Cooperative Dimerization of Fibroblast Growth Factor 1 (FGF1) upon a Single Heparin Saccharide May Drive the Formation of 2:2:1 FGF1·FGFR2c·Heparin Ternary Complexes*

The related glycosaminoglycans heparin and heparan sulfate are essential for the activity of the fibroblast growth factor (FGF) family as they form an integral part of the signaling complex at the cell surface. Using size-exclusion chromatography we have studied the capacities of a variety of heparin oligosaccharides to bind FGF1 and FGFR2c both separately and together in ternary complexes. In the absence of heparin, FGF1 had no detectable affinity for FGFR2c. However, 2:2:1 complexes formed spontaneously in solution between FGF1, FGFR2c, and heparin octasaccharide (dp8). The dp8 sample was the shortest chain length that bound FGFR2c, that dimerized FGF1, and that promoted a strong mitogenic response to FGF1 through FGFR2c. Heparin hexasaccharide and various selectively desulfated heparin dp12s failed to bind FGFR2c and could only interact with FGF1 monomerically. These saccharides formed 1:1:1 complexes with FGF1 and FGFR2c, which had no tendency to self-associate, suggesting that binding of two FGF1 molecules to the same saccharide chain is a prerequisite for subsequent FGFR2c dimerization. We found that FGF1 dimerization upon heparin was favored over monomeric interactions even when a large excess of saccharide was present. A cooperative mechanism of FGF1 dimerization could explain how 2:2:1 signaling complexes form at the cell surface, an environment rich in heparan sulfate.

The fibroblast growth factor (FGF) 2 family of proteins control the proliferation, migration, differentiation, and survival of a wide range of cell types (1). The FGFs are of fundamental importance during embryogenesis where they orchestrate the complex processes of organ formation and limb development (2). Dysregulated FGF expression is often associated with malignant transformation, because it facilitates the growth, survival, and metastatic spread of cancer cells (3). In humans 22 members of the FGF family have been described along with 4 FGF receptor tyrosine kinases (FGFR1-4). FGF signal transduction is achieved through ligand-dependent dimerization of FGFR subunits and their subsequent activation through transphosphorylation (4).
The FGFRs share a number of common structural features (5). All have an intracellular split tyrosine kinase domain, a transmembrane sequence, and an extracellular region composed of three immunoglobulin (Ig)-like domains (D1-D3). FGF binding is mediated by the D2 and D3 domains, with the D3 domain having the primary influence on ligand specificity (6,7). The D1 domain of the FGFRs does not take part in ligand interactions, and engineered receptors that lack this domain bind FGFs at least as strongly as full-length forms (8). For FGFRs1-3, alternative splicing of the D3 domain gives rise to two variants (termed IIIb and IIIc) with different ligand specificities (9). Therefore seven distinct FGF receptors can be expressed on the cell surface, each of which binds specifically to a subset of the FGFs. Significant redundancy is present within this system, because each receptor can be activated by several different FGFs. The FGFs themselves are also not limited to a single FGFR type, for example FGF1 (acidic FGF) is a universal ligand, recognized by all FGF receptors (10). On the other hand, FGF7 (keratinocyte growth factor) is only capable of binding FGFR2b (11). The D2 domain of the FGFRs contains a short stretch of basic amino acids referred to as the K18K loop (12). This region mediates interaction with heparin and HS (13) and is essential for receptor function. The FGFs are also heparin-binding proteins, and it is now appreciated that signal transduction requires the formation of a ternary complex at the cell surface formed between FGF, FGFR, and heparan sulfate (HS).
Heparin and HS are two structurally and functionally related glycosaminoglycans (GAGs (14 -16)). HS is expressed on core proteins in the extracellular matrix and on the surface of almost all mammalian cell types. In contrast, heparin is produced by mast cells where it is stored in granules for release as free chains. Biosynthesis occurs in the Golgi apparatus with the same enzymes involved for both GAGs (17). The saccharides are synthesized as 134 linked polymers of ␤-D-glucuronate (GlcA) and ␣-D-N-acetylglucosamine (GlcNAc). During chain elongation, enzymes substitute acetyl groups for sulfates to create ␣-D-N-sulfoglucosamine (GlcNS) residues (18). Regions of N-sulfation are then acted on by a variety of enzymes, which catalyze further modification steps, most commonly the epimerization of GlcA to ␣-L-iduronate (IdoA), 2-O-sulfation of IdoA residues and 6-O-sulfation of GlcNAc/ GlcNS (collectively, GlcNR). A key feature of heparin/HS synthesis is that not all available substrate sites are acted upon by each enzyme, resulting in a significant degree of structural heterogeneity (17). During HS synthesis an unknown mechanism targets the modification steps to discrete regions of the polysaccharide chain. This creates a domain organization where sulfated and unmodified regions of roughly equal length (14 -18 disaccharides) alternate in a relatively uniform pattern along the HS chain (19). Within the sulfated regions are stretches of IdoA(2S)-GlcNS with varying patterns of 6-O-sulfation (S-domains, 2-7 disaccharides long (20)), which are flanked by stretches of alternat-* This work was supported by funding from Cancer Research UK. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 1 To whom correspondence should be addressed. ing N-sulfated and N-acetylated disaccharides with highly variable patterns of O-sulfation (19). In contrast to HS, heparin chains are shorter and highly sulfated throughout. Heparin saccharides can be likened to the S-domains regions of HS, and the relative ease with which they can be prepared means that heparin is often used experimentally as a substitute for HS. The heterogeneity of sulfation present in HS, and to a lesser extent heparin, is key to their functions, because distinct patterns of sulfation define binding sites for specific protein ligands (14).
