Characterization of the Chemokine CXCL11-Heparin Interaction Suggests Two Different Affinities for Glycosaminoglycans*

Chemokines orchestrate the migration of leukocytes in the context of homeostasis and inflammation. In addition to interactions of chemokines with receptors on migrating cells, these processes require interactions of chemokines with glycosaminoglycans (GAGs) for cell surface localization. Most chemokines are basic proteins with Arg/Lys/His residue clusters functioning as recognition epitopes for GAGs. In this study we characterized the GAG-binding epitopes of the chemokine I-TAC/CXCL11. Four separate clusters of basic residues were mutated to alanine and tested for their ability to bind to GAGs in vitro and to activate the receptor, CXCR3. Mutation of a set of basic residues in the C-terminal helix (the 50s cluster, 57KSKQAR62) along with Lys17, significantly impaired heparin binding in vitro, identifying these residues as components of the dominant epitope. However, this GAG mutant retained nearly wild type receptor binding affinity, and its ability to induce cell migration in vitro was only mildly perturbed. Nevertheless, the mutant was unable to induce cell migration in vivo, establishing a requirement of CXCL11 for GAG binding for in vivo function. These studies also led to some interesting findings. First, CXCL11 exhibits conformational heterogeneity, as evidenced by the doubling of peaks in its HSQC spectra. Second, it exhibits more than one affinity state for both heparin and CXCR3, which may be related to its structural plasticity. Finally, although the binding affinities of chemokines for GAGs are typically weaker than interactions with receptors, the high affinity GAG binding state of CXCL11 is comparable with typical receptor binding affinities, suggesting some unique properties of this chemokine.

In a survey of the relative affinity of chemokines for heparin, we noted that I-TAC (IFN-␥-inducible T-cell ␣-chemoattractant)/CXCL11 binds to heparin-Sepharose more tightly than most chemokines with the exception of CCL5 (6) and were therefore interested in examining the GAG binding epitopes on this chemokine. CXCL11 was first discovered in astrocytes through cDNA sequencing (23) and was subsequently shown to be induced by IFN-␥ in a wide variety of additional cells, including bronchial epithelial cells, basal keratinocytes, neutrophils, and endothelial cells (24 -26). CXCL11 signals through a single receptor, CXCR3, which it shares with two additional chemokines, IP-10 (IFN-␥-inducible 10-kDa protein)/CXCL10 and MIG (IFN-␥-induced monokine)/CXCL9. CXCR3 is expressed on activated T-lymphocytes, mainly of the Th1 subtype, as well as on a subset of B-cells, monocytes, and NK cells (27). Although these three chemokines are agonists of CXCR3, they have all been described as natural antagonists of CCR3 in Th2type responses (28). Furthermore, CXCL11 was recently shown to bind to a second receptor, CXCR7, but it does not elicit classic signaling responses, such as calcium flux or cell migration (29).
To define the GAG binding epitopes of CXCL11, we first mutated four clusters of basic residues to alanine: a BBXB motif located in the N-terminal region ( 5 KRGR 8 ) and three clusters located in the 40s loop ( 46 KENKGQR 52 ), in the 50s loop ( 57 KSKQAR 62 ), and in the 60s loop toward the C terminus ( 66 KKVERK 71 ). In silico investigation by modeling with an 11-mer oligosaccharide, also suggested the involvement of Lys 17 , which was combined with the 50s mutations due to its spatial proximity. Several methods have been used to define GAG binding sites, but they do not always give consistent results, and typically multiple methods are best used in concert for the most definitive answers (30 -32). We therefore analyzed the CXCL11 mutants by affinity chromatography on heparin-Sepharose, saturation binding assays on immobilized heparin, mass spectrometry identification of a protected fragment from a tryptic digestion of a heparin-CXCL11 complex, and last, the ability to recruit cells in vitro and in vivo. Consensus data from both the in vitro and in vivo experiments suggest that the major epitope for GAG binding consists of a relatively diffuse site, including the 50s loop and Lys 17 . The classical BBXB heparin binding motif in the N-terminal region did not contribute to GAG binding but was shown to be involved in receptor binding, together with a contribution from the 40s loop. Last, analysis of these mutants in an equilibrium competition binding assay on solid-phase heparin with soluble heparin as competitor identified an unusually high affinity (picomolar) binding site for CXCL11, which was subsequently found to also exist for CCL5 but not for CCL2 or CXCL12. Future studies will address whether this high affinity binding site is present in other GAG families/chemokines as well as the structural nature and biological relevance of the interactions.

EXPERIMENTAL PROCEDURES
Reagents-Unless stated otherwise, all chemicals were purchased from Sigma. Commercial CXCL11 was obtained from PeproTech. The heparin used in the assays was unfractionated heparin sodium salt supplied by Sigma (5-30 kDa; catalogue no. H3393) or low molecular weight heparin sodium salt (3 kDa; catalogue no. H3400). Enzymes were purchased from New England Biolabs, and chromatographic material was from GE Healthcare. EC 50 and IC 50 values were calculated using GraphPad Prism software, employing the equation, sigmoidal dose response (variable slope), with Y ϭ bottom ϩ (top Ϫ bottom)/(1 ϩ 10ˆ((log EC 50 Ϫ X)*Hill slope), where X is the logarithm of concentration and Y is the response. For the double binding site curves, GraphPad Prism software was used, employing the equation for a biphasic dose response.
