The Interactions of Hepatocyte Growth Factor/Scatter Factor and Its NK1 and NK2 Variants with Glycosaminoglycans Using a Modified Gel Mobility Shift Assay

Full-length hepatocyte growth factor/scatter factor interacts with both heparan and dermatan sulfates and is critically dependent upon them as cofactors for activation of the tyrosine kinase receptor Met. Two C-terminally truncated variants (NK1 and NK2) of this growth factor also occur naturally. Their glycosaminoglycan binding properties are not clear. We have undertaken a comparative study of the heparan/dermatan sulfate binding characteristics of all three proteins. This has entailed the development of a modified gel mobility shift assay, utilizing fluorescence end-tagged oligosaccharides, that is also widely applicable to the analysis of many glycosaminoglycan-protein interactions. Using this we have shown that all three hepatocyte growth factor/scatter factor variants share identical heparan/dermatan sulfate binding properties and that both glycosaminoglycans occupy the same binding site. The minimal size of the oligosaccharide that binds with high affinity in all cases is a tetrasaccharide from heparan sulfate but a hexasaccharide from dermatan sulfate. These findings demonstrate that functional glycosaminoglycan binding is restricted to a binding site situated solely within the small N-terminal domain. The same minimal size fractions are also able to promote hepatocyte growth factor/scatter factor-mediated activation of Met and consequent downstream signaling in the glycosaminoglycan-deficient Chinese hamster ovary pgsA-745 cells. A covalent complex of heparan sulfate tetrasaccharide with monovalent growth factor is also active. The binding and activity of tetrasaccharides put constraints upon the possible interactions and molecular geometry within the ternary signaling complex.

Hepatocyte growth factor/scatter factor (HGF/SF) 1 is a large (ϳ90 kDa) growth and motility factor structurally related to plasminogen. In common with the latter, it is secreted as a single chain precursor that is subsequently cleaved into two covalently associated chains. This is the first step in the activation of HGF/SF. The larger N-terminal ␣-chain (69 kDa) has pronounced domain structure, being comprised of an N-domain followed by four repeats of a highly cross-linked Kringle domain. By comparison, the smaller C-terminal ␤-chain (34 kDa) is structurally related to the serine protease domain of plasmin (1). Interaction of HGF/SF with Met, its specific tyrosine kinase receptor, is believed to be mediated primarily by the first Kringle domain (K1) (2).
Two alternatively spliced, truncated variants of HGF/SF exist naturally. Both are single chain species that correspond to part of the ␣-chain only. They comprise the N-domain together with either the first one or the first two Kringle domains and are designated NK1 (25 kDa) and NK2 (35 kDa), respectively. Both truncated forms can interact with the Met receptor and were variously described as antagonists or partial agonists of HGF/SF activity, depending upon the cell type and the in vitro assay used (2)(3)(4)(5)(6). However, more recent work involving transgenic expression of NK1 (7) and NK2 (8) has suggested that in vivo they both can function as significant receptor agonists.
In addition to interacting with Met, it is now clear that HGF/SF also interacts with a number of glycosaminoglycans (GAGs), specifically heparin (9), heparan sulfate (HS) (10 -11), and dermatan sulfate (DS) (12)(13). Interactions with HS and DS are likely to be most important physiologically as they are widely expressed as components of proteoglycans, on the cell surface, or in the pericellular matrix and basement membranes of HGF/SF-responsive cells, e.g. epithelia/endothelia. The affinities of HGF/SF for these GAGs, i.e. K D values in the order of 10 Ϫ8 -10 Ϫ9 M (11)(12), may be lower than the corresponding binding to Met, but bearing in mind the high localized concentrations of HS and DS close to the cell surface, these interactions are clearly strong and of physiological relevance.
It has been demonstrated in vitro that interaction with HS/DS constitutes a second necessary activation step for HGF/ SF. Although HGF/SF can bind to Met in the absence of GAGs (14 -16), appropriate activation and signaling through Met require their presence as necessary co-receptors (17)(18)(19)(20). As such, HS and DS appear to be equipotent in their activity, which is * 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 1 The abbreviations used are: HGF/SF, hepatocyte growth factor/ scatter factor; GAG, glycosaminoglycan; HS, heparan sulfate; DS, dermatan sulfate; dp, degree of polymerization (i.e. number of monosac-charides in the oligosaccharide, e.g. dp4 ϭ tetrasaccharide); GMSA, gel mobility shift assay; CHO, Chinese hamster ovary; MAP, mitogenactivated protein; MAPK, MAP kinase; AMAC, 2-aminoacridone; HPLC, high pressure liquid chromatography; Erk, extracellular signalregulated kinase; PAPS, adenosine 3Ј-phosphate,5Ј-phosphosulfate. unusual by comparison with other GAG-dependent growth factors, which tend to be more HS-specific (for review, see Ref. 21).
