Total Conversion of Tissue Inhibitor of Metalloproteinase (TIMP) for Specific Metalloproteinase Targeting

Tissue inhibitors of metalloproteinases (TIMPs) are the endogenous inhibitors of the matrix metalloproteinases, the ADAMs (adisintegrin and metalloproteinase) and the ADAM-TS (ADAM with thrombospondin repeats) proteinases. There are four mammalian TIMPs (TIMP-1 to -4), and each TIMP has its own profile of metalloproteinase inhibition. TIMP-4 is the latest member of the TIMPs to be cloned, and it has never been reported to be active against the tumor necrosis factor-α-converting enzyme (TACE, ADAM-17). Here we examined the inhibitory properties of the full-length and the N-terminal domain form of TIMP-4 (N-TIMP-4) with TACE and showed that N-TIMP-4 is a far superior inhibitor than its full-length counterpart. Although full-length TIMP-4 displayed negligible activity against TACE, N-TIMP-4 is a slow tight-binding inhibitor with low nanomolar binding affinity. Our findings suggested that the C-terminal subdomains of the TIMPs have a significant impact over their activities with the ADAMs. To elucidate further the molecular basis that underpins TIMP/TACE interactions, we sculpted N-TIMP-4 with the surface residues of TIMP-3, the only native TIMP inhibitor of the enzyme. Transplantation of only three residues, Pro-Phe-Gly, onto the AB-loop of N-TIMP-4 resulted in a 10-fold enhancement in binding affinity; the Ki values of the resultant mutant were almost comparable with that of TIMP-3. Further mutation at the EF-loop supported our earlier findings on the preference of TACE for leucine at this locus. Drawing together our previous experience in TACE-targeted mutagenesis by using TIMP-1 and -2 scaffolds, we have finally resolved the mystery of the selective sensitivity of TACE to TIMP-3.

Tissue inhibitors of metalloproteinases (TIMPs) are the endogenous inhibitors of the matrix metalloproteinases, the ADAMs (a disintegrin and metalloproteinase) and the ADAM-TS (ADAM with thrombospondin repeats) proteinases. There are four mammalian TIMPs (TIMP-1 to -4), and each TIMP has its own profile of metalloproteinase inhibition. TIMP-4 is the latest member of the TIMPs to be cloned, and it has never been reported to be active against the tumor necrosis factor-␣-converting enzyme (

TACE, ADAM-17). Here we examined the inhibitory properties of the full-length and the N-terminal domain form of TIMP-4 (N-TIMP-4) with TACE and showed that N-TIMP-4 is a far superior inhibitor than its full-length counterpart. Although fulllength TIMP-4 displayed negligible activity against TACE, N-TIMP-4 is a slow tight-binding inhibitor with low nanomolar binding affinity. Our findings suggested that the C-terminal subdomains of the TIMPs have a significant impact over their activities with the ADAMs.
To elucidate further the molecular basis that underpins TIMP/TACE interactions, we sculpted N-TIMP-4 with the surface residues of TIMP-3, the only native TIMP inhibitor of the enzyme. Transplantation of only three residues, Pro-Phe-Gly, onto the AB-loop of N-TIMP-4 resulted in a 10-fold enhancement in binding affinity; the K i values of the resultant mutant were almost comparable with that of TIMP-3. Further mutation at the EF-loop supported our earlier findings on the preference of TACE for leucine at this locus. Drawing together our previous experience in TACE-targeted mutagenesis by using TIMP-1 and -2 scaffolds, we have finally resolved the mystery of the selective sensitivity of TACE to TIMP-3.
Tissue inhibitor of metalloproteinases (TIMP(s)) 1 are the endogenous regulators of the zinc-dependent metalloproteinases (MPs) of the matrix metalloproteinase (MMP), the ADAM (a disintegrin and metalloproteinase), and the ADAM-TS (ADAMs with thrombospondin repeats) families. There are four mammalian TIMPs identified to date (TIMP-1 to -4), and they are all small molecules of ϳ24 kDa in molecular mass. The TIMPs are 40 -50% identical in amino acid sequence (reviewed in Ref. 1), and structural analysis reveals that the molecules are composed of two very distinct N-and C-terminal domains. The N-terminal domain (N-TIMP) encompasses nearly the first two-thirds of the polypeptide, and the domain is made up of five-stranded pleated sheets in the shape of a conical ␤-barrel. The tertiary configuration of the domain is typical of that of an oligonucleotide/oligosaccharide-binding motif (2)(3)(4). It is within the N-terminal domain that the inhibitory activity of a TIMP resides. The C-terminal domain is ϳ8 kDa in mass. In contrast to the N terminus, this domain is less well defined (3,5). There are three disulfide bonds in each domain, but only the full-length and the N-terminal forms of TIMPs have been expressed so far and refolded from Escherichia coli inclusion bodies (6 -10). No success has yet been achieved in expressing the C terminus as an independent entity either in mammalian or in prokaryotic cells.
