Structural analysis of the alpha(2) integrin I domain/procollagenase-1 (matrix metalloproteinase-1) interaction.

Previous studies have established that ligation of keratinocyte alpha(2)beta(1) integrin by type I collagen induces expression of matrix metalloproteinase-1 (MMP-1) and that MMP-1 activity is required for the alpha(2)beta(1) integrin-dependent migration of primary keratinocytes across collagenous matrices. We now present evidence that MMP-1 binds the alpha(2)beta(1) integrin via the I domain of the alpha(2) integrin subunit. Using an enzyme-linked immunosorbent assay with purified human MMP-1 and recombinant alpha(2) integrin I domain, we showed that the alpha(2) integrin I domain specifically bound in a divalent cation-dependent manner to both the pro and active forms of MMP-1, but not to MMP-3 or MMP-13. Although both the I domain and MMP-1 bind divalent cations, MMP-1 bound, in a divalent cation-dependent manner, to alpha(2) integrin I domains containing metal ion-dependent adhesion sites motif mutations that prevent divalent cation binding to the I domain, demonstrating that the metal ion dependence is a function of MMP-1. Using a series of MMP-1-MMP-3 and MMP-1-MMP-13 chimeras, we determined that both the linker domain and the hemopexin-like domain of MMP-1 were required for optimal binding to the I domain. The alpha(2) integrin/MMP-1 interaction described here extends an emerging paradigm in matrix biology involving anchoring of proteinases to the cell surface to regulate their biological activities.

The extracellular matrix is not a static environment. Remodeling and degradation of the extracellular matrix is a vital component of physiological and pathophysiological processes, such as development and differentiation, cell migration, angiogenesis, wound healing, and metastasis. Matrix metalloproteinases (MMPs) 1 play a central role in the turnover of extracellular matrix components (1).
MMPs constitute a large family of metal-dependent endoproteases with varying substrate specificities for many extracellular proteins. The structure of native triple helical type I collagen makes it resistant to proteolysis, and only six MMPs, MMP-1, MMP-8, MMP-13, MMP-14 (MT1-MMP), MMP-18, and MMP-2, exhibit an ability to cleave native fibrillar collagen within its triple helical domain (2)(3)(4)(5)(6)(7)(8). Similar to most MMPs, the collagenases (MMP-1, MMP-8, and MMP-13) have several structural features in common, including an N-terminal prodomain, a catalytic domain, and a short proline-rich linker connected to a hemopexin-like domain at the C terminus (9). The catalytic domain contains a Zn 2ϩ -binding site that is conserved in all MMPs and is required for catalytic activity (10,11). The catalytic domain of the collagenases contains an additional structural Zn 2ϩ , as well as three structural Ca 2ϩ ions (12). The hemopexin-like domain contains a Ca 2ϩ and a Ca 2ϩ -Cl Ϫ ion pair (12).
The three collagenases differ in patterns of tissue expression. In humans, MMP-1, which is expressed by epithelium, endothelium, fibroblasts, chondrocytes, and macrophages, seems to be the enzyme principally responsible for collagen turnover in most tissues (13)(14)(15)(16)(17)(18). During cutaneous wound healing, human keratinocytes express MMP-1 while migrating over the type I collagen-rich dermis. This spatially and temporally confined expression of MMP-1 is induced by ligation of the ␣ 2 ␤ 1 integrin by dermal type I collagen (19). Both migration and MMP-1 expression induced by type I collagen are inhibited by function blocking antibodies against the ␣ 2 ␤ 1 integrin, but not by antibodies directed against the ␣ 1 ␤ 1 or ␣ 3 ␤ 1 integrins (19). Previous studies have established the role of the ␣ 2 ␤ 1 integrin as a receptor for type I collagen, as well as for type IV collagen, laminins, echovirus 1 and 8, and rotaviruses (20 -22).
