Structural and Functional Basis of the Serine Protease-like Hepatocyte Growth Factor (cid:1) -Chain in Met Binding and Signaling*

Hepatocyte growth factor (HGF), a plasminogen-re-lated growth factor, is the ligand for Met, a receptor tyrosine kinase implicated in development, tissue re-generation, and invasive tumor growth. HGF acquires signaling activity only upon proteolytic cleavage of sin-gle-chain HGF into its (cid:2) / (cid:1) heterodimer, similar to zymogen activation of structurally related serine proteases. Although both chains are required for activation, only the (cid:2) -chain binds Met with high affinity. Recently, we reported that the protease-like HGF (cid:1) -chain binds to Met with low affinity (Stamos, J., Lazarus, R. A., Yao, X., Kirchhofer, D., and Wiesmann, C. (2004) EMBO J. 23, 2325–2335). Here we demonstrate that the zymogen-like form of HGF (cid:1) also binds Met, albeit with 14-fold lower affinity than the protease-like form, suggesting optimal interactions result from conformational changes upon cleavage of the single-chain form. Extensive mutagenesis Construction, Expression, and Purification of Full-length HGF Pro- teins— Recombinant proteins were produced in 1-liter cultures of Chinese hamster ovary (CHO) cells by transient transfection (23). Amino acid changes were introduced by site-directed mutagenesis (24) and verified by DNA sequencing. The expression medium (F-12/Dulbecco’s modified Eagle’s medium) contained 1% (v/v) ultra low IgG fetal bovine serum (FBS) (Invitrogen). After 8 days the medium was harvested and supplemented with FBS to give a final content of 5–10% (v/v). Additional incubation for 2–3 days at 37 °C resulted in complete single-chain HGF conversion. This step was omitted for expression of scHGF, a noncleavable single-chain form, which has amino acid changes at the activation cleavage site (R494E) and at a protease-susceptible site in the (cid:1) -chain (R424A) (23). Mutant proteins were purified from the me- dium by HiTrap-Sepharose SP cation exchange chromatography (Am-ersham Biosciences) as described (23). Examination by SDS-PAGE (4–20% gradient gel) under reducing conditions and staining with Sim-ply Blue Safestain (Invitrogen) showed that all HGF mutants were (cid:1) 95% pure and were fully converted into (cid:1) / (cid:2) heterodimers except for scHGF, which remained as a single-chain form. Protein concentration for each mutant was determined by quantitative amino acid analysis. first x-rays. Data processing and reduction were performed using HKL (26) (HKL Research, Charlottesville, VA) and CCP4 (27). The structure was solved by molecular replacement using AMoRe (28) in space group P3 1 21, using parts of the protease domain of coagulation factor VIIa (29) as the search probe. Refinement was performed using X-PLOR98 (Accelrys, San Diego, CA) and REFMAC (30). Inspection of electron density maps and model manipulation was performed using XtalView (31) (Syrrx, San Diego, CA).

Hepatocyte growth factor (HGF), 1 also known as scatter factor, is the ligand for Met (1,2), a receptor tyrosine kinase encoded by the c-met protooncogene (3). HGF binding to Met induces phosphorylation of the intracellular kinase domain resulting in activation of a complex set of intracellular pathways that lead to cell growth, differentiation, and migration in a variety of cell types; several recently published reviews (4 -6) provide a comprehensive overview. In addition to its fundamental importance in embryonic development and tissue regeneration, the HGF/Met signaling pathway has also been implicated in invasive tumor growth and metastasis and as such represents an interesting therapeutic target (4,5,7,8).
We investigated the importance of the protease-like domain of HGF for molecular interactions with Met based upon a significant similarity to serine proteases and their activation process (9). In serine proteases, cleavage of the zymogen effects a conformational rearrangement of the so-called "activation domain" giving rise to a properly formed active site and the substrate/inhibitor interaction region. The activation domain constitutes three surface-exposed loops designated the [c140], [c180], and [c220] loops and the newly formed N terminus, which inserts into a hydrophobic pocket (17). We thus reasoned that rearrangements in the corresponding "activation domain" of the HGF ␤-chain might form a Met interaction site. In the homologous ligand/receptor pair macrophage-stimulating protein (MSP)/Ron, the serine protease-like MSP ␤-chain provides the main energy for receptor binding (18,19). This is reversed from the HGF/Met system where the high affinity receptor binding site for Met resides in the HGF ␣-chain (20,21).
