Interactions of Multiple Heparin Binding Growth Factors with Neuropilin-1 and Potentiation of the Activity of Fibroblast Growth Factor-2*

The hypothesis that neuropilin-1 (Npn-1) may interact with heparin-binding proteins other than vascular endothelial growth factor has been tested using an optical biosensor-based binding assay. The results show that fibroblast growth factor (FGF) 1, 2, 4, and 7, FGF receptor 1, hepatocyte growth factor/scatter factor (HGF/SF), FGF-binding protein, normal protease sensitive form of prion protein, antithrombin III, and Npn-1 itself are all able to interact with Npn-1 immobilized on the sensor surface. FGF-2, FGF-4, and HGF/SF are also shown to interact with Npn-1 in a solution assay. Moreover, these protein-protein interactions are dependent on the ionic strength of the medium and are inhibited by heparin, and the kinetics of binding of FGF-2, FGF-4 and HGF/SF to Npn-1 are characterized by fast association rate constants (270,000–1,600,000 m–1 s–1). These results suggest that Npn-1 possesses a “heparin” mimetic site that is able to interact at least in part through ionic bonding with the heparin binding site on many of the proteins studied. Npn-1 was also found to potentiate the growth stimulatory activity of FGF-2 on human umbilical vein endothelial cells, indicating that Npn-1 may not just bind but also regulate the activity of heparin-binding proteins.

It has since become apparent that Npn-1 interacts with other partners and has substantial biological functions outside the nervous system. Thus, Npn-1 is also a receptor for a heparin binding isoform of vascular endothelial growth factor (VEGF), VEGF 165 (6). The presence of Npn-1 apparently enhances the interaction of VEGF 165 with its VEGFR2 signaling receptor and potentiates both VEGF-stimulated mitogenesis and chemotaxis in cultured endothelial cells. However, it is not yet clear whether Npn-1 forms complexes directly with VEGFR2 or through the VEGF ligand (7,8), although it was recently reported that Npn-1 does form complexes directly with VEGFR1/ Flt-1 (9,10). Moreover, studies with monomeric and dimeric forms of soluble Npn-1 suggest that only the dimeric form can deliver VEGF 165 efficiently to the VEGFR2 receptor (11) and so the dimer may represent the agonist form of the molecule, the monomer acting as an antagonist.
Considerable data support a functional role for Npn-1 in regulating VEGF activities on endothelium. Studies with cultured endothelial cells have shown that semaphorin-3A can compete with VEGF 165 binding to Npn-1 and inhibit angiogenesis in vitro (12). Recently, a truncated form of Npn-1 mRNA encoding a soluble form of Npn-1 (sNpn-1) was identified in prostate carcinoma cells, heart, liver, kidney, skin, and placenta (6,13). In situ analysis of Npn-1 expression associated with blood vessels showed that only full-length Npn-1, and not sNpn-1, is expressed. Overexpression of sNpn-1 by prostate carcinoma cells inhibited tumor growth and reduced both the number and integrity of tumor blood vessels, suggesting that sNpn-1 inhibits VEGF 165 binding to Npn-1 or VEGFR2. This is supported by the recent finding that soluble monomeric Npn-1 suppressed angiogenesis in vitro (11). Npn-1 knock-out mice die at mid-gestation and, in addition to neural defects, exhibit transposition of large vessels, disorganized and insufficient capillary formation, and defects in heart development (14). In contrast, overexpression of Npn-1 leads to over stimulation of blood vessel formation (15). These in vivo studies suggest that Npn-1 expression is critical to normal vessel development, tumor, and wound angiogenesis (13)(14)(15)(16).
Npn-1 has also been shown to bind two other VEGF family members, VEGFB and placenta growth factor (17,18) as well as to a VEGF-like protein produced by the orf virus (19). All of these VEGF-related proteins also bind heparin. However, Npn-1 does not bind to VEGF 121 , the VEGF isoform that lacks the heparin binding site encoded by exon 7 (6). One interpretation of these results is that the site on Npn-1, which has been identified as responsible for interacting with heparin-binding VEGF family members, may represent a heparin mimic, which might act as a receptor organizing scaffold. Such a site would be relatively promiscuous and interact with heparin-binding proteins outside the VEGF family. We have tested this idea with a quantitative optical biosensor-based binding assay. The results show that Npn-1 interacts with a subset of the heparinbinding proteins tested, notably, fibroblast growth factor (FGF)-1, FGF-2, FGF-4, FGF-7, FGF receptor 1 (FGFR1), and hepatocyte growth factor/scatter factor (HGF/SF). In support of the idea that these interactions are due to a heparin mimetic site on Npn-1, they depend on ionic strength and are competed by heparin. Because Npn-1 potentiated the growth stimulatory activity of FGF-2 in human umbilical vein endothelial cells (HUVEC), the possibility is raised that Npn-1 may have a broad modulatory activity on a subset of growth factors, morphogens. and other heparin-binding proteins.
