|Home | About | Journals | Submit | Contact Us | Français|
Fibroblast growth factor (FGF) signaling regulates mammalian development and metabolism, and its dysregulation is implicated in many inherited and acquired diseases, including cancer. Heparan sulfate glycosaminoglycans (HSGAGs) are essential for FGF signaling as they promote FGF·FGF receptor (FGFR) binding and dimerization. Using novel organic synthesis protocols to prepare homogeneously sulfated heparin mimetics (HM), including hexasaccharide (HM6), octasaccharide (HM8), and decasaccharide (HM10), we tested the ability of these HM to support FGF1 and FGF2 signaling through FGFR4. Biological assays show that both HM8 and HM10 are significantly more potent than HM6 in promoting FGF2-mediated FGFR4 signaling. In contrast, all three HM have comparable activity in promoting FGF1·FGFR4 signaling. To understand the molecular basis for these differential activities in FGF1/2·FGFR4 signaling, we used NMR spectroscopy, isothermal titration calorimetry, and size-exclusion chromatography to characterize binding interactions of FGF1/2 with the isolated Ig-domain 2 (D2) of FGFR4 in the presence of HM, and binary interactions of FGFs and D2 with HM. Our data confirm the existence of both a secondary FGF1·FGFR4 interaction site and a direct FGFR4·FGFR4 interaction site thus supporting the formation of the symmetric mode of FGF·FGFR dimerization in solution. Moreover, our results show that the observed higher activity of HM8 relative to HM6 in stimulating FGF2·FGFR4 signaling correlates with the higher affinity of HM8 to bind and dimerize FGF2. Notably FGF2·HM8 exhibits pronounced positive binding cooperativity. Based on our findings we propose a refined symmetric FGF·FGFR dimerization model, which incorporates the differential ability of HM to dimerize FGFs.
The fibroblast growth factors (FGFs) (1, 2) comprise a family of secreted polypeptides that are encoded by 18 distinct genes (FGF1–FGF10 and FGF16–FGF23) in mammals. FGFs play pleiotropic roles in development as well as metabolism (3,–7). FGFs transmit their effects by binding and activating FGF receptor-tyrosine kinases (FGFRs) a subfamily within receptor tyrosine kinase superfamily (3, 8, 9).
The core homology domain of FGFs is about 120 amino acid long and adopts a β-trefoil fold consisting of 12 β-strands (10,–19). All FGFs interact with heparan sulfate glycosaminoglycans (HSGAGs)3 (20). The HSGAG-binding site of FGFs is composed of the β1–β2 loop and the region encompassing β10 through β12 (10, 21,–25). Based on their high sequence and structural homology to FGFs, four additional genes (FGF11–FGF14) are also regarded FGF family members. However, recent structural and biochemical data show that FGF11–FGF14 are functionally unrelated to FGFs (26,–28).
Four genes in mammalian organisms (FGFR1 (flg), FGFR2 (bek), FGFR3, and FGFR4) code for FGFRs (9). The ectodomain of a prototype FGFR comprises three Ig domains (D1–D3). A unique property of FGFR is the presence of a stretch of acidic residues in the linker connecting D1 to D2, referred to as the “acid box.” A wealth of structural and biochemical studies has demonstrated that D2, D3, and D2–D3 linker are necessary and sufficient for FGF binding (8, 29). Like FGFs, FGFRs are also HSGAG-binding proteins (30). HSGAG-binding site of FGFR resides in D2 and is composed of residues from g-helix A, β strands B and D, and gA–βA′ and βA′–βB loops (23, 24). D1 and the D1–D2 linker are dispensable for FGF binding and in fact negatively regulate FGFR signaling (31,–33).
Two major alternative splicing events take place in the ectodomains of FGFR1–FGFR3: one involving D1 and the D1–D2 linker, and the other D3 (9, 34). Alternative splicing of D1 and the D1–D2 linker serves as a mechanism to modulate receptor autoinhibition, whereas alternative splicing in D3 determines ligand binding specificity (35, 36). In FGFR1-FGFR3, the second half of D3 is encoded by two mutually exclusive exons (“b” and “c”) that are used in a tissue-specific fashion (37,–41). The “b” exon is used in epithelial tissues, whereas the “c” exon is preferentially used in mesenchymal tissues. As a result of this splicing event, the number of principal FGFRs is increased to seven isoforms namely FGFR1b, FGFR1c, FGFR2b, FGFR2c, FGFR3b, FGFR3c, and FGFR4. Importantly, alternative splicing of D3 sets a specificity barrier such that epithelially expressed FGFs activate FGFRc isoforms, whereas mesenchymally expressed FGFs activate FGFRb isoforms (42). Most FGFs bind and activate more than one of the seven principal FGFRs, although they do not cross the specificity barrier set by the D3 alternative splicing (43). For example, FGF2 exhibits high affinity to both FGFR1c and FGFR2c but does not bind to FGFR1b and FGFR2b. The affinity of FGF2 to FGFR3c and FGFR4 is also negligible (44). FGF1 is an exception, because it is capable of indiscriminately binding both “b” and “c” isoforms of FGFR1–3 and FGFR4 (42). Crystallographic studies of several FGF·FGFR complexes have shown that the D3 alternative splicing regulates FGF·FGFR binding specificity and promiscuity by modifying the primary sequences of the key ligand binding sites in D3 (14, 16).
Genetic analysis of mice and flies deficient in enzymes involved in heparan sulfate (HS) biosynthesis, and cell-based assays using cell lines devoid of HSGAG have established that FGF signaling requires HSGAGs (45,–47). HSGAGs impinges on FGF signaling through multiple mechanism, including promotion of FGF·FGFR binding and dimerization (8), control of FGF diffusion, and gradient formation in the extracellular matrix (18, 48, 49), providing thermal stability and protection from proteolytic degradation (50, 51).