Molecular Modeling-Docking calculations for heparin oligosaccharides with the NMR structure of CXCL11 (Protein Data Bank code 1RJT) were carried out using a validated protocol for the identification of heparin binding sites on protein surfaces (33). This program allows limited flexibility in the ligand structure but keeps the protein rigid. Three oligosaccharide structures were investigated, all taken from the NMR structure of heparin (Protein Data Bank code 1HPN). Two structures were pentasaccharide substructures, with iduronate residues in the 1 C 4 and 2 S 0 conformations, respectively. All exocyclic bonds other than the glycosidic linkages were allowed to rotate. The third oligosaccharide was the full endecasaccharide, which was maintained in a rigid conformation with no rotatable bonds. The Protein Data Bank entry for CXCL11 contains coordinates for 10 structures compatible with the NMR data, which vary primarily in the flexible N-terminal region (34). All three oligosaccharides were docked with the first of the 10 structures, whereas the endecasaccharide was docked with all 10 structures in the ensemble. Docking was performed as described (33), using the program Autodock version 2.4. Specifically, an 84-Å cubical grid, with a grid spacing of 0.7 Å centered on the mean of the protein coordinates, was used so that the entire protein surface was available to the ligand. For the docking procedure, 128 simulated annealing runs of 300 cycles were performed. No clustering was carried out because no analysis of binding modes was required.
Generation of CXCL11 Mutants-Mutagenesis was performed by PCR. The mutants were created in one or two PCR steps, depending on the relative position of the mutations. A pET24d plasmid with the CXCL11 sequence was used as the template; in addition to the WT sequence, it contains the mutation F73W, which was introduced for protein quantification, and an N-terminal tag (MKKKWP) followed by a caspase 8 cleavage site (LETD) to produce the native N terminus. The mutations K17A and K38A were made in the context of the 50s loop mutant using the QuikChange mutagenesis kit (Stratagene).
Protein Purification and Characterization-The mutant plasmids were transformed into competent BL21 Star Escherichia coli cells. 5 liters of LB medium containing kanamycin were inoculated and, after induction with 1 mM isopropyl-␤-Dthiogalactopyranoside at A 600 ϭ 0.6, allowed to grow for 3.5 h at 37°C before harvesting. The cells were resuspended in lysis buffer (50 mM Tris/HCl buffer, pH 8, containing 10 mM MgCl 2 , 5 mM benzamidine/HCl, 1 mM dithiothreitol, 0.1 mM phenylmethylsulfonyl fluoride, and 20 mg/liter DNase). Cells were broken by three passages through a French press. The suspension was then centrifuged at 10,000 ϫ g for 60 min at 4°C. The inclusion body pellet containing the recombinant protein was solubilized at 1 g/20 ml in 0.1 M Tris/HCl, pH 8.0, containing 6 M guanidine/HCl and 1 mM dithiothreitol and stirred for 30 min at 60°C. The proteins were renatured by dropwise 10-fold dilution into 0.1 M Tris/HCl buffer, pH 8.0, containing 0.01 mM oxidized glutathione and 0.1 mM reduced glutathione, with stirring overnight at 4°C. The pH was adjusted to 4.5 with acetic acid, and the conductivity was lowered below 20 millisiemens by dilution with distilled H 2 O. The solution was applied to an SP Sepharose column (16/10) previously equilibrated in 50 mM sodium acetate, pH 4.5. Protein was eluted with a 0 -2 M gradient of NaCl in the same buffer. Fractions containing CXCL11 or mutant were pooled and dialyzed twice against 1% acetic acid and finally against 0.1% trifluoroacetic acid. Insoluble material was removed by centrifugation at 10,000 ϫ g for 30 min, and the supernatant was lyophilized.
In order to remove the leader tag sequence, the lyophilized proteins were dissolved at 1 mg/ml in H 2 O and applied to PD-10 columns previously equilibrated with caspase 8 cleavage buffer (25 mM Tris, pH 7.5, containing 15% glycerol, 150 mM NaCl, and 2 mM EDTA). The MKKKWPLETD leader sequence was cleaved by incubation with caspase 8 (enzyme/substrate, 1:10, w/w) for 30 min at 37°C. The cleaved proteins were separated from uncleaved protein by reverse phase HPLC (Alliance HPLC, Waters; column: EC 250/4 Nucleosil 300-7 C8, Macherey-Nagel) using a linear gradient of 25-50% acetonitrile in 0.1% trifluoroacetic acid over 20 min at a flow rate of 2 ml/min. Protein sequences were verified by matrix-assisted laser desorption ionization mass spectrometry, which confirmed the formation of the two disulfide bonds. The function of the WT protein was confirmed by receptor binding and chemotaxis assays as described below.
Proteolytic Footprinting-For GAG footprinting experiments, CXCL11 was proteolytically digested with trypsin in the presence and in the absence of heparin. For this purpose, CXCL11 (20 pmol) was incubated for 1 h at 20°C with a 10-fold molar excess of heparin in 50 mM NH 4 HCO 3 , pH 8.0, containing 150 mM NaCl to suppress nonspecific interactions. After the addition of trypsin (sequencing grade unmodified, Roche Applied Science) in an enzyme/substrate ratio of 1:10 (w/w), proteolytic cleavage was carried out at 37°C for 18 h. The digest was stopped by adjusting the pH to 2.5 with formic acid. The resulting peptide fragments were identified by nano-HPLCtandem mass spectrometry as described previously (35). Potential GAG binding epitopes on the chemokine were identified as non-digested (protected) peptides observed in the presence of heparin but missing in the absence of heparin.
