Direct Binding of Integrin αvβ3 to FGF1 Plays a Role in FGF1 Signaling*

Integrins play a role in fibroblast growth factor (FGF) signaling through cross-talk with FGF receptors (FGFRs), but the mechanism underlying the cross-talk is unknown. We discovered that FGF1 directly bound to soluble and cell-surface integrin αvβ3 (KD about 1 μm). Antagonists to αvβ3 (monoclonal antibody 7E3 and cyclic RGDfV) blocked this interaction. αvβ3 was the predominant, if not the only, integrin that bound to FGF1, because FGF1 bound only weakly to several β1 integrins tested. We presented evidence that the CYDMKTTC sequence (the specificity loop) within the ligand-binding site of β3 plays a role in FGF1 binding. We found that the integrin-binding site of FGF1 overlaps with the heparin-binding site but is distinct from the FGFR-binding site using docking simulation and mutagenesis. We identified an FGF1 mutant (R50E) that was defective in integrin binding but still bound to heparin and FGFR. R50E was defective in inducing DNA synthesis, cell proliferation, cell migration, and chemotaxis, suggesting that the direct integrin binding to FGF1 is critical for FGF signaling. Nevertheless, R50E induced phosphorylation of FGFR1 and FRS2α and activation of AKT and ERK1/2. These results suggest that the defect in R50E in FGF signaling is not in the initial activation of FGF signaling pathway components, but in the later steps in FGF signaling. We propose that R50E is a useful tool to identify the role of integrins in FGF signaling.

Fibroblast growth factors (FGFs) 2 constitute a family of heparin-binding polypeptides involved in the regulation of biological responses such as growth, differentiation, and angiogenesis (1)(2)(3)(4). The FGF family currently consists of 22 members with FGF1 (acidic FGF) and FGF2 (basic FGF) the most extensively studied. The biological effects of FGFs are mediated by four structurally related receptor tyrosine kinases designated FGFR1-4. The binding of FGF to its receptor results in receptor dimerization and subsequent autophosphorylation of specific tyrosine residues within the cytoplasmic domain (1)(2)(3)(4). Activation of the receptor allows proteins containing Src homology 2 or phosphotyrosine binding domains to bind to sequence recognition motifs in the FGFR, resulting in phosphorylation and activation of these proteins (5). This leads to the activation of intracellular signaling cascades. The main signaling cascade activated through the stimulation of FGFR is the Ras/MAPK pathway.
FGF signaling enhances multiple biological processes that promote tumor progression (6). Therapies targeting FGF receptors and/or FGF signaling not only affect the growth of the tumor cells but also modulate tumor angiogenesis (7). FGF1 and FGF2 are responsible for resistance to chemotherapeutic agents in cancer (8 -11) and are also pro-inflammatory growth factors that play a role in pathological angiogenesis in chronic inflammatory diseases (12). Thus FGF signaling is a potential therapeutic target for cancer and pathological angiogenesis in chronic inflammatory diseases.
It has been proposed that cross-talk between integrins and growth factor receptors plays a critical role in growth factor signaling (13), but the specifics of the cross-talk are unclear. Integrins are a family of cell adhesion receptors that recognize extracellular (ECM) ligands and cell-surface ligands (14). Integrins are transmembrane ␣-␤ heterodimers, and at least 18 ␣ and 8 ␤ subunits are known (15). Integrins play an important role in anchorage-dependent cell survival and proliferation (16). Integrin-stimulated pathways are very similar to those mediated by growth factor receptors and are intimately coupled with them. Many cellular responses to soluble growth factors, such as epidermal growth factor, platelet-derived growth factor, and thrombin, are dependent upon the adherence of the cell to ECM ligands via integrins. FGF2-induced angiogenesis requires integrin signaling from the ECM (cross-talk between integrins and FGFRs). Indeed mAb against integrin ␣v␤3 blocks FGF2-induced angiogenesis (17,18).
It has been reported that substrate-bound FGF2 promotes endothelial cell adhesion by directly interacting with integrin ␣v␤3 (19) and induces endothelial cell proliferation, motility, and the recruitment of FGFR1 in the cell substrate contact (20). However, because heat-denatured FGF2 still supports integrin binding (20), it is unclear whether this interaction is biologically relevant, or how integrins interact with FGF2. It has also been reported that FGF2 binds to fibrinogen, and this interaction enhances FGF2-mediated endothelial cell proliferation and subsequent co-localization of ␣v␤3 and FGFR1 (21,22). However, because FGF1 does not bind to fibrinogen (23), the FGFintegrin cross-talk model, mediated by FGF-fibrinogen binding, cannot be applied to FGF1 and also perhaps not to other members of the FGF family. Thus we hypothesized that there may be an alternative model by which FGF1 might be able to cross-talk with the integrin signaling pathway.
In this paper, we show that integrin ␣v␤3 directly and specifically binds to FGF1. We found that the integrin-binding site overlaps with the heparin-binding site by using docking simulation and mutagenesis. One integrin-binding defective mutant (R50E) showed markedly reduced ability to induce cell proliferation and migration, whereas R50E induced FGFR1 phosphorylation, FRS2␣ phosphorylation, and AKT and ERK1/2 activation. We propose that the direct binding of integrins to FGF is a mechanism of integrin-FGFR cross-talk.

