Interactions of Hepatocyte Growth Factor/Scatter Factor with Various Glycosaminoglycans Reveal an Important Interplay between the Presence of Iduronate and Sulfate Density*

Hepatocyte growth factor/scatter factor (HGF/SF) has a cofactor requirement for heparan sulfate (HS) and dermatan sulfate (DS) in the optimal activation of its signaling receptor MET. However, these two glycosaminoglycans (GAGs) have different sugar backbones and sulfation patterns, with only the presence of iduronate in common. The structural basis for GAG recognition and activation is thus very unclear. We have clarified this by testing a wide array of natural and modified GAGs for both protein binding and activation. Comparisons between Ascidia nigra (2,6-O-sulfated) and mammalian (mainly 4-O-sulfated) DS species, as well as between a panel of specifically desulfated heparins, revealed that no specific sulfate isomer, in either GAG, is vital for interaction and activity. Moreover, different GAGs of similar sulfate density had comparable properties, although affinity and potency notably increase with increasing sulfate density. The weaker interaction with CS-E, compared with DS, shows that GlcA-containing polymers can bind, if highly sulfated, but emphasizes the importance of the flexible IdoA ring. Our data indicate that the preferred binding sites in DS in vivo will be comprised of disulfated, IdoA(2S)-containing motifs. In HS, clustering of N-/2-O-/6-O-sulfation in S-domains will lead to strong reactivity, although binding can also be mediated by the transition zones where sulfates are mainly at the N- and 6-O- positions. GAG recognition of HGF/SF thus appears to be primarily driven by electrostatic interactions and exhibits an interesting interplay between requirements for iduronate and sulfate density that may reflect in part a preference for particular sugar chain conformations.

which sulfated GAG domains may bind both ligand and receptor in close proximity. This crucial involvement of GAGs is akin to its acknowledged role in the formation of supramolecular FGF/HS/FGF receptor signaling complexes (21)(22)(23)(24). In addition to a co-receptor role, PGs may also be important in localizing the signaling complex to the basolateral surface of cells (11,16), restricting diffusion, forming gradients, and protecting HGF/SF from proteolysis.
HGF/SF is relatively unusual among heparin-binding proteins in displaying comparable high affinities for both HS and DS (6,9,10). Both GAGs also show similar potencies as activators of the HGF/SF-MET system in vitro (17). These two GAGs occupy overlapping, if not identical, binding sites within HGF/SF (7). A major heparin-binding site has been located within the N-terminal domain by NMR titration of the truncated NK1 splice variant with heparin (25), by crystallography of a NK1-heparin complex (18), and by heparin-affinity chromatography of a recombinant N-domain (26). The N-domain also binds DS by gel mobility shift assay (GMSA) (26). A hexasaccharide segment of heparin appears to fully occupy the NK1 binding site (18). However, tetrasaccharides are in fact sufficient for binding and activation of HGF/SF in vitro (7,8), and increasing length does not significantly change the binding kinetics (8).
HS/heparin and DS belong to two different structural subgroups of the GAG family. The polymeric backbone of HS/heparin is composed initially of alternating ␤1-4 glucuronic acid (GlcA) and ␣1-4 N-acetylglucosamine (GlcNAc), whereas that of DS is ␤1-3 GlcA and ␤1-4 N-acetylgalactosamine (Gal-NAc). Both then experience extensive post-polymeric modifications; some of which are common to both and some are GAG-specific. In particular, the fate of the uronate residue is identical, in that a proportion of GlcA are converted to ␣-Liduronate (IdoA), its C4 epimer, and some of these can then become O-sulfated at C2. In mammals this generally occurs to a greater extent in HS/heparin than in DS. By contrast, the distinct hexosamines of HS/heparin versus DS diverge even further under the action of specific sulfotransferases. In mammalian DS, GalNAc residues become predominantly O-sulfated at C4. By comparison, HS uniquely experiences N-sulfation of a proportion of GlcNAc residues forming predominantly clustered N-sulfoglucosamines (GlcNS). These then become foci for adjacent GlcA to IdoA conversions (and subsequent 2-Osulfations), as well as O-sulfation at C6 (also occur on some GlcNAc residues) and, rarely, at C3. Occasionally, unsubstituted, positively charged glucosamine residues also occur, although these are generally rare in most HS species, and their mechanism of formation is not well understood (27).
Mature HS and heparin differ principally in the degree of modification by sulfation and epimerization, with major consequences for the overall chain organization. Heparin, being extensively modified to predominantly trisulfated IdoA(2S)-GlcNS(6S) disaccharides, is relatively evenly sulfated throughout. By contrast, HS is modified to a more limited extent, but in a co-ordinated way, yielding a complex co-polymeric structure in which domains of relatively high IdoA content and sulfate density (S-domains), bounded by short transition zones (TZs) of intermediate sulfation and uronate epimerization, are inter-spersed among largely unmodified, non-sulfated, N-acetylated domains (NA-domains) (see 28). Despite these differences, the greater commercial availability of heparin has fostered its widespread experimental use as a structural surrogate of the protein-binding S-domains of HS.
Mature DS chains also demonstrate a co-polymeric domain structure in which IdoA and GlcA residues are predominantly segregated, although in this GAG the relatively homogeneous sulfation of backbone GalNAc residues provides for a lessmarked difference in sulfate densities between the alternating IdoA-rich and GlcA-rich domains (2-O-sulfation of IdoA being the major variable).