Variations in the distribution of basic amino acids upon the surface of the FGFs lead to markedly different heparin-binding properties (21). FGF2 (basic FGF) will bind to heparin/HS pentasaccharides (dp5) that are devoid of 6-O-sulfate groups as long as they contain one or more critically positioned N-and 2-O-sulfates (22). In contrast, FGF1 binding to heparin/HS oligosaccharides is thought to require N-, 2-O-, and 6-Osulfation (23). Heparin adopts a helical secondary structure in solution, leading to the sulfate groups of contiguous disaccharide units alternating between opposite sides of the helical axis (24). This conformation allows protein ligands to dimerize through interactions with opposite faces of the same short saccharide sequence. Crystal structures of 2:1 complexes formed between FGF1 and heparin dp10 show this trans binding arrangement (25). Whether FGF dimerization can occur in vivo on HS chains has important implications for how ternary complexes form and what stoichiometry these complexes adopt.
Two crystal structures of FGF⅐FGFR⅐heparin ternary complexes have been published. Schlessinger, Mohammadi, and colleagues (26) describe a putative 2:2:2 complex formed between FGF2, FGFR1c, and heparin dp10. In each half of this structure a heparin chain bound both FGF2 and FGFR1c, augmenting the ligand-receptor interaction. The two 1:1:1 complexes were then held together through direct FGFR1c-FGFR1c contacts, as well as secondary FGF2-FGFR1c and heparin-FGFR1c interactions. In contrast, Pellegrini et al. (27) identified a possible 2:2:1 FGF1⅐FGFR2c⅐heparin dp10 structure in which a single heparin chain bridged the gap between two FGF1⅐FGFR2c complexes, making additional contacts with one of the receptor subunits. The two ternary complex architectures place differing limitations on where they can form upon HS chains (28). Steric hindrance would only allow the 2:2:2 structure (26) to form between FGF⅐FGFR complexes on the very end of separate HS chains. In contrast, the heparin-bridged 2:2:1 complex (27) could form internally upon a HS chain but, because the two FGF⅐FGFR half-complexes do not interact directly, they are reliant on the probability that they become aligned upon opposite faces of the same short stretch of HS. A recent study using both analytical ultracentrifugation and mass spectrometry suggests that, in solution, 2:2:1 ternary complexes can be formed in which the FGF⅐FGFR subunits are arranged in either a Pellegrini-or Schlessinger/Mohammadi-like architecture (29). Expanding on these studies, we have used a variety of heparin fragments in a size-exclusion chromatography (SEC) system to investigate the role that heparin plays in the assembly of FGF1⅐FGFR2c complexes.
High Performance SEC-A Hewlett Packard 1100 Series High Performance Liquid Chromatography (HPLC) system was used to equilibrate a prepacked Superdex 200 HR column (10 ϫ 300 mm, Amersham Biosciences) at 0.5 ml/min in 150 mM NaCl, 10 mM HEPES buffer (pH 7.2). The same column and HPLC system were used for all runs. Proteins and/or heparin fragments were preincubated for 30 min at room temperature in 240 l of column running buffer prior to column loading. These samples were then eluted isocratically at 0.5 ml/min, and the absorbance at 280 nm (aromatic amino acid side chains) or 232 nm (unsaturated bonds of heparin fragments) was recorded. The elution of AMAC-labeled heparin dp12 was monitored by fluorescence detection (excitation max., 425 nm; emission max., 520 nm). All elution profiles shown are representative of at least two repeats.
The Superdex 200 column was calibrated using gel filtration HMW and LMW protein calibration kits (Amersham Biosciences). The column void (V o ) and total (V t ) volumes were determined using blue dextran 2000 and sodium dichromate, respectively. The elution volumes (V e ) of protein standards were converted in to a calibration chart of K av against molecular mass [K av . A line of best fit was fitted to the calibration data using Microsoft Excel, and the equation of this line was used to estimate mass according to observed K av . We also plotted the square root of Ϫlog 10 (K av ) against the Stokes radius for the protein standards. The equation for the line of best fit on this chart was used to calculate the Stokes radius of eluted species.
