Hepatocyte Growth Factor (HGF)/NK1 Is a Naturally Occurring HGF/Scatter Factor Variant with Partial Agonist/Antagonist Activity*

Hepatocyte growth factor/scatter factor (HGF/SF) stimulates cell proliferation, motility, and morphogenesis by activation of its receptor, the c-Met tyrosine ki- nase. HGF/SF is structurally related to plasminogen, in-cluding an amino-terminal hairpin loop, four kringle domains, and a serine protease-like region. A truncated HGF/SF isoform, designated HGF/NK2, which extends through the second kringle domain and behaves as a competitive HGF/SF antagonist, was previously shown to be encoded by an alternative HGF/SF transcript. In this study, we describe a second naturally occurring HGF/SF variant, HGF/NK1, consisting of the HGF/SF amino-terminal sequence and first kringle domain. This product is encoded by a 2-kilobase alternative transcript containing intronic sequence that was contiguous with exon K1b. Analysis of baculovirus-expressed HGF/ NK1 revealed that this isoform possesses the heparin binding properties of HGF/SF and modest mitogenic and scattering activity relative to HGF/SF. However, at a 40-fold molar excess, HGF/NK1 inhibited HGF/SF-dependent DNA synthesis. HGF/NK1 stimulated tyro- sine phosphorylation of Met, and covalent affinity cross-linking demonstrated a direct HGF/NK1-receptor interaction. These findings establish

Hepatocyte growth factor/scatter factor (HGF/SF) 1 is a multifunctional protein that acts as a mitogen, motogen, and/or morphogen depending on the cellular target and context (1)(2)(3). It stimulates the proliferation not only of hepatocytes but a wide variety of epithelial cells as well as melanocytes and endothelial and hematopoietic cells (4 -7). It promotes the dispersion of certain epithelial and endothelial cells when they are seeded onto plastic surfaces, hence the term "scatter factor" (8,9). Cells can be induced to invade collagen gels in response to HGF/SF (10), where in some instances they give rise to tubular structures in a manner analogous to branching morphogenesis in vivo. (11,12). Paradoxically, some sarcoma and carcinoma cells undergo cytotoxic changes following administration of HGF/SF (13,14). All of these responses appear to be mediated by the c-Met receptor tyrosine kinase (15)(16)(17)(18)(19)(20).
HGF/SF resembles plasminogen in that the two molecules share 38% amino acid sequence identity and several structural motifs (21,22). Each is synthesized as an inactive monomer, which is proteolytically processed at a conserved site to generate a functional, disulfide-linked heterodimer (23)(24)(25)(26). The heavy chain of the dimer (ϳ60 kDa in HGF/SF) is derived from the amino terminus of the precursor and contains multiple kringle domains (four in HGF/SF and five in plasminogen). These ϳ80-amino acid domains have a characteristic folding pattern defined by three internal disulfide bonds and additional conserved residues (27). For plasminogen and tissue plasminogen activator, specific kringles have been shown to participate in protein-protein interactions (28). The light chain of HGF/SF (32)(33)(34), like that of plasminogen, has the structure of a serine protease. However, two of the three amino acid residues of the catalytic triad have been replaced with nonconservative substitutions, and HGF/SF is apparently devoid of proteolytic activity (21).
Previous studies identified an alternatively spliced, HGF/SF transcript in which a kringle 2 exon was joined to an exon with an in-frame termination signal (29,30). Purified HGF/NK2, as the corresponding protein was designated, lacked intrinsic mitogenic activity but specifically blocked the action of HGF/SF in [ 3 H]thymidine incorporation assays (30). Cross-linking studies indicated that HGF/NK2 and HGF/SF competed with each other for binding to Met (30). Thus, HGF/NK2 is a competitive antagonist of HGF/SF mitogenic action. In the course of our analysis of HGF/SF gene expression, we also observed a 2-kb transcript. We report here that this transcript encodes a second naturally occurring, truncated form of HGF/SF. This molecule extends only a few amino acids beyond the first kringle (K1) domain and exhibits biological properties distinct from those of HGF/SF or HGF/NK2. Although it retains the avid heparin binding affinity of HGF/SF and interacts with Met, this novel polypeptide, designated HGF/NK1, exhibits partial agonist/antagonist properties. Thus, the HGF/SF gene encodes three dif-* 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.