Immunoprecipitation-One g of Npn-1 was incubated overnight at 4°C with protein G-Sepharose in 1 ml of 0.5 M NaCl, 20 mM sodium phosphate, pH 6.8. At lower concentrations of NaCl, the Npn-1 selfassociated, which prevented the efficient interaction of the Fc fusion protein with the protein G (result not shown). After centrifugation at 13,000 rpm for 5 min, the supernatant (S1) was collected, and protein was precipitated with 10% trichloroacetic acid. The pellet was washed with 1 ml of PBS and incubated for 3 h at room temperature with growth factor (500 ng FGF-2 and FGF-4, 1.5 g HGF/SF) in 250 l of PBS. The sample was centrifuged at 13,000 rpm for 5 min. The supernatant (S2) was precipitated with 10% trichloroacetic acid, and the pellet was washed with 1 ml of PBS. Supernatants S1 and S2 and the final pellet were adjusted in volume, and polypeptides were separated by SDS-PAGE followed by silver nitrate staining or Western blot using a rabbit antibody to FGF-2 (26,27).
Binding Assays-Npn-1 was immobilized on aminosilane surfaces using bissulfosuccinimidyl suberate (BS 3 ; Perbio, Chester, UK) as the cross-linker following the manufacturer's recommendations (Thermo electron, Basingstoke, UK). To ensure low coverage by the immobilized ligand and prevent self-association of Npn-1 (Table I, Fig. 1) from blocking binding sites, a low concentration of Npn-1 (Յ1 g/ml) was added to the BS 3 -activated surface, and no more than 200 arc s Npn-1 was immobilized on the surface (600 arc s ϭ 1 ng of protein/mm 2 ). Because this self-association reaction was inhibited by high concentrations of NaCl, in some experiments Npn-1 was immobilized (again no more than 200 arc s) in 2 M NaCl buffered with 10 mM Na 2 HPO 4 , pH 7.2. Binding assays were carried out in PBS (140 mM NaCl, 10 mM Na 2 HPO 4 , pH 7.2) supplemented with 0.02% (v/v) Tween 20 (PBST) at 20°C following previously described methods (26 -29) with minor mod-ifications. A single binding assay consisted of adding the soluble binding partner or ligate in 5 l of PBST to a cuvette containing 25 or 45 l of PBST. The association reaction was followed over a set time, usually 180 s, by which time binding was within 90% of the maximum. The cuvette was then washed 3 times with 50 l of PBST to initiate the dissociation of bound ligate. The Npn-1-derivatized surface was regenerated by washing twice with 50 l of 2 M NaCl, 10 mM Na 2 HPO 4 , pH 7.2, and 20 mM HCl, which removed 98 -100% of bound ligate. Binding parameters were calculated from the association and dissociation phases of the binding reactions using the non-linear curve fitting FASTfit (Thermo Electron).
Each binding assay yielded four binding parameters, which are the slope of initial rate of association, the on-rate constant (k on ) and the extent of binding, all calculated from the association phase, and the off-rate constant (k off , equivalent to the dissociation rate constant, k d ), calculated from the dissociation phase. In competitive binding assays, 5 l of PBST-containing polysaccharide was added to a Npn-1 derivatized cuvette containing 40 l of PBST. Ligate was then added, and the association reaction was followed for 210 s.
Data Analysis-The determination of binding kinetics in optical biosensors can be prone to artifactual second phase binding sites, due either to rates of diffusion of soluble ligate approaching or exceeding the rate of association or to steric hindrance at the binding surface (29). Binding assays were designed to avoid such artifacts (26,27,29). Thus, k on was only determined at low concentrations of ligate, whereas k off was measured at higher concentrations of ligate to avoid steric hindrance and rebinding artifacts, respectively. Competing soluble Npn-1 could not be added to the dissociation buffer to avoid rebinding artifacts (29), since it was found to interact with the immobilized Npn-1 in a self-association reaction (Fig. 1). A single site binding model fitted the association and the dissociation data at least as well as a two-site binding model in both the competitive binding assays and the kinetic experiments. Therefore, the binding reaction between the ligates and immobilized Npn-1 was deemed to be monophasic, and a single site model was used to calculate all binding parameters. The equilibrium dissociation constant (K d ) was calculated both from the ratio of the k d and k a and from the extent of binding to provide an estimate of the self-consistency of the results.
