Demystifying Heparan Sulfate–Protein Interactions

FGF1.

Many members of the FGF family interact with HS and form dimers (47, 48). The cocrystal structure of FGF1 and a heparin-derived decasaccharide [degree of polymerization (dp)10] revealed that protein–HS interactions drive dimerization in the absence of protein–protein interactions at the dimer interface (Figure 2a) (49). A recent isothermal calorimetry (ITC) study suggested positive cooperativity during FGF dimer formation, probably because binding of the first FGF1 monomer limits the conformational flexibility of the oligosaccharide and presents the preferred orientation of charged groups on the opposite side of the chain for binding to a second monomer (50). Importantly, the two FGF1 monomers bind HS asymmetrically due to the inherent asymmetry of the HS molecule (reducing and nonreducing ends). The same dimerization interface observed in the FGF1 dimer has been observed in two cocrystal structures of FGF1, FGF receptor 2 (FGFR2), and dodecasaccharides derived from heparin (51, 52). As discussed below, other crystal variants suggest that heparin oligosaccharides can also act as a template to oligomerize FGF with FGFR, independently of the formation of FGF dimers.

RAGE.

The receptor for advanced glycation end products (RAGE) becomes activated in many inflammatory conditions via ligands such as S100 proteins, high-mobility group protein B1, and other damage-associated molecular pattern proteins (DAMPs) released from activated cells or necrotic tissue. Oligomerization of RAGE is essential for signal transduction, which depends on engagement with ≥dp12 oligosaccharides. Binding to heparin or HS precedes ligand binding and presumably assembles RAGE into a functionally ready state for signaling prior to engagement of its cognate protein ligands. ITC experiments demonstrated a 2:1 stoichiometry for RAGE V-C1 subunits to oligosaccharide. The crystal structure of the RAGE hexamer obtained in the presence of heparin-derived dodecasaccharide demonstrates a trimer of dimers in which each dimer was stabilized by one dodecasaccharide bound at docking sites that were defined by mutagenesis (Figure 2b) (53). In addition, small-angle X-ray scattering indicates that the hexamer structure is retained in solution. The RAGE dimer has a small hydrophobic dimerization interface next to the HS-binding site with a buried surface area of ~300 Ų. In spite of its small size (too small to support a stable dimer by itself under normal conditions), the hydrophobic interface is essential for the formation of stable dimers. The hydrophobic interactions at the protein–protein interface and the electrostatic interactions mediated by HS are energetically coupled to promote RAGE dimerization. Importantly, recent data demonstrate that RAGE oligomerization induced by HS is required for signaling (53, 54).

Amyloid precursor protein.

Amyloid precursor protein (APP) is the precursor for amyloid-β, which accumulates in Alzheimer disease (55). APP and its homologs amyloid precursor–like proteins 1 and 2 (APLP-1 and APLP-2) are single-transmembrane proteins that play important (but not fully understood) functions in synapse organization (56–58). All APP family proteins bind HS, which is essential for APP dimerization at the cell surface (59, 60). APP is unusual in that it possesses two HS-binding sites in two different domains, E1 and E2 (61). Apparently, both domains can bind HS independently with micromolar association constants (62, 63). Quite remarkably, both E1 and E2 undergo HS-dependent dimerization with a 2:1 stoichiometry of protein:oligosaccharide (64, 65). The interaction between HS and the higher-affinity HS-binding domain E2 was characterized by cocrystallization (65). Similar to the RAGE dimer, the HS-binding site is located next to the dimerization interface (Figure 2c). Although the hydrophobic dimerization interface of the E2 domain (1,400 Ų) is much larger than that of RAGE, the E2 domain remains predominantly monomeric in solution (K_d for monomer–dimer equilibrium = 161 μM) (62). Clustering of positively charged residues near the dimerization interface presumably generates electrostatic repulsion and thus destabilizes the dimer. These positive charges, many of which are part of the HS-binding site, are effectively neutralized by bound HS, which consequently shifts the equilibrium toward dimer formation (K_d = 7.8 μM in the presence of heparin) (62).