The molecular basis for the requirement for GAGs in facilitating Met activation by HGF/SF is still unclear. Studies with covalently cross-linked conjugates of HGF/SF and GAG oligosaccharides (either HS or DS), in a 1:1 stoichiometry, have suggested that such a "monomeric" complex is the minimal Met-activating unit (20). Importantly, it also demonstrates that activation in vivo is likely to involve a ternary complex of GAG⅐HGF/SF⅐Met, in which HGF/SF binds simultaneously to both GAG and Met. It has also been demonstrated recently that Met possesses heparin/HS affinity in vitro (22)(23)(24), which raises the additional possibility that the GAG component may also bridge both protein components within the ternary complex.
In common with HGF/SF, both NK1 and NK2 are known to interact with heparin, and with similar affinities (4,16,25). They are also similarly dependent upon GAG binding for their agonist activities (6,25). GAG binding to all three proteins is to be expected, considering what is already known about the likely localization of the GAG-binding site within HGF/SF. Deletion and mutation analyses have consistently implicated the ␣-chain in GAG binding, with no apparent contribution from the ␤-chain (2,26,27). Indeed, most studies have suggested that the N-terminal hairpin loop structure is the major GAG-binding region (16,28). More recently, the x-ray crystal structures of NK1, both alone (29 -31) and in complex with a heparin dp14 oligosaccharide (22), have placed the major heparin interaction within the N-domain. However, there have also been some indications from affinity chromatography of domain-deleted (28) and point-mutated (16) forms of HGF/SF that Kringle 2 may contribute to the heparin binding properties of full-length HGF/SF, and therefore presumably NK2.
Recently it was shown that a heparin oligosaccharide as small as a tetrasaccharide was able to bind with high affinity to HGF/SF, as measured on a biosensor, and to elicit a biological response to HGF/SF in the HaCaT human keratinocyte cell line made deficient in sulfated GAGs by treatment with chlorate (32). This suggests that the necessary GAG interaction with HGF/SF does not involve a particularly extended binding site, in contrast, for example, to the interaction of HS/heparin with fibronectin (33) or endostatin (34). However, because the threedimensional solution structure of full-length HGF/SF is unknown, it is not possible to infer that the bioactivity of HGF/SF is regulated by GAG association with the N-terminal domain alone or to infer whether additional domains, e.g. K2, are required. A likely consequence of the latter would be differences in GAG binding behavior between NK1 on the one hand and NK2 and HGF/SF on the other.
We have undertaken a comparative study of the GAG binding properties of all three protein variants, i.e. NK1, NK2, and full-length HGF/SF, in an attempt to address the following questions. Is the unusual dual HS/DS binding selectivity of HGF/SF a common property of all three proteins? What is the minimal binding oligosaccharide size for both the natural GAG ligands, HS and DS, and is this a common feature for all three proteins? Do these minimal binding sequences activate HGF/ SF? To help address these questions, a significant modification of the gel mobility shift assay (GMSA; described originally by Wu et al. (35)) has been developed. The improved technique is sensitive, rapid, and easy to use. It also does not require radioactivity, and importantly, it can be generally used with all GAG species with equal efficacy and with a variety of GAG-binding proteins.

Methods
HS and DS Oligosaccharide Purification-HS (100 mg) dissolved in 10 ml of 50 mM sodium acetate, 5 mM calcium acetate, pH 7.0, was digested with 0.1 IU of heparinase III at 37°C for 18 h. The digest was then freeze-dried, dissolved in 1 ml of distilled water, and applied to a Bio-Gel P10 gel filtration column (1.6 ϫ 180 cm) eluted with 0.2 M NH 4 HCO 3 at a flow rate of 10 ml/h. Fractions of 2 ml in volume were collected, and peaks corresponding to oligosaccharides ranging from diup to dodecasaccharides were individually pooled on the basis of absorbance at 232 nm as described previously (10). Oligosaccharides were quantified using a molar extinction coefficient at 232 nm of 5200 M Ϫ1 cm Ϫ1 .
DS (50 mg) in 20 ml of 75 mM NaCl, 10 mM sodium phosphate, pH 7.0, was first digested with 0.6 units of chondroitinase ACI at 37°C for 48 h. This DS is considerably resistant to chondroitinase ACI, so after this treatment it was then partially digested with 0.1 unit of chondroitinase ABC at 37°C over a total period of 100 min. Aliquots of 1 ml were removed every 5 min, and enzyme action was stopped by boiling. These aliquots were recombined, freeze-dried, redissolved in 1 ml of distilled water, and applied to two Bio-Gel P10 gel filtration columns (each 1.6 ϫ 180 cm) connected in series. Chromatography was performed at a flow rate of 9 ml/h in 0.2 M NH 4 HCO 3 with collection of 3-ml fractions. Size-separated oligosaccharides were pooled on the basis of absorbance at 232 nm (12). Heparin oligosaccharides were prepared by the fractionation of low molecular weight heparin as described previously (36).