Disaccharide Composition Analysis of Heparin Oligosaccharides-Heparin saccharides (200 g) were suspended in heparinase buffer (50 mM NaAc, 0.5 mM CaAc 2 , 0.1 mg/ml bovine serum albumin, pH 7.0). Aliquots of heparinases I, II, and III were added to a concentration of 100 mIU/ml, and digests were incubated at 37°C for 16 h. Disaccharides were purified from the heparin digests using a Superdex peptide column (10 ϫ 300 mm, Amersham Biosciences) running at 0.5 ml/min in 0.1 M NH 4 HCO 3 , while monitoring the absorbance of the eluate at 232 nm. Heparin disaccharides were then repeatedly lyophilized to remove NH 4 HCO 3 until their conductivity in 1 ml of dH 2 O, pH 3.5, was Ͻ200 S.
A single ProPac PA-1 strong anion-exchange HPLC column (4.6 ϫ 250 mm, Dionex, Camberley, UK) was linked to a Hewlett Packard 1100 series high-performance chromatography system. The column was run at a flow rate of 1 ml/min in MilliQ ultrapure water, pH 3.5. Lyophilized disaccharides were resuspended in 1 ml of this buffer and loaded onto the column prior to the application of the gradient (45 min, 0 -1.0 M NaCl). The eluate was monitored for absorbance at 232 nm, and peaks were identified by comparison to heparin disaccharide standards.
Mitogenesis Assay-BaF-32 cells are a lymphoblastoid cell line deficient for HS synthesis (33). These cells have no endogenous expression of the FGFs or FGFRs, but were transfected with FGFR2c. FGF1-stimulated proliferation of the transfected cells required the addition of HS or heparin saccharides to the culture medium. BaF-32 cells were routinely maintained in interleukin-3-conditioned medium as described before (34). For the mitogenesis assay, cells were serum-starved for 2 days, washed 3 times in phosphate-buffered saline then seeded at a low density (2.5 ϫ 10 5 cells/ml) onto a 96-well plate in RPMI 1640 supplemented with 10% (v/v) horse serum. Cells were incubated for 4 days at 37°C in medium containing FGF1 (100 ng/ml) and sized heparin oligo-saccharides (0.1-2.0 g/ml) or porcine mucosal HS (10 g/ml). The CytoLite TM kit (Packard BioScience) was used as a chemiluminometric assay of viable cell number.

RESULTS
Size-exclusion Chromatography of FGF1, FGFR2c, and Heparin dp12-In this study we applied the technique to investigate the formation in solution of ternary complexes between FGF1, FGFR2c, and heparin. A Superdex 200 HR 10/30 column was connected to an HPLC system and equilibrated in a buffer of physiological ionic strength. A range of protein standards were used to calibrate the column so that the Stokes radius (R S ) and apparent mass could be determined from the K av values of eluted species (see "Experimental Procedures"). Human FGF1, the ligand-binding domain (D2-D3) of human FGFR2c, and a heparin dodecasaccharide sample were then applied separately to the calibrated column ( Fig. 1). FGF1 eluted with a peak at 36.1 min, which corresponds to an apparent mass of 16 kDa, very close to its known mass of 15.8 kDa. FGF1 has a compact globular structure (35) and therefore eluted close to its predicted K av value on the calibrated column. The ectodomain of FGFR2c had an apparent molecular mass of 33 kDa, larger than its actual mass of 24.5 kDa. The receptor protein consists of two contiguous Ig-like motifs rather than a single globular domain (27), and this was reflected in a larger than expected hydrodynamic volume. Our FGFR2c protein was purified from a bacterial expression system (27) and was therefore non-glycosylated. Glycosylation may have an important role in FGFR activity in vivo (36) but does not appear to affect FGF-binding ability (37,38). Both Schlessinger et al. (26) and Pellegrini et al. (27) used Escherichia coli-expressed FGFR proteins, and therefore the lack of glycosylation should not bias the formation of ternary complexes with a particular stoichiometry. The fully sulfated heparin 12-mer (dp12) had a molecular mass of just 3.44 kDa; however, this oligosaccharide adopts a rigid helical structure in solution (24), which resulted in a large observed Stokes radius (21 Å). When run on a Superdex 200 column calibrated with protein standards, the dp12 sample had an apparent molecular mass of 22 kDa. Data on these and all subsequent major observed peaks are presented in TABLES ONE and TWO.
