Analysis of the fibroblast growth factor receptor (FGFR) signalling network with heparin as coreceptor: evidence for the expansion of the core FGFR signalling network

Authors

  • Ruoyan Xu,

    1. Department of Structural and Chemical Biology, Institute of Integrative Biology, University of Liverpool, UK
    Search for more papers by this author
  • Timothy R. Rudd,

    1. Department of Structural and Chemical Biology, Institute of Integrative Biology, University of Liverpool, UK
    2. Diamond Light Source Ltd, Harwell Innovation Campus, Didcot, UK
    Search for more papers by this author
  • Ashley J Hughes,

    1. Department of Structural and Chemical Biology, Institute of Integrative Biology, University of Liverpool, UK
    2. Diamond Light Source Ltd, Harwell Innovation Campus, Didcot, UK
    Search for more papers by this author
  • Giuliano Siligardi,

    1. Department of Structural and Chemical Biology, Institute of Integrative Biology, University of Liverpool, UK
    2. Diamond Light Source Ltd, Harwell Innovation Campus, Didcot, UK
    Search for more papers by this author
  • David G. Fernig,

    1. Department of Structural and Chemical Biology, Institute of Integrative Biology, University of Liverpool, UK
    Search for more papers by this author
  • Edwin A. Yates

    Corresponding author
    • Department of Structural and Chemical Biology, Institute of Integrative Biology, University of Liverpool, UK
    Search for more papers by this author

Correspondence

E. A. Yates, Department of Structural and Chemical Biology, Institute of Integrative Biology, University of Liverpool, Crown Street, Liverpool L69 7ZB UK

Fax: +44 (0) 151 795 4406

Tel: +44 (0) 151 795 4429

E-mail: eayates@liv.ac.uk

Abstract

The evolution of the fibroblast growth factor (FGF)–FGF receptor (FGFR) signalling system has closely followed that of multicellular organisms. The abilities of nine FGFs (FGF-1 to FGF-9; examples of FGF subfamilies 1, 4, 7, 8, and 9) and seven FGFRs or isoforms (FGFR1b, FGFR1c, FGFR2b, FGFR2c, FGFR3b, FGFR3c, and FGFR4) to support signalling in the presence of heparin, a proxy for the cellular heparan sulfate coreceptor, were assembled into a network. A connection between two FGFRs was defined as their mutual ability to signal with a particular FGF. The network contained a core of four receptors (FGFR1c, FGFR2c, FGFR3c, and FGFR4) with complete connectivity and high redundancy. Analysis of the wider network indicated that neither FGF-3 nor FGF-7 was well connected to this core of four receptors, and that divergence of a precursor of FGF subgroups 1, 4 and 9 from FGF subgroup 8 may have allowed expansion from a three-member FGFR core signalling system to the four-member core network. This increases by four-fold the number of possible signalling combinations. Synchrotron radiation CD spectra of the FGFs with heparin revealed no overall common structural change, suggesting the existence of distinct heparin-binding sites throughout the FGFs. The approach provides a potential method of identifying agents capable of influencing particular FGF–FGFR combinations, or areas of the signalling network, for experimental or therapeutic purposes.

Abbreviations
FGF

fibroblast growth factor

FGFR

fibroblast growth factor receptor

GlcNS

N-sulfaminoglucosamine

GlcNS(6S)

N-sulfamino, 6-O-sulfated glucosamine

HBS

heparin-binding site

HS

heparan sulfate

HSPG

heparan sulfate proteoglycan

PDB

Protein Data Bank

SRCD

synchrotron radiation CD

ΔUA

4,5-unsaturated uronate residue

ΔUA(2S)

4,5-unsaturated, 2-O-sulfated uronate residue

Introduction

The fibroblast growth factors (FGFs) regulate many aspects of embryonic development and adult homeostasis. The 22 human FGFs are usually divided into seven subfamilies according to their sequence similarities: subfamilies 1, 4, 7, 8, 9, 11, and 19. These correspond broadly to certain functions, although the relationships of these remains a matter of debate, and the FGF-11 subfamily is intracellular and does not interact with extracellular receptors [1, 2]. In humans, the 22 fgf genes encode the ligands and five fgfr genes encode the cognate membrane receptors, FGF receptors (FGFRs). In simpler organisms, less elaborate FGF–FGFR signalling systems are evident. For example, in Caenorhabditis elegans, only two fgf genes and one fgfr gene have been identified, indicating that the FGF ligand–receptor system has expanded during the evolutionary process from primitive metazoa to vertebrates [1, 2].

