Oligomerization through hemopexin and cytoplasmic domains regulates the activity and turnover of membrane-type 1 matrix metalloproteinase.

The formation of multimeric complexes by membrane-type 1 matrix metalloproteinase (MT1-MMP) may facilitate its autocatalytic inactivation or proMMP-2 activation on the cell surface. To characterize these processes, we expressed various glutathione S-transferase/MT1-MMP fusion proteins in human HT-1080 fibrosarcoma cells and SV40-transformed lung fibroblasts and analyzed their effects on MT1-MMP activity and potential homophilic interactions. We report here that MT1-MMP is expressed on the cell surface as oligomeric 200--240-kDa complexes containing both the active 60-kDa and autocatalytically processed 43-kDa species. Overexpression of a glutathione S-transferase/MT1-MMP fusion protein containing the transmembrane and cytoplasmic domains of MT1-MMP inhibited the phorbol 12-myristate 13-acetate-induced autocatalytic cleavage of endogenous MT1-MMP to the 43-kDa species, but not proMMP-2 activation. On the other hand, a similar fusion protein with the hemopexin, transmembrane, and cytoplasmic domains inhibited proMMP-2 activation in a dominant-negative fashion. These results suggest that both the autocatalytic cleavage of MT1-MMP and proMMP-2 activation may be regulated by oligomerization through the cytoplasmic and hemopexin domains. Indeed, either domain, when attached to the cell membrane by a transmembrane domain, formed stable homophilic complexes. Copurification of MT1-MMP with these fusion proteins correlated with their cell-surface co-localization. Thus, MT1-MMP oligomerization through the hemopexin, transmembrane, and cytoplasmic domains controls its catalytic activity.

Spatial and temporal control of proteolytic activities is critical to the coordinated degradation of the extracellular matrix (ECM) 1 during various physiological tissue remodeling processes (1). Either excessive or deficient proteolysis is associated with a number of pathological conditions. For example, during tumor invasion and metastasis, neoplastic cells utilize proteolytic and invasive mechanisms in a spatially and temporally controlled, but abnormally regulated fashion (2). In normal tissues, a considerable molar excess of proteinase inhibitors either derived from plasma or secreted by the same tissue cells protects the surrounding ECM from degradation (3). Focal proteolytic activity may be achieved when proteinases are compartmentalized and activated in the immediate pericellular environment where proteinase inhibitors have limited access. Thus, cell-surface localization of proteinases via a transmembrane domain or binding to cell-surface receptors represents an important regulatory mechanism to target and concentrate proteolytic activity.
Membrane-type matrix metalloproteinases are a subset of MMPs that are associated with cell membranes by type I or II transmembrane domains or by a glycosylphosphatidylinositol anchor (4 -6). Membrane-type MMPs are effective ECM-degrading proteases (7)(8)(9), which may also serve as receptors and activators for secreted proteinases, including proMMP-2 (10,11). Therefore, changes in their cell-surface expression, distribution, and interactions may efficiently regulate pericellular proteolysis during tissue remodeling processes (12)(13)(14)(15). Compared with secreted proteases, membrane-type MMPs can be more precisely targeted at the cell surface and regulated cooperatively by cellular pathways involved in cytoskeletal rearrangement, signaling, and cell adhesion (16 -19). Indeed, targeting of MT1-MMP (MMP-14) activities by its transmembrane and cytoplasmic domains is crucial for cell invasion and migration, with both processes dependent upon the interplay between adhesive and proteolytic events (20 -22).
Based on current knowledge, proMMP-2 activation requires the formation of a ternary complex between MT1-MMP, TIMP-2, and MMP-2, followed by an initial cleavage of the MMP-2 prodomain by an adjacent TIMP-2-free MT1-MMP molecule (11,(23)(24)(25). In HT-1080 fibrosarcoma cells, proMMP-2 activation correlates with the N-terminal cleavage of active 60-kDa MT1-MMPs to proteolytically inactive 43-kDa membrane-bound forms (26 -28). As a result, a single soluble ϳ20-kDa fragment, which contains the catalytic site, is released by this autocatalytic cleavage (21). Recently, MT1-MMP dimers or oligomers, which may facilitate proMMP-2 activation and au-tocatalytic MT1-MMP inactivation processes, have been observed in platelets and HT-1080 cells (22,29). Disulfide bridges between cytoplasmic tails have been suggested to covalently link MT1-MMP molecules to dimers on the MCF-7 cell surface (22). In this work, we present evidence that MT1-MMP activates proMMP-2 and inactivates neighboring MT1-MMP in oligomeric complexes on the cell surface and that MT1-MMP mutants devoid of the catalytic domain have a dominant-negative effect on proMMP-2 activation by interfering with the oligomerization of wild-type MT1-MMP molecules.