Crystallographic analyses of FGF·FGFR·heparin ternary complexes have provided two conceptually different models by which HSGAG promote FGF·FGFR dimerization. According to a symmetric model (Mohammadi model, FGFR1·FGF2·heparin (decasaccharide), PDB entry 1FQ9) (23) HS promotes formation of a symmetric 2:2:2 FGF·FGFR signaling complex. The dimer interface is mediated by protein-protein contacts between the two adjacent FGF·FGFR halves and is strengthened by interaction of heparin with FGF and FGFR. The dimer interface comprises direct receptor-receptor contacts mediated by D2, and interactions of FGF from one FGF·FGFR half with D2 of FGFR in the second FGF·FGFR half (referred to as secondary FGF·FGFR interface). At the membrane distal end of the dimer, the individual HS-binding sites of two FGFs and FGFRs unite into one HS-binding canyon into which two HS chains bind. By engaging the HS-binding sites of FGF and FGFRs, HS augments FGF·FGFR affinity within each FGF·FGFR half as well as promotes the dimer interface (8, 52).
In this structure, only the first six sugar units at the non-reducing end of the decasaccharide are in contact with the protein (FGFR or FGF). The remaining four sugar units of one of two decasaccharides are disordered. In the other decasaccharide, two additional sugar units are visible due to the favorable crystal contacts they make with an adjacent FGF molecule. Thus, based on the symmetric model, a hexasaccharide should be sufficient for dimerization and hence biological activity (8, 52).
A different model has been proposed by Blundell and coworkers (Blundell model, FGFR2c·FGF1·HS, PDB entry 1E0O) (24). This model displays a 2:2:1 FGF·FGFR·HS stoichiometry in which a single HS chain bridges two FGFs in trans and each FGF binds to one FGFR only. In contrast to the symmetric model, there are no direct protein-protein contacts between the two FGF·FGFR halves in this model. In other words, the asymmetric dimer is held together solely by the ability of HS to dimerize FGFs, and consequently this mode of dimerization is strictly HS-dependent.
The entire co-crystallized decasaccharide is visible in this structure. Two sugar units at the non-reducing end of oligosaccharide engage the D2 domain of one of the two FGFR chains while the HSGAG-binding site of the other FGFR chain remains unoccupied. Seven sugar rings bridge the two FGFs in trans. Based on this model, the shortest biologically active HS would be an octasaccharide, although maximal activity would require a dodecasaccharide, because it will be long enough such that it could engage D2 of both FGFRs (24). It should be noted, however, that an alternative interpretation of this crystal structure also leads to a model similar to the symmetric one (52, 53).
Numerous in vitro and cell-based studies have been attempted to test the salient features of each model. Most data support the physiological relevance of the symmetric dimerization model. For example, mutations that ablate interactions of FGF from one FGF·FGFR half to D2 of FGFR in the second FGF·FGFR half diminish the ability of the mutated FGF to signal while these mutated FGFs retain the ability to form a 1:1 FGF·FGFR complex (54). A further unbiased piece of evidence in support of a symmetric model comes from analysis of the naturally occurring A172F mutation in FGFR2. This gain-of-function mutation, responsible for the Pfeiffer syndrome, maps to the direct FGFR·FGFR interface and confers gain-of-function by promoting direct D2–D2 contacts and hence receptor dimerization (54).
Biological studies with size-fractioned heparin oligosaccharides have also been used to test the validity of each model. However, no consensus has been reached with regard to the minimal oligosaccharide length needed for FGF signaling. Some studies show that a hexasaccharide and even smaller sugars are capable of promoting FGF signaling (55,–57, 78). In contrast, other data show that a hexasaccharide has either poor or no activity at all and that an octasaccharide is the shortest biologically active heparin (53). In all these prior studies, the heparin oligosaccharides were prepared by either enzymatic hydrolysis or chemical cleavage of heparin isolated from natural sources. Therefore, a potential reason for the disparity between these data could be differences in the homogeneity of oligosaccharide preparations used. Our ability to de novo synthesize homogeneously sulfated heparin oligosaccharides of various degrees of polymerization, including hexasaccharide (HM6), octasaccharide (HM8), and decasaccharide (HM10) (Fig. 1A), provided us with the unique opportunity to revisit the minimum oligosaccharide length requirement for FGF signaling. Using a BaF3 cell line overexpressing a chimeric FGFR4, we show differences in the abilities of these HM to promote FGF1 and FGF2 signaling through FGFR4. All three HM have comparable capacity to stimulate FGF1-FGFR4 signaling. In contrast, in the case of FGF2, a major activity difference in signaling is seen for the transition from HM6 to HM8/10.
To understand the molecular basis for the observed prominent activity differences between HM in promoting FGF1- and FGF2·FGFR4 signaling, we used NMR spectroscopy, ITC, and SEC to study the interactions of FGF1 and FGF2 with the isolated D2 of FGFR4 (the HSGAG-binding domain of FGFR4) in the presence and absence of HM6 and HM8. Because HM8 and HM10 displayed similar receptor activation level in biological assays, our biophysical studies were restricted to HM8. NMR chemical shift mapping, signal attenuation, and T2 relaxation studies show the existence of a secondary FGF·FGFR D2 interaction site as well as direct D2–D2 interaction site corresponding to those seen in 1FQ9 thus providing direct evidence that the symmetric mode of dimerization occurs in solution. Our data show that the two HM make identical sets of contacts with FGF and FGFR D2 in the 2:2 FGF·D2 dimers. Therefore, the higher activity of HM8 relative to HM6 is not due to the additional contacts of the two extra sugar rings of HM8 with FGF or FGFR D2. Rather, our data show that differences in the binding affinity of HM for FGFs along with differences in the ability of HM to dimerize FGFs correlate with the differences in the biologic activities of the FGFs. Based on our findings we propose a refined symmetric model that takes into account the differential abilities of HM to dimerize FGF2 in a cooperative fashion.