Heparin-Sepharose and S-Sepharose (Cation Exchange) Chromatography-As an initial screen to identify residues comprising the GAG-binding epitope, the amount of salt required to elute WT and mutant CXCL11 from a heparin-Sepharose column was determined. 100 g of either CXCL11 or mutant protein was loaded onto a heparin-Sepharose column (5/5), equilibrated in 50 mM Tris/HCl, pH 7.5, and eluted with a linear gradient of 0 -2 M NaCl in the same buffer, as described previously (21). To measure the contribution of various residues to the specificity of the interaction, the amount of salt required to elute mutants from a nonspecific S-Sepharose column (SP column 5/5) was also determined using the same protocol and compared with the elution from the heparin-Sepharose column. The specificity index is related to ⌬⌬NaCl as calculated from the formula below, where the superscripts Hep and S refer to elution from the heparin-Sepharose or S-Sepharose column, respectively (19), Saturation Binding Assay to Immobilized Heparin-In order to test the ability of WT CXCL11 and the mutants to bind to heparin, a colorimetric enzyme-linked immunosorbent assay on the surface of microtiter plates was used (Epranex Plates, Plasso Technology Ltd., Sheffield, UK). The plate was coated with heparin by incubation with a heparin solution (25 g/ml) in PBS (200 l/well) overnight at room temperature and in the absence of light, as described by the supplier. The liquid was discarded, and the plate was washed three times with standard assay buffer (SAB) (100 mM NaCl, 50 mM NaOAc, 0.2% (v/v) Tween 20, pH 7.2). 250 l of blocking solution (0.2% (w/v) gelatin in SAB) was then added per well, and the plate was incubated for 1 h, protected from light. Dilution series ranging from 10 Ϫ6 to 10 Ϫ12 M CXCL11 and mutant proteins were prepared. 100-l samples were then transferred in duplicate to the heparin-coated Epranex plate and incubated for 2 h at room temperature with gentle shaking. The wells were washed three times with SAB. For detection, a biotinylated anti-human CXCL11 antibody (R&D Systems; dilution 1:500 in SAB) was used. 200 l of antibody solution was added to each well and incubated for 1 h at room temperature. The plate was then washed three times with SAB and incubated with ExtrAvidin-AP (Sigma) in blocking solution (1:10,000). After three washes with SAB, 200 l of the development reagent (p-nitrophenyl phosphate liquid substrate system, Sigma) was added to each well. The plate was incubated for 40 min at room temperature, and the absorbance was measured at 405 nm.
Immobilized Heparin Competition Binding Assay-Competition experiments were performed in 96-or 384-well filter plates (Millipore MultiScreen, 0.22-m pore size, low protein binding) in a total volume of 100 or 40 l/well, respectively. Each well contained 0.1 nM [ 125 I]chemokine, heparin-Sepharose beads, or binding buffer as a background control and 0 -2 mg/ml heparin. This corresponds to 0 -667 M for the fractionated low molecular mass 3-kDa heparin, H3400. Using an average molecular mass of 17.5 kDa for unfractionated heparin, H3393, which consists of chains ranging from 5 to 30 kDa, the corresponding concentration range is 0 to ϳ115 M. The mass of heparin on the beads corresponded to 0.125 or 0.05 g/well for 96-or 384-well plates, respectively. The plates were incubated by shaking at room temperature for 4 h in binding buffer (50 mM Tris, pH 7.4, containing 0.5% bovine serum albumin, 5 mM MgCl 2 , and 1 mM CaCl 2 ). The beads were washed three times with 200 l of binding buffer containing 0.15 M NaCl by vacuum filtration. For CCL5, it was necessary to add 0.15 M NaCl to the binding buffer during the incubation and 0.5 M NaCl to the binding buffer for the washes to avoid nonspecific binding to the filter plate. The filters were air-dried, 30 l of scintillation fluid was added to each well, and the radioactivity was measured in a Wallac Microbeta counter. Each experiment was performed in triplicate.
Equilibrium Competition Receptor Binding Assays-The assays were carried out on membranes from CHO transfectants expressing CXCR3, using a scintillation proximity assay (SPA) with [ 125 I]CXCL11 as the tracer. Serial dilutions of the unlabeled chemokines, covering the range indicated in the figure, were prepared in binding buffer (50 mM Tris/HCl, pH 7.2, containing 1 mM CaCl 2 , 5 mM MgCl 2 , and 0.5% bovine serum albumin). Wheat germ SPA beads (Amersham Biosciences) were solubilized in PBS at 50 mg/ml and diluted in binding buffer to a concentration of 10 mg/ml, and membranes of CHO transfectants expressing CXCR3 were solubilized at 20 g/ml in binding buffer. Equal volumes of membrane and SPA bead solutions were mixed before adding them to the assay. The final membrane concentration in the assay was 5 g/ml, and the concentration of [ 125 I]CXCL11 was 0.05 nM. The plates (Corning, 96-well, flat and clear bottom) were incubated at room temperature under agitation for 4 h. Radioactivity was counted with a Microbeta counter for 1 min/well.
Chemotaxis Assays-Chemotaxis assays were carried out in 96-well microplates (Neuro Probe ChemoTx) with L1.2/ CXCR3 cell transfectants. The cells were cultured in RPMI 1640 medium containing 5% inactivated fetal calf serum, L-glutamine (2 mM), 25 mM HEPES, 0.05 mM ␤-mercaptoethanol, and 0.8 mg/ml Geneticin G-418. The day before the assay, n-butyric acid (5 mM) was added to the culture medium. The cells were collected by centrifugation (600 ϫ g) and resuspended at a concentration of 0.85 ϫ 10 6 /ml in RPMI 1640 medium containing 5% heat-inactivated fetal calf serum without phenol red. A dilution series of CXCL11 and mutants was prepared covering the range of 10 Ϫ6 to 10 Ϫ12 M in RPMI medium without phenol red, and 30 l were placed in the wells of the lower plate. A filter membrane (8-m pore size) was placed over the plate. 25 l of the cell solution (2.5 ϫ 10 4 cells) was deposited on top of the membrane of each well. The chamber was incubated for 5 h at 37°C under 5% CO 2 . The cell suspension was then washed with PBS, and the filter was removed. The migrated cells in the lower wells were transferred to a black plate using a 96-well funnel adaptor and frozen for at least 2 h at Ϫ80°C. The plate was thawed, and a solution of CyQUANT dye/cell-lysis buffer mix (Molecular Probes, 200 l) was added to each well. The fluorescence was measured using a Victor 2 Wallac plate reader.