Cell Cultures-Human HT-1080 fibrosarcoma cells (CCL-121, American Type Culture Collection, Manassas, VA), human embryonic lung fibroblasts (CCL-137, American Type Culture Collection), and SV40transformed human WI-38 lung fibroblasts (VA-13, CCL-75.1) were cultivated and treated with chemicals (20 nM PMA, 50 g/ml ConA, and 1 M BB-3103) as described (27). Cell-conditioned media and cell lysates were prepared as described (27). When a low background of endogenous MT1-MMP was desired, the transfection experiments were carried out in VA-13 cells, which express significantly lower levels of endogenous MT1-MMP. In studies in which the function of endogenous MT1-MMP was analyzed, we used mainly HT-1080 cells, which express easily detectable levels of the protein.
cDNA Constructs-The cloning of constructs ⌬Cyt and ⌬TM as well as full-length MT1-MMP cDNA has been described (7,21,30). A 801-bp fragment corresponding to coding sequence for Schistosoma japonicum GST was amplified by PCR using the pAcSecG2T baculovirus transfer vector (Pharmingen, San Diego, CA) as a template and ligated to the pSignal eukaryotic expression vector (31) derived from pcDNA3 (Invitrogen), generating the pc3GST plasmid encoding GST. Fragments of MT1-MMP cDNA (bases 1433-1906 and 1005-1906) were amplified by PCR and cloned into the pc3GST plasmid. The resulting ⌬Cat/Pex and ⌬Cat constructs encode non-catalytic fusion proteins of GST and amino acid residues 296 -582 and 438 -582 of MT1-MMP, respectively (illustrated in Fig. 1). For TM/Cyt, a fragment containing bases 1692-1868 (corresponding to amino acids 525-582) of MT1-MMP and a hemagglutinin epitope sequence at the 5Ј-end was amplified by PCR and ligated to pSignal. All constructs were verified by sequencing.
Expression of Recombinant MT1-MMP Proteins in HT-1080 and VA-13 Cells-Cells were cultured in six-well plates until they reached 50 -80% confluence and transfected using 2 g of plasmid DNA and 5 l of FuGENE 6 transfection reagent (Roche Molecular Biochemicals, Mannheim, Germany) for each transfection according to the manufacturer's instructions. Stable transfectants were selected with 0.6 mg/ml G418 (Invitrogen) in complete medium. Expression of recombinant MT1-MMPs by transfected cell clones was analyzed by immunoblotting.
SDS-PAGE, Immunoblotting, and Gelatin Zymography-SDS-PAGE was carried out using 4 -15% gradient or 7.5 and 10% standard Laemmli polyacrylamide gels (Bio-Rad). The proteins were then transferred to nitrocellulose membranes, and their immunodetection was performed as described (26). Aliquots of the cell-conditioned medium were analyzed by gelatin zymography and reverse zymography as described (26).
Immunofluorescence Analysis-Cells were cultured on glass coverslips, treated with PMA and BB-3103, washed with phosphate-buffered saline, and fixed. The cells were then labeled with the indicated anti-MT1-MMP or anti-GST antibodies followed by rhodamine-or fluorescein isothiocyanate-conjugated secondary antibodies (Jackson Immu-noResearch Laboratories, Inc., West Grove, PA), and the fluorescence images were obtained as described (21).
Cell-surface and Metabolic Labeling, Cross-linking, Immunoprecipitation, and Purification of GST/MT1-MMP Fusion Proteins-Cell-surface biotinylation and metabolic labeling were carried out as described (27). For cross-linking experiments, metabolically labeled (50 Ci/ml [ 35 S]methionine) or surface-labeled (0.2 mg/ml sulfosuccinimidobiotin) cells were incubated with the non-permeant thiol-cleavable cross-linker 3,3Ј-dithiobis(sulfosuccinimidyl propionate) (1 mM) for 1 h on ice, washed, and lysed. Aliquots of the lysates were subjected to immunoprecipitation with Ab-2 (raised against the cytoplasmic domain of MT1-MMP), followed by SDS-PAGE under nonreducing or reducing (50 -100 mM dithiothreitol) conditions and detection by fluorography or horseradish peroxidase-conjugated streptavidin as described (27). Alternatively, the cell-surface proteins were cross-linked with the non-cleavable cross-linker bis(sulfosuccinimidyl) suberate (0.5 mg/ml). The cells were then washed and lysed, and aliquots of cell lysates were fractionated by SDS-PAGE under reducing conditions and analyzed by immunoblotting as indicated. To analyze the processing time courses of cell surface-biotinylated MT1-MMP proteins, the cells were surface-biotinylated as described above, shifted to 37°C for increasing time periods (0 -180 min) to allow processing, lysed, and subjected to immunoprecipitation. For GST pull-down analyses, the non-catalytic GST/MT1-MMP fusion proteins were captured from the cell lysates with glutathione-Sepharose and washed. The bound proteins were eluted with Laemmli sample buffer, fractionated by SDS-PAGE under nonreducing or reducing conditions, and analyzed by immunoblotting with anti-MT1-MMP or anti-GST antibodies. Aliquots of cell lysates corresponding to equal number of cells were used for SDS-PAGE, immunoprecipitation, and GST pull-down analysis.