Multiple lines of evidence have revealed that the I domain of the ␣ 2 integrin subunit mediates the binding of the ␣ 2 ␤ 1 integrin to its ligands (20). For example, the ␣ 2 ␤ 1 integrin-binding site for collagen and the epitopes recognized by antibodies that block type I collagen binding to the integrin map to the I domain of the ␣ 2 integrin subunit (20). In addition, recombinant ␣ 2 integrin I domain specifically binds all known ligands of the ␣ 2 ␤ 1 integrin, including collagens, laminin, the C-terminal propeptide of type I collagen, and echovirus 1 and 8 (20,22). In the accompanying paper (23), we demonstrated that, in addition to inducing expression of MMP-1, the ␣ 2 ␤ 1 integrin also directly binds MMP-1. In this paper, we show that MMP-1 binds to the I domain of the ␣ 2 integrin subunit, and we define the structural basis of the interaction.

EXPERIMENTAL PROCEDURES
Cloning and Expression of the ␣ 2 Integrin I Domain-The cloning and expression of the ␣ 2 integrin subunit I domain has been described elsewhere (24). Briefly, the cDNA of full-length ␣ 2 integrin was used as a template in the polymerase chain reaction to produce a fragment that encodes Ser 124 through Met 349 of the published ␣ 2 sequence (25). The PCR primers were designed such that the amplification product would contain a BglII site at the 5Ј-end and a XhoI site at the 3Ј-end. The products were digested, purified by agarose gel electrophoresis, and ligated into a BglII-XhoI-digested pGEX-5x-1 vector (Amersham Pharmacia Biotech), which is a glutathione S-transferase fusion protein expression vector. The cDNA of the full-length ␣ 1 integrin (Dr. Eugene E. Marcantonio, Columbia University) was used as a template to amplify the ␣ 1 I domain. As with the ␣ 2 construct, this cDNA encodes for Ser 124 through Met 349 of the published ␣ 1 sequence (26). Like the ␣ 2 integrin I domain, the PCR primers were designed with a 5Ј BglII site and a 3Ј XhoI site, and the resulting product was cloned into the pGEX-5x-1 vector.
Expression and Purification of the GST-I Domain Fusion Proteins-The purification of the ␣ 1 and ␣ 2 integrin I domains have been described elsewhere (24). Briefly, DH5␣ Escherichia coli containing the appropriate plasmid were grown at 37°C in 500 ml of 2ϫ yeast extract/tryptone buffer until A 550 reached 0.3-0.4. The cultures were induced with 1 mM isopropyl-␤-D-thiogalactopyranoside and returned to the incubator for 3 h. The cells were recovered by centrifugation at 2,600 ϫ g for 10 min, washed twice with 10 ml of ice-cold phosphate-buffered saline, and stored at Ϫ70°C until needed. The GST-I fusion proteins were expressed, purified, and characterized exactly as recently described (24).
Construction of the MIDAS Motif Mutant-The glutamate at position 254 or 151 of the ␣ 2 integrin I domain sequence was replaced with an alanine in a single mutagenesis reaction using pBluescript/␣ 2 I. The 966-base pair BST BI-XhoI fragment of pBluescript and the 4669-base pair XhoI-BST BI fragment of pGEX-5x-1/␣ 2 I were isolated and ligated to create pGEX-5x-1/D151A or D254A. The sequences of the cDNAs used in this study, including all of the chimeras, were determined using the BigDye terminator cycle sequencing method (PerkinElmer Applied Biosystems, Foster City, CA) and were compared with the published ␣ 2 sequenced (25,26).
Purification and Activation of MMP-1-Human pro-MMP-1 was purified by two-step chromatography from conditioned medium of interleukin-1-treated dermal fibroblasts cultured as described (27). To isolate prodomain and activated MMP-1, autolysis of pro-MMP-1 was induced with 1 mM 4-aminophenylmercuric acetate (Sigma) for 1 h at room temperature, and the two fragments were separated by size selection dialysis in 0.5 ml of 0.05 M Tris-HCl, 0.01 M CaCl 2 using a 10,000 Molecular Weight Cut-Off Slide-A-Lyzer (Pierce).