Fluorescence End Labeling of Oligosaccharides-Fluorophore-labeled GAGs were produced essentially as described by Jackson (37). Briefly, heparin, HS, or DS oligosaccharides (10, 100, and 100 g, respectively) were dried down on a centrifugal evaporator, redissolved in 40 l of a solution containing 0.1 M 2-aminoacridone (AMAC) in 85% Me 2 SO, 15% acetic acid (v/v), and incubated at room temperature for 15 min. Subsequently, 40 l of 1.0 M sodium cyanoborohydride (in distilled water) was added to each reaction, and tubes were incubated at 37°C for 16 h. Samples were then concentrated for 1 h on a centrifugal evaporator. Labeled oligosaccharides were recovered from the free AMAC label by precipitation with 9 volumes of ethanol (prechilled to Ϫ20°C) for 15 min, followed by centrifugation at 11,000 ϫ g for 5 min. Precipitates were air-dried and then dissolved in 50% (v/v) glycerol in distilled water containing a trace of Phenol Red to give a final GAG concentration of 1 mg/ml.
Gel Mobility Shift Assay-This electrophoretic assay is essentially as described by Wu et al. (35). Briefly, oligosaccharides were combined with HGF/SF, NK1, or NK2 (at the concentrations indicated in the figure legends) in a total volume of 7 l of phosphate-buffered saline containing 25% (v/v) glycerol for 15 min at room temperature. Samples were then applied to the wells of a non-denaturing 6% polyacrylamide gel (2.6% cross-linker) in 10 mM Tris-HCl, pH 7.4, containing 1 mM EDTA. Electrophoresis was performed at 200 V for 10 min in a Bio-Rad Mini-Protean II system using an electrophoresis buffer comprising 40 mM Tris/acetic acid, pH 8.0, containing 1 mM EDTA. Immediately after electrophoresis, the fluorescent oligosaccharides were visualized on a UVItec gel analysis system coupled to a UVIphoto photographic imager (UVItec Ltd., Cambridge, UK).
When used with a mixture of heparin oligosaccharides, the GMSA was performed instead on a 36% polyacrylamide gel (2.6% cross-linker). Electrophoresis was performed as described above, but for an extended time of 2 h.
Preparation of Cross-linked HGF/SF⅐Heparin dp4 Conjugates-A fully sulfated heparin tetrasaccharide (dp4) was isolated by strong anion exchange-HPLC from a pool of heparin dp4s and then labeled with AMAC (as described earlier). The AMAC-labeled dp4 was then cross-linked to HGF/SF using a zero cross-linking strategy as described previously by Lyon et al. (20). After cross-linking, the sample was applied to a TSK-G4000PW XL (TosoHaas; 7.8 mm ϫ 30 cm) size exclusion HPLC column eluted with 1.5 M NaCl, 10 mM phosphate, pH 7.2, at 0.5 ml/min. The high molecular weight peak corresponding to the HGF/ SF⅐dp4 cross-linked conjugate, i.e. displaying both UV absorbance at 210 nm and fluorescence, was pooled and then desalted on a PD10 column eluted in water.
Erk MAP Kinase and Akt Activation Assays in GAG-deficient Cells-CHO pgsA-745 cells were grown in 12-well cluster plates in 2 ml/well of RPMI 1640 medium supplemented with 5% fetal bovine serum, 2 mM glutamine, penicillin (100 IU/ml), and streptomycin(100 g/ml) in a humidified atmosphere of 5% CO 2 in air at 37°C. Supernatants were replaced with 2 ml of serum-free RPMI for 2 h before addition of the same medium containing HGF together with oligosaccharides at the indicated concentrations. After 5 (for Akt) or 20 (for Erk) min of treatment, the supernatants were removed, and the cells were solubilized in 100 l/well of boiling Laemmli SDS sample buffer (non-reducing for Erk or reducing for Akt). After boiling for 2 min, 20-l aliquots of cell extracts were then electrophoresed on a 15% SDS-polyacrylamide gel (2.6% cross-linker) at 200 V for 40 -50 min. Proteins were electrotransferred to nitrocellulose membranes (Scientific Laboratory Supplies, Nottingham, UK) on a semidry blotter at 200 mA for 1 h. The membrane was blocked for 1 h with either phosphate-buffered saline containing 10% (w/v) nonfat milk powder (for Erk) or Tris-buffered saline containing 0.1% (v/v) Tween 20 and 5% (w/v) nonfat milk powder (for Akt) before being washed three times with either phosphate-buffered saline, 0.1% (w/v) Tween 20 or Tris-buffered saline, 0.1% (v/v) Tween 20, respectively. Membranes were probed with mouse antiphospho-Erk 1/2 (1:1000 dilution) in phosphate-buffered saline containing 0.1% (v/v) Tween 20 and 3% (w/v) nonfat milk powder for 1 h or with rabbit antiphospho-Akt (0.5 g/ml) in Tris-buffered saline, 0.1% (v/v) Tween 20, 5% (w/v) bovine serum albumin for 16 h. Subsequent incubations with horseradish peroxidase-conjugated goat anti-mouse or anti-rabbit IgGs (1:1000 dilution) were for 1 h. Visualization was by enhanced chemiluminescence detection (ECL) according to the manufacturer's (Amersham Biosciences) instructions.