DNA encoding the human FGF1 (Gly21–Asp155), FGF2 (Pro10–Ser155), and FGFR4 D2 (Asn138–Leu242) was amplified by PCR and cloned into Escherichia coli vector pETTEV (N-terminal His tag followed by a tobacco etch virus protease cleavage site) or pET21b (C-terminal His tag) (Novagen). The identity of the clones was verified by sequencing, and the E. coli expression plasmids were transformed into BL21(DE3) CodonPlus-RIPL (Stratagene) cells. Expression and purification of FGF1 (58) and FGF2 (59) was performed as previously described. The successful expression, purification, and refolding of the D2 domain is similar for FGFR1, FGFR2, and FGFR4 with minor modifications (60, 61).
The extracellular domain of FGFR4 and its mutated transmembrane parts (LALAVLLLAGLELLLLRWQ to LALAVLLLLRWQ amino acid x-y) were fused to the intracellular domain of hMpl and cloned into a modified pEF6/V5-His (C-terminal HA tag) (Invitrogen). The construct with chimeric FGFR4-hMpl (pEF6mut-HA FGF-R4αIIIcmut-hMpl2) was stably integrated by electroporation into murine BaF/3 cells genome. The cell line was selected with 20 ng/ml FGF2 (R&D, 234-FSE-025) and 10 ng/ml heparin (Sigma, H3149).
BaF/3 expressing FGFR4-hMpl was grown in a supplemented RMPI 1640 medium. Proliferation of cells was quantified by luminescence ATP measurement with a cell titer glow luminescent cell viability assay (Promega, G7571).
NMR spectra for all proteins and compounds were recorded at 298 K and were referenced to internal 3-trimethylsilyl-2,2,3,3-tetradeuteropropionate sodium salt (TSP). The experiments were carried out on a Bruker three-channel DRX600 and on a Bruker four-channel DRX800 spectrometer using standard pulse programs. Specific parameters, including buffer conditions and concentrations, are summarized in the figure legends.
ITC measurements were carried out on a VP-ITC ultrasensitive titration calorimeter (MicroCal LLC) and data analyses were performed as described previously (62).
For the synthesis of hexameric HM6, we have developed a strategy based on chain elongation from a common disaccharide building block (63). The elongation may then be repeated multiple times until the fully protected hexasaccharide is obtained. As in previous heparin mimetics synthesis, acyl groups were employed to protect the hydroxyl groups to be sulfated, benzyl ethers were used for the free hydroxyl groups, and azido groups masked the amino groups to be sulfated. We have used a similar strategy for the synthesis of oligosaccharides HM8 and HM10, and their preparation will be published elsewhere.
The murine pre-B cell line BaF3 requires interleukin-3 (IL-3) to proliferate. This cell line does not naturally express FGFRs and HSGAG. However, when transfected with FGFRs, BaF3 cells acquire the ability to proliferate in response to exogenous FGF and soluble heparin. Thus, the FGFR-overexpressing BaF3 cell lines are ideal for comparing the biological activity of heparin mimetics to promote FGF·FGFR signaling. To compare the efficacy of HM6, HM8, and HM10 (Fig. 1A) in promoting FGF-dependent dimerization of FGFR we established a stable cell line expressing chimeric FGFR4 (Fig. 1, B and C), termed FGFR4-hMpl, in which the ectodomain of FGFR4 was fused to transmembrane and intracellular domain of thrombopoietin receptor (hMpl). Addition of FGF and heparin causes the ectodomain of the FGFR4-hMpl to dimerize, which in turn juxtaposes the intracellular hMp1 regions, leading to subsequent Mpl activation and ultimately evoking a proliferation response. Therefore, cellular proliferation can be used as a readout to compare the capacity of FGFs and/or HM to dimerize the ectodomain of FGFR4.
FGFR4-hMpl-expressing BaF3 cells were deprived of IL-3 and treated with increasing concentrations of FGF1 or FGF2 in the presence or absence of 100 nm of HM6, HM8, or HM10. In the absence of HM, both FGF1 and FGF2 induced proliferation of the FGFR4-hMpl cells with a half-maximal effective concentration (EC50) of 16.6 ng/ml and 11.8 ng/ml, respectively (Fig. 1C). HM6/8/10 modestly decreased the EC50 of FGF1. In contrast, HM8/10 led to a 10-fold decrease in the EC50 values for FGF2 (Fig. 1C), suggesting that HM8/10 are more potent than HM6 in assisting FGF2 to dimerize and activate FGFR4-hMpl.
The observation that FGF1 and FGF2 are able to dimerize FGFR4-hMpl in the absence of HM is intriguing and can only be reconciled by the symmetric FGF·FGFR·heparin dimerization assembly model (Mohammadi model). In contrast, because the asymmetric mode of dimerization (Blundell model) lacks protein-protein contacts between the two FGF·FGFR halves, the ability of FGF1 and FGF2 to dimerize and activate FGFR4-hMpl in the absence of HM cannot be reconciled with this model. On the other hand, the higher efficacy of HM8/10 compared with HM6 fits better with the asymmetric model, because according to the symmetric model a hexasaccharide should be sufficient to promote FGF·FGFR binding and dimerization.
We decided to use NMR spectroscopy techniques to investigate the molecular basis for the higher potency of HM8 over HM6 in promoting FGF2·FGFR4 dimerization and signaling. Crystal structures of several FGF·FGFR complexes have established that both D2 and D3 are needed for FGF binding (8). In contrast to the well resolved Ig domains D1 (64) and D2 (60, 65), the D3 domain of all investigated FGFR (FGFR1/2/4) constructs (D1D2D3, D2D3, and D3) lead to line broadening beyond detection in solution. The structural integrity of the investigated D3 domain was confirmed by x-ray analysis (successful crystallization of FGFR2 D2D3·FGF1). Therefore, our NMR study focused on the Ig domain 2, the heparin binding domain of the receptor.