To better understand the binding requirements and mechanism for Met activation and signaling, we expressed and purified various forms of HGF ␤. Consistent with our recently published data (22), we find that the wild type protease-like HGF ␤ binds to Met. A zymogen-like single-chain form of HGF ␤ (scHGF ␤) also binds to Met, but with lower affinity, suggesting that this Met binding site is influenced by a conformational rearrangement upon cleavage. Met binding residues were identified by mutagenesis studies on both full-length HGF and the HGF ␤-chain and were mapped onto the crystal structure of HGF ␤, which was determined to a resolution of 2.53 Å. The mutagenesis results define a distinct Met binding site, which consists of residues of the "active site region" and "activation domain" of the serine protease-like HGF ␤-chain. Models of Met dimerization and activation by two-chain HGF are discussed.

EXPERIMENTAL PROCEDURES
Expression and Purification of HGF ␤ Proteins-All HGF ␤ proteins (Val 495 [c16] to Ser 728 [c250]), which contained a His 6 C-terminal tag, were expressed and purified to homogeneity (Ͼ95% purity) as described previously (22 Tranfection of Sf9 cells on plates in ESF 921 media (Expression Systems LLC, Woodland, CA) was carried out using the BaculoGold TM Expression System according to the manufacturer's instructions (Pharmingen). Virus amplification, cell culture, and purification was carried out essentially as described previously except that the elution buffer for the nickel-nitrilotriacetic-agarose column contained 500 mM imidazole instead of 250 mM imidazole (22). Proteins were analyzed by 12% SDS-PAGE stained with Coomassie Blue. Mutations were verified by DNA sequencing and mass spectrometry. Protein concentration was determined by quantitative amino acid analysis. N-terminal sequencing revealed a single correct N terminus present for scHGF ␤ and HGF ␤. Purified proteins showed the correct molecular mass on SDS-PAGE; multiple bands observed were likely due to heterogeneous glycosylation, consistent with the mass spectrometry data having molecular masses ϳ2 kDa higher than predicted from the sequence.
Construction, Expression, and Purification of Full-length HGF Proteins-Recombinant proteins were produced in 1-liter cultures of Chinese hamster ovary (CHO) cells by transient transfection (23). Amino acid changes were introduced by site-directed mutagenesis (24) and verified by DNA sequencing. The expression medium (F-12/Dulbecco's modified Eagle's medium) contained 1% (v/v) ultra low IgG fetal bovine serum (FBS) (Invitrogen). After 8 days the medium was harvested and supplemented with FBS to give a final content of 5-10% (v/v). Additional incubation for 2-3 days at 37°C resulted in complete single-chain HGF conversion. This step was omitted for expression of scHGF, a noncleavable single-chain form, which has amino acid changes at the activation cleavage site (R494E) and at a protease-susceptible site in the ␣-chain (R424A) (23). Mutant proteins were purified from the medium by HiTrap-Sepharose SP cation exchange chromatography (Amersham Biosciences) as described (23). Examination by SDS-PAGE (4 -20% gradient gel) under reducing conditions and staining with Simply Blue Safestain (Invitrogen) showed that all HGF mutants were Ͼ95% pure and were fully converted into ␣/␤ heterodimers except for scHGF, which remained as a single-chain form. Protein concentration for each mutant was determined by quantitative amino acid analysis.
Binding of HGF ␤ to Met in a Competition Binding ELISA-To develop a competition ELISA, we first determined the direct specific binding of the HGF ␤-chain to Met in a plate ELISA. Microtiter plates were coated with Met-IgG fusion protein (25) as described (22) and incubated with wild type HGF ␤-chain. Bound HGF ␤ was detected using penta-His horseradish peroxidase conjugate (Qiagen, Valencia, CA) followed by addition of SureBlue TMB peroxidase substrate and TMB STOP (Kirkegaard & Perry Laboratories, Gaithersburg, MD). The determined effective concentration to give half-maximal binding (EC 50 ) was 320 Ϯ 140 nM (n ϭ 6).
