CXCL1/MGSA Is a Novel Glycosaminoglycan (GAG)-binding Chemokine

In humans, the chemokine CXCL1/MGSA (hCXCL1) plays fundamental and diverse roles in pathophysiology, from microbial killing to cancer progression, by orchestrating the directed migration of immune and non-immune cells. Cellular trafficking is highly regulated and requires concentration gradients that are achieved by interactions with sulfated glycosaminoglycans (GAGs). However, very little is known regarding the structural basis underlying hCXCL1-GAG interactions. We addressed this by characterizing the binding of GAG heparin oligosaccharides to hCXCL1 using NMR spectroscopy. Binding experiments under conditions at which hCXCL1 exists as monomers and dimers indicate that the dimer is the high-affinity GAG ligand. NMR experiments and modeling studies indicate that lysine and arginine residues mediate binding and that they are located in two non-overlapping domains. One domain, consisting of N-loop and C-helical residues (defined as α-domain) has also been identified previously as the GAG-binding domain for the related chemokine CXCL8/IL-8. The second domain, consisting of residues from the N terminus, 40s turn, and third β-strand (defined as β-domain) is novel. Eliminating β-domain binding by mutagenesis does not perturb α-domain binding, indicating two independent GAG-binding sites. It is known that N-loop and N-terminal residues mediate receptor activation, and we show that these residues are also involved in extensive GAG interactions. We also show that the GAG-bound hCXCL1 completely occlude receptor binding. We conclude that hCXCL1-GAG interactions provide stringent control over regulating chemokine levels and receptor accessibility and activation, and that chemotactic gradients mediate cellular trafficking to the target site.

Chemokines, a large family of small soluble proteins, are highly versatile and play fundamental roles in diverse functions, from combating infection and initiating tissue repair to regulating metabolism and organ development (1,2). Common to these various functions is the directed movement of various cell types to distal and remote locations. Cellular trafficking must be highly coordinated to elicit the required biological function, and its dysregulation could be detrimental, resulting in disease. There is now increasing evidence that the ability of a chemokine to reversibly exist as monomers and dimers and binding glycosaminoglycans is coupled and plays important roles in mediating these functions (3,4).
Glycosaminoglycans (GAGs) 2 such as heparan sulfate (HS) are highly sulfated polysaccharides. They are expressed ubiquitously by many cell types, are anchored to the cell surface by covalent attachment to membrane proteins, and form non-covalent complexes with proteins in the extracellular matrix (5)(6)(7). Animal models and cellular studies have established that GAG interactions dictate chemokine concentration gradients and that these gradients orchestrate cellular trafficking (8 -10). GAGs are acidic, and chemokines are basic or contain clusters of basic residues, indicating that electrostatic and H-bonding interactions play a prominent role in mediating the binding process.
On the basis of conserved cysteines, chemokines are broadly classified into CXC and CC families. They can be divided further into subclasses. For example, a set of seven human CXC chemokines characterized by the N-terminal "ELR" motif are agonists for the CXCR2 receptor and also share the properties of monomer-dimer equilibrium and GAG binding (11)(12)(13). The chemokine melanoma growth stimulatory activity (MGSA), an ELR chemokine, is also known as Gro␣, by the systematic nomenclature as CXCL1, and in mice as KC. MGSA was one of the earliest chemokines identified and, as the name implies, was observed in melanoma-related activities (14,15). Subsequently, ELR chemokines have been shown to play an important role in recruiting neutrophils during microbial infections and injury (16 -18).
Solution structures of human CXCL1 (hCXCL1) and its dimerization properties are known (Fig. 1), and we have shown recently that both monomers and dimers are potent agonists for the CXCR2 receptor (19 -21). However, knowledge of the structural basis or the molecular mechanisms underlying GAG interactions has been lacking. Comparison of the ELR chemokine sequences reveals conserved lysines and arginines but also sequence-specific differences. In this study, using NMR spec-troscopy, we show that the dimer is the high-affinity GAG ligand and, most interestingly, that GAG binding residues are located in two non-overlapping domains. Although one of the domains has been observed for the related chemokine CXCL8/ IL-8, the presence of a second domain is novel. Residues from both GAG-binding domains are also involved in receptor interactions, indicating that GAG-bound hCXCL1 cannot activate the receptor. We propose that two independent GAG binding domains impart better control over fine-tuning chemokine concentration gradients and receptor activation for orchestrated cellular trafficking to the target site.