The nucleotide sequence(s) reported in this paper has been submitted to the GenBank TM /EMBL Data Bank with accession number(s) U46010.
§ Recipient of a fellowship from the  1 The abbreviations used are: HGF/SF, hepatocyte growth factor/ scatter factor; K1, kringle one of HGF/SF; PCR, polymerase chain reaction; 3Ј-UT, 3Ј-untranslated sequence of pH46; PAGE, polyacrylamide gel electrophoresis; kb, kilobase(s); HPLC, high pressure liquid chromatography. ferent molecules, each unique in its structural and biological properties.

EXPERIMENTAL PROCEDURES
cDNA Library, Probes, and Restriction Enzymes-An M426 cDNA library (31) was screened with probes corresponding to either the heavy or light chain region of HGF/SF, as described previously (30). Restriction enzyme analysis of selected cDNA clones was performed with PstI, XbaI, and EcoRV (New England Biolabs, Inc.) according to the manufacturer's instructions.
Northern Blot Analysis-Poly(A) ϩ RNA was isolated from M426 and SK-LMS-1 cells as described (33). Samples (5 g each) were electrophoresed in a 1% denaturing formaldehyde agarose gel and transferred to nitrocellulose filters. Filters were prehybridized for 2 h at 42°C in Hybrisol (Oncor; 40% formamide, 10% dextran sulfate, 1% SDS, 6 ϫ SSC, and blocking agents) and then hybridized for 15 h in the same solution with [ 32 P]dCTP-labeled randomly primed probes corresponding either to the heavy chain of HGF/SF or to a segment of the 3Ј-UT of pH46. The heavy chain probe was generated by PCR using HGF/SF cDNA as template with sense primer 5Ј-GGACAAAGGAAAAGAA-GAAATACAATT-3Ј and antisense primer 5Ј-ATTGTCAGCGCAT-GTTTTAATTGCACA-3Ј corresponding to sequences 91-117 and 847-873, respectively numbering according to Ref. 22. The 3Ј-UT probe was prepared using pH46 as template with sense primer 5Ј-GGTAAATA-AACCTGAATGCCA-3Ј and antisense primer 5Ј-TTTCTGTGGAAG-CAGGTGCTG-3Ј corresponding to sequences 679 -699 and 1057-1077, respectively (numbering based on HGF/NK1 cDNA sequence submitted to GenBank). Filters were washed twice (30 min each) in 2 ϫ SSC, 0.1% SDS at room temperature and twice (30 min each) in 0.1 ϫ SSC, 0.1% SDS at 50°C and exposed to Kodak X-Omat AR film for 24 h.
Baculovirus Production of Recombinant HGF/NK1 and HGF/SF-The HGF/NK1 coding sequence tagged with BamHI restriction sites was subcloned into the BamHI site of the baculovirus vector pVL941 (Pharmingen) (34). Recombinant baculovirus was produced by cotransfecting Sf9 (Spodoptera frugiperda) insect cells with HGF/NK1-pVL941 and AcNPV (Autographa californica) baculovirus DNA by the calcium phosphate method as suggested by the manufacturer (BaculoGold transfection kit, Pharmingen). Similarly, viral plaque purification, amplification, stock production, and infections were performed according to protocols provided by the manufacturer.
For production of HGF/NK1 protein, 2 ϫ 10 8 Sf9 cells were seeded in a 175-cm 2 T flask containing Sf 900 medium (Life Technologies, Inc.) plus 10% fetal bovine serum (Biofluids). After a 1-h incubation to facilitate cell attachment, the medium was replaced with fresh medium containing recombinant virus at a multiplicity of infection of 10:1. 1 h later, the culture was aspirated, and fresh medium was added. After 3 days, conditioned medium was harvested and either frozen at Ϫ20°C or directly loaded onto a heparin-Poros HPLC column (2.7 ml of bed volume; Perseptive Biosystems) at a flow rate of 5 ml/min. After washing the column with 20 mM phosphate buffer, pH 7.4/0.3 M NaCl, protein was eluted with a linear gradient of 0.3-1.5 M NaCl. Fractions containing HGF/NK1 were identified by immunoblotting.