Dual Polarization Interferometry-An Analight Bio200 (Farfield Sensors, Salford, UK) was used to probe the Npn-1 immobilized on amine surfaces, since unlike first generation optical biosensors, this instrument measures absolute changes in refractive index and so determines quantitatively the amount of adsorbate (30,31). An aminosilane chip (Farfield Sensors) was calibrated by injecting a solution of 80% ethanol, 20% H 2 O (both w/w) twice over the chip surface followed by an injection of H 2 O. BS 3 was used to cross-link Npn-1 to the amine surface; BS 3 was injected onto both channels on the chip surface at a concentration of 10 mM twice for 3 min per injection followed by a brief wash with PBS at a flow rate of 50 l min Ϫ1 . Npn-1 (3.3 gm/l) was loaded in 2 M NaCl to prevent self-association onto channel 1, and a high concentration of neuropilin, 33 g/ml, was loaded in PBS onto channel 3. Contact time for the reaction with BS 3 was 15 min. Unreacted BS 3 was then blocked with Tris-HCl (1 M, pH 8.0) followed by a wash with 20 mM HCl to remove non-covalently adsorbed Npn-1. Heparin (10 g/ml) and FGF-2 (3.3 g/ml) were injected in PBS at a flow rate of 50 l min Ϫ1 per channel for 3 min.
Size-exclusion Chromatography-Npn-1 (200 l) was applied to a 150-ml column of Superose 12 (Amersham Biosciences) on an ACTA Purifier 10 (Amersham Biosciences) at 0.5 ml/min. Npn-1 was injected at concentrations of 3.33 g/ml in PBS with PBS as the mobile phase and of 10 g/ml in 2 M NaCl buffered with 10 mM Na 2 HPO 4 , pH 7.2, with 2 M NaCl in the same buffer as the mobile phase. A greater concentration of Npn-1 was in the buffer containing 2 M NaCl to compensate for the increased background absorbance from the NaCl. Gel filtration standards (Bio-Rad) were used under the same conditions to assess the hydrodynamic volume of the eluted peaks.
HUVEC DNA Synthesis Assays-HUVEC were routinely cultured in complete large vessel endothelial medium (TCS CellWorks) containing 2% (v/v) fetal calf serum, as recommended by the manufacturer, and in humidified 5% CO 2 , air. Cells were passaged just before confluence using a 1:3 split and used at passage 3. For the determination of the effect of FGF-2 and Npn-1 on growth, cells were suspended in complete medium and plated into a 96-well tissue culture plate (Corning Costar, Sigma) at 10 4 cells per well and allowed to adhere overnight. The cells were then washed 3 times with sterile PBS over a period of 10 min and then incubated in large vessel basal medium (100 l) for 2 h. This was then replaced with 100 l of basal medium containing 10% (v/v) dialyzed fetal calf serum, gentamycin/amphotericin supplement, and 2 L. Duchesne, and D. G. Fernig, manuscript in preparation.
FGF-2 and Npn-1, as indicated in the figure legend. After a further 48 h of incubation, 10 l of 100 M 5-bromo-2Ј-deoxyuridine was added to each well, and the plate was reincubated for a further 3 h. Incorporated 5-bromo-2Ј-deoxyuridine was then measured using a colorimetric immunoassay kit (Roche Diagnostics) following the manufacturer's instructions.

RESULTS
Characterization of Npn-1 in Solution-To identify the monomeric and oligomeric species of Npn-1 in solution under the conditions used for immobilization on biosensor surfaces, Npn-1 was subjected to size-exclusion chromatography in different concentrations of NaCl. Npn-1 chromatographed as two peaks, the first corresponding to a hydrodynamic size equivalent to thyroglobulin (670 kDa) and a second corresponding to bovine ␥ globulin (158 kDa). These presumably represent an oligomeric species and a single Npn-1 Fc fusion protein, respectively. When 3.33 g/ml Npn-1 was applied in PBS, 62% of the protein eluted in the first peak and 38% in the second peak. However, in 2 M NaCl (10 g/ml Npn-1), just 7% of the Npn-1 eluted as the larger species, and 93% eluted as the smaller species. These results support the notion that Npn-1 undergoes a NaCl-dependent self-association and show that in PBS at 3.3 g/ml, 38% of the Npn-1 is likely to be in the form of a single Fc fusion protein.