Peritoneal Cellular Recruitment-Because CXCR3 is expressed on activated Th1 cells, mice were sensitized to activate T cells prior to the peritoneal recruitment assay. On day 0, 8 -12-week-old female Balb/c mice were sensitized by five subcutaneous injections (4 ϫ 50 l into each limb and 1 ϫ 100 l into the scruff of the neck) of 10 nM CpG-ODN (Microsynth) mixed with 100 g of ovalbumin (Sigma, Grade V) in sterile PBS. Cellular recruitment was induced on day 6 by intraperitoneal injection of 10 g of CXCL11 or mutant diluted in 0.2 ml of sterile, LPS-free NaCl (0.9%). Mice were sacrificed with aerosolized CO 2 4 h later. Peritoneal lavage was performed with 3 ϫ 5 ml of PBS, and the lavages were pooled. Cells were centrifuged at 1500 rpm for 5 min and resuspended in 1 ml of PBS, and total leukocytes were counted with a hemacytometer.
NMR Spectroscopy-15 N-Labeled CXCL11 was made by bacterial expression in minimal media as described previously (36). All spectra were run on Bruker Avance II 600-MHz spectrometer with a TCI CryoProbe. 1 H-15 N HSQC spectra were recorded under several conditions of pH and temperature and were screened for spectral quality, but the protein consistently showed conformational heterogeneity. For the representative spectra shown in Fig. 7, the sample was prepared at a concentration of 1 mM in 10 mM acetate buffer, pH 5.3, and the data were collected at 23°C. For the pulsed field gradient diffusion measurements, the 15 N-labeled CXCL11 sample was prepared in 50 mM acetate-d 6 , pH 5.6, and the experiments were conducted using the pulse sequence ledbpgpprwg2s with the Bruker macro DOSY. Values for the diffusion time, d20 (⌬) and the gradient pulse length, p30 (␦*0.5) were 150 and 2.0 ms, respectively. The self-diffusion coefficients (D s ) were calculated using the Bruker program T 1 /T 2 relaxation with manual integration for peaks at 7.0, 3.0, 2.0, and 0.7 ppm for each proton spectrum. The resulting decay curves were fit, and the D s values were calculated using the equation, where I(0) is 4.167 ϫ 10 Ϫ1 , ␥ is 4.258 ϫ 10 3 Hz/G, g was calibrated at 5.784 gauss/mm, ␦ was set to 4.0 ms, and ⌬ was set to 150 ms. The theoretical change in D s from a monomeric to dimeric form of a protein may be estimated by approximating the monomer-monomer interaction as a hard sphere molecular contact. Using the expected change in the frictional coefficient in conjunction with the Stokes-Einstein equation, the theoretical ratio D s,dimer /D s,monomer is 0.75 (37). Each sample was run in triplicate, and the D s values were averaged over the three experiments.

RESULTS
Because the primary GAG binding sites of many chemokines have been frequently localized to residues close in sequence, such as BBXB and BBXXB motifs, we initiated our characterization of the GAG site on CXCL11 by individually mutating four clusters of basic residues. Within each cluster, Lys and Arg residues were replaced by Ala because they are frequently found to be the dominant amino acids involved in heparin binding. The mutants are referred to as the 10s ( 5 AAGA 8 ), the 40s ( 46 AENAGQA 52 ), the 50s ( 57 ASAQAA 62 ), and the 60s ( 66 AAVEAA 71 ) mutants. Following purification and renaturation, they were all found to be Ͼ95% pure as determined by SDS-PAGE and reverse phase HPLC (data not shown). The formation of the two disulfide bonds was confirmed by a shift in the reverse phase HPLC retention time and a loss of 4 Da by mass spectrometry (data not shown).
Molecular Modeling-In parallel, molecular docking calculations were used to predict amino acids that might be involved in GAG binding. A published protocol was used, which has been optimized for rapid identification of heparin binding sites on the surfaces of small proteins (33). In this protocol, docking of pentasaccharide ligands with flexible exocyclic bonds is used to identify the essential core of the heparin binding site on the protein, whereas docking with a rigid endecasaccharide can identify extended heparin binding sites. The protocol does not provide simulations of the bound complexes at atomic resolution. Conformational flexibility of the protein partner in this docking procedure cannot be addressed directly, but when an ensemble of protein structures is available from an NMR study, as in this case, it is possible to assess the effects of conformational flexibility on the location of the heparin binding site by docking to each set of coordinates in the ensemble. CXCL11 is a basic protein, with most of the arginine and lysine residues concentrated on one face, forming two strongly basic surface patches in which basic amino acids distant in sequence are close together in space. One of the patches includes Lys 17 , Lys 38 , Arg 52 , Lys 57 , Lys 59 , and Arg 62 . All 10 of the NMR structures were predicted by the docking calculations to interact with heparin by means of some residues in this patch, and for those structures in which the N-terminal BBXB motif is in close proximity, three basic residues (Lys 5 , Arg 6 , and Arg 8 ) are also predicted to be involved. Fig. 1 shows one of the CXCL11 NMR structures with the heparin endecasaccharide docked in the 10 lowest energy conformations. The second basic patch involves the residues Lys 20 , Lys 66 , Lys 67 , Arg 70 , and Lys 71 , and for most of the predicted complexes, the endecasaccharide reaches Lys 66 and sometimes Lys 70 , but not Lys 71 and Lys 20 . Partially flexible heparin pentasaccharides docked with the first NMR structure occupied the first basic patch and the N terminus. Based on these docking results, we therefore made an additional mutant containing the 50s cluster plus K17A to more completely investigate the first basic patch. Fig. 1B shows the amino acid composition of each of the mutants compared with WT.