We used chemical shift perturbations (CSPs) and changes in the signal intensities of one 15N-labeled protein component upon addition of the second, unlabeled component to test if the isolated D2 domain of FGFR4 is capable of binding FGF. Backbone assignments of FGF1 and FGF2 have been reported previously (58, 59) and were used in this study. Consistent with published FGF·FGFR x-ray structures, no CSPs in two-dimensional 1H,15N-correlation spectra of 15N-labeled FGF1 or FGF2 were observed upon titration with unlabeled FGFR4 D2. We also prepared uniformly 15N-labeled D2 and titrated it with unlabeled FGF1 and FGF2. No CSPs were observed in HSQC of 15N-labeled D2 at protein concentrations of 50–100 μm. These data are consistent with published crystal structures indicating that both D2 and D3 are needed for FGF binding (8).
The crystal structures of FGF·FGFR·heparin ternary complex show that D2 mediates binding of FGFR to HM. In both these structures heparin interacts simultaneously with HM binding on D2 of FGFR1 (23) and FGFR2 (24) and the HSGAG binding site of FGFs. This observation suggests that, although D2 alone is not sufficient for high affinity FGF binding by itself, the presence of heparin increases the binding affinity of FGF to D2. Indeed a pervious ITC study has shown that the presence of sucrose octasulfate, a heparin analog, enhances binding of FGFR2 D2 to FGFR1 (60).
To test if HM6 and HM8 can promote binding of FGF1 and FGF2 to D2 of FGFR4 we first confirmed that isolated D2 of FGFR4 is capable of binding heparin. Addition of HM6 and HM8 to 15N-labeled FGFR4 D2 led to CSPs in two dimensional 1H,15N-correlation spectra (supplemental Fig. 1). To identify the HSGAG binding residues of D2 we then assigned the NMR backbone resonances of FGFR4 D2 using standard triple resonance NMR experiments on a 13C,15N-labeled sample.4 D2 showing CSPs include Lys158, Leu159, Val168, and Lys169. These D2 residues are homologous to the heparin-binding residues of FGFR1 and FGFR2 D2 seen in the x-ray structures of ternary complexes (PDB entries 1FQ9 (23) and 1E0O (24)). Interestingly, in addition to CSPs close to the HM binding site, strong CSPs and severe line broadening were observed for D2 residues Ser141 and Tyr142 located at the N-terminal tip of D2 in the presence of either HM.
Next, we tested whether HM can support binding of FGF1 and FGF2 to D2. Two complementary titration experiments were performed: In the first experiment, uniformly 15N-labeled FGFR4 D2 was titrated with a 1:1 mixture of unlabeled FGF with HM6 or with HM8. In the second experiment, 1:1 mixtures of 15N-labeled FGF and HM were titrated with unlabeled FGFR4 D2. In both these titration experiments, a major reduction in signal intensity of backbone resonances of the 15N-labeled component (FGF1, FGF2, or D2) was observed suggesting that indeed, in the presence of HM, D2 is capable of binding FGF. Analysis of the line width of individual backbone amide resonances of D2 and FGF allowed us to identify residues in FGF and D2 that are involved in protein-protein and protein-HM interactions (supplemental Table I).
No structure of FGFR4 D2 has been reported so far. However, the sequence identity between the extracellular domains of FGFR1, FGFR2, and FGFR4 is very high (over 50%). Therefore, NMR results on FGFR4 D2 can be analyzed using the available ternary complex structures of FGFR1 and FGFR2. The regions in FGF1 and D2 that are affected in the FGF1·D2·HM8 titration experiment are mapped onto both x-ray structures (the symmetric 2:2:2 complex 1FQ9 (Fig. 2) and the asymmetric 2:2:1 complex 1E0O (Fig. 3)). FGFR4 D2 residues that undergo signal attenuation upon titration with an unlabeled FGF1·HM8 mixture include the predicted HSGAG-binding site of FGFR4 D2 and the predicted primary ligand binding site of D2. These two regions are identical between the two crystallographic models. Interestingly, other affected D2 residues include Gly165, Asn166, Thr167, and Glu212 and correspond to the D2 regions that are expected to mediate the direct D2–D2 interface (Figs. 2 and and3,3, cyan circles) and the secondary FGF·FGFR D2 interface (Figs. 2 and and3,3, yellow circles) observed in the symmetric model.
Residues in FGF1 that undergo signal attenuation upon titration with unlabeled D2 correspond to the FGF1 HSGAG-binding site and primary receptor binding site (supplemental Table I). Interestingly, in addition CSPs and/or signal attenuation were observed for the residues mapping to the region on FGF1 that is predicted to mediate the secondary FGF·FGFR D2 interface (Figs. 2 and and3,3, yellow circles) based on the symmetric dimerization model. This secondary FGF·D2 interaction site was also described by Kochoyan and co-workers for a complex between FGFR1 D2 and FGF1 in the absence of heparin (65). Notably, the NMR data show that HM6 and HM8 generally induce perturbations and signal attenuation of the same residues in FGFR4 D2 and in FGF1 and FGF2 (FGFR4 D2·FGF2·HM6/8 mapped on the 1FQ9 x-ray structure (supplemental Figs. 2 and 3)). We therefore conclude that NMR data show the symmetric dimerization model in solution.
Strong line broadening of D2 and FGF residues indicates that the FGF·D2·HM complex is highly dynamic with exchange rates between different oligomerization states on the microsecond to millisecond timescale. Thus, we carried out T2 NMR relaxation measurements to gain more insights into the dynamics and stoichiometry of the FGF·D2·HM complexes. T2-derived apparent molecular weights of FGF1/2 and FGFR4 D2 in the absence and presence of either HM6 or HM8 are summarized in Table 1. Consistent with the results of NMR titration experiments described above in the absence of HM, the predicted molecular weights for monomeric FGF1, FGF2, and FGFR4 D2 were observed. In the presence of HM, however, higher molecular weight species were detected (Table 2). The highest molecular weights were observed for FGF1 and FGFR4 D2 in the presence of HM8. The measured apparent molecular mass of FGF1 increased from 21 to 44 kDa for 15N-FGF1 in complex with unlabeled FGFR4 D2 and HM8. At the same time, the apparent molecular mass of FGFR4 D2 increased from 21 to 37 kDa for 15N-FGFR4 D2 in the complex. For a stable 2:2 dimeric FGF·D2 complex, values of ~60 kDa would have been expected for both complexes. Therefore, smaller apparent molecular masses together with different apparent molecular masses for FGF and FGFR4 D2 suggest that both proteins are in a dynamic equilibrium between the free form and higher molecular mass complexes with a stoichiometry greater than 1:1:1 FGF·D2·HM (theoretical molecular mass of 32 kDa) but smaller than 2:2:2 in solution at concentrations of 50 μm. The increase in T2-derived apparent molecular weights was greater for FGF1 than for FGF2, and greater with HM8 than with HM6 indicating that the equilibrium between ternary complex and lower molecular weight species in solution is shifted prominently toward the ternary complex for FGF1 versus FGF2 and for HM8 versus HM6.