Characterization of Npn-1 of Surfaces and Immobilized Npn-1-Preliminary results in the IAsys biosensor showed that if the concentration of Npn-1 used to derivatize the surface was 13.3 g/ml in PBS the amount of immobilized Npn-1 (Ͼ600 arc s) was far greater than warranted by simple mass action (1 mg/ml protein required to immobilize ϳ600 arc s). Such a high level of immobilization of Npn-1 was presumably due to the self-association of the Npn-1 (Fig. 1). A dual polarization interferometer was used to obtain information regarding the density of Npn-1 immobilized on aminosilane surfaces and the ability of the immobilized Npn-1 to bind heparin. The addition of 33 g/ml Npn-1 to a BS 3 -activated aminosilane surface resulted in the immobilization of 5.19 ng of protein/mm 2 of sensor surface with a thickness of 4.8 nm (Table I). Although the structure of Npn-1 is not known, given the modular form of the protein and the presence of a C-terminal Fc fusion, the thickness of the immobilized protein suggests an asymmetric shape in which the longer sides are parallel to the surface and the shorter side (4.8 nm) is normal to the sensor surface. The theoretical maximum coverage for Npn-1 oriented in this way is 7.5 ng of protein/mm 2 of sensor surface. Therefore, the immobilized Npn-1 is close to a monolayer. The dual polarization interferometer readily measures the binding of polysaccharides as well as proteins, but on the surface with a high density of Npn-1, specific binding of heparin (10 g/ml) could not be detected (result not shown). Moreover, in both the IAsys and the dual polarization interferometer, Npn-1 immobilized at these high densities did not bind FGF-2 (result not shown).
In the IAsys biosensor the level of immobilized Npn-1 matched that expected from simple mass action (Յ200 arc s or Յ0.3 ng/protein/mm 2 of sensor surface; "Experimental Procedures") when a lower concentration of Npn-1 (Յ1 g/ml) was used or 3.3 g/ml Npn-1 was added in 2 M NaCl. Immobilization of Npn-1 in the dual polarization interferometer under the latter conditions resulted in the immobilization of just 0.16 ng of protein/mm 2 of sensor surface (Table I), which is within the range observed with the IAsys biosensor. This level, commensurate with that expected from simple mass action, represents a protein density equivalent to 2% of a monolayer. Npn-1 immobilized at a low density in the interferometer did not bind 10 g/ml heparin (result not shown), but such Npn-1 did bind FGF-2 in both instruments.
Taken together with the size-exclusion chromatography, the results suggest that Npn-1 is immobilized at a high density as a consequence of self-association of the Npn-1 and that in this form the Npn-1 fails to bind heparin or FGF-2. In contrast,

FIG. 1. Interaction of proteins with immobilized Npn-1 determined in an optical biosensor and in solution.
A, proteins were added as soluble ligates in 5 l of PBST to a Npn-1-derivatized cuvette containing 25 l of PBST. The association data were analyzed with a one-site binding model ("Experimental Procedures") to determine the extent of binding: VEGF 165 , 2 g/ml; FGF-1, 3.3 g/ml; FGF-2, 3.3 g/ml; FGF-4, 1.7 g/ml; FGF-7, 1.7 g/ml; FGFR1-ST, 60 g/ml; FG-FBP, 8.3 g/ml; HGF/SF, 1.7 g/ml; Npn-1, 3.33 g/ml; PrP c , 7.65 g/ml; antithrombin III (ATIII), 16.7 g/ml; CyPB, 16.7 g/ml; tumor necrosis factor ␣ (TNFalpha), 16.7 g/ml. Results are duplicates of one of at least two separate experiments carried out on different Npn-1derivatized surfaces. Errors are the S.E. of duplicates of the extent of binding calculated from fitting the data to a one-site model. B and C, protein G-Sepharose loaded with Npn-1 was incubated with growth factors for 3 h and then collected by centrifugation. S1 is the supernatant obtained after centrifugation of protein G-Sepharose following overnight loading with Npn-1. S2 is the supernatant after centrifugation of protein G-Sepharose loaded with Npn-1 after incubation with growth factor. P is the pellet corresponding to S2. S1, S2, and P were collected and analyzed by SDS-PAGE followed by silver staining or Western blotting ("Experimental Procedures"). B, incubation with HGF/SF, silver-stained. C, incubation with FGF-4 and FGF-2; FGF-4 is silver-stained, FGF-2 is silver-stained (protein) or Western blotted (WB) with an antibody to FGF-2. Molecular weight markers are shown on the left-hand side of the panel B; one of two experiments is shown. Npn-1 immobilized at a low density is likely to consist of single, well separated protein units; the covalent link to the surface prevents subsequent self-association of these molecules. In this form Npn-1 binds FGF-2 and a range of other proteins ( Fig. 1) but still does not bind heparin, which may be due to basic residues in Npn-1 that are involved in binding heparin being obscured, perhaps due to one or more reacting with the BS 3 .