The full-length coding region of HGF/SF was subcloned into the baculovirus vector pVL941 essentially as described for HGF/NK1. Recombinant HGF/SF protein was generated and purified as outlined for HGF/NK1, using either a heparin-TSK (ToyoHaas) or heparin-Poros column. Typically, HGF/SF preparations were ϳ90% pure and contained varying proportions of both monomeric and heterodimeric forms.
Physical Detection of HGF/NK1 Protein-For immunoblotting, proteins were resolved in 12.5% polyacrylamide-SDS gels under reducing or nonreducing conditions and transferred to Immobilon (polyvinylidene difluoride) filters (Millipore). Blocking and detection of proteins with diluted, GammaBind (Pharmacia Biotech Inc.)-purified antiserum to HGF/SF (4) were as described previously (35). Silver staining of protein resolved by SDS-PAGE was performed with Silver Stain Plus (Bio-Rad) following the manufacturer's protocol.
Biological Assays-DNA synthesis by B5/589 human mammary epithelial cells was measured by [ 3 H]thymidine incorporation as described (36). Epidermal growth factor (recombinant murine) was from Collaborative Research. The scatter assay was performed with a subclone of Madin-Darby canine kidney cells, kindly provided by Dr. Robert Furlong, according to published methods (37).
After lysing cells with Hepes solubilizer buffer (50 mM Hepes, pH 7.4, 1% Triton X-100 (v/v), 100 mM NaF, 2.5 mM sodium orthovanadate, 10 mM sodium pyrophosphate, 2 mM phenylmethylsulfonyl fluoride, 10 g/ml aprotinin, 10 g/ml leupeptin), proteins were immunoprecipitated with a rabbit polyclonal antiserum (50 g/ml) to the carboxylterminal 28 amino acid residues of human Met protein in the absence or the presence of competing peptide (1 g/ml). Immunoprecipitated proteins were pelleted with immobilized protein-G (GammaBind, Pharmacia) and eluted with Laemmli buffer. Following separation in 6% SDS-PAGE, gels were dried and exposed to Kodak X-Omat AR film or exposed to a storage screen and analyzed using a PhosphorImager (Molecular Dynamics). Alternatively, cells were solubilized directly in SDS and boiled for 3 min in the presence of 100 mM ␤-mercaptoethanol, and lysates were subjected to electrophoresis for autoradiography as above.

Isolation and Characterization of Human Fibroblast cDNA
Clones Encoding a New Truncated HGF/SF Variant-In the course of studies that identified HGF/NK2 as the product of a 1.3-kb alternative HGF/SF transcript, we detected several cDNA clones in an M426 library that hybridized to a HGF/SF heavy but not light chain oligonucleotide probe. Restriction enzyme analysis indicated multiple distinct patterns among these clones. One was typified by a 1.2-kb insert that encoded HGF/NK2, as previously reported (30). A different pattern was exhibited by three other clones that contained a 2-kb insert and lacked EcoRV and XbaI sites present in HGF/SF (Fig. 1A). Sequence analysis of one of these clones, designated pH46, revealed a distinct, 210-amino acid truncated version of HGF/SF consisting of the signal peptide and the amino-terminal and K1 domains. The coding sequence of HGF/NK1 terminated immediately downstream of the K1 domain with two additional amino acids and a translational stop codon not found in the corresponding region of HGF/SF. The open reading frame was flanked by 54 base pairs of 5Ј-untranslated sequence previously observed in HGF/SF transcripts as well as a unique ϳ1.2-kb 3Ј-UT containing a polyadenylation signal (AATAAA).