Construction of MMP Chimeras-The construction of these chimeras has been described elsewhere (28). Briefly, cDNAs for MMP-1, MMP-3, or MMP-13 cDNAs in the pET3a (Novagen, Madison, WI) vector were used as templates for overlapping PCR. To generate the N-terminal portion of the chimera, PCR was performed with the sense T7 promoter primer and an antisense chimeric junction primer containing the N-terminal MMP sequence at the 5Ј-end and a portion of the C-terminal MMP sequence at the 3Ј-end. To construct the C-terminal portion of the chimera, PCR was performed with an antisense primer containing plasmid sequence and a sense chimeric junction primer containing C-terminal MMP sequence at the 5Ј-end and N-terminal MMP sequence at the 3Ј-end; this sequence overlapped with the previous chimeric junction primer. These two fragments were then used as template in a final PCR reaction, along with the sense T7 promoter primer and the antisense pET3a primer to construct the full-length chimera. The PCR product was cut with NdeI and BamHI, cloned into pET3a vector, and sequenced at the Biotech Facility at the University of Kansas Medical Center. The MMP chimeras were expressed and purified as described previously. Briefly, the plasmid was transformed into E. coli BL21(DE3) competent cells, and protein production was induced upon addition of 0.4 mM isopropyl-␤-D-thiogalactopyranoside. The proteins were resuspended from inclusion bodies in 8 M urea and purified over a High Q Support Anion Support column (Bio-Rad). The proteins were refolded in dialysis as described previously (28), and folded proteins were further purified by chromatography on Green A Dyematrex column (Amicon, Beverly, MA).
Enzyme-linked Immunosorbent Assay-The wells of a 96-well microtiter plate (Immulon 2, Dynex Technologies, Chantilly, VA) were coated overnight at 4°C with 0.1 ml/well of 30 g/ml type I collagen from calf skin (Sigma) or BSA or 10 g/ml pro-MMP-1, activated MMP-1, propeptide from MMP-1, or chimeric MMP. The wells were then washed twice with 0.15 ml of Tris-buffered saline (50 mM Tris-HCl, 150 mM NaCl, pH 7.4) and then blocked for 1 h at room temperature with 0.15 ml of 300 g/ml BSA in Tris-buffered saline. Purified recombinant I domain proteins were diluted to the desired concentration in wash buffer (Trisbuffered saline containing 0.05% Tween 20, 30 g/ml BSA, and 2 mM MnCl 2 , MgCl 2 , EDTA, or CaCl 2 ). The wells were washed once with 0.15 ml of the appropriate wash buffer, and then 0.1 ml of each recombinant protein was added and allowed to interact with the collagen or MMP substrate at room temperature for 1.5 h. The wells were then washed with 0.15 ml of the appropriate wash buffer, and 0.1 ml of a 1:8000 dilution of anti-GST antiserum in the appropriate wash buffer was added for 1 h at room temperature. Again the wells were washed three times, and then a 1:20,000 dilution of pig anti-goat IgG secondary antibody-horseradish peroxidase conjugate (Roche Molecular Biochemicals) in the appropriate wash buffer was added per well for 1 h at room temperature. The wells were again washed three times, and 0.1 ml of tetramethylbenzidine dihydrochloride, prepared according to the manufacturer's instructions, was added per well. After 10 min of substrate conversion, the reactions were stopped with 0.025 ml of 4 N H 2 SO 4 , and the plates were read at 450 nm.
Co-immunoprecipitation-Human monocytes were isolated from blood and differentiated in culture to macrophages as described (27). Cell lysates were prepared as described (23). The ␣ 1 integrin subunit was immunoprecipitated using a monoclonal antibody from Chemicon International Inc. (Temecula, CA). MMP-1 was immunoprecipitated from the cell lysate using MAB1346 (Chemicon). The products were resolved through a 7.5% polyacrylamide gels and transferred to membranes. Immunoprecipitated and co-immunoprecipitated MMP-1 was detected with a rabbit polyclonal antibody (19) as described (23).

