The N-terminal Domain of Hepatocyte Growth Factor Inhibits the Angiogenic Behavior of Endothelial Cells Independently from Binding to the c-met Receptor*

Hepatocyte growth factor (HGF) is a pleiotropic factor that plays an important role in complex biological processes such as embryogenesis, tissue regeneration, cancerogenesis, and angiogenesis. HGF promotes cell proliferation, survival, motility, and morphogenesis through binding to its receptor, a transmembrane tyrosine kinase encoded by the MET proto-oncogene (c-met). Structurally speaking, HGF is a polypeptide related to the enzymes of the blood coagulation cascade. Thus, it comprises kringle domains that in some other proteins have been shown to be responsible for the anti-angiogenic activity. To check whether the isolated kringles of HGF were able to inhibit angiogenesis, we produced them as recombinant proteins and compared their biological activity with that of the recombinant HGF N-terminal domain (N). We showed that (i) none of the isolated HGF kringle exhibits an anti-angiogenic activity; (ii) N is a new anti-angiogenic polypeptide; (iii) the inhibitory action of N is not specific toward HGF, because it antagonized the angiogenic activity of other growth factors, such as fibroblast growth factor-2 and vascular endothelial growth factor; and (iv) in contrast with full-length HGF, N does not bind to the c-met receptor in vitro, but fully retains its heparin-binding capacity. Our results suggest that N inhibits angiogenesis not by disrupting the HGF/c-met interaction but rather by interfering with the endothelial glycosaminoglycans, which are the secondary binding sites of HGF.

for mature hepatocytes and as a cytokine capable of inducing the dissociation (scattering) of epithelial cells. HGF has now been proven to be a pleiotropic factor that acts on a wide array of target cell types, including epithelial, endothelial, neuronal, and hematopoietic cells (3). HGF promotes cell proliferation, survival, motility, and morphogenesis through binding to its receptor, a transmembrane tyrosine kinase encoded by the MET proto-oncogene (c-met) (4,5). The signaling cascade triggered by the binding of HGF to c-met contributes to complex biological processes, such as embryogenesis, tissue regeneration, and cancerogenesis (3,6). Studies performed in the recent years have shown that HGF is also a potent angiogenic molecule (7)(8)(9)(10).
HGF is a polypeptide structurally related to the enzymes of the blood coagulation cascade. In the cell it is synthesized as a biologically inactive single chain precursor that is then cleaved by specific serine proteases yielding a fully active disulfidelinked heterodimer composed of ␣and ␤-chains (11,12). The ␣-chain consists of an N-terminal domain (N) followed by four kringle modules and mediates the binding of HGF to the c-met receptor (13,14). Additional structure-function studies indicated that the N-terminal domain and first kringle (K1) are primarily involved in this interaction (15). The ␣-chain is also involved in the high affinity binding of HGF to heparin. Based on deletion analysis, the hairpin loop in the N-terminal domain and the second kringle (K2) were shown to be implicated in heparin binding (16,17). Recently x-ray crystal structures were determined for the complexes of heparin with NK1, a naturally occurring HGF variant containing N domain and K1, which suggested that some heparin binding activity is also located within K1 (18). Up to date, no discernible function has been demonstrated for the third (K3) and fourth (K4) kringles of HGF.
The kringle domain is an 80-amino acid triple-loop structure maintained by three intramolecular disulfide bonds highly conserved between different kringle containing proteins (19). Kringles are thought to play an important role in regulating the nature and strength of protein-protein interactions. Several reports have documented the anti-angiogenic activity of isolated kringles of different origin. The most known example is angiostatin, the three-to four-kringle containing fragment of human plasminogen (20,21), and plasminogen kringle 5 (22). More recently the prothrombin kringle-2 domain has been * This work was supported in part by grants from the Ligue Nationale Contre le Cancer and from the Association pour la Recherche sur le Cancer (to G. T.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
ʈ Recipient of a fellowship from the Association Franco-Chinoise pour la Recherche Scientifique & Technique. 1 The abbreviations used are: HGF, hepatocyte growth factor; N, N-terminal domain of HGF; K1, K2, K3, K4, the first, second, third, and fourth kringles of HGF, respectively; NK1, an HGF variant containing N domain and K1; NK2, an HGF variant containing N domain and the first two kringles; IPTG, isopropyl ␤-D-1-thiogalactopyranoside; NF heparin, non-fractionated heparin; LMW, low molecular weight; NHS, normal human serum; FBS, fetal bovine serum; FGF-2, fibroblast growth factor 2; VEGF, vascular endothelial growth factor; PDGF BB, platelet-derived growth factor BB; c-met/Fc, chimerical protein com-posed of the HGF receptor (c-met) extracellular domain fused to human IgG 1 Fc; HUVEC, human umbilical vein endothelial cell; hASMC, human aortic smooth muscle cell; NHEK, human normal epidermal kera-shown to inhibit endothelial cell growth and angiogenesis in the chorioallantoic membrane of chick embryos (23); a recombinant apolipoprotein (a) containing 18 kringle repeats has been implicated in the reduction of tumor microvessel density in transgenic mice (24). The molecular mechanism by which the kringles suppress blood vessel formation is still unknown. We hypothesized that one or more of the kringles of HGF might inhibit angiogenesis and that the mechanism of the anti-angiogenic activity of a kringle domain could be related to its ability to antagonize the HGF binding to c-met. To verify this hypothesis, we produced each of the five HGF ␣-chain individual modules as recombinant proteins and analyzed their binding properties and biological activity in different assays related to angiogenesis.
