Peptide Inhibition of Catalytic and Noncatalytic Activities of Matrix Metalloproteinase-9 Blocks Tumor Cell Migration and Invasion*

Migration of invasive cells appears to be dependent on matrix metalloproteinases (MMPs) anchored on the cell surface through integrins. We have previously demonstrated an interaction between the integrin (cid:1) -subunit I domain and the catalytic domain of MMP-9. We now show that there is also an interaction between the integrin (cid:2) subunit and MMP-9. Using phage display, we have developed MMP-9 inhibitors that bind either to the MMP-9 catalytic domain, the collagen binding domain, or the C-terminal hemopexin-like domain. The C-termi-nal domain-binding peptide mimics an activation epitope in the stalk of the integrin (cid:2) chain and inhibits the association of MMP-9 C-terminal domain with (cid:1) V (cid:2) 5 integrin. Unlike other MMP-9 binding peptides, it does not directly inhibit catalytic activity of MMP-9, but still prevents proenzyme activation and cell migration in vitro and tumor xenograft growth in vivo . We also find an association between MMP-9 and urokinase-plasmino-gen activator receptor and find that urokinase-plasmin-ogen activator receptor is cleaved by MMP-9. Collec-tively, we have defined molecular details for several interactions mediated by the different MMP-9 domains. Peptide Synthesis— The peptides using peptide syn- thesis done using -(9-fluorenyl)methoxycarbonyl) the purity and integrity The peptides the CRVYGPYLLC (CRV) and DDGW 50 M M the The


Migration of invasive cells appears to be dependent on matrix metalloproteinases (MMPs) anchored on the cell surface through integrins.
We have previously demonstrated an interaction between the integrin ␣-subunit I domain and the catalytic domain of MMP-9. We now show that there is also an interaction between the integrin ␤ subunit and MMP-9. Using phage display, we have developed MMP-9 inhibitors that bind either to the MMP-9 catalytic domain, the collagen binding domain, or the C-terminal hemopexin-like domain. The C-terminal domain-binding peptide mimics an activation epitope in the stalk of the integrin ␤ chain and inhibits the association of MMP-9 C-terminal domain with ␣ V ␤ 5 integrin. Unlike other MMP-9 binding peptides, it does not directly inhibit catalytic activity of MMP-9, but still prevents proenzyme activation and cell migration in vitro and tumor xenograft growth in vivo. We also find an association between MMP-9 and urokinase-plasminogen activator receptor and find that urokinase-plasminogen activator receptor is cleaved by MMP-9. Collectively, we have defined molecular details for several interactions mediated by the different MMP-9 domains.
Matrix metalloproteinases 2 and 9 (MMP-2 and -9), 1 also known as gelatinases, play an important role in cell migration and tissue remodeling during development but also in pathological conditions such as inflammation and cancer (1). We have identified a highly selective peptide inhibitor of gelatinases, CTTHWGFTLC (CTT) by phage display (2), whereas others have developed gelatinase-selective small molecule inhibitors (3) to specifically target these enzymes.
The unique structural feature of the gelatinases is the collagen-binding domain (CBD) within the catalytic domain (4). The CBD is composed of three fibronectin type II repeats and is an intriguing target to develop gelatinase-specific compounds.
Like most MMPs, the gelatinases also contain a C-terminal hemopexin/vitronectin-like domain (C domain or PEX), which contains the binding site for tissue inhibitors for matrix metalloproteinases (TIMPs) and is responsible for the dimerization of MMP-9 (5).
Although MMP-2 and MMP-9 are closely related enzymes, they do have differences in the regulation of expression, activation, and glycosylation and in substrate selectivity (1,4). Of these two enzymes, MMP-2 has been investigated in a more detail. For example, the activation of pro-MMP-2 has been thoroughly characterized and involves interactions of TIMP-2, MT1-MMP, and ␣ V ␤ 3 integrin on the cell surface (6,7). MMP-9 has not been found to be activated via the same mechanism, and several proteinases including the plasmin/MMP-3 cascade (8) and trypsin-2 (9) can activate MMP-9 in vitro.
Relatively little is known about the molecular details of the MMP-9 interactions on the cell surface and how these regulate cell migration. MMP-9 has been found to interact with the ␣ 5 ␤ 1 integrin, the ␣ 2 chain of type IV collagen, and the hyaluronan receptor CD44 (10,11). We have recently identified the leukocyte specific ␤ 2 -integrins as a binding partner for pro-MMP-9. The phage display peptide ADGACILWMDDGWCGAAG (DDGW) competed with pro-MMP-9 binding to the ligand-binding I domain of ␣ M integrin subunit and inhibited migration of leukocytes (12). Here we have isolated MMP-9 binding peptides, which inhibit either substrate binding or proenzyme activation, leading to an inhibition of cell migration and invasion. Using these peptides, we identify MMP-9 interaction sites in fibronectin, vitronectin, and ␣ V ␤ 5 integrin.