Chemokines.

With very few exceptions, all chemokines have a strong tendency to oligomerize, and oligomerization plays an essential role in chemokine biology (44, 66, 67). Most chemokines exist as either monomers or dimers at physiological concentrations, but their interaction with HS drives the equilibrium toward the formation of dimers or higher-order oligomers (67, 68). The overall folds of most chemokines are very similar (a long N-terminal loop followed by a 3¹⁰ helix, a β-sheet formed by three β-strands ending with a C-terminal α-helix). However, the mechanism of oligomerization and HS binding varies for different chemokines (69). Many chemokines in the CXCL family have a dimer interface formed by the first of the three β-strands. However, the HS-binding sites can differ significantly within this family. The HS-binding sites of interleukin (IL)-8 and platelet factor 4 (PF4, or CXCL4) are located on opposite sides of the dimerization interface (Figure 2d) (70, 71), whereas the HS-binding site of stromal cell–derived factor 1 (SDF-1, or CXCL12) is in close proximity to and just above the dimer interface (Figure 2e) (72). In contrast, most CCL chemokines use a different dimerization interface that involves interaction of largely unstructured N-terminal loops (73). The HS-binding site of this family is also adjacent to the dimerization interface. Interestingly, when a CXCL chemokine (such as PF4) and a CCL chemokine, monocyte chemoattractant protein 1 (MCP-1, or CCL2), form tetramers, both dimerization interfaces are used (71, 73). In these cases, the HS-binding sites from all four subunits form a ring of positive charges that surrounds the whole tetramer. The physiological significance of chemokine oligomerization is not yet fully understood. The most likely role of chemokine oligomerization involves better retention at the luminal endothelial surface because of the higher affinity of oligomers for HS, better stability and longer half-life because of protection of oligomers from proteolytic digestion, and an added layer of regulation based on monomer–oligomer equilibrium (67).

Other HSBPs that depend on HS for oligomerization include FGF2 (74), thrombospondin (75), hepatocyte growth factor/scatter factor (HGF/SF) (76), the membrane proteins neuropilin-1 and -2 (77), and the receptor protein tyrosine phosphatase σ (RPTP-σ) (78). Many other as-yetuncharacterized HSBPs are also likely to undergo HS-induced oligomerization.

Antithrombin/thrombin.

Thrombin occupies a prominent position in the blood coagulation cascade because it directly converts fibrinogen into fibrin, the major component of blood clots (80). Several serine protease inhibitors termed serpins regulate the enzymatic activity of thrombin; AT is the best-known member of this family (81). In the absence of HS, AT is not an efficient inhibitor of thrombin; however, its inhibitory rate constant increases more than three orders of magnitude when HS is present (82). The underlying reason for this dramatic rate enhancement is HS’s ability to act as a scaffold to approximate AT and thrombin, both of which are HSBPs (83). By binding simultaneously to both enzyme and substrate, HS greatly increases the chance of successful engagement of thrombin and the flexible reactive central loop (RCL) of AT (84). Interestingly, investigators initially observed that oligosaccharides of ≥dp18 are required for maximal rate enhancement (85). However, only an appropriately sulfated pentasaccharide suffices for AT binding, and a sulfated hexasaccharide is adequate for thrombin binding (86, 87). Two cocrystal structures of the ternary complex of AT, thrombin, and a synthetic HS mimic (84, 88) provided the explanation for the requirement of an oligosaccharide of ≥dp18. The distal locations of the HS-binding sites of AT and thrombin require an arrangement of two NS domains with an intervening inert linker of dp8 (Figure 3a). This arrangement is reminiscent of many scaffold proteins that are organized such that distinct protein-binding domains are separated spatially by an extended disordered region (79).