TIMP inhibits the enzymatic function of an MP by inserting the wedge-shaped edge of its N-terminal domain, the so-called "MMP-binding ridge," into the catalytic groove of the target MP to form a tight but essentially noncovalent 1:1 stoichiometric complex (3,5). The MMP-binding ridge, by its very definition, is composed of the N terminus, the AB-loop, CD-loop, and the EF-loop of the molecule. Despite the high degree of similarity in their tertiary structures, each TIMP has its own range of metalloproteinase (MP) targets, and the profiles of MP inhibition among the TIMPs are acutely variable between the species. For instance, with the exception of several cell surface-bound membrane-type MMPs and MMP-19, all secreted MMPs are sensitive to TIMP-1 inhibition (11,12). TIMP-2, on the contrary, inhibits membrane-type MMPs as well as MMP-19 at subnanomolar affinities (11,12). The ADAMs have slightly different sensitivity profiles in comparison to the MMPs. ADAM-10, for example, is inhibited by TIMP-1 and -3 and not TIMP-2 (13). TACE, in contrast, is selectively inhibited by TIMP-3 (14). Notwithstanding the successful delineation of two co-crystal structures of TIMP⅐MMP complexes, namely TIMP-1/stromelysin-1 (PDB code 1UEA) and TIMP-2/MT1-MMP (PDB code 1BUV), the molecular basis that underpins TIMP/MP selectivity is still not fully understood.
Unlike its counterparts TIMP-1, -2, and -3, the inhibitory characteristics of TIMP-4 with the MP have so far not been investigated in detail. Because it is the last TIMP to be isolated, our knowledge of its functions, biochemical or cellular, is admittedly still fairly limited. Our laboratory is interested in elucidating the molecular basis that governs MP/TIMP interactions, above all members of the cell surface-associated MPs that are of clinical importance, such as the tumor necrosis factor-␣-converting enzyme (TACE). TACE is unique among the MPs because of its versatility in the shedding of a broad range of membrane-bound bioactive molecules (reviewed in Ref. 15). Furthermore, it is selectively inhibited by TIMP-3 and not the other TIMPs (14). Biochemically, it is both interesting as well as challenging to unlock the molecular basis that underpins TACE/TIMP selectivity, and no TIMP is more challenging than TIMP-4. Hence, we chose TACE as the subject of our investigation in this study and TIMP-4 as the prototype for mutagenesis.
In line with the design and progress of the project, the results are divided into three sections. In the first section, we compare the inhibitory characteristics of full-length and the N-terminal domain forms of TIMP-4 (N-TIMP-4) with TACE. In particular, the discrepancy between their affinities and association rates will be examined. In the second part of this paper, we describe the mutagenesis process leading to the creation of a potent, slow, and tight-binding TIMP-4 inhibitor. The relevance of the TIMP-3 AB-loop to the current project will be discussed in detail. The last section under "Results" deals specifically with residue Leu 101 , the pivotal amino acid on the EF-loop that is essential for TACE recognition. Combining our previous TIMP/ TACE mutagenesis findings by using TIMP-1 and -2 as the scaffolds, we can now reveal the molecular basis that underpins the nature of the selective inhibition of TACE by TIMP-3.

EXPERIMENTAL PROCEDURES
Materials-TACE (ADAM-17) enzyme (the catalytic domain of TACE or TACE-473) was the kind gift from Dr. J. David Becherer, Glaxo-SmithKline, Research Triangle Park, NC (16). Human TIMP-4 cDNA was given by Dr. Rob Nuttall and Dr. Dylan Edwards, University of East Anglia, Norwich, UK. Native full-length TIMP-4 was purchased from R&D Systems, Minneapolis, MN. All chemicals and reagents in this study were purchased from Sigma unless otherwise stated. Vent DNA polymerase for mutagenesis and the restriction enzymes for subcloning were obtained from New England Biolabs (Hitchin, Hertfordshire, UK). The fluorescent substrate for TACE assay ((7-methoxycoumarin-4-yl)-acetyl-Ser-Pro-Leu-Ala-Gln-Ala-Val-Arg-Ser-Ser-Ser-Arg-Lys-2,4-dinitrophenyl-NH 2 ) was synthesized and purified by Dr. Graham Knight, Department of Biochemistry, University of Cambridge, as reported in our early papers (17,18).
Construction and Site-directed Mutagenesis of Full-length and N-TIMP-4 Mutants-Full-length and the N-terminal domain forms of TIMP-4 (N-TIMP-4; corresponding to residue Cys 1 to Gly 128 ) were amplified from the human cDNA template by PCR before being subcloned into the E. coli expression plasmid pRSET-c (Invitrogen). A hexahistidine tag was added to the C termini of the TIMP-4 constructs during PCR amplification for the ease of downstream purification. Mutagenesis was achieved by either forward or reverse oligonucleotides, depending on the loci of the mutated sites. All clones have been sequenced to confirm that no unintended mutations have been introduced during the mutagenesis process.