Peptide Synthesis-The phage peptides were initially prepared in a recombinant form using intein fusions (12,15). Chemical peptide synthesis was done using Fmoc (N-(9-fluorenyl)methoxycarbonyl) chemistry, and the purity and integrity of the peptides were verified by mass spectroscopy (15). The peptides were dissolved in water, except the CRVYGPYLLC (CRV) and DDGW peptides, which were dissolved in 50 mM NaOH at a 10 mM concentration and then diluted into PBS to neutralize the pH. The TTPNSLLVSWQPPRARIT and ADIMINF-GRWEHGDGYPF peptides were synthesized on a cellulose membrane. The membrane was blocked with 3% BSA in TBS plus 0.05% Tween 20 and incubated with 0.2 g/ml biotinylated CBD. Bound CBD was de-tected using peroxidase-conjugated streptavidin (1:10,000 dilution; Pierce) and chemiluminescence detection.
Gelatin and CBD Binding Assays-Recombinant CBD or human plasma fibronectin (Calbiochem) (0.2 g/ml in TBS) was immobilized in microtiter wells. The wells were saturated with 1% BSA-PBST. Biotinylated gelatin (0.2 g/ml in 1% BSA-PBST) was added with or without peptides at the concentrations indicated or with an excess of unlabeled gelatin (10 g/ml) and allowed to bind for 1 h. Bound gelatin was detected with streptavidin-peroxidase. CBD binding to immobilized fibronectin, the 110-kDa fragment of fibronectin (Upstate Biotechnology, Inc., Lake Placid, NY), or urea-denaturated human plasma vitronectin (17) (1 g/well) was studied using biotinylated CBD (5 g/ml) in 1% BSA-PBST in the presence or absence of 20 M peptides.
Dimerization of the MMP-9 C Domain-Recombinant C domain or CBD at a 5 g/ml concentration in PBS were coated on microtiter wells followed by blocking with 1% BSA-PBST. 125 I-Labeled C domain was preincubated with the peptides for 30 min in 1% BSA-PBST and then added to the wells. After a 2-h incubation, the wells were washed. Bound radioactivity was eluted with 1% SDS and measured with a ␥-counter.
Cell Adhesion-HT1080 cells were allowed to adhere on vitronectin or fibronectin (2 g/ml) in the presence or absence of peptides (200 M), proteins (40 g/ml), or a monoclonal anti-␣ V ␤ 5 integrin antibody P1F6 (Chemicon) or control antibody (25 g/ml). The adhesion was quantified as described (14). Adhesion of THP-1 cells to immobilized KIM127 or control monoclonal antibodies (2 g/ml) was done in the presence or absence of soluble proteins (50 g/ml) or antibodies (25 g/ml).
Cell Migration and Invasion-The cell migration assay was conducted using transwell migration chambers (8-m pore size; Costar) in 10% serum-containing medium (2,14). Briefly, the membranes were coated on both sides with 40 g/ml GST or with the ␤ 2 integrin ligand peptide CPCFLLGCC-GST fusion (GST-LLG-C4) and blocked with complete medium. THP-1 cells (50,000 cells/100 l) were preincubated with the peptides for 1 h in serum-containing medium. The cells were allowed to migrate for 16 h and were then stained with crystal violet and counted (14). The HT1080 (20,000 cells/100 l) invasion assay was performed as the THP-1 migration, except that matrigel-coated transwells (BD Biosciences) were used.
Pericellular Proteolysis-Microtiter wells were coated with a mixture of fibronectin (10 g/ml) and fluorescein isothiocyanate-labeled gelatin (100 g/ml) followed by saturation with 1% BSA in PBS. HT1080 cells (50,000 in 100 l of 0.1% BSA/Dulbecco's modified Eagle's medium) were incubated in the presence of 20 nM 4␤-phorbol-12,13-dibutyrate (PDBu) (Sigma) and the peptides or the MMP-2/MMP-9-selective inhibitor InhI (Calbiochem). As a control, nonactivated cells and medium without the cells were used. Gelatinolysis after 48 h was measured as the increase of fluorescence from a 50-l aliquot of the conditioned medium using a Wallac Victor 2 reader.