FGF1 Is Dimerized by Heparin dp12 and Forms 2:2:1 FGF1⅐FGFR2c⅐ dp12 Ternary Complexes-We applied the SEC technique to investigate, in solution, the formation of complexes between heparin saccharides and FGF1/FGFR2c. Protein elution could be followed by absorbance at 280 nm. However, proteins also absorb strongly at 232 nm; therefore, the heparin component could not be monitored at this wavelength. We investigated the use of 2-aminoacridone (AMAC) as a fluorescent tag for the reducing end of our heparin dodecasaccharide. This allowed us to monitor simultaneo-  Elution times, Stokes radii, and mass calculations for the major observed peaks Elution times shown are the points of maximal absorbance for the peaks indicated. The apparent mass of each peak is calculated from the elution time as described ("Experimental Procedures"). Over multiple runs elution times typically varied by Ϯ0.1 min, which corresponds to Ϯ1.5 kDa for the FGF1⅐dp12 complex, and Ϯ3 kDa for the FGF1⅐FGFR2c⅐dp12 ternary complex. Predictions on the stoichiometry of the complexes that each peak represents are presented. Based on each predicted stoichiometry, both the expected apparent mass and actual mass are included. Expected mass is the sum of the apparent masses of each component. The actual mass is the sum of the known masses of each component.  usly the elution of protein (absorbance at 280 nm) and saccharide (fluorescence) from the Superdex 200 column. The presence of the AMAC-tag had no significant effect on the elution position of the FGF1⅐dp12 and FGF1⅐F-GFR2c⅐dp12 complexes observed. The sensitivity of this approach also allowed us to detect the same complexes at concentrations below that which protein absorbance could be monitored (Ͻ200 pmol, data not shown). SEC with fluorescent end labeling of saccharides could be used in future studies to investigate protein⅐heparin complex formation at physiologically relevant concentration levels. A 2:1 incubation of FGF1 and AMAC-dp12 led to the formation of a single complex (R S ϭ 32.6 Å) in which all of the FGF1 and saccharide was present (Fig. 2A). The crystal structure of a 2:1 FGF1⅐dp10 complex has been previously published (25) and has a predicted hydrodynamic radius of 29.1 Å (using the program Crysol (39)). These data indicate that the observed FGF1⅐dp12 complex has a 2:1 stoichiometry. Such a complex would have a mass of 35 kDa, however, the apparent mass of the peak calculated from the calibration data was 57 kDa. This could be explained by the large apparent mass of the heparin dp12. The sum of the apparent masses of the components in the 2:1 complex was 55 kDa (16 ϩ 16 ϩ 22 kDa).
Using SEC we failed to detect any interaction between FGF1 and FGFR2c in the absence of heparin (Fig. 2B). However, when these two proteins were incubated in a 2:2:1 ratio with heparin dp12, a single major peak was observed with a much earlier elution position. All of the AMAC-tagged dp12 was present within this complex indicating that a 2:2:1 complex was formed. The Stokes radius of this ternary complex (41.4 Å, using heparin dp12) was also closer to that predicted from the crystal structure of the 2:2:1 Pellegrini complex (40.9 Å, dp10), than that of the 2:2:2 Schlessinger/Mohammadi complex (37.7 Å, dp10, calculated using Crysol (39)). An apparent mass of 113 kDa was calculated for the ternary complex. The discrepancy between this mass and the predicted 122-kDa apparent mass of a 2:2:1 FGF1⅐FGFR2c⅐dp12 complex is likely to be a consequence of the reduced hydrodynamic volume of the components when they are brought closely together into a multicomponent complex. In particular, the small dp12 saccharide is surrounded by protein within the ternary complex and so probably contributes less to the overall hydrodynamic volume (and therefore apparent mass) than would be predicted from its relatively large Stokes radius. Ternary complexes prepared under exactly the same conditions have been prepared by Harmer et al. (29) (using heparin dp10), with a 2:2:1 stoichiometry confirmed by mass spectrometry and analytical ultracentrifugation.
Octasaccharides Are the Shortest Heparin Fragments Capable of Dimerizing FGF1 and Forming 2:2:1 FGF1⅐FGFR2c⅐Heparin Complexes-We investigated the ability of a range of sized heparin saccharides to bind FGF1 and FGFR2c. Incubation with heparin dp4 had little effect on the elution position of FGF1 (Fig. 3A). In contrast FGF1 was split equally by dp6 into heparin-bound and -unbound peaks (Fig. 3A). This would be expected for a 1:1 FGF1⅐dp6 complex, because incubations were carried out in a 2:1 ratio of FGF1 to saccharide. When FGF1 was incubated with heparin dp8, a major peak was observed with a significantly larger hydrodynamic volume (R S ϭ 29.2 Å) than the FGF1⅐dp6 complex (23.9 Å, Fig. 3A). This peak is likely to represent a 2:1 association of FGF1⅐dp8 as the apparent mass (42 kDa) was 16 kDa larger than that observed for the 1:1 FGF1⅐dp6 complex (28 kDa). In contrast to FGF1, the elution position of FGFR2c was unaffected by heparin dp6, indicating that no stable complex formed (Fig. 3B). However, when incubated with heparin dp8 a peak consistent with a 1:1 FGFR2c⅐dp8 complex was observed. An octasaccharide has previously been reported as the minimal heparin length for binding to FGFR1c (40) and FGFR4 (41). The FGFR2c⅐dp12 complex also appeared to have a 1:1 stoichiometry (R S ϭ 33.1 Å, apparent mass 57 kDa, data not shown). were incubated with (thick black line) or without (thick gray line) 1 nmol of AMAC-tagged heparin dp12 with the protein absorbance at 280 nm monitored. Additionally, elution of AMAC-dp12 in the FGF1 incubation was followed by fluorescence detection (thin black line). B, 2 nmol of FGF1 and FGFR2c were incubated with (thick black line) or without (thick gray line) 1 nmol of AMAC-tagged heparin dp12 with the protein absorbance at 280 nm monitored. Additionally, elution of AMAC-dp12 in the FGF1 incubation was followed by fluorescence detection (thin black line).