We also studied the interaction of FGFR4 D2 with FGF1 and FGF2 in the absence and presence of HM using ITC. In the absence of HM, titrations of FGFR4 D2 into FGF1 or FGF2 resembled control titrations of FGFR4 D2 into buffer (supplemental Fig. 4). Titration of D2 into a 1:1 mixture of FGF1·HM6 and FGF2·HM6 also generated no heat signal. In contrast, titration of FGFR4 D2 into a 1:1 mixture of FGF1·HM8 and FGF2·HM8 showed a binding event with a KD of ~1 μm and a stoichiometry near unity (Table 1 and supplemental Fig. 4). These data indicate that HM8 forms a ternary complex with FGF and D2, whereas complexes mediated by HM6 are not stable at these protein concentrations (10 μm). The prominent dilution heat signal that was observed in the titrations of FGFR4 D2 into buffer possibly indicates dissociation of protein oligomers; a tendency for self-association of FGFR4 D2 was also observed by NMR spectroscopy (discussed below).
Taken together, NMR and ITC experiments indicate differences between HM6 and HM8 in their ability to promote FGF·D2 complex formation and dimerization. However, NMR CSP and signal attenuation studies show HM6 and HM8 generally induce effects on the same residues in FGFR4 D2 and in FGF1 and FGF2 suggesting that the greater efficacy of HM8 relative to HM6 to promote dimerization cannot be explained by additional contacts of two extra sugar units of HM8 with FGFR D2 or FGF. These findings imply that differences in the binary interactions of HM with FGF and/or FGFR D2 may underlie the observed differences in the biological potency of these two HM.
It has been reported previously that heparin can induce dimerization/higher order oligomerization of FGF1 or FGF2 (66, 67). Therefore, we compared the binding affinities of HM for FGF1 and FGF2 and the abilities of HM to induce oligomerization of FGF1/2 using NMR spectroscopy. As expected, titration of HM6 and HM8 to both 15N-labeled FGF1 and FGF2 (NMR backbone assignments (58, 59)) caused strong CSPs and line broadening of the amino acid residues comprising the HSGAG-binding site of the ligands (supplemental Fig. 5 and Table II). Compared with HM6, HM8 induced stronger CSPs and signal attenuation of 15N-labeled FGF1 and FGF2 than HM6 in agreement with higher affinity of HM8 over HM6 to FGF1 and FGF2.
Next we used NMR 1H T2 relaxation measurements to study HM-induced FGF dimerization. The 1H T2-derived apparent molecular weights of the FGF·HM complexes are given in supplemental Table III. The data show that both HM6 and HM8 are capable of dimerizing FGF. However, the smaller apparent molecular weight of the FGF dimer in the presence of HM6 pointed to dynamic monomer-dimer equilibrium on the NMR timescale. The dynamic equilibrium in the case of HM6 was also confirmed by size exclusion chromatography (SEC) (supplemental Fig. 6 and Table IV). In the presence of HM6, FGF eluted at a retention time midway between the expected retention times for monomeric and dimeric FGF. In contrast, upon addition of HM8, the FGF2 peak shifted to a retention volume, which correspond to the FGF2 dimer (supplemental Table IV).
Stepwise titration of HM to either FGF followed by NMR 1H T2 relaxation measurements allowed us to further elucidate the dimerization mechanism. The T2-derived apparent molecular weight as a function of the HM:FGF ratio was measured (Fig. 4E). Maximum dimerization was achieved at a HM:FGF2 ratio of 0.5. This finding indicates an HM:FGF2 stoichiometry of 1:2, suggesting a two-step binding model (Fig. 4F). At ratios higher than 0.5, the apparent molecular weight of FGF2 decreased again, in line with a partial dissociation of the HM·FGF dimer in the presence of an excess of HM. Although dimer association at low HM:FGF ratios was similar for both HM, dimer dissociation was more pronounced for HM6 than for HM8.
Theoretical fitting of the HM6 titration into FGF2 resulted in a two-step binding model with similar KD values for the two binding events (Fig. 4E, open triangles and black line; KD1 = 160 nm, KD2 = 120 nm). In contrast, the HM8·FGF2 interaction showed a pronounced positive cooperativity (Fig. 4E, filled triangles and black line; KD1 = 100 nm and KD2 = 5.8 nm). Taken together, the NMR data suggest that for the HM8·FGF2 interaction, the recruitment of the second FGF2 to form the dimeric complex HM8·(FGF2)2 occurs with higher affinity than the first step indicating positive cooperativity of binding, in agreement with previous reports (68).
The ability of HM to dimerize FGFs was also compared using ITC at 10-fold lower protein concentration than in the NMR 1H T2 relaxation measurements (Fig. 4, C and D). Both HM bound to FGF with sub-micromolar affinities (Table 1). Within experimental error, HM6 exhibited similar affinity for FGF1 and FGF2. Compared with HM6, HM8 bound 2-fold tighter to FGF1. In the case of FGF2 the difference in binding affinities between the HM was substantially larger: HM8 bound with 15-fold higher affinity than HM6 (54 versus 843 nm) to FGF2.