Pro-MMP-9 and Gelatin Binding to Leukocyte ␣ M Integrin-Pro-MMP-9 binding to the ␣ M I domain in the presence of peptides was studied as described (12). Gelatin binding to the pro-MMP-9⅐␣ M ␤ 2 integrin complex was studied by immobilizing the integrin ␣ M ␤ 2 (12) or ␣ IIb ␤ 3 as a control (Enzyme Research Laboratories, South Bend, IN) (1 g/well) in TBS plus 1 mM CaCl 2 and 1 mM MgCl 2 followed by saturation of the wells with 1% BSA in PBST. Pro-MMP-9 (100 ng/well) was incubated for 2 h, and the unbound pro-MMP-9 was washed away. Biotinylated gelatin (2.5 g/ml) was allowed to bind for 30 min at room temperature. Bound gelatin was detected with streptavidin-peroxidase.
Immunofluorescence-HT1080 cells were allowed to adhere on vitronectin (10 g/ml) in serum-free Dulbecco's modified Eagle's medium. Directional migration of the cells was stimulated by overlaying the cells with 0.5% agarose in Dulbecco's modified Eagle's medium and adding 5 l of fetal bovine serum with PDBu (20 nM final concentration) to the one end of the wells. Overnight cultured cells were washed with PBS, fixed with paraformaldehyde, permeabilized, and stained with the monoclonal anti-uPAR antibody (Ab3937, 2 g/ml; American Diagnostica) or anti-␤ 5 integrin IA9 (2 g/ml (21)) and polyclonal MMP-9 antibodies (H-129; 10 g/ml). The primary antibodies were detected with anti-mouse Alexa Fluor 488 and anti-rabbit Alexa Fluor 555 antibodies.
uPAR Cleavage-0.5 g of recombinant soluble human uPAR (R&D Systems) was digested with 50 ng of trypsin-activated MMP-9 in 50 mM Tris-HCl (pH 7.5), 5 mM CaCl 2 , 1 M ZnCl 2 , 0.02% NaN 3 , 10 g/ml aprotinin with or without 10 mM EDTA. Chymotrypsin cleavage was done without aprotinin. The samples were incubated for 16 h in 37°C and separated in a nonreducing 12% SDS-PAGE followed by Western blotting with anti-uPAR antibodies (399R; 1:1000 dilution). uPAR cleavage on the surface of HT1080 cells or THP-1 cells was studied in a serum-free medium with or without 20 nM PDBu for 48 h in the presence or absence of 20 M InhI, 200 M CTT or W3 A CTT control peptide, 25 g/ml aprotinin, or 20 M benzamidine. The cells were washed three times with PBS, incubated with 50 mM glycine HCl (pH 3.0) plus 100 mM NaCl to extract cell surface-bound urokinase-plasminogen activator (uPA) and MMPs, and neutralized with 500 mM Hepes (pH 7.5) plus 100 mM NaCl. Membrane proteins were enriched by Triton X-114 extraction (22), and 30 g (HT1080 cells) or 10 g (THP-1 cells) of protein was separated on 12% SDS-PAGE and analyzed for uPAR as above. Gelatinases were analyzed from the acid eluates with gelatin zymography and uPA with plasminogen/milk powder zymography (23).
Binding Assay-MMP-9 C domain, MMP-2 C domain, CBD, ␤ 5 I-EGF2ϩ3, or vitronectin (2 g/ml) were immobilized in microtiter wells. Biotinylated ␤ 5 I-EGF2ϩ3 fragment (2.5 g/ml) was added to the wells, which were preincubated with the competitors for 30 min in 1% BSA-PBST and then incubated further for 1 h. Bound biotinylated protein was detected with streptavidin-peroxidase. Statistical Analysis-Statistical significance was calculated with the t test or with log rank test in Kaplan-Meier survival analysis.

Identification of Peptide Probes to Different
Domains of MMP-9 -In order to understand gelatinase-mediated cell migration in depth, we searched for putative MMP-9-binding proteins by phage display of random peptide libraries. The pro-MMP-9 and its recombinant domains were used in biopanning, since the active MMP-9 primarily bound peptides with a WGF motif (2). Two groups of pro-MMP-9 binding peptides were found (Table I). Group I had a motif, CGArGRAr(S/ Q)PPC, where Ar represents an aromatic amino acid. These peptides show similarity to sequences found in the gelatinase substrates fibronectin and vitronectin (4). Group II had a CRX-YGPXXXC motif. In this group, the CRVYGPYLLC peptide was obtained by biopanning with pro-MMP-9, whereas the other sequences were obtained with a recombinant C-terminal domain. The CGYGRFSPPC (PPC) and CRV peptides were chosen for further studies as representatives of the two groups.