The presence of novel sequence downstream from the K1 domain suggested that HGF/NK1 mRNA resulted from alternative processing of the nascent HGF/SF transcript. To determine the mechanism responsible for the generation of HGF/ NK1, we examined the genomic structure near exon K1b (38,39). PCR analysis of total cellular and pH46 DNA was performed with various pairs of oligonucleotide primers corresponding to different regions of the HGF/NK1 cDNA sequence. As shown in Fig. 1 (A and B), amplification of sequences extending from the 3Ј end of K1 to the middle or downstream end of the 3Ј-UT (Fig. 1B, lanes 3-6) and within the 3Ј-UT (Fig. 1B,  lanes 7 and 8) yielded fragments of the same apparent size when either genomic or HGF/NK1 cDNA was used as template.
To rule out the possibility of plasmid contamination of the genomic DNA preparation, PCR was carried out with primers (P1 and P2) to exons spanning Ͼ10 kb of genomic sequence ( Fig. 1A and Refs. 38 and 39). Although a fragment of the expected length was obtained with pH46 DNA as template, no product was observed with the genomic DNA preparation (Fig.   1B, lanes 1 and 2). These data indicated that the unique sequence at the 3Ј end of the HGF/NK1 cDNA was contiguous with exon K1b in the human genome. This finding was reinforced by the fact that the first 10 nucleotides of the 3Ј-UT were identical to the published intronic sequence (38) located immediately downstream from exon K1b (Fig. 1C). As illustrated in Fig. 1C, the 3Ј-UT of pH46 corresponds to the 5Ј end of the ϳ6.6-kb intron separating exons K1b and K2a in the HGF/SF gene (38).
Although retention of intronic sequence in pH46 raised the possibility that the HGF/NK1 clone might have resulted from incomplete or aberrant RNA processing, this cDNA possessed a polyadenylation signal approximately 30 base pairs upstream from a poly(A) tail. Moreover, using a probe derived from the unique 3Ј-UT, Northern blot analysis of poly(A) ϩ RNA from M426 fibroblasts and SK-LMS-1 cells revealed a 2-kb transcript, corresponding to a faint band seen with an HGF/SF heavy chain probe (Fig. 2). Taken altogether, these results established that HGF/NK1 was a naturally occurring variant encoded by an alternative HGF/SF transcript.
Recombinant Expression of HGF/NK1-In preliminary experiments, the HGF/NK1 coding sequence was placed into an MMTneo vector (40) and introduced into NIH/3T3 cells using standard calcium phosphate transfection methodology. Immunoblot analysis of conditioned medium from transfected cells revealed the presence of a ϳ20-kDa HGF/SF cross-reactive protein that was absent from the medium of control cells (data not shown). This protein bound avidly to heparin-Sepharose and eluted with 0.9 -1.0 M NaCl, comparable with conditions employed for HGF/SF (4). A highly purified preparation was obtained by subsequent sizing chromatography and ion exchange or reverse-phase HPLC (data not shown). However, the amounts recovered (typically a few g/liter conditioned medium) were not sufficient to perform extensive biological analysis.
As an alternative, we expressed the protein in Sf9 insect cells with a baculovirus vector. This approach had proven successful in generating HGF/SF with biological activity comparable with that of the naturally occurring factor (18). The chromatographic, electrophoretic, and immunologic properties of baculovirus-expressed HGF/NK1 matched those of the recombinant NIH/3T3-derived material. Benefitting from the higher level of expression in the baculovirus system, we were able to obtain a highly purified preparation of recombinant HGF/NK1 with a one-step purification process based on heparin affinity chromatography (Fig. 3). The yield was approximately 40 g of HGF/ NK1 from 1 liter of Sf9-conditioned medium. Like HGF/SF and HGF/NK2 (4, 30), HGF/NK1 migrates as a more compact mol-  3) and M426 (lanes 2 and 4) were fractionated, blotted, and hybridized with probes corresponding to the heavy chain of HGF/SF (H) or the 3Ј-UT of pH46 as described under "Experimental Procedures." The dashes on the left indicate transcripts recognized by heavy chain probe; the arrow on the right identifies HGF/NK1 transcript encoded by pH46. ecule in SDS-PAGE when its disulfide bonds are intact rather than reduced (Fig. 3).
