Interactions of the Chemokine CCL5/RANTES with Medium Sized Chondroitin Sulfate Ligands

Summary

Interactions of the chemokine CCL5 (RANTES) with glycosaminoglycans (GAGs) are crucial to the CCL5-mediated inflammation process. However, structural information on interactions between CCL5 and longer GAG fragments is lacking. In this study, the interactions between oligosaccharides derived from chondroitin sulfate and a dimeric variant of CCL5 were investigated using solution NMR. The data indicate that, in addition to the BBXB motif in the 40s loop, GAGs also contact residues in the N-loop in a manner similar to interactions between chemokine and the receptor N-terminus, and leading to possible stabilization of the dimer. Using TEMPO-tagged hexasaccharides, the binding orientation of the hexasaccharides was shown to be highly dependent on the sulfation pattern of the GalNAc groups. Finally, a model of the CCL5 dimer complexed to CS hexasaccharides was constructed using paramagnetic relaxation enhancement and intra- and inter-molecular NOEs constraints.

Introduction

CCL5 (RANTES) is a pro-inflammatory chemokine that plays an important role in controlling the migration and activation of leukocytes. It functions by binding receptors from the chemokine family of G Protein-Coupled Receptors (GPCRs) on the leukocytes. But almost as important as CCL5’s interaction with GPCRs are its interactions with cell surface glycosaminoglycans (GAGs), a class of linear, sulfated polysaccharides that often act as coreceptors for extracellular signaling molecules (Bishop et al., 2007; Lindahl, 1999; Mulloy and Linhardt, 2001).

Although GAGs are composed of only two types of monosaccharides arranged in alternating fashion, the structures of GAGs are among the most complex in biology. This is because the monosaccharides in GAGs are often modified through sulfation and epimerization, which are carried out in a template-free fashion. GAGs are categorized by the type of monosaccharides in their unmodified backbones: heparin (Hp) and heparan sulfate (HS) are built from a backbone containing N-acetyl glucosamine (GlcNAc) whereas chondroitin sulfate (CS) is built from a backbone containing N-acetyl galactosamine (GalNAc). Both glucuronic acid (GlcA) and its C5 epimer iduronic acid can be found in Hp and HS, but CS contains mostly GlcA. GlcNAcs in Hp and HS can undergo N- and 6-O-sulfations whereas GalNAc in CS can only undergo 4- and 6-O-sulfations. Iduronic acids in these GAGs can also be O-sulfated.

CCL5 is known to interact with a number of different GAGs carrying molecules, including proteoglycans that typically contain HS or CS, but not the more highly sulfated Hp. These CCL5-GAG interactions facilitate the formation of chemotactic gradients on endothelial cell surfaces and in the extracellular matrix, and effectively provide directional cues for leukocytes to follow. Similar to other types of protein-GAG interactions, CCL5’s affinity for GAGs is mediated by strong electrostatic interactions. Disruption of such interactions has been shown to effectively attenuate the migration of leukocytes in vivo (Appay et al., 1999; Hoogewerf et al., 1997; Proudfoot et al., 2003), suggesting that inhibiting CCL5-GAG interactions might prove to be a viable anti-inflammatory strategy. (Sheng et al., 2013). However, rational design of modulators of such interactions must be built on a clear understanding of the biophysical interactions among components in these protein-carbohydrate complexes, yet obtaining detailed information on the structure and dynamics of protein-GAG complexes is challenging. In particular, out of the more than 100,000 structures deposited in the PDB data bank, only ~ 90 are structures of protein-GAG complexes. In addition, these structures involve only ~ 30 unique GAG-binding proteins, the majority of which were complexed to heparin-derived oligosaccharides.

