Structural basis of chemokine interactions with heparan sulfate, chondroitin sulfate, and dermatan sulfate

Chemokines play diverse roles in human pathophysiology, ranging from trafficking leukocytes and immunosurveillance to the regulation of metabolism and neural function. Chemokine function is intimately coupled to binding tissue glycosaminoglycans (GAGs), heparan sulfate (HS), chondroitin sulfate (CS), and dermatan sulfate (DS). Currently, very little is known about how the structural features and sequences of a given chemokine, the structure and sulfation pattern of a given GAG, and structural differences among GAGs and among chemokines impact binding interactions. In this study, we used solution NMR spectroscopy to characterize the binding interactions of two related neutrophil-activating chemokines, CXCL1 and CXCL5, with HS, CS, and DS. For both chemokines, the dimer bound all three GAGs with higher affinity than did the monomer, and affinities of the chemokines for CS and DS were lower than for HS. NMR-based structural models reveal diverse binding geometries and show that the binding surfaces for each of the three GAGs were different between the two chemokines. However, a given chemokine had similar binding interactions with CS and DS that were different from HS. Considering the fact that CXCL1 and CXCL5 activate the same CXCR2 receptor, we conclude that GAG interactions play a role in determining the nature of chemokine gradients, levels of free chemokine available for receptor activation, how chemokines bind their receptors, and that differences in these interactions determine chemokine-specific function.

Humans express ϳ50 chemokines that show a remarkable repertoire of functions from combating microbial infection to homeostasis and developmental biology. Fundamental to these diverse functions is the migration of immune and nonimmune cells to proximal and remote locations. Most, if not all, chemokines share the following properties: they exist as monomers and dimers, activate receptors that belong to the GPCR class, and interact with tissue glycosaminoglycans (GAGs) (1-3). 2 Biophysical, cellular, and animal model studies have shown that GAG interactions orchestrate leukocyte trafficking by regulating chemokine gradients, availability, how GAG-bound chemokines are presented to their receptors, and endothelial activation (4 -12).
GAGs heparan sulfate (HS), chondroitin sulfate (CS), and dermatan sulfate (DS) are linear sulfated polysaccharides, which are covalently attached to core proteins and are the glycan part of a family of proteins called proteoglycans (PGs). Proteoglycan structures are quite diverse, are expressed by many cell types, and are also an integral component of the connective tissue and extracellular matrix (12)(13)(14)(15). For instance, syndecans, a major class of membrane-spanning PGs, are of varying molecular weight, are expressed in different cell types, and carry varying number of HS and CS chains (15). Whereas syndecan-1 (Sdc-1) and Sdc-3 contain HS and CS, Sdc-2 and Sdc-4 contain only HS. Sdc-1 and Sdc-4 are expressed on the endothelium and epithelium, and both syndecans have been implicated in chemokine-mediated neutrophil recruitment and are essential for successful resolution of inflammation (16 -18). HS chains are located close to the N terminus of the ectodomains and so more accessible for interacting with the receptors on the blood leukocytes, whereas CS chains are closer to the membrane surface and less accessible. Extracellular matrix (ECM) PGs such as perlecan, versican, and decorin contain varying levels of HS, CS, or DS and exist as noncovalent macromolecular complexes with ECM proteins such as collagen and laminin. GAGs and ectodomain of PGs also exist in the free form because they are cleaved by sheddases during neutrophil recruitment. These observations collectively suggest that the chemokine functional response depends on GAG structural features, their location within a PG, and the nature and location of PG in the tissue.
The basic building blocks of HS consist of repeating disaccharide units of D-glucuronic acid (GlcA) and GlcNAc (Fig. 1). HS has a modular structure with sulfated sequences separated by nonsulfated regions and is also more diverse because of differential N-sulfation and O-sulfation. In addition to N-sulfation, glucosamine can have 6-O-sulfation, and GlcA can have 2-O-sulfation and epimerize at C5 to L-iduronic acid (IdoA).