The major structural differences between HS/heparin and DS make it difficult to draw obvious conclusions on the specific structural requirements for GAG binding and activation of HGF/SF. Interaction is critically dependent upon the presence of sulfation (11,16), but the influence of positional variations in sulfation around the sugar rings is unclear. In DS, N-sulfates are absent, and we have shown that 2-O-sulfates are not absolutely required (9). In the case of HS, 2-O-sulfation also seems not to be essential (29), and affinity correlates broadly with the level of O-sulfation of GlcNS/NAc (5), although such O-sulfation occurs on C6 in HS, and not C4 as in DS. Conversely, it has been suggested by others that binding of HS/heparin requires the clustering of at least two disulfated (30), or even trisulfated (31), disaccharides.
We have attempted to clarify this situation by screening an expanded panel of natural and modified GAGs, differing both in their density and disposition of sulfate groups and their uronate composition, for their potential as both binding ligands for NK1 and functional co-receptors for HGF/SF in vitro.
sham Biosciences (Chalfont St. Giles, UK). The CHO pgsA-745 and pgsE-606 mutant cell lines were a generous gift from Prof. J. Esko (University of California at San Diego, CA). All cell culture media and sera were from Invitrogen.
Preparation of Ascidian DS-DS species were extracted, as peptidoglycans, from the marine tunicates Ascidia nigra and Styela plicata, and purified as previously described (33,34). Their structures were characterized by disaccharide analyses after digestion with chondroitinase ABC (as described below).
Resonant Mirror Biosensor Analysis of HGF/SF Binding to A. nigra DS-Binding reactions between immobilized A. nigra DS peptidoglycan and soluble HGF/SF were performed in an IASys resonant mirror biosensor (Neosensors, Sedgefield, UK), as previously described for other GAGs (6, 8 -10). DS peptidoglycan (40 g of DS in 100 l of phosphate-buffered saline (PBS)) was biotinylated on its peptide portion by incubation for 2 h with 10 g of succinimidyl-6-(biotinamido) hexanoate (Perbio Science, Cramlington, UK) in 10 l of Me 2 SO. Biotinylated DS peptidoglycan was recovered by DEAE-Sephacel anion-exchange chromatography, and then immobilized by capture on a streptavidin-derivatized planar aminosilane surface, according to the manufacturer's instructions.
All HGF/SF binding reactions were performed in 30 l of PBS, pH 7.2, containing 0.02% (v/v) Tween 20, at 20°C. Surface binding was monitored until it reached a maximum, then, after a rapid wash with PBS, 0.02% (v/v) Tween 20, pH 7.2, the dissociation of HGF/SF was followed. The DS surface was regenerated between binding experiments by washing twice with 2 M NaCl, 10 mM sodium phosphate, pH 7.2, which removed all bound HGF/SF. Three independent sets of binding reactions at five different HGF/SF concentrations (1-10 nM) were performed. Dissociation was measured only at high HGF/SF concentrations (10 nM and 20 nM, five replicates) to minimize rebinding artifacts. Association and dissociation rate constants, k a and k d , respectively, were calculated from each set of association and dissociation curves, using the non-linear curve-fitting FastFit software (Neosensors) supplied with the instrument.
Single Desulfations of GAGs-For de-N-sulfation/re-Nacetylation of heparin, the pyridinium salt of heparin was solvolytically de-N-sulfated in 95% (v/v) Me 2 SO/5% (v/v) methanol at 50°C for 90 min (35). After extensive dialysis, the product was re-N-acetylated using a large excess of acetic anhydride in 0.25 M sodium phosphate buffer, pH 7.5, at 0°C, with maintenance of the pH at Ͼ7 by titration with NaOH (36). The final product was dialyzed against distilled water, concentrated, further desalted on a PD-10 column eluted with water, and then dried (disaccharide analysis revealed a 92% reduction in N-sulfation with no significant loss of 2-O-or 6-O-sulfates).
To effect de-2-O-sulfation of heparin or DS, the GAG was dissolved in 0.2 M NaOH and then freeze-dried (37). The product was re-dissolved in distilled water, neutralized with 20% (v/v) acetic acid, desalted on a PD-10 column eluted with distilled water, and then dried (disaccharide analysis revealed a 96% reduction in 2-O-sulfation of heparin with no significant loss of N-or 6-O-sulfates).
To selectively de-6-O-sulfate heparin, its pyridinium salt was heated at 90°C for 24 h in 90% (v/v) N-methylpyrrolidinone/ 10% (v/v) water (38). After extensive dialysis, the recovered GAG was re-N-sulfated by addition of twice the weight (relative to GAG) of both sodium carbonate and trimethylamine:sulfur trioxide complex (Sigma) and then heated at 55°C for 24 h. After passage through Amberlite IR 120 (H ϩ form) cation-exchange resin at 4°C, the effluent was titrated to pH 9.5 using 0.1 M NaOH, dialyzed against distilled water, concentrated, desalted on a PD-10 column in water, and dried (disaccharide analysis revealed a 90% reduction in 6-O-sulfation with no significant loss of N-sulfates, but with a 29% loss of 2-O-sulfates as an expected side-effect). In all cases the selectivity and extent of individual desulfations were determined by comparing disaccharide analyses of native and modified species, after complete enzymatic digestion (see below).
Isolation of "Transition Zone" Sequences from HS-Ten milligrams of porcine mucosal HS (NV Organon) in 3 ml of 50 mM sodium acetate/0.5 mM calcium acetate, pH 7.2, was digested with 0.1 IU of heparinase I at 22°C until the absorbance at 232 nm reached a plateau. The digest was resolved on a Superdex Peptide HPLC column (Amersham Biosciences) eluted with 0.2 M NH 4 HCO 3 at a flow rate of 0.5 ml/min and monitored at 232 nm using an in-line UV detector. The larger, heparinase I-resistant material (Նdp14) in the void volume was collected, freeze-dried, and then digested with bacteriophage K5 lyase (10 g) in 2 ml of 25 mM Tris acetate, pH 8.5, for 90 min at 37°C, after which time no further digestion was detectable (by periodic monitoring of aliquots by size exclusion on Superdex Peptide column). The remaining TZs were collected from the void volume of a Superdex Peptide column, digested with chondroitinase ABC to remove any contaminating CS/DS, and then recovered again on the Superdex Peptide column.