Recombinant Protein Expression and Purification-The isolated HGF ␣-chain domains were produced using the pET-based bacterial expression system (Novagen). The DNA encoding different HGF polypeptides was generated by PCR using Pwo DNA polymerase and the pGEM/HGF-TM construct as a template (a generous gift of Dr. Pascale Briand at Institut Cochin de Génétique Moléculaire) (25). This construct encodes for an HGF splice variant lacking a pentapeptide FLPSS in the first kringle domain (the amino acids from 162 to 166, according to the sequence deposited in the Swiss-Prot data base under accession number P14210) but which preserves all the biological activity of full-length variant (26,27). The primers used for PCR were: N forward, 5Ј-ATATCATATGCAAAGGAAAAGAAGAAATACAA-3Ј and N reverse, 5Ј-ATATGGATCCTAGTTTCTAATGTAGTCTTTGTTTT-3Ј; K1 forward, 5Ј-ATATCATATGAACTGCATCATTGGTAAAGGA-3Ј and K1 reverse, 5Ј-ATATGTCGACTTCTGAACACTGAGGAATGTC-3Ј; K2 forward, 5Ј-ATATGGATCCAGTTGAATGCATGACCTGCA-3Ј and K2 reverse, 5Ј-ATATGTCGACGTCAGCGCATGTTTTAATTGC-3Ј; K3 forward, 5Ј-ATATGGATCCAACTGAATGCATCCAAGGTC-3Ј and K3 reverse, 5Ј-ATATGTCGACCATATCACAGTTTGGAATTTG-3Ј; and K4 forward, 5Ј-ATATGGATCCAGATTGTTATCGTGGGAATGG-3Ј and K4 reverse, 5Ј-ATATGTCGACACCTTCACAACGAGAAATAGG-3Ј. The purified PCR fragment encoding the HGF N-terminal domain (amino acids from 32 to 127) was cloned into NdeI and BamHI sites of pET15b, in-frame with N-terminal His Tag. The DNA for K1 (amino acids from 127 to 208) was cloned into NdeI and SalI sites of pET21b(ϩ), in-frame with C-terminal His Tag, and the DNA for K2 (amino acids from 209 to 290), K3 (amino acids from 303 to 385), and K4 (amino acids from 390 to 471) were all cloned into BamHI and SalI sites of pET21b(ϩ), inframe with N-terminal T7 Tag and C-terminal His Tag. All PCRgenerated constructs were checked for polymerase fidelity by sequencing (Génome Express, Grenoble, France).
All plasmids encoding HGF kringles, or the plasmid encoding the HGF N domain were electroporated, respectively, into E. coli strain BL21(DE3), or into E. coli BL21(DE3) based strain pRI (kindly provided by Dr. Janne L. Simonsen at Arhus University) carrying two additional plasmids: one encoding a tRNA Ile specific for the rare isoleucine codon AUA and another for the LacZ repressor RP4. Bacteria were grown at 30°C in EZMix 2ϫ YT medium containing 1 M glucose and appropriate antibiotics. Protein expression was induced with 1 mM IPTG for 4 h at 25°C. Bacterial cells were harvested by centrifugation and stored at Ϫ20°C.
Bacterial pellets (5 g wet weight) were resuspended in 100 ml of lysis buffer (50 mM Tris-HCl, pH 7.4, containing 0.2 M NaCl, 10 mM imidazole, 10% glycerol, 1% Nonidet P-40, 2 mM ␤-mercaptoethanol, 1 mM phenylmethylsulfonyl fluoride, and 0.2 mg/ml lysozyme) and incubated for 20 min at room temperature with rotation. The suspensions were sonicated and centrifuged at 10,000 ϫ g for 30 min. All HGF kringles were recovered in the supernatants and purified by cobalt-chelate affinity chromatography on a 2-ml TALON column under non-denaturing conditions. The pellet containing insoluble N was resuspended in 30 ml of isolation buffer (20 mM Tris-HCl, pH 8, containing 0.5 M NaCl, 2 M urea, and 2% Triton X-100) and subjected to three successive rounds of brief sonication, centrifugation (10,000 ϫ g for 10 min), and resuspension. The supernatants from all three steps were pooled and applied to a TALON column equilibrated with isolation buffer. To achieve the N domain renaturation, the column was washed with a 2 to 0 M gradient of urea in IMAC buffer (20 mM Tris-HCl, pH 8, and 0.5 M NaCl) containing 10 mM imidazole. Renatured N was eluted by IMAC buffer containing 0.5 M imidazole.