Heparin-mediated FGF1 Dimerization and Signal Complex Formation
The same sized heparin oligosaccharides were tested for their ability to promote ligand-receptor binding (Fig. 4). Heparin dp6, which bound FGF1 in a 1:1 ratio (23.9 Å, Fig. 3A), formed a complex with a larger hydrodynamic volume when incubated with both FGF1 and FGFR2c (30.6 Å). This indicates that, although FGFR2c cannot bind to FGF1 or dp6 alone, it can bind to FGF1⅐dp6 to form a 1:1:1 complex. We observed no tendency for these 1:1:1 ternary complexes to associate with one another to form structures with a 2:2:2 stoichiometry, like that reported by Schlessinger et al. (26). The elution profile for the FGF1⅐FGFR2c⅐dp8 incubation was clearly resolved into two peaks with different hydrodynamic properties (Fig. 4). The apparent mass estimates (84 kDa) of the early eluting peak suggests that it represents a 2:2:1 FGF1⅐FGFR2c⅐dp8 ternary complex (TABLE ONE). A small subpopulation of our octasaccharides may have lacked one or more critical sulfate groups. The later eluting peak could represent an FGF1⅐FGFR2c⅐dp8 complex to which one or more of the protein com-ponents was unable to bind. The heparin dp10 sample formed one major peak with a Stokes Radius of 40.6 Å (Fig. 4), almost identical to the 40.9 Å predicted using Crysol (39). This structure has been extensively characterized previously as 2:2:1 FGF1⅐FGFR2c⅐dp10 (27,29).
Octasaccharides Are the Shortest Heparin Fragments Capable of Stimulating FGF1-mediated Mitogenesis through FGFR2c-Heparin dp8 was the shortest oligosaccharide capable of forming 2:2:1 FGF1⅐FGFR2c⅐heparin complexes (Fig. 4). In contrast, heparin dp6 formed ternary complexes with a 1:1:1 stoichiometry. To investigate the relevance of these complexes to FGF1 signaling we used a viable cell assay to quantify the effects of various sized heparin oligosaccharides on FGFR2c-mediated cell proliferation (Fig. 5). BaF-32 cells were transfected to express FGFR2c as their sole FGF receptor. These cells were deficient for HS synthesis, and therefore mitogenesis in response to FGF1 only occurred when exogenous HS or heparin was added to the media. At the heparin concentrations used in this assay both dp4 and dp6 had little effect on FGF1-mediated cell proliferation. In contrast, FGF1 with either heparin dp8 or dp10 raised cell numbers significantly in comparison to FGF1 alone. At a 2 g/ml concentration both dp8 and dp10 were almost as effective as the HS control (10 g/ml) at promoting FGF1 activity. These data suggest that if 1:1:1 FGF1⅐FGFR2c⅐heparin complexes form at the cell surface, they are incapable of mediating FGF1 signal transduction.
Selectively Desulfated Heparin 12-mers Bind FGF1 Monomerically and Form 1:1:1 FGF1⅐FGFR2c⅐Heparin Ternary Complexes-To investigate how the sulfation pattern of heparin contributed to its binding of FGF1 and FGFR2c, we prepared a range of heparin dodecasaccharides that were deficient for sulfation at either the N-, 6-O-, or 2-O-positions. The Stokes radii of these saccharides on the Superdex 200 column (19.1-20.7 Å, TABLE TWO) were close to that of the untreated dp12 sample (21.0 Å), showing that the hydrodynamic properties of these saccharides were largely unaffected by selective desulfation. Their disaccharide composition was assessed to determine the selectivity and efficiency of the desulfation procedures (data not shown). The N-de-  sulfated/N-reacetylated (DNS) dp12 had its N-sulfated disaccharide content reduced from 96% to just 2% with negligible loss of O-sulfation. Similarly, the 2-O-desulfated (D2S) sample had its 2-O-sulfate content reduced from 90% of disaccharides to just 5%. However, whereas N-sulfation was essentially unaffected in the D2S dp12, there was some loss of 6-O-sulfates (from 88% to 49%). The partially 6-O-desulfated (D6S) dp12 saccharide had its 6-O-sulfate content reduced from 88% to 21% (ϳ24% of the level of the parent heparin) with negligible loss of N-and 2-O-sulfation.