Disaccharide Analysis of GAGs-CS/DS species were completely digested to disaccharides using 2 mIU chondroitinase ABC in 50 l of 50 mM Tris acetate, pH 7.5, at 37°C for 12 h. HS/heparin species were digested with a mixture of 2 mIU each of heparinases I, II, and III in 50 l of 0.1 M sodium acetate, 0.1 mM calcium acetate, pH 7.0, at 37°C for 12 h. Resulting CS/DS or HS/heparin disaccharides were then applied to either a ThermoQuest Hypersil SAX-HPLC column (4.6 ϫ 240 mm, ThermoQuest Hypersil Division, Runcorn, UK) or a Dionex ProPac PA1 SAX-HPLC column (4 ϫ 250 mm, Dionex, Camberley, UK), respectively. Both columns were equilibrated in water acidified to pH 3.5 with HCl. After a wash with pH 3.5 water, the columns were eluted with linear gradients of 0 -1 M NaCl, pH 3.5, at a flow rate of 1 ml/min over 1 h. Disaccharides were detected by in-line UV absorption at 232 nm, and identified by comparison with elution positions of known disaccharide standards.

Soluble GAGs as Competitors for the Binding of 125 I-HGF/SF
to Cell Surface GAGs-HGF/SF, with its GAG-binding site protected by immobilization on heparin-agarose, was 125 I-labeled using Na 125 I (PerkinElmer Life Sciences) and chloramine T, as previously described (11).
Madin-Darby canine kidney (MDCK) cells were seeded into a 96-well plate (10 4 cells per well) in Dulbecco's modified Eagle's medium containing 5% (v/v) fetal calf serum. After 20 h, the culture medium was removed and the cells washed with PBS. Cells were then incubated at 4°C for 4 h with 0.2 ml/well of Dulbecco's modified Eagle's medium, 1% (v/v) serum, 1% (w/v) bovine serum albumin, containing 125 I-HGF/SF, with or without various GAGs. After five washes with the same solution (without HGF/SF and GAGs), the cells were solubilized in 1 M NaOH, and the released 125 I was measured on a ␥-counter.
MDCK cells lacking cell surface sulfated GAGs were prepared by culturing in the presence of 20 mM sodium chlorate, as previously described (11).
Assay for Activation of ERK MAPKs in GAG-deficient Cells-Trypsinized CHO-pgsA-745 cells were seeded at high density into a 24-well plate in 1 ml/well of RPMI with 5% (v/v) heatinactivated donor calf serum. The adherent monolayer was washed with serum-free medium and then incubated in the same for 3 h at 37°C. HGF/SF, with or without a sub-optimal level of GAG, was then added for 15 min. The culture medium was discarded and the cells lysed in 75 l of boiling, non-reducing Laemmli SDS-sample buffer. Equivalent loadings of lysates were electrophoresed by SDS-PAGE, electroblotted, and the blots probed with a mouse monoclonal antibody against duallyphosphorylated (Thr 183/202 /Tyr 185/204 ) ERK-1/2 MAPKs (Santa Cruz Biotechnology Inc., Santa Cruz, CA) or a rabbit monoclonal antibody against total ERK-1/2 MAPKs (Sigma), as described in Lyon et al. (17). Visualization was by enhanced chemiluminescence (ECL, Amersham Biosciences).
Transwell Cell Migration Assay-Freshly trypsinized CHO-pgsA-745 or MDCK cells in 1 ml of RPMI or Dulbecco's modified Eagle's medium, respectively, with 5% (v/v) donor calf serum, were seeded at 10 5 cells per well into the top of a 12-well Transwell plate (Costar, High Wycombe, UK) with 12-m pore polycarbonate membranes. Bottom chambers received 1 ml of medium containing 5% (v/v) donor calf serum, with or without HGF/SF and GAG. Cell migration after 5 h was quantified as previously described (17).
Gel Mobility Shift Assays-GMSAs were performed on preincubated mixtures of NK1 and equimolar amounts of 2-aminoacridone (AMAC)-tagged GAG species in a total volume of 10 l (7). Samples were run on 1-2% (w/v) agarose gels (substituted for the previously described polyacrylamide gels, but using the same electrophoresis buffers) for 20 min at 100 V. GMSA was also performed in a competitive mode, whereby unlabeled GAG, at various concentrations, was co-incubated with NK1 and a fixed concentration of a standard AMAC-heparin tetrasaccharide.
Size-exclusion Chromatography of NK1-GAG Complexes-NK1 (100 g) was incubated with ϳ2 molar equivalents of GAG in 100 l of 0.25 M NaCl, 50 mM Tris-HCl, pH 7.2, for 30 min. The mixture was then resolved on either Superdex 75 GL or Superdex 200 GL (10 ϫ 300 mM, Amersham Biosciences) size-exclusion HPLC columns eluted with 0.25 M NaCl, 50 mM Tris HCl, pH 7.2, at a flow rate of 0.5 ml/min. Protein elution was detected by in-line UV absorption at 280 nm.

DS Affinity Is Independent of the Position of Hexosamine
Sulfation-A possible rationalization of the similar behavior of HS/heparin and DS may be that, even though they have different polymer backbones, their inherent flexibilities may allow a similar spatial presentation of potentially critical hexosamine O-sulfates, even though they occur on different ring positions. To test this hypothesis, we initially exploited an unusual invertebrate DS species that differs from mammalian DS in possessing hexosamine O-sulfates at the same ring position as HS/heparin (i.e. C6 versus C4; see Table 1). This DS might therefore display substantially altered HGF/SF-binding properties.