cro ARTICLE
The basic building blocks of CS and DS consist of D-GlcA and GalNAc, and on average, the disaccharide unit has one sulfate with O-sulfation either at C4 or C6 of GalNAc (Fig. 1). DS can be considered a variant of CS with the difference arising from C5 epimerization of GlcA to IdoA.
Chemokines are classified into two major (CXC and CC) families. A subset of seven CXC chemokines, including CXCL1 and CXCL5, characterized by the N-terminal ELR residues, function as agonists for the CXCR2 receptor (19). These chemokines recruit neutrophils across the endothelium and ECM to the infected and injury site and have also been implicated in nonimmune physiological functions (20 -22). CXCL1 and CXCL5 are differentially expressed in different tissues and clinical symptoms, indicating nonredundant roles that are highly context-dependent (23)(24)(25). CXCL1 and CXCL5 share 51% sequence similarity, possess the same structural fold, and reversibly exist as monomers and dimers ( Fig. 1) (26 -29). Very little is known regarding how structural features and sequence of a chemokine, structure and sulfation pattern of a GAG, differences between GAGs, and differences between chemokines impact binding interactions and function.
In this study, we characterized the binding of CXCL1 and CXCL5 to HS, CS, and DS using solution NMR spectroscopy. We observe pronounced differences in binding geometries that could be attributed to differences in participation of a few select conserved and chemokine-specific residues and differences in GAG structures. Shared properties include higher affinity of the dimer for all three GAGs and higher affinity for HS compared with CS and DS. Considering that CXCL1 and CXCL5 activate the same CXCR2 receptor, we propose that diversity of GAG interactions play a role in determining the nature of chemokine gradients, levels of free chemokine available for receptor activation, how chemokines bind their receptors, and thereby impact function in a chemokine-specific manner.

Results
NMR chemical shifts are exquisitely sensitive to local structural changes, and so the binding surfaces were defined from chemical shift perturbation (CSP) on titrating GAGs to 15 Nlabeled proteins. Considering that CXCL1 and CXCL5 exist as monomers and dimers (monomer-dimer equilibrium constant, 1-4 M) (27,29), we initially characterized binding at low concentrations where both monomers and dimers are present. During GAG titration, four species will exist in solution: both monomer and dimer in the free and GAG-bound forms. Because of coupled equilibria, peak intensity changes reflect relative populations and binding constants.
For both chemokines and all three GAGs, the peaks corresponding to the monomer disappear on GAG binding (Fig. 2), indicating that the dimer is the high-affinity ligand. Therefore, all subsequent studies were carried out at 100 -120 M, where CXCL1 and CXCL5 exist as dimers. The chemical shifts in the

Chemokine-GAG interactions
GAG-bound form were assigned by collecting a series of spectra on adding small increments of GAG until essentially there were no changes in chemical shifts. We observe only one set of peaks during the titration, indicating that the free and boundforms are in the fast exchange regime in the NMR time scale.
We had previously shown that the length of a 14-mer (dp14) is more than sufficient to cover all of the chemokine-binding interface from heparin-binding studies (29 -31). For our cur-rent studies, we used CS and DS 14-mers. HS has a modular structure with sulfated domain separated by nonsulfated regions, and so we used a HS 30-mer to capture the modular HS structure. We also characterized binding to a heparin 26-mer and observed that the CSP profiles of backbone amides on binding the 14-mer and 26-mer were no different. The CSP profiles on binding HS and heparin were essentially identical, indicating that binding interactions of HS sulfated domain is similar to The monomer peaks disappear on GAG binding, indicating that the dimer is the high-affinity ligand. The spectra were collected using a ϳ10 M chemokine sample in 50 mM sodium phosphate, pH 6.0.

Chemokine-GAG interactions
heparin. For both chemokines, the CSP profiles indicate that basic residues located in the N-loop, 40s loop, and C-terminal helix mediate binding. However, the profiles vary between the chemokines, and also profiles on CS and DS binding were quite different from HS for a given chemokine.

CXCL1-HS interactions
HS binding resulted in perturbation of basic residues Arg-8, His-19, Lys-21, Lys-45, Arg-48, Lys-49, Lys-60, Lys-61, and Lys-65. The perturbation profile on HS binding is very similar to what was previously observed for heparin (Figs. 3 and 4). Lys-29 showed negligible perturbation but is implicated in binding from mutagenesis, functional, and NMR studies tailored to detect lysine side chain chemical shifts (9,31). Mapping these residues reveals that His-19, Lys-21, Lys-45, Lys-60, Lys-61, and Lys-65 constitute one binding domain (defined as the ␣-domain), and Arg-8, Lys-29, Arg-48, and Lys-49 constitute a second binding domain (defined as the ␤-domain) (Fig. 5). The CXCL1 structure reveals that ␣and ␤-domains are located on opposite faces of the protein and that a single GAG chain cannot simultaneously bind both domains. Isothermal titration calorimetry studies also show a stoichiometry of two heparins binding one CXCL1 dimer (Fig. S1). Some neighboring and buried residues also show significant CSP, but these are unlikely to be involved in direct binding and are likely due to indirect binding-induced structural changes.