Interactions of Proteins with Npn-1-Surfaces with a low density of immobilized Npn-1 (Յ200 arc s) were used to measure the interactions of proteins with Npn-1 in the IAsys biosensor. Ten proteins whose interactions with heparin/heparan sulfate (HS) are known to depend on different combinations of structural elements within the polysaccharide were tested for their ability to bind to Npn-1. The binding of three other proteins to immobilized Npn-1 was also determined: VEGF 165 , a known HS and Npn-1-binding protein, Npn-1 itself, and tumor necrosis factor ␣, which in our hands failed to bind to heparin (result not shown). None of the proteins showed significant (Ն1 arc s) binding to underivatized control aminosilane surfaces, indicating that under the conditions of this assay there was no significant nonspecific binding (result not shown). In addition, when the Fc fusion partner of the Npn-1 was immobilized alone on aminosilane surfaces, no binding was observed (result not shown). Because Npn-1 binds HS (32), it was important to exclude the possibility that the Npn-1 protein used in these experiments was associated with heparin/HS carried over from the expression system or purification. Four lines of evidence exclude the possibility that the interactions observed in this study are due to heparin/HS associated with the Npn-1 protein being immobilized on the sensor surface. First, Npn-1-derivatized surfaces were always subjected to a regeneration cycle before use in a binding assay so as to remove any traces of non-covalently bound polysaccharide. Second, the Npn-1 surfaces lost ligate binding activity after 48 -72 h (result not shown), which is probably due to the denaturation of the immobilized Npn-1. In contrast, polysaccharide-derivatized sensor surfaces are stable for many months (28,29) since these polymers cannot denature. Third, storing Npn-1-derivatized surfaces in PBST containing 1 g/ml heparin prolonged the growth factor binding activity of the surfaces to more than a week. Fourth, if high levels of Npn-1 were immobilized on the surface (Ͼ600 arc s in the IAsys, 5.19 ng of protein/mm 2 in the Analight), proteins did not bind the immobilized Npn-1, in all probability due to self-association of the immobilized Npn-1 blocking the binding site of the proteins.
As expected, VEGF 165 bound the immobilized Npn-1 (Fig. 1). Strong binding to Npn-1 was observed with the four FGFs tested, FGF-1, FGF-2, FGF-4, and FGF-7. A set of structurally unrelated proteins also bound Npn-1 relatively strongly, FGFBP, HGF/SF, PrP c , and Npn-1 itself (Fig. 1). A lower level of binding that required higher concentrations of ligate was observed with FGFR1-ST and antithrombin III, whereas an interaction between CyPB or tumor necrosis factor ␣ with Npn-1 could not be detected at the concentrations used in the assay (Fig. 1).
A limited number of interactions between Npn-1 and heparin-binding proteins were also measured in solution by pull down. Loading of protein G with the Npn-1 Fc fusion protein only occurred efficiently when the NaCl concentration was increased to 500 mM, presumably because at lower concentration of electrolytes the Npn-1 self-associated, which obscured the Fc fusion protein. Protein G-Sepharose alone failed to pull down HGF/SF, FGF-4, or FGF-2 (result not shown). After pull down with protein G loaded with Npn-1, some HGF/SF, but no detectable FGF-4 or FGF-2, was found in the supernatant S2 (Figs. 1, B and C). The pellet from the HGF/SF pull down contained bands at 25 and 75 kDa, corresponding to the ␣ and ␤ chains of the growth factor (Fig. 1B). Some SDS and mercaptoethanol resistant HGF/SF ␣␤ dimer was also observed at 85 kDa. Npn-1 consistently stained poorly and migrated as a single band corresponding to 150 kDa. FGF-4 and FGF-2 were similarly pulled down by Npn-1-loaded protein G-Sepharose, and in the case of FGF-2 this was confirmed by Western blotting (Fig. 1C).
