Structure and domain-domain interactions of the gelatin binding site of human 72-kilodalton type IV collagenase (gelatinase A, matrix metalloproteinase 2).

We have shown previously that all three fibronectin type II modules of gelatinase A contribute to its gelatin affinity. In the present investigation we have studied the structure and module-module interactions of this gelatin-binding domain by circular dichroism spectroscopy and differential scanning calorimetry. Comparison of the T values of the thermal transitions of isolated type II modules with those of bimodular or trimodular proteins has shown that the second type II module is significantly more stable in the trimodular protein coll 123 (T = 54°C) than in the single-module protein coll 2 (T = 44°C) or in the bimodular proteins coll 23 (T = 47°C) and coll 12 (T = 48°C). Analysis of the enthalpy changes associated with thermal unfolding of the second type II module suggests that it is stabilized by domain-domain interactions in coll 123. We propose that intimate contacts exist between the three tandem type II units and they form a single gelatin-binding site. Based on the three-dimensional structures of homologous metalloproteases and type II modules, a model is proposed in which the three type II units form an extension of the substrate binding cleft of gelatinase A.

We have shown previously that all three fibronectin type II modules of gelatinase A contribute to its gelatin affinity. In the present investigation we have studied the structure and module-module interactions of this gelatin-binding domain by circular dichroism spectroscopy and differential scanning calorimetry. Comparison of the T m values of the thermal transitions of isolated type II modules with those of bimodular or trimodular proteins has shown that the second type II module is significantly more stable in the trimodular protein coll 123 (T m ‫؍‬ 54°C) than in the single-module protein coll 2 (T m ‫؍‬ 44°C) or in the bimodular proteins coll 23 (T m ‫؍‬ 47°C) and coll 12 (T m ‫؍‬ 48°C). Analysis of the enthalpy changes associated with thermal unfolding of the second type II module suggests that it is stabilized by domain-domain interactions in coll 123. We propose that intimate contacts exist between the three tandem type II units and they form a single gelatin-binding site. Based on the three-dimensional structures of homologous metalloproteases and type II modules, a model is proposed in which the three type II units form an extension of the substrate binding cleft of gelatinase A.
Proteolytic degradation of constituents of the extracellular matrix and basement membranes plays an important role in tissue restructuring processes such as those accompanying cell migration, morphogenesis, wound healing, angiogenesis, and tumor invasion (Liotta et al., 1991). The proteinases implicated in tumor invasion include components of the urokinase receptor/urokinase/plasminogen system and several members of the metalloproteinase family (He et al., 1989;Matrisian, 1992;Vassalli and Pepper, 1994). Enzymes capable of degrading type IV collagen are crucial for tumor metastasis as this collagen is a major component of basement membranes which have to be penetrated during migration of tumor cells. Significantly, secretion of type IV collagenases is well correlated with metastasis and transformation (Librach et al., 1991;Stetler-Stevenson et al., 1993).
Studies on the primary structures of 72-kDa and 92-kDa type IV collagenases (gelatinase A and B) have revealed that they contain a catalytic domain, a hemopexin-like domain, and three tandem homology units closely related to the type II domains of fibronectin (Collier et al., 1988;Wilhelm et al., 1989).
The hemopexin-like domain of gelatinase A has been shown to be involved in modulation of its activity by TIMPs, the tissue inhibitors of metalloproteases (Murphy et al., 1992;Fridman et al., 1992). This region is also required for the activation of progelatinase A by a membrane-associated activator, suggesting that it may interact with some component of the cell membrane (Murphy et al., 1992;Vassalli and Pepper, 1994).
Studies on the substrate specificity of gelatinases have shown that they are able to degrade type IV, type V, type VII, and type X collagens, fibronectin, elastin, and all types of denatured collagens (Murphy et al., 1991;Senior et al., 1991;Matrisian, 1992).
Gelatinases are unique among metalloproteinases in having pronounced affinity for denatured collagens. Recently it was shown that the gelatin-binding sites of gelatinases reside in their fibronectin-related regions. Recombinant proteins corresponding to the fibronectin-related regions of gelatinase A and B were found to have high affinity for gelatin (Bá nyai and Patthy, 1991;Collier et al., 1992;Bá nyai et al., 1994). Conversely, recombinant gelatinase A lacking the fibronectin type II units was shown to be devoid of affinity for gelatin or type I and type IV collagens, indicating that the fibronectin-like domain is the sole site of collagen binding (Murphy et al., 1994;Allan et al., 1995).
