Identification of the tissue inhibitor of metalloproteinases-2 (TIMP-2) binding site on the hemopexin carboxyl domain of human gelatinase A by site-directed mutagenesis. The hierarchical role in binding TIMP-2 of the unique cationic clusters of hemopexin modules III and IV.

Cell surface activation of progelatinase A occurs in a quaternary complex with the tissue inhibitor of metalloproteinases-2 (TIMP-2) and two membrane-type matrix metalloproteinases. We have mutated the unique cationic clusters found in hemopexin modules III and IV of the carboxyl domain (C domain) of human gelatinase A to determine their role in binding TIMP-2. Twelve single, double, and triple site-directed mutations were produced that exhibited different TIMP-2 binding properties. Notably, single alanine substitutions at Lys547 and Lys617 reduced TIMP-2 binding by an order of magnitude from that of the recombinant wild-type C domain. Mutations that completely disrupted the C domain.TIMP-2 interaction were K558A/R561A, K610T/K617A, and K566A/K568A/K617A. A triple mutation, K566A/K568A/K575A, having TIMP-2 binding indistinguishable from the wild-type C domain (Kd 3.0 x 10(-8) M), showed that simple reduction of net positive charge does not reduce TIMP-2 affinity. Because the double mutation K566A/K568A also did not alter TIMP-2 binding, these data do not confirm previously reported chimera studies that indicated the importance of the triple lysine cluster at positions 566/567/568 in TIMP-2 binding. Nonetheless, a subtle role in TIMP-2 interaction for the 566/567/568-lysine triad is indicated from the enhanced reduction in TIMP-2 binding that occurs when mutations here were combined with K617A. Thus, these analyses indicate that the TIMP-2 binding surface lies at the junction of hemopexin modules III and IV on the peripheral rim of the gelatinase A C domain. This location implies that considerable molecular movement of the TIMP-2. C domain complex would be needed for the bound TIMP-2 to inhibit in cis the gelatinase A active site.

Connective tissue remodeling is important for growth, healing, and functional adaptation of tissues. In these processes, activation of matrix metalloproteinase (MMP) 1 zymogens is a key control step in the degradation of extracellular matrix proteins (reviewed in Refs. 1 and 2). Gelatinase A (EC 3.4.24.24) (MMP-2) is a pivotal MMP in the remodeling of basement membrane, pericellular, and cell attachment proteins. Cellular activation (3,4) and other cell membrane binding properties (5,6) of gelatinase A are central to the regulation and function of this enzyme (reviewed in Ref. 7). The tissue inhibitors of metalloproteinases (TIMPs) form essentially irreversible 1:1 molar inhibitory complexes with active MMPs (8,9) and so also control MMP activity. Inhibition includes critical interactions between the MMP catalytic Zn 2ϩ ion and Cys 1 of the TIMP inhibitory NH 2 -domain (10,11). However, outside the active site TIMP-2 (12,13), TIMP-3, 2 and TIMP-4 (14) also bind progelatinase A in a 1:1 molar complex on the hemopexin carboxyl-terminal domain (C domain) of the enzyme. That is, active gelatinase A can simultaneously bind two molecules of TIMP-2: one at the active site, to inhibit the enzyme, and another on the C domain. Similarly, TIMP-1 binds the progelatinase B C domain (15), but two molecules of TIMP-1 cannot bind gelatinase A (16).
We initially demonstrated that Con A induced the endogenous cellular activation of progelatinase A (3). Paradoxically, the progelatinase A C domain and TIMP-2 carboxyl domain interaction appears essential for enzyme activation on the cell membrane in this process (4,17,18). Upon binding TIMP-2, a trimolecular complex of progelatinase A first forms with an activator proteinase, membrane type (MT)-MMP (18 -20). In this interaction, the ihibitory NH 2 domain of TIMP-2 binds and inhibits the MT-MMP active site (20 -23). We (24) also have proposed that a quaternary activation complex then forms with a second MT-MMP that cleaves the prodomain of progelatinase A at Asn 37 -Leu 38 . The domain binding interactions of these proteins and their roles in the activation complex are not yet fully resolved. However, deletions and use of isolated domains have shown that TIMP-2 binds progelatinase A via the C domain of both the enzyme (14,16,17,20,25,26) and the inhibitor (26,27). 3 Nonetheless, the localization of the respective contact surfaces on these two protein domains and the important molecular determinants of the binding sites have yet to be identified. In this regard, Willenbrock et al. (26) proposed that the highly charged anionic peptide extension ( 186  ) present at the carboxyl terminus of TIMP-2, but not of TIMP-1, mediates binding to the gelatinase A C domain. Of note, TIMP-4 also contains a similar sequence ( 187 KEFVDI-VQP 195 ) (28) and binds to the gelatinase A C domain (14). Thus, these carboxyl-terminal tails may be crucial in forming the gelatinase A binding site on these TIMPs. In addition, an alternate interaction of the gelatinase A C domain with ␣ v ␤ 3 integrin has been described (5) but not confirmed (22), and a lower affinity interaction also occurs between the NH 2 -domain of TIMP-2, but not that of TIMP-1 or TIMP-3, and the gelatinase A C domain (24,27). 3 Finally, the MMP-independent growth factor effects of TIMP-2 (29) have been ascribed to the C domain of the inhibitor. These may be masked upon binding the gelatinase A C domain. Thus, locating the TIMP-2 binding site on the gelatinase A C domain and generating mutations that disrupt this interaction will be invaluable in dissecting the mechanistic aspects and relative importance of the different cell membrane binding and activation mechanisms proposed for progelatinase A. These mutant proteins will also be useful in assessing the consequences of TIMP-2 binding to the gelatinase A C domain on the growth factor properties ascribed to TIMP-2.