Materials and Methods
Recombinant hCXCL1 was expressed and purified as described previously (22). For NMR experiments, 15 N-labeled hCXCL1 was produced essentially in the same fashion, but the cells were grown in minimal medium containing [ 15 N]ammonium chloride. The heparin oligosaccharides dp8 and dp14 were purchased from Iduron.
NMR Titration Experiments-Titrations of heparin oligosaccharides to 15 N-labeled hCXCL1 WT and R8A mutant and of the CXCR2 N-domain peptide to 15 N-labeled hCXCL1 WT were carried out in 50 mM sodium phosphate (pH 5.7) containing 1 mM 2,2-dimethyl-2-silapentanesulfonic acid, 1 mM sodium azide, and 10% D 2 O (v/v). NMR spectra were acquired at 40°C on a Bruker Avance III 800 MHz (equipped with a TXI cryoprobe) or 600 MHz (with a QCI probe) spectrometers. The chemical shifts of the WT hCXCL1 dimer are available at pH 5.5 and 30°C. The assignments at pH 5.7 and 40°C were similar and confirmed using 15 N-NOESY and 15 N-TOCSY experiments. Aliquots of heparin oligosaccharides (ϳ8 mM) prepared in the same buffer were added to ϳ150 M hCXCL1, and a series of 1 H, 15 N HSQC spectra was collected. The final hCXCL1:GAG molar ratio was 1:4. In the case of receptor titrations, aliquots of the CXCR2 N-domain (1 mM) were added to ϳ100 M hCXCL1, and a series of 1 H, 15 N HSQC spectra was collected. The final hCXCL1:CXCR2 molar ratio was 1:3. In the case of CXCR2 N-domain titration to the heparin-bound hCXCL1, aliquots of the CXCR2 N-domain (1 mM) were added to dp14-bound hCXCL1, and a series of 1 H, 15 N HSQC spectra was collected. The final hCXCL1:GAG:CXCR2 molar ratio was 1:4:6. The chemical shift perturbation (⌬␦ obs ) was calculated as a weighted average of 1 H (⌬␦ H ) and 15 N(⌬␦ N ) chemical shift changes (⌬␦ obs ϭ [(⌬␦ H ) 2 ϩ (⌬␦ N /5) 2 ] 1 ⁄2). To determine relative GAG affinities of the WT hCXCL1 monomer and dimer, dp14 was titrated to ϳ15 M hCXCL1. At this concentration, both monomer (ϳ8%) and dimer (ϳ92%) peaks were observed. 1 H, 15 N NOE Experiment-Steady-state 15 N heteronuclear NOEs were measured using a gradient-selected, sensitivity-enhanced pulse sequence (23). The heteronuclear NOE values were calculated as a ratio of peak intensities with and without proton saturation.
Docking of hCXCL1-Heparin Complexes-Molecular docking of heparin oligosaccharides to hCXCL1 WT and the R8A dimer was carried out using high ambiguity driven biomolecular docking (HADDOCK) (24,25), as described previously for the CXCL8-heparin complexes (26). We used the hCXCL1 dimer (PDB code 1MGS) and dp8-and dp14-mer structures generated from a heparin 12-mer (PDB code 1HPN) as the starting structures (19,27). NMR chemical shift perturbations (CSPs) were used as ambiguous interaction restraints to drive the docking process. The topology and parameter files for heparin oligosaccharides were generated using the PRODRG server (28). In total, 3000 complex structures were generated during the initial rigid body docking. The top 1000 structures that had the best intermolecular energies were then subjected sequentially to semiflexible simulated annealing and explicit solvent refinement during which the oligosaccharide and protein interface residues were allowed to have flexibility. The pair-wise "ligand interface root-mean-square deviation matrix" over all structures was calculated, and final structures were clustered using a cutoff value of 7.5 Å. The clusters were sorted using root-mean-square deviation and "HADDOCK score" (the weighted sum of a combination of energy terms).