RESULTS
We wished to investigate in parallel the abilities of the three different variant proteins, NK1, NK2, and HGF/SF, to bind with high affinity to different sizes of both HS and DS oligosaccharides and also to ascertain the relationship between the oligosaccharide size required for activation, as opposed to binding, in the HGF/SF system. A sensitive high throughput binding assay was therefore required that could utilize small quantities of the same non-radiolabeled oligosaccharides that were to be used in activation assays in vitro. The gel mobility shift assay introduced recently by Wu et al. (35) is an attractive methodology in that it allows analysis of HS/heparin oligosaccharide-protein interactions that take place under solution conditions of physiological pH and ionic strength. The equilibrium mixture of oligosaccharide⅐protein complexes and free oligosaccharides/proteins can then be easily separated by rapid electrophoresis through a low percentage native (i.e. non-reducing and non-denaturing) polyacrylamide gel. Visualization was by autoradiographic detection of radiolabeled oligosaccharides after drying of the gel. Speed of analysis was thus dependent upon a high specific activity of incorporation of a relatively high energy radioisotope, i.e. 35 S, into the parent heparin/HS species. 35 S was incorporated into the GAG by enzymatic transfer in vitro from the radiolabeled sulfate donor [ 35 S]PAPS using specific HS sulfotransferases.
We substantially modified this procedure by replacing the need for radiolabeling and adopting instead the fluorescence end labeling of GAG oligosaccharides with 2-aminoacridone. This also avoids changes to the internal native GAG structure.
This labeling procedure has been used previously for analytical separations of non-GAG oligosaccharides (37), as well as hyaluronan-and chondroitin/dermatan sulfate-derived disaccharides (38) and larger hyaluronan oligosaccharides (39). GAG oligosaccharides are easily labeled using AMAC. We also found that the ethanol solubility of free AMAC provides an easy postlabeling clean-up and recovery step. AMAC-labeled oligosaccharides can be recovered, essentially free of excess AMAC, by selective ethanol precipitation of the labeled GAG. As AMAC is also neutral under the standard electrophoretic conditions employed here, any residual trace of free AMAC does not enter a polyacrylamide gel sufficiently to cause problems. Also, oligosaccharide mobility is not significantly altered by conjugation with AMAC (not shown). An alternative fluorophore, 8-aminonaphthalene-1,3,6-trisulfonate (ANTS), which has also been used in the past for oligosaccharide labeling, failed to label sulfated GAG oligosaccharides.
AMAC labeling was tested using a range of oligosaccharide size fractions derived from Bio-Gel P10 fractionation of specific enzyme digests of both HS and DS. Labeling efficiencies were broadly similar, and AMAC oligosaccharides of dp Ն 6 mostly migrated as single bands and at approximately the same rate through a porous 6% polyacrylamide gel (Fig. 1A). Oligosaccharides of dp2-4 showed both a relative retardation compared with dp Ն 6 and resolution into multiple bands (Fig. 1A). Although molecular heterogeneity in HS oligosaccharides (and probably DS) increases with increasing fragment length (40), there is a wider charge variation and a lower overall charge/ mass ratio within the smaller dp2-4 populations, which could explain their distinctive behavior. The minimal level of detection of AMAC-labeled oligosaccharides was ϳ10 pmol (Fig. 1B).
AMAC-labeled GAG oligosaccharides were therefore used as ligands for interaction and subsequent GMSA with NK1, NK2, and HGF/SF. In the original published GMSA method the HS/heparin binding properties of antithrombin III (35) and later FGF-1 (41) were investigated. In both cases, the protein⅐oligosaccharide complexes were anionic and thus migrated in the same direction as free oligosaccharides, but with the much larger protein⅐GAG complex, which has a lower FIG. 1. Electrophoresis of AMAC-labeled GAG oligosaccharides. A, electrophoresis of individual AMAC-labeled HS and DS oligosaccharide fractions ranging in size from dp2 to dp12. Samples (approximately 2 nmol of each) were run on a 6% polyacrylamide gel. B, sensitivity of detection of AMAC-labeled HS oligosaccharides of dp2 and dp12 over a dilution range from 1 g down to 8 ng.