Plasmid Construction, Protein Expression, and Purification of the WT and Mutant FGF1
The human FGF1 cDNA was amplified using PCR with human placenta library as a template. A BglII restriction site was introduced at the 5Ј end and an EcoRI site at the 3Ј end of the cDNA fragment with following primers: 5Ј-GCAGATCT-TTTAATCTGCCTCCAGGGAAT-3Ј and 5Ј-GCGAATTCT-TAATCAGAAGAGACTGGCAG-3Ј. The resulting fragments were digested with BglII and EcoRI and then subcloned into the BamHI/EcoRI sites of the pGEX-2T (Amersham Biosciences) vector. Site-directed mutagenesis was performed using the QuickChange method (28). The presence of the mutations was verified by DNA sequencing. The WT FGF1 and its mutants were expressed in Escherichia coli BL21 (DE3) and purified by glutathione affinity chromatography as described by the manufacturer's instructions (GE Healthcare). To remove endotoxin, GST-FGF1 fusion protein on glutathione-agarose was extensively washed with 1% Triton X-114 in PBS. WT, the R50E mutant, and the 3xA (E101A/Y108A/N109A) FGF1 mutant were further purified using a heparin-Sepharose column (Amersham Biosciences) after removing the GST tag by thrombin. The 4xE (K127E/K128E/K133E/R134E) mutant was purified by gel filtration. The FGF1 preparations were more than 90% homologous in SDS-PAGE.

Synthesis of the FGFR1 D2D3 Fragment
A DNA fragment encoding amino acid residues 140 -365 of the immunoglobulin-like D2 and D3 domains of FGFR1 was amplified by PCR with the full-length human FGFR1 cDNA in the pcDNA3 vector (gift from Ann Hanneken, the Scripps Research Institute, La Jolla, CA) as a template. A BamHI restriction site was introduced at the 5Ј end and an XhoI site at the 3Ј end of the cDNA fragment with the following primers: 5Ј-GCGGATCCACAGATAACACCAAACCAAACC-3Ј, 5Ј-GCCTCAGTCACCTCTCTTCCAGGGCTTCC-3Ј. The resulting cDNA fragment was subcloned into the BamHI/XhoI sites of the vector pET21a, and transformed into BL21 (DE3). The protein was expressed as an insoluble protein and refolded as described (29). The refolded protein was purified by affinity chromatography on heparin-Sepharose to enrich the properly folded protein. Bound protein was eluted with 1 M NaCl.

Surface Plasmon Resonance (SPR) Analysis
Heparin-FGF Interaction-Biotinylated heparin (Sigma) in Biacore HBS-EP buffer (0.01 M Hepes, pH 7.4, 0.15 M NaCl, 3 mM EDTA, and 0.0005% of surfactant P20) was immobilized by a streptavidin sensor chip (Biacore SA Chip) by injecting at 2 l/min for 15 min. One response unit difference represents about 1 pg/mm 2 of the analytes on the surface matrix of the sensor chip. 5-Fold serially diluted WT, R50E, and 3xA ranging from 800 nM to 51.2 pM and 2-fold serially diluted 4xE ranging from 1.6 M to 50 nM in HBS-EP were injected at 50 l/min for 3 min.
Integrin-FGF Interaction-Soluble ␣v␤3 was immobilized on the CM5 sensor chip using a standard amine coupling procedure (30). The WT and R50E FGF1 were individually 2-fold serially diluted from 8 M to 125 nM in HBS-P buffer (0.01 M Hepes, pH 7.4, 0.15 M NaCl, and 0.0005% of surfactant P20) with 1 mM of Mn 2ϩ , and the 3xA FGF1 was 2-fold serially diluted from 6 M to 93.75 nM in the same buffer. Samples were injected at 50 l/min for 1.8 min. The HBS-P buffer with 1 mM of Mn 2ϩ was then injected at 50 l/min for 3 min to allow the bound FGF1s to dissociate from the integrin.
FGF1-FGFR1 D2D3 Interaction-About 1500 response units of FGFR1 D2D3 was immobilized to a sensor chip CM5 by using a standard amine coupling method. WT or mutant FGF1 was 2-fold serially diluted from 800 to 50 nM in HBS-P buffer containing 1 mM of Mn 2ϩ and injected at a flow rate 30 l/min for 3 min. The same buffer was injected at the same flow rate for 3 min to measure the dissociation of the bound FGF1 from FGFR1 D2D3.