We also used ITC to compare the efficacy of HM to dimerize FGF1 and FGF2. ITC experiments showed that HM6 bound to FGF with a 1:1 binding stoichiometry, whereas a stoichiometry of 0.5:1 was observed reproducibly for HM8·FGF binding (Table 1). This finding indicates that one molecule of HM8 bound two FGF molecules at half-saturation, whereas HM6 bound only one FGF molecule. However, we cannot exclude the possibility that HM6 also induces the formation of a transient dimer, which dissociates upon titration of excess ligand; this scenario would require two comparable KD values, i.e. no or only weak cooperative binding. HM8 displayed a binding enthalpy ΔH that was on average 2-fold higher than the binding enthalpy of HM6. This feature is indicative of positive cooperativity and is in agreement with NMR data (described above). Apparently, with HM8, the binding process of two FGF molecules titrated simultaneously, resulted in an additive binding enthalpy ΔH1 + ΔH2. However, the symmetric sigmoid-shaped titration curves did not allow deconvolution of two separate binding events with a two-site binding model. Therefore, only apparent macroscopic KD values could be determined (Table 1).
Finally, the interaction of FGF and HM was also monitored using differential scanning fluorometry. Both HM6 and HM8 increased the melting temperature of FGF1 and FGF2, (supplemental Fig. 5 and Table V). HM6 had a similar effect on both FGFs, leading to an increased melting temperature Tm = 54.29 °C (FGF1, Tm = 42.55 °C) and Tm = 53.03 °C (FGF2, Tm = 46.48 °C). HM8 led to a small increase of the melting temperature in the case of FGF1 (Tm = 57.99 °C). However, a much stronger shift to higher Tm values was observed for HM8 with FGF2 (Tm = 68.89 °C). These observations support the finding that HM8 has a higher affinity especially toward FGF2. Taken together, our NMR, ITC, and SEC data consistently show that both HM can dimerize FGF, but dimerization is more efficient for HM8 compared with HM6.
Based on the published FGF-heparin crystal structures FGF binds to an internal trisaccharide binding motif (21, 22). Thus we decided to study binding of HM to both FGF1 and FGF2 by monitoring the anomeric proton resonances of HM in 1H one-dimensional NMR spectra (Fig. 4, A and B). Conveniently, the anomeric protons of the oligosaccharides fall in the region between 5.0 and 5.5 ppm where no protein signals occur. At an HM:FGF ratio of 1:1, the anomeric resonances of all sugar units of HM6 and HM8 become broadened beyond detection indicating the presence of multiple bound conformations, which interchange on the microsecond to millisecond timescale. No difference could be detected in the line broadening between the terminal and non-terminal sugar units. Based on the FGF-heparin crystal structures we would have expected to observe differences in line broadening between the internal trisaccharide binding motif and the terminal units. The lack of difference in the extent of line broadening between the terminal and internal sugar units indicate that all sugar units participate in ligand binding. Thus, based on our NMR data, FGF interacts in a degenerate fashion with all sugar units of HM with exchange of the units between the free and different protein-bound states on a time scale of microseconds to milliseconds.
Mathematical formulation of the degeneracy in the FGF2·HM interaction shows that an increased degeneracy of FGF binding sites for HM8 relative to HM6 alone is not sufficient to explain the differences in the affinity of HM6 and HM8 for FGF2. Based on a trisaccharide binding motif (see below), HM8 offers three degenerate binding sites compared with only two binding sites on HM6. Supplemental Derivation I demonstrates that the increased number of FGF binding sites results for HM8 in a higher affinity for both binding events (1.5-fold for binding of the first FGF; 1.3-fold for the second FGF). This theoretical formulation agrees well with the experimental data for the first HM·FGF interaction. However, for the second binding event, a 20-fold difference in affinity is observed experimentally for HM8 compared with HM6. In addition, binding site degeneracy alone results in a lower affinity for the second FGF·HM interaction compared with the first one (HM6: 4-fold higher affinity; HM8: 4.5-fold) and therefore cannot explain the observed positive cooperativity.
We also studied the ability of HM to induce dimerization/oligomerization of FGFR4 D2. T2 experiments indicate an increase in apparent molecular mass from 21 ± 3 kDa in the absence of HM8 to 28 ± 7 kDa in the presence of HM8 (expected monomeric molecular mass for FGFR4 D2: 15 kDa). These findings provide structural evidence for a FGF-independent, HM-induced dimerization of FGFR4, which was previously indicated by biological data (69).
Binding of HM to FGFR4 D2 was also measured by ITC (supplemental Fig. 8). For both HM, similar affinities in the low micromolar range (HM6: 2.4 μm, HM8: 3.4 μm) and stoichiometries near unity were obtained (Table 1). In a previous study, similar affinities of 0.3–0.4 μm have been reported for the HM·FGFR4 interaction (70, 71). Lower stoichiometries for HM8 than for HM6 (Table 1) possibly indicate a partial HM8-induced dimerization as observed by SEC (supplemental Table IV) and NMR. Binding thermodynamics differed strongly for the two HM. HM8 had a much larger binding enthalpy compared with HM6 (−9.3 versus −1.2 kcal/mol). In contrast, binding of D2 to HM6 was primarily entropy-driven, indicating a different binding mode for HM6 as compared with HM8. It has been reported previously that a heparin octamer constitutes the minimal chain length for stable HM binding to FGFR4 (70). The deviating binding thermodynamics observed with HM6 could therefore point to an alternative, possibly nonspecific binding mode. Taken together, our data show that HM6 and HM8 differ greatly in their affinity for FGF1/2 and have different ability to dimerize FGFs, whereas both HM exhibit comparable affinities for D2 and have poor ability to dimerize FGFR4 D2.
Using organic chemistry methods we were able to de novo synthesize chemically pure sulfated heparin hexasaccharide (HM6), octasaccharide (HM8), and decasaccharide (HM10). We showed that HM8 is significantly more potent than HM6, while HM10 displays similar potency as HM8 in promoting FGF2-mediated FGFR4 signaling. Interestingly, the prominent effect in biological activity of the octasaccharide relies on its interaction with FGF2, because FGF1·HM6/8/10 activation only resulted in relatively small changes in FGFR4 signaling.