Results
Characterization of GAG-hCXCL1 Interactions-We characterized the structural basis of GAG-heparin interactions from

Novel Glycosaminoglycan Binding Sites on CXCL1/MGSA
binding-induced CSPs in the WT hCXCL1 dimer. A series of HSQC spectra was collected on titrating dp8 and dp14 until essentially no changes in chemical shifts were observed. We used heparin oligosaccharides because they are nearly fully sulfated, are available commercially, and have been shown to capture endogenous interactions (4). Considering that hCXCL1 exists as monomers and dimers (monomer-dimer equilibrium constant, ϳ4 M), we initially characterized binding at low concentrations. The HSQC spectrum at 15 M showed peaks corresponding to monomers and dimers. Upon adding dp14-mer, the peaks corresponding to the monomer disappear, indicating that the hCXCL1 dimer is the high-affinity GAG ligand (Fig. 2).
For both dp8 and dp14 titrations, we observed one set of peaks that broadened out at molar ratios around 1:1 but recovered upon further GAG addition (supplemental Fig. 1). The CSP profiles for both dp8 and dp14 were similar, so we will confine our discussion to the 14-mer (Fig. 3). We observed significant CSPs of a number of basic residues that span the entire length of the protein: Arg-8 from the N terminus; His-19 and Lys-21 from the N-loop; Lys-45 and Arg-48 from the 40s turn; Lys-49 from the third ␤-strand; and Lys-60, Lys-61, and Lys-65 from the C-helix. On the basis of sequence analysis and previous studies on the ELR chemokine CXCL8, perturbations of the N-loop and C-terminal helical residues were expected. However, CSPs of the N-terminal, 40s turn, and third ␤-strand residues are novel. Some of the buried residues also show significant CSPs. Considering that buried residues cannot be involved in direct binding, their perturbations must be due to indirect interactions, such as binding-induced structure formation or structural rearrangement. For instance, CSPs of the helical residues Ile-58 and Ile-62 must be due to rearrangement of the ␣-helix, and a similar rearrangement of the helix on GAG binding to CXCL8 has also been observed (26).
We measured 1 H, 15 N heteronuclear NOEs of hCXCL1 in the free and dp8-and dp14-bound forms (Fig. 4). The NOE profiles for dp8 and dp14 were similar. Structured residues tend to have high NOE values (Ͼ0.8), and GAG binding will minimally influence their NOE. Conversely, residues that are unstructured and dynamic have low NOE values that could increase significantly to those of structured residues upon GAG binding. N-terminal Arg-8 is unstructured in the free form and shows a low NOE value that increases significantly on binding, indicating that it is more structured in the bound form. A significant increase for the N-loop residues, including Lys-21, also indicates that these residues are dynamic in the free but structured in the bound form. Higher NOE values are also observed for the 30s loop residues Cys-35 and Ala-36. Chemical shifts of Cys-35 on 14-mer binding were also perturbed (Fig. 3B). Cys-35 is buried, and Ala-36 is not likely to be involved in binding, indicating that NOE changes must be coupled to structure induction of N-terminal residues. N-terminal and 30s loop residues are linked via a disulfide, and, therefore, it is likely that the 30s residues are also dynamic because of coupled motions. Previous studies in CXCL8 have shown that mutations in the 30s loop residues could have a profound influence on function, providing further evidence that these regions are coupled structurally and functionally (29).