Docking Simulation
In the AutoDock 3.05 program, the ligand is presently compiled to a maximum size of 1024 atoms. The solvent-exposed Mg 2ϩ octahedral vertex was left empty in the model during docking calculations. Atomic solvation parameters and frac-tional volumes were assigned to the protein atoms by using the AddSol utility, and grid maps were calculated by using Auto-Grid utility in AutoDock 3.05. A grid map with 127 ϫ 127 ϫ 127 points and a grid point spacing of 0.603 Å included the whole metal ion-dependent adhesive site-containing face of the I-like domain of ␤3 and the ␤-propeller domain containing repeats 2-4, which are large enough to accommodate the FGF1 structure. Kollman "united-atom" charges were used. AutoDock 3.05 uses a Lamarckian genetic algorithm (LGA) that couples a typical Darwinian genetic algorithm for global searching with the Solis and Wets algorithm for local searching. The LGA parameters were defined as follows: the initial population of random individuals had a size of 50 individuals; each docking was terminated with a maximum number of 1 ϫ 10 6 energy evaluations or a maximum number of 27,000 generations, whichever came first; mutation and crossover rates were set at 0.02 and 0.80, respectively. An elitism value of 1 was applied, which ensured that the top ranked individual in the population always survived into the next generation. A maximum of 300 iterations per local search was used. The probability of performing a local search on an individual was 0.06, whereas the maximum number of consecutive successes or failures before doubling or halving the search step size was 4. This set of parameters was used for all dockings.

Integrin Binding Assays
Cell adhesion and soluble integrin binding assays were performed as described previously (24). Native or heat-denatured (70°C for 10 min) FGF in 0.1 M carbonate buffer, pH 9.4, was incubated in a polystyrene 96-well non-tissue culture plate overnight at 4°C. Unbound FGF was removed, and 200 l of 0.1% BSA in PBS was added and incubated for 60 min at room temperature. The wells were washed with PBS, and soluble integrin ␣v␤3 in 50 l in Hepes-Tyrode's buffer supplemented with 1 mM Mn 2ϩ was added to the wells and incubated at room temperature for 60 min. Then non-bound soluble integrin was removed by rinsing the wells with the same buffer. Horseradish peroxidase (HRP)-conjugated anti-His tag mouse IgG was added to the wells and incubated for 60 min. Non-bound antibodies were removed by rinsing the wells with the same buffer, and bound integrins were quantified by measuring the absorbance at 450 nm developed from adding the substrate 3,3Ј,5,5Јtetramethylbenzidine of HRP.

DNA Synthesis
DNA synthesis was measured by BrdUrd incorporation. Balb 3T3 cells were plated on sterile coverslips in 6-well culture plates and serum-starved in DMEM supplemented with 0.4% FCS for 48 h. They were stimulated with 5 ng/ml WT FGF1 and its mutants for 24 h in the presence of 5 g/ml heparin. BrdUrd (10 g/ml) was added to the medium for the last 6 h of the incubation. Cells were then fixed with 70% ethanol and incubated with 2 N HCl. After the medium was neutralized with 0.1 M borate buffer (pH 0.5), the cells were incubated with anti-BrdUrd antibody (Pharmingen). BrdUrd-incorporated cells were stained with HRP-conjugated secondary antibody (Bio-Rad) and metal-enhanced 3,3Ј-diaminobenzidine substrate kit (Pierce). 3,3Ј-Diaminobenzidine-positive and -negative cells were counted from the digital images of three independent fields.