charge/mass ratio, having a greatly reduced mobility compared with the free oligosaccharide. However, in the case of NK1, NK2, and HGF/SF, the complexes with GAGs remained slightly cationic. Consequently, a standard GMSA run with migration toward the anode only revealed the depletion, or not, of free oligosaccharides (compared with positive controls in the absence of protein). Such a comparison of the binding of NK1, NK2, and HGF/SF to HS oligosaccharides revealed identical oligosaccharide binding patterns ( Fig. 2A). dp2s failed to bind at all, whereas dp6s were fully bound by an excess of any of the three proteins. In the case of dp4 there clearly was also binding, but here there was a clear selectivity of response within the dp4 population, with binding being restricted to the most mobile (i.e. the most sulfated) component of the dp4s in the case of all three proteins. Results with larger oligosaccharides of dp8 -12 are not shown, as they were fully bound and essentially identical to the results shown with dp6. To confirm direct binding to protein rather than just an indirect inference from the loss of free oligosaccharide, a parallel electrophoresis was performed under reversed polarity to solely visualize the cationic protein⅐oligosaccharide complexes. As can be seen in Fig. 2B, complexes were formed between NK1 and either dp4 or dp6, but not dp2, in agreement with the interpretation of Fig. 2A (identical results were seen with NK2 and HGF/SF; not shown). The presence of size binding selectivity also rules out the possibility that protein binding is being nonspecifically mediated by AMAC, rather than the specific nature of the oligosaccharide itself (this was also confirmed by the ability of unlabeled oligosaccharides to efficiently compete for the binding of AMAC-labeled ones; data not shown). Thus, the minimal size HS oligosaccharide with measurable affinity for all three proteins is a dp4.
We also investigated the binding of heparin to NK1 using a more simplified procedure whereby the protein is incubated with a mixture of AMAC-labeled heparin dp2-10 species. Alone, these oligosaccharides conveniently resolve into a ladder of bands when electrophoresed through a 36% native polyacrylamide gel (Fig. 3). Prior incubation with an excess of NK1 to selectively pull out the binding species gives a radically depleted series of bands, corresponding to the non-binding species. dp6 -10 species are totally removed; dp4 is significantly depleted, whereas dp2 is apparently unaffected (Fig. 3). This confirms that dp4 is the smallest heparin oligosaccharide that binds.
We described previously the DS affinity of full-length HGF/SF (12)(13). However, the ability of NK1 and NK2 to also bind this GAG has not been investigated previously. Preliminary experiments with GMSA using large AMAC-labeled DS oligosaccharides demonstrated that these two proteins also bind DS (not shown). We therefore further investigated the minimal size of DS oligosaccharides required for interaction with all three proteins. Fig. 4 proves that again, all three proteins behaved identically. However, there were differences from the parallel situation with HS. Again, dp2 did not bind, and dp6 was fully bound (as were all larger oligosaccharides; data not shown). However, in the case of DS there was no evidence that dp4 bound. This was also confirmed by the results of a reverse polarity experiment, where oligo-saccharide⅐protein complexes were only formed with dp Ն 6 ( Fig. 4B). Thus the minimal size DS oligosaccharide with affinity is a dp6 for all three proteins.
It is not known whether the unusual dual GAG binding specificity of HGF/SF and now also of its truncated forms results from a single unique GAG-binding site, which is able to accommodate either HS or DS, or from the presence of two separate and distinct binding sites with different GAG specificities. GMSA allows us to answer this question by testing the ability of minimal HS and DS oligosaccharides to compete for binding to the protein. An AMAC-labeled DS dp6 binds fully to NK1. Increasing concentrations of unlabeled HS dp4 compete in a dose-dependent manner for the binding of the fluorescent DS dp6 (Fig. 5).
Minimal size HS and DS binding oligosaccharides were then tested for their ability to function as activators of HGF/SF signaling via Met. For this we assayed, in parallel, the diphosphorylation of the Erk-1/2 MAP kinases as well as the phosphorylation of Akt. These are two independent and essential FIG. 2. GMSA of AMAC-labeled, size-fractionated HS oligosaccharides in the presence of NK1, NK2, or HGF/SF. A, comparative GMSA using NK1 (50 g), NK2 (50 g), or HGF/SF (10 g) incubated with AMAC-labeled HS oligosaccharide size fractions of dp2-6 (1-3 g, respectively, with NK1/NK2 and 0.2-0.6 g with HGF/SF). Samples were run on a 6% native polyacrylamide gel. Bands corresponding to "free" HS oligosaccharides that migrate toward the anode are visualized. B, gel as in A, but using NK1 only and run under reverse polarity. The single band near the top of the gel corresponds to an NK1⅐HS oligosaccharide complex migrating slightly toward the cathode.
FIG. 3. Single run GMSA to determine the minimal size of binding heparin oligosaccharides. A mixture of AMAC-labeled heparin oligosaccharides, ranging in size from dp2 to dp10 (0.1 nmol of each), were incubated with or without an equivalent molar quantity of NK1 (0.5 nmol). Samples were run on a 36% native polyacrylamide gel. The bands correspond to free AMAC-labeled oligosaccharides. downstream signaling events triggered by Met activation that lead to the stimulation of cell proliferation. The sulfated GAGdeficient CHO pgsA-745 cell line was used as it possesses Met but is unresponsive to HGF/SF without an exogenous source of appropriate GAG (20). It was found that in the case of HS oligosaccharides, a dp4 was the minimum size species required for either Erk or Akt activation (Fig. 6), whereas for DS oligosaccharides a dp6 was needed in both cases (Fig. 6). dp4 was also the minimum size required for binding and activity when the more highly sulfated oligosaccharides from heparin were tested (not shown).