Protein C inhibitor/thrombin.

Protein C inhibitor (PCI) is another serpin that acts on thrombin. Although the structure of PCI is similar to that of other serpins, including AT, the HS-binding site is located in a completely different region (89). The HS-binding site of PCI is situated much closer to the PCI/thrombin interface than that found in AT/thrombin complexes, enabling the formation of a composite HS-binding site composed of domains from both PCI and thrombin (Figure 3b) (90). The close proximity between the two HS-binding sites enables shorter but fully sulfated oligosaccharides (≥dp14) to act as a scaffold, in contrast to the dp18 oligosaccharide required for bridging AT and thrombin. This arrangement suggests that a single NS domain may suffice to facilitate the formation of a PCI/thrombin complex.

PCI actually has procoagulant activity by inhibiting activated protein C (APC), another serine protease involved in coagulation (81). PCI inhibition of APC also requires HS to form a ternary complex (91). Interestingly, the HS-binding site of APC is distinct from the one in thrombin and is located on the opposite side of the molecule (28), which poses a steric problem for binding to HS (Figure 3c). The HS-binding site of PCI has substantial plasticity, which allows HS to bind in a quite different orientation to bridge the PCI/APC complex, in contrast to the orientation observed in the PCI/thrombin complex (92). What emerges from these various studies is HS’s enormous versatility as a scaffold.

FGF/FGFR.

The interactions between FGFs and their receptors have been extensively studied and crystal structures of eight different FGF/FGFR complexes have been solved over the years (47). All FGFR isoforms bind HS with moderate to high affinity (93–95). FGF and FGFR can form a complex in solution, but these complexes tend to dissociate during size-exclusion chromatography (96, 97). However, if heparin is included in the mixture, the FGF/FGFR/heparin ternary complex is much more stable and can easily be purified by chromatography (97). Interestingly, two cocrystal structures revealed two different ways that HS interacts with the FGF/FGFR complex that vary in stoichiometry and orientation of the subunits; these data may reflect different purification methods and crystallization conditions (Figure 3d,e). Despite the differences, both crystal structures showed that the HS-binding site of FGFR, which is located in the D2 domain, merges with the HS-binding site of FGF to form a large composite docking site for HS (Figure 3d,e). Although the exact orientation differs, such composite HS-binding sites are observed in virtually all FGF/FGFR combinations (51, 52, 96, 98). The bridging function of HS, mediated by a relatively short, highly sulfated oligosaccharide sequence, resembles the arrangement of HS in the PCI/thrombin complex. This mode of scaffolding might be relevant to other ligand–receptor interactions, including those between Slit and Robo (99) and between VEGF and neuropilin (77).

Antithrombin.

All serpins have an RCL that acts as bait to achieve inhibition of target serine protease (81). AT is unique among serpins in that its RCL adopts a more constrained conformation, which makes it less accessible for certain proteases (Factors Xa and IXa) and is responsible for the poor inhibitory activity of native AT (102). However, upon binding of a unique pentasaccharide sequence containing a rare 3-O-sulfated glucosamine, the RCL constraint is removed through an allosteric mechanism that is now well understood (102, 103). The conformation change starts with the elongation of helix D at its C terminus upon pentasaccharide binding, which pushes the first two β-strands toward the center of the four-strand A-sheet; in turn, these β-strands squeeze out the N-terminal portion of the RCL to allow it to adopt a more flexible conformation (Figure 4a). This flexible conformation enables two critical electrostatic interactions between Factor Xa (as well as Factor IXa) and AT (86, 104). The HS-binding site of AT plays a pivotal role in triggering the conformational change. Comparison between the pentasaccharide-bound and native AT structures reveals the key structural elements of the HS-binding site that undergo significant rearrangement and allow the side chains of the HS-binding amino acid residues to move by 5 to 17 Åto accommodate the pentasaccharide (Figure 4b) (103).

VCP.