Proliferation Assay Using BaF3 Cells Expressing the Human FGFR1c Isoform
Mouse pro-B BaF3 cells that express human FGFR1c (BaF3-FR1c, kindly provided by David Ornitz, Washington University, St. Louis) were maintained in a medium containing 0.5 ng/ml IL-3 as described (31). For proliferation assays, cells were maintained with WT or mutant FGF1 instead of IL-3, and cell proliferation was measured by MTS assays.

In Vitro Wound Healing Assay
Balb 3T3 cells were plated into 6-well cell culture plates. Cells were allowed to grow in DMEM containing 10% FCS overnight, and then cells were washed with serum-free medium and starved for 24 h. A scratch was made across the cell layer using a pipette tip. After washing with serum-free medium twice, DMEM containing 10 ng/ml WT or mutant FGF1 together with 5 g/ml heparin was added to the cells. The wounded areas were quantified by using ImageJ.

Chemotaxis
A polycarbonate filter of 8 m pore size of the Transwell insert was coated with 10 g/ml fibronectin (Sigma) overnight at 4°C. After washing, the insert was placed into a 24-well cell culture plate, and the lower portion of the plate was filled with 600 l of serum-free DMEM containing 5 ng/ml WT or mutant FGF1. Balb 3T3 cells (10 5 cells/filter) were plated on the filter and incubated at 37°C for 24 h, and cells were visualized by crystal violet staining (0.5% crystal violet in 50 mM borate, pH 9.0, and 2% ethanol). The uncoated side of each filter was wiped with a cotton swab to remove cells that had not migrated through the filter. Chemotaxed cells were counted from the digital images of the stained cells, determining the mean num-