We have shown previously that HGF/SF that is covalently cross-linked to intact HS or DS, or to oligosaccharides of dp8 -12 in size, is biologically active in CHO pgsA-745 cells (20). In light of our dp4 binding data we now attempted to cross-link AMAC-tagged dp4s to HGF/SF to test for biological activity of the complex. For this we chose the fractionated and structurally uniform hexasulfated heparin dp4. A population of AMAC-labeled HGF/SF⅐dp4 complexes were obtained, free of noncovalently linked dp4, by size exclusion HPLC in high salt. The apparent molecular mass of the complexes by non-reducing SDS-PAGE (not shown) was consistent with the presence of cross-linked monomers of HGF/SF, probably in a 1:1 stoichiometry with dp4. When tested in CHO pgsA-745 cells with no additional exogenous GAG, these complexes proved to be biologically active, as assessed by downstream phosphorylation of Erk MAP kinase (Fig. 7). The complexes were also much more potent than the same amount of free HGF/SF given in combination with an excess of free dp4 (50 ng of complexed HGF/SF will carry Ͻ1 ng of dp4). DISCUSSION HGF/SF is of considerable clinical interest (i) because of its potential benefit in promoting wound healing and organ regeneration (42) and (ii) as a target in cancer therapy, where upregulation of the HGF/SF-Met system appears to be a hallmark of many tumors and positively correlates with metastasis and poor prognosis (for a recent review, see Ref. 43). The highly truncated forms of HGF/SF, such as the naturally occurring NK1 and NK2, and also the more recently described artificial construct, NK4 (44), have generated interest as potential antagonists of HGF/SF. However, there is still some contention over the extent to which these forms, especially NK1 and NK2, mimic or antagonize the action of HGF/SF and over what functionality may be lost from full-length HGF/SF by concomitant truncation of its ␣-chain and loss of its ␤-chain. Most information on the binding of GAGs and on their role in growth factor activation has been obtained from studies of full-length HGF/SF. In this respect HGF/SF is relatively unusual in that it possesses high affinities for both HS and DS, which are structurally dissimilar GAGs. The two highly truncated forms, NK1 and NK2, have been less investigated in this respect, although both contain the N-domain that is thought to be the major heparin-binding site in HGF/SF. Indeed, the structure of a heparin-binding site in NK1 has been elucidated from an x-ray crystallographic analysis of NK1 complexed with a heparin dp14. However, it is not known whether NK1 has identical GAG binding properties to HGF/SF. We have tried to address this question with a comparative study of the size and struc- FIG. 4. GMSA of AMAC-labeled, size-fractionated DS oligosaccharides in the presence of NK1, NK2, or HGF/SF. A, comparative GMSA using NK1 (50 g), NK2 (50 g), or HGF/SF (10 g) incubated with AMAC-labeled DS oligosaccharide size fractions of dp2-6 (1-3 g, respectively, with NK1/NK2 and 0.2-0.6 g with HGF/SF). Samples were run on a 6% native polyacrylamide gel. Bands corresponding to free DS oligosaccharides that migrate toward the anode are visualized. B, gel as in A, but using NK1 only and run under reverse polarity. The single band near the top of the gel corresponds to an NK1⅐DS oligosaccharide complex migrating slightly toward the cathode.

FIG. 5. Competition between minimal sized HS and DS oligo-
saccharides for binding to NK1. NK1 (5 g) was incubated with AMAC-labeled DS dp6 oligosaccharides (250 ng) in the presence of increasing quantities (0 -5 g) of unlabeled HS dp4 oligosaccharide. Samples were run on a 6% native polyacrylamide gel. Free AMAClabeled DS dp6 is visualized (compare migration of AMAC-labeled DS dp6 alone).
FIG. 6. Size dependence of Erk and Akt activation in CHO pgsA-745 cells by HGF/SF in combination with sized GAG oligosaccharides. CHO pgsA-745 cells were incubated with HGF/SF (20 ng/ml) in combination with oligosaccharide size fractions ranging from dp2 to dp12 derived from HS (125 nM each) or DS (125 nM each). HGF/SF without a GAG addition and HGF/SF with intact GAG (heparin, HS, or DS) were used as negative and positive controls, respectively. After incubation for 5 min (for Akt) or 20 min (for Erk), cell lysates were prepared and run on a 15% non-reducing SDS-polyacrylamide gel. The subsequent Western blot was probed with specific antibodies against diphosphorylated Erk-1/2 (A) or phosphorylated Akt (B). tural specificity of GAG binding and activation of HGF/SF, NK1, and NK2.