Heparin Binding Assays of CXCL11 Mutants Highlight the 50s Cluster as the Principle GAG Binding Epitope-Two assays were initially used to examine the GAG binding ability of the CXCL11 mutants. In both cases, we used heparin as a model GAG because it has been successfully used for many years for identifying GAG binding epitopes. Furthermore, as shown here and elsewhere (7,21,31,38), the agreement between the in vitro and in vivo assays validates the use of heparin for this purpose. In the first assay, the NaCl concentration required to elute the mutant proteins from a heparin-Sepharose column was determined and compared with WT, yielding ⌬NaCl Hep . Values of 0.38 and 0.49 for the 50s and 50s ϩ K17A, respectively, suggest that these epitopes are more important contributors than the 60s, 40s, and 10s, which had values of 0.21, 0.17, and 0.08, respectively ( Table 1), results that are largely in agreement with the modeling studies. The concentration of NaCl required for elution from a nonspecific S-Sepharose column was also determined in comparison with WT yielding ⌬NaCl S . These values confirm the relative ordering of the mutants, as indicated by ⌬NaCl Hep . ⌬NaCl S was then subtracted from ⌬NaCl Hep to eliminate nonspecific electrostatic contributions, yielding ⌬⌬NaCl, which effectively provides a measure of the specificity of a given amino acid for heparin. Thus, a greater effect of the mutations on specific heparin binding is reflected by more positive ⌬⌬NaCl values. Both the 50s epitope and 50s ϩ K17A had ⌬⌬NaCl values of 0.09, the 60s epitope had a ⌬⌬NaCl of 0.08, and the 10s and 40s epitopes had a ⌬⌬NaCl of 0.04, suggesting the highest specificity for the 50s cluster (Table 1). Although the 60s was a close second, the overall affinity as defined by ⌬[NaCl] Hep and ⌬[NaCl] S suggests that it is less important than the 50s.
The second assay consisted of direct binding measurements of the proteins to heparin immobilized on specialized plates, with bound protein detected by a polyclonal antibody. The EC 50 values for the WT, 10s, and 60s mutants were very similar FIGURE 1. Structure of CXCL11 and CXCL11 mutant sequences. A, molecular modeling of the first NMR structure of CXCL11 in complex with a heparin endecasaccharide. Heparin is shown in yellow, and the basic residues predicted to be in contact with heparin are shown as green sticks. Predictions for the other NMR structures differ only in detail; all involve the 50s cluster, some involve the 60s cluster, and those (like the complex shown) where the unstructured N terminus happens to be in a favorable position also involve the N-terminal BBXB sequence. B, amino acid sequence of CXCL11 and the mutants. The mutated clusters are shown in red. Residues protected from tryptic digestion of the CXCL11-heparin complex are shown in blue. The additional residue predicted to participate in GAG binding by the molecular modeling, Lys 17 , is shown in green, and was used in combination with the 50s cluster to generate 50s ϩ K17A.

TABLE 1 Heparin affinity chromatography
The concentration (molar) of NaCl required for elution of the GAG binding mutants from heparin-and S-Sepharose columns compared with WT CXCL11 is indicated by ⌬NaCl Hep and ⌬NaCl S , respectively. ⌬⌬NaCl is the specificity index as described under "Experimental Procedures." Interestingly, as shown in Fig. 2, the 50s mutant showed a 50% reduction in binding capacity, and when combined with K17A, the binding capacity decreased to 30%. Possible reasons for this behavior will be discussed later. Nevertheless, these data further support a dominant contribution to GAG binding by the combination of the 50s ϩ K17. In parallel with the above studies, a proteolytic protection assay was undertaken. Proteolytic footprinting allows for the identification of regions of a protein that bind to GAGs without the requirement for mutagenesis. The protein of interest, complexed to a GAG, is subject to digestion with trypsin. The resulting peptides are subsequently analyzed by mass spectrometry and compared with those obtained after a digest of the free protein (35). Regions that are complexed with the GAG are protected from digestion and are thus identified in comparison with digests in the absence of heparin (or other GAG). This method was used to confirm the GAG binding site of CXCL11 as being principally located in a region encompassing the 50s loop ( 53 CLNPKSKQAR 62 ) because this region was digested in the absence of heparin but remained intact in the presence of heparin (supplemental Fig. 1, A and B). Detected ions had masses of 1145.45 (single charged) and 573.20 (double charged), which correspond to molecular weights of 1144.45 (single charged) and 1144.40 (double charged) and compare well with the calculated molecular weight of 1144.36. Strong protection was not observed anywhere else. Although our data suggest the importance of Lys 17 , one would not necessarily expect to be able to detect protection of this residue in a proteolytic footprinting assay. It is not associated with a large sequential cluster, which might be required to observe significant protection, and it is distant in sequence from the 50s epitope.

CXCL11 Heparin-Sepharose S-Sepharose ⌬NaCl
Competition Receptor Binding Assays Indicate that the 50s Cluster plus Lys 17 Contribute Little to Receptor Binding Affinity and Receptor Activation-The ultimate tests of the relevance of GAG binding and the identification of binding epitopes are in vivo assays of cell recruitment. As a prelude to these studies, however, it is critical to determine whether mutations that affect GAG binding also impact receptor binding and activation for accurate interpretation of the in vivo results. Accordingly, the receptor binding affinities of the mutants for CXCR3 were determined in comparison with the WT protein using an SPAbased competition binding assay with CHO membranes expressing CXCR3 (Fig. 3A). Our recombinant WT CXCL11 was compared with the commercially available CXCL11 protein obtained from PeproTech, and both proteins yielded similar results, with IC 50 values of 0.3 and 0.6 nM (results not shown). The mutants were all able to compete for the iodinated WT protein albeit with varying affinities. The most pronounced loss of affinity was seen for the 10s mutant, which had an IC 50 of 102 Ϯ 61 nM, representing a 225-fold loss compared with WT (based on the mean WT value of 0.45 nM). This observation is consistent with the importance of the amino terminus in receptor binding (and signaling) for all chemokines reported to date. More specifically, the basic residue, Arg 8 , in the N terminus of CXCL10 has been shown to play a role in binding to CXCR3 (39), and this residue is also an Arg in CXCL11. For the 40s mutant, the IC 50 value was determined to be 13.7 Ϯ 2.75 nM, which corresponds to a 30-fold loss of affinity, suggesting that the 40s loop is involved in receptor binding. The 50s and 60s mutants showed little loss of affinity, with IC 50 values of 1.3 Ϯ 1.0 and 2.6 Ϯ 0.5 nM, respectively, corresponding to only 3-and  6-fold drops. Introduction of K17A into the 50s mutant caused a further 3-fold reduction of affinity with an IC 50 value of 3.0 Ϯ 3.9 nM.