All three selectively desulfated dp12 saccharides were found to form complexes with FGF1 (Fig. 6A). However, the Stokes radii of these complexes (24.8 -27.6 Å, TABLE ONE) were all significantly less than that of the 2:1 complex formed between FGF1 and the native dp12 (32.6 Å, Fig.  2A). It is likely that these saccharides are binding FGF1 in 1:1 complexes and that, unlike with the parent heparin dp12 saccharide, this interaction precludes the binding of a second FGF1 molecule. The affinity of FGFR2c for the desulfated dp12 saccharides was also assessed, but in each case, unlike with the parent heparin dp12, there was no evidence for a stable interaction (data not shown).
Having determined that the selectively desulfated heparin dp12s failed to dimerize FGF1 and were unable to bind FGFR2c, we investigated whether they could still mediate binding between FGF1 and FGFR2c. The DNS-dp12 sample formed 1:1 complexes with FGF1 of which the Stokes radius was 25.3 Å (Fig. 6B and TABLE ONE). When incubated with both FGF1 and FGFR2c a larger complex was observed (34.7 Å). This complex eluted much later than the 2:2:1 complex formed by the unmodified parent dp12, and had an apparent mass consistent with a 1:1:1 ternary complex (64 kDa, Fig. 6B and TABLE ONE). Equivalent results were also obtained with the D2S and D6S heparin dp12 samples (data not shown).

The Effect of Oligosaccharide Concentration on FGF1 Dimerization and Ternary
Complex Formation-Our data suggest that the formation of a 2:1 FGF1⅐saccharide complex is necessary for subsequent receptor dimerization. However, saccharide (HS) at the cell surface will be much more abundant than the FGF1 ligand. Therefore we investigated the effect that heparin dp12 concentration had on the stoichiometry of FGF1⅐dp12 complexes (Fig. 7). As observed previously ( Fig. 2A), the incubation of FGF1 and dp12 at a 2:1 molar ratio resulted in full recruitment of the FGF1 into 2:1 complexes (R S ϭ 32.6 Å). At a 1:1 ratio of incubation there should have been sufficient saccharide present to bind all of the FGF1 in 1:1 complexes; however, the Stokes radius of the observed peak (32.4 Å) suggests that 2:1 FGF1⅐dp12 complexes still predominated. Although for this incubation no 1:1 peak was present, we observed that the major peak was broader and eluted slightly later from FIGURE 5. A heparin octasaccharide is necessary to stimulate mitogenic signaling of FGF1 through FGFR2c. BaF-32 cells transfected with FGFR2c were serum-starved for 2 days, washed three times with phosphate-buffered saline, then seeded at a low density (2.5 ϫ 10 5 cells/ml) in 96-well plates (see "Experimental Procedures"). Cells were then incubated for 4 days with 100 ng/ml FGF1 in the presence of various concentrations of heparin dp4 (white columns), dp6 (gray columns), dp8 (striped columns), or dp10 (black columns). Viable cell number was assayed using the CytoLite TM kit (100% ϭ CytoLite TM incorporation for FGF1 with porcine mucosal HS [10 g/ml], 0% ϭ CytoLite TM incorporation for FGF1 alone). This chart represents one of three experiments with each data point performed in triplicate. the Superdex 200 column compared with the 2:1 incubation peak. These effects were further apparent when the molar concentration of heparin dp12 was increased above that of the FGF1 (1:4 ratio FGF1 to dp12). The Stokes radius of this peak (31.2 Å, TABLE ONE) was also much too large to be considered a 1:1 complex (the 1:1 complex with DNS⅐dp12 was 25.3 Å, Fig. 6A) and still had considerable overlap with the 2:1 FGF1⅐dp12 elution profile. Therefore, even with eight times the required heparin dp12 present, the 2:1 FGF1⅐dp12 complex predominated.

DISCUSSION
The FGF signaling species at the cell surface is a ternary complex formed between the FGF ligand, its receptor, and heparin/HS (42). Two separate models of FGF ternary complexes have been published. Schlessinger, Mohammadi, and colleagues presented the crystal structure of a symmetrical dimer formed by two 1:1:1 FGF2⅐FGFR1c⅐decasaccharide complexes (26). In contrast, Pellegrini and colleagues (27) described an asymmetrical structure in which two FGFR2c subunits were bridged by an FGF1 dimer formed upon a single heparin decasaccharide. This 2:2:1 complex was present in crystals as part of a higher oligomeric structure defined by intermolecular crystal packing interactions (28). The higher oligomeric structure also allowed a further 2:2:1 complex to be tentatively identified as a candidate for a structure existing in solution; this involved direct interactions between the two FGF⅐FGFR half-complexes, resembling those of the 2:2:2 complex described by Schlessinger et al. (26,28). Further analysis suggested that, by varying the conditions of preparation, both architectures can be formed from the same proteins (29). However, in solution the complex following the Schlessinger preparation contains just one heparin chain (29). Therefore, there appears to be two potential architectures for the 2:2 FGF⅐FGFR signaling complex, both of which can be formed upon a single HS chain in vivo. We have expanded on these studies by using a variety of defined heparin fragments in a SEC approach to investigate the role that heparin plays in the assembly of 2:2 FGF⅐FGFR complexes.
We observed that FGF1⅐FGFR2c⅐heparin ternary complexes formed spontaneously in solution. The Stokes radius and apparent mass calculations from SEC both suggest that these complexes had a 2:2:1 stoichiome-try. In addition, by simultaneously measuring protein absorbance (280 nm) and saccharide fluorescence (AMAC tagging), we demonstrated that at a 2:2:1 FGF1:FGFR2c:dp12 incubation ratio all of the protein and saccharide was colocalized in the ternary complex (Fig. 2B). Ternary complexes pooled from an identical column were recently characterized by both analytical ultracentrifugation and mass spectrometry as having a 2:2:1 stoichiometry (29). These data offer compelling evidence that, when formed in solution, 2:2:1 FGF1⅐FGFR2c⅐heparin oligosaccharide complexes predominate. FGF1 has previously been crystallized in a 2:1 complex with heparin decasaccharide (25). Again, using a combination of Stokes radius, apparent mass calculation, and ratio of incubation (with AMAC-tagged saccharide, Fig. 2A) we demonstrated that a 2:1 FGF1⅐dp12 complex predominated in solution.
A range of size-defined heparin oligosaccharides were tested both for their ability to dimerize FGF1 and their ability to bind FGFR2c in ternary complexes with FGF1. In all cases, the heparin fragments that dimerized FGF1 formed 2:2 complexes between FGF1 and FGFR2c. The shortest saccharide capable of inducing the formation of FGF1 dimers was found to be a heparin dp8 (Fig. 3A), as observed previously (25). This saccharide was also the shortest capable of interacting with FGFR2c alone (1:1 complex, Fig. 3B) and the shortest that formed 2:2:1 complexes (Fig. 4) (29). Notably, octasaccharides were also the shortest heparin fragments that efficiently induced FGF1 mitogenesis in a HSdeficient FGFR2c-transfected cell line (Fig. 5). These data suggest that the ability of a saccharide to dimerize FGF1 may be a critical factor in determining whether it can promote FGF1 signaling by bringing together two FGFR2c subunits.
The heparin dp6 sample bound quantitatively to FGF1 in a 1:1 stoichiometry (Fig. 3A). A study of the heparin-protein interactions in the 2:2:1 Pellegrini-type complex correctly predicted that the hexasaccharide would be the shortest heparin fragment capable of forming 1:1:1 FGF1⅐FGFR2c⅐heparin complexes (Fig. 4) (28). FGFR2c was recruited into 1:1:1 complexes despite showing no stable binding to either FGF1 or heparin dp6 alone (Figs. 2B and 3B). In common with the heparin hexasaccharide, various selectively desulfated heparin dp12 samples also bound FGF1 monomerically (Fig. 6A) and showed no affinity for FGFR2c. In each case these saccharides bound FGF1 and FGFR2c in an apparent 1:1:1 stoichiometry (Fig. 6B). Therefore, saccharides that are incapable of binding FGFR2c alone will bind this receptor in ternary complexes provided they have an affinity for FGF1. This observation suggests that FGF1 binding precedes that of FGFR2c, as has been suggested by other studies of this system using different techniques (43). In the 2:2:1 Pellegrini complex, the FGFR2c subunit interacted with the three sulfate groups on the disaccharide at the non-reducing end of the heparin dp10 (28). Our data show that there could be some flexibility in these interactions as all three selectively desulfated heparin dp12s formed 1:1:1 complexes with FGF1 and FGFR2c (Fig. 6B). Alternatively, the FGF1-FGFR2c interaction may not require a direct heparin interaction with the receptor. One of the FGFR2c subunits in the ternary complex described by Pellegrini et al. (27) was bound to an FGF1 molecule without additional heparin contacts. Heparin-HS interactions with FGF have previously been reported to reduce the conformational flexibility of the protein ligand (44,45). It is possible that binding to heparin saccharides might induce a subtle conformational change within FGF1 that stabilizes a favorable topology of the FGFR binding site.