To effectively analyze GAG binding to these three proteins we developed a modification of the published GMSA procedure (35). As described originally, this utilized HS/heparin oligosaccharides labeled with 35 S transferred from [ 35 S]PAPS in an in vitro labeling procedure using specific HS sulfotransferases. Although this is useful in certain scenarios, as a widely applicable technique it suffers from the disadvantage that it requires access to, or synthesis of, specialized reagents, i.e. [ 35 S]PAPS and HS sulfotransferase enzymes. The latter also limits its application to heparin/HS unless an even wider range of GAG-specific sulfotransferases are available as reagents. It also involves specific modification of the native GAG structure, which may be unwanted in some cases. The alternative procedure described here utilizes fluorescence tagging of the reducing end with AMAC, which is easy and economical to achieve. It is also sensitive and does not alter the internal GAG structure, and the results of its use can be visualized immediately upon termination of electrophoresis (even without dismantling of the gel cassette). Importantly, the method is equally applicable to oligosaccharides derived from any GAG species, as long as a free reducing terminal is available. It was therefore ideal for comparing the protein binding behavior of both HS and DS oligosaccharides.
In practice, the combination of GMSA with AMAC-labeled oligosaccharides can be used with any protein. In addition to HGF/SF, NK1, and NK2, the procedure also works well with FGF-2 and endostatin (not shown). Likewise, proteins that are known to not bind to GAGs, e.g. angiostatin and an anti-HGF/SF IgG, show no interaction, as expected (data not shown). Whether the protein⅐GAG complexes migrate in the same direction as the free oligosaccharides or in the opposite direction does need, however, to be initially assessed on an individual protein basis. An alternative approach that may be preferred is that GMSA can also be performed on a flat bed electrophoresis system using 2% agarose as the support and a central loading well. Identical results can be obtained (data not shown), but this system does have the advantage that migration in both directions can be visualized simultaneously.
Once a GAG interaction has been demonstrated, the GMSA procedure can be set up in such a way as to provide an efficient and economical screen for identifying the size dependence of GAG binding of any protein. It had been reported already that the smallest heparin oligosaccharide with high affinity for HGF/SF was a dp4 (32). Fig. 3 illustrates an attempt to reproduce this result but with just one GMSA run and using only a single sample of NK1. The 36% gel used here possesses a much more restricted pore size than is normally used for a GMSA assay; however, here a differential mobility of different oligo-saccharides on the basis of size is being encouraged, and also no protein mobility into the gel is required. This experiment confirmed that a heparin dp4 is the smallest binding species. The incomplete depletion of dp4s may reflect an inadequate ratio of NK1 to oligosaccharides for complete binding, coupled to a more efficient competitive binding of the larger (Նdp6) oligosaccharides, although some positive selection may also be occurring (although heparin dp4s are less structurally heterogeneous than HS dp4s). Nevertheless, this example demonstrates the utility of the simplified procedure for a rapid assessment of oligosaccharide size selection by a protein.
Our application of the GMSA assay to the HGF/SF system allowed three fundamental questions to be answered. Firstly, it is clear from the individual behaviors of NK1, NK2, and HGF/SF that all three proteins display essentially the same GAG binding properties; i.e. they all show the same GAG species selectivity in that they bind both HS and DS, and with the same size dependences. Secondly, we have identified the minimal sizes of high affinity oligosaccharides for both of the physiological GAG ligands, HS and DS. In the case of HS, this minimal size species is a tetrasaccharide. Surprisingly, in the case of DS, the minimal size turned out, instead, to be a hexasaccharide. However, all three proteins shared the same GAG binding properties. Lastly, GMSA allowed us to confirm whether or not binding to these two structurally dissimilar GAGS involves a single GAG-binding site with dual GAG specificity or two separate binding sites with differing individual specificities. The use of minimal size-binding oligosaccharides (i.e. a labeled DS dp6 and an unlabelled HS dp4) in a competition binding experiment lessens the scope for misinterpretation arising from indirect steric hindrance occurring between oligosaccharides binding to two separate, but close, binding sites, that could more readily occur with larger ligands. The apparently efficient competition between these two small oligosaccharides (Fig. 5) indicates that both HS and DS species are indeed likely to be binding to one common and unique site on the protein.