VCP is a viral HSBP that helps Vaccinia virus evade host defense by downregulating complement pathways (105). This protein is particularly interesting to immunologists because it bears high structural homology to two host proteins that regulate complement activation, Factor H (FH) and C4b-binding protein, both of which bind to HS (106, 107). FH and C4b-binding protein play essential roles in distinguishing host and pathogen by downregulating complement pathways at the host cell surface. A cocrystal structure of VCP and heparin decasaccharide revealed the unexpected finding that HS binding to the C-terminal short consensus repeat (SCR) domain causes a significant conformational change that involves domain movement (Figure 4c) (107). In its native conformation, the last two SCRs of VCP, SCR-3 and SCR-4, form an angle of ~120° (108). When HS binds to the tip of SCR-4, the angle between SCR-3 and SCR-4 declines to ~80°, bringing SCR-3 much closer to SCR-4. HS may selectively bind and stabilize this closed conformation (107). Because SCR-4 contains several residues that interact with complement C3b (107), the changes in its relative orientation to SCR-1 and SCR-2 (which move with SCR-3 as a rigid body) would alter C3b-binding affinity and kinetics. HS promotes binding of FH to C3b in solution, probably by stabilizing the closed conformation of SCR domains of FH (109). Thus, HS appears to induce positive cooperativity toward C3b binding.

Interestingly, the vast majority of HSBPs do not show any noticeable backbone conformational changes upon HS binding on the basis of available structures of the bound and unbound proteins. However, the lack of observable conformational changes in crystals does not mean they do not occur, because crystallization inevitably biases proteins toward conformations that are more conducive to crystal formation and therefore may obscure conformational changes that would occur in solution. Also, as stated above, a conformational change at the target-binding site is not required for HS to have an allosteric effect. Presumably, HS binding can have a significant effect on the dynamic fluctuations in the structure of HSBPs and thus affect their interaction with a second molecule at a distal site. The most direct way to test this hypothesis experimentally is to use ITC. In one study, Brown et al. (50) examined FGF1 binding to FGFR2 in the presence or absence of heparin-derived hexasaccharide. Although the binding of hexasaccharide did not cause any conformation change of FGF1, the binding affinity of the hexasaccharide/FGF1 complex to FGFR2 (0.3 μM) was ~14-fold higher than the affinity between FGF1 and FGFR2 (4.3 μM), suggesting that the hexasaccharide induced positive cooperativity.

Modification-specific interactions.

As mentioned in the Introduction, each disaccharide unit can be modified at five different positions, which could give rise to up to 48 different disaccharides; however, only ~20 are commonly observed in HS (10). With this restriction, the number of permutations resembles the number of primary amino acid sequences found in proteins (21 amino acids). This observation speaks to the enormous coding capacity of HS, which prompted many researchers to hypothesize that distinct combinations of disaccharides, namely HS sequences, represent a code and dictate the specificity of HS–HSBP interactions (43). Because the term code can imply a template-driven process similar to that involved in transcription/translation, we prefer not to use this terminology but rather focus on specificity and selectivity based on structure.

The well-characterized HS–AT interaction supports this hypothesis. The high-affinity interaction between AT and HS requires a rare 3-sulfo-N-sulfoglucosamine residue (GlcNS3S) in a very specific context. Binding studies suggested that the 3-O-sulfate group contributes ~60% of the binding free energy and that removal of this modification causes a 10⁵-fold reduction in binding affinity to AT (112). Mechanistic studies determined that the 3-O-sulfate group, along with its amino acid partner in AT (Lys114), occupies a pivotal position in organizing both ionic and nonionic interaction networks that are responsible for the AT–HS interaction (Figure 5a). The surrounding sugar residues also play important roles in promoting this interaction. For example, there must be a nonsulfated glucuronic acid at the nonreducing side of the GlcNS3S unit (Figure 5a). If this residue is replaced with IdoA or 2-sulfo-IdoA (IdoA2S), then the oligosaccharide shows greatly reduced binding to AT (118). Although only a few proteins are known to depend on 3-O-sulfation, there will probably be many others because of the large number of HS 3-O-sulfotransferases (10, 119). Whether these proteins exhibit high selectivity, similarly to AT, awaits their identification and characterization.