Based on these results, a competition binding ELISA was developed as described (22) using wild type HGF ␤ biotinylated with 20-fold molar excess of biotin-maleimide (Pierce). Briefly, plates were coated with Met-IgG fusion protein and incubated with a mixture of 250 nM biotinylated wild type HGF ␤ and various concentrations of unlabeled HGF ␤, HGF ␤ mutants, or scHGF ␤. After incubation for 1 h at room temperature, the amount of biotinylated wild type HGF ␤ bound on the plate was measured by using horseradish peroxidase-neutravidin (Pierce). IC 50 values were determined by fitting the data to a fourparameter equation (Kaleidagraph, Synergy Software, Reading, PA).
Binding of HGF Mutants to Met-Biotinylated HGF (two-chain full length) was prepared using the Sigma immunoprobe biotinylation kit (Sigma). Microtiter plates were coated with rabbit anti-human IgG Fc-specific antibody as described above. Plates were washed in PBS containing 0.05% (v/v) Tween 20 followed by a 1-h incubation with 0.5% (w/v) bovine serum albumin, 0.05% Tween 20 in PBS, pH 7.4, at room temperature. After washing, 1 nM biotinylated HGF and 0.2 nM Met-IgG fusion protein (25) together with various concentrations of HGF mutants were added to the wells and incubated for 2 h. After washing, bound biotinylated HGF was detected by addition of diluted (1:3000) streptavidin horseradish peroxidase conjugate (Zymed Laboratories Inc., South San Francisco, CA) followed by SureBlue TMB peroxidase substrate and stop solution TMB STOP (Kirkegaard & Perry Laboratories). The A 450 was measured, and IC 50 values were determined as described above. Relative binding affinities are expressed as the IC 50 (mutant)/IC 50 (wild type HGF).
HGF-dependent Phosphorylation of Met-The kinase receptor activation assay (KIRA) was carried out as follows. Confluent cultures of lung carcinoma A549 cells (CCL-185, ATCC, Manassas, VA), maintained previously in growth medium (Ham's F-12/Dulbecco's modified Eagle's medium 50:50 (Invitrogen) containing 10% FBS (Sigma)), were detached using Accutase (ICN, Aurora, OH) and seeded in 96-well plates at a density of 50,000 cells per well. After overnight incubation at 37°C, growth medium was removed, and cells were serum-starved for 30 -60 min in medium containing 0.1% FBS. Met phosphorylation activity by HGF and HGF mutants was determined from addition of serial dilutions from 500 to 0.2 ng/ml in medium containing 0.1% FBS followed by a 10-min incubation at 37°C, removal of media, and cell lysis with cell lysis buffer (Cell Signaling Technologies, Beverly, MA) supplemented with protease inhibitor mixture set I (Calbiochem). HGF ␤-chain studies were carried out similarly starting at 5 g/ml. Cell lysates were analyzed for phosphorylated Met via an electrochemiluminescence assay using a BioVeris M-Series instrument (BioVeris Corp., Gaithersburg, MD). Anti-phosphotyrosine antibody 4G10 (Upstate Biotechnology, Inc., Lake Placid, NY) was labeled with ORI-TAG via Nhydroxysuccinimide ester chemistry according to the manufacturer's directions (BioVeris). Anti-Met extracellular domain antibody 1928 (Genentech, Inc.) was biotinylated using biotin-X-N-hydroxysuccinimide (Research Organics, Cleveland, OH). The BV-TAG-labeled 4G10 and biotinylated anti-Met antibody were diluted in assay buffer (PBS, 0.5% Tween 10, 0.5% bovine serum albumin), and the mixture was added to the cell lysates. After incubation at room temperature with vigorous shaking for 1.5-2 h, streptavidin magnetic beads (Dynabeads, BioVeris) were added and incubated for 45 min. The beads with bound material (anti-Met antibody/Met/anti-phosphotyrosine antibody) were captured by an externally applied magnet. After a wash step, the chemiluminescent signal generated by the light source was measured as relative luminescent units on a BioVeris instrument. For each experiment, the Met phosphorylation induced by HGF mutants was expressed as a percentage of the maximal signal obtained with two-chain HGF.