The finding that the minimal binding species is a dp4 agrees with the observation that in the crystal structure of the NK1⅐heparin dp14 oligosaccharide complex there is a sequence of four monosaccharide units that provide the major contacts with the protein, specifically within the N-domain (22). At present there is no proven explanation for the larger size of the minimal binding oligosaccharides obtained with DS, compared with HS/heparin, especially as they all appear to be occupying the same binding site on the protein. However, it is interesting that not all HS dp4 species bind (Fig. 2), and thus there does appear to be some structural selectivity operating. If it were only necessary to achieve a certain critical level of overall sulfation for high affinity oligosaccharide binding, then dp4s of both HS and DS should bind similarly. Most of the species within the HS dp4 population are disulfated species, although with heterogeneity in positions of sulfation (data not shown). Similarly the predominant species within the more homogeneously sulfated DS dp4 population is also disulfated (data not shown). This suggests that other more specific binding criteria, e.g. positions of sulfation or the nature of the uronate, are critical. Further work is proceeding to try to elucidate the required structural determinants for binding to a dp4 species. In DS, it may be the case that these criteria can only be satisfied within a dp6 species. It is interesting, in this context, that the crystal structure of the NK1⅐heparin dp14 complex did reveal, in some cases, an additional contact with a fifth monosaccharide (22). This would allow for additional interactions, if necessary, to occur beyond a dp4, but within a dp6.
Additional longer range contacts were also seen with the K1 FIG. 7. Activity of cross-linked complexes between HGF/SF and heparin tetrasaccharides. An AMAC-tagged hexasulfated heparin dp4 was zero-cross-linked to HGF/SF, and the complex then was purified by size exclusion HPLC (see "Experimental Procedures"). The activity of the cross-linked complex was assessed in CHO pgsA-745 cells by assaying downstream phosphorylation of Erk. The activity of the complex (50-ng equivalents of HGF/SF) was compared with that of free HGF/SF (50 ng) in combination with a range of concentrations of free heparin dp4. domain of the adjacent NK1 partner in the NK1 crystal dimer when complexed with the long heparin dp14 (22). If NK1 dimerization, either independently or GAG-mediated, does naturally occur in solution, then stronger interactions may take place with oligosaccharides Ͼ dp4 -6. However, mutations of relevant positively charged residues within the K1 domain did not interfere with heparin binding in solution, although mutations targeted to residues within the N-domain did (22). This clearly indicates that any interactions with K1 are secondary to those with the N-domain.
Interestingly, we have also shown that for both HS and DS, the minimal activatory species also corresponds to the minimal binding species. This concurs with the previous demonstration that heparin dp4s activate HGF/SF to stimulate the proliferation of chlorate-treated (i.e. sulfated GAG-deficient) human HaCaT keratinocytes (32). It has been suggested that the activation of HGF/SF by GAGs may involve dimerization of HGF/SF upon GAG (6,15,25) or GAG bridging of HGF/SF and Met (22)(23)(24). We have demonstrated previously that 1:1 covalently cross-linked complexes of monomeric HGF/SF with large HS or DS oligosaccharides, devoid of free HGF/SF, are biologically active in vitro (20). We now have extended this previous study to show that cross-linked complexes with HS dp4 species are also active. This suggests that GAG-mediated dimerization of HGF/SF may not be the primary mechanism by which GAGs facilitate HGF/SF activity but that the GAG can, and probably does, form part of the activatory complex with Met. Indeed, stabilization of the GAG⅐HGF/SF complex by covalent cross-linking potentiates subsequent receptor activation (Fig. 7). However, the fact that oligosaccharides as small as dp4 are activatory suggests that if Met is also required to interact with the GAG when presented as a GAG⅐HGF/SF complex, then it must be quite constrained in its likely mode of association. A dp4 already bound to HGF/SF would be unable to accommodate another binding protein in a cis configuration, leaving a trans arrangement of proteins as the only likely possibility. Of course, in the context of an intact GAG chain in vivo, this constraint would not necessarily apply, and it is possible that alternative productive modes of complex formation may occur in the two different situations of polysaccharide versus small oligosaccharide.
The similar GAG binding properties of HGF/SF and its truncated forms suggest that the smallest variant, NK1, possesses all that is structurally required to mediate a productive interaction with both HS/DS through a single common binding site and that extension of the protein beyond the first Kringle domain does not alter or add additional GAG binding functionality. This suggests that the previously identified GAG-binding site associated with the N-domain (and thus present in all three proteins) is indeed likely to be the main site of physiological GAG binding. If any other regions of the protein, e.g. K1 or K2, contribute further to GAG binding, then it is likely that they only strengthen the major interaction described here and are unlikely to alter its basic GAG selectivity and dependence. Indeed it is possible that the previously described loss of heparin binding upon complete deletion of K2 from HGF/SF (28) does not in fact reflect the loss of a significant binding determinant located in K2. Instead it may be the result of a major conformational change in the remaining modular protein leading to significant masking of the principal GAG-binding site in the N-terminal domain. However, the reduction in heparin affinity seen upon the more subtle site-directed mutagenesis of four specific basic residues within K2 (16) does suggest that K2 has additional GAG binding capability.
In conclusion, we have identified the binding of minimal length tetra-and hexasaccharides from the physiologically rel-evant HS and DS to all three HGF/SF variant proteins. The functional GAG-binding site is restricted to the N-terminal region of the protein, as encompassed by NK1. The activity of such short oligosaccharides, as either free or cross-linked ligands of HGF/SF, has implications for the structure and mechanism of action of the HGF/SF⅐Met signaling complex.