DS peptidoglycan from the marine tunicate Ascidia nigra was prepared as described by Pavão et al. (33). In this DS the uronate component is IdoA with no observable GlcA. Disaccharide analysis by strong anion-exchange-HPLC, after complete digestion with chondroitinase ABC, confirmed that it was comprised predominantly of di-2,6-O-sulfated disaccharides  with the remaining disaccharides being mostly mono-6-O-sulfated (Table 1), in accordance with a previous study (33) using paper chromatographic and NMR analyses. Thus, 96% of disaccharides bear 6-O-sulfate groups, and 4-O-sulfates are absent. By comparison, porcine intestinal mucosal DS is predominantly mono-4-O-sulfated with a high IdoA content that is only infrequently 2-O-sulfated (Table 1). Preliminary experiments indicated that 3 H-end-labeled A. nigra DS bound to HGF/SF coupled to Affi-Gel (data not shown). To provide quantitative binding data, we analyzed the kinetics of binding of soluble HGF/SF to immobilized, A. nigra DS peptidoglycan using an optical biosensor (Fig. 1A). Binding was specific and did not occur in the absence of immobilized DS peptidoglycan (data not shown). HGF/SF interaction was described by a one-site binding model, as evidenced by a random distribution of association data around a one-site model (supplemental Fig. S1). A linear relationship was also obtained between the observed on-rate, calculated using a one-site model, and HGF/SF concentration (Fig. 1B). The association binding reaction was not limited by bulk HGF/SF diffusion as a plot of the slope of initial rate was linearly dependent on HGF/SF concentration (Fig. 1C). Analyses of three sets of binding experiments gave a fast association rate constant, k a , of 1.6 (Ϯ0.2) ϫ 10 6 M Ϫ1 s Ϫ1 (correlation coefficient of linear regression of 0.9232). The dissociation rate constant, k d , was independently calculated as the mean of values obtained from analyses of five separate dissociations done specifically at high HGF/SF concentrations to avoid re-binding artifacts (39), and was 0.052 (Ϯ0.0005) s Ϫ1 . The ratio k d /k a yielded an equilibrium dissociation constant, K D , of 33 (Ϯ4) nM, similar to the K D of 20 (Ϯ5) nM previously determined for mammalian DS (9).
Does this high affinity for HGF/SF translate into an ability to promote HGF/SF activation of MET? The MAPKs ERK1/2 (p44/p42) become dually phosphorylated downstream of MET activation. Western blotting for dual phosphorylation of ERK was thus performed to test if A. nigra DS can act as a functional co-receptor for HGF/SF. Cells used in this assay were the mutant CHO-pgsA-745 cells, which are deficient in active xylosyltransferase and thus do not initiate synthesis of endogenous sulfated GAGs (40). Optimal MET activation in these cells has previously been shown to be dependent upon the coaddition of exogenous GAGs, such as HS/heparin or mammalian DS (7,17).
In the absence of HGF/SF and exogenous GAG, there is little activation of ERK-1/2 in CHO-pgsA-745 cells ( Fig. 2A). HGF/SF alone stimulated a low level of ERK phosphorylation, which may represent either an inefficient GAG-independent FIGURE 1. Evanescent wave optical biosensor analysis of HGF/SF binding to A. nigra DS. The binding kinetics of HGF/SF to A. nigra peptidoglycan, biotinylated and immobilized on a streptavidin-derivatized biosensor surface, were measured as described under "Experimental Procedures." Data are from one experiment representative of three; calculated binding parameters are from all three experiments. A, HGF/SF added at different concentrations to a cuvette with A. nigra peptidoglycan immobilized on the surface. Association was followed for 170 s, by which time at least 90% of the fitted binding curve was covered. Dissociation was measured at high concentrations of HGF/SF (10 nM and 20 nM) to avoid rebinding artifacts; only one of five dissociation curves (at 10 nM) is shown for clarity. B, linear relationship between k on , determined from a one-site model, and HGF/SF concentration; the slope yields the k a . C, linear relationship between the slope of initial rate of association and HGF/SF concentration. The distribution of the data around a one-site model for each concentration is shown in supplemental Fig. S1. activation, or an incomplete inactivation of xylosyltransferase catalytic activity in the CHO mutant leading to biosynthesis of a trace amount of sulfated GAG cofactor (40). However, coaddition of increasing amounts of A. nigra DS elicited highly elevated levels of ERK activation, comparable to that achieved using mouse skin DS ( Fig. 2A).
A more demonstrable cellular response was assessed by measuring the chemotactic, transmembrane migration of CHO-pgsA-745 and MDCK cells in the Transwell system. MDCK cells express a normal complement of sulfated GAGs at their cell surface, and thus respond well to HGF/SF alone, although they still show an enhanced migratory response upon further addition of exogenous HS or DS (11). Both cell types showed little migration in the absence of HGF/SF, with or without added GAG. HGF/SF alone induced a limited cell migration, but this was enhanced in a dose-dependent way, in both cell types, by simultaneous addition of exogenous A. nigra DS. The number of migrating cells increased by 100 and 147% using 100 g and 55 g of A. nigra DS with the CHO-pgsA-745 and MDCK cells, respectively (Fig. 2B). By comparison, equivalent amounts of murine DS enhanced migration by 33 and 71%, respectively (Fig. 2B).  (9). Quantification of the cell surface-bound 125 I-HGF/SF under standardized binding conditions, and over a range of soluble competitor GAG concentrations, allows comparative competition curves for different GAGs to be generated. As a control, binding was also assessed using MDCK cells cultured in the presence of 20 mM chlorate (as described in Ref. 11), an inhibitor of sulfation, that therefore lack sulfated, protein-binding motifs in their expressed GAGs. Chlorate-treated MDCK cells did not bind 125 I-HGF/SF to any significant extent (data not shown), demonstrating that cell surface GAGs were predominantly responsible for the binding of 125 I-HGF/SF.