Early work by two groups showed that mutations of a cluster of basic amino acids in the 40s loop of CCL5 (R44, K45, R47) dramatically impaired the affinity of CCL5 for Hp, establishing these residues as the primary epitope for GAG-binding (Martin et al., 2001; Proudfoot et al., 2003). Subsequent crystal structures of wild type (WT) CCL5 bound to a Hp disaccharides validated the biochemical experiments and provided the first glimpse of CCL5-GAG interactions at atomic resolution (Shaw et al., 2004). However, no high resolution structural information on the interactions of CCL5 with medium or long fragments of Hp, HS, or CS exists to date, although low resolution models were proposed based on surface plasmon resonance (SPR) -derived binding kinetics between Hp and an N-terminally truncated form of CCL5, which is monomeric in solution.(Vivès et al., 2002). Information on interactions of oligomeric CCL5 with GAGs longer than disaccharides, especially those of HS and CS, is important, given the fact that cell surface GAG chains typically contain at least 25 disaccharide units, and that CCL5 exists as a oligomer in solution (Appay et al., 1999). Indeed, chemokine oligomerization and GAG-binding are synergistic. Earlier study by Hoogewerf et al. showed GAGs easily triggered CCL5 oligomerization both in vitro and on cell surfaces (Hoogewerf et al., 1997). Moreover, understanding the interactions of larger GAGs with chemokines will also aid in the development of potential anti-inflammatory agents, including GAG mimetics (Sheng et al., 2013).

Here we describe the interaction of CCL5 with homogeneous hexsaccharides derived from chondroitin sulfate (CS). CS is commonly found on the surface of mammalian cells and in the extracellular matrix (ECM). Although heparan sulfate is often considered the major GAG species that interacts with chemokines, both CS and dermatan sulfate (DS) also have submicromolar affinities for CCL5 (Mizumoto et al., 2013), establishing CS as a relevant target for investigation. In fact, CS has been shown to be important for the CCL5/CCR5-mediated induction of T cell apoptosis (Murooka et al., 2006), and for CCL5-induced monocyte arrest (von Hundelshausen et al., 2005). The sulfation pattern of CS and DS are also simpler than heparan sulfate (HS) and Hp, making them more convenient materials for the isolation of oligomers with defined structure. Most importantly, the complexes formed when CCL5 binds CS are more soluble than complexes with other types of GAGs, an essential characteristic for structural investigations of GAG-CCL5 interactions in solution. Another technical complication that has hindered study of CCL5 is the protein’s high tendency to oligomerize (Appay et al., 1999), which severely degrades the quality of the NMR signal. However, by using a dimeric E66S mutant of CCL5, we were able to define the interaction interface for the CCL5-CS complex using a novel series of NMR experiments. In particular, we found that, in addition to the previously identified BBXB cluster in the 40s loop, residues around R17 in the "N-loop" (residues 12 to 20) and at the N-terminus of CCL5 have extensive interactions with CS dp6 fragments. Moreover, the GAG interactions with the N-loop and N-terminus involve segments from different monomers in the dimer, suggesting that either the GAG stabilizes the dimer and/or that the CCL5 dimer enhances GAG binding. Using novel TEMPO-labeled GAG fragments, we showed that the interactions of CCL5 with these GAGs are dynamic and that a combination of interaction modes can coexist. Furthermore, the sulfation pattern of CS significantly influences the interaction, suggesting that specific sulfation patterns play a role in regulating CCL5-GAG interactions.

Results

The binding epitopes of CS dp6 on CCL5

CCL5 is an avid GAG-binding protein and its interaction with GAGs usually results in precipitation of the complex if the protein concentration is sufficiently high. For example, 50 µM E66S CCL5 precipitates readily in the presence of one molar equivalent of Hp dp6 (degree of polymerization 6, i.e. hexasaccharide) or DS dp6. In both cases, no changes in chemical shifts or relative intensities of the remaining resonances were seen, an observation consistent with aggregated complexes too large to be detected by NMR and slow ligand exchange between soluble and precipitated CCL5. However, we found that complexes of CCL5 E66S with CS dp6 are soluble at pH 4.5 even when the protein concentration is around 200 µM and no salt is present in the buffer. As mentioned previously, CS is important for the function of CCL5 (Murooka et al., 2006; von Hundelshausen et al., 2005), and structural information on CCL5-CS complexes can shed light on its interaction with other GAGs such as HS as well as revealing insights into the specificity of these interactions.

