Differential Inhibition of Membrane Type 3 (MT3)-Matrix Metalloproteinase (MMP) and MT1-MMP by Tissue Inhibitor of Metalloproteinase (TIMP)-2 and TIMP-3 Regulates Pro-MMP-2 Activation*

The membrane type (MT)-matrix metalloproteinases (MMPs) constitute a subgroup of membrane-anchored MMPs that are major mediators of pericellular proteolysis and physiological activators of pro-MMP-2. The MT-MMPs also exhibit differential inhibition by members of the tissue inhibitor of metalloproteinase (TIMP) family. Here we investigated the processing, catalytic activity, and TIMP inhibition of MT3-MMP (MMP-16). Inhibitor profile and mutant enzyme studies indicated that MT3-MMP is regulated on the cell surface by autocatalytic processing and ectodomain shedding. Inhibition kinetic studies showed that TIMP-3 is a high affinity inhibitor of MT3-MMP when compared with MT1-MMP (Ki = 0.008 nm for MT3-MMP versus Ki = 0.16 nm for MT1-MMP). In contrast, TIMP-2 is a better inhibitor of MT1-MMP. MT3-MMP requires TIMP-2 to accomplish full pro-MMP-2 activation and this process is enhanced in marimastatpretreated cells, consistent with regulation of active enzyme turnover by synthetic MMP inhibitors. TIMP-3 also enhances the activation of pro-MMP-2 by MT3-MMP but not by MT1-MMP. TIMP-4, in contrast, cannot support pro-MMP-2 activation with either enzyme. Affinity chromatography experiments demonstrated that pro-MMP-2 can assemble trimolecular complexes with a catalytic domain of MT3-MMP and TIMP-2 or TIMP-3 suggesting that pro-MMP-2 activation by MT3-MMP involves ternary complex formation on the cell surface. These results demonstrate that TIMP-3 is a major regulator of MT3-MMP activity and further underscores the unique interactions of TIMPs with MT-MMPs in the control of pericellular proteolysis.

The matrix metalloproteinases (MMPs), 1 a multidomain family of zinc-dependent endopeptidases, degrade all struc-tural components of the extracellular matrix (ECM) and many bioactive molecules, thereby playing essential roles in many physiological and pathological processes (1)(2)(3)(4). Based on structural organization and subcellular localization, the MMP family is divided into secreted and membrane-anchored enzymes (1,5). The membrane type-MMPs (MT-MMPs) comprise six members of plasma membrane-tethered MMPs, which include four type I transmembrane enzymes: MT1-, MT2-, MT3-, and MT5-MMP, and two glycosylphosphatidylinositol-anchored enzymes: MT4-and MT6-MMP (4, 6 -8). Although there is some tissue-specific expression of MT-MMPs, there is also significant overlap both in expression and in function raising the question of how cells command the repertoire of MT-MMPs at their disposition during proteolytic events. A major mechanism for controlling MT-MMP activity in the pericellular space is mediated by the action of tissue inhibitors of metalloproteinases (TIMPs). However, as opposed to soluble MMPs, which are almost equally inhibited by all TIMPs (9,10), the MT-MMPs are highly selective when it comes to TIMP interactions. For example, the type I transmembrane MT-MMPs are poorly inhibited by TIMP-1 but are relatively well inhibited by TIMP-2, TIMP-3, and TIMP-4. In contrast, the glycosylphosphatidylinositol-anchored MT-MMPs are inhibited by both TIMP-1 and TIMP-2 (11). The differential sensitivity to TIMP inhibition among members of the MT-MMP family may facilitate the control of MT-MMP activity in the pericellular space.
In addition to being potent ECM-degrading enzymes, the type I transmembrane MT-MMPs can also initiate a cascade of zymogen activation on the cell surface. MT-MMPs activate the zymogenic form of MMP-2 (pro-MMP-2 or pro-gelatinase A) (6,8), which in turn can activate pro-MMP-9 (pro-gelatinase B) (12). MT1-MMP also promotes the activation of pro-collagenase 3 (MMP-13) (13), a potent collagenolytic protease. Because the gelatinases are efficient gelatinolytic enzymes, the MT-MMP/ MMP-13/gelatinase axis represents a well adapted proteolytic system designed to promote coordinated and complete collagen degradation in the pericellular space. Although the type I transmembrane MT-MMPs have been shown to activate pro-MMP-2 (6), the specific mechanism(s) by which each member of this subgroup accomplishes pro-MMP-2 activation and the accessory molecules involved in this process are not completely understood. In the case of MT1-MMP, pro-MMP-2 activation requires the participation of TIMP-2, which acts as a molecular link between an active MT1-MMP on the cell surface and pro-MMP-2 (14,15). The N-terminal region of TIMP-2 binds to the active site of MT1-MMP, whereas the C-terminal region of the inhibitor binds to the hemopexin-like domain of pro-MMP-2 forming a so-called "ternary complex" (16,17). The propeptide of the bound pro-MMP-2 is then cleaved at the Asn 37 -Leu 38 peptide bond by a neighboring TIMP-2-free active MT1-MMP molecule. This is followed by a second cleavage event in which the intermediate MMP-2 form is cleaved at the Asn 80 -Tyr 81 peptide bond by a fully active MMP-2 in an autocatalytic intermolecular manner resulting in full activation (13). TIMP-2dependent pro-MMP-2 activation occurs only at low TIMP-2 concentrations relative to MT1-MMP (9,15,18). This permits availability of enough TIMP-2-free active MT1-MMP to hydrolyze the prodomain of pro-MMP-2 bound in the ternary complex. Thus, under limited conditions, TIMP-2 promotes zymogen activation, whereas high levels of TIMP-2 relative to MT1-MMP inhibit activation by blocking all free MT1-MMP molecules (14, 18 -20). Besides its role in regulation of pro-MMP-2 activation, TIMP-2 also controls the turnover of active MT1-MMP on the cell surface by inhibiting its autocatalytic processing, which in turn can positively influence MT1-MMP activity (18). Paradoxically, reversible synthetic MMP inhibitors, which also inhibit the autocatalytic turnover of MT1-MMP, can enhance pro-MMP-2 activation by MT1-MMP in the presence of TIMP-2 (21).