Inspection of the hCXCL1 structure reveals that the N-loop and C-helical residues constitute one GAG-binding domain that we define as the ␣-domain. The N-terminal, 40s turn, and third ␤-strand residues constitute a second GAG-binding domain that we define as the ␤-domain. The structure reveals that these domains are located on opposite faces of the protein and that a single dp14 oligosaccharide cannot simultaneously bind residues at both domains. To determine whether binding to the ␣or ␤-domains is coupled or independent, we characterized the binding of heparin dp14 to the ␤-domain R8A mutant. Interestingly, Arg-8 is also part of the ELR motif, and we have shown that mutating this residue results in substantial loss of function (13). The CSP profile was strikingly different because only the ␣-domain residues show significant perturbation (Fig. 5). Furthermore, the extent of perturbation of the ␣-domain residues was similar to the WT, indicating that binding to the ␣and ␤-domains is independent.
Can GAG-bound hCXCL1 Bind the Receptor?-Our observation that the N-loop and N-terminal residues are involved in GAG binding has a direct effect on function, considering that these residues have also been implicated in receptor binding. Functional studies of various chemokines, including hCXCL1, have shown that receptor activation involves two interactions: between the chemokine N-loop and receptor N-domain residues and between the chemokine N-terminal and receptor extracellular loop residues (30). We have shown previously that the structural basis of chemokine binding to the receptor N-domain can be studied outside of the context of the intact receptor by characterizing binding to N-domain peptides (21,31). Therefore, we characterized the binding of the CXCR2 N-domain peptide to the WT hCXCL1 dimer by NMR spectroscopy (supplemental Fig. 1). CSP shows a prominent role for the N-loop residues (Fig. 6). Comparison of the GAG and receptor binding profiles indicates that a number of N-loop residues that are perturbed on receptor binding are also perturbed on GAG binding, suggesting that the GAG-bound chemokine cannot The monomer peaks disappear on dp14 binding, indicating that the dimer is the high-affinity GAG ligand. The spectra were collected using a 15 M hCXCL1 sample in 50 mM sodium phosphate (pH 6.0) at 40°C. FEBRUARY 19, 2016 • VOLUME 291 • NUMBER 8 bind the receptor. Indeed, we observed that the GAG-bound chemokine is unable to bind the CXCR2 N-domain peptide ( Fig. 7A and supplemental Fig. 1). A schematic of the GAG and receptor-binding residues and the extent of overlap are shown in Fig. 7B. CD studies also indicated that the GAG-bound hCXCL1 cannot bind the receptor (supplemental Fig. 2).

Novel Glycosaminoglycan Binding Sites on CXCL1/MGSA
Structural Models of Heparin Oligosaccharides Binding to the hCXCL1 Dimer-Analysis of the GAG binding sites indicated that binding to the two sites was mutually exclusive. To stringently define the binding modes, we carried out HADDOCKbased docking that used CSP data. We obtained essentially similar results for both dp8 and dp14, so we will discuss our results for the longer oligosaccharide. We first modeled the binding of heparin dp14 to a monomer of the dimer by providing constraints to only one monomer (run I). The docking exercise resulted in three major clusters. Cluster 1 consisted of the ␣-domain residues C-helix Lys-60, Lys-61, and Lys-65 and N-loop Lys-21; cluster 2 consisted of the ␤-domain residues Arg-8, Arg-48, and Lys-49; and cluster 3 consisted of the ␣-domain Lys-65 and Lys-21 and ␤-domain Arg-48.
We then modeled the binding by providing constraints to both monomers of the dimer (run II). These resulted in two clusters corresponding to binding to the ␣-domain and ␤-domain (Fig. 8). In the first cluster, the GAG bound to N-loop His-19 and Lys-21; 40s turn Lys-45; and C-helix Lys-60, Lys-61, and Lys-65 residues of both monomers of the dimer. In the second cluster, the GAG bound to N-terminal Arg-8, 40s turn Lys-48, third ␤-strand Lys-49, and, interestingly, Lys-29 of the first ␤-strand and both monomers of the dimer (Fig. 8). We also modeled simultaneous binding of two dp14-mers to a dimer  15 N HSQC spectra showing the overlay of WT hCXCL1 in the free (black) and dp14-bound (red) forms. Arrows indicate the direction of the peak movement. B and C, histograms of chemical shift changes in the hCXCL1 dimer on binding heparin dp8 (B) and dp14 (C). The basic residues Lys, Arg, and His are shown in blue, and buried residues (ASA Ͻ 20%) are shown in black. The CSP of Lys-21 is truncated, and the actual CSP is 1.80 ppm. The horizontal line at 0.1 ppm represents the cutoff for a residue to be considered perturbed. The spectra were collected using a 100 M hCXCL1 sample in 50 mM sodium phosphate (pH 5.7) at 40°C.