FGF2 is another well-characterized HSBP. Formation of FGF2/HS complexes depends critically on IdoA2S and N-sulfated glucosamine (GlcNS) (113, 120). The cocrystal structures of FGF2 and hexasaccharide show that the major interaction occurs via one disaccharide that makes contact with the so-called high-affinity HS-binding site of FGF2 (121). This disaccharide makes nine salt bridges and hydrogen bonds with FGF2; of these nine, six are contributed by the IdoA2S residue and three by the N-sulfate group, which explains the importance of these two modifications (Figure 5b). In contrast, the three 6-O-sulfate groups of the hexasaccharide do not contact FGF2. The crucial role of IdoA2S and GlcNS in FGF2/HS interaction is further confirmed by the cocrystal structure of FGF2/FGFR1/decasaccharide, in which 8 out of 10 hydrogen bonds between FGF2 and decasaccharide involve these two sugar residues (96). Interestingly, in this structure 6-O-sulfates contribute two hydrogen bonds with FGF2, suggesting that a certain degree of flexibility is allowed in the interaction. Although 6-O-sulfate groups may be dispensable for interacting with FGF2, they are indispensable for promoting FGF2/FGFR signaling due to their extended hydrogen bonding with FGFR (122, 123).

A survey of several heparin oligosaccharide/HSBP complexes shows that many sulfate groups in the oligosaccharides do not contact the protein; therefore, most probably do not contribute to the binding free energy. Apparently, many HSBPs tolerate these unbound negative charges, which would explain the successful use of highly sulfated heparin-derived oligosaccharides in many studies. However, the use of highly sulfated heparin can be problematic. One study, which tested the binding specificity of FGF7 by using heparin-derived octasaccharides, found that moderately sulfated octasaccharides with 7 to 8 sulfates showed higher affinity to FGF7 than did fully sulfated octasaccharides with 11 to 12 sulfates (124). Thus, fully or nearly fully sulfated sequences (such as the commonly used heparin-derived oligosaccharide) may create artifacts and potentially obscure a binding site in less-sulfated domains. In other words, the natural sequences presented in HS may be more selective than the fully sulfated sequence in heparin. Recent advances in chemical and chemoenzymatic synthesis of structural defined oligosaccharides should provide the reagents needed to decode the relevant binding sequences in HS (125).

Domain-specific interactions.

A distinct structural feature of HS is that it is arranged as alternating sulfated NS domains, rarely sulfated NAc domains, and mixed NAc/NS domains (10). Analyses by heparin lyase and nitrous digestion suggest that the length of NS and NAc domains varies considerably, with NS domains that range from dp6 to dp16, NAc domains as large as dp18, and NAc/NS domains that account for as much as one-quarter of the length of the chain (8, 9, 126). Investigators initially assumed that proteins interacted with one sulfated NS domain, which is the case for many HSBPs. However, later research showed that many proteins, mostly in the chemokine and cytokine families, bind to two adjacent NS domains simultaneously. A few well-characterized examples include IL-8, PF4, MIP-1α, and IFN-γ (114, 115, 127, 128).