The functional significance of the gelatin-binding site is suggested by the observation that although deletion of this domain does not affect the catalytic properties of gelatinase A on small synthetic substrates (Ye et al., 1995), activity on gelatin is drastically reduced and the cleavage pattern of type IV collagen is altered (Murphy et al., 1994;Ye et al., 1995). On the basis of these observations, it has been proposed that the fibronectinlike domain of gelatinase A specifically orientates the enzyme on type I gelatin or type IV collagen, thus enhancing the rate of substrate cleavage (Murphy et al., 1994).
To clarify the role of the collagen-binding domain, we have initiated studies to define its interaction with substrates. In a previous paper (Bá nyai et al., 1994), we described the expression of type II domains of the fibronectin-related region of gelatinase A in Escherichia coli and the characterization of their type I gelatin-binding properties. We have shown that although each of the three type II domains binds gelatin, proteins containing all three tandem type II domains of gelatinase A have significantly higher affinity than any of the constituent units: all three type II units contribute to gelatin binding. In the present work, we have studied the collagen-binding domain by circular dichroism spectroscopy and differential scanning calorimetry to elucidate the structure and interactions of the three type II modules.

EXPERIMENTAL PROCEDURES
Recombinant Proteins-Recombinant proteins DEL␤galcoll 1, DEL␤galcoll 2, DEL␤galcoll 3, DEL␤galcoll 12, DEL␤galcoll 23, and DEL␤galcoll 123 containing different segments of the collagen-binding site of human gelatinase A and a 37-amino acid-long N-terminal peptide derived from the ␤-galactosidase moiety of the expression vector (Bá nyai et al., 1991) were produced in E. coli. The properly folded, functionally homogeneous recombinant DEL␤galcoll proteins were isolated by binding to gelatin-Sepharose 4B columns and elution with a 0 -8 M urea gradient as described previously (Bá nyai et al., 1994). We have shown previously that the ␤-galactosidase moiety may be removed by limited elastase digestion, whereas the regions containing the properly folded type II domains are resistant to elastase (Bá nyai et al., 1994). In the present work, the N-terminal peptides of the DEL␤galcoll fusion proteins were removed as follows: the DEL␤galcoll proteins (1 mg/ml) were dissolved in 0.1 M ammonium bicarbonate, pH 8.0 buffer and were incubated with 0.01 mg/ml elastase (Serva) at 25°C for 60 min. Reaction was arrested with 2 mM phenylmethylsulfonyl fluoride (Serva), the digest was applied to a gelatin-Sepharose 4B column, and the bound protein was eluted with a urea gradient. Protein eluted at the same urea concentration as the starting material was pooled, dialyzed and lyophilized. SDS-polyacrylamide gel electrophoresis analysis of the elastase-digested recombinant proteins (coll proteins) indicated a 5-kDa reduction in molecular mass compared with that of DEL␤galcoll proteins (data not shown). Sequence analyses have shown that elastase cleavage occurred at Ala-34 of the ␤-galactosidase region of the DEL␤gal fusion proteins.
Reduced-alkylated DEL␤galcoll 123 was prepared as follows: DEL␤galcoll 123 (0.025 mM) was dissolved in 0.1 M Tris-HCl, 6 M guanidinium HCl, 5 mM EDTA, 50 mM dithiothreitol, pH 8.0, and the solution was incubated at 25°C for 30 min. The reduced-denatured protein was alkylated with iodoacetic acid (110 mM). The alkylated protein was desalted by gel filtration on a Sephadex G-25 column equilibrated with 0.1 M ammonium bicarbonate, pH 8.0, and the protein was lyophilized. For some experiments, reduction-alkylation of DEL␤galcoll 123 was carried out in a similar way, except that guanidinium HCl was omitted from the reaction mixture.
The concentration of the recombinant proteins was determined spectrophotometrically using extinction coefficients determined according to a described method (Mach et al., 1992). Protein samples were analyzed by SDS-polyacrylamide gel electrophoresis using 11-22% or 16 -22% linear polyacrylamide gradient slab gels (Laemmli, 1970). The gels were stained with Coomassie Brilliant Blue G-250.
Circular Dichroism Spectroscopy-CD spectra were measured over the range of 190 -250 nm by using a JASCO J-720 spectropolarimeter thermostatted with a Neslab RT-100 water bath. The CD spectropolarimeter and the optical cells were calibrated with recrystallized d-10camphorsulfonic acid. The measurements were carried out in 1-mm path length cells and protein solutions of approximately 0.5-1.0 mg/ml. All spectra were measured at 25°C with a 16-s time constant and a scan  (Rost and Sander, 1994): capitals denote positions with an expected accuracy higher than 82%; E, ␤-strand; L, not helix or ␤-strand. In the case of the second type II domain of PDC-109 (PDC2) for which the solution structure has been determined by NMR spectroscopy (Constantine et al., 1992), the segments known to form ␤-sheets are underlined. rate of 10 nm/min. The spectral slit width was 1.0 nm. All measurements represent the computer average of five scans. Secondary structure of recombinant proteins was estimated from their CD spectra with the J-700 program for Windows Secondary Structure Estimation Ver.1.10.00, JASCO.
