Mapping ligand binding domains in chimeric fibroblast growth factor receptor molecules. Multiple regions determine ligand binding specificity.

Fibroblast growth factors (FGFs) mediate essential cellular functions by activating one of four alternatively spliced FGF receptors (FGFRs). To determine the mechanism regulating ligand binding affinity and specificity, soluble FGFR1 and FGFR3 binding domains were compared for activity. FGFR1 bound well to FGF2 but poorly to FGF8 and FGF9. In contrast, FGFR3 bound well to FGF8 and FGF9 but poorly to FGF2. The differential ligand binding specificity of these two receptors was exploited to map specific ligand binding regions in mutant and chimeric receptor molecules. Deletion of immunoglobulin-like (Ig) domain I did not effect ligand binding, thus localizing the binding region(s) to the distal two Ig domains. Mapping studies identified two regions that contribute to FGF binding. Additionally, FGF2 binding showed positive cooperativity, suggesting the presence of two binding sites on a single FGFR or two interacting sites on an FGFR dimer. Analysis of FGF8 and FGF9 binding to chimeric receptors showed that a broad region spanning Ig domain II and sequences further N-terminal determines binding specificity for these ligands. These data demonstrate that multiple regions of the FGFR regulate ligand binding specificity and that these regions are distinct with respect to different members of the FGF family.

Fibroblast growth factors (FGFs) mediate essential cellular functions by activating one of four alternatively spliced FGF receptors (FGFRs). To determine the mechanism regulating ligand binding affinity and specificity, soluble FGFR1 and FGFR3 binding domains were compared for activity. FGFR1 bound well to FGF2 but poorly to FGF8 and FGF9. In contrast, FGFR3 bound well to FGF8 and FGF9 but poorly to FGF2. The differential ligand binding specificity of these two receptors was exploited to map specific ligand binding regions in mutant and chimeric receptor molecules. Deletion of immunoglobulin-like (Ig) domain I did not effect ligand binding, thus localizing the binding region(s) to the distal two Ig domains. Mapping studies identified two regions that contribute to FGF binding. Additionally, FGF2 binding showed positive cooperativity, suggesting the presence of two binding sites on a single FGFR or two interacting sites on an FGFR dimer. Analysis of FGF8 and FGF9 binding to chimeric receptors showed that a broad region spanning Ig domain II and sequences further N-terminal determines binding specificity for these ligands. These data demonstrate that multiple regions of the FGFR regulate ligand binding specificity and that these regions are distinct with respect to different members of the FGF family.
FGFs range in molecular mass from 18 to 29 kDa and show 13-71% amino acid identity (pairwise comparisons of mouse or rat FGFs). Most members of the family share 28 highly conserved amino acid residues, and all FGFs have six identical positions. All members of the FGF family have a high affinity for heparin. Significantly, heparin or heparan sulfate can enhance the high affinity binding of FGFs to their receptors by forming a trimolecular complex with FGF and the FGFR and by decreasing the rate of ligand dissociation (22)(23)(24)(25)(26)(27)(28). The formation of this complex results in receptor dimerization, transphosphorylation, and downstream signaling.
During development, FGFs are involved in several and sometimes opposing functions requiring tight regulation of their activity and specificity. This may be achieved by regulating spatial and temporal expression of FGFs and FGFRs and by regulating binding specificity (21, 29 -34). A major determinant of ligand binding specificity is alternative splicing in the C terminus of Ig domain III of FGFRs 1-3. This splicing event is tissue-specific and is likely to regulate important signaling events across epithelial/mesenchymal boundaries (35)(36)(37)(38). The "b" (epithelial) splice form of FGFR2 (FGFR2b) can be activated by FGF7 and FGF10, ligands produced in mesenchymal tissue. These ligands show no activity toward c (mesenchymal) spliced receptors. Conversely, FGF8 activates c-spliced FGFR2 (FGFR2c) but shows no activity toward FGFR2b (21,39). 2 Notably, FGF8 expression is often restricted to epithelial tissue such as the apical ectodermal ridge of the developing limb bud (40,41).
The C-terminal region of Ig domain III is clearly important for ligand binding and shows specificity toward different ligands. For example, specific mutations in this region in FGFR2 can decrease the binding of FGF2 without affecting the binding of FGF1 or FGF7 (42). The b splice form of FGFR3 (FGFR3b) also has unique properties in that it can only be activated by FGF1, which shows little specificity toward any receptor, and FGF9, which shows no activity toward FGFR1b and FGFR2b (21,43,44). Alternative splicing of Ig domain III can also lead to truncation of the transmembrane and intracellular regions ("a" splice form) creating a secreted FGF-binding protein (45). The physiological relevance of this form of the receptor is not known; however, recent studies demonstrated that in transgenic mice overexpression of a soluble FGFR extracellular domain can result in dramatic developmental defects (46). Another major alternative splicing event truncates Ig domain I. Receptor forms lacking Ig domain I have a higher affinity for some FGF ligands, although it is not known if ligand binding specificity is affected by this splicing event (47)(48)(49). Truncation of Ig domain I also correlates with the progression of several tumors toward malignancy, suggesting a functional difference between long and short receptors (50,51).