DS Affinity Is Significantly Enhanced by the Presence of Even
An initial competition assay using native mammalian GAGs (Fig. 3A) demonstrated that HS was a better competitor (and thus binder of HGF/SF) than porcine DS, with respective IC 50 values of 750 nM and 25 M (GAG concentrations expressed as molar concentrations of constituent disaccharide repeats). These relative values mirror the difference between the published K D values of ϳ1 nM and 20 nM for the direct binding of HGF/SF to HS (6) and DS (9) species, although valency of binding will also contribute in the present assay. By comparison, a mixture of chondroitin 4-O-and 6-O-sulfate isomers displayed no significant competition over the concentration range studied, confirming previous demonstrations that it does not bind HGF/SF (9,11).
Native and chemically modified DS species were then tested for their competitive activities. A. nigra DS was a slightly more effective competitor than porcine DS (IC 50 (Table 2), slightly reduced competitive ability (IC 50 of 4 M), bringing it closer still to that of mammalian DS (Fig. 4B). However, the competitive binding of the porcine DS was dramatically reduced (to an IC 50 of Ͼ100 M) by a proportionately lesser reduction in its already low level of 2-O-sulfation (reduced from 3.8% to 1.4% (Table 2)) ( Fig. 3B).
An alternative DS species, from the ascidian Styela plicata, contains predominantly di-2,4-O-sulfation (Table 1) (Fig. 3B). As expected, a total de-O-sulfation of DS abolished competitive ability completely (not shown). By comparison with all DS species, porcine HS was a far more effective competitor (IC 50 of 80 nM).
ERK activation assays revealed that all the various DS species promoted HGF/SF signaling to varying degrees, reflecting their relative affinities. Thus the A. nigra DS was particularly potent, but de-2-O-sulfated porcine and A. nigra DS variants still retained activity, albeit weaker (Fig. 3C). A totally de-O-sulfated DS species was inactive (Fig. 3C).    (Table 2). In all cases, where N-sulfates were lost they were replaced by N-acetyl groups. For HGF/ SF-binding comparisons, two alternative methodologies were employed, namely HPLC size-exclusion chromatography (SEC) (24) and GMSAs (7). These two direct binding methods allowed better comparisons over a wider range of affinities than the MDCK competition binding assay. They were also complementary in that weaker affinity interactions not detectable by SEC could still be followed using the more sensitive GMSA. As more protein was required for these techniques, HGF/SF was replaced by the truncated NK1, which could be readily expressed in yeast; this has previously been shown to have essentially the same GAG-binding properties as full-length HGF/SF (7). ERK activation assays conducted in parallel continued to use HGF/SF. Incubation of NK1 with heparin, and rapid chromatography on a Superdex S200 column, yielded a broad, high M r complex that eluted much earlier than free NK1 (monitored by absorbance at 280 nm to detect only protein without contribution from the GAG) (Fig. 4A). Incubation of NK1 with singly de-N-, de-2-O-, or de-6-O-sulfated heparins all continued to lead to formation of apparently stable complexes, although with slightly later elution positions than with native heparin (this may be due to changes in hydrodynamic properties as a consequence of the desulfation and/or the attendant risk of a small degree of depolymerization during the solvolytic desulfation procedures) (Fig. 4A). ERK activation assays indicated that all three singly desulfated heparins remained as efficient at activating HGF/SF as unmodified heparin (Fig. 4B).

HS/Heparin Affinity Is Primarily Determined by Density
Subsequently, heparin was doubly desulfated in two alternative combinations, namely de-N-/de-2-O-sulfation or de-2-O-/ de-6-O-sulfation, to yield purely 6-O-sulfated heparin or N-sulfated heparin, respectively ( Table 2). The third possible combination of de-N-/de-6-O-sulfation was not attempted, because a significant loss of 2-O-sulfates is an inevitable consequence of extensive de-6-O-sulfation. Both the resulting, singly sulfated heparin species now failed to support stable complex formation with NK1 under SEC conditions, i.e. even in their presence, NK1 eluted essentially as free protein (Fig. 5A). GMSA, however, revealed that weak binding of these two modified heparins does occur. In these experiments, the unlabeled heparins competed for the binding to NK1 of a fluorescently tagged, fully sulfated, AMAC-heparin tetrasaccharide (Fig. 5B). shown). In cell assays, both mono-sulfated heparin species elicited similar levels of ERK phosphorylation (Fig. 5C). By comparison, completely desulfated heparin was inactive (not shown). The heparin-like sequences that can occur within the S-domains of HS may represent the highest affinity binding sites for NK1 within HS chains. However, the demonstrable activities of the partially desulfated heparin species suggest that there may be scope for less sulfated, native HS sequences, such as those present in the shorter transition zones (TZs), to be active, although maybe to a lesser extent. These sequences have not traditionally been accessible as isolated fragments, and thus we have used a novel dual-enzyme approach to enrich for them. HS chains were exhaustively digested with heparinase I to cleave at IdoA(2S) residues, thereby largely fragmenting the 2-O-sulfated core of the S-domains (Fig. 6A). The larger, resistant fragments, primarily expected to comprise NA-domains bounded on either side by TZs terminating in small S-domain remnants, were then digested with bacteriophage K5 lyase. This enzyme cleaves specifically within the extended GlcA-GlcNAc sequences of the NA-domains (Fig. 6A). The resulting fragments resistant to both enzymes were recovered by HPLC gel filtration (data not shown). These essentially comprise TZs of variable length, flanked on one side by a short NA-domain remnant (probably up to three GlcA-GlcNAc disaccharides in length, as the minimal effective substrate size for K5 lyase is at least four disaccharides long (28)) and on the other side by a short S-domain remnant that may terminate in a 2-O-sulfated uronate (Fig. 6A). Analysis of their sulfated disaccharides revealed a slightly reduced level of N-sulfation (73% versus 81%), a slightly elevated 6-O-sulfate content (38% versus 34%), and a significantly reduced 2-O-sulfation (22% versus 34%), compared with the intact HS. A marked change was the noticeable switch of 6-O-sulfation from N-sulfated disaccharides (reduced from 19% to 13%) to N-acetylated disaccharides (increased from 16% to 25%). This relative enrichment of GlcNAc 6-O-sulfates, probably occurring in alternating N-sulfated/N-acetylated sequences, combined with a depletion of 2-O-sulfates, would be diagnostic of TZ structures. These excised TZs efficiently bound NK1 by GMSA (Fig. 6B) and also activated HGF/SF, although less strongly than the intact HS (Fig. 6C).