In structural terms, FGFRs consist of two or three extracellular immunoglobulin domains, linked by a single transmembrane region to a cytoplasmic domain, which contains a tyrosine kinase in FGFR1 to FGFR4 [1, 3]. The latter becomes phosphorylated following binding of an FGF to the receptor together with its obligatory coreceptor, heparan sulfate (HS), which is the carbohydrate portion of cell surface HS proteoglycans (HSPGs). Phosphorylation of the cytoplasmic domain leads to downstream signalling events. The FGFR signalling system has been implicated in both normal developmental and aberrant processes, and the possibility of controlling it remains of great interest for the study and eventual treatment of many diseases, and to promote tissue repair [4-6]. Alternative splicing produces receptors with differences in the third immunoglobulin domain of FGFR (D3) and/or shortened versions, in which the acid box between D1 and D2 is absent. According to the crystal structures of FGFs with receptor domains D2 and D3, FGFs bind to D2 and the proximal regions of D3, and this may explain the specificity for FGFs between splice variants [7, 8]. FGFR1 and FGFR2 show distinct types of binding to heparin [9], and the crystal structures of complexes of FGF and FGFR in the presence of heparin-derived oligosaccharides provide evidence of direct interaction of the sugar with each protein component, albeit in two distinct arrangements [10, 11]. The stimulation of cell proliferation requires the formation of a ternary complex of the FGF, the FGFR, and the HS coreceptor, a requirement that has been demonstrated in vivo [12]. The HS coreceptor has at least one other key function, as it controls the transport of FGFs from sources to targets [13].

HS, as well as its proxy heparin, are glycosaminoglycan polysaccharides with a common underlying disaccharide backbone structure, comprising a uronic acid, α-l-iduronic or β-d-glucuronic acid, linked alternately through a 1–4-linkage to an α-d-glucosamine residue. Biosynthesis first generates a polysaccharide chain containing 50–100 repeating disaccharides of 1,4-linked β-d-glucuronic acid and α-linked GlcNAc, which is then modified enzymatically in the Golgi, first through N-deacetylation and N-sulfation by N-sulfotransferases acting concomitantly to produce clusters of N-sulfated glucosamines in the chain. These provide sites for the action of subsequent enzymes, including the epimerization of d-glucuronate acid to l-iduronate, and O-sulfation of C-2 on iduronate residues and of C-6 and C-3 on glucosamine residues [14]. The result is that modifications tend to be clustered and resemble distinct domains of varying dimensions known, respectively, as: nonsulfated (N-acetylated) domains; intermediate domains, which have approximately one in two d-glucosamine units modified by N-sulfation; and sulfated domains, in which glucosamine is N-sulfated. A branched arrangement of this biosynthetic pathway, based on a disaccharide unit and commensurate with the abundances of experimentally observed structures, has been suggested recently [15]. The more highly sulfated regions of HS resemble heparin, but contain fewer O-sulfate groups, and are generally considered to provide protein-binding sites [14].

The interactions of some FGFs with HS and its derivatives have been subjected to scrutiny in vitro, but their levels of specificity and selectivity are actively debated (e.g. reference [16]). A recent biophysical analysis suggested that different FGFs bind to heparin in distinct ways to different, but overlapping, motifs in the polysaccharide [17]. It has been proposed that a higher degree of specificity may exist at the level of the full ligand–receptor system [16, 18]. In vitro, some specificity is evident, in particular regarding the preferences of FGFRs for FGFs. However, the majority of FGF subfamilies can support signalling through several receptors in the presence of heparin, as a proxy for HS [18, 19]. This is likely to identify all of the possible signalling combinations, because, generally, the extensively sulfated heparin activates all pathways in which only a restricted set of forms of HS are active. In vivo ternary complexes containing the FGF, HS and FGFR can be formed in situ only on HS polysaccharide coreceptors from particular tissue compartments [20]. Thus, a higher level of selectivity is likely in vivo, owing to the lower sulfation and restricted sets of structures in HS.