Production, Refolding, and Activity Assessment of Full-length and N-TIMP-4 -The protocols for the production, refolding, and titration of full-length and N-TIMP-4 were almost identical to those of N-TIMP-2 elaborated in our previous papers with the exception that 0.6 M Larginine was included in the refolding solution (17,18). The concentration of active TIMPs in each preparation was determined by titration against a known amount of gelatinase-A (MMP-2) and/or collagenase-3 (MMP-13) as described before (19).
Inhibition Constant Measurement (K i )-Inhibition (K i ) and association (k on ) assays were performed at 27°C constant temperature in fluorescence assay buffer (50 mM Tris-HCl, pH 7.5, 10 mM CaCl 2 supplemented with 0.05% Brij-35, 1% Me 2 SO, 0.02% NaN 3 ) with a PerkinElmer Life Sciences LS-50B spectrofluorimeter equipped with thermostatic cuvette holders (17)(18)(19). TACE (0.22 nM) was preincubated at room temperature with TIMP-4 or its mutants, ranging from 0.1 to Ͼ1000 nM, for a minimum of 2 h prior to steady state (V S ) measurement at 27°C. Reactions were initiated by adding quenched fluorescent peptides to a final concentration of 1 M. All data were fitted into competitive tight binding equations with the computer program Grafit to obtain an estimation of the K i values with Equation 1 (20), Where V o is the rate in the absence of inhibitor; E t is the total enzyme concentration, and I t is the total inhibitor concentration. Association Rate Constant Measurement (k on )-k on measurements were performed by adding TIMP-4 or its mutants (up to 2000 nM) to 80 pM TACE. The rate of inhibition was monitored using a continuous fluorimetric assay at 27°C until steady state was reached. The progress curve was analyzed by using Equation 2 (20), where P is the product concentration; V o is the initial velocity; V S is the steady state velocity; and k is the apparent first-order rate constant of equilibrium between enzyme and TIMP complex. k on values were calculated by linear regression of k on TIMP concentrations.

N-TIMP-4 Is a Far Superior TACE Inhibitor Than
Fulllength TIMP-4 -TIMP-4 is the latest TIMP variant to be cloned, and our understanding of its inhibitory functions with the MMPs, the ADAMs, and the ADAM-TS proteinases in general is therefore very limited. To date, there have been several reports on the inhibitory activity of TIMP-4 with a number of MMPs and ADAM-TSs (21-23) ,but none has so far looked at the ADAMs. As a first step toward delineating ADAM/TIMP-4 interactions, we compared the binding affinity and the association rate of full-length TIMP-4 and its N-terminal form, N-TIMP-4, with the ADAM proteinase, TACE. The results are listed in Table I. As shown, N-TIMP-4 is functionally much more active than full-length TIMP-4; its affinity (K i 8.7 nM) was at least an order of magnitude better than that of the full-length species (K i ϳ120 nM). To rule out the caveat that the poor affinity of refolded full-length TIMP-4 is caused by misfolding in the C-terminal domain, we repeated the measurement using native TIMP-4 expressed and purified from the eukaryotic expression system (K i Ͼ180 nM). The findings are unequivocal, i.e. full-length TIMP-4 is a much poorer inhibitor than N-TIMP-4. Further study on the association profile revealed that N-TIMP-4 is clearly a slow, tight binding inhibitor, whereas full-length TIMP-4 is not. The contrast between the two forms of TIMP-4 is best illustrated by their k on profiles, as shown in Fig. 1 (k on of N-TIMP-4 ϳ6.4 ϫ 10 Ϫ3 M Ϫ1 s Ϫ1 ; whereas no k on could be determined for full-length TIMP-4 because of its low inhibitory activity).
Strategy for Creating a Fully Active, Potent N-TIMP-4 TACE Inhibitor-In contrast to its counterparts the N-TIMP-1 (K i 356 nM) and the N-TIMP-2 (K i 893 nM), N-TIMP-4 is a compara-TABLE I The inhibitory constants (K i ) and association rates (k on ) of full-length and N-TIMP-4 against TACE Native human TIMP-4 was included alongside the refolded full-length TIMP-4 to show that the poor activity of refolded species was not due to mis-folding in the C-terminal domain. tively better inhibitor in terms of TACE inhibition. On the other hand, its potency against TACE is still lagging behind that of the N-TIMP-3 (K i 0.22 nM) ( Table II). As part of our effort in the elucidation of the molecular basis that governs TACE/TIMP selectivity, we wondered if N-TIMP-4 could ever be engineered to be as potent as N-TIMP-3, and if so, which part(s) of the molecule should be targeted for engineering? The most likely candidates, in our conjecture, are the amino acids located at the MMP-binding ridge as these are the parts of the molecule that would be in direct physical contact with the target enzyme upon the formation of TACE⅐TIMP complex. Furthermore, we have demonstrated previously (17)(18)(19) that the selectivity profile of a TIMP variant could be re-modeled by adjusting the amino acid composition of its MMP-binding ridge. Fig. 2 compares the primary sequences of N-TIMP-1, -2, -3, and -4. Clearly, the most conspicuous difference between the four TIMPs is the AB-loop region, and to a lesser extent residues 2 (Ser 2 ) and 4 (Ala 4 ) at the N terminus and residue 100 (His 100 ) on the EF-loop. If our hypothesis on TIMP/MP selectivity is correct, these would be the residues worthy of investigation if the activity profile of N-TIMP-4 is to be tuned against TACE. In this work, our strategy was to divide our mutagenesis targets into two categories. The first category consisted of mutants at the N terminus (i.e. residue Ser 2 and Ala 4 ) and the EF-loop (i.e. residue His 100 ). The second category, on the other hand, was exclusively AB-loop-related. The residues that were deemed to be important for TACE selectivity were either retained or replaced by amino acids that are, according to our modeling prediction, capable of enhancing the affinity with TACE. The goals of the current project are 2-fold. In the short term, we intend to enhance the activity of N-TIMP-4 specifically against TACE for the purpose of finding out the part(s) of the TIMP molecule that is/are involved in TACE recognition and selectivity. If the objective can be attained, our long term goal would be to create a potent, TACE-inhibiting "N-TIMP-4 probe" to answer the many contentious issues surrounding the biological and cellular functions of TIMPs.