To delineate the underlying molecular basis for higher efficacy of HM8 relative to HM6 we used NMR, ITC, and SEC to characterize binding interactions of FGF1 and FGF2 with isolated Ig-domain 2 (D2) of FGFR4 in the presence of heparin mimetics, and binary interactions of FGFs and D2 with HM. Our NMR data in solution support the symmetric dimerization model proposed by Mohammadi and colleagues (23) for HM6 and HM8 in contrast to the conclusion of Goodger and coworkers (53) who proposed an asymmetric ternary complex formation (24) for octasaccharides and larger heparin fragments. Our observed symmetric complex in solution is transient, and HM8 (compared with HM6) shifts the equilibrium more strongly toward the ternary complex for both FGF1 and FGF2. Importantly, the NMR data show that the higher efficacy of HM8 relative to HM6 is not due to the ability of HM8 to make additional contacts with FGF or D2.
Analysis of the binary interaction of HM with the D2 domain of FGFR4, which harbors the heparin binding site, shows that HM6 and HM8 bind D2 with ~1 μm KD, in agreement with previous reports (70, 71). Therefore, the differences in affinity appear too small to account for the observed differences in potency between HM6 and HM8. A tendency of HM8 to dimerize the receptor already in the absence of FGF could be observed, an interesting finding in light of the reported heparin sensitivity of FGFR4 (69).
In contrast to the findings of Goodger and coworkers (53) that an octasaccharide is the shortest heparin fragment to form a 2:1 FGF·heparin complex, in our study both HM6 and HM8 were able to dimerize FGF1 and FGF2, above a protein concentration of 50 μm. NMR-based epitope mapping revealed closely overlapping binding sites of HM6 and HM8 on FGF1 and FGF2. The observed binding epitopes correspond to the heparin binding sites observed in the published crystallographic studies (21, 22).
Importantly, FGF2 dimerization upon HM8 addition reveals strong positive cooperativity, in agreement with previous studies on both FGF1 (68) and FGF2 (53). Furthermore, HM8 has a stronger propensity to dimerize FGF than HM6 and binds with 15-fold higher affinity to FGF2. By contrast, for FGF1 the affinity for HM8 is only 2-fold higher than for HM6.
There are two potential mechanisms that could account for the differences in affinities of HM6 and HM8 toward FGFs. Firstly, the affinity can be increased due to a degeneracy effect observed in multivalent interactions. Secondly, the interactions of HM with both FGF and FGFR are highly dynamic, as observed by NMR line broadening experiments. Our findings are in line with recent surface plasmon resonance data (72), where rapid association and dissociation kinetics for the FGF·HM complex and rapid dissociation for the receptor·HM complex were demonstrated. As the oligosaccharides offer multiple binding motifs for FGF, the fast dynamics result in multivalent binding. As a consequence, for both HM6 and HM8 all sugar units are involved in the interaction with at least one FGF molecule in agreement with a previous NMR study (73). The presence of multiple, overlapping binding sites in the longer oligosaccharide could result in apparent higher FGF binding affinities (57). However, a mathematical formulation of the influence of multivalency on the affinities of the hexameric versus octameric oligosaccharide (supplemental Derivation I) shows that the degeneracy effect can explain neither the observed cooperativity nor the difference between FGF1 and FGF2. Rather, we detect strong cooperativity for the second binding event only in the case of FGF2 with HM8.
To gain insights into the molecular basis of the observed cooperativity in FGF2·HM8 binding on the basis of the reported structures, we analyzed the interactions of FGF1 and FGF2 with heparin oligosaccharides in the published crystal structures (PDB-entries 2AXM, 1E0O, 1FQ9, 1BFB, and 1BFC). Striking differences between the binding modes of FGF1 and FGF2 to heparin oligosaccharides were observed. Based on the structures of FGF1 in complex with a decasaccharide (22) (PDB entry 2AXM) or FGFR2 D2D3-decasaccharide (24) (PDB entry 1E0O), FGF1 interacts with the HM sulfate groups of the sequence GlcN-IdoA-GlcN (GIG) (supplemental Fig. 9, left). In contrast to FGF1, the available crystal structures of FGF2 in complex with heparin differ in their sugar interaction pattern: in the FGF2·FGFR1c-decasaccaride structure (1FQ9) FGF2 binds to a IdoA-GlcN-IdoA (IGI) motif, whereas, in the FGF2-tetra/hexasaccharide structure (1BFB and 1BFC (21)), FGF2 interacts with a IdoA-GlcN (IG) motif. Based on our NMR CSP mapping data (Gly28, see supplemental Table I) we suggest that FGF2 binds to the IGI motif for FGF2 in solution as observed in 1FQ9 (23) (supplemental Fig. 9, right).
Different symmetries are observed for the interactions of FGF1 and FGF2 with HM in the ternary complexes. The conformations of the sugars bound to FGF1 and FGF2 are therefore also different (supplemental Fig. 9). The affinity for the first binding event of HM8 is similar for FGF1 and FGF2. Because the structure of free HM (PDB entry 1HPN) differs from the conformation of bound HM, the first binding event of both FGF1 and FGF2 must change the conformation of the HM. In structural terms, positive cooperativity is consistent with the reported data that binding of the first FGF induces a kink (74) in the structure of HM, which then, together with entropic factors, facilitates binding of a second FGF. However, the affinity of the second binding event for FGF1 is in the same KD range as the first step. Therefore, there is no indication for a difference in the binding mode for the first and second step for FGF1. In stark contrast, for FGF2 the second step affinity is 17-fold stronger than the first association of FGF2 to HM8. Thus, according to our data, the strong ability of HM8 to bind and dimerize FGF2 molecules in a cooperative fashion correlates with the higher efficacy of HM8 relative to HM6 to promote FGF2 signaling. Recent studies of heparin dendrimers (75) induced FGF-oligomers, synthetically polymerized FGFs (76), and covalently cross-linked FGF-dimers (77) suggest that oligomerization of FGF is required for an agonistic effect of HM on FGFR signaling. Therefore, the oligomerization of FGF in the presence of longer HM such as HM6 and HM8 needs to be considered in the formation of FGF·FGFR cell surface signaling assembly. Based on our findings we propose a refined symmetric FGF·FGFR dimerization model (Fig. 5), which incorporates the differential ability of HM to dimerize FGFs and the presence of a symmetric 2:2:2 ternary complex in solution (Fig. 5, A and B).