One of the most intriguing findings, observed in five separate experiments, was that both the commercial CXCL11 and our recombinant protein had 30 -50% lower [ 125 I]CXCL11 bound to the membrane-coated beads compared with the mutants at the lowest picomolar competitor concentrations used in Fig.  3A. This suggested the presence of an even higher affinity binding mode for the WT protein that was eliminated by the mutations. We therefore extended the competition binding experiments to lower concentrations of cold competitor, including a control with no cold chemokine added. Importantly, these curves all converge at the lowest concentrations, and they do indeed show the presence of a subpicomolar interaction that represents ϳ30 -50% of the total binding for the WT protein (Fig. 3B).
The ability of the CXCL11 mutants to activate CXCR3 was further assessed with a chemotaxis assay using L1.2 cells expressing CXCR3. The 50s, 60s, and 50s ϩ K17A mutants demonstrated equivalent potency to the WT, with a maximal response at 0.3 nM, consistent with a minimal perturbation of receptor binding. In accordance with its loss of affinity for the receptor, the 10s mutant showed significantly impaired recruitment of CXCR3-expressing cells with a 50% loss in efficacy and 400-fold loss of potency such that 125 nM ligand was required to achieve the maximal response (Fig. 4). Paralleling the 30-fold loss of affinity for CXCR3, the 40s loop mutant required 7 nM ligand for the maximal response (20-fold lower than the WT), and the efficacy was 50% of WT. Considering all of the data, the 50s and 50s ϩ K17A mutants showed the least perturbation on receptor binding and activation but the largest effect on GAG binding, making them particularly suitable for evaluating the role of GAG binding and the importance of these epitopes in vivo.
An in Vivo Peritoneal Cellular Recruitment Assay Confirms the Importance of GAG Binding and the 50s Epitope for the Chemotactic Activity of CXCL11-The ability of the cluster mutants to attract cells in vivo relative to WT CXCL11 was examined in a T cell peritoneal recruitment assay. Because CXCR3 is only expressed on activated T cells, the mice were sensitized with CpG prior to the assay (40). Interestingly, most of the mutants showed impaired cell recruitment compared with WT CXCL11, although the small reduction observed for the 60s was not statistically significant and indicates that it contributes little to GAG binding (Fig. 5). The inability of the 10s and 40s mutants to recruit cells in the in vivo assay was expected, based on the in vitro binding and chemotaxis assays. However, because the 50s loop mutant showed a WT chemotaxis profile and little difference in affinity for the receptor relative to WT CXCL11, the inability to recruit cells in vivo can only be attributed to the defect in GAG binding. In fact, although the 40s mutant was significantly impaired in the in vitro chemotaxis assay relative to the 50s mutant, it was capable of recruiting more cells than the 50s mutant, further emphasizing the relevance of the 50s region and not the 40s on GAG binding. These in vivo data are also remarkably consistent with the in vitro data, despite the use of heparin as a surrogate for heparin sulfate or other GAGs for identification of the GAGbinding epitopes.
An Immobilized Heparin Assay Defines a High Affinity GAG Site-In principle, the picomolar binding site identified in the equilibrium receptor binding assay may represent a high affinity site on CXCR3, another receptor, or a GAG site. To investigate the potential role of a high affinity GAG site, we conducted a solid phase binding assay on heparin beads, extending the concentration range of the heparin from 10 Ϫ6 to 10 Ϫ15 M. Surprisingly, a high affinity binding site was in fact observed (Fig.  6A). Using average molecular weights as outlined under "Experimental Procedures," the high affinity sites correspond to an IC 50 of ϳ100 pM for the 3-kDa heparin (H3400) and 13 pM for unfractionated heparin (H3393, 5-30 kDa). These numbers are obviously only approximations and rest on the additional assumption that there is only one binding site per chain (which is probably true for 3-kDa heparin (ϳ10 monosaccharides) but not unfractionated heparin); thus, the data are plotted as mg/ml. Nevertheless, although approximate, they do suggest an unusually high affinity of heparin for CXCL1l. In view of this unexpected observation, we tested CCL5 in this assay, and observed the same phenomenon: a high affinity site with an IC 50 of 20 pM for 3-kDa heparin and 2 pM for unfractionated heparin, in addition to the well characterized low affinity site FIGURE 4. In vitro chemotaxis assay. In accordance with its inability to bind the receptor, the 10s mutant shows significantly impaired chemotactic potency, 400-fold lower than WT. The maximum of the chemotactic response for the 40s mutant is shifted to 20 times higher concentration relative to WT CXCL11. The 50s, 60s, and 50s ϩ K17A mutants, on the other hand, showed chemotactic behavior comparable with that of WT CXCL11. Error bars, S.D. with an IC 50 of 0.15 M (Fig. 6B). However, no obvious high affinity binding sites were observed for CCL2 and CXCL12, where the IC 50 values for competition by heparin were more typical, at 0.34 and 0.14 M, respectively (Fig. 6C).
NMR Studies of CXCL11 Indicate That It Forms a Dimer and Is Conformationally Heterogeneous in Solution-To corroborate the above results, we intended to use chemical shift perturbation experiments with heparin fragments of defined size and composition. To this end, we prepared 15 N-labeled CXCL11 and recorded HSQC spectra at various temperatures and pH conditions. However, unlike most other chemokines, CXCL11 showed evidence of multiple species in slow exchange, by the presence of more cross-peaks than one would expect, with most peaks at least doubled and in some cases tripled (Fig. 7). Pulsed field gradient diffusion NMR experiments (41) indicate that CXCL11 is a dimer under the conditions used for the HSQC (Fig. 7B). These results and the uniform size of the cross-peaks suggest that the extra peaks are not due to the presence of both monomer and dimer but rather alternate conformations of the chemokine. This behavior is consistent with a previous report on the NMR structure of CXCL11, which was determined at low pH (4.5) and high temperature (40°C) because only these conditions gave a single species (34). Because of the complexity of the spectra under most conditions, we abandoned the chemical shift perturbation experiments, which would require challenging spectral assignments. Nevertheless, the results are intriguing and suggest some unique conformational plasticity for this chemokine, which could be related to the observation of  high affinity binding sites in the SPA receptor binding assay and in the immobilized heparin assay.