RESULTS
FGF1 Specifically Binds to Integrin ␣v␤3-We tested whether FGF1 directly interacts with integrins in vitro. We found that recombinant soluble ␣v␤3 bound to immobilized WT FGF1 in a dose-dependent manner but did not bind to heat-denatured FGF1 in ELISA-type integrin binding assays (Fig. 1a). These results suggest that FGF1 binds to integrin ␣v␤3, and this interaction requires an intact three-dimensional structure of FGF1. EDTA and mAb 7E3 (anti-␤3) blocked the binding of soluble ␣v␤3 to FGF1, suggesting that binding to FGF1 is cation-dependent and specific to ␣v␤3 (Fig. 1b). An SPR analysis of the FGF1-␣v␤3 interaction showed that FGF1 binds to ␣v␤3 at a high affinity (see below).
We next tested whether cell-surface ␣v␤3 binds to FGF1. We found that K562 erythroleukemic cells (␣5␤1 ϩ ) expressing exogenous ␣v␤3 (␣v␤3-K562, ␣5␤1ϩ/␣v␤3ϩ) cells adhered to FGF1, but mock-transfected K562 cells showed only weak adhesion to FGF1 (Fig. 1c). This suggests that FGF1 binds to ␣v␤3, but not to ␣5␤1, on the cell surface under the current assay conditions. mAb 7E3 blocked, but mAb KH72 (anti-␣5) did not significantly block, the adhesion of ␣v␤3-K562 cells to FGF1 (Fig. 1d). Consistent with this finding, cyclic RGDfV, which is an antagonist specific to ␣v␤3 (32), blocked the adhesion of ␣v␤3-K562 cells to FGF1 (Fig. 1e). K562 cells that express several ␤1 integrins only weakly adhered to FGF1 (Fig.  1c). These results suggest that ␣v␤3 is a predominant, if not the only, integrin that binds to FGF1 in our assay conditions. Because FGF1-␣v␤3 interaction is blocked by mAb 7E3 that has been mapped in the ligand-binding site of the ␤3 subunit (33,34), and blocked by cyclic RGDfV, it is suggested that FGF1 binds to the ligand-binding site of ␣v␤3. We previously reported that a small disulfide-linked loop in the I-like domain of ␤3 plays a critical role in determining ligand specificity of ␣v␤3 (27). To test if FGF1 binding requires the specificity loop of ␣v␤3, we used a ␤1 mutant in which a disulfide-linked loop of ␤1 (the specificity loop, residues 187-193) was swapped with the corresponding amino acid residues of ␤3 (the CYDMKTTC sequence) (the ␤1-3-1 mutation) (27). ␣v␤1-3-1 has been shown to bind to several ␣v␤3 ligands, including vitronectin, fibrinogen, von Willebrand factor (27,35), and a viral surface protein (36). Nevertheless function-blocking anti-␤1 mAbs such as AIIB2 blocked ligand binding to ␣v␤1-3-1 (27). We found that CHO cells that express ␤1-3-1 (␤1-3-1-CHO) strongly adhered to FGF1, whereas those expressing WT ␤1 (␤1-CHO) did not (Fig. 1f). Furthermore, anti-␤1 mAb AIIB2 FIGURE 2. Identification of the integrin-binding site in FGF1. a, model of FGF1-integrin interaction. Docking simulation of the interaction between FGF1 (PDB code 1AXM) and integrin ␣v␤3 (PDB code 1L5G) was performed as described under the "Experimental Procedures" using AutoDock3. Model a of 1AXM was used for docking. The headpiece of 1L5G was used as a receptor. The pose in the cluster 1 with the lowest docking energy Ϫ26.3 kcal/mol is shown. This pose represents the most stable pose of FGF1 when FGF1 interacts with integrin ␣v␤3. b, positions of amino acid residues that are selected for mutagenesis at the predicted interface between FGF1 and ␣v␤3. Several amino acid residues within the predicted integrin-binding site in FGF1 were selected for mutagenesis. c, positions of the amino acid residues (E102A, Y109A, and N110A) at the FGFR-binding site selected for mutagenesis. Note that the predicted integrin-binding site is distinct from the FGFR-binding site.
effectively blocked the binding of ␤1-3-1-CHO cells to FGF1. This suggests that the specificity loop is critical for FGF1 binding to ␣v␤3 and that FGF1 interacts with ␣v␤3 in a manner similar to those of known ␣v␤3 ligands (e.g. vitronectin). These results also suggest that ␣v␤1 and ␣v␤5 that are expressed in parent CHO and ␤1-CHO cells do not appear to be important for FGF1 binding.
It is possible that integrin ␣v␤3 may not interact with FGF1 in the presence of plasma adhesive proteins, because they may compete with FGF for binding to ␣v␤3. We found that the presence of 10% fetal bovine serum did not suppress ␣v␤3-specific cell adhesion to FGF1 (Fig. 1g). This consistent with the previous reports that integrin-binding sites in serum-adhesive proteins are cryptic (see "Discussion").
Generation of Integrin-binding Defective FGF1 Mutants-To localize the integrin-binding site in FGF1, we used a docking simulation program and site-directed mutagenesis. AutoDock is a set of docking tools widely used for predicting the pose of small ligands bound to receptors (37), and the methods are being extended to predict protein-protein complex poses (38). We performed 50 dockings of the FGF1-integrin ␣v␤3 interaction, each one starting with a random initial position and orientation of FGF1 (PDB code 1AXM) with respect to the integrin ␣v␤3 headpiece (PDB code 1L5G). The results were clustered together by positional root mean square distance into families of similar poses (data not shown). Many of the docking poses clustered well with the lowest docking energy, 26.1 kcal/ mol (cluster 1). These results predict that the docking pose of cluster 1 (Fig. 2a) may represent the most probable stable FGF1 pose upon binding to ␣v␤3. This model predicts that the integrin-binding interface of FGF1 with integrin ␣v␤3 is  . Effect of FGF mutations on the binding to integrin ␣v␤3. a, R50E mutation of FGF1 reduced its binding to integrin ␣v␤3. Amino acid residues in the integrin-binding site or in the FGFR-binding site were mutated individually or in groups. WT FGF1 and the R50E mutant were coated to a plastic plate at various concentrations as indicated, and the binding of soluble ␣v␤3 was determined as described in Fig. 1a. The results show that the R50E mutation reduced ␣v␤3 binding. b, summary of the effects of FGF1 mutations on integrin binding. The results suggest that mutations in the predicted integrin-binding site of FGF1 reduced the binding of soluble ␣v␤3 to FGF1, but mutations in the FGFR-binding site did not. c and d, binding of WT FGF1 (c) and R50E FGF1 (d) to ␣v␤3 in SPR. Soluble ␣v␤3 was immobilized on a CM5 sensor chip. WT and R50E FGF1 were individually 2-fold serially diluted from 8 M to 125 nM in HBS-P buffer with 1 mM of Mn 2ϩ , and the 3xA FGF1 was 2-fold serially diluted from 6 M to 93.75 nM in the same buffer. 4xE did not show any binding to integrin (data not shown).