RESULTS
In our accompanying paper (23), we used co-immunoprecipitation and platelet adhesion to demonstrate that MMP-1 binds to the ␣ 2 ␤ 1 integrin. The 226-amino acid I domain of the ␣ 2 subunit mediates binding of all identified ligands of the ␣ 2 ␤ 1 integrin. As a GST fusion protein, recombinant I domain provides a reliable reporter in a solid phase binding assay. Wells were coated with collagen or with pro or activated forms of MMP-1, MMP-3, or MMP-9, enzymes that are expressed by keratinocytes in wounded skin (13,29). In the presence of Mn 2ϩ , which shifts ␣ 2 ␤ 1 to a more active state (20), soluble I domain bound to both pro and active MMP-1 (Fig. 1). The level of binding to MMP-1 was equivalent to that detected for collagen, used as a positive control. Of note, we found that the I domain still bound appreciable levels of pro and active MMP-1 in the presence of EDTA, which abolished binding to collagen (Fig. 1).
Next, we sought to establish whether the pro or active forms of MMP-1 were ligands for the ␣ 2 integrin I domain and whether the prodomain bound directly. Fig. 2A demonstrates that the ␣ 2 integrin I domain binds to both pro-MMP-1 and activated MMP-1 in a concentration-dependent manner. No binding to BSA was observed. To determine whether the slight difference observed between the binding of ␣ 2 integrin I domain to pro-MMP-1 and active MMP-1 was a result of a binding interaction between the I domain and the propeptide, we also investigated the binding of I domain to the propeptide of MMP-1. Although activated MMP-1, which lacks the propep- tide, bound to of the ␣ 2 integrin I domain, the purified propeptide failed to bind (Fig. 2B). Thus the propeptide of MMP-1 is not directly involved in the interaction between the ␣ 2 integrin I domain and pro-MMP-1. This conclusion is further supported by the data from experiments with MMP chimeras shown in Fig. 6. The small difference observed between the binding of pro-MMP-1 and activated MMP-1 is most likely due to differences in coating efficiency, because the difference is in the amount of ␣ 2 integrin I domain bound at saturation. Additionally, the binding data were best described by a single binding site with half-maximal binding for both active and proMMP-1 occurring between 10 and 40 nM, depending upon the experiment. There was no significant difference between the halfmaximal binding of ␣ 2 I domain to pro or active MMP-1. These data indicate that both the pro and active forms of MMP-1 are ligands for the ␣ 2 integrin I domain, but, as suggested in our accompanying paper (23), the physiologic ligand may be limited to pro-MMP-1.
To further characterize MMP-1 binding to the ␣ 2 integrin I domain, we studied the divalent cation dependence of this interaction. The different ligands for the ␣ 2 ␤ 1 integrin differ in their dependence upon divalent cations for binding (20). Both Mn 2ϩ and Mg 2ϩ support I domain binding to collagen, whereas Ca 2ϩ does not (20). Predictably, EDTA completely inhibits the binding of collagens to the I domain (20). On the other hand, binding of echovirus 1 to the ␣ 2 integrin I domain is divalent cation-independent (30). The metal dependence of I domain interaction with ligands is mediated by the divalent cationbinding site called the MIDAS motif (31). The catalytic and hemopexin domains of MMP-1 also bind divalent cations (12). Because both MMP-1 and the ␣ 2 integrin I domain contain divalent cation-binding sites, any divalent cation dependence of MMP-1 binding to the ␣ 2 integrin I domain could arise from either or both the MMP and the I domain. Fig. 3 demonstrates that both 2 mM Mn 2ϩ and 2 mM Mg 2ϩ supported binding