Attempts were made to determine if CXCL11 oligomerizes further in solution in the presence of heparin, as observed for CCL2 (31) and CCL27 (62). However, the addition of heparin octasaccharide caused immediate precipitation of the complex at substoichiometric concentrations of GAG to chemokine (supplemental Fig. 2). Although this behavior makes it difficult to characterize the oligomeric state of CXCL11 in the complex, similar behavior was observed for CCL2, although it did not precipitate as vigorously, and what was left in solution was tetrameric (62). Heparin octasaccharide has also been shown to stabilize oligomerization of several other CCR2 ligands as both homo-and hetero-oligomers (42). Thus, it seems likely that GAGs stabilize the CXCL11 dimer and induce higher order oligomerization.

DISCUSSION
The interactions of chemokines with cell surface GAGs, in addition to high affinity interactions with their receptors on leukocytes, play an essential role in cell recruitment. This has been demonstrated in vivo using chemokine mutants that have a compromised ability to bind GAGs in vitro but intact receptor binding; although such GAG binding-deficient mutants are generally competent to recruit cells in vitro in simple trans-well assays, they are incapable of inducing migration in vivo (7,38). The requirement for the GAG interaction is thought to reflect, in part, the need for the sequestration of chemokines on cell surfaces, where they provide directional cues for migrating cells. Indeed, a mutant of RANTES, 44 AANA 47 -RANTES, rapidly accumulates in serum when injected intraperitoneally, in contrast to the WT protein (43). Notably, this mutant was able to block the in vivo cell recruitment to the WT protein and was shown to be effective in three murine models of inflammation (thioglycollate-and ovalbumin-induced cell recruitment and MOG-induced experimental autoimmune encephalomyelitis). Similarly, a GAG-binding mutant of MCP-3/CCL7 antagonizes chemokine-mediated cell recruitment in an air pouch in vivo model and in vitro to the synovial fluid from patients with rheumatoid arthritis (38). Finally, a GAG-binding mutant of CCL2 inhibits recruitment of cells to WT CCL2 in an in vivo peritoneal recruitment assay. 4 Because there are ϳ50 ligands in the human chemokine system, it remains to be seen if GAG binding is important for the vast majority of the ligands. The chemokine family also provides an extraordinary opportunity for investigating the extent to which GAG interactions add to the specificity of their function. Finally, the ability of GAG mutants to antagonize chemokine function offers the potential for the development of novel therapeutics.
For these reasons, in this study, we set out to define the GAG binding determinants of the chemokine I-TAC/CXCL11. CXCL11 mutants were prepared and tested in several in vitro assays of GAG binding, including the commonly used method of heparin affinity chromatography in conjunction with cation exchange chromatography, and in an Epranex plate assay. These assays provide a relative quantitative measure of affinity rather than absolute binding constants, which would require a completely homogeneous GAG with a single or known number of binding sites, entities that are not readily available. However taken together, these assays defined Lys 57 , Lys 59 , and Arg 62 in the 50s region and Lys 17 as forming the dominant GAG-binding epitope. Furthermore, the protease protection assay confirmed the importance of the 50s epitope.
Lortat-Jacob et al. (44,45) describe the existence of typical GAG binding sites for each chemokine class. They state that for most CXC chemokines, the residues of the C-terminal helix together with residues connecting the first N-terminal ␤-strand with the N terminus are involved. We examined the positions of the mutated residues of the 50s mutant and Lys 17 and found that they exactly coincide with the binding site predicted (supplemental Fig. 3). Matrix metalloproteases 8, 9, and 12 process CXCL11 at both the N and C terminus to generate CXCL11(5-73), CXCL11(5-63), and CXCL11 , where processing of the amino terminus results in increased GAG binding, and C-terminal truncation results in the loss of GAG binding (46). Although one cannot exclude an indirect effect from misfolding, the results from C-terminal truncation correlate with our findings that the 50s cluster ( 57 KSKQAR 62 ) is involved in GAG binding because two of the basic residues are removed.
The ultimate test of the importance of a GAG binding motif is in vivo recruitment. For reliable interpretation of these experiments, a requirement is that the mutations do not significantly perturb the ability of the chemokine to bind and activate its receptor. As seems to be the case for many chemokines, although there is often some overlap in GAG and receptor binding sites, one can usually identify mutations that impair GAG binding without significantly interfering with receptor binding and signaling. The interaction of CCL5 with CCR5 is a good example, where complete decoupling of the receptor and GAG-binding epitopes are observed, although this is not the case for CCL5 and CCR1. CCL2 is another good example, where GAG-binding and receptor-binding epitopes can be distinguished because only a 20-fold loss in affinity for CCR2 was observed with the GAG-binding mutant 18 AA 19 -MCP-1, and the mutant showed equal efficacy to the WT chemokine in cell migration (4). Perhaps the best example is a GAG-binding mutant of CCL7, which actually showed slightly higher affinity for the receptor compared with the WT chemokine (38). Similarly, the CXCL11 50s ϩ Lys 17 mutant, which bound heparin poorly in vitro, showed only a small decrease in receptor binding, and the efficacy and potency for inducing cell migration was equivalent to the WT protein. Consequently, the reduced ability of the 50s mutant to induce cell recruitment into the peritoneal cavity establishes these residues as bona fide contributors to GAG binding. By contrast, the inability of the 10s mutant to recruit cells is due to the substantial effect of the mutations on receptor binding and activation, and by contrast, these mutated residues contribute little to GAG binding, despite conforming to a classic BBXB motif. The 40s cluster also contributes more to receptor binding/signaling and less to GAG binding compared with the 50s epitope and 50s ϩ Lys 17 . Although it remains to be determined, the ability to fully or partially decouple GAG binding from signaling may have impli-cations for whether or not chemokines can simultaneously bind to GAGs and receptors or whether the interactions are mutually exclusive. These in vivo observations confirm the results of our in vitro assays and furthermore validate the use of heparin as a suitably representative GAG for defining GAG epitopes.