The (FGF2)2·HM dimer is structurally not part of the symmetric ternary crystal complex, which therefore implies a dissociation of the (FGF2)2·HM for the assembly of the ternary crystal complex. Due to the cooperativity of the FGF2·HM8 dimerization, it is unlikely that a FGF2·HM8 (1:1) monomer interacts with the receptor. This discrepancy is addressed in Fig. 5 by postulating a 2:1:1 intermediate state (Fig. 5B, IVb). This model could explain the transition from the (FGF2)2·HM dimer to the symmetric 2:2:2 complex.
Assuming that FGF2 binds to an IGI motif (supplemental Fig. 9), the initial 1:1 FGF2·HM6 complex can form through two alternative ways for HM6 (Fig. 5A, IIa and IIb). Binding of a second FGF molecule leads to one unique dimeric species (Fig. 5A, III). This dimeric species cannot directly bind to the receptor because of potential steric clashes between FGF2 and the receptor (Fig. 5A, IVa, Fig. 6A). Therefore, the 1:1:1 FGF2·FGFR·HM6 complex (Fig. 5A, IVb) can only be formed when the binding of (FGF2)2·HM6 to the receptor, and dissociation of the second FGF2 are synchronous or from the FGF2·HM6 complex with a single FGF2 bound to HM6 directly. In contrast to HM6, HM8 can form three different 1:1 FGF2·HM8 complexes (Fig. 5B, IIa, IIb, and IIc). Presuming the conformation of the oligosaccharide in the symmetric ternary complex, the dimerization step can either lead to an FGF dimer in analogy to the HM6 complex (Fig. 5B, IIIa), or the FGF molecules bind one to the middle binding site and one to the reducing end site of the saccharide (Fig. 5B, IIIb). The cis-binding mode where both FGF2s are at the outer binding sites can be excluded, because it would lead to direct FGF2·FGF2 contacts that are not observed in two-dimensional NMR experiments. The dimerization step has a significantly higher affinity compared with the first FGF2 association step for HM8. Because HM6 is chemically contained within HM8, it can be assumed that the 2:1 FGF2·HM8 (Fig. 5B, IIIa) has a similar structure and is equally stable as the corresponding complex with HM6 (Fig. 5A, III). Therefore, the high cooperativity upon HM8-induced dimerization is likely a result of complex formation involving the additional sugar units of HM8 at the reducing end (Fig. 5B, IIIb). We speculate that cooperativity is mediated by a different binding mode of the second FGF2 binding to HM8 leading to a GIG binding motif shown in IIIb to be the favored dimer. The crystal structure of FGF2·FGFR1·decasaccharide (1FQ9) shows a GIG binding motif for the second FGF2 at the reducing end of decasaccharide (23). The fact that eight sugars are necessary for the biological effect further supports that the binding mode of the formation of IIIb is according to a GIG motif. Additionally, the contacts of HM to the second FGF2 are formed by three sulfate groups (GIG) instead of two sulfates and a carboxyl group for the IGI motif, which could also explain higher affinity for the second FGF2 binding. The direct binding of the (FGF2)2·HM8 (Fig. 5B, IIIa) complex in analogy to the (FGF2)2·HM6 (Fig. 5A, III) complex to the receptor would also lead to the same steric clashes as for HM6 (Fig. 5A, IVa, Fig. 6A). The favored (FGF2)2·HM8 complex, however, can form a 2:1:1 FGF2·FGFR·HM8 complex (Fig. 5B, IVb) as an intermediate state. The ability of the (FGF2)2·HM dimer to interact directly with FGFR constitutes the key difference between the octameric as compared with the hexameric HM. The formation of a symmetric ternary assembly of stoichiometry 2:2:2 in analogy to PDB entry 1FQ9 would require the dissociation of one FGF2 and subsequent dimerization of two 1:1:1 complexes. Alternatively, we propose that the resulting complex could dimerize directly to a 4:2:2 FGF2·FGFR·HM8 complex. As depicted in Fig. 6B and 6C, the proposed complex is sterically possible. This mechanism could explain the increased potency of HM8 versus HM6 based on a symmetric complex, because HM-induced dimerization is one of its steps, and therefore this step can also be rate-limiting. On the basis of our model, it remains unclear if the 4:2:2 complex induces signaling directly or if it constitutes an intermediate state.
Our biophysical studies provide a detailed view on the interaction of FGF, FGF receptors, and heparin. The striking biological observation that oligosaccharide length and specificity toward FGF2 translates into differences in signaling is based on unique dimerization properties of the FGF2·HM8 complex due to both multivalent, dynamic interaction of FGF with HM and the positive cooperativity only observed for FGF2·HM8. Given the fact that there is significant interest in pharmaceutical development of mimetics of HSGAG as potential drugs to modulate FGF signaling, our data should facilitate the rational design of heparin mimetics as agonists of FGF2·FGFR signaling.
4K. Saxena, U. Schieborr, O. Anderka, E. Duchardt-Ferner, B. Elshorst, S. L. Gande, J. Janzon, D. Kudlinzki, S. Sreeramulu, M. K. Dreyer, K. U. Wendt, C. Herbert, P. Duchaussoy, M. Bianciotto, P.-A. Driguez, G. Lassalle, P. Savi, M. Mohammadi, F. Bono, and H. Schwalbe, unpublished results.
3The abbreviations used are: