM8 and Glu 202 of MMP-3. Charge repulsion may also be responsible for the greater separation between C␣ of Cys 1 and Zn 2ϩ in the modeled TM8⅐MMP-3 complex (3.0 Å) as compared with N-TIMP-2⅐MMP-3 complex (2.4 Å) ( Table 6).
The S1Ј subsite of MMP-1 is shallower than that of MMP-3 because of the presence of Arg at position 214, whereas the corresponding residue in MMP-1 is Leu (32). In the modeled complexes, Asp 2 of TM8 has more contacts with residues in the S1Ј pocket than Ser 2 of wild-type N-TIMP-2 particularly with Arg 214 , Tyr 237 , and Ser 239 of MMP-1 (Table 6 and Fig. 3). This appears to result from favorable electrostatic interactions between the negatively charged Asp 2 of the inhibitor and Arg 214 of the MMP that facilitate closer contact between the two proteins, including increased interactions with the N-TIMP-2 A-B loop. In the modeled complexes of MMP-3 with wild-type and mutant N-TIMP-2, there are fewer contacts between the protease and residues 1-5 of the mutant, as well as greatly reduced interactions with the A-B loop, C-D connector,

TABLE 5 K i values for inhibition of different MMPs by N-TIMP-2 variants corresponding to TIMP-2 mutants selected by phage display
The K i values are adjusted for the active concentrations of TIMPs determined by titration. In the case of weak inhibitors where titration was not possible, the K i values were adjusted to using the active fraction of bacterially expressed N-TIMPs of ϳ45% (27,40). P.I. means partial inhibition; ND means no inhibition detected.

CSCS (WT)
CDCA (TM8)  CDCS  CSCA  TM8 -12  TM8 -23  TM8 - and E-F loop in comparison with wild-type N-TIMP-2. The overall effect of the two substitutions is to weaken contacts between the two proteins. Although charge repulsion involving Asp 2 of the mutant appears to be a major factor in the change in specificity, the smaller Ala side chain in position 4 of TM8 appears also to contribute because it makes no van der Waals interactions with MMP-3, whereas Ser 4 of N-TIMP-2 makes nine contacts.

Comparison of the van der Waals interactions of MMP-3 with N-TIMP-2 as compared with TM8 mutant in modeled complexes
The superscript numbers are the number of van der Waals interactions, and the numbers in brackets are the totals.