Characterization of MT1-MMP Oligomers on the Cell Sur-
face-To test the hypothesis that MT1-MMP functions through a multimeric complex, we first analyzed cell-surface MT1-MMP by biotinylation and cross-linking with non-permeant reagents, followed by immunoprecipitation and detection with horserad-  [2][3][4]. Minor protein bands of ϳ150 kDa were also detectable under these conditions. In contrast, cross-linked endogenous MT1-MMP on the surface of CCL-137 fibroblasts appeared to be predominantly in ϳ150-kDa complexes with or without PMA treatment ( Fig. 2A, upper panel, lanes 5 and 6), which are incapable of activating proMMP-2 (26,27). Dissociation of the cross-linked complexes by reduction revealed that the cell-surface complexes of PMAtreated or MT1-MMP-overexpressing HT-1080 cells were composed of both the active 60-kDa and proteolytically inactive 43-kDa species ( Fig. 2A, lower panel, lanes 2-4), arguing that these MT1-MMP species are present on the cell surface as high molecular mass complexes.
To resolve the composition of these complexes, we labeled MT1-MMP-overexpressing HT-1080 cells with [ 35 S]methionine and performed cross-linking and immunoprecipitation as described above. We then fractionated the precipitated proteins by SDS-PAGE (7.5%) under nonreducing conditions and subsequently subjected a gel lane to a second SDS-PAGE (10%) under reducing conditions. In addition to the 200 -240-kDa complexes, the precipitates contained 63-, 60-, and 43-kDa species representing the pro-, active, and inactive MT1-MMP monomers, presumably from intracellular compartments that are inaccessible to cross-linking (Fig. 2B, horizontal lane, ϪDTT). Reduction caused dissociation of the cross-linked complexes mainly to the 60-and 43-kDa monomeric forms of MT1-MMP (Fig. 2B, ϩDTT). The 60-and 43-kDa forms were released from the ϳ240-kDa complexes, whereas the ϳ200-kDa complexes consisted mainly of the inactive 43-kDa forms (Fig.  2B). No other major protein bands were detected from the dissociated complexes. These data suggest that MT1-MMP molecules assemble into homo-oligomeric complexes on the HT-1080 cell surface, where the 43-kDa forms are generated.
Competitive Inhibition of Autocatalytic Processing of MT1-MMP to the 43-kDa Species by the Transmembrane/Cytoplasmic Domains-Identification of MT1-MMP oligomers on the cell surface led us to hypothesize that interactions between cytoplasmic or extracellular domains of adjacent MT1-MMP molecules could regulate the autocatalytic intermolecular processing in trans. To test this hypothesis, we designed GST fusion constructs encoding N-terminally truncated non-catalytic MT1-MMP mutants (Fig. 1). All functional extracellular domains were deleted from ⌬Cat/Pex, which contains amino acid residues 438 -582, corresponding to the transmembrane and cytoplasmic domains of MT1-MMP fused to the C terminus of GST. A similarly constructed fusion protein (⌬Cat) contains residues 296 -582 of MT1-MMP, corresponding to the hinge, hemopexin-like, transmembrane, and cytoplasmic domains fused to GST (Fig. 1). The plasmids coding for ⌬Cat/Pex and ⌬Cat and a control plasmid coding for GST were transfected into VA-13 and HT-1080 cells, and stable cell clones were generated by G418 selection and dilution cloning. Positive cell clones were identified by immunoblotting. Expression of TIMP-2 and MMP-2 was examined by reverse and direct gelatin zymography, respectively (data not shown). Three independent clones for each construct, which expressed similar levels of TIMP-2 and MMP-2, were used in the experiments.