In addition to alternative splicing, differential binding specificity also exists when similarly spliced receptors are compared. For example, FGF8, FGF17, and FGF18 can activate FGFR2c and FGFR3c but show little activity toward FGFR1c (21,39). 3 FGF6 can activate FGFR1c and FGFR2c but shows little activity toward FGFR3c (21,52). Although much is known about how alternative splicing affects ligand binding specificity, the molecular basis of binding specificity between similarly spliced receptors is not known. In this study we examine determinants of ligand binding specificity by comparing the binding activity of FGFR1c and FGFR3c. We have generated a series of soluble chimeric receptors and have assayed these molecules for binding to FGF1, FGF2, FGF8, and FGF9. These data demonstrate that the FGFR contains two ligand binding regions. FGF1 binds either region in either receptor, FGF2 preferentially recognizes distal sequence (Ig domain II-III)  Soluble Receptor Constructs-Fusion proteins between the extracellular domain of FGFR1 with human placental alkaline phosphatase (AP) were constructed previously (54). An XhoI fragment from FRAP (54) containing the entire FGFR1 extracellular domain fused to AP was subcloned into the SalI site of Bluescript SK and then cut with HindIII and XhoI and cloned into the pCDNA I expression vector (Invitrogen Inc.). A similar chimeric protein for FGFR3 was described previously (15). To construct chimeric molecules, the 5Ј end from FGFR1 or FGFR3 was used as a template for PCR amplification or as a source of restriction fragments. In all oligonucleotide sequences listed, the FGFR3 sequences are represented with uppercase letters and FGFR1 sequences are in lowercase letters. The 3Ј-flanking region of each FGFR chimera was generated by using primers DO51 for FGFR1 (5Ј-cggaagatctctccaggtagagcg-3Ј) and DO81 (5Ј-CGGAAGATCTGCCAGCCTCAT-CAGT-3Ј) for FGFR3. Both 3Ј primers contain a common BglII restriction site (underlined) for subcloning in frame with AP (15,22). The various fragments, generated by PCR amplification and ligation, were subcloned into a T vector (Promega Inc.), sequenced, and then subsequently cloned between the HindIII and BglII site of the FRAP pcDNA I expression vector. FGFR-AP fusion proteins were transiently expressed in COS-7 cells as described below (55).
Soluble Receptor Forms Containing Ig Domain II and III-The nucleotide sequences 3Ј to the signal peptide and 5Ј to the acid box were deleted to engineer FGFR1⌬ and FGFR3⌬ (see N-terminal deletion, Fig. 1 and 2A). This deletion corresponds to the naturally occurring two Ig domain splice form of FGFR1 (45). The internal primers used to make FGFR1⌬ are QL78 (antisense), 5Ј-gtgcatcttgttcgggcaaggt-3Ј, and QL77 (sense), 5Ј-cgaacaagatgcactcccatcc-3Ј. The internal primers used to make FGFR3⌬ are QL79 (antisense), 5Ј-GAGCATCCTGCTCTGGACC-AGG-3Ј, and QL80 (sense), 5Ј-AGAGCAGGATGCTCCATCCTCA-3Ј. A common primer within vector sequence was used to generate the 5Ј fragments containing the appropriate signal peptide.
Point Mutagenesis of FGFR3-Point mutations, introduced into the C terminus of FGFR3 Ig domain III (constructs C-13, C-14, C-15, and C-16), were engineered using PCR amplification and appropriate mutant oligonucleotide primers. All mutations were confirmed by DNA sequencing.
Production of FGFR-AP Conditioned Media-COS-7 cells were transfected by the DEAE-dextran method (56) with 20 g of plasmid/ 10 6 cells. Conditioned media were collected after 3 days and replaced with fresh media. Subsequently, media were collected every 2 days over a period of 9 days. The conditioned media were assayed for AP enzyme activity to normalize for receptor number in binding assays (55). AP enzyme activity was determined by transferring 50 l of conditioned media to a flat bottom microtiter plate and adding 50 l of a 2ϫ AP assay solution (2 M diethanolamine, 1 mM MgCl 2 , 20 mM homoarginine, 12 mM p-nitrophenol phosphate (Sigma)) and measuring the change in optical density at 405 nm in a kinetic microtiter plate reader (Molecular Devices Inc.).