Unfolding of recombinant proteins containing the fibronectin type II domains of gelatinase A was monitored by the changes in their CD spectra at 224 nm. To monitor thermal unfolding of type II units, circular dichroism was recorded at 224 nm during the course of heating the solutions from 15°C to 90°C at a heating rate of 50°C/h. The experiments were conducted at pH 8.0, 10 mM Tris-HCl, the protein concentration was 0.5-1.0 mg/ml. Melting temperatures were determined by derivative processing of the CD changes using the J-700 program for Windows Standard Analysis Ver.1.30.00.7, JASCO.
Differential Scanning Calorimetry-Calorimetric measurements were carried out on the differential scanning microcalorimeter DASM-4 microcalorimeter at a heating rate of 1 K/min and a solution concentration of 1-5 mg/ml. Experiments were conducted at pH 8.0, in 10 mM Tris-HCl buffer. Buffer base lines were obtained under the same conditions and subtracted from sample tracings.
Secondary Structure Prediction-Secondary structures of type II domains were predicted by the method of Chou and Fasman (1978). Predictions based on multiple alignments of type II modules were carried out by the procedure of Rost and Sander (1994) using the PHD program.
Protein Modeling-Homology modeling of the catalytic domain and type II domains of gelatinase A was carried out at the Glaxo Institute for Molecular Biology SA using the Swiss-Model Automated Protein Modelling Service (Peitsch, 1995). Protein structure prediction by the Swiss-Model is based on the principle that homologous proteins with a high sequence similarity are characterized by distinct structural similarity. Swiss-Model first searches the Brookhaven Protein Data Bank for proteins that show a significant sequence similarity to the target protein. The framework structure of the model is produced by aligning the target sequence with the selected template sequences using a combination of sequence alignment tools and three-dimensional superposition. Gaps in the framework structure are then filled by structural similarity searches through the Brookhaven Data Bank (Peitsch, 1995). In the present work, the catalytic domain of gelatinase A was modeled using the coordinates of fibroblast collagenase, MMP-1 1 (Lovejoy et al., 1994a(Lovejoy et al., , 1994bSpurlino et al., 1994), neutrophil collagenase, MMP-8 (Stams et al., 1994), and stromelysin, MMP-3 (Gooley et al., 1994). The three type II domains of gelatinase A were modeled using the coordinates of the second type II domain of PDC-109 (Constantine et al., 1992).

RESULTS
The far UV CD spectra of DEL␤galcoll 123 (containing the entire fibronectin-related part of human gelatinase A), as well as the CD spectra of recombinant proteins containing single type II modules are characterized by maxima at 224 nm and minima at 198 nm (Fig. 1). Removal of the ␤-galactosidase fusion peptides had no major effect on the CD spectra of these proteins (cf. DEL␤galcoll 123 and coll 123 in Fig. 1). Analysis of the CD spectra has shown that the fibronectin-related domain of gelatinase A consists of about 32% ␤-sheet, 19% ␤-turn with no detectable ␣ helix. Recombinant proteins coll 1, coll 2, and coll 3 were estimated to contain 30%, 30%, and 31% ␤-sheet, 28%, 27%, and 23% ␤-turn structures, respectively.