Under physiological conditions, IL-8 binds to oligosaccharides of at least dp18. Detailed structural analyses of the bound HS suggest that the active oligosaccharide contains two dp4–dp6 NS domains at the ends and one intervening NAc domain (up to dp14) (115). Each NS domain binds to one subunit of the IL-8 dimer. PF4, which binds HS as a tetramer, requires a much larger HS fragment for interaction (dp42), with a domain structure resembling the bipartite structure of the IL-8-binding fragment. The dp42 fragment consists of one dp10–dp14 or two dp6–dp8 NS domains at each end and one dp14–16 NAc domain (127). The HS-binding sequence of the MIP-1α dimer is also large (~dp34) and is composed of two long NS domains (dp12–dp14) at both ends and a relatively small NAc domain (dp8) (128). This arrangement is almost the opposite of that observed for IL-8. Finally, the preferred binding sequence of the IFN-γ dimer is ~dp46 (114) and is organized as two terminal NS domains of typical size (dp6–dp8) and an unusually long intervening NAc domain with a length of dp32. The utility of such a long linker, which is commonly regarded as an inert structure that has no interaction with protein, remains obscure.

A requirement for bipartite binding domains in HS may make suitable sites for binding somewhat rare. A recent study demonstrated that changing the domain structure of HS by inactivation of Hs2st (which causes an increase in glucosamine N-sulfation and 6-O-sulfation by an unknown compensatory mechanism) increased by 10-fold the number of binding sites for IL-8 on endothelial HS and increased acute inflammatory responses in mice (129). This finding raises the possibility that regulation of N-sulfation could regulate the capacity of HS to present chemokines and cytokines. Some data suggest that inflammation alters the expression of different enzymes in the pathway, but whether these changes alter the binding of chemokines is unknown (130, 131).

Charge-based interactions.

Many HSBPs interact with HS in a more flexible manner. Thrombin has often been cited as a classical example of a nonspecific charge-based interaction partner of HS. Early binding studies revealed that the thrombin–HS interaction showed a strong dependence on salt concentration, showed a low dependence on nonionic interactions, and demonstrated an increase in apparent binding affinity with increasing length of oligosaccharide (27). These characteristics contrast with the behavior of AT–HS and FGF2–HS interactions, which show mild dependence on salt concentration, have large nonionic contributions to binding, and have a binding affinity that is relatively indifferent to the size of the oligosaccharide (29, 112). The cocrystal structures of thrombin and a heparin octasaccharide showed opposite orientations for the oligosaccharide. Although the amino acids involved in binding remain the same, any given residue can bind to different sulfate groups in the two orientations. For instance, residue Lys236, which contributes the most to the binding free energy, interacts with N-sulfate and carboxylate groups in one structure, whereas in another structure, it interacts with 2-O-sufate and a different carboxylate group. Thus, the arrangement of positively charged residues of the thrombin–HS–binding site enables it to tolerate different negative-charge patterns in HS, as long as the interaction satisfies the required five to six ionic interactions (87).

HGF/SF is a large growth factor that relies on HS for activation of its cognate receptor, the tyrosine kinase MET. The extent of HGF/SF–HS interaction strongly correlates with the overall sulfation of the chain, but the interaction does not depend on the specific location of the sulfate groups; selectively de-N-, de-2-O-, and de-6-O-sulfated heparins bind HGF/SF equally well (132). Remarkably, more extensively desulfated heparin lacking both N- and 2-O-sulfates or both 2-O- and 6-O-sulfates can associate with HGF/SF in gel mobility shift assays and promote HGF/SF signaling (132). Cocrystal structures of HGF/SF/decasaccharide complexes show that several nonidentical decasaccharide binding modes exist. Among the seven amino acid residues that form the HS-binding site, only three (Thr61, Lys63, and Arg73) make invariant contacts with HS in all binding modes, whereas interactions with the other four basic residues are more variable. The nonspecific nature of the HS-binding site of HGF/SF also explains its capacity to interact with DS with an affinity comparable to HS (132).

Note that this type of charge-based, nonspecific interaction does not imply low affinity binding or lack of biological impact. It should be considered simply as another way that some HSBPs can exploit the structure of HS. Low selectivity can also ensure binding across different cell types and HS presentations, which could prove beneficial for housekeeping proteins, DAMPs, and receptors for viral or microbial infection. Also, a general feature of this type of HSPG–HS interaction is that binding strength increases with increasing level of sulfation. This observation has prompted some researchers to hypothesize that the purpose of graded levels of sulfation of HS chains is to enable a so-called modulated functional response, which may well be one of the most fundamental ways in which HS regulates various biological systems (111).