CXCL1-CS and CXCL1-DS interactions
The CSP profiles for CS binding were dramatically different than that for HS. CS binding resulted in selective perturbation of the basic residues Arg-8, His-19, Lys-21, and Arg-48. Lack of perturbation of the helical residues Lys-60, Lys-61, and Lys-65 is striking (Figs. 3 and 4). Of the perturbed residues, His-19 and Lys-21 are from the ␣-domain, and Arg-8 and Arg-48 are from the ␤-domain. We define this new interface as the ␥-domain.
Whereas both ␣and ␤-domains span the dimer interface, the ␥-domain exists within a monomer (Fig. 5). The CSP profile for DS was similar but for some differences (Figs. 3 and 4). Com-

Chemokine-GAG interactions
pared with CS, perturbation of Arg-48 was less profound, and in addition, the CSP profile indicates that Lys-45 and Lys-49 could be involved in binding. Mapping these residues also suggests that DS binds the ␥-domain. For both CS and DS, like in HS, some neighboring and buried residues show significant CSP that can be attributed to indirect interactions. Dissociation constants were determined by fitting the binding-induced chemical shift changes as described previously (32). Affinities of CXCL1 for CS and DS were weaker than that for HS (K d 50 Ϯ 10 versus 4 Ϯ 2 M). We show sections of the spectra that highlight GAG binding-induced chemical shift changes and also K d plots for a select set of peaks (Fig. S2).

CXCL5-HS interactions
HS binding resulted in perturbation of the basic residues His-23, Lys-25, Lys-49, Lys-52, Lys-64, Lys-65, Lys-69, and Lys-76. The perturbation profile on HS binding is very similar to that for heparin (Figs. 6 and 7). Modeling and NMR studies that detect lysine side chain chemical shifts have shown that Lys-76 is not involved in binding, indicating that its chemical shift change must be due to indirect structural changes (30,31). The remaining residues form a contiguous surface within one monomer of the dimer (Fig. 8A). Chemical shift changes were also observed for a few buried residues that can be attributed to indirect interactions.
Sequences reveal that the same residues that mediate HS binding in CXCL5 also mediate binding in CXCL1 (Fig. 1). However, in CXCL1, additional interactions are observed for Arg-8, Lys-29, and Lys-49. Arg-8 is absolutely conserved among all seven CXCR2-activating chemokines and is essential for receptor activation. Lys-29 is unique to CXCL1, and Lys-49 is quasiconserved and observed only in four of the seven chemokines (3). In CXCL5, the residues corresponding to Lys-49 and Lys-29 are Glu and Phe, respectively, indicating that the strate-

Chemokine-GAG interactions
gic location and participation of a few conserved and chemokine-specific basic residues determine very different HS binding geometries.

CXCL5-CS and CXCL5-DS interactions
Perturbation profiles for CS and DS are similar but very different compared with HS ( Figs. 6 and 7). Chemical shift changes of basic residues were observed only for His-23, Lys-25, Lys-49, and Lys-52. Similar to CXCL1, C-terminal helical basic residues are not perturbed. However, perturbation of several N-loop and 30s loop nonbasic residues that are not in the vicinity of a basic residue is striking. Mapping the perturbed residues suggests a binding interface that spans the dimer that is very different than in CXCL1 (Fig. 8B). Binding affinities of CXCL5 for CS and DS are also weaker compared with that for HS (K d ϭ 30 Ϯ 10 versus 3 Ϯ 2 M). For all three GAGs, we performed three different HAD-DOCK runs to ensure that the input constraints did not bias specific structural models and that all possible binding geometries within a monomer and across the dimer interface were considered. We modeled binding of one GAG chain with constraints to only one monomer of the dimer, binding of two GAG chains with constraints to both monomers of the dimer, and binding of one GAG chain with constraints to both monomers of the dimer. For all three GAGs, all runs resulted in only one major cluster with the GAG binding to a monomer of a dimer (Fig. 9).