To induce autocatalytic processing of endogenous MT1-MMP to the 43-kDa forms, we treated VA-13 cells and stable VA-13 cell clones expressing GST, ⌬Cat/Pex, or ⌬Cat with PMA. We purified these GST/MT1-MMP fusion proteins using glutathione-Sepharose and characterized glutathione-Sepharosebound and -unbound fractions of the cell lysates by immunoblotting with Ab-2 (raised against the cytoplasmic domain) (Fig. 3A, lanes 2-8). Ab-2 detects both endogenous MT1-MMP and the fusion proteins. GST was detected using anti-GST antibodies (Fig. 3A, lane 1). The ⌬Cat/Pex and ⌬Cat proteins bound to glutathione-Sepharose migrated as ϳ42and 65-kDa protein bands, respectively (Fig. 3A, lanes 3 and 4). In the lysates of PMA-treated control cells and cells expressing GST, endogenous MT1-MMP was detected as active 60-kDa and inactive 43-kDa forms (Fig. 3A, lanes 5 and 8). In contrast, overexpression of either of the mutants prevented PMA-induced processing of endogenous 60-kDa MT1-MMP to the 43-kDa forms (Fig. 3A, lanes 6 and 7). Low levels of ⌬Cat/Pex migrating right below the 43-kDa form and ⌬Cat migrating just above the 60-kDa form of MT1-MMP were retained in the corresponding glutathione-Sepharose-adsorbed cell lysates (Fig. 3A, lanes 6 and 7).
To further analyze the effects of ⌬Cat/Pex on the processing of cell-surface MT1-MMP, we performed a time course experiment. Cell-surface proteins were labeled with sulfosuccinimidobiotin on ice, and the cells were shifted to 37°C for up to 180 min to allow processing of biotinylated proteins (Fig. 3B). At the indicated time points, the cells were lysed, and the lysates were immunoprecipitated with Ab-2. As described previously (27), control cells expressed MT1-MMP as the 60-kDa mature form on the cell surface (Fig. 3B, upper left panel). PMA treatment caused processing of this form to the N-terminally cleaved 43-kDa species within 40 min (Fig. 3B, lower left pan- el). In contrast, the majority of the 60-kDa species remained unprocessed in cells expressing the ⌬Cat/Pex mutant under similar conditions (Fig. 3B, upper and lower right panels). Together, these results indicate that overexpression of the transmembrane/cytoplasmic domains is sufficient to interfere with the processing of MT1-MMP to the inactive ϳ43-kDa form.

MT1-MMP Mutant Lacking the Catalytic Domain Has a Dominant-negative Effect on MT1-MMP-mediated proMMP-2
Activation-We have reported previously that the autocatalytic processing of MT1-MMP to the 43-kDa form correlates with proMMP-2 activation in fibroblastic cells (26,27). The inhibition of the autocatalytic processing by ⌬Cat/Pex and ⌬Cat suggested that these mutants could interfere also with MT1-MMP-mediated proMMP-2 activation. To test this hypothesis, we treated HT-1080 cells overexpressing these mutants with PMA or ConA and analyzed their impact on proMMP-2 activation by zymography. PMA-or ConA-treated HT-1080 cells converted proMMP-2 to the intermediate 64-kDa and active 62-kDa forms as expected (Fig. 4, lanes 5 and 6). Unexpectedly, ⌬Cat/Pex fusion protein expression had only a minimal effect on PMA-induced proMMP-2 activation: it affected only the second autocatalytic cleavage of the intermediate 64-kDa form that leads to the generation of the fully activated 62-kDa enzyme (Fig. 4, lane 2 versus lane 5). However, the ⌬Cat fusion protein, which also contains the hemopexin domain, completely prevented MT1-MMP-mediated proMMP-2 activation in PMAtreated cells (Fig. 4, lane 8). In contrast, the ⌬Cat mutant failed to inhibit proMMP-2 activation in ConA-treated cells (Fig. 4, lane 9), presumably because ConA stimulates clustering of cell-surface glycoproteins, including MT1-MMP. Because competitive inhibition of MT1-MMP hemopexin domain interactions by the ⌬Cat mutant inhibited PMA-induced proMMP-2 activation, we consider these interactions critical for proMMP-2 activation.
Non-catalytic MT1-MMP Mutants Interfere with Endogenous MT1-MMP Activity by Forming Stable Complexes on the Cell Surface-To assess whether ⌬Cat/Pex and ⌬Cat interact directly with endogenous MT1-MMP, we purified the GST fusion proteins from ⌬Cat/Pexand ⌬Cat-expressing HT-1080 cells using glutathione-Sepharose. The bound proteins were separated by SDS-PAGE under reducing and nonreducing conditions and analyzed by immunoblotting. We used anti-GST antibodies to detect only the fusion proteins, Ab-2 (raised against the cytoplasmic domain of MT1-MMP) to detect both fusion proteins and membrane-bound MT1-MMP species (data not shown), Ab815 to detect ⌬Cat mutants and endogenous MT1-MMP, and Ab-3 to detect only endogenous MT1-MMP. Treatment of the cells with PMA or ConA had minor effects on the levels of purified ⌬Cat/Pex (Fig. 5A, lanes 1-4). However, endogenous MT1-MMP was copurified with ⌬Cat/Pex in PMA-or ConA-treated cells (Fig. 5A, lanes 6 -8), but not in untreated cells (lane 5). In contrast, endogenous MT1-MMP was copurified with ⌬Cat under all conditions tested (Fig. 5B, lanes 1-5,  Ab815 and Ab-3). PMA treatment slightly enhanced the copurification of endogenous MT1-MMP (Fig. 5B, lane 2, Ab815 and Ab-3), whereas simultaneous treatment with PMA and BB-3103 significantly enhanced the levels of both ⌬Cat and copurified endogenous MT1-MMP (lane 4). From these results, we concluded that interactions of both the transmembrane/cytoplasmic and hemopexin domains regulate MT1-MMP oligomerization. Under nonreducing conditions, ⌬Cat/Pex and ⌬Cat migrated mainly as ϳ90and 130-kDa bands, respectively, corresponding to disulfide-bonded dimers (Fig. 5A, lower panel,  lanes 1-4; for ⌬Cat, data not shown). A minor ϳ100-kDa band was detected above the 90-kDa ⌬Cat/Pex dimers by anti-GST antibodies in PMA-and ConA-treated cells (Fig. 5A, lower  panel, lanes 2-4), in parallel to the detection of endogenous MT1-MMP as an ϳ100-kDa band (lanes 6 -8), arguing the formation of covalent dimers between endogenous MT1-MMP and ⌬Cat/Pex.
The dominant-negative effects of the ⌬Cat/Pex and ⌬Cat mutants on MT1-MMP-mediated proteolytic events by forming complexes with full-length MT1-MMP suggested that they would co-localize at the sites of potential proteolytic activity. To test this hypothesis, we carried out double immunofluorescence analysis of non-permeabilized HT-1080 cells expressing the GST fusion proteins ⌬Cat/Pex and ⌬Cat with anti-GST and anti-MT1-MMP antibodies (Fig. 6A). Endogenous MT1-MMP partially co-localized with ⌬Cat, whereas no significant colocalization with ⌬Cat/Pex was detected at the cell surface (Fig.  6A, upper panels). Treatment of cells with PMA to enhance MT1-MMP synthesis and with BB-3103 to inhibit autocatalytic MT1-MMP activity increased the levels of the mutants ( and led to the co-localization of MT1-MMP with both ⌬Cat and ⌬Cat/Pex on the cell surface (Fig. 6A, lower panels).
Next we characterized the GST/MT1-MMP fusion protein complexes on the cell surface by cross-linking with the noncleavable reagent bis(sulfosuccinimidyl) suberate. Aliquots of the cell lysates were fractionated by SDS-PAGE under reducing conditions to exclude the formation of disulfide-bonded GST fusion protein complexes. Cross-linked endogenous MT1-MMP was detected in HT-1080 cells as ϳ240-kDa complexes by immunoblotting with Ab-2 (raised against the cytoplasmic domain of MT1-MMP) (Fig. 7, lanes 1, 2, and 4). Ab-3 (raised against the synthetic peptide corresponding to the exposed loop in the catalytic domain of MT1-MMP) failed to detect the crosslinked complexes, apparently due to loss of antigen recognition after acylation of the exposed lysine residue in the epitope sequence by the cross-linker (data not shown). Treatment of control as well as ⌬Catand ⌬Cat/Pex-expressing HT-1080 cells with PMA and BB-3103 increased the levels of crosslinked complexes (Fig. 7, lanes 2, 4, 6, and 8), whereas in untreated cells, similar complexes were detectable only with long ECL exposure times (lane 1). The levels of cross-linked complexes increased only slightly in cells expressing ⌬Cat/Pex, but significantly in cells expressing ⌬Cat compared with the control cells (Fig. 7, lanes 6 and 8 versus lane 10), suggesting the involvement of these mutants in complex formations. Expression of ⌬Cat/Pex also slightly increased the levels of endogenous MT1-MMP monomers in the lysates of cross-linked HT-1080 cells compared with the control cells (Fig. 7, lanes 5  and 6 versus lanes 3 and 4). No high molecular mass complexes were detected in the corresponding uncross-linked samples (Fig. 7, lanes 9 -14), except an ϳ130-kDa band in PMA-and BB-3103-treated cells that expressed ⌬Cat (lane 14), most likely corresponding to partially reduced ⌬Cat dimers. These co-localization, copurification, and cross-linking data argue strongly that ⌬Cat and ⌬Cat/Pex form stable complexes with MT1-MMP on the cell surface, offering an explanation for the dominant-negative effects of these non-catalytic mutants on PMA-induced MT1-MMP activity.