How many lysines and arginines are usually involved?

Electrostatic interactions dominate the interaction of most HSBPs with heparin/HS. The vast majority of characterized HS-binding sites (20 out of 22) contain four to seven lysine or arginine residues. Interestingly, the number of basic residues does not correlate with the affinity of HSBPs for heparin, which also includes the contribution of hydrogen-bonding, van der Waal, and hydrophobic interactions. For example, the HS-binding sites of FGFR1, thrombin, RAGE, and RPTP-σ contain seven basic residues, yet their affinities for heparin range from 0.3 nM to 7 μM (54, 78, 87, 96). In contrast, HS-binding sites containing only four basic residues can have high binding affinity, such as FGF2 and VEGF (121, 133). Note that many HSBPs bind HS as oligomers (more than 50%), which increases valency and the number of basic residues involved in binding. For example, each monomer of PF4 contains six basic residues that interact with HS; thus, a PF4 tetramer docks with HS by way of 24 basic residues (71).

Annexins, which play a role in the fibrinolytic pathway, represent a notable exception to the general rule of four to seven basic residues per HS-binding site. These proteins bind HS in a calcium-dependent manner. The cocrystal structures of annexin A2 and heparin-derived oligosaccharides (up to an octasaccharide) suggest that only two basic residues participate in HS binding (134). Two calcium ions occupy two negatively charged pockets in close proximity to the two basic residues and interact either directly or indirectly with the sulfate and carboxylate groups of the oligosaccharide. Annexin A5, another member of the annexin family, similarly interacts with HS (135). Other examples of calcium-dependent HSBPs include L-selectin, P-selectin, and type V collagen (136–138). However, a direct interaction between calcium and HS has not yet been demonstrated in complexes of these proteins with heparin.

Other polar residues form hydrogen bonds.

Amino acids other than lysine and arginine often contribute to and, in some cases, may be critical for binding HS. Histidine is not commonly found in HS-binding sites (Table 2) but has been demonstrated by mutagenesis in the HS-binding sites in MCP-1 and APLP-1 (59, 139) and in the crystal structures of thrombin and annexin A2 (87, 134). The HS-binding site of APLP-1 is exceptional because it contains five histidines (59). All five residues create either direct or water-mediated hydrogen bonds with sulfate groups (65). Consistent with the crystallization data, mutation of each of the five histidines significantly reduced the salt concentration required for elution from heparin/Sepharose beads, suggesting that the histidines substantially contribute to the binding energy (59).

Polar residues, especially asparagine and glutamine, often make hydrogen bonds with sulfate groups—for example, in FGF1 and FGF2. In both cases, three polar residues participate in hydrogen bonding with sulfate groups (49, 121). A thermodynamic study of HS–FGF2 interaction demonstrated that ionic interactions contribute only 30% of the binding free energy and that hydrogen bonding contributes much of the remaining binding energy (29). The contribution of asparagine and glutamine residues to binding is probably underestimated in most HS-binding sites because these residues are rarely targeted for study. One way to identify potential polar residues in HS-binding sites would be to first define the boundary of HS-binding sites by mutation of lysine and arginine residues, and then to target conserved polar residues within this region by mutagenesis.

What is the size of a heparan sulfate–binding site?

At first glance, it may appear that the size of HS-binding sites varies considerably, given that the buried surface areas range from 200 to 1,050 Ų (Table 2). However, in many cases, these areas reflect the sum of several individual HS-binding sites (referred to as HS-binding units below) that are present in the oligomeric form of the HSBPs. The vast majority (19 out of 22) of monomeric HS-binding sites have a surface area between 200 to 360 Ų. In some proteins, such as FGF1, HGF, and AT, only one HS-binding unit provides high-affinity binding to HS (49, 103, 132). In contrast, other proteins, especially chemokines and cytokines, use more than one HS-binding unit to assemble a composite HS-binding site. These HSBPs usually interact with HS in their oligomeric form (69).