MT3-MMP (MMP-16), which was originally cloned from human melanoma tissue and human placenta (30), is expressed in a variety of normal (30 -33) and tumor (7,(33)(34)(35) tissues. Studies with a purified active MT3-MMP lacking the transmembrane domain have shown that the truncated enzyme cleaves a variety of ECM components and is inhibited by TIMP-2 and TIMP-3 but not by TIMP-1 (29). However, the kinetics of MT3-MMP inhibition by TIMPs was not determined. Functionally, expression of MT3-MMP in human WM1341D melanoma cells (36), but not in monkey kidney COS-1 cells (37), facilitated in vitro collagen I invasion. Also, expression of MT3-MMP in hamster CHO-K1 and canine Madin-Darby canine kidney cells induced the expression of a fibrin invasive phenotype (38). Expression of MT3-MMP in various cell lines (30, 37, 39 -41) promoted pro-MMP-2 activation and this activity was inhibited by addition of exogenous TIMP-2 (30, 42) or synthetic MMP inhibitors (40,42). However, the role of TIMPs in MT3-MMP-mediated pro-MMP-2 activation has not been determined. To gain insight into the regulation of MT3-MMP, we expressed it in mammalian cells and investigated its processing, inhibition by TIMPs, and its ability to activate pro-MMP-2 in the presence of TIMPs.
Recombinant Proteins, Protease Inhibitors, and Antibodies-Human recombinant pro-MMP-2, TIMP-2, and TIMP-1 were expressed in HeLa S3 cells and purified to homogeneity, as previously described (44). Human TIMP-4 and TIMP-3 were purchased from R&D Systems Inc. (Minneapolis, MN) and their concentrations were determined by activesite titration using active MMP-2 with known concentration, as previously described (45). The catalytic domains of MT1-MMP (MT1-MMPcat ) and MT3-MMP (MT3-MMP cat ) were purchased from Calbiochem (San Diego, CA). N-TIMP-2, an N-terminal inhibitory domain of human TIMP-2 ending at Cys 128 was expressed in mammalian cells and purified as previously described (46). N-TIMP-3 comprising the N-terminal region of mature human TIMP-3, residues Cys 1 to Asn 121 , with a His tag attached to the C terminus was expressed in Escherichia coli and purified as described (47). The hydroxamate-based MMP inhibitors batimastat (BB94) and marimastat (BB2516) were obtained from British Biotech (Oxford, United Kingdom). A rabbit polyclonal antibody against human MT3-MMP was purchased from Calbiochem. The monoclonal antibody to TIMP-2 (CA-101) has been previously described (48).
Recombinant Vaccinia Viruses-The generation of the recombinant vaccinia virus expressing bacteriophage T7 RNA polymerase (vTF7-3) has been described by Fuerst et al. (49). Recombinant vaccinia viruses expressing MT1-MMP (vT7-MT1), pro-MMP-2 (vT7-GelA), or TIMP-2 (vSC59-T2) were generated by homologous recombination as previously described (18,48,49). To generate a recombinant vaccinia virus expressing the wild-type human MT3-MMP enzyme, the full-length human MT3-MMP cDNA was amplified by PCR using specific primers containing the appropriate restriction sites, and the resulting fragment was cloned into the NcoI and BamHI sites of the pTF7EMCV-1 expression vector, under the control of the T7 promoter (49). After sequence verification of the insert in both directions, the resulting pTF7-MT3 plasmid was used to generate a recombinant vaccinia virus (vT7-MT3) by homologous recombination with wild-type vaccinia virus, as previously described (48,49).
Generation of MT3-MMP Mutants-To generate a catalytically inactive MT3-MMP (E/A-MT3-MMP), Glu 247 was substituted for Ala using the QuikChange TM site-directed mutagenesis kit (Stratagene, La Jolla, CA) and the wild-type MT3-MMP cDNA in pTF7EMCV-1 as a template. A cytosolic tail (CT) deletion mutant (⌬CT-MT3-MMP) was constructed by introducing a termination codon at Gln 586 . Likewise, a transmembrane domain and cytosolic tail deleted MT3-MMP (⌬TM/CT-MT3-MMP) was constructed by introducing a termination codon at Ala 564 . The fidelity of the inserts was verified by DNA sequencing in both directions. Recombinant viruses expressing these MT3-MMP mutants were generated by homologous recombination with wild-type vaccinia virus, as previously described (48,49).
Expression of MT3-MMP and TIMP-2 by Co-infection-To express MT3-MMP, cells (various types) in six-well plates were co-infected with 10 plaque-forming units (pfu)/cell each of vTF7-3 and vT7-MT3 viruses (encoding for either wild-type or mutant enzymes) for 45 min in infection media (DMEM supplemented with 2.5% fetal bovine serum and antibiotics) at 37°C. As a control, the cells were infected only with the vTF7-3 virus. To co-express MT3-MMP with TIMP-2, and to modulate the level of inhibitor expression, BS-C-1 cells were co-infected with 5 pfu/cell each of vTF7-3 and vT7-MT3 viruses and increasing amounts (0 -20 pfu/cell) of the TIMP-2-expressing vaccinia virus vSC59-T2 as described (18). The infected cells were then used for pro-MMP-2 activation and MT3-MMP processing studies by gelatin zymography and immunoblot analysis, respectively, as described (50). In some experiments, the media of cells co-expressing MT3-MMP and increasing amounts of TIMP-2 were collected and analyzed for TIMP-2 concentration by enzyme-linked immunosorbent assay (catalog number QIA40, Oncogene Research Products, San Diego, CA) as described by the manufacturer.
Pro-MMP-2 Activation by Membrane-bound and Soluble MT3-MMP-Cells (various types) in six-well plates were co-infected to express MT3-MMP as described above. After these procedures, the media were aspirated and cells were rinsed once with serum-free DMEM, followed by a 4-h incubation with serum-free DMEM. The media were then aspirated and replaced with serum-free DMEM (1 ml/well) supplemented with pro-MMP-2 without or with various amounts of TIMP-2, N-TIMP-2, TIMP-3, TIMP-4, or marimastat. At varying times, the media were collected for gelatin zymography and the cells were lysed on ice with cold lysis buffer (25 mM Tris-HCl, pH 7.5, 1% IGEPAL CA 630 (Sigma), also known as Nonidet P-40, 100 mM NaCl, 10 g/ml aprotinin, 1 g/ml leupeptin, 2 mM benzamidine, and 1 mM phenylmethylsulfonyl fluoride). Media and lysates were analyzed by gelatin zymography for pro-MMP-2 activation and by immunoblot analysis for MT3-MMP forms, as previously described (50). To assess the ability of soluble MT3-MMP to process pro-MMP-2, BS-C-1 and TIMP-2Ϫ/Ϫ cells were grown in a 100-mm tissue culture dish each and co-infected (vTF7-3 and vT7-MT3 viruses) to express MT3-MMP or infected with vTF7-3 virus alone, as described above. After infection, the cells were washed and incubated overnight at 37°C in serum-free media (6 ml/dish). Then, the media were collected, clarified by centrifugation, and concentrated ϳ30 times with Centricon Plus-20. The concentrated media were then subjected to ultracentrifugation at 100,000 ϫ g for 1 h at 4°C. The supernatants were collected and 20 l of this fraction were incubated (24 h, 37°C) with various concentrations of pro-MMP-2 (25, 100, and 500 nM) in collagenase buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 5 mM CaCl 2 , 0.02% Brij-35) in a final reaction volume of 50 l. Pro-MMP-2 processing was monitored by gelatin zymography.
Enzyme Inhibition Studies-The enzymatic activity of MT1-MMP and MT3-MMP catalytic domains (MT1-MMP cat and MT3-MMP cat ) was monitored using the synthetic fluorogenic peptide (7-methoxycoumarin-4-yl)acetyl-L-prolyl-L-glycyl-leucyl-(N 3 -(2,4-dinitophenol)-L-2,3-diaminopropionyl)-L-alanyl-L-arginine amide (MOCAcPLGLA2pr(Dnp)AR-NH 2 from Peptides International, Louisville, KY), as described (51), on a Photon Technology International spectrofluorometer, at excitation and emission wavelengths of 328 and 393 nm, respectively. The concentration of MT1-MMP cat and MT3-MMP cat was determined using a solution of recombinant TIMP-2 with known concentration. All of the kinetic assays were carried out in a buffer consisting of 50 mM HEPES (pH 7.5), 150 mM NaCl, 5 mM CaCl 2 , 0.01% Brij-35, and 1% Me 2 SO (1% v/v) (buffer R), at 25°C, in a thermostatted cuvette holder. Less than 10% hydrolysis of the fluorogenic substrate was monitored (51). The kinetic parameters k cat and K m for the reaction of MT3-MMP cat with the fluorogenic substrate were determined from fitting the substrate concentration dependence of the initial rates of substrate hydrolysis to the Michaelis-Menten equation, using the program Scientist (MicroMath, Salt Lake City, UT). MT1-MMP cat and MT3-MMP cat inhibition by TIMPs was carried out essentially as described (52). The concentrations of functional full-length TIMPs and N-TIMP-3 were determined by active site titration using active MMP-2 with known concentration, as previously described (45). MT1-MMP cat or MT3-MMP cat (0.5 nM) was added to a mixture of synthetic substrate (7 M) and various inhibitor concentrations (up to 16 nM), in buffer R, in acrylic cuvettes with stirring, and hydrolysis of the substrate was followed for 30 min. Analysis of these progress curves yielded the association and dissociation rate constants (k on and k off ) and the inhibition constant values (K i ϭ k off /k on ), as described (52). For TIMP-3, progress curves for enzyme inhibition were done with N-TIMP-3 because of adsorption of the wildtype inhibitor to the acrylic cuvettes used, which precluded kinetic analysis of the data. Of note, TIMP-3 titration with MMP-2, an end point experiment, was still feasible, as described above. The k off values were independently determined from the enzyme activity recovered after dilution of a solution of pre-formed enzyme-inhibitor complex, obtained by incubating enzyme (25 and 50 nM) with inhibitor (50 and 100 nM, respectively) for 1 h, at room temperature. The complex was diluted (400-fold) into 2 ml of buffer R, containing 10 M of the fluorogenic peptide substrate. Complex dissociation was monitored for up to 1 h. The dissociation curves were analyzed as described previously (52) to yield the k off values. To monitor MT3-MMP cat inhibition by marimastat and batimastat, initial rates were obtained by adding the enzyme (0.5 nM) to a solution containing the fluorogenic substrate (7 M), and various concentrations of the inhibitor (up to 40 nM), in buffer R, in semi-microquartz cuvettes, and monitoring substrate hydrolysis for 10 min. The initial velocities were determined by linear regression analy-sis of the fluorescence versus time traces using FeliX TM software. Analysis of the initial rate dependence on the inhibitor concentration, according to a competitive model of inhibition, yielded the K i values, as previously described (52).
Cell Surface Biotinylation and Plasma Membrane Isolation-To detect MT3-MMP forms on the cell surface, CV-1 cells were infected/ transfected to express either wild-type or ⌬CT-MT3-MMP followed by surface biotinylation, as described (18). Briefly, cells in six-well plates were infected with 10 pfu/cell of vTF7-3 virus in infection media for 45 min followed by transfection of plasmids containing either wild-type or ⌬CT-MT3-MMP cDNA, using Effectene (Qiagen, Valencia, CA) as described by the manufacturer. Control cells were infected with vTF7-3 virus but received no plasmid DNA. Eighteen hours later, the cells were rinsed with cold phosphate-buffered saline (PBS) containing 0.1 mM CaCl 2 and 1 mM MgCl 2 and half of the dishes were incubated with 0.5 mg/ml sulfo-NHS-biotin as described (53), whereas the other half of the dishes received no biotin. The biotinylated and non-biotinylated cells were lysed with lysis buffer, followed by addition of streptavidin beads. After 12 h at 4°C, the beads were washed four times with harvest buffer (0.5% SDS in 60 mM Tris-HCl, pH 7.5, 2 mM EDTA) supplemented with 2.5% Triton X-100 (final concentration) followed by one wash with Tris-buffered saline (50 mM Tris-HCl, pH 7.5, 150 mM NaCl). The biotinylated proteins bound to the streptavidin beads were eluted with Laemmli SDS sample buffer, boiled, and resolved by reducing 10% SDS-PAGE followed by transfer to a nitrocellulose membrane. The biotinylated MT3-MMP forms were detected with anti-MT3-MMP as described (54). Plasma membrane fractions were isolated from BS-C-1 cells infected to express MT3-MMP or from human fibrosarcoma HT1080 cells, which express natural MT3-MMP. Briefly, HT1080 cells were grown to confluence in complete media. BS-C-1 cells were coinfected to express MT3-MMP as described above. Infected BS-C-1 cells and HT1080 cells (three 150-mm dishes each) were washed with cold PBS, scraped, and centrifuged at 1,000 ϫ g for 5 min. The cells were re-suspended in 25 mM Tris-HCl (pH 7.4) containing 8.5% sucrose, 50 mM NaCl, and protease inhibitor mixture (Roche Diagnostics) followed by homogenization in a Dounce homogenizer. The homogenates were centrifuged at 3,000 ϫ g for 10 min in a refrigerated centrifuge. The resulting supernatants were centrifuged at 100,000 ϫ g for 2 h at 4°C and the pellets were re-suspended in a 25 mM Tris-HCl (pH 7.4), 50 mM NaCl buffer containing protease inhibitors. The fractions were separated further on a discontinuous sucrose gradient (20,30,50, and 60% sucrose), and centrifuged at 100,000 ϫ g for 2 h at 4°C. The plasma membrane band (30/50% sucrose interface) was collected, pelleted at 100,000 ϫ g for 2 h, and stored at Ϫ80°C. The protein concentration was determined using the BCA assay (Pierce). Aliquots of the plasma membrane fractions were subjected to immunoblot analysis for detection of MT3-MMP and MT1-MMP using the appropriate antibodies as described (50).
Affinity Chromatography-Generation of a trimolecular complex between MT3-MMP, TIMP-2/TIMP-3, and pro-MMP-2 was investigated using the method developed by Bigg et al. (22). Briefly, human recombinant pro-MMP-2 (750 nM) was incubated (2 h, 4°C) with 750 nM each of either human recombinant TIMP-2 or TIMP-3 in PBS. MT3-MMP cat (375 nM) was then added to the mixture (35 l final volume, molar ratio of pro-MMP-2:TIMP-2/3/MT3-MMP cat ϭ 1:1:0.5) for an additional 2-h incubation at 4°C. As a control, some samples contained one or two components (in various combinations) of the reaction proteins in the same amounts as described above. The samples were then loaded onto a column packed with gelatin-agarose beads (column volume of 100 l) that was pre-equilibrated with PBS (for TIMP-2) or PBS supplemented with 0.025% (v/v) Brij-35 (for TIMP-3). The loaded samples were then incubated for 1 h at 4°C followed by a brief spin and the flow-through was collected (unbound fraction). The column was then washed three times with 500 l each of PBS with 0.025% (v/v) Brij-35 and the bound proteins were dissociated from the beads with 50 l of 4ϫ Laemmli SDS sample buffer. The columns were then briefly centrifuged and the eluates were subjected to 12% SDS-PAGE under reducing conditions. The proteins were visualized by staining the gels with the Gelcode SilverSNAP Stain Kit II (Pierce).

RESULTS
Expression and Processing of MT3-MMP-Human MT3-MMP was expressed in TIMP-2 Ϫ/Ϫ fibroblasts (Fig. 1A, lane 2), BS-C-1 cells (Fig. 1A, lane 3), and COS-1 cells (Fig. 1A, lane 4) and was detected as a major form of ϳ63 kDa and two minor species of ϳ61 and ϳ59 kDa. In addition, a ϳ35-kDa form, which specifically reacted with the polyclonal antibody to MT3-MMP, was clearly detected in the TIMP-2 Ϫ/Ϫ fibroblasts (Fig.  1A, lane 2). BS-C-1 and COS-1 cells showed lower levels of the low molecular weight species, which was usually detected as a doublet of ϳ35,000 -37,000 (Fig. 1A, lanes 3 and 4). Although the polyclonal antibody to MT3-MMP exhibits significant crossreactivity with proteins in the range of 50 -40 kDa, proper controls indicated the specificity of the species described. The profile of the MT3-MMP species in the cell lysates and on the cell surface (shown in Fig. 1C) was undistinguishable even when cells were infected with very low virus levels (data not shown) indicating that the level of MT3-MMP expression had no significant effect on cellular distribution. The appearance of the ϳ35-kDa form of MT3-MMP in the lysate of TIMP-2 Ϫ/Ϫ cells prompted us to examine the cell culture medium for the presence of shed MT3-MMP forms. A ϳ32-kDa soluble protein reactive with the polyclonal antibody to MT3-MMP was specifically identified in the media of TIMP-2 Ϫ/Ϫ fibroblasts expressing MT3-MMP (Fig. 1B, lane 2), which was absent in the control cells (Fig. 1B, lane 1). Surface-biotinylated CV-1 cells expressing MT3-MMP showed that the 63-, 59-, and 35-kDa species were present on the cell surface, albeit at different levels (Fig. 1C). CV-1 cells expressing ⌬CT-MT3-MMP exhibited on the cell surface forms of ϳ62, 58, and 33 kDa, consistent with a truncation at the C-terminal end. However, at similar levels of expression, ⌬CT-MT3-MMP showed much greater levels of the low molecular weight form when compared with the wild-type enzyme (Fig. 1C, lane 2). Purified plasma membrane fractions isolated from BS-C-1 cells co-infected to express MT3-MMP (Fig. 1D, lane 3) or from HT1080 cells (Fig. 1D, lane 2) contained mostly the 63-kDa species of MT3-MMP. The 59-kDa species of MT3-MMP was also detected in the plasma membranes of BS-C-1 cells (Fig. 1D, lane 3). The 35-37-kDa form was barely detected, suggesting that it is not retained after plasma membrane isolation. As expected, the plasma membranes of HT1080 cells contained the 57-kDa form of MT1-MMP (Fig. 1D, lane 1). However, higher amounts of membrane protein were required to detect MT3-MMP (10 g) when compared with MT1-MMP (5 g), indicating the low level of MT3-MMP expression in HT1080 cells.
We expressed various MT3-MMP mutants including a catalytic inactive mutant (E/A), a cytosolic tail deletion mutant (⌬CT), and a transmembrane domain and cytosolic tail-deleted mutant (⌬TM/CT) ( Fig. 2A). In general, all of these mutants exhibited enzyme forms consistent with the presence of the 63-, 61-, and 59-kDa species. However, the relative electrophoretic migration and amount of these forms varied accordingly with the type of mutation/truncation. The catalytic inactive MT3-MMP (E/A-MT3-MMP) did not produce the 35-37-kDa species indicating that this form is the product of autocatalytic degradation. In contrast, cells expressing ⌬CT-MT3-MMP showed greater levels of the low molecular weight form ( Fig. 2A). The ⌬TM/CT-MT3-MMP showed lower levels of the 35-37-kDa form when compared with the wild-type enzyme suggesting that MT3-MMP autolysis requires membrane insertion. In TIMP-2 Ϫ/Ϫ fibroblasts expressing wild-type MT3-MMP, processing to the 35-kDa species was dose dependently inhibited by marimastat (Fig. 2B). Marimastat also caused a slight, but readily detectable accumulation of the 61-and 59-kDa species (Fig. 2B). However, quantitatively, disappearance of the 35-37 kDa did not correlate with an equivalent accumulation of the 59-kDa species, probably because of a differential reactivity of Eighteen hours after transfection, the cells were surface-biotinylated (Biotin) as previously described (50). Parallel plates of cells were similarly infected/transfected but the cells were not biotinylated (No biotin). The cell lysates were then precipitated with streptavidin beads and the bound proteins were resolved by reducing 10% SDS-PAGE, transferred to a nitrocellulose membrane, and detected with the polyclonal antibody against MT3-MMP. D, plasma membranes isolated from HT1080 cells (lanes 1 and 2) (Fig. 3A, lane 2) suggesting that the first cleavage site in the pro-domain of pro-MMP-2 is specifically mediated by MT3-MMP and occurs independently of TIMP-2 expression. Immunoblot analysis showed the characteristic 63-, 61-, 59-, and 35-37-kDa species of MT3-MMP (Fig. 3B). However, the level of the 35-37-kDa forms decreased as a function of TIMP-2 expression. In contrast, high levels of TIMP-2 correlated with a slight but noticeable accumulation of the 59-kDa species of MT3-MMP (Fig. 3B, lanes 5-9). Thus, MT3-MMP undergoes an autocatalytic degradation process that is regulated by TIMP-2 and synthetic MMP inhibitors.
Shed MT3-MMP Can Process Pro-MMP-2-We next examined whether the soluble ϳ32-kDa fragment of MT3-MMP (shown in Fig. 1B, lane 2) is catalytically active, as described under "Experimental Procedures." These studies demonstrated that pro-MMP-2 was specifically processed when exposed to the conditioned media of cells (BS-C-1 and TIMP-2 Ϫ/Ϫ cells) expressing MT3-MMP (Fig. 3D). In contrast, media from cells infected with vTF7-3 virus alone, as a control, had no effect (Fig. 3D, lanes 1, 3, 5, and 7). Processing of pro-MMP-2 by the MT3-MMP-containing media was dependent on zymogen concentration. At doses of 25 and 100 nM pro-MMP-2 (Fig. 3D,  lanes 2 and 4, respectively), the intermediate form of MMP-2 was detected most. However, full activation was detected at pro-MMP-2 concentrations of 500 nM (Fig. 3D, lanes 6 and 8), consistent with the effect of the pro-MMP-2 concentration on zymogen activation via the second MMP-2 autocatalytic step, as previously reported (55). The ϳ32-kDa species is a true soluble form because its activity was completely recovered in the supernatant fraction after ultracentrifugation. Thus, MT3-MMP sheds a ϳ32-kDa soluble form that is catalytically active. gous TIMP-2-deficient cells. However, in the absence of exogenous TIMP-2, active MMP-2 was only observed in the heterozygous TIMP-2 ϩ/Ϫ cells, which, as expected, express endogenous TIMP-2 (21). Dose dependent studies in the homozygous TIMP-2 Ϫ/Ϫ cells showed maximal pro-MMP-2 activation with 5-10 nM TIMP-2 (Fig. 4B). At high concentrations (50 -200 nM), TIMP-2 was inhibitory consistent with a biphasic effect of TIMP-2 on pro-MMP-2 activation, as reported for MT1-MMP (14,15,18,19). Although full pro-MMP-2 activation was only achieved in the presence of TIMP-2, formation of the intermediate form by MT3-MMP did not require exogenous TIMP-2, as previously reported with MT1-MMP (22,23). Activation of pro-MMP-2 in the presence of TIMP-2 (10 nM) and increasing concentrations of N-TIMP-2 (up to 200 nM) caused a dose-dependent inhibition in the generation of the fully active form of MMP-2 but had no effect on the formation of the intermediate form (Fig. 4C). These results suggest that a ternary complex may be involved in the activation of pro-MMP-2 by MT3-MMP. Consistently, administration of N-TIMP-2 alone (1-200 nM), which cannot bind to the hemopexin-like domain of pro-MMP-2, did not support pro-MMP-2 activation by MT3-MMP but inhibited formation of the intermediate form, consistent with inhibition of MT3-MMP (Fig. 4D).

TIMP-2 Is Required for Full Pro-MMP-2 Activation by MT3-MMP-Because
Marimastat and TIMP-2 Enhance Pro-MMP-2 Activation by MT3-MMP-We previously reported that pretreatment of cells expressing MT1-MMP with reversible hydroxamic-based synthetic MMP inhibitors resulted in enhanced pro-MMP-2 activation after exposure to TIMP-2 (21). Here we report that this effect is also evident in cells expressing MT3-MMP. As shown in Fig. 5 Table I shows that TIMP-2 binds MT3-MMP cat with a ϳ2-5-fold lower affinity than MT1-MMP cat due to a faster dissociation (k off ) rate because the association rates are comparable (k on ϭ 1.78 ϫ 10 6 and 2.74 ϫ 10 6 M Ϫ1 s Ϫ1 , respectively). It should be mentioned that, although not clearly evident in the values shown in Table I, the difference in dissociation (k off ) rates between the MT3-MMP cat ⅐TIMP-2 and MT1-MMP cat ⅐TIMP-2 complexes may be even more pronounced, because, under the experimental conditions used, the direct k off determination for the MT1-MMP cat -TIMP-2 interaction was not feasible, and the value listed is an estimate based on a 10-fold difference between the slopes of the linear portions of the dissociation curves for the complexes of MT1-MMP cat with N-TIMP-2 (steady-state rate) and fulllength TIMP-2, as previously described (21). In contrast, dissociation of the MT3-MMP cat ⅐TIMP-2 complex was readily observed yielding a k off value between 3 and 6 ϫ 10 Ϫ4 s Ϫ1 . Thus, the differential affinity of TIMP-2 toward MT1-MMP and MT3-MMP may account, at least in part, for the observed differences in pro-MMP-2 activation. (24) and TIMP-3 (28), like TIMP-2, are known to form a non-covalent complex with pro-MMP-2. Therefore, we examined whether TIMP-4 and TIMP-3 could promote pro-MMP-2 activation by MT3-MMP in TIMP-2 Ϫ/Ϫ cells. As shown in Fig. 7A, TIMP-3 promoted pro-MMP-2 activation. However, as opposed to TIMP-2, the active enzyme produced in the presence of TIMP-3 was mostly detected in the cell lysates, whereas the culture medium exhibited the latent and intermediate forms of MMP-2 (Fig. 7A). TIMP-3 enhancement of pro-MMP-2 activation was also dose dependent and at a concentration of 100 nM, pro-MMP-2 processing was inhibited in both the media and the cells (Fig. 7B). At the highest levels (40 -100 nM) of TIMP-3, more pro-MMP-2 was also associated with the cells. As shown with TIMP-2 (Fig. 4), formation of the intermediate form of MMP-2 was also TIMP-3 independent (Fig. 7B). N-TIMP-3 (1-30 nM) had no enhancing effect on pro-MMP-2 activation by MT3-MMP but inhibited formation of the intermediate form (data not shown). Furthermore, TIMP-3 had no effect on the activation of pro-MMP-2 by MT1-MMP (data not shown), in agreement with a previous study (28). TIMP-4 did not support the activation of pro-MMP-2 by MT3-MMP and, in fact, inhibited the formation of the intermediate form of MMP-2 at the highest concentrations (Fig. 7C).

TIMP-3 but Not TIMP-4 Supports Pro-MMP-2 Activation by MT3-MMP-TIMP-4
We studied the kinetics of inhibition of MT1-MMP cat and MT3-MMP cat by TIMPs and synthetic MMP inhibitors. N-TIMP-3 was used in these studies because of nonspecific adsorption of wild-type TIMP-3 to the acrylic cuvettes. As shown in Table I, N-TIMP-3 exhibits the typical kinetic characteristics of a tight, slow-binding, reversible inhibitor of both enzymes. However, N-TIMP-3 inhibited MT3-MMP cat with a significantly higher affinity (ϳ20-fold) than that for MT1-MMP cat (K i values of 0.008 and 0.16 nM, respectively), due to a lower dissociation rate (k off ϭ 2.9 ϫ 10 Ϫ5 and 1.7 ϫ 10 Ϫ4 s Ϫ1 , respectively) because the k on values were practically indistinguishable. TIMP-4 exhibited a comparable lower affinity for both MT3-MMP cat and MT1-MMP cat (K i ϭ 0.3 nM) ( Table I).
The affinity of TIMP-4 for MT1-MMP cat is in good agreement to that previously reported by Troeberg et al. (25). TIMP-1 interaction with MT1-MMP cat has been shown to be practically negligible (22,55) and, at concentrations up to 9 nM, showed no inhibition of MT3-MMP cat (data not shown). In contrast to the TIMPs, the synthetic hydroxamate inhibitors were unable to discriminate between the two MT-MMPs and both marimastat and batimastat inhibited the MT-MMPs competitively, with similar but considerably lower affinity (K i values in the low nanomolar range) (Table I).
TIMP-3 and TIMP-2 each Form a Trimolecular Complex with Pro-MMP-2 and MT3-MMP cat -Because both TIMP-3 and TIMP-2 promoted pro-MMP-2 activation by MT3-MMP, we examined the ability of these proteins to form a trimolecular complex using gelatin affinity chromatography as previously described (22). Pro-MMP-2 was incubated with TIMP-3 or TIMP-2 to form zymogen-inhibitor complexes, which then were incubated with MT3-MMP cat as described under "Experimental Procedures." The mixtures were subjected to gelatin affinity chromatography and the bound (Fig. 8, A and B) and unbound (Fig. 8, C and D) fractions were analyzed by silver-stained SDS-PAGE. As shown in Fig. 8, most of the MT3-MMP cat was specifically recovered in the bound fraction with TIMP-3 (Fig.  8A, lane 4) or with TIMP-2 (Fig. 8B, lane 4) only in the samples containing pro-MMP-2 consistent with the formation of trimolecular complexes. In contrast, when MT3-MMP cat , TIMP-3, and TIMP-2 were incubated, alone or in combination, with gelatin beads in the absence of pro-MMP-2, the proteins were recovered in the unbound fraction (Fig. 8, C and D). Some processing of pro-MMP-2 to the intermediate form occurred when MT3-MMP cat was incubated with pro-MMP-2 in the absence of TIMPs, which was recovered in the bound fraction, as expected (Fig. 8, A and B, lanes 5). However, MT3-MMP cat did not bind to the beads under these conditions and was recovered in the unbound fraction, even in the presence of TIMP-3 or TIMP-2 (Fig. 8, C and D, lanes 3 and 6). TIMP-3 and TIMP-2 were also recovered in the bound fraction when complexed with pro-MMP-2 (Fig. 8, A and B, lanes 7), as expected.  (21). b Estimated value based on a 10-fold difference between the slopes of the linear portions of the dissociation curves for the complexes of MT1-MMP cat with N-TIMP-2 (steady state rate) and wild type TIMP-2 (21).

DISCUSSION
Accumulating evidence indicates that type I transmembrane MT-MMPs, as opposed to the secreted MMPs, are uniquely regulated by TIMPs (1,6,56). The TIMPs not only act in inhibition of catalysis but also have profound effects on MT-MMP turnover (processing and internalization), which eventually determines the level of mature enzyme on the cell surface (6, 56 -58). In the case of TIMP-2 and MT1-MMP, their interaction can also promote pericellular proteolysis by supporting the activation of pro-MMP-2 via formation of ternary complexes (14,22). The present study shows that these properties can also be applied to the interactions of MT3-MMP with TIMP-2 and with TIMP-3. TIMP-3 is a member of the TIMP family of metalloproteinase inhibitors and is mostly bound to the ECM (9, 10, 59 -62). Although only a few quantitative studies on the kinetics of TIMP-3 inhibition have been published and the number of MMPs that were examined is rather small, the consensus is that TIMP-3 is an efficient MMP inhibitor (9). In regards to MT-MMPs, TIMP-3 inhibition of MT1and MT2-MMP is similar to that of TIMP-2 (27,55). In addition, TIMP-3 is a potent inhibitor of several members of the closely related family of metalloproteinases, the ADAMs (a disintegrin and a metalloproteinase domain) (47,55,(63)(64)(65), further demonstrating the unique properties of TIMP-3. Our inhibition kinetic studies revealed that N-TIMP-3 binds MT3-MMP cat with significant higher affinity than MT1-MMP cat (K i ϭ 0.008 and 0.057/0.16 nM, respectively), mainly because of a lower rate of dissociation (2.9 ϫ 10 Ϫ5 and 1.7 ϫ 10 Ϫ4 s Ϫ1 , respectively). Moreover, N-TIMP-3 inhibits MT3-MMP cat considerably more strongly than TIMP-1 (K i Ͼ 9 nM), TIMP-2 (K i ϭ 0.17/0.31 nM), and TIMP-4 (K i ϭ 0.34 nM). Thus, it seems reasonable to propose that, among the members of the TIMP family of inhibitors, TIMP-3 is the most likely candidate to act as an MT3-MMP inhibitor, under physiological conditions. Although MT3-MMP is an activator of pro-MMP-2 (29,30,40,42), the role of TIMPs in this process has not been established. Here we showed for the first time that both TIMP-2 and TIMP-3 enhance the activation of pro-MMP-2 by MT3-MMP. However, there were significant differences between TIMP-3and TIMP-2-mediated activation of pro-MMP-2 by MT3-MMP. First, TIMP-3-mediated pro-MMP-2 activation was much less efficient than that elicited in the presence of TIMP-2 and second, active MMP-2 generated by MT3-MMP in the presence of TIMP-3 was mostly detected in the cell lysate, whereas with TIMP-2, MMP-2 was readily detected in both the cell lysate and in the medium (20). At present the mechanism(s) regulat-ing the release of active MMP-2 from the cell surface is unknown but it is likely to be a consequence of the interactions and affinities among the components of the activation complex (MT3-MMP, TIMP-3, and MMP-2). It is also possible that active MMP-2 remains cell associated via TIMP-3 bound to polyanionic molecules on the cell surface (28,66) or to the pericellular ECM (60,62). When compared with MT1-MMP, the activation of pro-MMP-2 by MT3-MMP in the presence of TIMP-2 was significantly less efficient. Although several reasons may account for this difference, TIMP-2 exhibits a slightly reduced affinity (ϳ2-5-fold) for MT3-MMP when compared with MT1-MMP, which is because of a faster dissociation rate. A recent study using chimeric MT-MMPs demonstrated that the catalytic domains of MT1-and MT3-MMP, which are responsible for TIMP-2 binding, are major determinants for the lower efficiency of MT3-MMP in promoting pro-MMP-2 activation (67). Together, these observations suggest that under conditions of MT1-MMP and MT3-MMP co-expression in cells (35,68), activation of pro-MMP-2 is likely to proceed preferentially via the MT1-MMP/TIMP-2 axis.
The role of TIMP-2 in pro-MMP-2 activation by MT1-MMP is to facilitate the assembly of a ternary complex on the cell surface (14,22). Our data suggest that a similar mechanism may apply for the activation of pro-MMP-2 by MT3-MMP in the presence of TIMP-2 and TIMP-3. First, the effect of TIMP-2 on activation was dose dependent and consistent with a biphasic effect because of titration of free active MT3-MMP on the cell surface, analogous to the process reported with MT1-MMP and TIMP-2 (14,15,18,19). Second, N-TIMP-2, which cannot bind to the hemopexin-like domain of pro-MMP-2, did not support activation and inhibited this process in the presence of TIMP-2. Third, TIMP-3, like TIMP-2, which binds pro-MMP-2 (28) and inhibits MT3-MMP with high affinity (shown here), supported the activation of pro-MMP-2 by MT3-MMP. In contrast, N-TIMP-3 did not support activation (data not shown). And fourth, pre-formed pro-MMP-2⅐TIMP-3 and pro-MMP-2⅐TIMP-2 complexes bound to the catalytic domain of MT3-MMP in solution yielding complexes that were specifically recovered in the bound fraction after gelatin affinity chromatography. Together, these results strongly suggest that activation of pro-MMP-2 by MT3-MMP in the presence of TIMP-2 or TIMP-3 proceeds via formation of ternary complexes on the cell surface, as reported for the activation of pro-MMP-2 by MT1-MMP in the presence of TIMP-2 (14,15,22). Structurally, the ability of TIMP-2 to support pro-MMP-2 activation via a ternary complex has been attributed at least in part on the anionic FIG. 8. Affinity chromatography of TIMP-3/TIMP-2 and MT3-MMP cat with pro-MMP-2. Pro-MMP-2⅐TIMP-3 (A and C) and pro-MMP-2⅐TIMP-2 (B and D) complexes were generated and incubated with MT3-MMP cat as described under "Experimental Procedures." As a control, some samples contained one or two components (in various combinations) of the reaction proteins (indicated by ϩ or Ϫ). The samples were subjected to gelatin affinity chromatography and the bound (A and B) and unbound (C and D) fractions were subjected to reducing 12% SDS-PAGE. The proteins were detected by silver staining.
character of its C-terminal tail, which may account for the high affinity interaction with the positively charged hemopexin-like domain of pro-MMP-2 (28, 69 -71). The binding affinity and conformation of the pro-MMP-2⅐TIMP-2 complex allows the N-terminal region of TIMP-2 to dock into the active site of MT1-MMP yielding trimolecular complexes either in solution (22) or on the cell surface (14). Because the C-terminal tail of TIMP-3 has a net charge of zero, based on amino acid sequence (70), it has been suggested that the interaction of TIMP-3 with the positively charged hemopexin-like domain of pro-MMP-2 may be further weakened (70). In addition, studies using gelatin affinity chromatography suggested a lower affinity of pro-MMP-2 for TIMP-3 when compared with TIMP-2 (28). Therefore, the ability of TIMP-3 to support pro-MMP-2 activation via a ternary complex with MT-MMPs has been questioned (70). Although the affinities between TIMP-3/TIMP-2 and pro-MMP-2 in the presence and absence of MT3-MMP need to be measured directly in future studies, the results presented here could not have been predicted solely from the amino acid sequence and charge distribution within the C-terminal tail of TIMP-3 and the hemopexin-like domain of pro-MMP-2, which under physiological conditions may very well be fully solvated. Solvation of charged amino acids on the surfaces of proteins may at times be so effective that their involvement in interactions with other structural partners may be minimized. As such, it is entirely conceivable that other structural elements such as strong hydrogen bonding or hydrophobic interactions may also contribute to the binding affinities and assembly of ternary complexes. On the other hand, the reported weaker interaction of pro-MMP-2 with TIMP-3 when compared with TIMP-2 (28) may partly explain why TIMP-3 was less efficient than TIMP-2 in promoting pro-MMP-2 activation by MT3-MMP. Considering that TIMP-3 can bind tightly to the pericellular ECM, it is possible that zymogen activation may also ensue as a consequence of docking of pro-MMP-2 via its hemopexin-like domain to ECM-bound TIMP-3, providing the interaction of pro-MMP-2 with MT3-MMP on the cell surface is not compromised. However, under the experimental conditions used in this study, this is not a likely scenario because TIMP-3 did not support pro-MMP-2 activation by MT1-MMP (our data and Ref. 28), further demonstrating the unique relationship between MT3-MMP and TIMP-3.
Expression of MT3-MMP in a variety of cells yielded a complex pattern of enzyme forms including four cell-associated forms of 63, 61, 59, and 35-37 kDa and a soluble form of ϳ32 kDa. Although determination of the true nature of these species awaits isolation and sequencing, based on molecular mass, the 63-and 59-kDa forms may represent the zymogen and active forms, respectively. The 63-kDa form of MT3-MMP was consistently detected on the cell surface and in the plasma membrane fraction of HT1080 cells and in infected cells. If this species is the zymogen form, this observation would imply that pro-MT3-MMP may also traffic to the cell surface, as has been reported with pro-MT1-MMP (18,72,73). Zymogen activation may then be accomplished on the cell surface by furin, which was also reported to be capable of trafficking to the extracellular milieu (74). The 59-kDa species, on the other hand, may represent the active enzyme because its relative amount increased in cells exposed to TIMP-2 or marimastat, consistent with stabilization of the mature enzyme by metalloproteinase inhibitors, as reported for the mature form of MT1-MMP (18,72,75). The 35-37-kDa form is an autocatalytic degradation product because of its sensitivity to TIMP-2 and synthetic MMP inhibitors and its absence in cells expressing the inactive E/A-MT3-MMP mutant. Thus, the level of MT3-MMP on the cell surface is regulated by autolysis of the active enzyme, as reported in previous studies (41,76). Here we showed that autolysis of MT3-MMP to the 35-37-kDa species was significantly enhanced after deletion of the cytosolic tail. In MT1-MMP, the cytosolic tail is required for endocytosis via clathrincoated vesicles, a process that is regulated by dynamin (77,78). Unpublished evidence from our laboratory using a dominantnegative mutant of dynamin 1 (dynamin K44A) show that co-expression of dynamin K44A with MT3-MMP results in enhanced pro-MMP-2 activation in the presence of TIMP-2. This result suggests that, analogous to MT1-MMP, the cytosolic tail of MT3-MMP regulates enzyme internalization and, indirectly, the rate of autocatalytic processing. The ability of MT1-and MT3-MMP to undergo autocatalytic processing may be usurped by reversible synthetic MMP inhibitors to produce unwarranted proteolytic effects because in the presence of TIMP-2, exposure to low doses of marimastat resulted in enhanced pro-MMP-2 activation. This process has been attributed to inhibition of autocatalytic turnover and/or to displacement of reversible synthetic MMP inhibitors such as marimastat from the active site by TIMP-2, which may promote formation of ternary complexes resulting in enhanced pro-MMP-2 activation (6,21). Further studies are required to establish the precise mechanism behind this paradoxical effect of reversible synthetic MMP inhibitors on MT-MMP regulation of catalytic activity. Finally, the soluble 32-kDa species found in the media of cells expressing MT3-MMP is a functional protease because it promoted the processing of pro-MMP-2 in solution. Thus, ectodomain shedding is also a property of MT3-MMP, which may regulate pericellular proteolysis at the cell surface and in the extracellular space. In summary, our data contribute to the emerging amount of evidence demonstrating the unique properties of the members of the MT-MMP subfamily and their differential regulation by TIMPs, which may endow cells with the ability to elicit the most effective proteolytic program at the pericellular microenvironment in both physiological and pathological conditions.