Kinetics of Interaction-These results suggested that Npn-1 is able to interact specifically with many, but not all heparinbinding proteins. To gain further understanding of these interactions, the kinetics of interaction of FGF-2, FGF-4, and HGF/SF with Npn-1 were then analyzed. The validity of the analyses is evidenced by the data from one of the three experiments with HGF/SF, which are presented in detail. As the concentration of HGF/SF was increased, the extent of binding increased, as did the gradient of the association curves, indicating that the interaction was dependent on the concentration of the ligate (Fig. 2A). Analysis of the association data with a single site model (Fig. 2, C-I) shows that at all concentrations of HGF/SF, the distribution of the data around the one site model was random. Analysis of the association data with a two-site model did not improve the goodness of fit (result not shown). These results, therefore, indicate that a one-site binding model is a suitable description of the interaction between HGF/SF and Npn-1. The slope of a plot of the initial rate of HGF/SF binding to Npn-1 against the concentration of HGF/SF increased linearly with concentration (Fig. 2J), which indicates the binding of HGF/SF is not diffusion-limited (33). Further evidence for the appropriateness of a one-site model is the observation that k on , calculated from a one-site model, also increased linearly with concentration of HGF/SF (Fig. 2K). The dissociation of HFG/SF from immobilized Npn-1 (Fig. 2B) was also best fitted by a single-site curve (Table II). The interactions of FGF-2 and FGF-4 with Npn-1 were also best described by a single site binding model (results not shown).
The association reaction for the interaction of Npn-1 with the three selected growth factors was characterized by a fast k ass , which ranged from 270,000 M Ϫ1 s Ϫ1 for FGF-2 to 1,600,000 M Ϫ1 s Ϫ1 for FGF-4 and HGF/SF (Table II). The k diss of these interactions ranged from 0.012 s Ϫ1 for FGF-4 to 0.036 s Ϫ1 for FGF-2 (Table II). Consequently, FGF-4 possessed the highest affinity for Npn-1 (2.2 nM) followed by HGF/SF (19 nM) and then FGF-2, which bound Npn-1 with considerably lower affinity, 130 nM. Because the values of K d calculated from the kinetic parameters are similar to those calculated from the extent of binding when this had reached a maximum (Table II), it is likely that these binding parameters represent the intrinsic values.
Competition for Npn-1 Binding by Heparin and Heparinderived Hexasaccharides-In all cases tested (Fig. 1) proteins that bound Npn-1 could be dissociated by increasing the ionic strength, since 2 M NaCl was found to be sufficient to regenerate at least partially the Npn-1 binding surfaces (result not shown) and was, therefore, included in the surface regeneration washes ("Experimental Procedures"). This observation suggested that electrostatic interactions might be an important driving force for the interactions between Npn-1 and its binding proteins. Ionic interactions drive the binding of the same growth factors to heparin/HS. To test whether the heparin binding sites of the growth factors may be involved in binding to Npn-1, we examined whether heparin and a heparin-derived hexasaccharide close in size to the minimal binding site for some of the proteins, e.g. FGF-2 and HGF/SF (26,27), might act as competitors for the interaction between FGF-2 and Npn-1. Heparin (20 g/ml) completely inhibited this interaction, whereas 10-fold more hexasaccharide reduced the binding of FGF-2 to Npn-1 by 80% (Fig. 3A). The binding of HGF/SF to Npn-1 was similarly competed by heparin and a hexasaccharide, although in this case the hexasaccharide was 100-fold less effective as a competitor (Fig. 3B). The ability of heparin and a hexasaccharide to compete for the binding of six other proteins with Npn-1 was then examined. Heparin effectively competed for the binding of FGF-4, FGF-1, FGF-7, FGFR1-ST, PrP c , and Npn-1 itself to Npn-1 (Fig. 3C). In contrast, the hexasaccharide inhibited the binding of FGF-1, FGF-7, and FGFR1-ST to Npn-1 but was a much less effective inhibitor of the binding of FGF-4 and PrP c and did not inhibit the self-association of Npn-1 (Fig. 3C).
Effect of Npn-1 on FGF-2 Stimulation of DNA Synthesis-In the appropriate assay conditions Npn-1 has been shown to potentiate the proliferative activity of VEGF 165 on endothelial cells (6). It was of interest to determine if Npn-1 might have a similar effect on some of the partners identified in the present study. FGF-2 and HUVECs were used as a test system because Npn-1 is known to regulate endothelium in vivo, and FGF-2 is known to be an important component of the angiogenic switch (34). FGF-2 induced approximately a 10-fold increase in  a The S.E. of the k ass was derived from the deviation of the data from a one-site binding model, calculated by matrix inversion using the FastFit software provided with the instrument. No evidence was found for a two-site model of association, and so the growth factor binding sites on Npn-1 were homogenous in this respect. Three independent sets of k on were measured, and the three resulting values for k ass and their errors were combined.
b The correlation coefficient of the linear regression through the k on values. c The k diss is the mean Ϯ S.E. of five values, obtained at high concentrations of the respective growth factor. No evidence was found for a two-site model of dissociation, and so the growth factor binding sites in Npn-1 were homogenous in this respect as well.
d The K d (kinetic) was calculated from the ratio of k diss /k ass , and its S.E. is the combined S.E. of the two kinetic parameters. e The K d (equilibrium) was calculated from the extent of binding at or near equilibrium, and its S.E. is the combined S.E. of three independent determinations of K d .