Biological Properties of Purified, Recombinant HGF/NK1-HGF/NK1 exhibited mitogenic activity as determined by [ 3 H]thymidine incorporation in B5/589 human mammary epithelial cells (Fig. 4A). However, it was significantly less potent than baculovirus-expressed HGF/SF tested under the same conditions. Even at a concentration as high as 8 nM, HGF/NK1 stimulated only 20 -25% of the maximal DNA synthesis elicited by HGF/SF at 0.5 nM. Moreover, HGF/NK1 behaved as a specific antagonist of HGF/SF in the same assay. A 40-fold molar excess of HGF/NK1 reduced the mitogenic activity of 0.1 nM HGF/SF by ϳ70%, whereas no inhibition of epidermal growth factor activity was observed under the same conditions (Fig.  4B). At high concentrations (5-10 nM), HGF/NK1 also promoted scattering of Madin-Darby canine kidney cells, with an effect comparable with that of HGF/SF at a 50-fold lower molar concentration (data not shown).
Recombinant HGF/NK1 Binds Directly to Met and Stimulates Met Tyrosine Phosphorylation-To establish that the activities of HGF/NK1 were attributable to a direct interaction with the high affinity HGF/SF receptor, a series of covalent cross-linking experiments were performed. Following incubation of B5/589 cells with 125 I-labeled HGF/NK1 and crosslinking agent, cell lysates were immunoprecipitated with a Met-specific peptide antiserum in the presence or the absence of competing synthetic peptide. When the immunoprecipitates were resolved by SDS-PAGE, autoradiography revealed a single major Met peptide-specific band corresponding in size to a complex consisting of [ 125 I]HGF/NK1 and the Met ␤ subunit at a stoichiometry of 1:1 (Fig. 5A). Cross-linking in the presence of either excess unlabeled HGF/NK1 or HGF/SF suggested that the affinity of HGF/NK1 for Met was within an order of magnitude of that of the full-length growth factor (Fig. 5B).
To further study the interaction of HGF/NK1 with its receptor, we examined the ability of this ligand to stimulate tyrosine phosphorylation of Met. Using concentrations corresponding to those employed in the bioassays described above, a significant increase in Met tyrosine phosphorylation was detected in response to HGF/NK1 (Fig. 6). In fact, the intensity of the phosphotyrosine signal induced by HGF/NK1 at 8 nM was comparable with that observed with HGF/SF at 0.5 nM. These results demonstrated that HGF/NK1 and HGF/SF were capable of stimulating a similar level of Met tyrosine phosphorylation, even though HGF/NK1 elicited a weaker mitogenic response. DISCUSSION In the present study, we identified a new naturally occurring truncated form of HGF/SF. This molecule, designated HGF/ NK1, is encoded by a 2-kb alternative HGF/SF transcript, which results from retention of a portion of the intron separat- ing exons K1b and K2a. The biological properties of HGF/NK1 are different than those of the other previously characterized HGF/SF isoforms. In contrast to HGF/NK2, HGF/NK1 has intrinsic agonist activity in the B5/589 DNA synthesis bioassay. However, its potency was considerably less than HGF/SF. Moreover, a 40-fold molar excess HGF/NK1 inhibited the mitogenic activity of HGF/SF. Chemical cross-linking experiments with B5/589 cells demonstrated that HGF/NK1 bound directly to Met, with an apparent affinity estimated to be within an order of magnitude that of HGF/SF. Thus, HGF/NK1 behaves as a partial agonist/antagonist of HGF/SF mitogenic activity on these human mammary epithelial cells.
The different biological effects of HGF/SF and HGF/NK1 cannot be attributed simply to the level of ligand-induced Met autophosphorylation. Under conditions in which they elicited a similar degree of Met tyrosine phosphorylation, HGF/NK1 was less efficient in stimulating DNA synthesis. However, whereas the total phosphotyrosine content of Met following exposure to each isoform was comparable, the distribution of phosphotyrosine residues in the receptor sequence may vary. The major sites of Met autophosphorylation (Tyr 1234 and Tyr 1235 ), which reside in the catalytic domain and are required for tyrosine kinase activity (41,42), presumably are phosphorylated in response to both ligands. However, other tyrosine residues, Tyr 1349 and Tyr 1356 , near the carboxyl terminus of Met are believed to be critical for docking of effector molecules such as phosphatidylinositol 3-kinase, phospholipase C␥, pp60c-src, and the GRB-2-Sos complex (43). Site-directed mutagenesis of the above-mentioned as well as other tyrosine residues in Met had either a positive or negative effect on specific cellular responses (44). Thus, variation in phosphorylation of these residues is likely to have a significant impact on signaling pathways and could account for the unique patterns of activity associated with the different HGF/SF isoforms. In addition, differential regulation of Met tyrosine kinase activity by serine phosphorylation (45)(46)(47) or Met-associated tyrosine phosphatase activity (48) also might contribute to the particular effects of the various HGF/SF isoforms.