The E66S mutant of CCL5 was chosen for study as it is largely dimeric compared to WT CCL5, which forms large but apparently ordered polymers (Wang et al., 2011). Furthermore, the E66S mutation is not expected to affect the GAG interaction because of its location in the C-terminal helix, and a CCL5 dimer should be sufficient to capture the essence of GAG interactions with larger oligomers since the dimers are believed to be the building blocks of larger oligomers (Wang et al., 2011). For these studies, two CS dp6 fragments were obtained by partial depolymerization of CS with hyaluronidase. One fragment contained only 4-O-sulfated GalNAc (CS444) while the other fragment contained a 6-O-sulfated GalNAc at the non-reducing end (CS644) (Figure 1). Structures of both fragments have been verified using NMR and mass spectrometry. To prevent chemical shift doubling due to α/β isomerization we reduced the residue at the reducing end, which is commonly done in NMR studies of GAG structures. The affinities of the CS fragments for E66S were measured by NMR titration (Figure 2A), which showed that the protein is completely saturated after the CS444-to-CCL5 ration has reached 1:1. This implies the Kd of the interaction is well below 1 µM for CS444 (Table 1 & Figure S1). These residues also demonstrated intermediate exchange as manifested in the broadening of many signals when the CS444-to-CCL5 ratio was 1:2. This observation is consistent with the relatively strong interaction between CCL5 and CS444. To accurately estimate the value of the Kd, isothermal titration calorimery (ITC) was performed. The ITC results showed the interaction is driven by enthalpic energy and Kd of the interaction is approximately 0.25 ± 0.07 µM (Figure S2). As expected for interactions relying mainly on electrostatic interactions, the affinities of the CS dp6 for CCL5 are highly dependent on ionic strength. In particular, increasing the ionic strength from 20 mM to 150 mM NaCl resulted in an increase in Kd from <1 µM to ~ 1 mM. Chemical shift mapping of the changes induced by CS444 showed a number of residues in CCL5 to be highly perturbed (Figure 2B). Specifically, residues in the 40s loop, at the N-terminus and in the N-loop showed chemical shift changes that were significantly higher than average. Residue S5 at the N-terminus also broadened beyond detection upon the addition of CS fragments, indicating changes in their dynamics in the presence of the GAGs. The perturbation of the N-loop residues is similar to that reported for the chemokine CCL4 (MIP-1β) (McCornack et al., 2003), but has not been observed in previous CCL5 studies (Proudfoot et al., 2001; Shaw et al., 2004). Perturbation of the N-terminus has also not been reported for either CCL4 or CCL5. In all cases, the binding curves fit best with a model that assumes a stoichiometry of one CS444 per CCL5 monomer (Figures S1 and S2). When CS444 was used as the ligand, all chemical shift movements remained linear and correlated throughout titration, and Kds obtained using residues in different regions were mutually consistent, indicating the sites either act cooperatively or have identical Kds. However, this was not the case for the titration of E66S with CS644. Although CS644 produced similar patterns of chemical shifts changes in backbone amides as CS444, the magnitudes of the change induced by CS644 were considerably smaller compared to CS444. Furthermore, the direction of CS644 induced chemical shift changes observed in some N-loop residues (R17 & L19) changed significantly after the ratio of CS644:CCL5 exceeded 1:1, implying that, once the concentration of CS644 exceed that of CCL5, multiple CS644 ligands may bind to CCL5, or that multiple conformations exist at higher ligand concentrations. Using the chemical shift of three residues showing changes with a single mode of interations, the global Kd of the CCL5’s interaction with CS644 was determined to be approximately 2 µM We also titrated CCL5 with a mixture of CS dp10 fragments. The patterns of chemical shift changes observed in this titration are similar to CS dp6 ligands, indicating that the interactions between CCL5 and these fragments are similar. ¹⁵N-edited HSQCs of the CS dp10 titration are shown in Figure S3 and binding curves of specific residues from these titrations can be found in Figure S1.

Figure 1.

Structure of CS444 and CS644. The difference in sulfation position at the non-reducing end GalNAc is highlighted in bold.

Figure 2.

A) ¹⁵N-HSQCs of E66S-CCL5 in the presence of different concentrations of CS444. The concentration of protein is 40 µM and the concentrations of CS444 are 0, 20, 40, 80, 120, and 160 µM. The colors of the contours of HSQCs represent 0 (dark green), 20 (brown), 40 (yellow), 80 (pink), 120 (light green), 160 (light blue). Detailed sections showing movements of representative residues from the N-loop and the 40s loop are shown on the right. B) Scaled and combined chemical shift changes of each residue. The combined chemical shift changes were calculated using the formula $\Delta {\delta }_S=\sqrt{\Delta {\delta }_{H,HZ}^2+{(1.7\Delta {\delta }_{N,HZ})}^2}$. Positions of S5 and R45, which broadened beyond detection during the titration, are indicated by stars. Other residues missing from the graph are prolines.

Table 1.