To identify the binding sites of these peptide motifs, we carried out phage binding experiments. Binding of PPC peptide-bearing phage to pro-MMP-9 was inhibited by a soluble recombinant 18-mer ADGACGYGRFSPPCGAAG (PPC) peptide and gelatin, but not with CTT or a recombinant ADGACRVYGPYLLCGAAG (CRV) peptide (Fig. 1A). Conversely, binding of the CRV-phage was inhibited by CRV and not by PPC, CTT, or gelatin, indicating nonoverlapping binding sites for these peptides. Inhibition of PPC phage binding by gelatin implied that that PPC binds to the CBD of MMP-9. Furthermore, phage selection with the MMP-2 CBD has also yielded a PPC-like peptide ACGYTYHPPCARLT (25). The PPC peptide, but not CTT or CRV, inhibited gelatin binding to immobilized CBD in a dose-dependent manner (Fig. 1B) but had no effect on gelatin binding to fibronectin (data not shown), suggesting that PPC is specific for the fibronectin type II repeats of gelatinases. In a gelatin degradation assay, PPC inhibited both MMP-9 and MMP-2 activity (Fig. 1C), the scrambled control peptide ADGACPSYGPRFGCGAAG (scr. PPC) having no effect. The CRV peptide did not inhibit gelatinase activity, consistent with the inability of gelatin to compete with CRV. The PPC peptide was a weaker gelatinase inhibitor than CTT, which completely inhibits gelatin degradation at a 100 M concentration in this assay (15).
To study whether the PPC-like sequences of fibronectin and vitronectin bind MMP-9, we examined the binding of CBD to these proteins in a solid phase binding assay. CBD bound to both fibronectin and vitronectin, but not to the 110-kDa fragment of fibronectin lacking the C-terminal heparin-binding domain and thus the suspected gelatinase-binding site (Fig.  1D). PPC, but not the scrambled peptide, inhibited the CBD binding. Similar results were obtained in a pepspot membrane assay, where biotinylated CBD bound to the PPC-like fibronectin peptide TTPNSLLVSWQPPRARIT but not to an 18-mer control peptide (Fig. 1D, inset).
When different MMPs were compared, the CRV phage showed a CRV peptide-inhibitable binding only to pro-MMP-9, and not to pro-MMP-2 or pro-MMP-3 ( Fig. 2A). The scrambled CRV peptide CGYLPLRYVC had no effect. MMP-9 selectivity was also observed with the recombinant MMP-9 and MMP-2 C domains. The CRV phage recognized the MMP-9 C domain strongly in comparison with the MMP-2 C domain (Fig. 2B).  TIMP-1 could not compete with the CRV phage binding to the MMP-9 C domain (Fig. 2C) or pro-MMP-9 (not shown). The CRV phage did not bind to the CBD (Fig. 2C) or a pro-MMP-9 lacking the hinge region and the C-terminal domain (pro-MMP-9-⌬HC; not shown). We next examined the effect of CRV on the dimerization of MMP-9 C domain. 125 I-labeled C domain was preincubated with CRV or scrambled peptide and then added to wells coated with unlabeled C domain. Dimerization of the C domain was inhibited by CRV but not by the scrambled peptide (Fig. 2D).
Cell Migration and Invasion Are Inhibited by Blocking the Domain-specific Interactions of the Gelatinases-We studied the role of the gelatinase domains in cell migration and invasion using the CTT, PPC, and CRV peptides. The binding site of CTT maps to the catalytic domain, but not to CBD (Fig. 1B  (12)). 2 As indicated above, PPC and CRV are probes for the CBD and the C domain, respectively. All three peptides inhibited HT1080 fibrosarcoma invasion into matrigel. At a 200 M concentration of CRV or CTT, 50% inhibition was observed. The PPC peptide required a 500 M concentration to achieve the same efficacy (Fig. 3A). The scrambled control peptides were inactive. Similar results were obtained with THP-1 monocytic cells, which migrate on a synthetic GST-LLG-C4 substratum (14) in a ␤ 2 integrin-and gelatinase-dependent manner. PPC, CRV, and CTT, but not the scrambled peptides, had an inhibitory effect (Fig. 3B). The inhibition of cell migration was not due to toxicity as there was no effect on cell viability when the cells were cultured for 48 h with the peptides at a 500 M concentration (not shown). Surprisingly, CRV inhibited peri-cellular gelatinolysis similarly as did CTT and PPC, as measured by a release of fluorescent gelatin fragments into the conditioned medium (Fig. 3C). In this assay, HT1080 cells were cultured for 48 h in the presence of PDBu on a fibronectin/ fluorescein isothiocyanate-labeled gelatin coating. The gelatinase-selective small molecule inhibitor (Inh1) also inhibited gelatinolysis, but the scrambled peptides did not. These results indicated that not only the direct MMP enzyme inhibitors but also CRV affects cell migration and pericellular proteolysis. We also tested that the CRV and PPC peptides do not affect the interaction of MMP-9 with the leukocyte ␣ M integrin I domain, which is blocked by DDGW (Fig. 3D). In fact, PPC stabilized pro-MMP-9 binding to the I domain as shown by typically 20 -50% higher binding in the presence of PPC. Antibody binding to pro-MMP-9 in the absence of the I domain was not affected by PPC (not shown). The data suggested that the ␣ M ␤ 2 integrin-bound MMP-9 could bind its substrates using CBD to generate a triple molecular complex between an integrin, MMP-9, and a ligand/substrate. To directly test this, pro-MMP-9 was allowed to bind to immobilized ␣ M ␤ 2 integrin, and binding of biotinylated gelatin, an MMP-9 substrate, was examined. Gelatin bound to the pro-MMP-9⅐␣ M ␤ 2 integrin complex but not the ␣ M ␤ 2 integrin alone. The platelet integrin ␣ IIb ␤ 3 did not support pro-MMP-9/gelatin binding (Fig. 3E).