Binding Assays-FGF1, -2, and -9 were iodinated using the chloramine T method to a specific activity of 800 -3000 Ci/cpm/fmol as described previously (57). 1-3 g of FGF was incubated with 1 mCi of Na 125 I (Amersham Pharmacia Biotech), 43 g/ml chloramine T (Eastman Kodak Inc.), and 143 mM HEPES, pH 7.4, in a volume of 70 l for 2 min at 23°C. 100 l of 20 mM dithiothreitol was then added, and the mixture was then incubated for an additional 10 min. The mixture was then applied to a heparin-agarose column (200-l bed volume) that had been pre-washed with 20 mM HEPES, pH 7.4, 0.2% bovine serum albumin, and 0.4 M NaCl. Labeled growth factor was eluted with 20 mM HEPES, pH 7.4, 0.2% bovine serum albumin, and 3 M NaCl. Labeled FGF1 and FGF9 were used in the binding studies within 48 h of labeling. FGF2 was used up to 2 weeks after labeling. The labeled growth factors were stored at Ϫ70°C. Binding assays were set up by adding components at 4°C in the following order: Dulbecco's modified 3 J. Xu and D. M. Ornitz, manuscript in preparation.
Eagle's medium (Life Technologies, Inc.) with 0.1% bovine serum albumin; 30 l of a 2ϫ slurry of anti-alkaline phosphatase monoclonal antibodies coupled to Sepharose (55); 10 l of 25 g/ml heparin; FGFR-AP conditioned media containing specific soluble FGFRs (300 optical density units of AP activity) (54); unlabeled FGF as a competitor (up to a concentration of 80 nM or approximately an 800-fold molar excess); and 125 I-FGF (30,000 -50,000 cpm) in a volume of 250 l. The reaction was gently rotated for 120 min at 4°C. Bound receptor and FGF were recovered by centrifugation (10 s at 12,000 rpm (4000 ϫ g) 4°C in a microcentrifuge). Free ligand concentration was determined independently by measuring an aliquot of supernatant following the sedimentation of the bound FGF-FGFR-AP-Sepharose complex. After washing two times with 750 l of ice-cold phosphate-buffered saline, 125 I-FGF bound to FGFR-AP was quantified directly in a ␥ counter (Beckman Inc.). Nonspecific binding, 200 -400 cpm, was subtracted from total cpm bound.
To calculate binding constants, binding reactions were carried out in the presence of increasing amounts of unlabeled FGF. FGF concentrations were calculated by dividing the measured cpm bound or cpm free by the calculated specific activity (specific activity times dilution factor). Binding curves were then calculated and fit to either a two-ligand binding is bound ligand and F is the concentration of free ligand, K 1 and K 2 are the binding constants for first and second step binding, respectively), or to a single ligand binding equation (B ϭ Bmax F/(K d ϩ F)). Binding constants were determined by non-linear least squares analysis (using the Fit Program, E. Di Cera, Washington University). This method of determining dissociation constants is essentially independent of receptor concentration (58,59) and shows very little dependence on the precise specific activity of the ligand (for example a 40% error in the value of the specific activity corresponds to an 8% error in the calculated K d ). The mean ligand activity, X m (defined as (K 1 K 2 ) Ϫ1/2 ) for a two binding site model, reflects the overall binding affinity and can be compared with K d in a one binding site model (58). Scatchard plots were used to demonstrate graphically cooperative binding of some receptors. However, because of the dependence of the Scatchard plot on receptor concentration and the observed non-linearity (positive cooperativity) in some cases, these plots were not used to calculate binding constants.
Western Blot Analysis-Soluble receptors (40 optical density units of AP activity) were immunoprecipitated using wheat germ agglutinin-Sepharose (Amersham Pharmacia Biotech). The immune complexes were then eluted in electrophoresis sample buffer (62.5 mM Tris-HCl, pH 6.8, 2% SDS, and 10% glycerol) and analyzed on 6% polyacrylamide gels with SDS (60) and were electrophoretically transferred to a polyvinylidene difluoride membrane. Blots were blocked with 10% nonfat dry milk in PBS-T and then incubated with primary antibody (monoclonal anti-alkaline phosphatase) (Genzyme Inc.) for 2 h in PBS-T (phosphate-buffered saline with 0.5% Tween 20) (1:1000 dilution). The blots were washed three times for 10 min each with 25 ml of PBS-T and then incubated with a 1:1000 dilution of peroxidase-conjugated secondary antibody for 2 h at room temperature. After three washes for 10 min with PBS-T, protein bands were visualized by enhanced chemiluminescence (ECL) detection (Tropix Inc.) (61).

RESULTS
Sequence comparison of the FGFR1 and FGFR3 extracellular regions (c splice form) shows a high degree of similarity and several highly conserved regions of identity (Fig. 1). The most highly conserved sequences are in Ig loop II, the linker region between Ig loop II and III (II-III linker region) (C), Ig loop III and through the N-terminal part of the juxta-transmembrane region (J). The signal peptide (SP), N terminus (A), and Ig loop I are less well conserved, and the I-II linker (B) shows some regions that are highly conserved. For the purposes of this study the Ig loop is defined as the sequence between the conserved disulfide-linked cysteine residues within each Ig domain. To examine the molecular basis for ligand binding specificity, deletions, chimeric molecules, and scanning mutations were constructed throughout the extracellular domains of FGFR1 and FGFR3. Fusion sites were chosen at residues that are identical between FGFR1 and FGFR3 to minimize nonspecific effects on protein structure. All molecules were tagged at the C terminus with human placental AP, which serves as an enzymatic marker to normalize protein concentration and as an epitope tag for immunoprecipitation binding assays.