of ␣ 2 integrin I domain to pro-MMP-1 and to collagen. As shown previously (24), 2 mM EDTA abolishes the binding of the ␣ 2 integrin I domain to type I collagen; however, a significant component (about 50 -60%) of the binding of ␣ 2 integrin I domain to pro-MMP-1 was retained in 2 mM EDTA, indicating a fundamental difference in the binding of the ␣ 2 integrin I domain to MMP-1 as compared with collagen type I. These results suggest that there are two components to the ␣ 2 integrin I domain MMP-1 interaction: a metal-dependent component and a metal-independent component. As expected, 2 mM Ca 2ϩ did not support binding of the ␣ 2 integrin I domain to collagen. In contrast, 2 mM Ca 2ϩ effectively supported binding of the ␣ 2 integrin I domain to pro-MMP-1. Because Ca 2ϩ does not bind to the I domain MIDAS motif (24), the metal enhancement of I domain binding to MMP-1 observed in the presence of Ca 2ϩ and the significant difference in metal cation dependence between the ␣ 2 integrin I domain-collagen interaction and the ␣ 2 integrin I domain/MMP-1 interaction suggest that the metal dependence of the MMP-1/␣ 2 integrin I domain interaction may arise from the MMP-1 rather than from the I domain. These results indicate that there are critical differences between the binding of ␣ 2 integrin I domain to collagen as opposed to pro-MMP-1.
To investigate further the metal dependence of the MMP-1/␣ 2 integrin I domain interaction, we utilized I domains with either of two point mutations in the MIDAS motif of the ␣ 2 integrin I domain. D256A and D151A both result in loss of essential metal coordinating side chains within the MIDAS motif of the ␣ 2 integrin I domain (31). As expected, these mutants do not bind collagen in either the presence or absence of divalent cations (Fig. 4A). However, the mutant I domains retained the ability to bind to pro-MMP-1 in the presence of either Ca 2ϩ or Mn 2ϩ . Divalent cation-dependent binding of the MIDAS motif mutants to pro-MMP-1 was comparable with that of wild-type I domain, even though the divalent cation chelation site of the I domain had been disrupted (Fig. 4B). Both the wild-type and mutant ␣ 2 integrin I domains bound to pro-MMP-1 to a lesser extent in the presence of EDTA, again indicating both a metal-dependent, as well as a metal-independent, component of binding. These results further demonstrate that the divalent cation requirement for the optimal binding of MMP-1 by the ␣ 2 integrin I domain derives from the MMP and not the I domain.
Like the ␣ 2 ␤ 1 integrin, the ␣ 1 ␤ 1 integrin is a cell surface collagen receptor that contains a structurally similar I domain (20). As with the ␣ 2 I domain, recombinantly expressed ␣ 1 I domain recapitulates the binding properties of the full-length ␣ 1 ␤ 1 integrin (20). Therefore, we also tested the binding of the ␣ 1 integrin I domain to pro-MMP-1. As observed with the ␣ 2 integrin I domain, Ca 2ϩ and Mn 2ϩ supported binding of pro-MMP-1 by the ␣ 1 I domain. In 2 mM EDTA, both I domains showed a significant degree of metal-independent binding to pro-MMP-1, again suggesting the presence of both metal-dependent and metal-independent interactions (Fig. 5A). The binding of pro-MMP-1 to the ␣ 1 ␤ 1 integrin was confirmed in intact cells by co-immunoprecipitation experiments. Monocytic U937 cells were differentiated to macrophages in culture, and then MMP-1 and ␣ 1 were immunoprecipitated from cell lysates. Immunoprecipitation with both anti-␣ 1 integrin and anti-MMP-1 antibodies precipitated nearly equivalent amounts of MMP-1 derived from the cell layer (Fig. 5B), suggesting that most of the MMP-1 was bound to the cell surface via ␣ 1 ␤ 1 .
As described above, pro-MMP-1 contains an N-terminal propeptide, a catalytic domain, and a short proline-rich linker connected to a hemopexin-like domain (12). Although the data shown in Fig. 2 suggest that the propeptide sequence of MMP-1 does not mediate the binding by the ␣ 2 integrin I domain, we were interested in determining in greater detail which domain(s) of MMP-1 are required for binding to ␣ 2 integrin I domain. To address this question, we used MMP chimeras in which domains of MMP-1 were replaced with the equivalent domains of either MMP-3 or MMP-13, neither of which binds the ␣ 2 integrin I domain (Fig. 6B). To name the chimeras, the principal domains (namely, the propeptide and catalytic domain, the linker, and the hemopexin-like domain) were assigned a number corresponding to the MMP from which they originated. Thus, pro-MMP-1 was designated 1-1-1, and proMMP-3 was 3-3-3 (Fig. 6A). Consequently, 3-1-1 was composed of the propeptide and catalytic domain from MMP-3, and the linker and hemopexin-like domain of MMP-1 (Fig. 6A).

integrin I domain MIDAS motif mutants bind to pro-MMP-1 but not to collagen.
A, the binding of the ␣ 2 integrin I domain (12.5 nM) and I domain mutants (12.5 nM) to type I collagen was measured in a solid phase binding assay. The microtiter plates were coated with collagen at a concentration of 30 g/ml. Binding was determined in buffers containing 2 mM Mg 2ϩ or 2 mM EDTA. B, the binding of the ␣ 2 integrin I domain (12.5 nM) and I domain mutants (12.5 nM) to pro-MMP-1, type I collagen, and BSA (control) was measured in a solid phase binding assay. The microtiter plates were coated with pro-MMP-1 at a concentration of 10 g/ml or collagen or BSA at a concentration of 30 g/ml. Binding was determined in buffers containing 2 mM of the indicated divalent cation or 2 mM EDTA. required for interaction of MMP-1 with ␣ 2 integrin I domain. This conclusion agrees with the results shown in Fig. 2. 3-3-1 and 13-13-1 bound less effectively to the ␣ 2 integrin I domain than did 1-1-1, indicating that the hemopexin-like domain of MMP-1 alone was not sufficient to support optimal binding of the ␣ 2 integrin I domain. Similarly, 1-1-13 did not support optimal binding of the ␣ 2 integrin I domain, indicating that the linker alone was also insufficient for optimal binding. Thus, both the linker and hemopexin-like domains were necessary to reconstitute efficient binding of ␣ 2 integrin I domain to MMP-1.
However, the degree of binding of 1-1-13, 13-13-1, and 3-3-1 were intermediate between the positive control (1-1-1) and the negative control (3-3-3 or 13-13-13). Thus, the presence of either the linker or the hemopexin-like domain of MMP-1 conferred readily detectable but suboptimal binding of the ␣ 2 integrin I domain. These findings suggest that two contact points are formed between the I domain and MMP-1: one with the linker domain and one with the hemopexin-like domain.

DISCUSSION
In this paper, we demonstrate that MMP-1, in both its active and pro forms, binds specifically to both the ␣ 1 and ␣ 2 integrin I domains. We also demonstrate that the interaction between the ␣ 2 integrin I domain and MMP-1 has both a divalent cation-dependent component and a divalent cation-independent component. In addition to Mn 2ϩ and Mg 2ϩ , Ca 2ϩ also supported binding of the ␣ 2 integrin I domain to MMP-1. However, because Ca 2ϩ has not been shown to support binding of the ␣ 2 integrin I domain to any other ligand (20), the metal dependence seemed to be a requirement of MMP-1, not of the I domain. This conclusion was confirmed with the use of ␣ 2 integrin I domain MIDAS motif mutants. As expected, these mutants failed to bind to collagen, but they did bind to MMP-1 in a divalent cation-dependent manner. These results indicate FIG. 5. The ␣ 1 integrin I domain also binds to pro-MMP-1. A, the binding of the ␣ 2 integrin I domain (12.5 nM) and the ␣ 1 integrin I domain to pro-MMP-1, type I collagen, and BSA (control) was measured in a solid phase binding assay. The microtiter plates were coated with pro-MMP-1 at a concentration of 10 g/ml or collagen or BSA at a concentration of 30 g/ml. Binding was determined in buffers containing 2 mM of the indicated divalent cation or 2 mM EDTA. B, human monocytes were isolated from blood and differentiated in culture to macrophages as described under "Experimental Procedures." The cell lysates were prepared as described under "Experimental Procedures." The antibodies used for immunoprecipitation are indicated above each lane; the ␣ 1 integrin subunit and MMP-1 were immunoprecipitated with monoclonal antibodies. In all lanes, immunoprecipitated and coimmunoprecipitated MMP-1 was detected with a rabbit polyclonal antibody as described under "Experimental Procedures." The migration of molecular mass standards is shown on the right (in kDa). The two bands represent the different glycosylated forms of pro-MMP-1.