An emerging theme in the chemokine field is that GAG binding epitopes are often not confined to linear epitopes like BBXB because some chemokines oligomerize on GAGs, making much more diffuse sites in the context of the quaternary structures (6). The 50s ϩ Lys 17 motif constitutes a diffuse site in the tertiary structure of CXCL11 but would present a different surface again in the context of the dimer or higher order oligomer. The principal GAG binding residues of another CXCR3 ligand, IP-10/CXCL10, are situated in the 20s and 40s loops (39), suggesting significant differences from CXCL11 and underlying specificity between these chemokines with respect to GAG binding. Although its NMR structure was solved as a monomer (47), human CXCL10 has been crystallized in three different tetrameric forms that have been suggested to be involved in GAG binding (48), and a recent structure of murine CXCL10 reveals yet a fourth tetrameric form (49). Although dimeric in solution, MCP-1/CCL2 forms tetramers, as shown by crystallography (50), and the tetramer is stabilized by the presence of GAGs (31). PF-4/ CXCL4 also crystallizes as a tetramer and forms tetramers in solution (51,52), whereas RANTES/CCL5 forms large oligomeric structures, which can be destabilized into dimers and tetramers by point mutations (53). The structure of CXCL11 was solved as a monomer by NMR because of the low pH conditions used for the sample preparation (34); however, this chemokine clearly dimerizes in solution under more physiological conditions (Fig. 7B) and, like other chemokines, such as CCL2, may oligomerize further in the presence of GAGs (supplemental Fig. 2). Reduced stabilization of oligomers by GAGs for the mutants relative to WT is the most likely explanation for the lower binding capacity of the GAG mutants in the solid phase heparin assay compared with WT in Fig. 2.
Although oligomeric structures of CXCL10 have been solved by crystallography, crystallizing CXCL11 has not been feasible. 5 The spectra shown in Fig. 7, however, provide an obvious explanation for why crystallization has been challenging; CXCL11 is conformationally heterogeneous. This heterogeneous behavior is likely to be relevant to its biological function involving different receptor and/or GAG interactions. Although most chemokines studied to date appear to adopt a single conformation based on their HSQC spectra, another example of structural promiscuity has been reported for the human chemokine lymphotactin/XCL1. Similar to CXCL11, the HSQC spectra of XCL1 are indicative of two species, and in fact, XCL1 exists in equilibrium between two entirely distinct folds: one that binds GAGs but does not bind or activate the receptor and one that has the complementary receptor binding ability but does not bind GAGs (54). CXCL11 may also be subject to dynamic structural interconversion, although it is not likely to be as extreme as lymphotactin because the CXCL11 structure is locked down by two disulfides, whereas XCL1 has only a single disulfide bond. Nevertheless, it is likely to be biologically relevant, and we are currently investigating the structural nature and functional significance of this behavior in CXCL11.
Two unusual observations made in the current study may be related to the conformational heterogeneity of CXCL11, however. The receptor binding studies revealed a consistent anomaly in at least five repetitive experiments, in that the WT protein showed a reduced binding capacity for CXCR3 compared with the mutants, as judged by the radioactivity counts at picomolar concentrations. We therefore used even lower concentrations of competitor and identified a subpicomolar interaction that represents ϳ30 -50% of the total. This interaction could involve CXCR3, another receptor, or a GAG interaction. The presence of more than one binding site could also reflect the structural heterogeneity of CXCL11, where one conformation displays a higher affinity than the other. To address this observation, we posed the question of whether GAG binding could account for the high affinity interaction, with an assay consisting of competition of radiolabeled CXCL11 to immobilized heparin, extending the concentration of unlabeled CXCL11 to femtomolar concentrations. Effectively, a very high affinity site was revealed. We then tested CCL2, CCL5, and CXCL12 in a similar assay and observed the same phenomenon for CCL5 but not for CCL2 and CXCL12, with the most pronounced effect observed for CXCL11. Because GAGs are heterogeneous, the valency of the interaction is unknown, and the molecular weights for the GAGs are average values. Therefore, it is not possible to obtain exact IC 50 values. However, there is no doubt that CXCL11 and CCL5 have unusually high affinities compared with other chemokines. Moreover it will be interesting to investigate whether the high affinity site exits for other GAGs, such as heparan sulfate, chondroitin sulfate, and dermatan sulfate, in addition to heparin.
In summary, it is clear that GAG binding plays an important role in CXCL11 function and that binding sites comparable in affinity to receptors do exist. In this study, we characterized a GAG-binding mutant of CXCL11 that consists of mutations that are largely separate from the receptor binding determinants. Although we have not tested the GAGbinding mutant as an inhibitor of cell migration in vivo or in disease models, the results described above for CCL5 and CCL7 GAG mutants suggest that one can anticipate that the 50s ϩ Lys 17 mutant will also be an effective antagonist of cell migration. The most widely appreciated chemokine-based antagonists consist of chemokines in which the N terminus is modified in some way by mutation, deletion, or extension (55,56). Non-oligomerizing chemokines have also been shown to be powerful inhibitors of cell migration in experimental autoimmune encephalomyelitis and rheumatoid arthritis (57,58). GAG binding-deficient chemokines offer a third possibility. A GAG-binding mutant of CCL5 functions at least in part by a dominant negative mechanism involving formation of heterodimers and preventing higher order oligomerization (43). As suggested in recent studies, GAGbinding mutants that retain their agonist properties may be 5 J. P. Shaw, personal communication.
particularly effective and even have broad spectrum effects due to homologous and heterologous desensitization (38,59,60). Herein we have defined a GAG-binding mutant of CXCL11 that retains its agonist activity and can be used in future studies to test its efficacy in the many inflammatory diseases where CXCR3 and its ligands have been implicated (61).