The size of the HS-binding site does not serve as a good indicator of the binding affinity. High-affinity interactions can be supported by either small HS-binding sites (e.g., FGF2, HGF, RPTP-σ, annexin A2) (76, 78, 121, 134) or large HS-binding sites (e.g., PF4, RAGE) (53, 140). Conversely, some HSBPs exhibiting low affinity for heparin or HS have either small (e.g., thrombin, IL-8) or large (e.g., FGFR1, MCP-1) HS-binding sites (73, 87, 96). It follows that the density of basic residues in the HS-binding site does not necessarily predict affinity as well. For instance, FGFR1 contains seven basic residues in an area of 350 Ų (per monomer) but has a K_d of only 3 μM (96), whereas RAGE has seven basic residues in an area of 525 Ų but has a K_d of ~3 nM) (54). The HS-binding sites of thrombin and AT have surface areas of ~300 Ų. Thrombin has a higher density of basic residues, but the affinity of AT for HS is three orders of magnitude greater than the affinity of thrombin for HS (86, 87).

Interestingly, most HS-binding sites have a rectangular or half-cylindrical shape. The short axis is usually between 10 and 16 Å. This axis accommodates HS well because it has a rod-like helical shape with a maximal diameter of 12 Å(Figure 1). Lysine and arginine residues spaced along the binding site typically make salt bridges and/or hydrogen bonds with sulfate groups located on both faces of the heparin helix. In most cases, the long axes of HS-binding sites accord reasonably well with the minimal length of oligosaccharides required for interaction. For example, dp4–dp6 (20–30 Åin length) is the minimal length required for interaction with FGF1, FGF2, and HGF/SF (49, 121, 132). Accordingly, the long axes of the HS-binding sites in these proteins are ~25 Å. The HS-binding site in RAGE has an axis length of 55 Å, which accommodates a dp12 oligosaccharide (54–60 Åin length), the minimal length required for inducing dimerization (53).

Secondary structural elements of heparan sulfate–binding sites.

Generally, HS-binding sites consist of residues contributed by two to four spatially separated structural elements (21 out of 22 proteins listed in Table 2). Loops appear to be the most common secondary structural element, which may reflect their flexibility. In fact, only one HS-binding site (APLP-1) is devoid of a loop (65). α-Helices and β-strands occur less frequently than loops, but in some HS-binding sites they can be the dominant structural elements (e.g., APLP-1, thrombospondin) (65, 75). Specificity for subsequences in HS does not seem to depend on particular secondary structural elements. HSBPs with high selectivity, including FGF2 (binding requires Ido2S), cyclophilin B, and AT (binding requires GlcNS3S), have HS-binding sites that are composed primarily of loops (86, 121, 141).

What is the best way to identify a heparan sulfate–binding site?

Several groups have attempted to find a formula to help identify HS-binding sites in HSBPs (142–145). However, these algorithms suffer from sampling errors introduced by considering only a limited number of HSBPs, from the inclusion of hypothetical HS-binding sites that were not tested experimentally, and from the assumption that a linear sequence of amino acids was responsible for HS binding. An analysis of the experimentally determined HS-binding sites listed in Table 2 makes clear that a single linear sequence does not define a HS-binding site and that two to three primary and secondary elements usually make up the binding site. Thus, topology rather than linear sequence defines most HS-binding sites. Therefore, we suggest that it is best to calculate the surface electrostatic potential by using popular software such as Adaptive Poisson–Boltzmann Solver. In the absence of a three-dimensional structure, homology modeling can be performed with structural threading software.