Ligand Binding Specificity of FGFR1 and FGFR3 Molecules Lacking Ig Domain I-The binding activity of the full-length FGFR1 extracellular domain was compared with a receptor in which the N-terminal Ig domain was deleted (FGFR1⌬). The deletion in the N terminus corresponded to that of a naturally occurring variant in which Ig domain I is skipped by alterna- The FGFR and the full-length extracellular domain of the FGFR fused to human placental AP requires heparin for optimal ligand binding (28,54). Wang and co-workers (48) have shown for FGFR1 expressed in SF9 cells that, for FGF1, one possible role for Ig domain I was to diminish the absolute requirement for heparin. It was therefore important to determine whether or not, in our experimental system, the dependence on heparin for ligand binding also required Ig domain I. The three Ig domain (FGFR1) and two Ig domain (FGFR1⌬) receptors, fused to AP, were therefore assayed for FGF2 binding in the presence or absence of exogenous heparin. For both receptors, binding of FGF2 was significantly increased by heparin (Fig. 2B). These data demonstrate that the heparin requirement for FGF2 binding to the soluble FGFR1 extracellular domain is independent of the N-terminal Ig domain. All remaining binding experiments in this study were in the presence of 2 g/ml heparin, a concentration of heparin that optimizes ligand binding for FGF1, FGF2, and FGF9 (22,44).
The ligand binding specificity of FGFR1⌬ and FGFR3⌬ was compared with that of the three Ig domain receptors. When iodinated FGF1, FGF2, and FGF9 were assayed for binding, no significant difference was observed between full-length (FGFR1 and FGFR3) and two Ig domain (FGFR1⌬ and FGFR3⌬) receptors (Fig. 2C). Furthermore, when a competition binding curve for FGF2 binding to FGFR1 or FGFR1⌬ was fit to either a one-site or two-site binding equation, there was no significant difference in the affinity constants (X m ϭ 2.9 Ϯ 0.7 ϫ 10 Ϫ10 and 2.2 Ϯ 0.3 ϫ 10 Ϫ10 M, respectively, see "Experimental Procedures" and Table I) or Hill coefficients (1.7 Ϯ 0.1, 1.5 Ϯ 0.1, respectively), indicating similar binding mechanisms (see below). However, the magnitude of ligand binding to either FGFR1 or FGFR3 varied with respect to FGF1, FGF2, and FGF9 and was consistent with previously published data for FGFR1-AP and FGFR3-AP binding specificity (15,44). Notably, FGF1 bound to a similar extent to both long and short forms of FGFR1 and FGFR3, FGF2 showed consistently 3-4fold more binding to both long and short forms of FGFR1 compared with FGFR3, and FGF9 bound both long and short forms of FGFR3 but showed no binding to either form of FGFR1 over background levels (Fig. 2C). These data demonstrated that the N-terminal Ig domain is not a determinant of ligand binding specificity and therefore localized the binding domain(s) to regions encompassing Ig domain II and III. To localize further receptor regions that determine binding specificity, a series of chimeric molecules were constructed throughout the extracellular sequences (B, II, C, III and J; defined in Fig.  1) of FGFR1 and FGFR3 and assayed for ligand binding.
Binding Properties of Soluble Chimeric FGF Receptors-To map regions that determine ligand binding specificity, two types of receptor mutations were constructed (Fig. 3). 1) Chimeric receptors, named 1,3 or 3,1 (Fig. 3B), consist of a simple fusion between FGFR1 and FGFR3 at sites of amino acid identity such as conserved cysteine residues. 2) Scanning mutations, named 1S or 3S (Fig. 3C), have a backbone structure from FGFR1 or FGFR3, respectively, and contain sub-regions derived from the other receptor, again fused at sites of amino acid identity.
Because mitogenic assays assessing ligand-receptor specificity demonstrated that FGF1 is the only member of the FGF family that can activate all receptors, regardless of splice form (21), the ability of FGF1 to bind each chimeric receptor was tested. All chimeric receptors bound FGF1 significantly over background (Fig. 3). Only two scanning mutations, 3Sa and 1Sa, showed 2-fold lower binding compared with most other molecules. These data demonstrate that the primary structure of the chimeric receptors is not significantly altered and that these chimeric molecules are therefore suitable for testing the specificity of other potentially more selective ligands.