The 1:1:1 FGF1⅐FGFR2c⅐heparin complexes that we observed showed no inherent affinity for one another. The formation of complexes in which FGFR2c is dimerized therefore appears to be dependent on the ability of a single heparin saccharide to act as a surface upon which FGF1, and subsequently FGFR2c, can be dimerized. This function of heparin is distinct from its ability to promote FGF1-FGFR2c interactions as demonstrated by our observation of 1:1:1 ternary complexes (Figs. 4 and 6). If 1:1:1 FGF1⅐FGFR2c⅐HS complexes form in vivo they may represent an important step toward the formation of full signaling complexes. Binary FGF1⅐FGFR2c half-complexes could migrate along or even between HS chains to sites where they can be dimerized. Such migration of FGF⅐FGFR binary complexes along HS has previously been suggested (40,46). The 2:2:2 FGF:FGFR:heparin structure presented by Schlessinger and colleagues (26) suggests that the dimerization of 1:1:1 "half-complexes" should be extremely favorable. However, saccharides that were incapable of dimerizing FGF1 to form 2:2:1 FGF1⅐FGFR2c⅐heparin complexes also failed to promote the formation of complexes with a 2:2:2 stoichiometry. In this situation, the formation of 1:1:1 complexes could actually be inhibiting the dimerization of the FGF1⅐FGFR2c units. A previous study using mass spectrometry suggested that the Schlessinger et al. (29) structure may well be an asymmetric 2:2:1 complex disordered in the crystals around the 2-fold axis, thus giving an apparent 2:2:2 stoichiometry. Our data agree that the dimerization of FGF⅐FGFR units is most favorable with only one heparin molecule. This contrasts with the "two-end" model of FGFR dimerization (43,47) and suggests that 2:2 FGF⅐FGFR complexes will form within HS chains at any position of appropriate length and sulfation pattern.
A signaling mechanism that requires two binary FGF⅐FGFR complexes to form on adjacent sides of the same stretch of sequence within an HS chain would not appear to be particularly efficient. However, this does not take into account factors that may drive the process of dimerization. We investigated how changing the concentration of heparin dp12 in our incubations affected the formation of 2:1 FGF1⅐dp12 complexes (Fig. 7). We found that even at an 8:1 molar ratio of heparin chains to FGF1, dimerization of FGF1 upon a single heparin saccharide was still the predominant mode of interaction. Excess saccharide caused the FGF1⅐dp12 peak to become broader and elute slightly later from the column; however, a 1:1 FGF1⅐dp12 peak was not apparent. Although the high saccharide concentration appeared to have an affect on the stability of the 2:1 FGF1⅐dp12 complex, the eluted material still had significant overlap with the 2:1 peak and the Stokes radius was much closer to that of a 2:1 complex than a 1:1 complex. These heparin dp12s were highly sulfated and so presumably most saccharide chains were capable of meeting the sulfate requirements for FGF1 binding. Despite this fact, heparin chains that were already occupied by a single FGF1 molecule appeared to have a higher affinity for free FGF1 than unoccupied chains. This could be evidence for a cooperative mechanism of FGF1 dimerization.
Crystal structures of FGF1 dimers formed upon heparin decasaccha-rides revealed no direct protein-protein contact (25). If cooperativity drives the formation of these complexes it must be mediated through changes in the conformation of the heparin chain. In solution heparin adopts a helical conformation with the sulfate groups of contiguous disaccharide units alternating between opposite "faces" of the helical axis (24). A pronounced kink in the helical axis of heparin oligosaccharides appears to be a common feature of its interaction with FGFs (21). If this kink is induced upon binding of the first FGF molecule, the distortion to the helical axis may realign sulfate groups on the opposite face of the heparin saccharide in such a way as to create a particularly high affinity binding site for a second FGF. This would explain the preference for FGF1 dimerization upon the same heparin chain when the molar concentration of heparin far outweighs that of FGF1 (Fig. 7). A cooperative mode of FGF dimerization could drive the formation of 2:2:1 FGF⅐FGFR⅐HS complexes at the cell surface where HS chains are abundant (Fig. 8).
In this study we describe the use of the SEC technique to study the heparin-mediated interaction of FGF1 with FGFR2c. This approach has several advantages over more established techniques (48). Hepa-rin⅐protein complexes were both formed and resolved in a physiologically relevant buffer solution. Direct visualization of protein by absorbance at UV wavelengths allowed us to keep our complexes free in solution without having to immobilize components or bind them with antibodies or other detecting agents. Alternatively, fluorescent tagging of saccharides can be used to detect complexes formed between components incubated at physiologically relevant concentrations. The SEC column proved fast and simple to run with good reproducibility. Calibration using protein standards gave mass estimates and Stokes radii for our heparin⅐protein complexes that, when combined with information on the molarity of the incubated components, allowed us to propose stoichiometries. We anticipate that this technique could be easily adapted to study a wide range of protein⅐GAG complex interactions and to screen potential inhibitors of such interactions.