The overall shape of the gelatinase A C domain is a squat cylinder composed of four ␤-sheets, each representing a hemopexin module (I-IV) and each forming a blade of the four bladed ␤-propeller structure (30,31). Each ␤-sheet is formed from four antiparallel ␤-strands. Analysis of these three-dimensional models has revealed cationic clusters in hemopexin module III as a striking feature. These largely result from contiguous stretches of lysine residues, Lys 547 -Asn-Lys-Lys 550 and Lys 566 -Lys-Lys 568 , located near the antiparallel ␤-strand turns at the rim and on the upper surface of the domain. Notably, Lys 550 , Lys 566 , and Lys 567 from these sequences also contribute to three consecutive BX 7 (32). Although the gelatinase A C domain binds heparin (33,34), which enhances progelatinase A activation (22,33), it does not bind hyaluronan. 4 However, these cationic clusters and BX 7 B repeats may be important in binding TIMP-2, possibly interacting through salt bridge formation (24). This hypothesis has been tested in the present report.
In the absence of a three-dimensional structure of the TIMP-2⅐gelatinase A complex, we adopted a mutagenesis approach to identify the TIMP-2 binding site on the gelatinase A C domain. Rather than individually mutate all 36 basic residues in the C domain, we devised a strategy of using multiple replacements to screen more efficiently for important amino acid sites. Further, these substitutions were only made in the third and fourth hemopexin modules, which contain the unique cationic clusters not found in other MMPs-in particular gelatinase B, which binds TIMP-1 but not TIMP-2. We reasoned that multiple substitutions should also lessen the risk of missing weaker binding residues that as single mutations may have their effects masked by the overall binding energy of the TIMP-2⅐C domain complex. Reported here are mutations that establish an important role in TIMP-2 binding for several basic residues in the gelatinase A C domain. Analysis of the effects of these mutations and their location on the three-dimensional structure of the C domain has also enabled us to define the approximate boundary of the TIMP-2 binding site. Thus, TIMP-2 binds the upper surface of the human gelatinase A C domain, on its outer rim at the junction of hemopexin modules III and IV. This location has a number of ramifications concerning the proposed actions of the bound TIMP-2.

EXPERIMENTAL PROCEDURES
Sequence and Three-dimensional Modeling Analyses-To devise a rational mutagenesis strategy, we first utilized the Clustal V method (35) for multiple sequence alignments of MMP C domains to locate unique and conserved cationic residues in the human gelatinase A C domain. The strategy was later refined using structural models of the domain. Until the three-dimensional structure was reported (30,31), we used a molecular model that was made with ProMod to predict the tertiary fold and CHARMM for energy minimization. As a template, the coordinates of the porcine collagenase-1 hemopexin domain (36) were used; they were most kindly provided by Dr P. Brick (Blackett Laboratory, Imperial College, London, United Kingdom) before publication.
Mutagenesis-Using the method of Kunkel et al. (37), 12 site-directed mutations of the human gelatinase A C domain (Gly 417 -Cys 631 ) were made, and the mutant cDNAs were fully sequenced by double strand dideoxy chain termination. As listed in Table I, the following sites were  mutated as indicated: K547A, K549A, K550A,   The mutations generated in the gelatinase A C domain are tabulated according to the sequence or BX 7 B motif thereby disrupted. The first, second, and third BX 7 B repeats are in hemopexin module III. The fourth BX 7 B repeat is in module IV. We rationalized that it was unlikely for TIMP-2 to bind both the upper and lower surfaces of the domain. Because 7 of the 10 basic residues in hemopexin module III are on the upper surface, TIMP-2 was not likely to bind to the lower surface. Therefore, all basic residues on hemopexin module III were mutated except Arg 538 , which is on the lower face of the domain, and Lys 567 in the Lys 566 -Lys-Lys 568 triad. To further limit the large number of mutants that could be made, only one of the two basic residues forming each BX 7 B motif was replaced at a time. In module IV we considered the more distantly placed residues on the lower surface of the domain (Lys 604 and Lys 625 ) to be unlikely sites for binding TIMP-2, and so these were not mutated. Amino acid numbering commences at the NH 2 terminus of the proenzyme. Although an alanine substitution was planned for Lys 610 , the mutagenesis reaction was not efficient and produced nucleotide changes in only a single clone that coded for Thr 610 . The location of the sites mutated on the upper face of the C domain are depicted in a three-dimensional structural model in Fig. 1.