The expansion of FGFs and FGFRs during evolution and the pivotal regulatory role played by the HS coreceptor suggest that specificity may be discernible from a consideration of the entire system rather than individual interactions. Thus, we have analysed the properties of the FGF–FGFR–HS signalling system from the point of view of a simple network, in which the ability of a particular FGF to support signalling is represented as connecting any two or more FGFRs through which it can signal. Alongside this, we have measured the changes in the secondary structure of selected FGFs that accompany their interaction with heparin, to determine whether these follow a simple pattern.

Results

Analysis of the FGFR signalling network

One important characteristic of the FGF–FGFR signalling system is its high degree of connectivity. Thus, in many cases, different FGFs are known to have the ability to interact with, and signal through, several FGFRs. The interactions of FGFs with FGFRs have been surveyed previously in the presence of heparin, in terms of both binding and the ability to support signalling in a cell-based assay [19]. Binding events as such do not necessarily equate to the formation of a signalling complex, and so the data reported for signalling in an experimental cell-based assay of mitogenesis for FGF-1 to FGF-9 and with receptors FGFR1b, FGFR1c, FGFR2b, FGFR2c, FGFR3c, and FGFR4, rather than binding, were analysed. This allows a preliminary network to be assembled, employing approaches from discrete mathematics (graph theory), in which two FGFRs (represented by vertices) that both signal with a particular FGF are shown joined (by edges) (represented by Table 1). The ensemble of these ‘connections’ records the ability of the FGF to affect multiple points in a network, and can be analysed in terms of both the individual FGFs and the receptors through which they signal in the presence of heparin but, more interestingly, also forms a graph depicting connectivity between receptors (Fig. 1A). This could represent, for example, a signalling situation that may exist between two cells supported by the same FGF (and HS structures), which express distinct receptors or combinations of receptors. It is evident that some receptors, e.g. FGFR2c, are more heavily connected than others, such as FGFR2b, and the level of redundancy is noteworthy. It can also be seen that there is a highly connected core network, involving FGFR1c, FGFR2c, FGFR3c, and FGFR4, which are all interconnected, and that the connections are heavily redundant (Fig. 1A). The network generated with this approach lends itself to analysis with tools from graph theory [21], which demonstrates that this is a substantially connected network (Table 1). It is known, however, that, beyond this immediate FGFR network, HS interacts with over 400 proteins [22], but there are few experimental data concerning those interactions and any relationship that they may have with FGF signalling.

Table 1. Relative mitogenic activities of individual exogenous FGFs on BaF3 cells lacking HS and in the presence of added heparin (2 μg·mL−1). Data are from [19], with the exception of the value for FGFR1c and FGF-8, which is from [23], are normalized relative to FGF-1, and are rounded to the nearest tenth. The FGF subfamily to which each FGF has been attributed is also shown. An expander is a graph in which each vertex (or node) is connected to each other vertex to produce a network with the same number of connections (edges) emanating from each. Such graphs are extremely robust, as breaking individual connections has little effect on their overall characteristics [21]. Eigenvalue analysis of the matrices representing the full and core networks can provide a measure of their qualities, in particular the difference between the first and second eigenvalues, i.e. the eigenvalue gap. Analysis of the product of the matrix shown in the table with its transpose reveals a difference in the moduli of the first and second eigenvalues of 17.6 − 2.1 = 15.5, which indicates a substantially connected graph [24]
 FGF-n
 123456789
Subfamily 1 1 3 4 4 4 7 8 9
FGFR
1b1.00.60.30.20.00.00.10.00.0
1c1.01.00.01.00.60.60.00.60.2
2b1.00.10.50.10.00.10.80.00.1
2c1.00.60.00.90.30.60.00.20.9
3b1.00.00.00.00.00.00.00.00.4
3c1.01.10.00.70.10.10.00.41.0
41.01.10.11.10.10.80.00.80.8
Figure 1.