MMP-9 Associates with the Urokinase-Plasminogen Activator Receptor-We next investigated the effects of the peptides on plasmin/MMP-3-mediated pro-MMP-9 activation in PDBuactivated HT1080 and THP-1 cells. The conditioned medium from the cells incubated in the presence of the peptides was analyzed by gelatin zymography. Of the three peptides, only CRV was capable of inhibiting pro-MMP-9 activation. In HT1080 cells, CRV peptide inhibited pro-MMP-9 activation strongly and the activation of pro-MMP-2 partially (Fig. 3F). The addition of plasminogen was sufficient in activating pro-MMP-9 in HT1080 cells, and pro-MMP-3 did not promote activation any further. In THP-1 cells, pro-MMP-9 activation required pro-MMP-3 and plasminogen added together, and the activation was blocked by CRV but not by the other peptides (Fig. 3G). In fact, pro-MMP-9 activation was augmented in the presence of PPC or DDGW, and there were higher levels of released MMP-9 as previously observed with DDGW (12). CRV did not inhibit the activation of purified pro-MMP-9 by MMP-3 in vitro (Fig. 3H).
Since the plasminogen activation cascade is involved in pro-MMP-9 activation, we considered the possibility that the urokinase receptor associates with MMP-9. Immunoprecipitations from PDBu-activated HT1080 cells showed that pro-MMP-9 co-precipitated with anti-uPAR antibodies but not with the control antibodies (Fig. 4A). The association of uPAR and pro-MMP-9 was similarly found in THP-1 cells and was not affected by prior PDBu activation (Fig. 4A). Several proteinases are able to cleave uPAR (26,27); we thus asked whether MMP-9 also does so. Using purified proteins, we observed that MMP-9 cleaved the domain 1 (D1) from uPAR similarly as does chymotrypsin (Fig. 4B). The uPAR cleavage by MMP-9 occurred in the presence of aprotinin and was inhibited by the metalloproteinase inhibitor EDTA. uPAR cleavage occurs on the surface of phorbol-ester-activated cells (26). To study the contribution of gelatinases in this process, we incubated HT1080 cells with proteinase inhibitors and analyzed the membrane protein-enriched lysates by Western blotting with antibodies to uPAR. The gelatinase-selective inhibitor InhI, but not the serine proteinase inhibitors aprotinin or benzamidine, inhibited uPAR cleavage (Fig. 4C). The inhibition of uPAR cleavage was accompanied with reduced gelatinase levels in the conditioned medium and on the cell surface. In the conditioned medium, 2 M. Björklund and E. Koivunen, unpublished data. MMP-9 occurred in higher levels than MMP-2, whereas the opposite was true for the cell surface. The cell surface-bound MMP-9 was in the latent form, as previously observed (28). In addition, the level of cell surface-bound uPA was reduced in the presence of InhI. uPAR cleavage on the THP-1 cells was similarly inhibited by InhI and CTT but not by the inactive W3 A CTT mutant peptide (15) or aprotinin (Fig. 4D). In the absence of PDBu, the THP-1 cells cultured in a serum-free medium expressed hardly detectable levels of uPAR.
The CRV Peptide Is a Mimic of an Integrin ␤ Chain Epitope-In nonleukocytic cells, uPAR is able to associate with ␤ 1 , ␤ 3 , and ␤ 5 integrins (29 -32). We thus investigated which integrin(s) could interact with MMP-9 in HT1080 cells. Immunoprecipitations were performed with antibodies against ␣ 2 , ␣ 3 , ␣ 5 , ␤ 3 , and ␤ 5 integrins. Pro-MMP-9 associated with the ␣ 5 and ␤ 5 integrins, indicating that ␣ 5 ␤ 1 and ␣ V ␤ 5 are the major integrins involved in pro-MMP-9 binding in HT1080 cells grown on a tissue culture-treated plastic (Fig. 5A). MMP-1 and -2 can interact with integrins through their C-terminal domains (6,33). Interestingly, a database search revealed that the CRV peptide bears a similarity to sequences found in the stalk of the integrin ␤ chains, in particular the ␤ 5 chain. Seven of the CRV amino acid residues had a matching or a similar residue in the ␤ 5 sequence (Fig. 5B). These sequences are located in the cysteine-rich I-EGF-like domain 2 and become exposed in the activated integrins, as shown by the reactivity of activation state-specific antibodies (34,35). Indeed, the antibody KIM127 epitope maps to the CRV-like sequence in the ␤ 2 integrin chain (34). To study whether MMP-9 binds to this integrin activation epitope, we first assessed the effect of the MMP-9 C domain on cell adhesion to vitronectin and fibronectin. Neither the C domain nor the pro-MMP-9-⌬HC (40 g/ml) or the CRV peptide (200 M) inhibited HT1080 cell adhesion to vitronectin or fibronectin (Fig. 5C). Adhesion to vitronectin occurred in a ␣ V ␤ 5dependent manner as demonstrated by inhibition with the ␣ V ␤ 5 integrin-blocking antibody P1F6 (25 g/ml). We did not observe specific adhesion of HT1080 cells to the immobilized C domain (not shown). These results indicated that the putative interaction site of the MMP-9 C domain in ␣ 5 ␤ 1 and ␣ V ␤ 5 is not the major RGD ligand-binding site or a cell adhesion determinant. This prompted us to express the I-EGF domains 2 and 3 (36) from the ␤ 5 integrin. Interestingly, biotinylated ␤ 5 I-EGF2ϩ3 protein specifically bound to the MMP-9 C domain in a CRV-peptide-inhibitable manner (Fig. 5D). The ␤ 5 I-EGF2ϩ3 fragment did not bind to MMP-9 CBD, vitronectin, or itself (Fig. 5D) or the C domain of MMP-2 (not shown). The binding was cation-independent (not shown) and could be inhibited with unlabeled ␤ 5 I-EGF2ϩ3. We next mutated the Lys 542 and Tyr 544 residues of the ␤ 5 I-EGF2ϩ3 to alanines to study the importance of the CRV-like sequence. This resulted in a decrease of activity, the K542A and Y544A proteins competing less efficiently for the binding of biotinylated ␤ 5 I-EGF2ϩ3 to the MMP-9 C domain (Fig. 5E). The Y544A mutation also decreased the ability of ␤ 5 I-EGF2ϩ3 to inhibit HT1080 invasion through matrigel (Fig. 5F). ␤ 5 I-EGF2ϩ3, MMP-9 C domain and MMP-2 C domain each inhibited HT1080 invasion with a similar potency, whereas GST had no effect.

FIG. 3. MMP-9 domain-specific inhibition of cell migration and invasion.
A, HT1080 fibrosarcoma invasion through matrigel-coated invasion chambers in the presence or absence of the peptides. All samples were assayed in triplicate in three independent experiments. B, transwells were coated with LLG-C4-GST or GST as a control. THP-1 cells were allowed to migrate overnight in the presence of peptides. C, gelatinolysis of HT1080 cells after a 48-h incubation with the peptides in the presence or absence of 20 nM PDBu. Data are means Ϯ S.E. from six samples. D, pro-MMP-9 binding to the ␣ M integrin I domain in the presence or absence of peptides. Bound MMP-9 was detected with a monoclonal anti-MMP-9 antibody. Statistically significant differences in t test are indicated with asterisks. *, p Ͻ 0.05; **, p Ͻ 0.001. E, binding of gelatin to pro-MMP-9⅐integrin complex. Biotinylated gelatin was detected with streptavidin peroxidase. Activation of MMP-9 in HT1080 (F) and THP-1 (G) cells is shown. The cells were incubated in serum-free medium in the presence of phorbol ester to stimulate MMP-9 expression. Plasminogen (2.5 g/ml) and pro-MMP-3 (0.5 g/ml) were added to promote MMP-9 activation. The peptides were To find further evidence for the MMP-␤ 5 integrin interaction, we studied the binding of MMP-9 C domain to ␣ V ␤ 5 expressing cells. 125 I-Labeled MMP-9 C domain showed a specific binding to ␤ 5 integrin-transfected, but not to the untransfected CS-1 melanoma cells (Fig. 6A). The binding was competed with unlabeled MMP-9 C domain, the ␤ 5 I-EGF2ϩ3 fragment, and to a lesser extent the ␤ 5 I-EGF2ϩ3 Y544A mutant. No competition was observed with the MMP-2 C domain or the GRGDSP peptide.