Cell Migration Assay-Breast cancer cells MDA-MB435 (HTB-129, ATCC, Manassas, VA) were cultured in recommended serum-supplemented medium. Confluent cells were detached in PBS containing 10 mM EDTA and diluted with serum-free medium to a final concentration of 0.6 -0.8 ϫ 10 6 cells/ml. This cell suspension (0.2 ml) was added in triplicate to the upper chambers of 24-well Transwell plates (8 m pore size) (HTS Multiwell TM Insert System, Falcon, Franklin Lakes, NJ) pre-coated with 10 g/ml of rat tail collagen type I (Upstate Biotechnology, Inc.). Wild type HGF or HGF mutants were added to the lower chamber at 1 nM in serum-free medium, unless specified otherwise. HGF ␤-chain was also tested at 0.95 M. After incubation for 13-14 h, cells on the apical side of the membrane were removed, and those that migrated to the basal side were fixed in 4% paraformaldehyde followed by staining with a 0.5% crystal violet solution. After washing and air drying, cells were solubilized in 10% acetic acid, and the A 560 was measured on a Molecular Devices microplate reader. Pro-migratory activities of HGF mutants were expressed as percent of HGF controls after subtracting basal migration in the absence of HGF. Photographs of stained cells were taken with a Spot digital camera (Diagnostics Instruments, Inc., Sterling Heights, MI) connected to a Leitz microscope (Leica Mikroskope & Systeme GmbH, Wetzlar, Germany). Pictures were acquired by Adobe Photoshop 4.0.1 (Adobe Systems Inc., San Jose, CA).
HGF ␤ X-ray Structure-Purified HGF ␤ was concentrated to 10 mg/ml using a Centriprep® YM-10 in 10 mM HEPES, pH 7.2, 150 mM NaCl, 5 mM CaCl 2 . Hanging drops (1 l of protein and 1 l of 30% PEG-1500) over a reservoir containing 500 l of 30% PEG-1500 (Hampton Research, Laguna Niguel, CA) yielded crystalline rods (ϳ25 ϫ 25 ϫ 500 m) during incubation at 19°C overnight. A crystal fragment was preserved directly from the mother liquor by immersion in liquid nitrogen. Data extending to 2.53 Å resolution were collected on a Quantum 4 ccd detector (ADSC, Poway, CA) at ALS beam line 5.0.2 with 1.0 Å wavelength x-rays. Data processing and reduction were performed using HKL (26) (HKL Research, Charlottesville, VA) and CCP4 (27). The structure was solved by molecular replacement using AMoRe (28) in space group P3 1 21, using parts of the protease domain of coagulation factor VIIa (29) as the search probe. Refinement was performed using X-PLOR98 (Accelrys, San Diego, CA) and REFMAC (30). Inspection of electron density maps and model manipulation was performed using XtalView (31) (Syrrx, San Diego, CA).

RESULTS
Binding of HGF ␤ to Met-Because single-chain HGF binds to Met with comparable affinity to that of two-chain HGF, but does not induce Met phosphorylation (11)(12)(13), we hypothesized that this may be due to the lack of a Met binding site in the uncleaved form of the ␤-chain. To test this hypothesis, we carried out competition binding ELISAs with expressed and purified HGF ␤ and scHGF ␤. scHGF ␤ is a zymogen-like form of HGF ␤ containing the C-terminal 16 residues from the HGF ␣-chain and a mutation at the cleavage site (R494E) to ensure that the single-chain form remained intact. Initial ELISA data showed that HGF ␤ directly bound to immobilized Met-IgG fusion protein, albeit with relatively low affinity. This is consistent with our recent data from surface plasmon resonance experiments with immobilized Met extracellular domain, where a K d of ϳ90 nM for the HGF ␤/Met interaction was found (22). In the HGF ␤/Met competition binding ELISA we determined IC 50 values of 0.86 Ϯ 0.17 and 11.6 Ϯ 1.8 M for HGF ␤ and the precursor form scHGF ␤, respectively, showing that scHGF ␤ bound Met with 14-fold lower affinity than HGF ␤ (Fig. 1A). Thus, although a Met binding site on the zymogenlike HGF ␤ does in fact exist, it is not optimal. The apparent affinity differences observed between K d (22) and IC 50 values are due to the different assays used; the higher IC 50 values reflect the higher concentrations of HGF ␤ necessary to compete with 250 nM biotinylated HGF ␤ for binding to Met.