In all the binding studies shown, the quantity of soluble receptor used was normalized to AP enzyme activity. To demonstrate that AP activity correlates with the amount of protein present, several of the preparations were also compared by Western blot analysis. Immunoprecipitation of 40 units of AP activity, followed by Western blot and detection with anti-AP monoclonal antibodies (Fig. 4A), showed similar band intensities and confirmed that enzymatic quantification of AP activity is a good measure of receptor protein concentration. Additionally, the predicted 10-kDa deletion in FGFR1⌬AP is demonstrated by directly analyzing both FGFR1AP and FGFR1⌬AP conditioned media by Western blot without prior immuno- precipitation (Fig. 4B).
Specificity of Chimeric Receptors for FGF2-FGF2 consistently binds FGFR1 to a greater extent than FGFR3 (Fig. 2, see Refs. 15 and 54) but nevertheless can similarly activate either receptor in a mitogenic assay (21). The FGFR extracellular sequence(s) responsible for this differential binding were mapped by analysis of FGF2 binding to the panel of chimeric receptor molecules shown in Fig. 3.
Chimeric receptors containing sequence from the N terminus of FGFR1, fused to the C terminus of the FGFR3 extracellular region (1,3-a; 1,3-b; 1,3-c), bind less FGF2 than full-length FGFR3 (Fig. 5, A and B) but similar amounts of FGF1 (Fig. 3B). These three molecules all have the FGFR3 Ig loop III and J region in place of FGFR1. In contrast, when Ig loop III is derived from FGFR1 (1,3-d and 3Se, Fig. 5) binding is similar to that of full-length FGFR1. Chimeric molecules containing the N terminus of FGFR3 and progressively more FGFR1 sequence from the C terminus bind FGF2 as well (3,1-b) or better (3,1-c, 3,1-d, and 3,1-e) than full-length FGFR1 once the Cterminal 30 amino acid residues of Ig loop III are derived from  FGFR1 (compare 3,1-a to 3,1-b or 3,1-c in Fig. 5B). These data demonstrate that the C-terminal half of FGFR1 Ig domain III (exon c) is important for FGF2 binding and that the I-II linker region of FGFR1 may inhibit FGF2 binding. Consistent with this, scanning mutations in which small regions of FGFR3 are replaced by corresponding sequence from FGFR1 bind FGF2 less well than full-length FGFR1 if the sequences exclude the The blots were hybridized with anti-AP antibody, and bands were visualized with a peroxidase-conjugated secondary antibody and ECL detection. The positions of molecular mass markers are indicated on the right. B, 20 units of FGFR1-AP and FGFR1⌬-AP conditioned media were directly loaded onto a 6% SDS-polyacrylamide electrophoresis gel that was electroblotted and probed with anti-AP antibody as described above.  C-terminal 30 amino acid residues of FGFR1 Ig loop III (3Sa to 3Sd). However, scanning mutations containing these 30 residues from FGFR1 bind FGF2 at least as well as full-length FGFR1 (3Se to 3Sg), again demonstrating the importance of this sequence.
Within the C-terminal 30 amino acid residues of Ig loop III, five differences exist between FGFR1 and FGFR3 (Fig. 6A). To determine whether single or multiple amino acid residues account for these differences, residues in FGFR3 were sequentially changed to residues used in FGFR1, and FGF2 binding was assayed. The H327R, S325H, or the double mutation (H327R/S325H) resulted in only a small increase in FGF2 binding (Fig. 6B). Adding a third point mutation, L321M (H327R/S325H/L321M) significantly enhanced FGF2 binding. However, levels of FGF2 binding comparable to that of FGFR1 were not attained until all five residues were changed (3SG, Figs. 5C and 6). These data suggest that the sequence context of this entire region is required for optimal binding of FGF2 and that no single amino acid difference can account for the decreased binding to FGFR3.
Several of the chimeric receptors consistently bind more FGF2 than the full-length FGFR1 extracellular domain (3,1-c,  3,1-d, 3,1-e, 3Se, and 1Sa in Fig. 5). These "activated" molecules all have in common sequences derived from the FGFR3 linker region between Ig loops I and II (I-II linker (B) in Fig. 1) and FGFR1 sequence derived from Ig loop III. These data suggest that the I-II linker region in FGFR3 may contribute to a second FGF2 binding region that, together with a primary binding region localized in FGFR1 Ig loop II-III, increases the FGF2 binding capacity. Alternatively, because FGFR1 already has two linked binding regions (see below), uncoupling these regions in chimeric receptors may increase binding by relieving constraints on the individual regions. The N-terminal binding region in FGFR3, along with an inactive FGFR3 Ig loop III binding region, may account for the decreased FGF2 binding observed in the full-length FGFR3 extracellular domain compared with the FGFR1 extracellular domain. To examine further these possibilities, the number of binding sites and binding affinity of chimeric receptor molecules were assessed.