To test whether the MMP-9 C domain is able to bind to the CRV-like site of the ␤ 2 integrin, we examined THP-1 cell binding to the immobilized KIM127 antibody. THP-1 cells bound to the KIM127 antibody but not to an anti-His 6 tag antibody (Fig.   5F). The C domain (50 g/ml) inhibited the cell binding by 40%, whereas CBD did not (Fig. 5F). The specificity of the binding is shown by competition with soluble KIM127 but not by a control antibody. The C domain had no effect on THP-1 binding to another ␤ 2 integrin-activating antibody R3F9C (not shown).
Double immunofluorescence stainings of HT1080 cells on vitronectin showed a partial colocalization for uPAR, MMP-9 and ␤ 5 integrin. MMP-9 was concentrated on the leading edge of the cells, where the colocalization with integrin and uPAR are evident (Fig. 7). Only nonspecific nuclear staining was observed with irrelevant control antibodies. Colocalization of uPAR with MMP-9 was also found on the surface of THP-1 cells (not shown).
As a final test of reactivity of the CRV peptide, we assessed its effect on tumor growth in vivo. Mice carrying HSC-3 tongue squamous cell carcinoma xenografts were treated with the peptide when the subcutaneous tumors were in an early phase and not yet visible. CRV, the scrambled peptide, or PBS was injected intravenously five times. At 31 days, a statistically significant inhibition of tumor growth by CRV was observed in comparison with the scrambled peptide or PBS (Fig. 8A). CRV increased the survival of the mice, and after 2 months all five CRV-injected mice were alive, whereas the mice given the scrambled peptide or PBS had been euthanized due to large tumors (Fig. 8B). The effect of CRV could at least partially be accounted for by inhibition of angiogenesis. The CRV-treated mice had a less developed tumor vasculature as shown by immunostaining of the endothelial marker CD31 (Fig. 8C). DISCUSSION We have developed domain-specific peptide probes to the gelatinases and examined molecular interactions important for these enzymes. Each of the domain-specific peptides inhibited cell migration, indicating that the three major domains of MMP-9 (the catalytic domain, CBD, and the C domain) each play a distinct role. We have previously shown that in leukocytes, pro-MMP-9 interacts with the ␣ M and ␣ L integrin I domains through the catalytic domain (12). Here we have found another integrin interaction for MMP-9, where the C domain of MMP-9 binds to the integrin ␤ subunit. In contrast to the I domain interaction, which occurs in the presence of calcium and presumably maintains pro-MMP-9 inactive, the C domain/␤ subunit interaction requires activated integrins and appears to play a dynamic role in mediating MMP-9 activation and pericellular gelatinolysis.
Of the MMP-9 binding peptides identified in this study, the CBD-binding peptide PPC functioned as an exosite inhibitor of MMP-2 and -9 inhibiting gelatin binding and degradation but had no inhibitory effect on the MMP-9 interactions with integrins. We identified a PPC-like sequence in the heparin-binding domain of fibronectin as a CBD recognition site. Vitronectin had a similar but apparently lower affinity binding site for CBD. Since PPC did not bind to the fibronectin type II repeats of fibronectin, it could serve as a lead compound for the development of highly specific gelatinase inhibitors.
The C-terminal domain-binding CRV peptide did not affect the enzymatic activity of MMP-9 but inhibited dimerization of the MMP-9 C domain, activation of the pro-MMP-9 via plasminogen/ MMP-3-dependent pathway, and pericellular gelatinolysis. Several findings indicate that CRV is a mimic of the activation epitope in the integrin ␤ subunit, preferentially the ␤ 5 subunit. The C domain of MMP-9 inhibited leukocyte adhesion to the KIM127 antibody, which recognizes the CRV homologous site in the ␤ 2 integrin. The recombinant ␤ 5 integrin I-EGF2ϩ3 fragment specifically bound to the C domain in a CRV-dependent manner, and the single alanine mutations of the Lys 542 and Tyr 544 residues in the ␤ 5 I-EGF2ϩ3 decreased its activity. The ␤ 5 integrin- transfected cells, but not the untransfected cells, bound the C domain of MMP-9. In HT1080 cells, pro-MMP-9 was co-precipitated with antibodies to ␤ 5 integrins, and the ␤ 5 I-EGF2ϩ3 fragment and the C domain both inhibited invasiveness of this cell line. MMP-9 and ␤ 5 integrins similarly localized to the leading edge of the HT1080 cells. However, we cannot exclude the possibility that the CRV peptide inhibits also other C domainmediated interactions.