Recombinant Protein Preparation-Recombinant human TIMP-2 was expressed in Chinese hamster ovary cells and purified from the conditioned medium as described previously (38). Recombinant human gelatinase A C domain and mutant proteins were expressed with an NH 2 -terminal His 6 tag (39) in 60-liter fermenter cultures of Escherichia coli strain Le392 using a 100-liter L & H fermenter. During culture, the temperature was controlled at 37°C, airflow was 7.5 liters/min, air saturation was controlled at 5%, and mixer agitation was 400 rpm (39). After 18 or 42 h of culture, recombinant protein was purified from the harvested cells (ϳ310 g) essentially according to Steffensen et al. (40) using procedures that have been characterized as generating correctly folded and biologically functional recombinant C domain (14,34). We have previously reported the essential role of the structural Ca 2ϩ ion in stabilization of the C domain (34). Therefore, to further optimize the refolding environment, 5 mM CaCl 2 was included in the 0.1 M Na 2 B 4 O 7 , pH 10, refolding buffer.
The purity of the wild-type and mutant domain proteins was assessed by SDS-polyacrylamide gel electrophoresis and Western blotting using two affinity purified antipeptide antibodies: ␣His 6 (34) and ␣72Ex12. Anti-72Ex12 antiserum was raised in rabbits against a KLHcoupled peptide from the sequence 561 RYNEVKKKMDPG 572 in the gelatinase C domain that encompasses the lysine triad at positions 566/567/568. To confirm the predicted molecular mass and homogeneity of the recombinant wild-type and mutant proteins, electrospray mass spectrometry was performed on a PESCIEX API 300 as before (14). Protein concentrations were calculated using the molar extinction coefficient (5.10 ϫ 10 4 M Ϫ1 cm Ϫ1 ) of the recombinant C domain determined previously from quantitative amino acid composition analysis (14). The Lys 3 Ala, Lys 3 Thr, and Arg 3 Ala mutant proteins were assumed to have the same extinction coefficients as the wild-type recombinant C domain. The purified recombinant proteins were stored in phosphatebuffered saline (PBS) (140 mM NaCl, 2.7 mM KCl, 4.3 mM Na 2 HPO 4 ⅐7H 2 O, 1.5 mM KH 2 PO 4 , pH 7.4), 0.05% Brij 35 at Ϫ70°C after snap freezing in liquid N 2 .
Binding Assays-Binding of the C domain and mutant proteins to TIMP-2 was measured using a solid phase microwell plate assay (14) with bovine serum albumin and myoglobin as negative control proteins. TIMP-2 was coated onto 96-well plates at 0.2 g per well in 15 mM Na 2 CO 3 , 35 mM NaHCO 3 , 0.02% NaN 3 , pH 9.6, for 18 h at 4°C. Wells were then blocked with 2.5% bovine serum albumin, 0.1% NaN 3 in PBS for 1 h at 21°C. Wild-type and mutant C domain proteins were serially diluted and added to the TIMP-2 coated wells under physiological conditions in 100 l of PBS for 2 h at 21°C. Unbound protein was removed by extensive washing with PBS, 0.05% Tween 20 using an automated plate washer. Binding of C domain or mutant proteins to TIMP-2 was measured using affinity purified anti-peptide ␣His 6 antibody followed by incubation with goat anti-rabbit alkaline phosphatase conjugated secondary antibody (Bio-Rad). After adding p-nitrophenyl phosphate disodium (Sigma) as substrate, absorbance measurements were made at 405 nm in the linear range of the assay using a Thermomax plate reader (Molecular Devices). Binding constants were determined by curve fitting using SigmaPlot on a Macintosh computer.
Affinity Chromatography-The NH 2 -terminal His 6 fusion tag was utilized to bind wild-type and mutant C domains with high affinity to Zn 2ϩ -or Ni 2ϩ -charged chelating-Sepharose 6B resin (Amersham Pharmacia Biotech) in 75-l minicolumns made from pipette tips (41). Following extensive washes in PBS to remove unbound protein, a Ͼ10-fold molar excess of TIMP-2 in PBS was applied to the column, which was then washed with PBS, 0.05% Brij 35. TIMP-2 elution was attempted with 1.0 M NaCl, then 10% Me 2 SO, and finally 50 mM EDTA. Elutes were electrophoresed on 15% SDS-polyacrylamide gels and stained with Coomassie Brilliant Blue R250 to detect TIMP-2 in the column fractions. Relative amounts of the EDTA-eluted C domain (or mutant proteins as appropriate) and TIMP-2 were quantitated by laser densitometry of the protein bands. Affinity chromatography was performed at least three times per protein.

RESULTS
To test the hypothesis that cationic residues in the gelatinase A C domain bind TIMP-2, possibly at the anionic terminal tail, multiple sequence alignment analysis was first used to devise a rational mutagenesis strategy. As shown in Fig. 2, 15 of the 26 basic residues in the C domain are grouped in hemopexin modules III and IV. Comparing the distribution of cationic residues in the C domains of gelatinases A and B, there are more nonconserved basic residues in modules III and IV than in I and II. Therefore, we predicted that the cationic sequences in hemopexin modules III and IV were more likely to be involved in binding TIMP-2. We first focused on the markedly cationic module III, which contains 10 basic residues compared with 5 in module IV, and designed mutations to disrupt the lysine-rich sequences Lys 547 -Asn-Lys-Lys 550 and Lys 566 -Lys-Lys 568 . Three point mutations were made to assess the TIMP-2 binding properties of the individual lysines in the Lys 547 -Asn-Lys-Lys 550 sequence. A double mutation (K566A/ K568A) was also used to test the effect of charge reduction in the Lys 566 -Lys-Lys 568 triad.