The binding affinities of FGFR1, FGFR1⌬, and chimeric receptors 1Sa, 1Sb, 3,1-c and 3,1-e, to FGF2, were determined by competition for binding of iodinated FGF2 with increasing amounts of unlabeled FGF2 up to an 800-fold molar excess ( Fig. 7 and data not shown). Interestingly, Scatchard plots (B/F versus B) of FGFR1, FGFR1⌬, and 1Sb showed a concave curve instead of a straight line (Fig. 7, A-C, right) suggesting the presence of more than one interacting binding region in FGFR1 (62,63). In contrast, chimeric receptors 1Sa, 3,1-c, and 3,1-e (which showed increased binding capacity) have linear Scatchard plots (Fig. 7, D-F, right) indicating single or identical and  6. FGF2 binding to chimeric point mutations in FGFR3 Ig loop III. A, amino acid sequence of FGFR1, FGFR3, and FGFR3 mutants. Only sequence differences between FGFR3 and FGFR1 are shown, and identical residues are indicated by a dash. Note that scanning mutant, 3Sg (Fig. 3C) is identical to FGFR1 in this region. B, FGF2 binding to control media (cos) and to the chimeric FGFRs shown in A. independent binding regions in the chimeric receptors. Based on this observation, the binding data were fit to a two-ligand binding ; B ϭ bound ligand, F ϭ free ligand) for FGFR1, FGFR1⌬, and 1Sb (Fig. 7, A-C). Binding constants (K 1 and K 2 ) using the two-site equation are shown in Table I. The mean ligand activity X m (defined as (K 1 ⅐K 2 ) 1/2 ) reflects the overall binding affinity that can be compared with the dissociation constant K d in a one binding site model (58). The Hill coefficient, n1 ⁄2 (defined as 2/(1 ϩ K 1 /K 2 ) 1/2 ) (58), is significantly greater than 1 for these three molecules, indicating positive cooperativity between two binding sites. Notably, X m and n1 ⁄2 values are not significantly different for these three receptors, which indicates that these receptors interact with FGF2 in a similar manner. In contrast to FGFR1, FGFR1⌬, and 1Sb, the binding data for chimeric receptors 1Sa, 3,1-c, and 3,1-e fit well to either a single site binding equation (B ϭ B max ⅐F/(K d ϩ F); B ϭ bound ligand; F ϭ free ligand) or the two-site equation (Table I). The Hill coefficients for these receptor binding domains are close to 1 providing further evidence for single or identical and independent binding sites in these chimeric molecules.
Specificity of Chimeric Receptors for FGF8 and FGF9 -FGF8 and FGF9 can bind and activate FGFR3c but not FGFR1c. Furthermore, unlike all other FGFs tested, only FGF1 and FGF9 can bind and activate FGFR3b (Refs. 21, 39, 43, and 44 and Fig. 2). Comparison of FGF8 and FGF9 activity shows that both ligands activate FGFR3c but not FGFR1c, but unlike FGF9, FGF8 cannot interact with FGFR3b. To evaluate the molecular basis for this specificity the ability of FGF8 and FGF9 to interact with chimeric receptors was assayed.
Because FGF8 and FGF9 have similar binding specificity with respect to FGFR3c and FGFR1c but discordant activity toward FGFR3b, the receptor binding regions for these two ligands were compared. The ability of FGF8 (which cannot be efficiently iodinated) to compete with 125 I-FGF1 binding to chimeric receptors was assayed. FGF1 was used because of its ability to activate all splice forms of all FGFRs and to bind all chimeric FGFRs (Ref. 21, Fig. 3). As predicted from the known specificity of FGF8 (39), it could compete efficiently (defined as greater than 10% competition) for binding to FGFR3 but not to FGFR1 (Fig. 9A). FGF8 failed to compete for binding to any chimeric receptor that contained FGFR1 derived sequence from the I-II linker, Ig loop II, or the N terminus of the II-III linker (1,3-a to 1,3-d, 3,1-d, 3,1-e, 3Sa, and 3Sc, Fig. 9B). Significantly, FGF9 showed a similar binding profile to that of FGF8 except for chimeras 3,1-d and 3Sc, molecules that bound FGF9 well but did not bind FGF8. This difference in binding suggests that the I-II linker, Ig loop II, and the N-terminal half of the II-III linker of FGFR3 are all important for FGF8 binding, whereas only the I-II linker and Ig loop II specify FGF9 binding. FGF8 bound poorly to chimeras 3,1-b, 3,1-c, 3,1-d, and 3Se compared with full- length FGFR3, 3,1-a, and 3Sd. Chimeras  3,1-b, 3,1-c, 3,1-d, and 3Se have in common FGFR1 sequence from Ig loop III suggesting that this region of FGFR1 is also unfavorable for FGF8 binding. DISCUSSION FGFs bind and activate a subset of the four known FGFRs. One well described mechanism that modulates the binding specificity of an individual FGFR is alternative splicing in Ig loop III (12-16, 18 -20). However, sequence differences between FGFRs must also regulate specificity. In this study we have mapped regions of the FGFR that are important determinants of ligand binding specificity of similarly spliced FGFRs. We have also addressed the role of Ig domain I in ligand binding.