TABLE 1 Amino acid residues at the predicted interface between FGF1 and integrin ␣v␤3
Amino acid residues of FGF1 or ␣v␤3 that are within a 6-Å distance from ␣v␤3 or FGF1, respectively, were selected using Swiss-pdb viewer (version 3.7). Amino acid residues of FGF1 (bold) were selected for mutagenesis.

FGF1
␣v ␤3 Binding of ␣v␤3 to FGF1 Plays a Role in FGF1 Signaling distinct from the FGFR-binding site (39) but is close to or overlapping with the heparin-binding site (Fig. 2, b and c). We introduced several mutations within the predicted integrin-binding site (Fig. 2b and Table 1) to localize the integrinbinding site in FGF1. We tested the ability of the mutants to bind to ␣v␤3 in ELISA-type binding assays. The Arg-50 to Glu (R50E) mutation effectively reduced the binding of soluble ␣v␤3 to FGF1 (Fig. 3a). Mutating positively charged residues (Lys-127, Lys-128, Lys-133, and Arg-134) in the predicted integrin-binding site to Glu in combination diminished the binding of soluble ␣v␤3 to FGF1 (Fig. 2b and Fig. 3b). Mutating all these residues simultaneously to Glu (designated the 4xE mutation) effectively diminished integrin binding. We predicted that Glu-102, Tyr-109, and Asn-110 are located in the FGFR-binding site from the FGF1-FGFR-2 complex structure (39) and mutagenesis data of FGF2 (40). Mutating these residues to Ala individu-ally or simultaneously (designated 3xA mutant) did not affect integrin binding (Fig. 2c and Fig. 3b), whereas the 3xA mutation effectively suppressed the binding to FGFR1 (below).
We assessed the ability of the FGF1 mutants to bind to ␣v␤3 in surface plasmon resonance (SPR) using a sensor chip immobilized with soluble ␣v␤3. WT FGF1 bound to ␣v␤3 at a high affinity (K D 1.1 M) (Fig. 3c). The R50E mutant showed much lower R max value and enhanced off-rate to ␣v␤3, whereas it has a K D value to ␣v␤3 (1.31 M) similar to WT FGF1 (Fig. 3d). The ability of the 3xA mutant to bind to ␣v␤3 was intact (K D 0.18 M), but the 4xE mutant did not show binding affinity to ␣v␤3 (data not shown)). These results are consistent with those obtained by ELISAtype binding assays.
We assessed the ability of these FGF1 mutants to interact with the FGFR1 fragment (domains 2 and 3, or FGFR1 D2D3) in SPR using a sensor chip immobilized with FGFR1 D2D3. WT FGF1 (K D 30 pM) and R50E FGF1 (K D 14 pM) had similar affinity to FGFR1 (Fig. 4, a and b). The 4xE mutant showed low affinity to FGFR1 (K D 5.5 M; data not shown). The 3xA mutant has a much lower affinity (K D 0.41 mM) (Fig. 4c).
We assessed the ability of the mutants to bind to heparin-Sepharose. WT FGF1, and the R50E and 3xA mutants, bound strongly to heparin-Sepharose and required 1.2 M NaCl for elution, but the 4xE mutant did not show affinity to heparin (Fig. 4d). In SPR R50E (Fig. 4e) had slightly lower affinity to heparin (K D 66 nM) than WT FGF1 (K D 16.6 nM) (Fig. 4f), whereas the slightly reduced affinity of R50E to heparin did not affect its ability to induce FGFR1 activation and ERK1/2 activation (see below). The 3xA mutant (data not shown) showed affinity to heparin (K D 11.5 nM) similar to that of WT FGF1, but the 4xE did not show any affinity to heparin (data not shown).
These results are consistent with the prediction that in mutations in the predicted integrin-binding site in FGF1 compromised integrin binding and heparin binding, but those in the FGFR-binding site affected FGFR binding but did not affect integrin binding. The R50E mutation was unique in that it reduced integrin binding and induced minimal effect on heparin or FGFR binding. Thus we propose that the R50E mutant is useful to determine the potential role of integrin ␣v␤3 in FGF signaling. The 4xE mutation was defective in integrin, heparin, and FGFR1 binding and could be used as a control together with the 3xA mutant. Positions of these mutations in FGF1 are shown in Fig. 2, B and C.