Structural models of GAG-bound CXCL1/CXCL5 complexes
Structural models reveal that all three GAGs adopt the same geometry. Ionic interactions were mediated by basic residues Arg-8, His-19, Lys-21, Lys-45, Arg-48, and Lys-49, and H-bonding and packing interactions by polar residues Gln-10, Asn-22, and Asn-46. For all three GAGs, CXCL1 engages two sulfates, two carboxylates, and two N-acetyl groups. Differences in GAG backbone structure and topology of the acidic groups did impact binding interactions. For instance, His-19 is involved in binding to CS6S and not CS4S or DS, Arg-8 engages a sulfate in CS6S and a carboxylate in CS4S and DS, and Lys-21 engages a sulfate in CS4S and DS and a carboxylate in CS6S.
NMR studies, which used naturally occurring CS that is a mixture of CS-4S and CS-6S, showed significant perturbation for Arg-8, His-19, Lys-21, and Arg-48. Lack of chemical shift changes for Lys-45 and Lys-49 can be attributed to chemical shift changes from direct and indirect interactions that are of

Chemokine-GAG interactions
opposite sign and similar magnitude and cancel out. A similar observation has been made for one of the lysines in heparin binding to CXCL1 and CXCL8 (29,33).

CS/DS-bound CXCL5 complexes
The rationale and the docking strategy were the same as described for CXCL1 complexes. We initially modeled binding of one GAG chain with constraints to both monomers of the dimer (run 1) and two GAG chains with constraints to both monomers of the dimer (run 2). For all three GAGs, both runs resulted in similar structures with one GAG binding across the dimer interface in run 1 and two GAG chains across the dimer interface on opposite faces of the protein in run 2 (Fig. 9). Structural models reveal that all three GAGs adopt the same geometry. Electrostatic interactions were mediated by basic residues His-23, Lys-25 Lys-49, and Lys-52 and H-bonding and packing interactions by nonbasic residues Gln-17, Thr-18, Gln-20, Met-26, Leu-48, Asn-50, Ile-35Ј, and Pro-37Ј (where Ј corresponds to residues from the second monomer). For all three GAGs, CXCL5 engages two sulfates, two carboxylates, one N-acetyl, and one hydroxyl group. However, differences in GAG structure result in some differences in residue-specific interactions. For instance, whereas Lys-49 binds a sulfate in CS-6S and a carboxylate in CS-4S and DS, Lys-25 and Lys-52 bind a sulfate in all three GAGs. NMR studies show significant CSP for all residues identified from modeling studies except for Pro-37, which is to be expected because prolines do not have a backbone amide proton. We also modeled binding of GAG chains with constraints given to only one monomer of the dimer (run 3). Run 3 resulted in two major clusters: one similar to that observed in run 1 despite the absence of explicit constraints to the second monomer, and a second cluster that involved binding only to the N-loop and 40s loop residues within a monomer. The latter model can be ruled out because it is inconsistent with the NMR data.

HS-bound CXCL1/CXCL5 complexes
NMR studies reveal that the CSP profiles of HS and heparin binding to both CXCL1 and CXCL5 are very similar. To mimic the diversity in HS structures, we modeled HS that is either missing N-sulfate, 2-O-sulfate, or 6-O-sulfate. For both CXCL1 and CXCL5, the binding geometries of all three variants were similar to heparin, suggesting that the binding interface is plastic, and the structural scaffold and distribution of sulfates and carboxylates dictate a similar binding geometry (Fig. 10).

Discussion
GAGs bind hundreds of proteins with diverse structural scaffolds and functions from growth factors and enzymes to chemokines and cytokines. Describing the molecular basis of GAG interactions has been challenging because structures of GAG-bound protein complexes have been hard to come by. Therefore, many studies have reported structural models of proteins bound to heparin on the basis of NMR backbone CSPbased methods. In recent years, lysine, arginine, and histidine side chain chemical shifts have also been shown to be quite sensitive and useful for identifying GAG-binding residues (31,34,35). Heparin has been and continues to be the GAG of choice for structural and biophysical studies because it is more uniformly sulfated and because of availability of size-defined oligosaccharides. Both HS and heparin share a repeating disaccharide unit composed of glucosamine and hexuronic acid, and it is assumed that heparin is a good surrogate for capturing HS interactions. Our studies for two related chemokines validate this premise by providing experimental evidence that the same basic residues mediate binding heparin and HS. We point out native HS polymer used in our studies is intrinsically heterogeneous because of differential N-and O-sulfation and that the NMR chemical shift changes are a sum of interactions to all of the different structures present in the polymer. To our knowledge, ours is the first NMR study that has characterized residue-specific binding interactions to heparin and HS polymers. Whether our observation holds for other HS-binding proteins needs to be determined. Very few studies have characterized the structural basis of CS and DS interactions or how the structural features of the same protein mediate binding different GAGs. We show that a given chemokine can have different binding surfaces for different GAGs, that related chemokines can have different binding surfaces for a given GAG, and that binding affinities can vary between GAGs, indicating that differences in chemokine sequences and GAG structure result in different binding interactions.