Sequence comparison of the extracellular domain of FGFRs 1 and 3 show that the N terminus (region A), Ig loop I, and parts of the I-II linker (region B) are poorly conserved (26% amino acid identity), whereas Ig loop II, the II-III linker (region C), Ig loop III, and parts of the juxta-transmembrane domain (region FIG. 7. Binding curves for soluble FGFRs. Binding curves were generated by incubating a constant amount of iodinated FGF2 and increasing amounts of unlabeled FGF2 with soluble FGFRs. cpm bound and cpm free were both measured and used to calculate concentrations of bound and free ligand. Left panels show binding curves (total ligand bound versus total ligand free) and right panels show Scatchard plots derived from the binding curves. Note that affinity constants (see Table  I J) are highly conserved (66% amino acid identity, Fig. 1). In addition to alternative splicing in Ig loop III, a second common alternative splicing event excludes the exon encoding Ig domain I (region A, Ig loop I, and part of the I-II linker). Comparison of binding to both the three Ig loop form of the receptor and the two Ig loop form showed no difference in binding for FGF1, FGF2, FGF8, or FGF9, suggesting that this region is not a determinant of ligand binding specificity. Furthermore, the affinity and cooperativity of FGF2 binding and the heparin requirement for binding to both long and short receptor splice forms of FGFR1 was similar, indicating that Ig domain I has little effect on the binding affinity of FGF2. These data are in contrast with data of Wang et al. (48) that show that receptors lacking Ig domain I have increased affinity for FGF1 and heparin. It remains possible that Ig domain I is an important determinant of the binding specificity of other members of the FGF family.
To map further the regions within the FGFR extracellular domain that determine ligand binding specificity, the binding of FGF1, FGF2, FGF8, and FGF9 to chimeric molecules be-tween FGFR1c and FGFR3c was assayed. These ligands were chosen because they have distinct differences in their ability to interact with these two receptors. Consistent with mitogenic assays and previous binding assays, FGF1 bound all chimeric molecules significantly over background levels (Refs. 15, 21, and 54 and Fig. 3). Binding assays with FGF2, which preferentially binds FGFR1, shows that FGF2 binds cooperatively to two sites within FGFR1, whereas FGF8 and FGF9 binding is most dependent on a single proximal region within FGFR3 (Fig. 10A). These data suggest that binding specificity is determined by more than one region on an FGFR.
A two binding region model (Fig. 10) is supported by the observed cooperativity of FGF2 binding. This is shown by a concave Scatchard plot and a Hill coefficient of 1.7. Additionally, the binding data fit a two-ligand binding equation much better than a one-ligand binding equation, also supporting the notion of two binding regions. Furthermore, cooperativity is lost in some chimeric molecules suggesting that specific amino acid side chains in the I-II linker and Ig domain II interact with residues in Ig domain III in FGFR1 and that these inter- FIG. 8. FGF9 binding to  actions are essential for cooperative FGF2 binding. Furthermore, chimeras 3,1-c, -d, and -e bind more FGF2 than FGFR1 alone, and chimeras 1,3-a, -b, and -c bind less FGF2 than FGFR3 alone. These observations support a model in which FGFR3 contains a proximal (I-II linker, Ig domain II) favorable binding region and a distal (Ig domain II-III) poor binding region and in which FGFR1 contains a proximal neutral (or interacting) binding region and a distal very favorable binding region (Fig. 10B). Despite the attractiveness of this model, the stoichiometry of binding of FGF2 to FGFR is not known and is controversial. By using isothermal titration calorimetry, Pantoliano and co-workers (23) identified a complex containing one FGF2 and two FGFR1 molecules. In contrast, several other groups have demonstrated a stoichiometry of 2:2 by receptor cross-linking (25), by surface plasmon resonance (64), and by crystallography (65). Our data are consistent with two FGF2 molecules binding cooperatively (in cis) to one or two FGFRs if there are two independent and non-interfering binding sites (Fig. 10C, top) or with two FGF2 molecules binding cooperatively to a receptor dimer (in trans) if the binding specificity regions on a single receptor exclude the binding of a second ligand (Fig. 10C, bottom). Interestingly, interaction studies focusing on the highly conserved II-III linker region and a recent crystal structure of FGFR1 complexed with FGF2 (65,66) demonstrate that receptor dimers can form independently of ligand binding. These studies suggest that cooperative ligand binding could occur in trans to a receptor dimer.
The hypothesis that two FGF binding regions exist within a single FGFR is further supported by studies in which either FGFR2 Ig domain II or Ig domain IIIb, when fused to the immunoglobulin heavy chain constant region, are capable of binding ligand (19). However, the immunoglobulin molecule itself may also contribute to ligand binding so it is still not clear whether an isolated FGFR Ig domain is capable of independently binding ligand. Additionally, a soluble chimeric receptor containing only region B, Ig loop II, and region C of FGFR3 fused to AP failed to bind FGF1 or FGF9 with high affinity (data not shown) suggesting that this region is not sufficient to form an FGF-binding site. Studies by Hou et al. (67) also strongly suggest that both Ig domain II and III are required to form an FGF-binding site and that Ig domain I cannot substitute for either Ig domain II or III.