(A) Representation of the signalling network between FGFRs. The numbers on the edges represent the number of FGFs that are able to support signalling between the respective FGFRs in the presence of heparin. The products of the published mitogenic activities are also higher in the core network formed by FGFR1c, FGFR2c, FGFR3c and FGFR4 than elsewhere. (B) The highly connected core network involving FGFR1c, FGFR2c, FGFR3c, and FGFR4. The core network can be differentiated into an earlier triangular subnetwork (left) comprising FGFR2c, FGFR3c, and FGFR4, which, upon divergence of FGF subfamilies 1, 4 and 9 from the FGF-8 subfamily, and the evolution of FGR1c, formed a four-member core network (right).

Analysis of the mitogenic signalling activity of a network comprising FGF-1 to FGF-9 with FGFRs of the core network

The data reporting the relative mitogenic activities of individual FGFs with FGFRs [19] (Table 1) can be analysed by determining which FGFs can individually support signalling through pairs of FGFRs, and so form a reciprocating signalling network. This was done by taking the products of the two mitogenic activities between the FGFRs and individual FGFs (Table 2). In this sense, it is clear that FGFR1b, FGFR2b and FGFR3b are all rather poorly connected to other FGFR members, and, even when they do support signalling, it is often to a relatively low extent. Notably, FGF-3 and FGF-7 signal almost exclusively through these FGFRs, forming a subnetwork that is able to utilize FGF-3 and FGF-7, which most of the other FGFRs are unable to do (with the exception of FGFR4 with FGF-3, but then only weakly). Although FGF-3 has been reported to bind FGFR4, no mitogenic activity was reported [25].

Table 2. A measure of signalling between FGFRs in the core network can be found from the product of the mitogenic activities of the two FGFRs (from Table 1) and the particular FGF. This analysis assumes that there is no direct influence exerted on signalling through an FGFR by any other FGFR
 FGF-n
 123456789
FGFR
1c and 2c1.000.600.000.900.180.360.000.120.18
2c and 3c1.000.600.000.810.030.060.000.080.90
3c and 41.001.210.000.770.010.080.000.320.80
1c and 41.001.100.001.100.060.480.000.480.16
2c and 41.000.660.000.990.030.480.000.160.72
1c and 3c1.001.100.001.100.000.480.000.240.16

Exclusion of the poorly connected FGFR1b, FGFR2b and FGFR3b, from Table 2 reveals a core network (Fig. 1A) with complete mutual interconnection (a complete graph) and redundancy. The significance of a complete graph in a signalling network is that it is extremely robust, and high redundancy further ensures the fidelity of the signals in this section of the network. These four FGFRs are generally more highly connected in terms of the number of FGFs that support signalling between FGFR pairs, averaging 6.8 inside the core network and 4.2 outside.

Subsequent to the work of Ornitz et al. [19], Zhang et al. [23] found that FGF-8 had mitogenic activity through FGFR1c, and this work also included additional members of the FGF-8 subfamily. Although there is inevitably some variation in cell proliferation assay data between these two reports, it is interesting that the data reported in reference [23] for the three members of the FGF-8 subfamily (FGF-8, FGF-17, and FGF-18) reveal very weak mitogenic activity for FGF-18 (< 5% of that of FGF-1) with FGFR1c, or weak to moderate activities for FGF-17 (23% of that of FGF-1) and FGF-8 (58% of that of FGF-1) with FGFR1c. Consequently, the products of mitogenic activities for combinations of FGFRs involving FGFR1c are the weakest three (with FGF-18), and three of the four weakest for both FGF-8 and FGF-17. This indicates relatively poor connectivity of FGFR1c via the other members of the FGF-8 subfamily, but particularly FGF-18, with FGFR2c, FGFR3c, and FGFR4.