The estimated secondary structures of type II domains of gelatinase A are in good agreement with the predictions based on their homology with the second type II domain of PDC-109. The solution structure of the latter protein has been solved by NMR spectroscopy (Constantine et al., 1992), and these studies have revealed the presence of two short antiparallel ␤-sheets connected by ␤-turns, with no evidence for ␣ helix. Secondary structure predictions by the method of Chou and Fasman 1 The abbreviation used is: MMP, matrix metalloproteinase.  (1978) also suggested the presence of ␤-turns and ␤-sheets but no ␣ helix in fibronectin type II domains (Patthy et al., 1984). This conclusion was confirmed and extended in the present work for type II modules of gelatinase A and B as well as PDC-109 (Esch et al., 1983), BSP A3 (Seidah et al., 1987), mannose 6-phosphate receptor (Morgan et al., 1987), blood coagulation factor XII (McMullen and Fujikawa, 1985), hepatocyte growth factor activator (Miyazawa et al., 1993), mannose receptor (Taylor et al., 1990), and phospholipase receptor (Ishizaki et al., 1994) using the multiple alignment-based method of Rost and Sander (1994). In agreement with our earlier findings (Patthy et al., 1984), the type II modules of these proteins are predicted to have four short ␤-strands in the vicinity of their four half-cystines and an additional ␤-strand is predicted in the N-terminal extension that is present only in some of the type II units (Fig. 2). It is noteworthy that the predicted ␤-strands align with the four ␤-strands determined experimentally for PDC-109 domain 2. The CD spectra of type II modules are similar to those of the related kringle domains inasmuch as they also have characteristic maxima near 225 nm (Castellino et al., 1986). In accordance with the homology of kringles and type II domains (Patthy et al., 1984), both are characterized by the presence of antiparallel ␤-sheets and ␤-turns (Constantine et al., 1992). The maxima at 224 nm are characteristic of the ordered, native structure of the type II modules of the gelatin-binding domain. For example, denaturation of DEL␤galcoll 123 eliminates this maximum (Fig. 1); the reduced-alkylated protein has a minimum at 200 nm, typical of unordered proteins (Chang et al., 1978). The disulfide bonds of type II units are essential for the integrity of their three-dimensional structure since reductive cleavage of S-S bonds of DEL␤galcoll 123 in the absence of denaturing agents leads to a CD spectrum characteristic of the fully denatured form and eliminates its affinity for gelatin-Sepharose 4B (not shown). (It must be pointed out that this finding contrasts with the conclusion of Steffensen et al. (1995); these authors suggested that reductive cleavage of the disulfide bonds has no effect on the functional integrity of these domains).
Unfolding of the recombinant proteins with urea or guanidinium HCl eliminates the maxima at 224 nm in a highly cooperative fashion. As shown in Fig. 3, the sensitivity of DEL␤galcoll 1, DEL␤galcoll 2, and DEL␤galcoll 3 to denaturant-induced unfolding is markedly different: the midpoints of unfolding with guanidinium hydrochloride are 0.3 M, 1.6 M, and 4.8 M for DEL␤galcoll 2, DEL␤galcoll 1, and DEL␤galcoll 3, respectively. The order of sensitivity of type II domains to urea-induced unfolding is similar: the midpoints of unfolding with this denaturant are 1.7 M and 6.5 M for DEL␤galcoll 2 and DEL␤galcoll 1, respectively. The most stable DEL␤galcoll 3 is not unfolded even in 8 M urea.
It may be pointed out that unfolding of DEL␤galcoll 123 with guanidinium HCl occurs in three steps, the midpoints of these steps coincide with unfolding of the individual type II units (Fig. 3A). In contrast with this, unfolding of DEL␤galcoll 123 with urea is not detectable below 3 M urea, even though unfolding of DEL␤galcoll 2 is practically complete at this concentration (Fig. 3B). It seems likely that interactions in DEL␤galcoll 123 increase the stability of the second type II domain to urea-induced unfolding.
Thermal unfolding was also monitored by changes in the CD spectra of type II modules at 224 nm. These studies have shown that the three type II modules show marked differences in their thermal stability (Fig. 4 and Table I). The least stable is the second type II module with a T m value of 44°C, the most stable is the third unit with a T m of 80°C. Differences in their stability are retained in the recombinant proteins that contain two or three type II domains, permitting the resolution of the thermal transitions into distinct components (cf. Fig. 4 and Table I). From a comparison of recombinant proteins containing the N-terminal fusion peptide (DEL␤galcoll series) with those lacking this segment (coll series), it is also clear that the N-terminal peptide does not have a major effect on the thermal transition of type II modules (Table I).
When the T m values of the second type II module of different recombinant proteins are compared (Table I), it is clear that neighboring modules have a marked influence on its thermal stability. This is most obvious in the comparison of the T m value (44°C) of type II unit 2 of coll 2 with its T m value (54°C) in full-length collagen-binding domain (coll 123): the presence of module 1 and module 3 increases its T m value by 10°C. The presence of module 1 or module 3 alone (in coll 12 or coll 23) causes a smaller but significant increase in thermal stability of module 2 (Table I). It thus seems likely that the second type II domain interacts tightly with both module 1 and module 3.