The presence of a proximal binding region is supported by the observation that the I-II linker sequence and Ig loop II from FGFR3 are both required for FGF9 binding and that the II-III linker sequence and Ig loop III can be derived from either receptor without affecting FGF9 binding specificity (Figs. 8 and  10). The irrelevance of the source of Ig loop III sequence for FGF9 binding provides a molecular explanation for why FGF9 binds to both alternatively spliced forms of FGFR3. However, as stated above, Ig loop III is still required to form an FGF9binding site because the I-II linker, loop II, and the II-III linker cannot by themselves bind ligand (data not shown). This suggests that there may be interdomain interactions that contribute to the FGF9-binding site.
Exchanging the C-terminal 8 amino acid residues of the J region (chimera 3,1-a), a sequence poorly conserved between FGFR1 and FGFR3, enhanced ligand binding by 1.4-(FGF1) to 2-fold (FGF9). This juxta-transmembrane sequence may therefore contribute to FGF binding or may be involved in orienting the extracellular domain relative to the transmembrane domain or, for the soluble FGFRs examined here, relative to AP.
Unlike FGF8 and FGF9 that show specificity for a region that includes Ig loop II of FGFR3 and FGF2 that recognizes two cooperatively linked binding regions in FGFR1, FGF1 binds all chimeric receptors and can activate any FGFR regardless of alternative splicing (21). This unique property of FGF1 suggests that it can recognize either proximal or distal FGF binding regions derived from any FGFR. FGF8 can compete for up to 45% of the bound 125 I-FGF1 to chimeric FGFRs (Fig. 9), whereas FGF1 can displace essentially all bound 125 I-FGF1 (data not shown). This observation suggests that FGF8 can only compete at one FGF1 binding region, whereas FGF1 can compete for binding at both regions.
The data presented here and diagrammed in Fig. 10C (top) suggests that there may be two binding sites on a single FGFR. In contrast, a recent crystal structure of FGF2 bound to a fragment of FGFR1 shows only one molecule of FGF2 in contact with a pocket formed by Ig domain II, the II-III linker, and Ig domain III (65). This binding pocket corresponds to the distal binding region defined here. However, the fragment of FGFR1 that was crystallized lacks 22 N-terminal amino acids (includ- Darkly shaded boxes are unfavorable for binding, and lightly shaded boxes are favorable for binding. The gradient filled FGF2 box indicates a binding site that is cooperatively linked (double arrow) to a primary FGF2binding site. B, FGF2 binding to chimeric receptors. // indicates uncoupled FGF2-binding sites in 3,1-and 1,3-chimeric molecules. Note that in 3,1-chimeras, favorable (lightly shaded) and good (open) binding regions are juxtaposed resulting in greater FGF2 binding (ϩϩϩ), whereas in 1,3-chimeric molecules, unfavorable (darkly shaded) and poor (solid) binding region are juxtaposed resulting in very poor FGF2 binding (---). C, models for cooperative FGF2 binding. Top, cis-binding model: ligand (L, solid circle) binding to two sites on a single receptor molecule (R, shaded). In this model, ligand can bind to either site 1 or site 2 and induce a change in the other site that increases its affinity for a second ligand. Bottom, trans-binding model: in this model ligand can bind to a single site with two specificity regions, 3 and 4, on a single receptor in a dimeric complex. Binding can then induce a change in the other site on the second receptor that increases its affinity for a second ligand.
ing the highly conserved acidic region) that are present in naturally occurring 2 Ig-domain splice forms of FGFR1 and in the FGFR1⌬ molecule used in this study. It is therefore possible that the fragment used for crystallography has lost a critical component of the proximal binding region. Another possibility that cannot be ruled out is cooperative binding in "trans" across a receptor dimer (Fig. 10C, bottom). Although this model would appear to be more consistent with the FGFR dimer observed in the crystal structure, which defines a binding site that is in close agreement with the distal binding region mapped here, the concentrations of receptor used in the binding assay make it unlikely that such a dimer would form under the conditions of the experiment.
Multiple binding regions in the FGFR may be important to allow four receptors to differentially recognize 19 or more FGF ligands. Additionally, cooperative linkage of multiple binding regions may make FGFR1 responsive to a threshold level of ligand, below which a response is not elicited. This may be an important mechanism regulating FGF2 activity during tissue repair and remodeling. FGF2, which is present on the surface and in the extracellular matrix of many cells, may have little activity under steady state conditions. However, following injury or during tissue remodeling, large amounts of FGF2 may be released from damaged cells, raising the threshold concentration of ligand to elicit a rapid response.