N-T2 NT2/MMP-1cd TM8/MMP-1cd NT2/MMP-3cd TM8/MMP-3cd
It is interesting that TM8, in contrast to WT N-TIMP-2, has no detectable inhibitory activity toward MT1-MMP (MMP-14). In the crystallographic structure of the TIMP-2⅐MT1-MMP complex (PDB code 1bqq), the side chain OH of Ser 2 of TIMP-2 forms an H-bond with O⑀2 of the catalytic Glu 240 but in the modeled TM8⅐MT1-MMP complex, this H-bond is absent. The distance between atoms O⑀1 of Glu 240 and O␥ of Ser 2 is 2.9 Å whereas the distance between atoms O⑀1 of Glu 240 and C␤ of Asp 2 in the modeled complex (TM8⅐MT1-MMP) is 4.0 Å. Again, this appears to arise from charge repulsion between the side chain carboxylates of Glu 240 (MT1-MMP) and Asp 2 (TIMP). The Asp 2 side chain is also distant from residues Phe 198 -Leu 199 -Ala 200 in the S1Ј subsite making unfavorable contacts with the S1Ј wall forming Phe 260 and Tyr 261 of MT1-MMP. This appears to tilt TM-8 away from the active site cleft perturbing interactions between the AB-loop of the inhibitor and the MT-loop of MT1-MMP. Ser 4 makes three favorable van der Waals contacts with MT1-MMP, whereas Ala 4 in the modeled complex with TM8 makes just a single contact with the S3Ј subsite. Ser 2 and Ser 4 of wild-type N-TIMP-2 make 14 and 3 contacts with the S1Ј and S3Ј subsites of MMP-3, respectively. Although the EF loop of WT N-TIMP-2 makes eight van der Waals interactions with MT1-MMP as compared with 18 interactions in the modeled TM8⅐MT1-MMP complex, altogether wild-type N-TIMP-2 makes 16 more interactions with MT1-MMP (1bqq) as compared with its modeled complex with TM8 ( Table 7).
The K i values of N-TIMP-2 mutants containing the single site Ser 2 to Asp and Ser 4 to Ala mutations with different MMPs suggest that the Ser 2 to Asp mutation is a major contributor to the selectivity for MMP-1 relative to MMP-2, -3, -7, and -14 ( Table 5). The Ser 4 to Ala substitution alone has smaller effects on selectivity but modulates the effects of the Ser 2 to Asp mutation. It appears from this that the interactions between the N terminus of the S2D mutant and S1Ј pocket of the protease in their complex are decreased relative to those in the WT N-TIMP-2⅐protease complex. For example the CO group of Pro 221 in the lining of the S1Ј pocket H-bonds with the amino group of Cys 3 in the modeled TM8⅐MMP-3 complex but not in the WT N-TIMP-2⅐MMP-3 complex, although the peptide NH of Tyr 223 forms a hydrogen bond with the CO group of Cys 3 in both models. In the crystallographic structure of the MMP-1⅐N-TIMP-1 complex, O␥ of Thr 2 of N-TIMP-1 makes a hydrogen bond with Pro 238 (3.1 Å), but the corresponding residue, Ser 2 , of N-TIMP-2 does not H-bond with Pro 221 in the corresponding modeled complex. Cys 3 is predicted to H-bond with Pro 221 in the TM8⅐MMP-3 complex indicating some rearrangement and possible unfavorable energetic effects arising from the conformation of the Asp 2 side chain. However, in the modeled TM8⅐MMP-3 complex Asp 2 forms four side chain H-bonds with Ala 165 (2) , Glu 202(1) , and Leu 164(1) as compared with three H-bonds formed by Ser 2 of WT N-TIMP-2.
TM8 is also a weak inhibitor of MMP-7, matrilysin, which has the smallest S1Ј pocket among all MMPs, being lined by bulky residues such as tyrosine and valine. The substitution of residues with large side chains at the P1Ј site of peptide substrates is less tolerated by the shallow S1Ј pocket of MMP-7 as compared with the deep S1Ј pocket of the gelatinases (32). TM8