Interestingly, this implies two distinct regions within the core network. The connections between FGFR2c, FGFR3c and FGFR4 all involve FGF-1, FGF-2 (subfamily 1), FGF-4, FGF-6 (subfamily 4), FGF-8 (subfamily 8) and FGF-9 (subfamily 9), whereas those with FGFR1c from all other FGFRs involve FGF-8 subfamily members much more weakly (Fig. 1A,B). This suggests that divergence from this FGF subfamily by a common ancestor of FGF-1, FGF-4, FGF-6 and FGF-9 coupled to, or followed by, coevolution of FGFR1c (followed by, because FGF evolution is thought to have preceded FGFR evolution [1]) was a key event in allowing expansion of an original three-member, triangular core network with three interconnections (Fig. 1B) to a four-member form, with six interconnections between FGFRs (Fig. 1A). This provides a fourfold increase in the number of possible signalling combinations (single, double, triple, and quadruple). The robust four-member form of the core network has a level of redundancy in each connection of 6 or 7 (from Table 1).

Synchrotron radiation CD (SRCD) spectral changes when different concentrations of heparin are added may relate to the number of heparin-binding sites (HBSs) of FGFs

In light of the above, it was of interest to examine the binary complexes of FGFs and heparin. The SRCD spectra of six FGFs and their heparin complexes were collected under identical conditions (Fig. 2A–F). The spectrum of FGF-18 (belonging to FGF subfamily 8) (Fig. 2E) showed significant changes when heparin was introduced, and FGF-1, FGF-2, FGF-7 and FGF-9 also showed some differences (Fig. 2A–D). On the other hand, FGF-21 (subfamily 19), which binds heparin only very weakly [17], showed very little spectral change (Fig. 2F), confirming the lack of binding and showing that higher levels of heparin do not unduly alter the CD spectrum.

Figure 2.

SRCD spectra of FGFs with different concentrations of heparin. Spectra of six FGFs and their heparin complexes at FGF/heparin (F : H) molar ratios of 5 : 1, 1 : 1, and 1 : 5. (A) FGF-1. (B) FGF-2. (C) FGF-7. (D) FGF-9. (E) FGF-18 (FGF-8 subfamily). (F) FGF-21 (FGF-19 subfamily). Spectra for the FGF alone and for the 1 : 5 molar ratio of FGF/heparin were originally published in reference [17]. © the American Society for Biochemistry and Molecular Biology.

Importantly, the SRCD spectra demonstrated that there was no simple trend in the structural changes observed in the FGFs as the concentration of heparin was increased (that is, there was not a single direction of change). This is illustrated in Fig. 3, where the normalized ellipticity at 203 nm is plotted for FGF-1, FGF-2, and FGF-18. Thus, although structural changes depend on the presence of heparin, higher concentrations do not necessarily produce larger changes in secondary structure in a linear direction (Fig. 4; Table S1). One trend that is apparent in the secondary structure data is evident for FGF-1, FGF-2, FGF-7, and FGF-9. As the heparin concentration increases, so does the content of unordered secondary structure (Fig. 4), but the opposite occurs for FGF-18, the protein becoming markedly more structured (both strands and helices); for FGF-21, there is no change, presumably because it does not interact with heparin. This may be related to the number of binding sites for the sugar in the proteins [17, 26]. When a small amount of heparin is present, most of it will bind to the canonical HBS (HBS-1), as this has the highest affinity, initiating a particular set of conformational changes in the FGF ligands. As more heparin is added, the canonical HBSs become saturated, and the heparin will start binding the lower-affinity HBS-2, HBS-3, and HBS-4, producing additional conformational changes. Whereas the canonical HBS-1 is the one engaged with the sugar in ternary receptor complexes [10, 11], the secondary binding sites may have other functions, including regulation of the transport of FGFs through extracellular matrices [13]. Therefore, the specificity and effects on FGF structure of heparin binding may be more convoluted than those simply required for forming a signalling complex, as they may reflect different functions, e.g. signalling and transport.

Figure 3.

Normalized change in ellipticity at 203 nm for FGF bound to heparin in different ratios. It can be seen that the interaction between heparin and the protein does not follow a simple linear relationship. The FGF and FGF–heparin (1 : 5) complex secondary structures and SRCD spectra were originally published in reference [17]. © the American Society for Biochemistry and Molecular Biology.

Figure 4.