We did not observe association of MMP-9 with ␣ V ␤ 3 in HT1080 cells, although a functional linkage between MMP-9 and the active ␣ V ␤ 3 integrin has been found (37). This may reflect the fact that HT1080 cells utilize the ␣ V ␤ 5 integrin for vitronectin adhesion. MMP-9 binding to ␣ V ␤ 5 may be physiologically more relevant, since ␣ V ␤ 5 and MMP-9 expression are under similar transcriptional regulation (8,38). In turn, ␣ V ␤ 3 and MMP-2 appear to be co-regulated (39). In our studies, the CRV peptide only weakly inhibited pro-MMP-2 activation, and the C domains of MMP-2 and MMP-9 did not compete with each other in binding assays.
The finding that CRV mimics an integrin activation epitope provides an explanation for the requirement of ligand-engaged integrins in pro-MMP-9 activation (37,40). We also demonstrate that uPAR, which is required for MMP-9 activation, associates with pro-MMP-9 in HT1080 and THP-1 cells. uPAR was a substrate for MMP-9 in vitro, and the cellular cleavage of uPAR was gelatinase-dependent. Cleavage by MMP-9 resulted in the release of the D1 domain of uPAR, which has also been observed with other MMPs such as MMP-12 (27). Functionally, uPAR cleavage causes loss of uPA binding and the dissociation of uPAR and integrins (41). Thus, MMP-9 not only regulates its own activation but also uPAR function. Interestingly, cooperation of MMP-9 and uPAR has been shown to be essential for the intravasation of tumor cells (42). Also, uPA/uPAR and gelatinases co-exist in transport vesicles in migrating cells (43,44).
Inhibition of tumor growth by CRV suggests an important function for the MMP-9/␣ V ␤ 5 pair in primary tumor growth and/or angiogenesis. However, increased tumor growth rather than inhibition is observed in both the ␤ 3 and ␤ 5 integrin knockout mice (45) and also in mice with low plasma levels of MMP-9 (46). The ability of MMP-9 to generate angiostatin or tumstatin (47) may explain these contradictory findings, and perhaps ␣ V ␤ 5 -bound MMP-9 is also used for angiostatin generation. Furthermore, the cleavage of uPAR by MMP-9 could also inhibit tumor spreading. Since tumor therapies aimed at direct inhibition of MMP activity have not been very successful, the noncatalytic means to inhibit MMPs may be more attractive (6). It is encouraging that our phage display-developed peptides specifically interfere with different integrin-mediated interactions blocking either the MMP catalytic or the C-terminal domain binding, suggesting that specific drugs can be developed that locally prevent gelatinase function but not the enzymatic activity. Supporting this conception, also the ␤ 2 -integrin ligand DDGW peptide, which blocks the ␣ M ␤ 2 integrin⅐pro-MMP-9 complex, is active in vivo, inhibiting neutrophil recruitment in an acute inflammation model in mice (48).
Our model of the MMP-9 interactions with integrins is based on a "peptidoscopic" view obtained with phage display peptides and suggests that pro-MMP-9 can interact with FIG. 5. Identification of the integrin ␤ 5 chain as a binding site for the MMP-9 C domain. A, rabbit antisera against the cytoplasmic domain of integrins were used for immunoprecipitation (IB) followed by Western blotting (WB) with anti-MMP-9 antibodies as in Fig. 4. B, schematic representation of the integrin ␤ chain. The sequence similar to the CRV peptide in individual ␤ chains is shown. The KIM127 antibody epitope in the ␤ 2 integrin is underlined. C, the CRV peptide or the MMP-9 domains do not block HT1080 cell adhesion to fibronectin or vitronectin. D, binding of biotinylated I-EGF2ϩ3 fragment of ␤ 5 integrin to the C domain or CBD of MMP-9, BSA, I-EGF2ϩ3 fragment, or vitronectin was assessed in the presence or absence of peptides or unlabeled EGF2 ϩ 3. **, p Ͻ 0.001 in Student's t test. E, competition of ␤ 5 I-EGF2ϩ3 fragment binding by the alanine mutants of ␤ 5 EGF2 ϩ 3. F, inhibition of HT1080 invasion through matrigel in the presence or absence of ␤ 5 integrin and MMP domains (50 g/ml). The data are means Ϯ S.D. from four samples. Statistically significant differences in t test are indicated with asterisks *, p Ͻ 0.05; **, p Ͻ 0.001. scr., scrambled; mAb, monoclonal antibody. integrins in two ways. In leukocytes, the interaction between the integrin I domain and the MMP-9 catalytic domain is dominant and apparently keeps pro-MMP-9 in an inactive form. However, our data do not exclude the possibility that both ␣ and ␤ subunit-mediated interactions occur at the same time. Ligand binding activates the integrin and exposes the activation epitope in the ␤ chain, which can act as a docking site for the C domain of MMP-9. MMP-9 may then be activated by proteases or becomes catalytically competent by direct binding to a substrate (49). In integrins that lack an I domain in the ␣ subunit, the MMP-9 C domain-directed interaction may be the dominant interacting site.