HUVEC DNA synthesis at 10 ng/ml (Fig. 4A). Npn-1 (25 ng/ml) significantly increased the FGF-2 stimulation of DNA synthesis by 35 and 52% at 3 and 10 ng/ml FGF-2, respectively (Fig.  4A). Because Npn-1 alone was without effect (Fig. 4A, 0 ng/ml FGF-2), these results suggest that the interaction between Npn-1 and FGF-2 or its receptor increases the proliferative activity of the growth factor. In the presence of 10 ng/ml FGF-2, a potentiation of the growth factor proliferative activity was apparent at 20 g/ml Npn-1, and this reached a maximum between 30 and 50 g/ml Npn-1 (Fig. 4B). DISCUSSION Npn-1 is a multifunctional protein that interacts with a range of partners that are specified in part by the tissue context. In the nervous system Npn-1 is a receptor for semaphorin 3A, and this interaction involves a basic region in semaphorin 3A and acidic regions in the first F5/8 b1 domain (Fig. 5A) of Npn-1 (35). In the endothelium, Npn-1 interacts with the heparin binding isoforms of VEGF. There is, however, no interaction between Npn-1 and VEGF 121 , the isoform that lacks the exon 7-encoded basic heparin-binding module. VEGF and semaphorin bind Npn-1 competitively, indicating that they interact with substantially the same acidic region of Npn-1 (35). These observations suggested to us that Npn-1 may have the ability to interact with HS-binding proteins other than mem-bers of the VEGF family. The results of the optical biosensorbased and solution binding assays show that this is indeed the case for some but not for all HS-binding proteins.
Thus, Npn-1 bound four members of the FGF family, FGF-1, -2, -4, and -7 as well as FGFR1-ST, FGFBP, HGF/SF, PrP c , antithrombin III, and Npn-1 itself. The four FGFs are known to bind to saccharide structures within heparan sulfate/heparin that possess different combinations of N-sulfate and 6-O-sulfate on the glucosamine, iduronic acid, and 2-O-sulfate on iduronate. Thus, the FGF-2 binding site in HS involves interaction of the protein with N-sulfate on glucosamine and 2-O-sulfate on iduronic acid with a minimum of four saccharides being required for binding (26,36,37). The other FGFs recognize motifs in heparin/HS that differ in their patterns of sulfation and length, allowing regulation of their activity by the polysaccharide (28,38,39). For example, 6-O sulfation of glucosamine in the context of a pentasaccharide is essential for the interactions of FGF-1 with the polysaccharide (40 -42), whereas the similar requirement for 6-O sulfation of FGF-4 is in the context of a decasaccharide (41). FGF-7 requires at least both 2-O-sulfated iduronate and 6-Osulfated glucosamine (43). HGF/SF binds to a similar subclass of sequences in HS, requiring 6-O-sulfate and iduronate (44), with a tetrasaccharide representing the minimal length of saccharide (27), whereas PrP c shows a strong preference for 2-O-sulfated iduronate (45). Iduronate residues contribute significantly to these interactions, which probably reflects the role of the flexibility of this sugar ring (46,47) in mediating protein interactions. Indeed, HGF/SF also interacts with dermatan sulfate but not chondroitin sulfate (48), including a dermatan sulfate, which probably contains just one iduronate per chain (49). Thus, these growth factors recognize binding sites that would be found in the classic S-domain of the polysaccharide (50,51). FGFR1 probably also recognizes a similar structural motif (52). In marked contrast antithrombin III and CyPB, which bind Npn-1 poorly or not at all, have additional structural requirements in the polysaccharide. The antithrombin III binding site consists of a pentasaccharide containing both 3-O sulfated glucosamine and glucuronic acid (53). CyPB recognizes a site in heparin/HS that contains a free amino group. 3 Taking the HS-binding specificities of these proteins together with their Npn-1 binding activity suggests that Npn-1 recognizes a site on proteins which itself binds motifs in heparin/HS containing various permutations of N-sulfate, 6-Osulfate on glucosamine, iduronate and 2-O-sulfate on iduronate. Sites in proteins that bind motifs in the polysaccharide containing other features, e.g. 3-O sulfate, glucuronic acid, a free amine in glucosamine, interact poorly with Npn-1. Hence, these results suggest that Npn-1 contains a binding site that acts as a mimic for particular motifs in HS/heparin. The motifs in the polysaccharide recognized by Npn-1 and FGFBP remain to be elucidated. However, the above argument and the observation that these proteins bind Npn-1 suggests that they possess a polysaccharide binding site containing some permutation of N-and 6-O-sulfated glucosamine, iduronate, and 2-Osulfated iduronate. Moreover, since a basic region of semaphorin 3A binds the same (or an overlapping) site on Npn-1 as VEGF 165 (35), semaphorin 3A may also be a heparinbinding protein, although this remains to be tested.