(run III) and observed binding to both domains. This indicated no steric clashes and that the two binding sites are independent (Fig. 8E). None of the run I structures were observed in run II or run III, indicating that they are energetically less favored compared with interactions to both monomers of the dimer. We also modeled the binding of dp14 to the R8A mutant dimer and observed one major cluster corresponding to binding to the ␣-domain.
In all docking runs, there was no evidence for binding to the 30s loop residues, indicating that the CSP and NOE changes must be due to indirect interactions. On the other hand, the chemical shifts of Lys-29 were not perturbed, although docking studies indicate otherwise. This suggests that chemical shift changes from direct and indirect interactions are of opposite sign and similar magnitude and, therefore, cancel out. We made similar observations earlier for CXCL8, where lysines known to be involved in GAG binding from functional studies showed minimal backbone chemical shift changes (26).

Discussion
Our NMR studies indicate that the molecular basis of GAG heparin binding to hCXCL1 is novel and that it is strikingly different from the closely related ELR chemokine CXCL8. We have shown recently that heparin-binding residues in CXCL8 can be classified as core and peripheral residues and that GAG can bind in different geometries (26). The core residues in CXCL8 are located in the N-loop and C-terminal helix, and these residues are also involved in hCXCL1-GAG interactions. However, in hCXCL1, there was no evidence for peripheral residues, but, instead, binding was observed to a second domain consisting of N-terminal, first ␤-strand, 40s turn, and third ␤-strand residues on the opposite face of the protein. Comparison of the ELR chemokine sequences reveals that the N-terminal arginine is conserved, but the 40s turn and third ␤-strand basic residues are observed only in CXCL1, CXCL3, and CXCL7 (Fig. 9). Other chemokines, including mouse KC and MIP2, have one of the two basic residues. Interestingly, CXCL5, CXCL8, and KC actually have an acidic residue at the third ␤-strand position. NMR studies of CXCL8 and KC show evidence of binding only to the ␣-domain (26,32). In addition, only CXCL1, CXCL2, and CXCL3 have a basic residue at the end of the first ␤-strand. These observations suggest that the strategic presence or absence of one or two basic residues in the context of chemokine dimer structure could elicit strikingly different GAG interactions.
We used dp8 and dp14, whose dimensions easily allow binding of a dimer. Naturally occurring GAGs are much longer, and, furthermore, heparan sulfate has a modular structure consisting of sulfated regions (defined as NS) interspersed with nonsulfated acetylated regions (defined as NA). Therefore, it is possible that two NS regions on the same chain bind ␣and ␤-domains. We propose a "clamp" model in which the hCXCL1 dimer is sandwiched between two NS domains (Fig. 10A). Previous biochemical studies have proposed a horseshoe model for binding of heparan sulfate to CXCL8, where GAG lies in the same plane with NS domains binding parallel to the helix that are linked by the NA domain (Fig. 10B) (33). It is also possible that hCXCL1 interacts with NS domains from two different GAG chains. Proteoglycans can have multiple GAG chains, and it may be that initial binding to one GAG chain promotes binding to the second GAG chain because of spatial proximity on the cell surface (Fig. 10C) (34).