Effect of the Mutations on FGF-induced DNA Synthesis and Cell
Migration-We examined the effect of the FGF1 mutations on FGF1-induced DNA synthesis in Balb 3T3 cells. The R50E, 4xE, and 3xA mutants induced much lower BrdUrd incorporation than WT FGF1 (Fig. 5a), suggesting that these mutants are defective in inducing DNA synthesis.
To test the ability of the FGF1 mutants to induce cell proliferation, we used BaF3 mouse pro-B cells that express human FGFR1c (designated BaF3-FGFR1c) (41), which have been used to sensitively detect cell proliferation that is dependent on exogenous FGF1. We found that WT FGF1 robustly induced proliferation of BaF3-FGFR1c cells, whereas the R50E, 4xE, and 3xA mutants did not induce detectable cell proliferation (Fig. 5b). This is consistent with the results that these mutants were defective in inducing DNA synthesis.
Because FGF1 is a potent inducer of cell migration and chemotaxis (42,43), we tested the ability of the mutants to induce cell migration. Using scratch wound healing assays with Balb 3T3 cells, we found that although WT FGF1 closed the wound by 59% in 24 h, the R50E, 4xE, and 3xA mutants did not close the wound in this time frame (Fig.  5c). We found that the R50E, 4xE, and 3xA mutants induced much lower chemotaxis than WT FGF1 (Fig. 5d). These results suggest that these mutants are defective in inducing cell migration.
Effect of the Integrin-binding Defective Mutations on FGF1 Signaling-Next we studied whether these FGF1 mutants induce FGF signaling inside the cells. The binding of FGF to FGFR leads to tyrosine phosphorylation of the docking protein FRS2␣, which functions as a major mediator of signaling via FGFR (1)(2)(3)(4). This results in recruitment of multiple Grb2-Sos complexes leading to activation of the Ras/MAPK signaling pathway (44). The mitogenic action of the FGFs is mediated in part by the activation of ERK1/2 (45). WT FGF1 induced phosphorylation of FGFR1, FRS2␣, AKT (protein kinase B), and ERK1/2. The 4xE and 3xA  mutants showed little or no ability to induce FGFR1, FRS2␣, or ERK1/2 activation as compared with WT FGF1 and the R50E mutant (Fig. 6), which is consistent with the finding that the 4xE and 3xA mutants are defective in binding to FGFR1. Notably, the R50E mutant induced FGFR1, FRS2␣, AKT, and ERK1/2 activation to the similar extent of WT FGF1 (Fig. 6). These results suggest that the integrin-binding defective R50E FGF1 mutant can induce the initial FGFR activation and subsequent FGF signaling and that the slightly reduced heparin binding to this mutant did not affect it.