Secondary structure changes of FGFs with different concentrations of heparin. The secondary structures (H1, ordered alpha helix; H2, disordered alpha helix; S1, ordered beta strands; S2, disordered beta strands; T, beta turns; U, unstructured) were determined for six FGFs and their heparin complexes at molar ratios (FGF/heparin) of 5 : 1, 1 : 1, and 1 : 5. (A) FGF-1. (B) FGF-2. (C) FGF-7. (D) FGF-9. (E) FGF-18 (FGF-8 subfamily). (F) FGF-21 (FGF-19 subfamily). Protein secondary structures were determined by use of the Dichroweb website, and the selcon3 algorithm with reference dataset 3 [45, 46]. The FGF and FGF–heparin (1:5) complex secondary structures and SRCD spectra were originally published in reference [17]. © the American Society for Biochemistry and Molecular Biology.

Discussion

The present work indicates that the coevolution of FGFR1c and the divergence of FGF subfamilies 1, 4 and 9 from FGF subfamily 8 may be linked to the expansion of a triangular core network involving FGFR2c, FGFR3c and FGFR4 into a four-member network. This core network is highly redundant, and this presents cells with the possibility of reliably stimulating the highly connected core network with particular FGFs in distinct tissues, while maintaining the fidelity of core signalling events. In graph terminology, this core network is a complete graph. Furthermore, it has good expander properties (Table 1); that is to say, each vertex is connected to each other vertex, and several times over, to produce a very robust system.

Members of FGF subfamily 8 and the other FGFs diverged early in the evolution of multicellular life

The analysis suggests that expansion of the three-member to a four-member core network occurred relatively early in evolution, at the stage of fgf expansion. Two-stages of fgf evolution have been postulated. The first involved fgf expansion in the early metazoan, from three to six genes, by duplication. The second, in early vertebrates, involved large-scale genome duplications. This is in contrast to FGFR family expansion, which occurred during the second phase only, but then achieved additional diversity through splice variants [1]. The analysis by Oulion et al. [2], based on a systematic analysis of recently discovered gene sequences, proposes a reclassification of the fgf gene family into eight, rather than seven, subfamilies, the notable difference being the allocation of FGF-3 to its own subfamily, in contrast to its usual allocation to subfamily 7. It is also interesting that the same analysis indicates that genes encoding the FGF-1 subfamily (FGF-1 and FGF-2) and FGF-8 subfamily (FGF-8, FGF-17, FGF-18, and FGF-24) were already present in the eumetazoan ancestor, indicating divergence of the FGF-8 subfamily from other subgroups at an early stage in evolution. Unlike those encoding glycosaminoglycan biosynthetic enzymes, fgf-like genes have not been identified in the unicellular choanoflagellates, which are postulated to represent the descendants of the last common unicellular ancestor of metazoa [22, 27]. However, two fgf-like genes are present in C. elegans (egl-17 and let-756) [28, 29], and three in Drosophila (branchless, pyramus, and thisbe) [30, 31], supporting their primordial origin.

The early divergence of FGF subfamily 8 from the other FGF subfamilies has been revealed by analysis of amino acid sequences [25], and, in light of the indications here of the importance of the divergence of FGF subfamily 8 for expansion of the core FGFR signalling network from three to four members, it is interesting that the FGFs identified in simple multicellular organisms resemble particular FGFs [32]. In C. elegans, the two FGFs, encoded by the genes egl-17 and let-756, resemble FGF-8 and FGF-9 respectively, whereas in Drosophila, products of both pyramus and thisbe resemble FGF-8.