͓4͔
Phe 107 (2) ͓2͔ Ala 202 (3) , Asn 208 (1) , and also with Tyr 423 , which the shorter Ser cannot reach. Ser 2 interacts with His 411 , whereas Asp 2 does not make any interactions with this S1Ј depth-defining residue. Modeling studies suggest that the deep S1Ј subsite of MMP-9 can accommodate the longer Asp 2 side chain. The increased tolerance of this site is consistent with the relatively small effect of a T2R mutation in N-TIMP-1 on its affinity for MMP-9 as compared with MMP-1 and MMP-7 (1).
Here, combinatorial libraries of human TIMP-2 were subjected to positive selection with MMP-1cd followed by a competitive negative selection with MMP-3cd to identify and isolate variants that selectively inhibit MMP-1. Selection from the L1 TIMP-2 library produced TM8 that contained substitutions of Asp for Ser 2 and Ala for Ser 4 . From a library in which residues Cys 1 -Gln 9 of the TM8 variant were randomized, clones TM8 -12, TM8 -19, TM8 -22, and TM8 -23 were selected. Further characterization of these clones showed that, overall, TM8 remained the most highly potent MMP-1-specific variant identified in this study (Table 5 and Fig. 2). TM8 selectively inhibited MMP-1 with a K i of 10 nM while being a weaker inhibitor of MMP-2, MMP-7, MMP-8, and MMP-13 and inactive against MMP-3 and MMP-14. It has been previously suggested based on mutagenesis and structural studies that the elongated ABloop in TIMP-2 is the key to MT1-MMP inhibition (39). However, the properties of TM8 and the single site Ser 2 to Asp mutant show that substitutions in the N-terminal region alone can drastically reduce its inhibitory activity toward MT1-MMP. This is consistent with our previous results where a mutant of N-TIMP-1 with a swapped A-B loop from TIMP-2 remained a weak inhibitor of MT1-MMP (1). The K i values of variants with the single substitutions S2D and S2A show that the former makes a major contribution to MMP selectivity, whereas the latter modulates the inhibition toward different MMPs in a more subtle manner. The properties of TM8 show that these two mutations, when combined, have nonadditive effects on the free energy of binding to various MMPs, emphasizing the advantages of employing phage display for identifying selective MMP inhibitors (Table 5). We have recently reported an investigation of the thermodynamic profiles for the interactions of N-TIMP-2 and N-TIMP-1 with MMP-1cd and MMP-3cd (41). The interaction of N-TIMP-2 with MMP-3cd was found to result in a 10-fold more unfavorable enthalpy change than for the interaction with MMP-1cd; both interactions are driven by a large increase in entropy that, in the case of the MMP-1cd interaction, appears to arise from the hydrophobic effect, reflected in a large negative heat capacity change (⌬Cp). For the N-TIMP-2/MMP-3cd interaction, the interaction appears to be driven by an increase in conformational entropy. For the interaction of MMP-3 with TM8, charge repulsion between Asp 2 and Glu 202 is likely to increase the unfavorable ⌬H, whereas interactions involving Asp 2 and Arg 214 of MMP-1cd would be expected to increase the stability of the complex.
Variants isolated from TM8-derived library showed increased selectivity toward several other MMPs. TM8 -12 (CVCVGVRPL) with S2V and S4V showed ϳ200-fold selectivity toward MMP-1 over MMP-3. It, however, maintained affinities to MMP-2, MMP-7, MMP-8, MMP-9, and MMP-13 with K i values ranging from 1.5 to 31 nM similar to that of wild-type N-TIMP-2. TM8 -12 was a poor inhibitor of MT1-MMP (235 nM) ( Table 5). The TM8 -23 (CDCAPVNPM) with S2D and S4A substitutions has a similar inhibitory profile to TM8 (CDCA). This suggests that the substitutions for residues 7 and 9 have relatively minor effects on the inhibitory profile. TM8 -24 (CGCSPMRPQ) showed that the Ser 2 to Gly mutation, which essentially eliminates interactions between the TIMP and the S1Ј pocket, has varying effects on the inhibition of different MMPs. TM8 -24 shows selectivity toward MMP-1 (ϳ60-fold reduction in affinity) relative to MMP-3 (ϳ800-fold reduction in affinity), but its affinity for most other MMPs was dramatically lowered. Surprisingly, it remained an effective inhibitor of MMP-9 and MMP-13 suggesting that interactions outside of the S1Ј pocket play a greater role in the interactions with these two MMPs.
Among the panel of MMPs that we probed with our N-TIMP-2 variants, MMP-1, -2, and -7 were selected as anticancer drug targets and MMP-8, -9, -3, and -14 as anti-targets (16). TM8 has achieved the desired selectivity against MMP-3 and MMP-14 but did not show specificity with respect to MMP-8 and -9. This would limit its potential therapeutic application. In this study, we only used a single MMP for counterselection (MMP-3) but achieved specificity with difference in K i values of over 3 orders of magnitude. We did not attempt to use other MMPs for counter-selection because of the limited availability of other MMPs in large quantity and, not surprisingly, did not reach a similar level of selectivity for them. A similar strategy could potentially be used to develop inhibitors that select between other members of the MMP family. Although this study suggests that L2 and L3 are not involved in selectivity between MMP-1 and MMP-3, it is possible that mutations in these regions could affect selectivity for other members of the MMP family. Therefore, it may be possible to achieve increased selectivity for TM8 against MMP-8 and -9 by counter-selection with these two proteins in libraries containing extensive L2 and L3 randomization.