DISCUSSION
Although it has been reported that integrins play a critical role in FGF signaling (13), the specifics governing their interaction in this process are still unclear. In this study, we showed that integrin ␣v␤3 directly bound to FGF1 at an affinity comparable with those of known integrin ligands. Heat denaturation of FGF1 destroyed integrin binding, and the amino acid residues of FGF1 that are critical for integrin-binding were discontinuous, indicating that proper folding of FGF1 is required for integrin binding. Because mAb 7E3 and cyclic RGDfV blocked the FGF1-␣v␤3 interaction, it is suggested that FGF1 binds to the ligand-binding site of ␣v␤3. We showed that the specificity loop plays a role in the binding of FGF1 to ␣v␤3, suggesting that FGF1 binds to ␣v␤3 in a manner similar to that of other ␣v␤3 ligands tested (27,35,36). We showed that 10% fetal bovine serum did not suppress ␣v␤3-mediated cell adhe-sion to FGF1. This is consistent with the reports that integrinbinding sites are cryptic in fibrinogen (46), fibronectin (47), and vitronectin (48). These findings suggest that the observed FGF1 binding to ␣v␤3 may happen in biological body fluid. It has been proposed that productive integrin-ligand interactions require ligand "activation" (e.g. by immobilization and proteolysis) in addition to activation of integrins (46). We recently showed that ␣v␤3 binds to the C-terminal fragment of the fibrinogen ␥-chain but not to whole fibrinogen in solution (49).
We localized critical amino acid residues of FGF1 for binding to ␣v␤3 using docking simulation and mutagenesis. We found that the integrin-binding site overlaps with the heparin-binding site. These findings suggest that docking simulation and mutagenesis are useful tools for studying how integrins bind to protein ligands. The R50E mutation affected integrin binding, although its effect on heparin and FGFR binding was minimal, suggesting that this mutation primarily affected integrin binding. The R50E mutant did not induce cell proliferation and migration, whereas R50E induced FGF1 signaling (FGFR1 phosphorylation, FRS2␣ phosphorylation, and ERK1/2). This suggests that the integrin binding may not be required for the early activation of ERK1/2 but is required for the later steps in FGF signaling that leads to cell proliferation and migration. Indeed it has been reported that that transient ERK1/2 activation is not integrin-dependent and not related to cell proliferation, but that sustained ERK1/2 activation is integrin-dependent and directly related to cell cycle entry (50,51). The R50E mutant of FGF1 may be a useful tool to study the role of integrins in FGF signaling. Further studies in this direction are under way.
We showed that WT FGF1 free in solution had a K D of 1.1 M to ␣v␤3 and a K D of around 1 pM to FGFR1. It is thus expected that FGF1 will bind preferentially to FGFR on the cell surface. We expect that the binding of FGF1 to FGFR will facilitate the binding of FGF1 to ␣v␤3 because FGF1 is concentrated on the cell surface through binding to FGFR. Our results predict that the integrin-binding site of FGF1 would be still exposed upon binding to FGFR. The results in this study suggest that integrin ␣v␤3 is a major, if not the only, integrin that binds to FGF1. It has been reported that ␣v␤3 is detectable in NIH3T3 cells (52) and that ␣v␤3 is markedly induced when the Ras-ERK1/2 pathway is activated by Raf oncogene in NIH3T3 cells (53). The R50E mutation is close to the FGF1-FGFR1 domain D2 interface as shown in the crystal structure of the FGF1-FGFR1 complex (PDB code 1EVT). We showed that the R50E mutation did not affect the binding of FGF1 to FGFR1 D2D3. Also we showed that this mutant induced phosphorylation of FGFR1 and FRS2␣ and activation of ERK1/2 and AKT. It is thus unlikely that the R50E mutation suppressed FGFR binding. We propose that the primary defect in the R50E mutant is in its ability to bind to integrins.
In current models of the role of integrins in growth factor signaling, integrins interact with ECM ligands, and the resulting integrin-mediated signaling merges with growth factor-mediated signaling (54). The present results suggest that the direct binding of FGF to integrin ␣v␤3 is critical for proper FGF signaling and that FGF1 itself acts as an integrin ligand. If this is the case, it is likely that integrin-FGFR cross-talk may be mediated by direct binding of FGF to integrins and FGFR. We . Effect of FGF1 mutations on FGF signaling. FCS, 10% FCS. Serumstarved NIH3T3 cells were stimulated with WT or mutant FGF1s (5 ng/ml) in the presence of 5 g/ml heparin for 10 min at 37°C. For FGFR phosphorylation (p-FGFR1), phosphorylated FGFR1 was first immunoprecipitated from lysates (750 l, total 2.6 mg of protein) using anti-phospho-Tyr antibodies and then blotted using anti-FGFR1 antibodies. Whole FGFR1 in cell lysates were detected with anti-FGFR1. For FRS2␣, ERK1/2, and AKT, cell lysates were analyzed by Western blotting with antibodies against phospho-FRS2␣, FRS2␣, phospho-ERK1/2, ERK1/2, phospho-AKT, and AKT, respectively. Results with the 4xE mutant were very similar to those with the 3xA mutant (data not shown).
showed that FGF1 binds to integrins in the presence of 10% serum, consistent with the previous reports that integrin-binding sites in major serum adhesive proteins (e.g. fibrinogen) are cryptic in solution as discussed above. We propose that FGFintegrin interaction occurs in body fluids. Also, we propose that FGF may be concentrated on the cell surface through high affinity binding to FGFR. This would facilitate FGF-integrin interaction on the cell surface. Because FGFR and integrins bind to distinct sites in FGF1, FGFR binding to FGF would not block FGF-integrin interaction. Thus we expect that FGF may interact with integrins on the cell surface under physiological conditions even if the affinity of ␣v␤3 to FGF in solution in SPR is low (K D 1 M). We showed that an integrin-binding defective FGF1 mutant (R50E) was defective in inducing cell proliferation and migration. Taken together, we propose that the direct FGF-integrin interaction on the cell surface is biologically relevant. The present results, however, do not rule out the possibility that ECM-integrin interaction is required for FGF signaling. Integrin antagonists or down-regulation of integrins block both integrin-ECM and integrin-FGF interactions, and do not distinguish the two interactions in FGF signaling. The R50E mutant of FGF1 is thus an important tool that selectively blocks FGF1-integrin interaction in FGF signaling.