HBSs on FGFs overlap with FGFR-binding sites

FGFs mediate their bioactivities by binding to their cell surface receptors, FGFR and HS, to form a signalling complex. It seems likely that, in vivo, FGF binds first to HS and then to the FGFR, as HS-binding sites are several orders of magnitude more abundant in the pericellular matrix and on the cell surface [13]. There are different numbers of HBSs evident in the FGF subfamilies [17]. A question of interest is whether there is any overlap between these and the FGFR-binding sites. FGF-2 has three HBSs [26, 33-36], and the residues interacting with the FGFR have been identified by site-directed mutagenesis [37] and X-ray crystallography (FGF-2–FGFR1) [Protein Data Bank (PDB): 1CVS] [8]. A comparison of these binding sites indicates that HBS-3, towards the N-terminus, overlaps with the FGFR-binding site at Lys30. In addition to FGF-2, other FGFs also have residues involved in both interactions. According to a recent analysis using a protect and label approach (Fig. 2 in reference [17]) and the crystal structures of heparin–FGF-1–FGFR2 (PDB: 1E0O) [10], Lys24 in HBS-3 overlaps with its FGFR. Lys24 in FGF-1 is at the equivalent position to Lys30 in FGF-2 when the secondary structures of these proteins are aligned [17]. According to sequence alignment, there is a possible HBS-3 at the N-terminus of FGF-7, Arg65, but, as this only contains arginines, it was not identified by the protect and label approach [17]. The sequence alignment of the FGF-7 subfamily and the FGF-10–FGFR2b crystal structure (PDB: 1NUN) [38] shows that Arg78 of FGF-10 (allocated to FGF subfamily 7) is at the equivalent position to Arg65 in FGF-7, and so might be part of an HBS-3 [17], which is involved in binding to FGFR [38]. These secondary HBSs are likely to impact on the transport of FGFs in the extracellular matrix [13]. The overlap of one of these, HBS-3, with part of the FGFR-binding site may imply that a rearrangement of FGF–sugar interactions is required in some instances to allow the ternary signalling complex to form.

The interactions between FGFs and heparin/HS derivatives and analogues depend on both their dimensions and the nature of their charge distribution, and evidence obtained with a variety of experimental methods demonstrates that charge, size and conformation are important in these interactions [39, 40]. Importantly, different sequences can, if the dimensions, shape and charge distribution are correct, act in the same way as heparin to provide a suitable coreceptor for signalling [41]. A consequence may be the observed modulatory effect of chrondroitin sulfate on FGF signalling [42]. Interestingly, the mechanisms for FGF-1 and FGF-2 have been found to be different; FGF-1 requires only a saccharide capable of providing stabilization, whereas FGF-2 requires induction of the appropriate secondary structural changes before being able to support signaling [17, 41]. The SRCD spectra (Figs 2 and 4) indicate considerable conformational change on the addition of heparin, confirming that interactions occur, and reveal differences between FGFs. Importantly, nonlinear changes are evident as the ratio of heparin is increased (Fig. 3). This has been studied in some detail elsewhere, and indicates the presence of several binding sites of different affinity, which have also been mapped with a protect and label strategy [17].

Appreciation of the networked signalling system will influence the choice of pharmaceutical targets and the nature of attempts to target this system

Assembling the signalling network in the present way suggests that it can be approached as an entire experimental system. There are increasing indications that the combinations of activities supported by particular sugar structures mayconstitute one way of influencing this and similar networks in a more effective way than targeting single interactions [e.g. references [43-45]). This might be particularly relevant, because it is apparent that there is a good deal of redundancy in terms of FGF signalling and sugar coreceptor specificity and selectivity. Employing the network to search for agents capable of particular combinations of selective intervention is an attractive experimental and practical possibility, as is the prospect of supporting signalling patterns distinct from those present in the extant system.

The FGF–FGFR–HS signaling network presented here is only a small proportion of a much richer, more complex network with which HS interacts. There will also be comparable (but probably subtly different) networks of FGF/FGFR connections with other HS structures, and the variability of HS structures offers an additional dimension through which the network can be viewed or influenced. Presumably, HS structures with lower sulfation levels will tend to induce restricted networks, although this remains largely untested. It will also be interesting to compare HBSs and their affinities with those of different HS structures.

Highly complex signalling networks exist at the heart of cell signalling, and both the FGF–FGFR system and HS are key players, interacting with many hundreds of other components. Such systems can be prone to noise, as well as being sensitive to small fluctuations causing wider disruption. It is important, therefore, for the network to have built-in stability against such sensitivity. If the core, highly connected network is viewed as serving to transmit a particular range of signals, rather than any individual signals, it is closely analogous to a communication network, and can be termed an expander graph [24], a term coined in the field of telecommunications, and which describes the degree to which a network is interconnected and reflects its resulting degree of stability. Such expanders conserve the fidelity of patterns of information transfer against noise, in this context ensuring that the correct set of coordinated signals is transmitted. Consequently, the redundancy noted above is likely to reflect the need for such robustness.