Fermenter conditions for each mutant protein were optimized. Expression conditions were mutant protein-specific, but in general, oxygen-limited conditions (5% aeration) proved optimal for the expression of the recombinant gelatinase A C domain and mutants thereof. The mass (Da) measured by electrospray mass spectrometry and the ⌬ mass from that predicted for the wild-type C domain protein and from the mass of the first round of mutant proteins expressed is shown in Table II. These data confirmed the homogeneity of the protein preparations, that NH 2 -terminal methionine processing occurred in the recombinant proteins, and that the correct amino acid substitution had been translated.
Mutagenic Analysis of the TIMP-2 Binding Properties of the Lys 547 , Lys 549 , and Lys 550 Cluster-A differential role in TIMP-2 binding was observed for alanine-substituted proteins at positions 547, 549, and 550 on the upper surface of hemopexin module III (Table III). An alanine mutation at Lys 547 raised the apparent K d of TIMP-2 interaction by approximately 1 order of magnitude (apparent K d 1.7 ϫ 10 Ϫ7 M) and reduced the amount of TIMP-2 bound by ϳ50% at saturation (Fig. 3A). Mutation of Lys 549 or Lys 550 to alanine did not significantly alter the apparent K d values of TIMP-2 interaction (7.8 ϫ 10 Ϫ8 M (Fig. 3A) and 4.8 ϫ 10 Ϫ8 M (Fig. 3B), respectively) from that of the wild-type domain (3.0 ϫ 10 Ϫ8 M). TIMP-2 binding and elution from these mutant proteins on affinity chromatography columns was also no different from the wild-type C domain (not shown). Because a cooperative effect on TIMP-2 binding by residues in this cluster was possible, we then made the triple alanine mutation (K547A/K549A/K550A). However, TIMP-2 binding by the K547A/K549A/K550A protein (apparent K d , 2.6 ϫ 10 Ϫ7 M) was essentially no different from that of the K547A mutant (Fig. 3A). This indicated that cooperative binding by these three residues does not occur and that of these residues, Lys 547 plays the most important binding role.
The enigmatic role of Lys 566 -Lys 567 -Lys 568 in TIMP-2 Binding-A striking feature of the C domain sequence is the unique cationic triad Lys 566 -Lys-Lys 568 on hemopexin module III (see Figs. 1 and 2). This triad also contributes to two BX 7 B motifs: [Lys 558 -Lys 566 ] and [Lys 567 -Lys 575 ] (see Table I). Rather than initially mutate each of the three residues separately, we first screened for the TIMP-2 binding properties of this cluster as a whole by charge reduction using a double alanine mutation. Sites 566 and 568 were selected, as these were nearer than Lys 567 to the other cationic residues in module III (see Fig. 1).
The mutant K566A/K568A protein showed TIMP-2 binding properties (apparent K d 4.7 ϫ 10 Ϫ8 M) (Fig. 3C) that were essentially identical to those of the wild-type C domain. Moreover, an affinity purified antibody raised against a peptide (NH 2 -RYNEVKKKMDPG-COOH) encompassing the lysine triad did not interfere with TIMP-2 binding to the wild-type C domain immobilized on a Zn 2ϩ -chelate affinity column (Fig.  4A). Finally, chromatography of TIMP-2 over a K566A/K568A protein affinity column revealed strong TIMP-2 binding to the mutant domain. Like the wild-type C domain (Fig. 4A), this interaction was not disrupted by 1.0 M NaCl or 10% Me 2 SOthe bound TIMP-2 and K566A/K568A mutant domain complex were eluted in 50 mM EDTA (not shown). We have previously shown that fibronectin and heparin binding by the gelatinase A C domain is Ca 2ϩ ion-dependent, being disrupted by divalent cation chelators (34). In microwell plate assays, the C domain was not dissociated from TIMP-2 by 50 mM EDTA (not shown). This indicates that the EDTA elution of C domain with TIMP-2 from the affinity columns was primarily due to chelation and removal of the ligating Zn 2ϩ ion from the metal chelate resin. Thus, the C domain and TIMP-2 in the 50 mM EDTA eluates represent the bound complex. Because the K566A/K568A double mutation did not alter TIMP-2 binding, these two residues were not individually substituted nor was Lys 567 mutated.
Of the basic residues in the Lys 547 -Asn-Lys-Lys 550 sequence, Lys 550 is topographically closest to the Lys 566 -Lys-Lys 568 triad (see Fig. 1). Therefore, to determine the effects of simultaneous mutations in these two sequences, we made the triple mutation K550A/K566A/K568A. In addition, this selection would also disrupt both the [Lys 550 -Lys 558 ] and [Lys 558 -Lys 566 ] BX 7 B repeats (Table I). In this regard, Lys 550 , rather than Lys 558 , was also the more rational selection because Lys 558 is on the underside of the domain. We considered this surface a less likely TIMP-2-binding region candidate given that 7 of the 10 basic residues in hemopexin module III are on the upper surface. As already discussed, the K550A and K566A/K568A mutations alone showed little alteration in apparent K d values for TIMP-2 compared with the wild-type C domain. However, as a triple mutation (K550A/K566A/K568A), a reduction in binding by approximately 1 order of magnitude resulted (apparent K d , 1.8 ϫ 10 Ϫ7 M) (Fig. 3B). This indicates that Lys 550 , and possibly Lys 566 /Lys 568 , forms contacts with TIMP-2. However, unlike Lys 547 , these appear weak and contribute little to the overall binding energy of the complex.
Mutation of Basic Residues at Positions 558, 561, and 575- The four antiparallel ␤-strands in the ␤-sheet of module III place Lys 558 and Lys 575 on the lower surface of the domain. Lys 575 lies between Lys 558 and Arg 561 , which is located on top of the domain on the outer rim near Lys 550 (see Fig. 1). Neither the single point mutation K575A (apparent K d , 5.2 ϫ 10 Ϫ8 M) nor the triple mutation K566A/K568A/K575A (apparent K d , 7.5 ϫ 10 Ϫ8 M) significantly perturbed the TIMP-2 interaction. Indeed, these two mutant proteins showed very similar binding properties to that of the wild-type C domain both in the plate assay (Fig. 3D) and by affinity chromatography (not shown). However, the mutation K558A/R561A showed almost total loss of TIMP-2 binding (Fig. 3D). Therefore, it is likely that in the K558A/R561A double mutant protein, Arg 561 has the major TIMP-2 binding role of this pair.
The Importance of Lys 617 in Module IV for TIMP-2 Binding-All the mutations described in the preceding sections were in module III. To investigate the contribution to TIMP-2 binding by basic residues in module IV, three mutant proteins were made. Lys 617 is on the outer rim of this module near the upper surface of the domain (see Fig. 1). The mutant protein K617A bound to TIMP-2 in a saturatable manner but with reduced affinity, as determined from the reduction in apparent K d by nearly 1 order of magnitude (1.1 ϫ 10 Ϫ7 M) (Fig. 3C). This result indicates an important role in the TIMP-2 interaction for Lys 617 and that TIMP-2 also contacts hemopexin module IV. As already discussed, the mutation K566A/K568A had little effect on TIMP-2 binding. However, when this double mutation was combined with K617A, a synergistic decrease in TIMP-2 affinity resulted that reduced binding to undetectable levels (Fig.  3C). Consistent with this, TIMP-2 did not bind to K566A/ K568A/K617A mutant protein affinity columns, with all the applied TIMP-2 being collected in the unbound fractions (Fig.  4B). In contrast, TIMP-2 bound to the K617A mutant protein and could be eluted with this protein in 50 mM EDTA (Fig. 4C). However, the relative amount of TIMP-2 bound to the mutant protein and recovered in the eluates was quantitated at ϳ0.5fold that of the amount of TIMP-2 bound to the wild-type C domain. The plate assays were consistent with this finding. The curve fitting analysis confirmed a near 50% reduction in the total amount of K617A protein bound to TIMP-2 at saturation compared with the wild-type C domain.
The importance of position 617 and module IV was further shown by the K610T/K617A mutation. This reduced TIMP-2 binding to undetectable levels both in the plate assay (Fig. 3B) and by affinity chromatography (Fig. 4B). Thus, Lys 610 would also appear to play an important role in TIMP-2 binding, but a confirmatory point mutation of this site is needed for an unequivocal demonstration. DISCUSSION Salt bridge formation between the unique cationic clusters in the hemopexin-like C domain of progelatinase A and the anionic carboxyl-terminal peptide tail of TIMP-2 is predicted to be important for complex formation between the gelatinase A C domain and TIMP-2. This hypothesis was tested by generating 12 single, double, and triple site-directed mutations of 10 basic residues in the human gelatinase A C domain in order to locate the TIMP-2 binding site. Our studies have shown that disruption of TIMP-2 binding to the C domain follows the introduction of an energetically unfavorable change in side chain polarity at several sites, in particular at Lys 547 in hemopexin module III and at Lys 617 in module IV. Two double mutants also strongly implicate Arg 561 and Lys 610 as playing an important role in TIMP-2 binding. Finally, analysis of several triple mutations indicates that a weak TIMP-2 interaction occurs with Lys 550 and reveals a subtle role in TIMP-2 binding for Lys 566 and Lys 568 . Our data do not reveal a canonical linear sequence that

domain proteins by electrospray mass spectrometry
The molecular masses of the wild-type C domain and the four lysine mutants generated in the first round of mutagenesis were measured by electrospray mass spectrometry. The typical error for these mass determinations is 1 in 10 4 Da. Because of the close agreement between the measured masses with those predicted, this analysis was not carried out on proteins generated in mutagenesis rounds two and three. The masses were measured after sample injection onto a PESCIEX API 300 mass spectrometer from a C18 high pressure liquid chromatography column. b The measured mass change from the predicted mass of the protein after N-terminal methionine processing. From the amino acid sequence, the predicted mass of the N-terminal methionine processed recombinant C domain is 25,925 Da. The predicted decrease in mass by a lysine to alanine substitution is 57 Da. forms the TIMP-2 binding site, nor do they show a critical role in TIMP-2 binding for the four BX 7 B motifs present in hemopexin modules III and IV of the domain. Only two of seven residues covered by these motifs are implicated in TIMP-2 binding, and they are not one of the BX 7 B pairs. Rather, these repeats, in combination with the periodicity of the turn of the antiparallel ␤-strands, serve to position lysines near to each other in the tertiary structure of the domain. This concentration of charged residues forms cationic clusters on the upper surface of the domain-some residues in each, but not all, we show to be involved in the TIMP-2 interaction. Thus, in the absence of the three-dimensional structure of TIMP-2 bound to the gelatinase A C domain, this mutagenesis approach provides convincing but nonetheless indirect evidence that these lysine and arginine residues form important elements of the interaction surface shown in Fig. 5.
Our data are consistent with those of Friedman et al. (25), who found that a C domain deletion mutant that terminates at Gly 454 , 15 residues down from Cys 440 at the start of the domain (see Fig. 2), lost TIMP-2 binding. A chimera of module I that replaced up to Pro 474 with the corresponding region from stromelysin-1 was recently found by Butler et al. (22) to retain TIMP-2 binding. Another chimera that replaced all of module IV and part of module III from Lys 568 to Cys 631 lost TIMP-2 binding (22). From these data, the authors concluded that the TIMP-2 binding site was located between Lys 568 and the carboxyl-terminal Cys 631 . However, a common problem of chimera studies is structural perturbations at the chimera interfaces, which may also have accounted for the loss of TIMP-2 binding. Indeed, we have found TIMP-2 binding residues outside the region implicated by the chimera studies. Thus, our data considerably refine these previous analyses by identifying the role played by individual basic residues in both modules III and IV and reveal an important contribution to TIMP-2 binding by residues amino-terminal to the Lys 566 -Lys-Lys 568 triad.
Unlike domain chimeras, in which entire ␤-sheet packing faces are replaced, alanine is considered an innocuous substitution that generally does not produce steric clashes (42). Indeed, one triple alanine mutation (K566A/K568A/K575A) showed TIMP-2 binding properties that were essentially identical to those of the wild-type C domain. Thus, the introduction of three mutations does not by itself globally destabilize the domain. Although structural causes cannot be definitively excluded as an explanation for the reduced TIMP-2 affinity observed at some mutation sites, the mutant alanines would not support an ionic interaction potential with TIMP-2 partner residues. If, as our data suggest, certain basic residues of the C domain are involved in salt bridge formation with anionic TIMP-2 residues, the removal of the cationic sidechain would leave an unpaired anion on the TIMP-2 surface during binding. This would impose a penalty in free energy for burial into the contact face of an acidic unpaired charge or an associated water molecule. Clearly, the nature of the salt bridges characterizing the TIMP-2 binding interaction is complex with cooperative and differential roles played by the individual basic residues. Thus, simply reducing the net positive charge of a cluster was not sufficient to weaken the TIMP-2 interaction. Rather, the role of individual basic residues in TIMP-2 recognition is position-sensitive. This is likely related to structural considerations and distance constraints imparted by the anionic carboxyl-terminal TIMP-2 tail. Potentially, some of the less important TIMP-2-binding basic residues may play more important roles in binding TIMP-4, which competes with TIMP-2 for gelatinase A C domain binding and possibly activation (14). This hypothesis is now under investigation in our laboratory.
Almost certainly, other surrounding nonbasic residues are required for TIMP-2 binding that will also contribute to the free energy of association. Indeed 1.0 M NaCl does not dissociate TIMP-2 from C domain-affinity columns, nor does it prevent TIMP-2 interaction from occurring in the plate assays (not shown). Thus, due to the large surface area of the TIMP-2 binding site, a pronounced effect on TIMP-2 binding by single mutations is predicted to be relatively small. Nonetheless, the observations that single point mutations at positions 547 and 617 produced such impressive shifts in the apparent K d values of the TIMP-2 interaction and that the two double alanine mutations of Lys 610 /Lys 617 and Lys 558 /Arg 561 alone rendered any TIMP-2 binding undetectable, together reveal the importance of salt bridges involving some of these residues in the TIMP-2 interaction. Additional mutagenesis is needed to further characterize the binding interaction and to unequivocally assign binding roles for all contact residues. However, the combinations of mutants utilized in this investigation have proven effective in identifying an important role for salt bridge formation in TIMP-2 docking to the gelatinase A hemopexin C domain and for refining the location of the TIMP-2 binding site.
A position-sensitive interaction with TIMP-2 was demonstrated for the cationic residues Lys 547 , Lys 549 , Lys 550 , Lys 558 , Arg 561 , and Lys 575 on module III. Residues in this cluster also form a potential heparin-binding site (34) and form the three BX 7 B motifs in module III. From characterization of the single alanine mutants at positions Lys 547 , Lys 549 , and Lys 550 , the lack of effect on TIMP-2 binding by K549A indicates that this residue lies at the border or just outside the TIMP-2 binding surface (see Fig. 5). The more marked alteration in TIMP-2 binding by K547A alone and by K550A in the triple mutant protein K550A/K566A/K568A indicates that Lys 547 forms the FIG. 4. Elution profiles of TIMP-2 chromatographed over wildtype and mutant C domain protein affinity columns. C domain (A) (260 g) and the mutant proteins indicated (B and C) (100 g) were bound to Zn 2ϩ -or Ni 2ϩ -chelate minicolumns as described under "Experimental Procedures." U, unbound fraction during loading. After extensive washes with chromatography buffer (W), affinity-purified antipeptide antibody ␣72ex12 was loaded and incubated on the column (A). The antibody bands in the 40 -60-kDa range and at the top of the gel (labeled) could be readily seen on elution with the bound complexes in 50 mM EDTA. TIMP-2 was applied to the affinity columns after washes (W). The unbound (TIMP-2) and PBS wash fractions were collected and analyzed on 15% SDS-polyacrylamide gel electrophoresis gels (W in A; PBS wash fractions 1-4 in B). Sequential elution with 1.0 M NaCl, 10% Me 2 SO (DMSO), and 50 mM EDTA followed as shown. TIMP-2 did not bind to the zinc chelate resin (not shown). M r , molecular mass marker proteins in Da as indicated; Std, recombinant C domain (rC) and TIMP-2 (T2) before chromatography. For some recombinant proteins, a lower molecular mass form without the His 6 NH 2 -terminal tag is an occasional minor component of the preparation after freeze-thawing (see rC Std in A). dominant TIMP-2 recognition residue in this cluster, with Lys 550 playing a minor role. Also in this cluster is Arg 561 , which when mutated to alanine together with Lys 558 , displays strongly reduced TIMP-2 binding. In contrast, the nearby Lys 575 does not appear to play a role in the TIMP-2 interaction. Three-dimensional structural analysis provides a possible explanation for these effects and suggests an important role in TIMP-2 binding for Arg 561 rather than Lys 558 . Lys 558 lies on the underside of the domain, and the ␤-strand commencing here makes a turn near Lys 566 -Lys-Lys 568 on the upper domain surface, where Lys 566 forms a BX 7 B motif with Lys 558 [Lys 558 -Lys 566 ]. The antiparallel ␤-strand then returns to Lys 575 on the underside of the domain and at the end of another consecutive BX 7 B repeat [Lys 567 -Lys 575 ]. The tertiary fold packs Lys 575 near Lys 558 on the lower domain surface. Although Arg 561 is 14 residues distant from Lys 575 in the primary structure, in the tertiary structure, these residues lie close to each other. As the lower panel in Fig. 5 shows, Arg 561 is on the outer rim of the upper surface of the domain. Lys 575 , which is not implicated in TIMP-2 binding, is interposed between it and Lys 558 on the lower surface. Thus, from analysis of the binding data for the K558A/R561A mutant protein in the context of this structural information, we consider that the primary effect on TIMP-2 binding by this double mutation is through substitution of the Arg 561 . Although not ruled out by these mutations, a contribution to TIMP-2 binding by Lys 558 is a possibility that is effectively discounted by these structural considerations.
The function of the Lys 566 -Lys-Lys 568 triad in TIMP-2 binding is enigmatic. These sites lie on the upper surface of the domain close to a turn connecting two antiparallel ␤-strands. The influence of this cationic triad on TIMP-2 binding was first studied by a double mutation (K566A/K568A), designed to reduce the overall charge of this cluster. We then added mutations of BX 7 B motifs onto the K566A/K568A background. Replacement of two of the lysine residues in Lys 566 -Lys-Lys 568 with alanine did not disrupt TIMP-2 binding. Although substitution of Lys 567 was not performed, an antipeptide antibody to this sequence also did not block the TIMP-2 interaction. Thus, our studies do not support the recent report that Lys 566 -Lys-Lys 568 is important in the TIMP-2 interaction (22). However, when the K566A/K568A mutations were combined with other substitutions (K550A and K617A), a marked reduction in TIMP-2 interaction resulted. This reveals a small contribution to TIMP-2 binding by Lys 566 -Lys-Lys 568 that is not strong enough to be manifest without other changes to the contact surface. From its topographical location in relation to Lys 550 , which also appears to only make a weak interaction with TIMP-2, this outlines a small zone of weak TIMP-2 contact that may account for the minor direct effects on TIMP-2 binding by mutations of these residues alone. As Fig. 5 shows, this weak binding zone (shown by the hatched pattern) is at one end of the main TIMP-2 binding surface that is defined by the more effective mutation sites (shown within the dotted perimeter).
Although heparin has been shown to bind the C domain (33,34) and to potentiate progelatinase A activation on the cell surface (22,43), the heparin binding residues and their relationship to TIMP-2 binding have yet to be determined. Understanding the nature of the simultaneous heparin and TIMP-2 binding interactions is likely to be complex. Further use of the mutants generated here will be invaluable for our ongoing studies to understand these cooperative interactions. Heparin interaction may involve the potential heparin binding sites Lys 547 -Asn-Lys-Lys 550 and 566 Lys-Lys-Lys 568 (34). Notably, our data reveal that two residues in the first motif (Lys 547 and Lys 550 ) appear to play a role in the TIMP-2 interaction and so may no longer interact with heparin after TIMP-2 has bound to the domain. This possibility supports the potential involvement in heparin binding of Lys 566 -Lys-Lys 568 to increase the rate of gelatinase A activation. The Lys 547 -Asn-Lys-Lys 550 sequence also contributes to the first BX 7 B motif, and Lys 566 -Lys-Lys 568 contributes to both the second and third BX 7 B motifs. In the cell adhesion molecule RHAMM (receptor for HA-mediated motility), the BX 7 B motif has been shown to bind hyaluronan (32), but recent evidence has cast doubt on this function (44). Indeed, we have found no evidence for hyaluronan binding by the C domain, 3 but this may be due to the presence of an acidic residue in each of these repeats that reduces hyaluronan binding. Because the BX 7 B motifs also do not appear to play a critical role in TIMP-2 binding, the significance of these motifs in the gelatinase A C domain remains unclear.
The marked clustering of positive charges to one face of the gelatinase A C domain introduces a considerable dipole moment to the molecule. Thus, a related role for these cationic residues may be to correctly orientate the TIMP-2 contact face of the C domain with the anionic surface of TIMP-2 immediately prior to protein contact. In the active enzyme, these ionic interactions may thereby favor TIMP-2 binding first to the C domain rather than to the active site potentially explaining how TIMP-2 can bind the C domain despite the stronger binding interaction with the active site (45). Indeed, the recently published three-dimensional structure of full-length TIMP-2 bound to the active site of MT1-MMP (46) shows that the TIMP-2 carboxyl-terminal tail exhibits considerable disorder. Such molecular flexibility of the anionic tail may allow initial contact and molecular alignment with the gelatinase A C domain as a first step prior to full protein contact and stabilization of the bound complex. In this regard, we observed an approximate 0.5-fold reduction in the relative amount of TIMP-2 bound with the K617A and K547A mutant C domains in addition to a reduction in apparent K d values. In a two-part binding reaction, this may indicate that one of the two bound states of TIMP-2 on the C domain is perturbed by these mutations. Indeed, biphasic kinetics for binding of TIMP-2 to progelatinase A have been reported (46).
Locating on the three-dimensional structure of the gelatinase A C domain the mutation sites that disrupted TIMP-2 binding, compared with those that did not, identifies, with higher precision than previous deletion and chimera studies, the general position of the TIMP-2 contact interface. Thus, TIMP-2 binds the upper surface of the gelatinase A C domain, at the junction of hemopexin modules III and IV, to form an elongated binding surface on the outside rim of the domain (Fig. 5). This region also includes a zone of weaker TIMP-2 interaction between Lys 550 and Lys 566 /Lys 568 (Fig. 5, hatched  pattern). Notably, this is unlike many ␤-propeller structures, which bind ligands at the top of the ␤-propeller near the pseudosymmetry axis (47,48). The asymmetrical localization of the TIMP-2 binding site away from the central axis of the C domain has several implications for the function of gelatinase A and the bound TIMP-2, particularly in terms of cellular binding and activation by MT-MMPs. By analogy with the full-length porcine collagenase-1 structure (36), our study indicates that the bound TIMP-2 would lie away from the active site, on the opposite side of the molecule from the fibronectin type II-like repeats that form the collagen binding domain of gelatinase A (40). This orientation would likely preclude the inhibition in cis of the gelatinase A active site by the C domain-bound TIMP-2. At the same time, it would permit interaction of the NH 2inhibitory domain of the bound TIMP-2 with the active site of MT-MMPs that can act as cell receptors for gelatinase A (20 -23). Free access by a second MT-MMP that is proposed to initially cleave the gelatinase A prodomain (24,43) at Asn 37 -Leu 38 would also not be impeded by this spatial arrangement. Inhibition of gelatinase A activity by TIMP-2 would therefore require a second TIMP molecule that can be simultaneously bound by the active site of gelatinase A (12,13), rather than through inhibition by the C domain-bound TIMP-2 as some authors suggest. This complex ternary interaction also requires stabilization of the active site-bound TIMP-2 with the C domain of gelatinase A (46) at a second site. Indeed, we have recently found that the NH 2 -domain of TIMP-2 binds with low affinity to the gelatinase A C domain (24,27). 3 In this regard, a gelatinase A C domain chimera generated by Butler et al. (22) that replaced most of hemopexin module I up to Pro 474 with the corresponding region of stromelysin-1 still retained TIMP-2 binding properties. In consideration of our studies, the stabilization site for the gelatinase A inhibitory TIMP-2 molecule might therefore lie on the rim of hemopexin module II of the gelatinase A C domain. Lastly, these studies are consistent with the notion that the carboxyl-terminal tail of TIMP-2 binds to the gelatinase A C domain. Whether TIMP-2 binding to the gelatinase A C domain masks the growth factor effects of the TIMP-2 carboxyl domain is an intriguing prospect that the mutant domains generated here should prove useful in helping to address.