We also measured the ability of HGF ␤ to stimulate Met phosphorylation of A549 lung cancer cells. Fig. 1B shows that HGF ␤ was completely inactive, even at concentrations that exceeded optimal phosphorylation activity by full-length HGF by Ͼ1000-fold. Similarly, in MDA-MB435 cell migration assays, HGF ␤ at concentrations of up to 0.95 M had no effect. These results are consistent with other studies demonstrating the inability of HGF ␤ to activate the Met receptor (11,32).

Effects on Cell Migration and Met Phosphorylation by HGF and HGF ␤-Chain
Mutants-To identify the Met binding site in the ␤-chain, we systematically changed residues in regions corresponding to the activation domain and the active site of serine proteases, herein referred to as the "activation domain" and "active site region" of HGF. Initial expression of HGF mutants in CHO cells yielded a mixture of single-and twochain HGF forms, exemplified by mutant HGF I623A ( Fig. 2A). Complete conversion of residual uncleaved HGF was accomplished by additional exposure of the harvested culture medium to 5-10% serum for several days (Fig. 2A) ) had 60 -80% of wild type activity. The remaining 21 mutants had activities Ͼ80% that of wild type and were considered essentially unchanged from HGF. As expected, scHGF did not stimulate cell migration (Fig. 2B). The complete inability of 1 nM R695A [c217] or G696A [c219] to promote cell migration is illustrated in Fig. 2C, showing that migration in the presence of either mutant is similar to basal migration in the absence of HGF.
To examine whether reduced activities in cell migration correlated with reduced Met phosphorylation, a subset of HGF mutants was examined in a kinase receptor assay (KIRA). For wild type HGF and HGF mutants, maximal Met phosphorylation was observed at concentrations between 0.63 and 1.25 nM (Fig. 3) (Table I). We also examined the cell migration activities of selected mutants at 10-and 50-fold higher concentrations; no increase in pro-migratory activity was observed (Table II). Therefore, the impaired function of HGF mutants is not due to reduced overall binding to Met, since an increase in concentration of up to 50-fold had no compensatory effect.
The poor correlation between HGF mutant binding to Met and either HGF-dependent cell migration or Met phosphorylation is likely due to the high affinity between Met and the HGF ␣-chain. The HGF ␣-chain dominates the overall binding and thus masks any altered Met interactions stemming from the low affinity HGF ␤-chain. Therefore, we made mutations in HGF ␤ itself to eliminate any ␣-chain effects. background to avoid any potential dimerization during purification, although this mutation had no effect on binding to Met (Fig. 4). All mutants had reduced binding affinity to Met and R695A [c217] and G696A [c219] did not compete for binding at all (Fig.  4). The binding affinities of the mutants were then normalized to HGF ␤, which had an IC 50 ϭ 0.55 Ϯ 0.38 M (n ϭ 16). The results are summarized in Table III and include binding data of previously characterized HGF ␤ mutants Y673A [c195] and V692A [c214] (22). We found that mutants R695A [c217], G696A [c219], and Y673A [c195] had the greatest loss in migration activity (as two-chain full-length HGF mutants) and also had the greatest loss in Met binding (as HGF ␤ mutants). Conversely, mutants with a small reduction of migration activity (Y619A [c143] and I699A [c221a]) also had a small (less than 10-fold) reduction in Met binding (Table III). Thus, the elimination of HGF ␣-chain binding contribution in this Met binding assay revealed that the reduced migration activity of full-length HGF mutants was due to an impaired binding interaction of the HGF ␤-chain with the Met receptor.
Location of the Met Binding Site in the Crystal Structure of HGF ␤-To better interpret Met binding and activity data from HGF mutants, we determined the HGF ␤ structure at 2.53 Å resolution. Data reduction, refinement statistics, and final model metrics appear in Table IV. HGF ␤ adopts the fold of chymotrypsin-like serine proteases, comprising two tandem distorted ␤-barrels. There are two poorly ordered and untraceable segments, His 645 -Thr 651 [c170a-c175] and the C-terminal region beginning with Tyr 723 [c245]. The active site region of HGF ␤ clearly differs from those of true proteases (Fig. 5A) (Fig. 5B). Furthermore, there are structural differences in the nominal "S1 pocket," where Gly 667 [c189] at the bottom of the pocket and Pro 668 [c190] are also distinct from residues found in serine proteases. Thus we have a structural basis to understand why mutations in HGF creating the Asp [c102]-His [c57]-Ser [c195] catalytic triad would be insufficient to impart catalytic activity.
HGF ␤ residues important for interactions with Met are shown in Fig. 5, C and D, according to their relative activities in cell migration assays. The Met binding site is compact and centered on the "active site region." This core region comprises "catalytic triad" residues (Gln 534 -621), and residues 514 and 537 (Fig. 5, C and D).
Recently, the HGF ␤/Met binding site in the crystal structure of HGF ␤ bound to the N-terminal portion of Met comprising the Sema and PSI domains was identified (22). Superimposition (33) of unbound HGF ␤ (present study) and HGF ␤ bound to Met gave a root mean square deviation value of 0.5 Å for the 203 C␣ pairs, which excludes residues Gly 638 -Asn 653 (some are not seen in the present study and others are strongly shifted in the Met complex), Gly 694 [c216] and Arg 695 [c217] which are shifted by about 2 Å, and six others far from the Met binding site. While keeping these differences in mind, we can use our HGF ␤ structure to delineate the structural binding site from the HGF ␤⅐Met structure in order to compare the functional and structural binding sites (Fig. 5D). Fig. 5D shows an almost complete overlap of the Met binding sites identified by functional and structural approaches. The four residues that lie outside the structural binding site, i.e.

DISCUSSION
HGF acquires biological activity only upon proteolytic conversion of the single-chain precursor pro-HGF into two-chain HGF (11)(12)(13)(14). Based on the structural similarity of HGF with chymotrypsin-like serine proteases and plasminogen in particular, we propose that this activation process is associated with structural changes occurring in the HGF ␤-chain. Here we provide evidence that the "activated" form of the HGF ␤-chain contains a distinct Met binding site located in a region that corresponds to the substrate/inhibitor binding site of chymotrypsin-like serine proteases.
HGF Binding Interactions with Met-Binding studies with purified HGF ␤-chains revealed that the "activated form" of HGF ␤ binds to Met with ϳ14-fold higher affinity than its precursor form, scHGF ␤, consistent with the view that optimization of the Met binding site is contingent upon processing of single-chain HGF. This suggested that the Met binding site includes the HGF region undergoing conformational rearrangements after pro-HGF cleavage, i.e. the "activation domain." Indeed, functional analysis of HGF variants with amino acid substitutions in the "activation domain" led to the identification of the functional Met binding site. However, HGF mutants with the greatest reduction in pro-migratory activities ( (20,21). Consistent with this, the reduced activities remained unchanged upon increasing the concentration of HGF mutants by more than 50-fold (Table II). Therefore, the reduced activities of HGF mutants were interpreted as resulting from perturbed molecular interactions of HGF ␤-chain with its specific, low affinity binding site on Met. In support of this, we found that the    b Values are relative to wild type two-chain HGF activity ( ϭ 100%) in the cell migration assay.
c Data were taken from Stamos et al. (22).
reduced biological activities of selected HGF mutants (twochain full-length) were well correlated with reduced Met binding of their corresponding HGF ␤ mutants in an assay that eliminated the binding contribution of the HGF ␣-chain (Table  III) ). Together, these residues define a region that bears a remarkable resemblance to the substrate-processing region of true serine proteases. The functional importance of the [c220] loop has precedent in the well described family of chymotrypsin-like serine proteases (15,16). The extended canonical conformation of substrates and inhibitors includes residues that can form main chain interactions from [c214] to [c216]. This region is also recognized as an allosteric regulator of thrombin catalytic activity (34) and as an interaction site with its inhibitor hirudin (35). In addition, residues in factor VIIa and thrombin that correspond to HGF Arg 695 [c217] are important for enzymecatalyzed substrate processing (36,37). In MSP, the closest structural homolog of HGF, this residue (Arg 683 [c217]) plays a pivotal role in the high affinity interaction of MSP ␤-chain with its receptor Ron (38). Arg 683 [c217] is part of a cluster of five surface-exposed arginines proposed to be involved in Ron binding (19). Although only Arg 695 [c217] and possibly Lys 649 [c173] are conserved in HGF, these residues are all located within the Met binding region of the HGF ␤-chain, suggesting that the Ron binding site on the MSP ␤-chain is highly homologous.
The functional binding site identified herein is in excellent agreement with the structural Met binding site revealed in the crystal structure of the complex of HGF ␤ bound to soluble Met Sema/PSI domain (22). The 17 identified functional binding residues are located within or proximal to the structural binding region (Fig. 5D), which mainly interacts with three sepa-rate loops from the Met Sema domain. Notably, residues of the functional "core" region, e.g. , also make the most important contacts to the Met receptor in the crystal structure. For example, Met residues Tyr 125 and Tyr 126 that are in the core of the binding interface pack against the HGF ␤-chain residue Arg 695 [c217]. Thus, the results derived from three different experimental approaches, functional studies with HGF mutants, Met binding assays with HGF ␤ mutants, and the HGF ␤⅐Met crystal structure, are consistent and provide strong evidence for a distinct Met binding site located at the active site region of the HGF ␤-chain.
Our findings with HGF Ala mutants agree with a previous study where Tyr 673 [c195] and Val 692 [c214] were each replaced by serine (12). The normal biological activity measured for HGF variant Q534H [c57] in two previous studies (12,39) may reflect functional compensation of Gln by His, a relatively close isostere. However, our results contrast with previous studies demonstrating that HGF ␤-chain itself did not bind to Met (11,32). In one instance, the HGF ␤-chain was different from ours, having extra ␣-chain residues derived from elastase cleavage of HGF, which could adversely affect Met binding. However, it is more likely that either the concentrations used, the sensitivity of the assays, or the extent of pro-HGF processing may have been insufficient to observe binding to this low affinity site. HGF ␤-chain has been reported to bind to Met although only in the presence of NK4 fragment from the ␣-chain (32).
Signaling Mechanisms-In principle, the existence of two Met binding sites, one high affinity and one low affinity, in one HGF molecule supports a 2:1 model of a Met⅐HGF signaling complex, analogous to the proposed 2:1 model of the related Ron⅐MSP system (19). As found with HGF, the individual ␣and ␤-chains of MSP bind to their receptor but do not induce signaling (18,38). High affinity binding is also dominated by one of the chains, although in the case of MSP it is the ␤-chain. Compared with full-length MSP, the MSP ␣-chain alone binds with ϳ100-fold reduced affinity. However, biochemical studies have not identified any 2:1 complexes of Met⅐HGF (40). This could be due to the low affinity interaction between Met and HGF ␤; perhaps more stable complexes are only found on cell surfaces with membrane-anchored Met or with additional contributions by heparin-like surface molecules. Further experiments may shed light on this possibility. Alternatively, the HGF ␤-chain may have critical functions in receptor activation beyond those involved in direct interactions with Met that would favor a 2:2 complex of HGF⅐Met.
Upon inspection of intermolecular contacts in the HGF ␤ crystal lattice, we observed that one of the dimer interfaces (green and blue molecules in Fig. 6A)  found in serine proteases and partially blocks the entrance to the "S1 pocket," which has a Gly 667 [c189] at the bottom. B, stereo view of active site regions of HGF ␤ (green) and plasmin (gray). The pseudo-substrate inhibitor Glu-Gly-Arg-chloromethyl ketone from the plasmin structure (yellow) fills the S1 pocket and interacts with its Asp [c189] side chain. The main chain amide nitrogen atoms that stabilize the oxyanion hole (blue spheres) are structurally conserved in HGF ␤. C, location of Met binding site on HGF ␤. Worm depiction of HGF ␤ showing mutated residues with Ͻ20% (red), 20 -60% (orange), 60 -80% (yellow), and Ͼ80% (blue) of wild type HGF pro-migratory activity data in Fig. 2B. The N terminus and three activation domain loops are in black. Residue Lys 649 [c173] would be colored yellow but is disordered in the crystal structure and is not depicted. D, solvent-accessible surface of HGF ␤ showing residues colored as in C. The dotted line depicts the Met binding region from the crystal structure of the complex of HGF ␤ with the Sema/PSI domains of Met (22). HGF ␤-chain dimer interaction is important for Met signaling, it would explain why the single-chain form lacks any biological activity, despite weak Met interactions through its incompletely formed "active site region." In this model the HGF ␤-chain interaction with Met would serve to properly orient the ␤-chain/␤-chain interaction site. Whereas this HGF ␤-chain/␤chain contact may be a crystallization artifact, the presence of the identical contact in the crystal lattice of the HGF ␤⅐Met complex offers some support (22). A dimeric arrangement of HGF ␤ modules in the HGF⅐Met signaling complex would favor a 2:2 model in which two individual HGF⅐Met complexes form a higher order signaling complex consisting of two HGF and two Met molecules (9). If this model is correct, then amino acid changes in this putative dimer interface may adversely affect Met-dependent functions. More extensive studies are needed to unequivocally support or reject this hypothesis.
Comparison of HGF ␤ with Plasmin and Other Proteins-Among proteins with reported molecular structures, the amino acid sequence of HGF ␤ is most homologous with that of plas-min/plasminogen, having 37% identity. Superimposition (33) of the plasmin protease domain 1BUI (41,42) with HGF ␤ yields a root mean square deviation of 1.2 Å for 192 C␣ pairs (out of 227 in our HGF ␤ structure). A structure-based sequence alignment with plasmin shows that HGF ␤ has single amino acid deletions immediately before and after the sequence 505 IGW-MVSLRYR 514 (Fig. 6B)  with two disulfide bonds. In HGF, the ␣-chain to ␤-chain link homologous to plasmin Cys 567 /Cys 685 [c122] has been proposed to form between Cys 487 and Cys 604 [c128] (9). However, this may not be the case since Cys 561 [c79] could also form a disulfide with Cys 487 as suggested recently (22).
The nonenzymatic "catalytic triad" of HGF is shared by the acute phase plasma protein haptoglobin (43), the Trypanosome lytic factor binding protein haptoglobin-related protein (44), and the blood coagulation cofactor protein Z (45). Like HGF, they retain the intact "catalytic triad residue" Asp [c102], but have changes in residues [c57] (Lys) and [c195] (Ala or Gly). MSP, the other plasminogen-related growth factor, also has a nonenzymatic "catalytic triad" in which residues [c57] and [c102] are each changed to Gln. Except for MSP, which uses the ␤-chain for a high affinity interaction with its receptor tyrosine kinase Ron, the role of these other nonenzymatic protease-like domains is not well understood. It is tempting to speculate that their function may involve activation-dependent formation of a protein binding site similar to that found on the ␤-chains of HGF and MSP.
Although zymogen forms of proteases are generally not catalytically competent, some are still capable of binding and even cleaving substrates. For example, single-chain forms of tissue plasminogen activator and urokinase-type plasminogen activator still have catalytic activity, albeit somewhat reduced from their activated forms (46,47). Thus, binding of the zymogenlike ␤-chain of pro-HGF to Met would not be without precedent; our binding data of scHGF ␤ to Met supports this idea.
Another HGF ␤-chain region with the potential for proteinprotein interactions corresponds to exosite I of thrombin (fibrinogen-binding exosite). Exosite I is present in zymogen and active forms of thrombin (48) and contains a positively charged patch centered at the [c70 -80] loop, which is involved in interactions with substrates, cofactors, and inhibitors (35). HGF ␤ also has a positively charged surface in this region, suggesting a potential role in protein interactions. Although two mutational changes introduced in this region (I550-E559 [c70-c77]) did not affect HGF function in cell migration assays, the possibility of this region interacting with cell surface co-stimulatory factors of Met signaling remains.
Conclusions-In conclusion, the results presented herein show that the ␤-chain of HGF contains a new interaction site with Met, which is similar to the active site region of serine proteases. Thus HGF is bivalent, having a high affinity Met binding site in the ␣-chain and a low affinity site in the ␤-chain. Other important interactions may occur between two HGF ␤-chains, two HGF ␣-chains (9), and as found with MSP/Ron (49), between two Met Sema domains (50). Furthermore, heparin also plays a key role in HGF/Met receptor binding. The identification of a distinct Met binding site on the HGF ␤-chain may further the design of new classes of Met inhibitors with therapeutic potential for cancer.