Iduronate Is Critical for Interaction, Except at High Sulfate Densities-It is notable that GAGs previously known to bind and activate HGF/SF are those that contain some IdoA. Indeed the presence of IdoA is what distinguishes an active de-2-Osulfated DS from an inactive CS. Is the presence of IdoA critical, in either CS/DS or HS/heparin, irrespective of sulfate density?
Chondroitin sulfate-E (CS-E) is a GlcA-containing CS species that contains primarily (ϳ60%) disulfated [GlcA-GalNAc (4,6S)] disaccharide repeats, with some additional 3-O-sulfation of GlcA (Table 1) (41). This more highly sulfated CS species showed a weak affinity for NK1 by SEC, although weaker than DS (Fig. 7A). NK1 in the presence of CS-E eluted slightly earlier than free NK1, although noticeably later than when chromatographed together with porcine mucosal DS (Fig. 7A). However, the extensive smear of NK1 preceding the main NK1 peak (but not extending into the void volume where free CS-E elutes) also suggests that complexes formed initially but were prone to dissociation during the chromatography run. CS-E also potentiated HGF/SF-mediated ERK activation, although to a lesser extent than HS (supplemental Fig. S2A).
The K5 polysaccharide, obtained from the capsule of the Escherichia coli K5 strain, is identical in backbone structure to the nascent HS precursor, i.e. repeating disaccharides of [GlcA-GlcNAc], but completely lacks sulfation. However, it can be chemically sulfated in vitro at the N-and/or various O-positions to give species resembling HS/heparin in sulfation though still remaining completely devoid of IdoA residues. A fully N-sulfated K5 polymer (Table 1) did not complex with NK1 under SEC conditions (Fig. 7B) or potentiate HGF/SF activity (supplemental Fig. S2B). In contrast, an N-sulfated and highly O-sulfated K5 derivative (Table 1) (42) formed stable complexes with NK1 (Fig. 7B) and effectively potentiated HGF/SF activity, being more active than HS (supplemental Fig. S2B).

The Uronate Residue Is Itself Dispensable If Sulfation Is
Sufficiently High-We also tested the properties of a GAG and two sulfated polysaccharides that naturally lack uronate residues. Keratan sulfate (KS) is the sole member of the GAG family that lacks uronate residues, being a polymer of [Gal-GlcNAc] disaccharide repeats in which sulfation can occur on C6 of GlcNAc, and additionally on C6 of Gal. In corneal KS (Table 1) ϳ38% of disaccharides are disulfated (43), and these tend to be clustered (44). Corneal KS did not show any indication of binding to NK1 by SEC (data not shown). However, it did show a weak ability to compete with an AMAC-labeled heparin tetrasaccharide for binding to NK1 by GMSA (Fig. 8A) and also a weak ability to potentiate HGF/SF activity in the phospho-ERK assay (supplemental Fig. S2C). Dextran sulfate is a homopolymer of ␣-1,6-Dglucose that is then chemically sulfated to varying degrees. A preparation containing 3-4 sulfates per disaccharide bound strongly to NK1 as revealed by SEC (Fig. 8B) and was more active than HS in the phospho-ERK assay (supplemental Fig.  S2D). Similarly, pentosan polysulfate, a complex polymer based upon a ␤-1,4-D-xylose backbone with approximately two sulfates/disaccharide, also bound NK1 by SEC (Fig. 8C).