FIG. 6. Binding of ␣ 2 integrin I domain to MMP chimera constructs. A, schematic representation of chimerical proteins. Each domain (propeptide-catalytic, linker, or hemopexin-like domain) is given a number corresponding to the MMP from which it came. In the chimeras, entire domains are replaced with corresponding domains from either MMP-3 or MMP-13. B, the binding of the ␣ 2 integrin I domain (12.5 nM) to chimeras 3-1-1, 3-3-1, 1-1-13, and 13-13-1, as well as controls pro-MMP-1, proMMP-3, pro-MMP-13, and BSA was measured in a solid phase binding assay. The microtiter plates were coated with one of the matrix metalloproteinases or chimeras or BSA at a concentration of 10 g/ml. Binding was determined in 2 mM Mn 2ϩ . that the metal dependence of the ␣ 2 /MMP interaction is conferred by the proteinase, whereas the metal dependence of the ␣ 2 -collagen binding is conferred by the integrin.
We used chimeric MMPs to begin to dissect the structural basis of the interaction between MMP-1 and the ␣ 2 integrin I domain. Although MMP-3 and MMP-13 have significant structural homology to MMP-1 (9), neither enzyme bound to the ␣ 2 integrin I domain. Thus, they provide suitable partners for domain swaps in chimera construction. These chimeras demonstrated that neither the propeptide nor the catalytic domains of MMP-1 are required for binding to the integrin I domains but that both the linker and hemopexin-like domain must be present for optimal binding. The presence of either the linker domain or the hemopexin-like domain of MMP-1 in a chimera is sufficient to confer detectable, albeit suboptimal binding. These findings suggest that there are at least two points of contact between the ␣ 2 integrin I domain and MMP-1, one with the linker and one with the hemopexin domain. It is interesting to note that the divalent cation experiments described above suggest that the interaction between MMP-1 and the ␣ 2 integrin I domain has both cation-dependent and cation-independent components. Perhaps the cation-dependent component of the interaction emanates from the hemopexin-like domain, which in the crystal structure has been shown to bind Ca 2ϩ , whereas the divalent cation-independent component of the interaction is derived from the linker domain, which does not bind any divalent cation in the crystal structure (12).
Localization of matrix metalloproteinases and related enzymes to the cell surface is not without precedent. Proteinases localized to the cell surface include MMP-2 by TIMP-2 (32, 33), MMP-2 by ␣ v ␤ 3 integrin (34,35), seprase by ␣ 3 ␤ 1 integrin (36), MMP-9 by CD44 (37,38), and MMP-7 by heparan sulfate proteoglycans (39). In this work, we add to the growing list of cell surface-associated proteinases by demonstrating that MMP-1 binds to the ␣ 2 integrin I domain, and our accompanying paper (23) demonstrates that pro-MMP-1 is bound to the cell surface by interaction with the ␣ 2 integrin. Anchoring proteinases to the cell membrane may afford the cell precise control over both proteinase activity and location. Binding of pro-MMP-1 to a collagen receptor could result in an enzyme that is properly positioned for cleavage of its substrate. In addition, binding of pro-MMP-1 to the ␣ 2 ␤ 1 integrin could result in activation of the enzyme, either by a second proteinase or by a conformational change of MMP-1 itself.