Our experiments established that the alternative transcript encoding HGF/NK1 resulted from retention of intronic sequence adjacent to exon 1Kb. Although this mechanism for generating alternative transcripts is unusual, other examples of retained introns have been described (49 -53). Of note, three cases involve transcripts that are expressed at relatively high levels in placenta: a soluble form of the HLA-G antigen (52), a variant of human growth hormone-V encoding a unique 104amino acid carboxyl terminus (51), and an isoform of human gonadotropin-releasing hormone (50). The relative preponderance of cases in placenta led to the hypothesis that this organ may contain specific factors that facilitate export of mRNA with introns into the cytoplasm (52). This may be relevant to the expression of HGF/NK1. Miyazawa et al. detected a relatively abundant ϳ2-kb alternative HGF/SF transcript in placenta (29). Although the extent of truncation had not been fully determined, this mRNA hybridized to the coding sequence of HGF/NK2 but not to probes for K3, K4, or the serine-protease domain. Given its size and hybridization pattern, we surmise that it probably corresponded to the HGF/NK1 transcript described here. Recently, gene targeting experiments revealed that loss of the HGF/SF gene resulted in death in utero due to placental insufficiency (54,55). This finding demonstrated that the HGF/SF gene is crucial for placental development and, in view of the observed expression pattern in placenta, raises the possibility that HGF/NK1 might participate in the development of this organ.
Our results concerning HGF/NK1 activity differ in some respects from the analysis of an artificially engineered version of HGF/NK1 (56). In the latter instance, HGF/NK1 behaved as a pure antagonist of HGF/SF mitogenic activity, showing no agonist activity at concentrations up to 100 nM in an assay using rat hepatocyte primary cultures. It also barely stimulated Met tyrosine phosphorylation when tested at 20 and 100 nM. There are several possible reasons for these differences. The artificially constructed version of HGF/NK1 was expressed in bacteria as a fusion protein containing a 10-amino acid FLAG epitope at its amino terminus to target secretion of the protein into the periplasmic space. The engineered protein also lacked the two amino acid extension to the K1 domain that is present in the naturally occurring isoform. Conceivably, either of these structural differences could affect biological activity by modifying receptor-ligand interactions. Recently, we expressed the naturally occurring HGF/NK1 sequence in bacteria and observed biological activity very similar to that described above for the baculovirus-expressed protein. 2 Thus, bacterial expression per se does not result in a molecule having activities at variance with the results obtained in the present study. However, our bacterial expression strategy included a series of steps to optimize protein refolding and fidelity of disulfide bond formation that could influence the activity of the final product. Independent of differences in HGF/NK1 preparations, the discrepancies in our data relative to the earlier report might be attributable to the cells used in our respective bioassays. For instance, proteoglycan composition, which varies enormously among cells (57) and affects signaling by fibroblast growth factors (58 -60) and HGF/SF (61, 62) 3 might account for the contrasting responses of different cell types to HGF/NK1. HGF/NK1 should prove to be a useful tool in the structurefunction analysis of HGF/SF. Although its affinity for both heparin and Met are similar to that of HGF/SF and it retains biological activity, the smaller size and lack of glycosylation sites render HGF/NK1 particularly suitable for crystallographic and NMR structural analysis. Systematic modification of HGF/NK1 should provide additional insight into the interaction of HGF/SF with its receptors. This, in turn, may lead to the development of more potent, clinically useful agonists or antagonists of HGF/SF signaling.