GAG-binding residues for various chemokines have been characterized using mutagenesis, which identified a BBXB motif in the 40s loop for CC chemokines and showed that different regions are involved in CXC chemokines such as the first ␤-strand in CXCL12 and C-helix and N-loop in CXCL8 (4, 35-37). However, structural characterization of chemokine-    FEBRUARY 19, 2016 • VOLUME 291 • NUMBER 8 GAG complexes has been challenging. X-ray structures could be determined only of a disaccharide-bound chemokine (38,39). Although NMR characterization of binding of disaccharides posed no problems, characterizing longer oligosaccharides seems to depend on the chemokine, with some giving good spectra and others precipitating even for a tetrasaccharide (26,32,40,41). We obtained excellent NMR spectra for hCXCL1, and, more significantly, our observation of two distinct non-overlapping GAG binding domains is novel.

Novel Glycosaminoglycan Binding Sites on CXCL1/MGSA
Animal model studies have shown that the ability to reversibly exist as monomers and dimers regulates recruitment (3,4,13). Under conditions of active neutrophil trafficking, the local chemokine concentration could vary by orders of magnitude, and, therefore, in principle, hCXCL1 could exist as monomers, dimers, or both at different times and locations. We unambiguously show that the hCXCL1 dimer is the high-affinity GAG ligand and that the GAG-bound hCXCL1 is occluded and not accessible for receptor binding. Using a disulfide-trapped dimer, we have shown that the hCXCL1 dimer is highly active for receptor function (21). On the basis of these observations, we propose that chemotactic gradients and not haptotactic gradients direct recruitment. During early stages of recruitment, when the local chemokine concentration is low, dimers exist in the GAG-bound form and monomers in the soluble form. At higher local concentrations, dimers in addition to the GAGbound form could also exist in solution. The amount of free chemokine monomers and dimers dictate the steepness of the chemotactic gradients and the extent of CXCR2 activation, which, in turn, dictates the flux and kinetics of cellular trafficking.
Chemokine function is also regulated by proteolytic cleavage (42). For instance, Streptococcus pyogenes protease inactivates hCXCL1 and hCXCL2 to dampen the innate immune response by cleaving the C-terminal helix at residues between Lys-61 and Lys-62 (43). Considering that these residues are buried in the GAG-bound form, GAG-bound hCXCL1 is much less susceptible to proteolysis, indicating that GAG-binding also plays a direct role in evading microbial infection. Endogenously pro- Residues that are solvent-exposed but occluded in the GAG-bound form (ASA Ͻ20%) were considered to be involved in binding. For receptor binding, residues that showed significant CSP and were solvent-exposed and N-terminal Glu-6 and Leu-7 (of the ELR motif) were considered to be involved in binding. duced CXCL1 has also been shown to be heterogeneous because of the N-terminal cleavage, and the isoforms have differential neutrophil activity (44). Our observation that the N-terminal residues are also involved in GAG binding suggests that GAG binding also indirectly influences receptor activation by regulating accessibility to proteases. In the case of CXCL12,  D). A and B, the hCXCL1 dimer is shown in ribbon representation, and GAG is shown as sticks. C and D, the geometry of the GAG chain and interactions of the GAG-binding residues. In the two monomers, GAG-binding residues are shown in light and dark blue, respectively. E, GAG can bind both domains without steric clashes. Two GAG chains are represented as spheres. We used the hCXCL1 dimer (PDB code 1MGS) and heparin 12-mer (PDB code 1HPN) structures to generate the HADDOCK models.

Novel Glycosaminoglycan Binding Sites on CXCL1/MGSA
a similar observation of competition between N-domain and GAG binding and proteolytic protection in the GAG-bound form has been reported (41).
Previous studies have shown that the CXCL8 dimer is the high-affinity GAG ligand (26) and that, for various CC chemokines, GAG binding and dimerization/oligomerization are coupled (45,46). Chemokine expression in vivo is complex, and both robust expression of multiple chemokines and selective expression of just one chemokine have been observed, depending upon the biological context. This study provides compelling evidence that small differences in sequences significantly influence the molecular basis of GAG binding, which, in turn, may dictate chemokine-specific fine-tuning of trafficking of different cell types to distal and remote locations.
Author Contributions-K. R. and K. M. S. designed the research and analyzed the data. K. M. S. performed the experiments, and K. R. wrote the paper with the assistance of K. M. S. Both authors reviewed the results and approved the final version of the manuscript.