In addition to regulating enzyme activity, binding of MMP-1 to the ␣ 2 ␤ 1 integrin may allow the cells to carefully control proteinase expression levels in their microenvironment. Ligation of the ␣ 2 ␤ 1 integrin by collagen induces expression of pro-MMP-1 in migrating keratinocytes (19). When MMP-1 cleaves type I collagen at physiological temperature, the triple helical molecule denatures. Denatured collagen is no longer a high affinity ligand for the ␣ 2 ␤ 1 integrin; consequently, migration and signaling are turned off (40,41). Thus cells have a mechanism to coordinate and integrate the expression of pro-MMP-1 and the ␣ 2 ␤ 1 integrin, the amount of type I collagen in the matrix, and the activity of MMP-1. Because the enzyme is surface-bound, diffusion of the active enzyme within the extracellular matrix may be low, facilitating focal cleavage of collagen, as well as allowing each cell to sense its microenvironment and adjust its MMP-1 levels independently of its neighbors. Expression of MMP-1 may also be sensitive to ␣ 2 ␤ 1 integrin levels, in terms both of ligated integrin available to signal and of integrin available to orient/activate the enzyme.
Functionally, localization of MMP-1 to the cell surface may have important effects upon cell migration. The ␣ 2 ␤ 1 integrin cannot bind to MMP-1-cleaved collagen fragments, which would slow migration across a modified surface (40). In wound healing, keratinocytes require activity of MMP-1 for migration across a collagen surface (19). It was suggested that MMP-1 may be required for direction finding to ensure that the cells progress across to the wound face toward full-length collagen (19). Reverse migration to the wound edge is prevented by turning a pro-migratory substrate into a poor migratory substrate (19). Interestingly, a recent report suggested that MMP-1 bound to the surface of smooth muscle cells is important for stimulating migration by enhancing rear release of the integrin from the degraded collagen (42). Although this work did not determine whether the MMP-1 was bound to the ␣ 2 ␤ 1 or ␣ 1 ␤ 1 integrins, the finding is entirely consistent with the model we have presented in this report. Indeed, it has been shown that an intermediate level of integrin expression is required for optimal migration (43). If levels are too low, the cell cannot gain traction; if levels are too high, presumably the cell becomes too firmly attached to the matrix. Similar considerations apply to ligand density within an adhesive substrate. Perhaps MMP-1 associates with the ␣ 2 ␤ 1 integrin to carefully regulate the concentration of ligand available for the integrin and thus regulate its avidity.
There are interesting parallels in our structure-function analysis and the studies carried out on the MMP-2/␣ v ␤ 3 integrin interaction (34,35). Specifically, both studies have highlighted the hemopexin-like domain as a point of interaction between the integrin and the MMP. In addition, we found that the linker domain in between the catalytic domain and the hemopexin-like domain was also important for binding to the ␣ 2 integrin I domain. At least a portion of the linker domain may also be important for interaction of MMP-2 with ␣ v ␤ 3 ; the hemopexin-like domain used in those experiments was generated by MMP-2 autocatalysis and may contain a piece of the linker in addition to the hemopexin-like domain. It is interesting to speculate that the hemopexin domain of other MMPs may also be important for their localization to the cell surface; mediated, perhaps, by binding other integrins.
In summary, we have shown that MMP-1 interacts with the ␣ 2 ␤ 1 integrin via the I domain of the ␣ 2 integrin subunit. We demonstrated that both the linker and hemopexin-like domains were required for binding but that the propeptide and catalytic domains were not. Our results indicate that there are two separate contact points between the I domain and MMP-1: one on the linker and one on the hemopexin domain. We also showed that the MMP-1/␣ 2 integrin I domain interaction had both a metal-dependent and a metal-independent component, and through the use of I domain mutants, we showed that the metal dependence was a function of the MMP.