First Category N-TIMP-4 Mutants, the N Terminus and the EF-loop-Three mutants were created in this category, namely Ser 2 to threonine (S2T) mutant, Ala 4 to serine (A4S) mutant, and Ala 4 to methionine (A4M) mutant (Table II). S2T and A4S mutants were made because the second and the fourth residues in TIMP-3, the only native inhibitor of TACE, are threonine and serine, respectively (see Fig. 2). On the other hand, Ala 4 was mutated to methionine (A4M) as our previous mutagenesis study with TIMP-3 showed that a bulky, hydrophobic side chain (such as methionine) at position 4 enhanced the affinity of N-TIMP-3 with TACE (24). The findings are shown in Table  II. Among the three, only S2T displayed possible signs of improvement in binding affinity (K i , S2T of 7.7 nM versus wildtype N-TIMP-4 of 8.7 nM). The A4X mutants, regardless of the biochemical nature of the amino acid substitutes, are poorer inhibitors than the wild-type protein (K i , A4S and A4M of 9.5-19 nM).
The reason His 100 was chosen for mutagenesis in this work was because of its close proximity to Leu 101 , a residue that, according to our hypothesis, could be of critical importance in TACE recognition (details will be elaborated below). Being adjacent to Leu 101 , His 100 could in theory exert significant influence upon the movement of the Leu 101 side chain or even the mobility of the EF-loop as a whole; hence the creation of H100T and H100G mutants to mimic the EF-loops of TIMP-1, -2, and -3, respectively (highlighted in Fig. 2). Of the two, H100G exhibited possible signs of improvement in affinity (K i 7.6 nM), whereas H100T mutation severely compromised the activity of N-TIMP-4 with TACE (K i ϳ170 nM).
Second Category N-TIMP-4 Mutants, Fine-tuning the ABloop-None of the mutations carried out at the N terminus and the EF-loop significantly improved the affinity of N-TIMP-4 with TACE. If the affinity of N-TIMP-4 for TACE is to be enhanced to a level on a par with that of the N-TIMP-3 (K i 0.22 nM), it is very unlikely that the mutations so far offer any realistic chance of success. In other words, the answer to the successful creation of a tight-binding N-TIMP-4 variant must lie elsewhere in the molecule. The next promising locus on the MMP-binding ridge was no doubt the AB-loop (Figs. 2 and 3). It is worth noting that the AB-loops of TIMP-2 and -4 are longer than those of the TIMP-1 and -3 by five to six residues (Fig. 3). Furthermore, the amino acid composition of the AB-loop is highly variable among the four TIMPs (Figs. 2 and 3). If the AB-loop is indeed the key that confers TACE-inhibiting activity upon TIMP-3, transplantation of the fragment onto TIMP-4 should generate a TIMP-4 inhibitor with improved affinity against the enzyme. Our modeling exercise on the TIMP-3 AB-loop suggests that the residues most likely to come into contact with TACE in a hypothetical TACE⅐N-TIMP-3 complex are Pro 33 , Phe 34 , and Gly 35 (Fig. 3). The equivalent residues in TIMP-4 Thr 38 , Glu 39 , Lys 40 , respectively, as highlighted in Fig.   3 (upper panel). Could these residues be the key to TACE recognition? As the first step toward delineating the molecular significance of the loop, we created an N-TIMP-4 triple mutant with three amino acid substitutions: T38P, E39F, and K40G. Indeed, the K i value of the resultant mutant (T38P/E39F/ K40G) was an order of magnitude better than that of the wild-type N-TIMP-4 (K i , T38P/E39F/K40G of 0.86 nM versus wild-type of 8.7 nM). Its association rate was almost 2.5-fold that of the wild-type inhibitor (T38P/E39F/K40G, k on of ϳ16 ϫ 10 Ϫ3 M Ϫ1 s Ϫ1 versus wild-type of 6.4 ϫ 10 Ϫ3 M Ϫ1 s Ϫ1 ) ( Table III).
The findings, although impressive, immediately prompted two further questions. 1) Which residue(s) within the three is/are responsible for the improvement in activity? 2) Given that TIMP-4 AB-loop is five amino acids longer than that of the TIMP-3, what would the effects be for shortening the length of the AB-loop of N-TIMP-4 mutant T38P/E39F/K40G to that of the TIMP-3 AB-loop? To address the first question, we dissected the triple T38P/E39F/K40G mutation into six individual single and double mutants as follows: T38P, E39F, K40G, T38P/E39F, T38P/K40G, and E39F/K40G (Table III). Collec-  tively, these six mutants should reveal the key residue(s) that is/are critical for TACE recognition. Among the three single mutants, only T38P demonstrated marginal signs of improvement in binding affinity (T38P, K i of 3.91 nM versus wild-type N-TIMP-4 of 8.68 nM). The affinities of the other two were either impaired (E39F, K i 9.86 nM) or not altered (K40G, K i 8.80 nM).
In contrast, all the double mutants displayed subtle enhancement in activities (K i T38P/E39F, T38P/K40G, and E39F/ K40G between 3.7 and 5.0 nM) (Table III). Most interestingly, the enhancement was noted even in the E39F/K40G double mutant (K i 4.90 nM) that was made up of two single mutations (E39F and K40G) that on their own have either negligible or deteriorating effects on TACE association.
To address the second question, we selected four N-TIMP-4 variants (i.e. wild-type N-TIMP-4, E39F single mutant, E39F/ K40G double mutant, and T38P/E39F/K40G triple mutant) and truncated five residues from their AB-loops to generate a group of "AB-loop-less" versions of the equivalent mutants. The five AB-loop residues selected for removal are Ala 33 , Asp 34 , Pro 35 , Ala 36 , and Asp 37 , chosen purely on the basis of sequence alignment (Fig. 3, upper panel). These AB-loop-less mutants are henceforth designated as follows: (i) ⌬AB, (ii) ⌬AB ϩ E39F, (iii) ⌬AB ϩ E39F/K40G, and (iv) ⌬AB ϩ T38P/E39F/K40G (Table III). The kinetic profiles of these ⌬AB mutants are listed in Table III. In short, all AB-loop-less mutants were considerably weaker than their AB-loop-intact counterparts not only in affinities but also in association rates (K i , ⌬AB mutants varied from 14 to 60 nM; k on ⌬AB mutants in the range of 2-5 ϫ 10 Ϫ3 M Ϫ1 s Ϫ1 versus wild-type N-TIMP-4 of 6.4 ϫ 10 Ϫ3 M Ϫ1 s Ϫ1 ). Notably, the K i values of these ⌬AB mutants were 3-15 times higher than those with intact loops, whereas the drop in k on

FIG. 3. The AB-loops of TIMP-1, -2, -3 and -4 as shown by the schematic (upper panel) and molecular modeling (lower panel).
Note the relatively short length of TIMP-1 and -3 AB-loops in comparison with those of TIMP-2 and -4. Although the AB-loop of TIMP-3 is 5-6 amino acids shorter than those of TIMP-2 and -4, it is, however, critical for metalloproteinase selection and inhibition. In this work, we show that the "Pro 33 -Phe 34 -Gly 35 triad" (highlighted by blue-red-purple; lower panel) of TIMP-3 is the key to effective TACE recognition. In order to convert TIMP-1, -2, and -4 into TACE-active inhibitors, the TIMPs must be equipped with this triad at their respective AB-loops. The upper panel was modified from the AB-loop diagram by Williamson et al. (31). values was only 2-3-fold, an indication that dissociation rate (k off ) is the kinetic factor most affected by the removal of the AB-loop.
Combining S2T and T38P/E39F/K40G to Generate a Potent N-TIMP-4 Mutant-T38P/E39F/K40G is by far the most potent N-TIMP-4 mutant in this study. With a K i value of just over 0.8 nM, its affinity is comparable with that of the N-TIMP-3 (0.22 nM). In an attempt to further increase its potency, we combined the triple mutant with S2T and H100G, the two suspected "good" mutations at the N terminus and EF-loop, singularly as well as collectively, and the results of the combination are shown in Table IV. Indeed, incorporation of S2T resulted in slight enhancement in affinity, the K i of the quadruple mutant S2T ϩ T38P/E39F/K40G (0.73 nM) was marginally better than its prototype, the T38P/E39F/K40G mutant (Table IV). The opposite is true for H100G. Incorporation of this mutation only impaired the affinity of the resultant mutants with TACE, a conclusion that could be clearly derived from the affinity values of the compounded H100G ϩ T38P/E39F/K40G (K i 1.80 nM) and S2T ϩ H100G ϩ T38P/E39F/K40G (K i 2.65 nM) mutants.
L101X Mutations-Leucine 101 is located at the turn of the EF-loop, right before the second disulfide bond (Cys 3 -Cys 102 ) that bridges the loop to the N terminus of TIMP (Fig. 4, disulfide bonds are shown in yellow stick format). The amino acid is conserved in TIMP-2 (Leu 100 ) and TIMP-3 (Leu 94 ) but not in TIMP-1 (Fig. 2). Its equivalent in TIMP-1, as shown in Fig. 2, is threonine (Thr 98 ). The reason the residue is of immense interest to us is because previous mutagenesis investigation using TIMP-1 and TIMP-2 scaffolds demonstrated that the choice of amino acid occupant at this locus has a profound impact on the ability of a TIMP to recognize and inhibit TACE (17,18). As a matter of fact, TIMP-1 could be rendered active against TACE simply by replacing its Thr 98 residue with leucine (T98L) (17). Further mutagenesis analysis with TIMP-1 and -2 scaffolds showed that the presence of a leucine residue at the position was indispensable for the generation of tight-binding, TACE-active TIMP mutants (17,18). Would the same rule apply to TIMP-4?
Using S2T ϩ T38P/E39F/K40G as the prototype, we mutated Leu 101 to all the existing amino acids, except cysteine. The kinetic properties of these L101X mutants are summarized in Table V. Indeed, leucine again fared best (K i S2T ϩ T38P/E39F/ K40G prototype of 0.73 nM), followed by methionine (K i S2T ϩ T38P/E39F/K40G/L101M 3.30 nM), isoleucine (K i S2T ϩ T38P/ E39F/K40G/L101I 3.44 nM), and tyrosine (K i S2T ϩ T38P/ E39F/K40G/L101Y 3.55 nM) (L101X mutations highlighted in boldface for easy recognition). Apart from these three, the majority of the other amino acids was reasonably well tolerated; their K i values ranged from 4 to 25 nM (Table V). The worst residues were no doubt proline (K i 105 nM), tryptophan (K i 185 nM), and aspartate (K i 144 nM); replacement of Leu 101 with these residues resulted in near complete elimination of the activities of the prototype with TACE. A conspicuous aspect worth noting is the discrepancy between the aspartate (K i 144 nM) and the glutamate (K i 13.3 nM) mutants. Despite the most modest difference in their biophysical properties, the residues have vastly dissimilar impact on the activities of the N-TIMP-4 prototype with TACE. Table VI summarizes the latest progress of our TIMP engineering projects aimed specifically at two of the clinically most important membrane-associated MP, i.e. MT1-MMP and TACE. To date, we have engineered two variants of TIMP-1 mutants that are superb inhibitors for MT1-MMP (V4A/P6V/ T98L, K i MT1-MMP 1.66 nM) and TACE (V4S/TIMP-3 AB-loop/ V69L/T98L, K i TACE 0.14 nM) (19,17). Furthermore, we have also generated several mutants based on TIMP-2 and TIMP-4 that are highly effective against TACE (TIMP-2 mutant S2T/ TIMP-3 AB-loop/A70S/V71L, K i 1.5 nM, and TIMP-4 mutant S2T ϩ T38P/E39F/K40G from this work, K i 0.73 nM) (Table VI) (18). What drives us to modify the inhibitory functions of a TIMP in such a specific manner? Above all, what could we gain by rendering TIMP-1, -2, and -4 active against TACE since the enzyme is already well inhibited by its natural inhibitor TIMP-3? There are two reasons for such an endeavor. First, we are interested in unlocking the molecular basis that governs TIMP/TACE selectivity, in particular, the reasons for the ineffectiveness of TIMP-1, -2 and -4 as TACE inhibitors. Notwithstanding the delineation of the NMR/co-crystal structures of TIMP-1 (PDB codes 1D2B and 1UEA) and TIMP-2 (PDB codes 2TMP and 1BUV), the molecular rationale that underpins the activity profiles of a TIMP is still a matter of conjecture. Whether a TIMP is capable of inhibiting a particular MP, be it a MMP, an ADAM, or an ADAM-TS, is a question that even today requires empirical measurement. Neither could a theory on TIMP/MP interactions be formulated by molecular simulation or modeling. For instance, it has been known for years that TIMP-1 is the only TIMP species that is not capable of inhibiting MT1-MMP, and yet the reason for such inactivity was not known until very recently (19). Likewise is the issue surrounding the selective inhibition of TACE by TIMP-3 (17, 18). It is against this background that our current series of TIMP engineering projects was conceived and initiated. As mentioned before, TACE was chosen as the target of our investigation not only because of its prominent roles in inflammatory diseases such as arthritis, but more so on the basis of its discriminate sensitivity to TIMP-3. Hence, the immediate objective of this FIG. 4. The formula for the successful generation of TACE-inhibiting TIMPs. To convert an otherwise inactive TIMP into a fully active one against TACE, there are four rules to be observed as follows: 1) a leucine residue on the EF-loop immediately before the second disulfide bond; 2) a leucine residue on the CDloop immediately before the first disulfide bond; 3) a threonine residue at the P1Ј (i.e. second) position; and 4) a proline-phenylalanine-glycine triad at the AB-loop.  project was to identify the unique amino acid composition that allows TIMP-3 to recognize and form a tight binary complex with TACE and, if such residues could be identified, the possibility of transferring them to another TIMP species. TACE is clinically relevant as it is involved in the ecto-domain shedding of a great variety of biologically active cell surface molecules (15). The ability to control its activities is of utmost importance in the maintenance of the general well being of the entire physiological system (25,26), hence the urgency to unravel the molecular basis of its inhibitory machinery. True, if this unique TACE-inhibiting gift of TIMP-3 could be fine-tuned or, better still, transplanted onto another TIMP species, the prospect of TIMP development as potential therapeutic agents against TACE-related diseases will no doubt be immeasurably enhanced.

DISCUSSION
The second but no less important reason is our keen interest in making use of these "designer" TIMPs to clarify the many ambiguities surrounding the biological functions of the molecules that may not be related to their inhibitory capacities. TIMP-3, for example, is also known for its ability to induce apoptosis in mammalian cells (27,28). The latest report by Ahonen et al. (29) showed that the process occurred through stabilization of the death receptors TNF-RI, FAS, and TRAIL-RI, but how? Does TIMP-3 inhibit the cell surface-associated proteinases that shed these death receptors? If the answer is yes, is TACE the sole culprit? If not, is TACE only one of the many proteinases that are involved? We hope to address these kinds of intractable issues with our panel of TACE-active designer TIMP-1, -2, and -4 mutants (Table VI). Apart from TIMP-3, similar conundrums also occur with other members of the TIMP family. A good example is the seemingly irreconcilable functions of TIMP-1 both as a growth factor as well as a proteinase inhibitor. TIMP-1 was originally identified as erythroid potentiating activity based on its ability to simulate the growth of erythroid progenitor cells (30), and yet the molecule is also an MMP inhibitor par excellence. An interesting question is: which part(s) of the TIMP-1 molecule is/are responsible for the cell growth activity? Would a TIMP-1 mutant with a modified MMP-binding ridge (such as the TIMP-1 mutant V4S/ TIMP-3 AB-loop/V69L/T98L, K i 0.14 nM with TACE, Table VI) still be a growth factor?
Here we show that N-TIMP-4 is a far better TACE inhibitor than its full-length counterpart. Indeed, the association profile of N-TIMP-4 with TACE fits neatly the trajectory of that of a slow tight-binding inhibitor (Fig. 1). The findings are markedly different from those of TIMP-3; our previous kinetic analysis with the N-terminal and full-length forms of TIMP-3 demon-strated that the C terminus of TIMP-3 has negligible impact on TACE inhibition (K i of N-and full-length TIMP-3 being equal at 0.2 nM) (8). Thus, our results support the notion that the C termini of TIMPs are also crucial in the determination of their MP inhibition profiles.
The findings of this investigation are, to a large extent, in agreement with the conclusions from our earlier TACE/TIMP mutagenesis studies that employed TIMP-1 and -2 scaffolds (17,18). The principles leading to the creation of slow, tightbinding inhibitors may be varied subtly among the TIMPs; however, the essence is identical. In fact, we are now able to explain the molecular basis of the selective sensitivity of TACE to TIMP-3. The rules that underpin TACE/TIMP recognition are herein summarized. First and foremost, in order to forge tight binding with TACE, the amino acid immediately preceding the second disulfide bond on the EF-loop (i.e. residue Thr-98 in TIMP-1, residue Leu 100 in TIMP-2, residue Leu 94 in FIG. 5. A hypothetical TACE⅐N-TIMP-3 binary complex. Where is the site of interaction between the Pro 33 -Phe 34 -Gly 35 triad and TACE? Given that TIMP-3 AB-loop is likely to be highly mobile, it is therefore difficult to pin-point the exact location where the triad interacts with the enzyme. The nonprimed (left) side of TACE is however rather hydrophobic in appearance, and we believe that the triad is most likely to exert its function by forming hydrophobic bonding with the surface of TACE at the nonprimed side of the molecule. The thin line across the middle of the figure divides the TACE enzyme into nonprimed (left) and primed (right) sides. The positions of S1Ј, S3Ј pockets (primed side), and the triad (nonprimed side) are highlighted by shaded boxes for easy recognition. VI "Designer TIMPs" to be used in addressing issues surrounding the cellular functions of TIMPs Throughout this series of TIMP engineering projects, we have designed and created a whole new generation of TIMP mutants that are capable of inhibiting MPs otherwise insensitive to the wild-type inhibitors. We have engineered a TIMP-1 mutant that inhibits MT1-MMP, as well as TIMP-1, TIMP-2, and TIMP-4 mutants that are highly active against TACE. The binding affinities (K i ) of the TIMPs before (wild type) and after (mutants) mutations are listed in 4th and 5th columns of the table. We were interested in making use of these mutants to address the many previously unanswered questions surrounding the biological functions of TIMPs that might not be related to their inhibitory functions; a good example is the ability of the TACE-active TIMP-1, TIMP-2, and TIMP-4 mutants to induce apoptosis in mammalian cells. A summary of the cellular functions of TIMPs can be found in the review by Baker et al. (32). TIMP-3, and residue Leu 101 in TIMP-4) of a TIMP must be none other than a leucine. For reasons not yet clear, leucine is capable of bridging the MMP-binding ridge of TIMP to the surface of TACE and bringing about the establishment of a tight enzyme-inhibitor binary complex. Isoleucine and methionine are tolerated at the locus, but the affinities of the resultant TIMP mutants are always somewhat poorer than those with leucine (17,18). On the other hand, occupation of the position by proline, tryptophan, and aspartate completely abrogates the activities of a TIMP toward TACE. Second, an identical stringency of amino acid requirement also occurred at the CD-loop, the site where TIMPs interact with the S2 pocket at the nonprimed side of the catalytic zinc (Fig. 5). The residue immediately before the first disulfide bond (i.e. the equivalent of valine 69 in TIMP-1, valine 71 in TIMP-2, leucine 67 in TIMP-3, and leucine 72 in TIMP-4) must again be leucine. Third, this rule relates to the preferred choice of amino acid at the P1Ј position (i.e. residue 2 at the very N terminus of the TIMPs). Occupation of the locus by threonine is always favorable to serine, the native residues of TIMP-2 and -4. The last and perhaps the most important rule that enables TIMP/TACE recognition is the absolute requirement of three residues, namely Pro-Phe-Gly, in concert, at the AB-loop. For the successful generation of a potent, TACE-active TIMP variant, the AB-loop of the TIMP candidate must be equipped with this "Pro-Phe-Gly" triad. Combination of this triad with the three aforementioned rules will greatly enhance the affinity of a TIMP with TACE to low nanomolar levels.
There are subtle differences however, between the TIMPs in the way they respond to mutagenesis at the EF-loop. N-TIMP-4, in general, is more pliable to mutagenesis as it tolerates mutations at the Leu 101 site far better than its N-TIMP-2 counterpart. With the exception of proline, tryptophan, and aspartate, the majority of the L101X N-TIMP-4 mutants (L101X ϩ S2T/T38P/E39F/K40G) in this work behaves like slow, tight-binding inhibitors, despite the significant reduction in binding affinities (Table V). The association rates of these L101X mutants with TACE were scarcely affected by the replacement. The opposite is true for N-TIMP-2. In our previous mutagenesis study with Leu 100 of the N-TIMP-2 prototype S2T/TIMP-3 AB-loop/A70S/V71L (K i 1.6 nM; k on 5.71 ϫ 10 Ϫ4 M Ϫ1 s Ϫ1 with TACE), replacement of the residue by the majority of the amino acids (L100X ϩ S2T/TIMP-3 AB-loop/A70S/ V71L) brought about severe reduction not only in affinities but also in association rates (K i of most L100X mutants Ͼ20 nM; k on unable to determine); the only exceptions were isoleucine (L100I) and methionine (L100M) (K i 7-10 nM; k on ϳ4 ϫ 10 Ϫ4 M Ϫ1 s Ϫ1 ) (17). Another aspect of discrepancy that differentiates TIMP-4 from TIMP-1 and -2 is the ideal length of the AB-loop required for TACE inhibition. When the Pro-Phe-Gly triad was grafted onto N-TIMP-1 and N-TIMP-2 in our previous studies, the AB-loops of the TIMPs were shortened to the length of that of TIMP-3 (17,18). The K i values of the best N-TIMP-1 and -2 mutants were in the low nanomolar range, as shown by our panel of N-TIMP-1 and N-TIMP-2 mutants in Table VI (K i of N-TIMP-1 mutant V4S/TIMP-3 AB-loop/V69L/T98L 0.14 nM; K i of N-TIMP-2 mutant S2T/TIMP-3 AB-loop/A70S/V71L 1.5 nM). TIMP-4, in contrast, prefers a longer AB-loop. Truncation of its loop reduced the affinity of the Pro-Phe-Gly triad mutant by as much as 16-fold. The reason for the discrepancy is not known at present.
Throughout this series of TIMP/MP engineering projects, we are able to demonstrate that the specificity of a TIMP could be fine-tuned against a specific MP backdrop. Furthermore, we have shown that the effects of the mutations are quantifiable, because in most cases the affinity of a TIMP toward its MP target could be progressively enhanced upon combination of positive mutations. So far, we have succeeded in fine-tuning TIMP-1 against MT1-MMP (19), created a TIMP-1 mutant that surpassed even TIMP-3 against TACE (17), converted TIMP-2 into an active TACE inhibitor (18), and in this project turned TIMP-4 into a potent TACE inhibitor. The scientific potential of these mutants is immense, and we intend to make use of these designer TIMPs to answer the many TIMP-related issues in our next phase of enterprise. Whether these mutants would be of any therapeutic use, however, has yet to be determined, but the very fact that there are stiff challenges to be overcome adds excitement, puts a premium upon sound judgment, and renders the mutants still more interesting.