Differential scanning calorimetry studies (e.g. Fig. 5) yielded T m values similar to those obtained by monitoring changes in CD spectra (Table I). The melting curves of the individual domains are each well described by single two-state transitions, consistent with their ⌬H cal /⌬H vH ratios close to unity (data not shown). Comparison of the enthalpy changes of the unfolding of type II domain 2 in coll 123 (⌬H cal ϭ 58 kcal/mol) and coll 2 (⌬H cal ϭ 49 kcal/mol) indicates that domain-domain interactions in coll 123 significantly stabilize the folded state of domain 2.

DISCUSSION
In the present investigation we have shown by analysis of the CD spectra of recombinant proteins that the type II domains of the gelatin-binding site of type IV collagenase have a protein-fold characterized by a high content of ␤-structures, with no ␣-helix. This observation is in harmony with structural information obtained on a related type II domain, the second unit of the bovine seminal fluid protein PDC-109 (Constantine et al., 1992). Secondary structure prediction from multiple alignments of type II units also supports the presence of ␤-structures but no ␣-helix (Patthy et al. (1984) and the present study). CD spectroscopy and differential scanning microcalorimetry of the collagen-binding domain has shown that the thermal stability of the second type II module is significantly increased by its interactions with the first and third modules: its T m value is increased from 44°C to 54°C. The fact that the stability of the second type II domain to urea-induced unfolding is increased by the presence of the other two type II units also suggests that these units are involved in tight interactions. It seems likely that these interactions between the tandem modules permit little flexibility in their relative orientation at physiological temperatures. In view of the fact that all three The lower figure shows a hypothetical model of gelatinase A consistent with the experimental data. In this model, the three type II units form a hydrophobic groove lined by the aromatic residues (purple) that are known to be involved in gelatin binding in the case of the type II domain of PDC-109 (Constantine et al., 1992). This groove may serve to bind and orient the substrate relative to the catalytic site, presenting the scissile bond to the active site of the enzyme. The figures were prepared with Insight II (Biosym Techn, San Diego). type II units contribute to gelatin binding (Bá nyai et al., 1994), we propose that their fixed arrangement facilitates the tight binding of a single substrate molecule.
The binding of the substrate by the type II domains may play a crucial role in the proper positioning of these substrates relative to the active site cleft of gelatinase A. Although the three-dimensional structure of gelatinase A has not yet been determined, structures of proteins homologous to its threeconstituent domains: the catalytic domain (Lovejoy et al., 1994a(Lovejoy et al., , 1994bBorkakoti et al., 1994;Gooley et al., 1994;Stams et al., 1994;Bode et al., 1994), hemopexin-like domain (Faber et al., 1995;Li et al., 1995), and fibronectin type II domain (Constantine et al., 1992) have been determined recently.
The catalytic domains of other members of the matrixin family of metzincins MMP-1, MMP-3, and MMP-8 (Lovejoy et al., 1994a(Lovejoy et al., , 1994bBorkakoti et al., 1994;Gooley et al., 1994;Stams et al., 1994;Bode et al., 1994;Stöcker and Bode, 1995) have been shown to have very similar three-dimensional structures. They all comprise a regularly folded "upper" subdomain consisting of a twisted five-stranded ␤-sheet flanked by two ␣ helices and connecting loops, and a smaller, less regularly folded, "lower" subdomain comprising two open loops and the C-terminal helix. The catalytic zinc ion residing at the bottom of the active-site cleft between these subdomains is bound by a characteristic helix-bend-loop structure comprising the HEXX-HXXGXXH consensus sequence. The catalytic zinc ion is coordinated by the imidazolyl N⑀2 atoms of the three consensus histidine residues and by a water molecule that is also bound to a glutamate (Bode et al., 1994;Stöcker and Bode, 1995).
On the basis of the close homology of gelatinase A (MMP-2) with other matrix metalloproteinases, it is safe to assume that its three-dimensional structure is also similar to these enzymes, except that three fibronectin-related type II units are inserted at the N-terminal boundary of the active site helix of gelatinase A, seven residues upstream of the HEXX-HXXGXXH consensus sequence motif. Homology modeling of the catalytic domain of gelatinase A indicates that the type II domains are inserted at the right-hand end of the active-site cleft, close to the SЈ 1 pocket of the enzyme (Fig. 6). In view of the proximity of the collagen-binding type II units to the active site cleft, it is conceivable that they are crucial for properly orienting the substrate relative to the catalytic site, presenting the scissile bond to the active site of the enzyme. Such a role of the collagen-binding domain is supported by the observation that the cleavage pattern of type IV collagen is altered in a mutant gelatinase A lacking this domain (Murphy et al., 1994). In accordance with the experimental data, we suggest a model in which the three type II units form an extension of the substrate binding cleft of gelatinase A (Fig. 6).