DISCUSSION
The structural basis for GAG co-receptor recognition by HGF/SF has been very unclear, if not contradictory, and the activation specificity is even less well understood. It had been suggested that clustering of at least two trisulfated disaccharides, i.e. own work had indicated that removal of N-sulfates from HS and their replacement by N-acetyls did not significantly compromise HGF/SF binding by affinity chromatography (5), and likewise the complete lack of 2-O-sulfates in HS derived from the HS 2-O-sulfotransferase Ϫ/Ϫ mouse did not impair either HGF/SF binding or activity in vitro (29). This confusing picture also fails to accommodate the binding and activating properties of mammalian DS (7,9,17), a GAG that has a different polymer backbone from HS, lacks N-sulfates, is frequently low in 2-Osulfates, and is O-sulfated on hexosamine at C4 not C6. The present study was initiated to try and resolve these structural recognition issues, which have potential mechanistic implications and will also be invaluable in the further design of GAG mimetics as inhibitors of HGF/SF activity (45). It is clear that the highest affinity for HGF/SF is displayed by heparin/HS, which is probably a consequence of the presence of trisulfated disaccharides in these GAGs. Experimentation with modified species demonstrates that reducing sulfation incrementally from three sulfates down to two sulfates, and then finally to one sulfate per disaccharide repeat, leads to a progressive weakening of affinity for NK1. Nevertheless, disaccharide monosulfation of this GAG backbone is sufficient to support weak binding and activation; in this respect it is broadly comparable to the behavior of mammalian DS, which is also largely monosulfated. However, most revealing of all, is that the binding and activity of less-sulfated heparin/HS species is relatively independent of the positioning of sulfation, suggesting that no single sulfate position is critical for recognition and activity, which helps to explain many of the previous observations on heparin/HS binding (5,29,30). The newly demonstrated ability of HGF/SF to bind and be activated by the distinctive, di-2,6-O-sulfated DS from A. nigra demonstrates that there is also similar flexibility in recognition of different positions of hexosamine sulfation upon a DS backbone.
The similar binding parameters for mammalian and A. nigra DS species superficially suggests that the predominantly disulfated structure of the latter does not confer any advantage over its predominantly monosulfated mammalian counterpart (unlike the difference between mono-and disulfated heparin species). However, this mammalian DS does contain a small proportion of disulfated disaccharides, which are likely to be clustered. It is notable that removal of most of these 2-O-sulfates (thereby moving closer to a true monosulfated species) significantly reduces HGF/SF affinity, whereas removal of a large part of the high 2-O-sulfation in A. nigra DS has very little impact. This suggests that only relatively few disulfated units within a predominantly mono-4-O-sulfated chain may be sufficient to constitute a higher affinity recognition site, especially as the minimum required is only a tetrasaccharide in length (7,8). Because HGF/SF is a large, modular, and elongated (46) growth factor, it is possible that relatively few HGF/SF ligands can be physically accommodated, simultaneously, on a single GAG chain, irrespective of the density of potential binding sequences it contains. Additional 2-O-sulfation beyond a sufficient level to provide a minimal number of binding sites may therefore be largely superfluous. A useful comparison here is with the endothelial dermatan sulfate proteoglycan endocan, which efficiently binds and activates HGF/SF even though the extent of GlcA to IdoA conversion within its single CS/DS chain is only around 6% (10), which is probably enough, if clustered, to provide a single binding site.
Interestingly, S. plicata DS, which is primarily di-2,4-O-sulfated, binds HGF/SF similarly to the 4-O-sulfated mammalian DS, but slightly weaker than di-2,6-O-sulfated A. nigra DS. This may suggest that, in general, sulfation of hexosamine at C6 may be slightly preferable to C4, possibly giving an additional binding advantage to HS/heparin over mammalian DS. Sulfation patterns can affect polysaccharide conformation and, consequently, the disposition in space of the sulfate groups (47). One notable difference between DS and heparin/HS is the hexosamine N-sulfate present in the latter, which may play a structural role, if not an essential direct role, in binding. GlcNS can form a hydrogen bond to the OH group at C3 of a neighboring IdoA, thereby rigidifying the glycosidic bond. This may be an additional contributory factor to the higher affinities of heparin/HS over DS species, although this needs further detailed investigation.
If relative affinity for HGF/SF is determined more by sulfate density than position, then do uronate residues contribute in a qualitatively different way from sulfate groups, or do their carboxyl groups just contribute additional anionic charge? A common feature of those GAGs that bind and activate HGF/SF is their IdoA content. IdoA clearly confers a binding advantage when sulfation density is relatively low (cf. DS versus CS). However, there is no absolute requirement for IdoA, and GlcA will suffice, if sulfation levels are high, as evidenced by the HGF/SFbinding and -activating properties of CS-E and, to a greater extent, a highly sulfated K5 polysaccharide, both of which contain purely GlcA and have Ն2 sulfates per disaccharide. However, the fact that CS-E (predominantly disulfated [GlcA-GalNAc(4,6S)]) still binds more weakly than disulfated DS species (containing Ido(2S) residues) again emphasizes the importance of the IdoA ring at comparable sulfate densities. The behavior of non-GAGs such as dextran sulfate and pentosan polysulfate show that even the presence of a uronate residue (or even a hexosamine) is not strictly required, should sulfate density be sufficiently high. This would explain previously published observations that highly sulfated malto-oligosaccharides bind and activate HGF/SF (13). Interestingly, KS, which is the sole GAG that naturally lacks uronates, but which can possess clusters of disulfated disaccharides, is a very weak binder and activator.
The critical role of IdoA is likely to relate to its flexible conformation, compared with GlcA, as well as its far greater tendency to bear additional sulfation at C2. Depending upon the positioning of IdoA within a GAG sequence, its possible 2-O-sulfation, and also the sulfation of adjacent hexosamines, IdoA residues can switch readily between almost equi-energetic 1 C 4 , 2 S 0 , and 4 C 1 conformers; by comparison GlcA adopts purely the 4 C 1 conformation (for reviews see Refs. 48,49). These dynamics lead in particular to marked changes in 2-O-sulfate orientation that may enhance GAG accommodation to the HGF/SF binding site. In the case of more highly sulfated polymers, IdoA may not be necessary, because a sufficient proportion of the sulfates present may be able to adopt an appropriate topological arrangement for effective binding purely through chain flexibility around the glycosidic bonds.