Chemokine-GAG interactions
Considering GAGs bind diverse protein classes (36 -39), chemokine sequence must play a major role in determining selectivity, affinity, and geometry for a given GAG and between GAGs. We observe for two related chemokines, a cluster of basic residues located in the N-loop, 40s loop, and C-terminal helix mediate HS binding via ionic interactions. Participation of additional residues from elsewhere in the sequence in CXCL1 results in a very different binding geometry. For both chemokines, binding interactions of CS and DS did not involve helical residues. This observation is quite striking, because helical residues in CXCR2-activating chemokines have long been assumed as important in GAG binding. NMR structural models for both chemokines suggest that N-and 2-O-sulfation in HS interact with helical lysines. Collectively, these data indicate that distribution of basic residues in the context of chemokine structure, differences in participation of a few select basic residues, and differences in GAG sulfation pattern can result in very different binding interactions. Knowledge of structures and molecular dynamics simulation studies are essential to fully appreciate how and which GAG sulfate/carboxylate groups mediate binding and how differences in GAG structures and chemokine sequence determine the large diversity of binding interactions. Previous NMR studies indicate that the binding interactions of CS, DS, and heparin hexasaccharide binding to

Chemokine-GAG interactions
CXCL8 are similar (40), which is different from our observations for CXCL1 and CXCL5.
Animal model studies have shown that the ability of chemokines to reversibly exist as monomers and dimers and GAG interactions regulate neutrophil recruitment (9,41,42). During active neutrophil trafficking, chemokine concentration can vary by orders of magnitude, and so it may exist as a monomer, dimer, or both at different times and locations. Our results indicate that both CXCL1 and CXCL5 dimers bind all GAGs with higher affinity. NMR studies for a number of chemokines have demonstrated that GAG binding and dimerization are coupled (43)(44)(45)(46)(47)(48)(49), suggesting that this is a shared property of all che-mokines for all GAGs. CXCL1 and CXCL5 are selectively expressed and/or under specific disease conditions and are also co-expressed in other conditions (22-25, 50 -52). Further, CXCL1 and CXCL5 KO mice show distinct differences in disease models, indicating that the unique phenotype of each chemokine is critical for successful resolution (23,24). CXCL1 and CXCL5 have also been implicated as major players in housekeeping functions including in developmental biology, suggesting that GAG interactions must play an important role in these highly spatiotemporally regulated processes (53,54). Additional studies similar to those reported here for other chemokines and other GAG-binding proteins with different

Chemokine-GAG interactions
structural scaffolds are essential to fully appreciate how differences in GAG interactions regulate and dictate functional response in health and disease.

Materials and methods
Recombinant 15 N-labeled CXCL1 and CXCL5 were expressed and purified as described previously (27). Heparin (dp26), HS (dp30), CS (dp14), and DS (dp14) were purchased from Iduron. According to the manufacturer, the oligosaccharides were purified using high-resolution gel filtration chromatography. The major disaccharide unit in heparin is IdoA, 2S-GlcNS,6S (ϳ75%), in chondroitin sulfate is GlcA-GalNAc sulfated at C6 or C4, and in dermatan sulfate is IdoA-GalNAc,4S (ϳ88%). The heparan sulfate polymer on an average has 1.75 sulfate groups (0.65 N-sulfate and 1.1 O-sulfate groups) per disaccharide. All GAGs contain uronic acid at the nonreducing end and a C4 -C5 double bond as a result of the endolytic action of heparinase or chondroitin ABC lyase.

Structural models of GAG-bound CXCL1 and CXCL5 complexes
Molecular docking of GAGs to CXCL5/CXCL1 dimer was carried out using high ambiguity driven biomolecular DOCKing (HADDOCK) using NMR CSP as structural constraints described previously (29,30,55,56). Docking simulations were carried out using CXCL1 and CXCL5 dimer NMR and various GAGs. HS, CS, and DS oligosaccharide structures generated using the GLYCAM web server (http://glycam.org/). 3 The topology and parameter files for all oligosaccharides were generated using the PRODRG server (57). After an initial rigid body docking, the top 1000 structures that had the best intermolecular energies were then sequentially subjected to semiflexible simulated annealing and explicit solvent refinement. The pairwise ligand interface root-mean-square-deviation matrix over all structures was calculated, and final structures were clustered using a cut-off value of 7.5 Å. The clusters were sorted using root-mean-square deviation and HADDOCK score (weighted sum of a combination of energy terms). The structures with best HADDOCK score were used for representation.