Hemopexin Domains Cooperate with the Cytoplasmic and Transmembrane Domains in MT1-MMP Oligomerization-Next we analyzed the roles of the cytoplasmic, transmembrane, and hemopexin domains in MT1-MMP oligomerization by carrying out a series of cotransfections in VA-13 cells as indicated (Fig. 8A). The non-catalytic GST fusion proteins ⌬Cat and ⌬Cat/Pex were captured onto glutathione-Sepharose, and the bound proteins were eluted and identified by immunoblotting with anti-GST and anti-MT1-MMP (Ab-3) antibodies as indicated. Full-length MT1-MMP was copurified with ⌬Cat/Pex, containing only the cytoplasmic and transmembrane domains of MT1-MMP (Fig. 8A, lane 1), whereas the ⌬Cyt and ⌬TM mutants, lacking the cytoplasmic and transmembrane/cytoplasmic domains, respectively, were unable to stably interact with ⌬Cat/ Pex (lanes 2 and 3). This suggests that the cytoplasmic domains were required for this interaction. On the other hand, ⌬Cyt, but not the soluble ⌬TM construct, was copurified with ⌬Cat, which contains also the hemopexin domain (Fig. 8A, lane 5 versus lane 6), although with substantially lower efficiency than full-length MT1-MMP ( lane 5 versus lane 4). The relative levels of the copurified ⌬Cyt protein compared with those of the corresponding wild-type MT1-MMP increased upon PMA treatment (Fig. 8A, lane 11 versus lane 10), suggesting that PMA treatment increases hemopexin domain interactions. On the other hand, the soluble ⌬TM construct was barely detectable in PMA-treated cells (Fig. 8A, lane 12). These results suggest that the hemopexin domains can form stable complexes independently of the cytoplasmic tail, but that membrane anchorage through the transmembrane domain is essential for the formation of these stable complexes.
The ⌬Cat/Pex mutant contains one-fourth of the MT1-MMP hemopexin domain, which is unlikely to form a structural unit of the hemopexin domain because the correctly folded structure of this domain is dependent on the intramolecular disulfide  5) and ⌬Cat-expressing cells (lanes 3 and 7) as well as the corresponding PMA-and BB-3103-treated cells (lanes 2, 4, 6, and 8) were lysed, and aliquots of the cell lysates were fractionated by SDS-PAGE (7.5%) under reducing conditions, followed by immunoblotting with anti-GST (lanes 1-4) 1-4, 9, and 10) and stable HT-1080 clones expressing ⌬Cat/ Pex (lanes 5, 6, 11, and 12) or ⌬Cat (lanes 7, 8, 13, and 14) were left untreated (lanes 1, 3, 5, 7, 9, 11, and 13)  bond between cysteine residues at both ends of this domain. To exclude the possibility that this part of the hemopexin domain would, however, contribute to the interactions between ⌬Cat/ Pex and wild-type MT1-MMP, we carried out cotransfection experiments with the TM/Cyt mutant (containing only the cytoplasmic and transmembrane domains of MT1-MMP) and the non-catalytic ⌬Cat/Pex and ⌬Cat mutants (Fig. 8B). The observed copurification of TM/Cyt with both the ⌬Cat/Pex and ⌬Cat mutants (Fig. 8B, lanes 2 and 4), in contrast to the absence of detectable signal from cells transfected with TM/Cyt alone (lane 5), confirms that the transmembrane and cytoplasmic domains are sufficient for stable complex formation.
To examine the effects of ⌬Cat/Pex and ⌬Cat on MT1-MMPmediated MMP-2 activation, we analyzed, by gelatin zymography, aliquots of the conditioned media of VA-13 cells cotransfected with wild-type MT1-MMP and ⌬Cat/Pex or ⌬Cat. Cells expressing wild-type MT1-MMP processed proMMP-2 to the 64-and 62-kDa forms as expected (Fig. 8C, lane 1). Coexpression of ⌬Cat/Pex with wild-type MT1-MMP reduced the processing of proMMP-2 to its intermediate and fully activated forms (Fig. 8C, lane 2), and cells expressing both wild-type MT1-MMP and ⌬Cat failed to activate MMP-2 (lane 3). Taken together, our results indicate that overexpression of non-catalytic mutants capable of interacting with wild-type MT1-MMP through the cytoplasmic, transmembrane, and extracellular hemopexin domains leads to competitive inhibition of MMP-2 activation by wild-type MT1-MMP molecules in a dominantnegative fashion.

DISCUSSION
The three main mechanisms controlling the activity of secretory MMPs include the regulation of 1) gene expression, 2) zymogen activation, and 3) inhibition by TIMPs (4). However, fibroblastic cells constitutively express MT1-MMP (26), which is constantly activated by furin-like proteinases (7,32). Furthermore, MT1-MMP is usually coexpressed with TIMP-2 and MMP-2 (33)(34)(35)(36). Therefore, alternative mechanisms for the regulation of MT1-MMP activity are required. Increasing evidence suggests that the cytoplasmic and transmembrane domains of MT1-MMP are important for regulating its activity on the cell surface by mechanisms involving its targeting to special cell-surface structures and dimerization (20 -22). Two proteolytic events mediated by MT1-MMP, viz. the cleavage of proMMP-2, which is bound by a neighboring MT1-MMP/ TIMP-2 receptor, to the intermediate form and the intermolecular autocatalytic cleavage to the inactive 43-kDa form, are presumably dependent on the close proximity of at least two MT1-MMP molecules. Here we demonstrate that both proteolytic events are regulated by interactions through the hemopexin-like, transmembrane, and cytoplasmic domains of MT1-MMP to form oligomers. The interactions between hemopexin domains are critical for the MT1-MMP-mediated MMP-2 activation, and those between its transmembrane/cytoplasmic domains regulate the turnover of the enzyme to the 43-kDa form.
Based on our current observations, we propose a model in which MT1-MMP forms oligomers on the surface of fibroblastic cells, potentially dimers and tetramers bound together by two types of interactions through the hemopexin and cytoplasmic domains designated as type I and II interactions, respectively (Fig. 9). By copurification analyses with the non-catalytic GST fusion proteins ⌬Cat/Pex and ⌬Cat, we demonstrated that both of these domains, when attached to the cell membrane by the MT1-MMP transmembrane domain, formed stable homophilic interactions. The primary evidence for the presence of oligomeric MT1-MMP complexes came from cross-linking experiments on endogenous and transfected wild-type MT1-MMP in HT-1080 cells with the identification of the cell-surface ϳ200 -240-kDa complexes composed mainly of the 60-and 43-kDa species. In addition, two sequential SDS-PAGE analyses of metabolically labeled and cross-linked complexes demonstrated that the 200 -240-kDa complexes contain mainly MT1-MMP molecules, providing evidence that they are oligomers, most likely tetramers. However, under nonreducing conditions, MT1-MMP monomers migrated on SDS-polyacrylamide gels slightly faster than under reducing conditions, 2 and the apparent molecular mass of oligomers on gels did not necessarily increase exactly according to the molecular masses of the monomers. The 200 -240-kDa complexes could therefore also correspond, for example, to pentamers or hexamers of MT1-MMP. Alternatively, they may contain other protein(s) that have long half-lives and that are not efficiently biotinylated.
In the MT1-MMP complexes, a portion of the MT1-MMP molecules presumably bind TIMP-2 through the catalytic domain, and this TIMP-2 binds proMMP-2, forming a previously described ternary complex (11). Three proteolytic events may then occur in these complexes ( Fig. 9): 1) MT1-MMPs cleave the proMMP-2 molecules bound to the neighboring MT1-MMP to the intermediate 64-kDa form (24,25); 2) two 64-kDa MMP-2 molecules cleave each other in an intermolecular autocatalytic event to generate the 62-kDa form (23,37); and 3) MT1-MMP cleaves the other TIMP-free MT1-MMP to the inactive 43-kDa form (21, 26 -28, 38). The effects of ⌬Cat/Pex and ⌬Cat on these proteolytic steps support this model (Fig. 9). ⌬Cat/Pex, which is capable of type II interactions, but not type I interactions, competitively inhibited the formation of normal oligomeric complexes and prevented the processing of MT1-MMP to the 43-kDa form. However, ⌬Cat/Pex could not prevent MT1-MMP dimerization through hemopexin domains (type I interactions) or proMMP-2 activation, although proMMP-2 activation was reduced to some extent. On the other hand, ⌬Cat prevented the formation of the normal oligomeric complexes by competing with both type I and II interactions in a dominant-negative fashion. Consequently, both MT1-MMP processing to the 43-kDa form and MMP-2 activation were completely blocked.
In our cross-linking experiments, we were unable to identify TIMP-2-and MMP-2-containing complexes on the cell surface ( Fig. 2). Biotinylated protein bands possibly corresponding to coprecipitated TIMP-2 and MMP-2 were, however, detectable with long exposures of the reduced samples in the cross-linking analyses (data not shown). Our immunoblot analyses of the immunoprecipitates were, however, not sensitive enough to identify them directly. Previously, we observed coprecipitation of MMP-2 with MT1-MMP from HT-1080 and CCL-137 cells without cross-linking by gelatin zymography (27). However, those experiments suggest that only a minor proportion of MT1-MMP molecules form ternary complexes with TIMP-2 and MMP-2 in these cells or that the half-life of such complexes is relatively short.
The formation of disulfide bonds between Cys 57 4 residues on the cytoplasmic tails of neighboring MT1-MMP monomers may stabilize type II interactions as reported recently (22). However, under our experimental conditions, without cross-linking, the cell surface-biotinylated MT1-MMP proteins migrated mainly as monomers even under nonreducing conditions, and only a minor proportion were detectable as dimers or other complexes (data not shown). The purified ⌬Cat/Pex⅐MT1-MMP and ⌬Cat⅐MT1-MMP complexes migrated under nonreducing conditions mainly as dimers, indicating stabilization by disulfide bonds. However, it is possible that these bonds were formed by spontaneous oxidation during the cell extraction and purification processes. Nevertheless, our results do not rule out the possibility that MT1-MMP would be regulated on the cell surface by transient intermolecular disulfide bond formation between the cytoplasmic domains.
It remains unclear whether the MT1-MMP molecules in the oligomeric complexes on the control HT-1080 cell surfaces engage in both types of interactions. In contrast to ⌬Cat, ⌬Cat/ Pex (which lacks most of the hemopexin domain) did not colocalize or copurify with endogenous MT1-MMP in these cells without PMA or Con A stimulation. Type I interactions through the hemopexin domain could also be multivalent, supporting complexing of more than two molecules. Therefore, it is possible that type II interactions through the cytoplasmic domain only transiently regulate the function of these complexes in stimulated cells. Another associated event such as binding or 2 K. Lehti and J. Keski-Oja, unpublished data. A, wild-type MT1-MMP (wt) exists mainly as oligomers, potentially tetramers bound together by two types of interactions on the HT-1080 cell surface. Homophilic type I (I) interactions through the hemopexin domains are cooperative with type II (II) interactions, which occur through the cytoplasmic domains. Type II interactions may involve regulation by disulfide bond formation between Cys 574 residues in the cytoplasmic domains as suggested recently (22). Type II interactions may also bring together two dimers to a tetrameric conformation, and they can apparently stabilize also dimers bound together by type I interactions. Type I interactions may also be multivalent. Some of the cell-surface MT1-MMP molecules bind TIMP-2 (T2) through the catalytic center, and this TIMP-2 binds proMMP-2, forming a ternary complex. The three proteolytic events occurring in the complexes are the following: 1) MT1-MMPs cleave the proMMP-2 molecules bound to the neighboring MT1-MMP to their intermediate 64-kDa forms ([1]); 2) two 64-kDa MMP-2 molecules cleave each other in an intermolecular autocatalytic event to the 62-kDa form ([2]); and 3) the TIMP-free MT1-MMP cleaves another free MT1-MMP to the inactive 43-kDa form ( [3]). Alternatively, autocatalytic MMP-2 maturation (event 2) may also involve interactions of intermediate MMP-2 forms with ␣ v ␤ 3 integrin in some cellular systems (16,42). B, these proteolytic events are affected by the non-catalytic MT1-MMP mutants as follows. ⌬Cat/Pex, which is capable of type II interactions, but not type I interactions, competes with the formation of normal oligomeric complexes and inhibits MT1-MMP processing to the 43-kDa form. However, ⌬Cat/Pex cannot prevent MT1-MMP dimerization through the hemopexin domains (I) or proMMP-2 activation. ⌬Cat, which can form complexes through both type I and II interactions, prevents the formation of dimeric and higher oligomeric complexes of wild-type MT1-MMP by competition in a dominant-negative fashion. Therefore, MMP-2 activation is completely prevented. release of another protein, phosphorylation, or other modification of the cytoplasmic tail may modulate type II interactions. Cytoplasmic domain interaction may then induce conformational changes in MT1-MMP, which may in turn affect its proteolytic activity. Alternatively, extracellular events such as association or dissociation of the ternary complex, interactions with ECM proteins, or autocatalytic cleavage of MT1-MMP to the 43-kDa form may induce changes, which in turn affect type II interactions.
In conclusion, we suggest that the type I interaction is the primary mechanism to bring adjacent MT1-MMP molecules into close contact on the cell surface and to enable proMMP-2 activation. In support of this, Itoh et al. (39) provided evidence for homophilic MT1-MMP hemopexin domain interactions and their importance for proMMP-2 activation and cell invasion. On the other hand, type II interactions may be more related to the intracellular events regulating enzyme targeting to the sites of activity (20,21) and to its inactivation and removal from the cell surface (21,27). Indeed, we recently observed that the cytoplasmic domain is critical not only for targeting MT1-MMP activity at the cell surface, but also for regulating its cell-surface expression by endocytosis through clathrin-coated pits (40). Therefore, it will be interesting to analyze these interactions in the context of MT1-MMP trafficking and turnover. Our current observations along with the results of Itoh et al. (39) suggest an intriguing role for the autocatalytically processed 43-kDa form (26), which lacks the catalytic site. Instead of being only a by-product as previously suggested (27,41), it may also regulate adjacent active enzymes. Although, we have focused on proMMP-2 activation in this study, MT1-MMP can also directly degrade ECM proteins. The significance of MT1-MMP oligomerization for this ECM remodeling remains to be resolved.