Clearly, more refined data, in which the strength of signalling with other interacting proteins [14] is included, will be required for a fuller treatment. A more complete description of this signalling system will need to incorporate the affinity of each component and take into account multiple, and potentially overlapping, binding sites. Future refinements will entail extension of the network to incorporate other proteins that interact with FGFs and FGFRs, either directly [e.g. [46]) or indirectly through HS. Importantly, more extensive networks would provide data concerning which other receptors may be influenced, and the likely capacity of the system to provide compensatory signalling pathways. A probable outcome of this approach will be the realization that it will be necessary to inhibit (or stimulate) several events simultaneously, and the network does provide an experimental framework for screening compounds to this end. This would increase the discovery rate of such compounds, which have intriguing activities in animal models and patents, but for which there are just a few examples [43-45].

Experimental procedures

FGF expression and purification

FGF-1 (UniProt Accession P05230; residues 16–155) and FGF-2 (UniProt Accession P09038-2; residues 1–155) were cloned into vector pET-14b (Novagen; Merck Chemical, Nottingham, UK) and purified as previously described [47, 48]. cDNAs encoding FGF-7 (Uniprot Accession P21781; residues 32–194), FGF-9 (Uniprot Accession P31371; residues 1–208), FGF-18 (zFGF5) (UniProt Accession O76093; residues 28–207) and FGF-21 (Uniprot Accession Q9NSA1; residues 29–209) were inserted separately into a modified pET-24b vector (pETM-11; kind gift of P. Elliott, University of Liverpool), and purified as previously described [17, 47, 48].

SRCD spectroscopy

FGFs were buffer exchanged into CD buffer (15.3 mm Na2HPO4, 2.2 mm NaH2PO4, pH 7.5). Six different FGFs (FGF-1, FGF-2, FGF-7, FGF-9, FGF-18, and FGF-21) at a concentration of 0.5 mg·mL−1 (FGF-1, FGF-2, FGF-7, FGF-18, and FGF-21) or 1 mg·mL−1 (FGF-9) and heparin (average molecular mass, 17 kDa; Celsus Lab, Cincinnati, OH, USA) [disaccharide composition (%): ΔUA-GlcNAc; 5.8, ΔUA-GlcNAc; 8.4, ΔUA-N-sulfaminoglucosamine (GlcNS); 2.1, ΔUA-N-sulfamino, 6-O-sulfated glucosamine [GlcNS(6S)]; 22.9, ΔUA(2S)-GlcNS; 9.0, ΔUA(2S)-GlcNS(6S); 49.5, ΔUA(2S)-GlcNAc; 0.6, ΔUA(2S)-GlcNAc-6-O-sulfate, 1.7] [where ΔUA indicates a 4,5-unsaturated uronate residue, and ΔUA(2S) indicates a 4,5-unsaturated, 2-O-sulfated uronate residue] (data from [49]) were mixed in different ratios (1 : 5 to 5 : 1) and then loaded into a quartz cuvette (Hellma UK, Southend on Sea, UK), with a path length 0.2 mm, and the SRCD spectra were acquired from 178 nm to 260 nm on beam line B23, Diamond Light Source [50], as previously described [17]. Heparin binds FGF-21 (subfamily 19) only very weakly, and showed very little spectral change (Fig. 2F) on addition of heparin at any of the concentrations tested. This confirms both the lack of binding and that higher levels of heparin do not provide significant CD spectral features in this experimental set-up. Protein secondary structures were determined by use of the Dichroweb website, and the selcon3 algorithm with reference dataset 3 [51, 52].

Acknowledgements

The authors acknowledge the support of the Cancer and Polio Research Fund and the North West Cancer Research Fund. The authors also acknowledge a Biomedical Research Unit Award from the National Institute of Health Research.

Ancillary