Further steps were all performed at 4°C. Selected column fractions corresponding to the peaks of purified N, K1, K2, K3, or K4 were pooled and diluted with 10 volumes of refolding buffer (PBS containing 10% glycerol, 1 mM reduced glutathione, and 0.2 mM oxidized glutathione). Following an overnight incubation with stirring, the protein solutions were concentrated using Ultrafree-15 Centrifugal Filter Devices and dialyzed against PBS containing 10% glycerol for 2 days with at least three changes of buffer. The final protein preparations were sterilized using 0.22-M Millex-GV filter units and stored at Ϫ80°C until required. The protein concentration was determined using the BCA protein assay. The purified recombinant HGF-derived proteins were analyzed by SDS-PAGE and by circular dichroism as described previously (28).
Cells-Human umbilical vein endothelial cells (HUVECs) were isolated by collagenase (Roche Diagnostics) digestion (29). Cells were routinely grown in flasks coated with gelatin, in M199 medium containing 2 mM glutamine, 50 units/ml penicillin, 50 g/ml streptomycin, 2.5 g/ml amphotericin B, 7.5% NHS, and 7.5% FBS (regular medium). HUVECs from the second or third passage were used. The human colorectal carcinoma cell line DLD-1, the human osteogenic sarcoma cell line TE 85, and the human pancreatic adenocarcinoma cell line Capan-1 were obtained from the American Type Culture Collection and were cultured in RPMI supplemented with 10% FBS. Human aortic smooth muscle cells (hASMCs) and the corresponding growth medium (SmGM-2), and the human normal epidermal keratinocytes (NHEK) and the corresponding growth medium (KGM) were purchased from Clonetics (BioWhittaker France).
Proliferation Assay-Cells that had been grown until confluence were harvested by trypsinization, and their proliferation was measured by use of a [ 3 H]thymidine incorporation assay. Cells were plated in 24-well plates: HUVECs at a density of 2 ϫ 10 4 cells/well in M199 medium supplemented with 2.5% FBS, hASMCs at a density of 2 ϫ 10 4 cells/well in SmGM-2 supplemented with 1% FBS, NHEKs at a density of 10 4 cells/well in KGM supplemented with 3 mg/ml AlbuMAX (Invitrogen), TE 85 cells at a density 5 ϫ 10 3 cells/well, DLD-1 and Capan-1 cells at a density of 10 4 cells/well in RPMI supplemented with 3 mg/ml AlbuMAX. Cells were allowed to attach for 4 h at 37°C, and proliferation was induced by adding 10% FBS or 10 ng/ml of FGF-2, VEGF 165 , HGF, or PDGF BB. The rising concentrations (10 -5000 nM) of purified recombinant HGF-derived proteins were added to some wells, and HUVECs were incubated for a further 24 h. [ 3 H]Thymidine (1 Ci/well) was added during the last 18 h of incubation. Cells were washed three times with PBS and treated with ice-cold 10% (w/v) trichloroacetic acid for 30 min. The resulting precipitates were solubilized with 0.3 N NaOH (0.5 ml/well), and the incorporated radioactivity was measured in a LS-6500 multipurpose scintillation counter (Beckman Coulter, Fullerton, CA).
Migration Assay-HUVEC migration was evaluated in a modified Boyden chamber assay. Transwell cell culture chamber inserts (BD Biosciences) with porous polycarbonate filters (8-M pore size) were coated with 0.2% gelatin in PBS. HUVECs suspended in M199 medium supplemented with 2.5% FBS were added to the inserts at 4 ϫ 10 4 cells per well. The inserts were placed over chambers containing a chemotactic stimulus (10 ng/ml HGF, FGF-2, or VEGF 165 ), and cells were allowed to migrate for 5 h at 37°C in a CO 2 incubator. For inhibition experiments one of the purified HGF ␣-chain-derived polypeptides (each at 1 M) was added to the lower chamber. The filters were then rinsed with PBS, fixed with 1% (w/v) paraformaldehyde, and stained with hematoxylin of Harris. The upper surface of the filters was scraped with a cotton swab to remove the non-migrant cells. The number of cells per high power field (ϫ200) was recorded. Each experimental point was performed in triplicate, and 10 fields per filter were analyzed.
"In Gel" Three-dimensional Collagen Culture-Type I rat tail collagen composed gels were formed according to the manufacturer's recommendations. Briefly, an ice-cold collagen solution was neutralized by addition of 1 N NaOH and mixed with 10ϫ PBS in the ratio 9:1. HUVECs (passage 2) isolated by trypsin-EDTA treatment and suspended in the M199 medium supplemented with 2.5% FBS, were added at a final concentration of 5 ϫ 10 5 cells/ml in a 1 mg/ml collagen solution. 0.4 ml of this suspension was poured into wells of a 24-well plate (Nunc, Roskilde, Denmark), and the gels were allowed to form at 37°C for 40 min. The gels were then overlaid with 0.5 ml of M199 medium supplemented with 2.5% fetal bovine serum and 10 ng/ml of the growth factors. When indicated, purified recombinant HGF domains (1 M) were added at the same time as the growth factors to the overlaying medium. Cultures were incubated for 48 h at 37°C in a humidified atmosphere containing 5% CO 2 . Digital images were captured using a Kappa CF11DSP charge-coupled device camera (KAPPA opto-electronics GmbH, Gleichen, Germany). Binding and displacement studies with the iodinated proteins were carried out on confluent HUVECs in 24-well dishes, at 4°C. Cells were washed with PBS and preincubated for 30 min with 0.5 ml of binding medium (M199 medium containing 0.1% AlbuMAX). To determine the binding parameters for N, increasing amounts of radiolabeled N were added to HUVECs, with or without a 100-fold molar excess of unlabeled N, and incubated for 6 h. For displacement experiments, constant amounts of radiolabeled HGF (1.5 nM), VEGF 165 (1.2 nM), or FGF-2 (1 nM) were added to HUVECs in the presence of increasing concentrations of unlabeled N for 4 h. At the end of incubation, cells were washed three times with binding medium and solubilized with 0.3 N NaOH. HUVEC-associated radioactivity was determined in a ␥-counter, and the results were analyzed using the program Ligand included in the KELL package (Biosoft, Cambridge, UK). All experiments were done in triplicate. The K d values represent the means Ϯ S.D. obtained from three or four experiments.

Radiolabeled Ligand Binding and Displacement Experiments-HGF
In Vitro Binding Studies by Surface Plasmon Resonance-The binding studies were performed on a BIACORE 2000 instrument. The analytes were routinely diluted in BIACORE running buffer (PBS, pH 7.5, containing 0.005% P-20) completed with 0.2 mg/ml BSA, and injected at 20 l/min. Between binding cycles the coated surfaces were regenerated by two injections of 1 M NaCl. In each case the first flow channel of the sensor chip did not contain any immobilized ligand and served as a reference surface. In one set of experiments, a non-fractionated heparin (NF heparin) was biotinylated with EZ-Link Biotin-LC-Hydrazide as described in a previous study (30) and about 600 RU were captured on the surface of streptavidin-coated SA sensor chips. The c-met/Fc chimera was captured on the surface of a CM5 sensor chip with immobilized protein A (600 RU). HGF and purified HGF domains were injected over immobilized heparin and c-met. In a second set of experiments, HGF and its domains were immobilized on the CM5 sensor chip surface (ϳ1000 RU) using the amine coupling kit according to the manufacturer's instructions. NF heparin, low molecular weight heparin (LMW heparin), and soluble c-met, were used as analytes. Binding curves were obtained over a range of analyte concentrations for each of the analyte/ligand pairs. The data were analyzed with the non-linear least squares algorithm implemented in the BIAevaluation 3.1 software package.
Statistics-Results were expressed as mean Ϯ S.E. Statistical significance was evaluated by analysis of variance followed by Bonferroni/ Dunn analysis.

Production and Purification of Recombinant HGF ␣-Chain
Structural Modules-The HGF kringles were all expressed in E. coli in a soluble form. Recombinant N-terminal domain was mainly accumulated in insoluble inclusion bodies and was easily solubilized with 2 M urea containing 2% Triton X-100. The proteins were purified to homogeneity using a simple one-step procedure on TALON metal affinity column (Fig. 1). They were subsequently subjected to oxidative folding using the glutathione-based oxido-shuffling system. The final protein preparations were analyzed in SDS-PAGE under reducing and nonreducing conditions to ensure the presence of disulfide bonds (Fig. 1). We observed very little or no difference in the mobility of the non-reduced and reduced N, although this protein possesses two disulfide bonds. In contrast, the non-reduced K1 exhibited a substantial mobility shift compared with the reduced sample. Similar results had been previously reported for N and K1 produced in E. coli (31). K2, K3, and K4 also demonstrated an increased mobility under non-reducing conditions. These results indicate that our HGF kringle preparations have a compact conformation under non-reducing conditions and that the disulfide bonds were formed during the refolding procedure.
The circular dichroism analysis of N, K1, K2, K3, and K4 demonstrated that these proteins were folded. The far-UV CD spectra (180 -260 nm) of N, K1, K2, K3, and K4 are all indicative of a high content in secondary structure. A broad band of positive ellipticity can be seen in the 230-to 260-nm region: it most probably originates from oxidized disulfides and aromatic residues involved in the characteristic hydrophobic core of the kringle-type folding. The near-UV CD spectra (240 -340 nm) show bands in the 275-to 290-nm region, which are characteristic of an asymmetric orientation of aromatic residues (Trp and Tyr). These signals are typical for the folded tertiary structure of globular proteins (data not shown).
Effect of Isolated HGF Domains on Cell Proliferation-Angiogenic response involves a surge in endothelial cell proliferation, which could be induced by several growth factors (32). Under our experimental conditions, the highest capacity of induction of HUVEC proliferation was demonstrated for FGF-2, whereas HGF was a very poor mitogen (about 750% stimulation against 25%; Fig. 2, A and B, respectively). Therefore, we first analyzed the effect of isolated HGF domains on the proliferation of HUVECs stimulated with 10 ng/ml FGF-2 ( Fig. 2A). In these experiments none of HGF-derived kringles exhibited any activity, whereas N inhibited HUVEC prolifera- tion in a concentration-dependent manner. The influence of N on HUVEC proliferation could be detected at a concentration of 100 nM (about 10% of inhibition). Depending on the umbilical cord specimen, a maximum inhibition effect (up to 100%) was observed for a concentration of N of 1-5 M.
Because our hypothesis was that kringles could modulate angiogenesis by competitive binding to the c-met receptor, we also studied the effect of recombinant HGF domains on the proliferation of HUVECs induced by HGF. The kringles of HGF had no effect on the proliferation of HUVECs in the presence of HGF (Fig. 2B). In contrast, N significantly inhibited the pro-liferation of HUVECs induced not only by HGF but also by two isoforms of VEGF: VEGF 165 and VEGF 121 (Fig. 2C). Thus, the anti-proliferative effect of N was not specific toward HGF.
We further tested the specificity of the anti-proliferative effect of N toward the endothelial cell type. In the preliminary experiments we showed that HGF was unable to stimulate the proliferation of hASM cells. In contrast, hASMC responded to PDGF BB (Fig. 2D), whereas PDGF BB had no effect on the proliferation of HUVECs (data not shown). The TE 85 and DLD-1 tumor cells did not respond to any of the growth factors tested in this study, and only DLD-1 cells could be slightly stimulated by 10% FBS (Fig. 2D). As reported in Fig. 2D, N had no effect on the proliferation of hASMC induced with PDGF BB, or on the growth of TE 85 and DLD-1 cells in the presence of 10% FBS. In contrast, N inhibited the proliferation of NHEK and Capan-1 cells induced by HGF or FGF-2 (Fig. 2E). No inhibition was observed when these cells were stimulated with HGF or FGF-2 in the presence of kringles. Only K1 demonstrated an inhibitory effect (about 50%) toward Capan-1 cells at higher concentration (5 M, data not shown). Taken together, our data demonstrate that the inhibitory activity of the HGF N-terminal domain is not strictly specific toward the endothelial cells. Nevertheless this activity is restricted to the cell types (presumably of epithelial or endothelial origin), which respond to the GAG-dependent growth factors.
Effect of Isolated HGF Domains on HUVEC Migration-To determine the effect of isolated recombinant HGF domains on endothelial cell migration, an essential component of the angiogenic response, we used a modified Boyden chamber assay. In this assay the cells seeded in the upper chamber of Transwell plates would migrate across the porous filter, if the chemoattractive stimulus is present in the lower chamber. In the preliminary experiments we determined that HGF had a relatively high chemotactic potential (4-to 6-fold stimulation over non-stimulated control), whereas FGF-2 was a less potent motogen (2.5-fold stimulation, data not shown). Therefore, we studied the effect of our recombinant proteins on the HGFinduced migration (Fig. 3A). Only isolated HGF N-terminal domain was capable of inhibiting the HUVEC chemotaxis toward HGF (more than 90% inhibition). Some kringles had a slight inhibitory effect, but this effect was not significant (Fig.  3A). In the migration assay, we also observed that the effect of recombinant N on HUVEC was not specific regarding HGF, as this polypeptide also inhibited the FGF-2 and VEGF 165 -induced HUVEC migration (Fig. 3B).

Isolated HGF N-terminal Domain Inhibits Angiogenesis in
Vitro-The results of two previous sets of experiments indicated that the HGF-derived kringles had no inhibitory activity, whereas the N-terminal domain appeared to be a potential anti-angiogenic molecule. To confirm the anti-angiogenic activity of N, we used a model known as "angiogenesis in vitro." In this model, the HUVECs were suspended in a three-dimensional gel composed of type I collagen and underwent a number of events leading to the formation of a network of capillary-like tubes. The model reproduces several steps of the angiogenesis such as motility, morphogenesis, and the production of lytic enzymes, which makes the penetration of the extracellular matrix possible (33). When grown in a three-dimensional collagen gel in the presence of FGF-2, VEGF 165 , or HGF, HUVECs rapidly underwent morphogenesis forming a network of cords and tubes. After 24 h, HUVECs showed a characteristic elongated phenotype and sprouting extensions. Some cells formed short structures made of several branched and associated elongated cells (data not shown). After 48 h, more cells were involved in structure formation, but some died (Fig. 4). After 72 h, virtually all the cells were dead. When 1 M N was added to HUVECs grown in three-dimensional collagen, no formation of structures was observed, whatever the growth factor and the length of time tested (Fig. 4, shown for 48 h). In the presence of the HGFderived kringles, the capillary-like structures formed normally (data not shown).

Recombinant HGF N-terminal Domain Prevents the Growth Factor Binding to HUVECs-Recombinant HGF N-terminal
domain clearly antagonized the angiogenic response induced by at least three different cytokines. This inhibition could occur at several levels: (i) by inhibiting the interaction with binding sites on the cell surface; (ii) by inhibiting the receptor phosphorylation; and (iii) by inhibiting some signaling pathways downstream of the receptor activation. To clarify the molecular mechanism of N inhibitory activity, we undertook the displacement experiments with radiolabeled HGF, VEGF 165 , and FGF-2 (Fig. 5).
The recombinant K1 of HGF, which had no effect on the HUVEC proliferation, migration, and morphogenesis, was not able to compete with HGF for binding to HUVECs (Fig. 5A). In contrast, N inhibited HGF, VEGF 165 , and FGF-2 binding to HUVECs in a concentration-dependent manner (Fig. 5). In the three cases, the displacement kinetics was best described by a simple bimolecular reaction model. The 27.52 nM, and 68.16 Ϯ 17.5 nM for HGF, VEGF 165 , and FGF-2 displacement, respectively. These values were in good agreement with a K d value of 67.9 Ϯ 11.07 nM, obtained from the binding experiments between radiolabeled N and HUVECs. Scatchard analysis of the direct binding of N demonstrated the existence of a single class of binding sites (Fig. 6). Taken together these experiments indicated that N shares common binding sites with HGF, VEGF 165 , and FGF-2 on HUVECs and that these sites are most probably of a low affinity nature.
SPR Analysis of the Interactions of HGF and Its Domains with Heparin and c-met-To gain insight into the molecular mechanism of the inhibitory activity of N, we compared the binding properties of this polypeptide with those of full-length HGF and of the four other isolated HGF ␣ chain domains, using surface plasmon resonance (SPR) assays.
We first studied the interaction of purified HGF domains with immobilized non-fractionated heparin. Representative results are shown in Fig. 7: all five HGF ␣-chain domains were able to bind heparin, but their binding characteristics presented major differences. The analysis of the binding curves indeed showed that N bound heparin with a 1000-fold higher affinity than any of the four kringles, mainly because its association rate constant k on was much higher than that of the other HGF domains (Table I). N therefore clearly is the principal HGF domain involved in heparin binding.
In a parallel set of experiments, the purified HGF ␣-chain domains were injected over the c-met/Fc chimera immobilized on the surface of CM5 sensor chip. No binding could be observed for the HGF-derived kringles, whereas full-length HGF interacted strongly with c-met (data not shown). Although a slight binding of N to the immobilized c-met could be seen, the high background of nonspecific adhesion made it difficult to determine whether it was significant.
Therefore, we decided to undertake a whole new series of binding assays, in which the roles of the binding partners were swapped, i.e. we analyzed the binding of soluble c-met or heparin molecules to HGF domains immobilized on the surface of CM5 sensor chips. Representative examples of the results obtained are shown in Fig. 8. Only full-length HGF was able to interact with c-met (Fig. 8A), whereas no detectable binding could be evidenced for the five HGF ␣ chain isolated subdomains (data shown only for N, Fig. 8B). The interaction of c-met with immobilized full-length HGF showed a complex kinetic behavior, which could be analyzed taking into account the dimeric (and therefore bivalent) nature of the c-met/Fc fusion. The kinetic parameters of the interaction were determined as follows: k on 1 ϭ 2.97 (Ϯ0.57) ϫ 10 4 M Ϫ1 s Ϫ1 ; k on 2 ϭ 2.11 (Ϯ0.40) ϫ 10 Ϫ3 RU Ϫ1 s Ϫ1 ; k off 1 ϭ 6.22 (Ϯ1.01) ϫ 10 Ϫ3 s Ϫ1 ; k off 2 ϭ 4.31 (Ϯ0.50) ϫ 10 Ϫ2 s Ϫ1 . By analyzing the concentration   dependence of the steady-state responses, a K d Ј of 50 (Ϯ4) nM could be determined for the interaction between c-met and HGF: this value considers the binding process globally without making any assumption on the mechanism of the interaction.
We finally studied the interaction with immobilized HGF and N of two different commercial heparin preparations: NF heparin and LMW heparin (data shown only for NF heparin, Fig. 8, C and D). The kinetic parameters of the interactions are shown in Table II: they were not significantly dependent on the amount of immobilized ligand (data not shown). Interestingly, NF heparin bound ϳ25 times better than LMW heparin, both to HGF and N (Table II): NF heparin showed both higher association rates, which could be due to the fact that its global electrostatic charge is higher (and therefore more favorable) than that of LMW heparin, and lower dissociation rates, which could be explained by the fact that NF heparin is statistically more multivalent (and therefore more avid) than LMW heparin. It should also be noted that the affinity of heparin for N is slightly stronger than that for full-length HGF (Table II): this 4-fold difference, only due to changes in k off , could be linked to slight conformational modifications in the heparin binding site(s) of N, whether this domain is isolated or in the context of full-length HGF. DISCUSSION HGF is a powerful endothelial cell mitogen, motogen, and morphogen (9). Abnormal HGF levels have been shown to contribute to the increased or impaired angiogenesis associated with several human cancers (34 -37) and with some other pathological situations, such as diabetes (38), diabetic retinopathy (39), and arthritis and osteoarthritis (40). Therefore, HGF and its receptor represent very attractive targets for the development of new pro-or anti-angiogenic therapies. Indeed, several reports have demonstrated the effectiveness of HGF-based strategies in the promotion of therapeutic angiogenesis (38,41,42) or in the prevention of pathological angiogenesis (43).
Understanding the relationship between the structure and function of HGF could provide a way of creating potent HGF antagonists and agonists useful for the treatment of angiogenesis associated disorders. At present, the significance of the modular structure of HGF and the mode of interaction of HGF with the c-met receptor are not completely understood. Protein engineering experiments and x-ray crystallography analysis have shown that the c-met receptor and heparin binding regions were located within the ␣ chain, and more precisely, within NK1 (15,16,18). NK1 was further identified as a partial c-met receptor agonist in vivo (44). Recent studies have demonstrated that K1 alone is sufficient to bind to and activate the c-met receptor (45). Interestingly, the addition of the following kringles to NK1 changed the biological activity of the resultant polypeptide. Thus, NK2, another naturally occurring HGF variant composed of N and the two first kringles, antagonized the HGF-induced cell growth in vivo but facilitated cell motility (46). Finally, NK4, an entire HGF ␣ chain, has been shown to compete successfully with the native HGF for the binding to the receptor and inhibit the diverse biological responses to HGF, including the angiogenic response (47)(48)(49). These observations suggested that an anti-angiogenic activity could be located within the HGF kringle domains. This inhibitory activity could depend on the ability to displace HGF from its receptor, and this mechanism could be common with anti-angiogenic kringles of other origin.
To understand the mode of interaction of HGF with its receptor and establish the regions with antagonistic activity, we produced all five HGF ␣-chain domains as recombinant proteins and tested their ability to inhibit the angiogenic response induced in HUVECs. Our results demonstrated that none of the isolated HGF kringles was capable of inhibiting the endothelial cell proliferation, migration, and morphogenesis in a three-dimensional collagen gel. In contrast, the N-terminal domain was identified as a new anti-angiogenic polypeptide.  The inhibitory activity of N was not specific toward HGF, because N successfully inhibited the HUVEC proliferation, migration, and morphogenesis induced by other growth factors, such as FGF-2 and VEGF 165 . Moreover, in contrast to the full-length HGF, the HGF N-terminal domain did not interact significantly with the c-met/Fc chimera in vitro. Therefore, the mechanism of anti-angiogenic activity of N was not dependent on the competitive binding to the HGF receptor. SPR studies showed that none of the isolated domains of the ␣ chain was capable to bind to the c-met receptor in vitro. This indicates that the c-met binding determinant in the native cytokine could involve several modules. Our results seem to be in disagreement with the findings that K1 is able to induce the c-met phosphorylation (45). Nevertheless, in the cited report, the authors showed that the c-met receptor activation in the presence of K1 was strongly dependent on the presence of heparin or heparan sulfate. The absence of K1 binding in our in vitro experiments using purified proteins could therefore be explained by the fact that heparan sulfate is required for the binding of K1 to the receptor in vivo.
We have demonstrated in this work that K1 and three other HGF kringles were able to bind heparin in vitro, albeit with a relatively low affinity. These results are in agreement with the recently published x-ray crystal structure of NK1-heparin complexes, which have shown that heparin is in contact with several amino acid residues in N and in K1 (18). The K d calculated from our SPR experiments for the binding of different HGF kringles to heparin did not differ significantly between the kringles but were three orders of magnitude lower than that of N. From these data K2 does not seem to contribute in a particular way to the HGF interaction with heparin, in contrast with the conclusion drawn from earlier experiments with HGF deletion mutants (16).
The demonstration that N preserved the heparin binding activity of full-length HGF but did not bind the c-met receptor in vitro ruled out the possibility of contamination of the cmet/Fc preparations with cellular glycosaminoglycans (GAGs). Therefore, the HGF binding to c-met/Fc chimera observed in our SPR experiments is GAG-independent. Nevertheless, we obtained a K d value that was substantially lower than that obtained in the earlier in vivo studies (nanomolar against picomolar order). This is consistent with the data showing that HGF is less efficient at inducing the receptor activation in the heparan sulfate-deficient cells (45). Taken together these results indicate that the cell surface GAGs could increase the affinity of HGF toward c-met.
Despite the progress in the understanding of the role of heparin and other GAGs in the growth factor receptor signaling, the mechanisms through which they regulate this process remain incompletely understood. GAGs are believed to be physiologically required co-receptors for a number of growth factors, by providing secondary binding sites that complement the interaction with the specific receptor and strengthen the adhesive forces (50). GAGs also promote dimerization/oligomerization of growth factors. This allows the monomeric ligands such as FGF to induce receptor dimerization, which is widely accepted as a prerequisite for the receptor activation. According to this model, a monomeric receptor remains inactive until the oligomerized ligand binds and oligomerizes the receptor, allowing autophosphorylation of receptor-associated kinases (51). Nevertheless, an increasing number of data suggest that GAGs do not act only by oligomerizing either ligand or receptor and that their role could be to induce the conformational changes in ligand and/or receptor (52,53).
Our results suggest that the anti-angiogenic activity of N is dependent on its high heparin binding capacity. The displace-ment experiments with HUVECs showed that N acts most probably at the level of growth factor binding to the secondary binding sites, presented by cell surface GAGs. Interestingly, our results demonstrated that N inhibited more efficiently the binding of HGF and VEGF 165 than that of FGF-2. This is in good agreement with the available data on the structure of the heparin binding sites of the three growth factors. Thus, the structure of the 55-residue heparin binding domain of VEGF 165 is remarkably similar to the structure of the hairpin-loop region within N domain of HGF, which forms part of the putative heparin binding site (17,54). On the contrary, the folding pattern of the FGF-2 heparin binding site is quite different from that of the two other growth factors (55). Previous studies had also indicated that the heparin sequence that binds FGF-2 differs from the one that binds HGF. Thus, 2-O-sulfates in iduronic acid, N-sulfates in glucosamine, and the carboxylate group in iduronic acid units of heparin are involved principally in the interaction with FGF-2, whereas 6-O-sulfates in glucosamine units seem to be of less importance (55)(56)(57). In contrast, 6-O-sulfates are the major determinants of the heparin-HGF interaction, with 2-O-sulfates playing minor role (18,58). Similarly, the VEGF 165 -heparin interaction involves principally the 6-O-sulfates and N-sulfates in glucosamine with a minor contribution of 2-O-sulfates in iduronic acid (59). Thus, our results are in agreement with the available data on the structural basis of the interaction of FGF-2, HGF, and VEGF 165 with sulfated polysaccharides.
N could also interfere with the pre-existing c-met/GAG interaction. Recent results suggest that the role of GAG in growth factor signaling is more complex than that predicted previously on the basis of ligand-GAG interaction alone. Thus, it has been shown that some growth factor receptors, including c-met, are themselves able to bind with high affinity to GAGs, which presumably could stabilize the pre-existing oligomeric receptor clusters (45,60). Further elucidation of the molecular mechanism of N inhibitory action could contribute to the understanding of the physiological significance of the c-met/GAG interaction and to the design of new anti-angiogenic molecules targeting the GAGs involved in receptor tyrosine kinase signaling.