The conjecture that Npn-1 possesses a heparin-mimetic site is reinforced by several other lines of evidence. First, the kinetics of binding of FGF-2, FGF-4, and HGF/SF to Npn-1 were characterized by a rapid k ass , a feature often associated with interactions involving substantial ionic bonding, such as those between proteins and HS (26 -28, 48, 49, 52, 54, 55), since ionic bonding acts over longer distances than other non-covalent bonding modes, e.g. van der Waals, H-bonding. Second, much of the bound protein could be removed by washing the Npn-1 derivatized cuvette with 2 M NaCl, again a characteristic of ionic interactions. Third, binding of FGF-1, FGF-2, FGF-4, FGF-7, FGFR1, HGF/SF, PrP c , and Npn-1 itself to Npn-1 was inhibited by heparin and in some cases by a heparin-derived hexasaccharide. The latter observations do not contradict previous ones which show that the interaction of VEGF 165 and other family members with Npn-1 is potentiated by heparin, e.g. Ref. 32, since analysis of the Npn-1 immobilized on aminosilane sensor surfaces indicated that the heparin binding site is not functional, probably due to its involvement in covalent coupling to the surface.
Where is the putative heparin-mimetic site in Npn-1? Semaphorin 3A and the heparin binding isoforms of VEGF compete for binding to Npn-1 (35). Basic regions in VEGF and semaphorin 3A are responsible for the interaction with Npn-1, the b1 domain (Fig. 5) being a major site of interaction (35). Deletion analysis of Npn-1 domains involved in VEGF binding cannot separate functionally b1 and b2 (32), but the basic isoelectric point of b2 (Fig. 5A) makes this a less favorable option. The b1 domain is to date the only one for which a structural model is available (35). Although b1 has a neutral isoelectric point (Fig. 5A), the model structure reveals that charged residues are segregated on its surface, such that it presents an electronegative face and an adjacent electropositive face (Fig. 5B). Thus, it seems likely that the electronegative face of b1 makes a major contribution to the heparinmimetic site in Npn-1, although the overall acidic character of the two CUB domains, a1 and a2 (Fig. 5A) means that these N-terminal parts of Npn-1 may have a role to play in binding the basic heparin binding sites of proteins. Intriguingly, the heparin binding site of Npn-1 has also been localized to the b1/b2 domain (32), and thus, the two adjacent faces of the b1 domain may be involved in interactions with heparin and he-parin-binding proteins, although the basic b2 domain might be expected to contribute to polysaccharide binding. The MAM domain (Fig. 5A) has been suggested to be involved in Npn-1 oligomerization. However, the observation that Npn-1 self-association of NP-1 is ionic in character, that self-associated Npn-1 in solution is oligomeric, and that when immobilized on a surface at high concentrations, conditions that promote selfassociation, Npn-1 does not interact with heparin-binding proteins, all suggest the b1 domain may also be involved in Npn-1 self-association.
The importance of the conclusion that Npn-1 binds a subset of heparin-binding proteins through a heparin mimetic site is reflected by the potentiation of the growth stimulatory activity of FGF-2 by Npn-1 on HUVECs. The majority of cytokines, chemokines, morphogens, and growth factors bind heparin/HS, and the activity of many of these regulatory proteins is dependent on or is modulated by the polysaccharide (56). Therefore, Npn-1 and perhaps other members of the neuropilin family (1) may have a far wider spectrum of activity than currently appreciated. This may explain the phenotype of the Npn-1 knock out mouse, which does not phenocopy to known functions of VEGF (14). Several fascinating aspects of Npn-1 function remain to be established, including how Npn-1 potentiates the activity of growth factor-receptor complexes, whether it has a general scaffold function, perhaps by a balance of self-association and hetero interactions, and what function, if any, do other members of the neuropilin family have with respect to heparinbinding proteins.