Physiologically, heparin is unlikely to be the major HGF/SF ligand. However, it is clear that, within the structurally heterogeneous HS polymer, the highest affinity binding sites for HGF/SF will likely reside in the most anionic, trisulfated, heparin-like sequences within the S-domains, although clearly various disulfated disaccharide sequences are adequate to support strong binding and activity. In support of this, the CHO pgsE-606 mutant cell line, which synthesizes HS with shorter and less-sulfated S-domains (50), is not compromised in its HGF/SF responsiveness (data not shown). Our study also reveals that the less-sulfated TZs in HS, which essentially comprise mixed mono-and disulfated disaccharides, also possess binding potential and activity. Thus HGF/SF may be able to initially bind anywhere within a sulfated domain from where it could then process to the highest affinity site within that domain. However, the significant attenuation of HGF/SF activity in cells transfected with HSulf-1 (51), an extracellular sulfatase that removes 6-Osulfates predominantly from trisulfated disaccharides in heparin/HS (52), suggests that there might be a significant physiological distinction between GAG sites of different sulfate density in some cellular contexts.
In most mammalian tissues both HS and DS will co-exist, although they are likely to have differential localizations and be engaged in substantially different repertoires of protein interactions. Purely in terms of the apparent correlation between GAG affinity and sulfate density, HGF/SF would be expected to preferentially bind to HS. The predominant plasma membrane integration of HS would favor its cofactor involvement in HGF/SF-MET signaling. However, although dermatan sulfate proteoglycans are predominantly extracellular, this may not in itself be an impediment to effective cofactor activity. We have previously shown that chloratetreated MDCK cells can respond to HGF/SF presented on an immobilized, extracellular HSPG substratum (11,16), demonstrating that cis-activation of MET is not an absolute mechanistic necessity and that trans-activation is feasible (as has been demonstrated for HS in vascular epidermal growth factor signaling (53)). Also, the potent activity of soluble DS chains in vitro (Figs. 3 and 4C) may be important during tissue remodeling, or invasive tumor growth, where extracellular matrix components are being degraded and solubilized. For comparison, it has been shown that substantial DS mobilization into wound fluid after injury can generate an effective soluble cofactor for FGF-2 activity (54), and HGF/SF is also known to be a very potent growth factor during wound healing. However, it is also the case that extracellular DS may primarily fulfill other roles by affecting diffusion paths, tissue storage, and bioavailability of HGF/SF.
In the above context, the interaction and activation of HGF/SF in vitro by non-mammalian CS/DS structures also has potential physiological relevance in mammalian systems. It is becoming increasingly clear that mammalian CS/DS species often contain a minor proportion of variable, more highly sulfated, and often clustered, structures equivalent to the major repeats of CS/DS species from lower animals. For example, CS-E structures (i.e. [GlcA-GalNAc(4,6S)] repeats) occur in the brain. They can interact with midkine, pleiotrophin, HB-EGF, and various members of the FGF family and can promote neurite-outgrowth in vitro (55). An antibody that recognizes decasaccharide sequences from CS-E (and also the [IdoA-GalNac(4,6S)] repeats of CS-H) exhibited specific staining within mouse brain, especially the cerebellum, and could inhibit neurite outgrowth upon a CS-E substratum (56). Similarly, an antibody that recognizes the characteristic disaccharide repeat of A. nigra DS (i.e. [IdoA(2S)-GalNAc(6S)]) also recognizes a highly regulated epitope in the brain, and the corresponding oligosaccharide epitopes have a similar growth factor binding profile to CS-E in vitro (57). Recently, Li et al. (58) demonstrated that HGF/SF has a high affinity for CS/DS rich in the disulfated [IdoA(2S)-GalNac(4S)] and [HexA-GalNAc(4,6S)] units, and that the neurite-outgrowth-promoting activity of this CS/DS can be markedly suppressed by an anti-HGF/SF antibody. It is also interesting that rat liver, an organ that is very responsive to HGF/SF, possesses CS/DS that contains a significant proportion (16%) of disulfated [HexA-GalNAc(4,6S)] disaccharides. 5 Overall, our data show that a variety of highly sulfated, IdoA-/GlcA-containing CS/DS structures have the potential to bind and activate HGF/SF, suggesting a putative role for these growth factor-GAG cofactor combinations, especially in neural development where they may be particularly abundant and developmentally regulated.
In conclusion, we have clearly demonstrated that the interaction of HGF/SF with GAGs is remarkably flexible and of relatively low sequence specificity, tolerating a variety of sulfation patterns that allow it to functionally accommodate both HS/heparin and various forms of CS/DS. The presence of IdoA appears to be critically important in supporting the weaker binding of GAGs with predominantly monosulfated disaccharide repeats, suggesting an important role for GAG conformation. However, HGF/SF affinity increases with increasing sulfate density and this, together with a possible slight preference for O-sulfation at the exocyclic C6 position rather than at the C4 ring position, gives heparin/HS a binding advantage over DS. At higher sulfate densities there appears to be no longer a strict requirement for IdoA, or indeed for any uronate residue, although the latter observation may have little physiological relevance. This apparent lack of a strict GAG sequence specificity mirrors a developing opinion within the FGF field, where it has been suggested that FGF binding to HS (59,60), and the formation of ternary complexes with FGF receptors (24,61), is dictated primarily by the length and overall charge density of sulfated domains rather than by specific dispositions and linear sequences of sulfate groups. There are also similarities with the apparent sulfate density-dependent interaction between heparin/HS and the HepII domain of the matrix protein fibronectin (62). The observation that the activities of different polysaccharides broadly mirror their HGF/SF affinities also has potential mechanistic implications for HGF/SF-MET complex formation. If GAG does interact simultaneously with both MET and HGF/SF in vivo, then this would suggest that any direct GAG-MET binding component should also be of relatively low sequence specificity. Alternatively, this may indicate that GAG may influence MET activation indirectly, solely as a consequence of a direct interaction with, and a consequent effect upon, HGF/SF.