Dissociation constants of GAG-CCL5 interactions as determined by NMR.

	N-terminus (S1)	N-loop (F12)	40s loop (N46)	Global
CS444a	< 1 µM	< 1 µM	< 1 µM	< 1 µM
CS644	N/A (non-linear shift)	N/A (non-linear shift)	4 ± 1 µM	2 ± 2 µMb
CS dp10 (100 mM NaCl)	7 ± 1 µM	8 ± 2 µM	8 ± 1 µM	8 ± 1 µM

^aKd was determined to be ~ 0.25 ± 0.07 µM by ITC (Figure S2)

^bFit included information from residues K46, H23, N46 and R47. See figure S8 for additional binding curves.

CCL5 dimerization and GAG interaction

To investigate the role of CCL5 dimerization, we capitalized on the fact that low concentrations (< 50 µM) of E66S CCL5 exists as discrete populations of monomer and dimers at 40 °C (Schnur et al., 2013) and investigated whether CS444 is capable of promoting dimerization of E66S. In the absence of CS444, the ratio of dimer-to-monomer in a 30 µM E66S CCL5 sample is approximately 1:1 at 40 °C (Figure S4). However, addition of 2 mole equivalents of CS444 collapsed the peaks to a single population, which possessed line width and transverse relaxation rates consistent with an E66S dimer (Figure 3 & S3). The data indicate that CS444 stabilizes the dimeric form of CCL5. We also investigated interactions of CS444 with a monomeric variant, P2 RANTES (Jin et al., 2010), which contains a mutated N-terminus that effectively destabilizes the core dimer of the WT CCL5 polymer. Our data indicate that P2-RANTES’s affinity for CS444 is weaker than the E66S CCL5 dimer (Kd ~ 5 µM, Figure S5) with residues in the 40s loop as well as the N-loop of P2 RANTES showing the most significant changes in chemical shifts. This indicates that the monomeric form of RANTES has similar GAG-interactions as dimeric RANTES. Surprisingly, P2-RANTES also showed the same tendency for GAG-induced aggregation as E66S CCL5. This is consistent with earlier CCL5-GAG interaction studies conducted with N-terminally truncated CCL5 (Vivès et al., 2002). One important difference maps to residues in the N-loop, which show additional chemical shift perturbations similar to the changes observed in the titration of E66S CCL5 with CS644. Specifically, the direction of chemical shift changes for these residues reversed at higher CS444-to-P2-RANTES ratios.

Figure 3.

Sections of ¹⁵N-HSQCs of 30 µM E66S-CCL5 at 40°C. Black: E66S-CCL5 without CS444, both monomer and dimer species are visible. Red: E66S-CCL5 with two molar equivalents of CS444. The assignments of signals representing monomeric and dimeric species are made based on (Schnur et al., 2013) and (Duma et al., 2007).

CCL5-GAG interactions investigated using TEMPO-tagged CS444 and CS644

GAGs are known to bind to proteins in multiple conformations (Carter et al., 2005; Tan et al., 2008). To gain insight into the structure and dynamics of the CCL5-GAG interactions, we probed E66S CCL5 with paramagnetically tagged CS dp6 fragments. To functionalize CS fragments with paramagnetic TEMPO, we covalently linked 4-amino-TEMPO to the reducing ends of these fragments by reductive amination. Paramagnetic moieties increase relaxation rates for nearby NMR active nuclei in a distance dependent fashion leading to paramagnetic relaxation enhancement (PRE). Because these electron-nuclear interactions are long range, they have proven to be effective at revealing dynamic interactions between biomacromolecules (Clore et al., 2007; Tang et al., 2008). The ¹H relaxation rates of the CCL5 backbone amide hydrogens before and after the reduction of the radical on the ligand were measured (Iwahara et al., 2007) and relaxation due to paramagnetic effects are shown in Figure 4. It is clear that CS444 produced the most relaxation around residue A22. This is consistent with the reducing end of the CS444 being close to the BBXB cluster in the 40s loop. Compared with CS444, CS644 was able to produce relaxation in a larger number of residues, including Y3, A16 and R21. These residues span the N-terminus, the N-loop and the 20s loop. These data indicate that the reducing end of CS644 is as likely to be near the N-loop and N-terminus as the 20s loop, making the binding orientation of CS644 far more heterogeneous than that of CS444. The differences seen in CS444 and CS644 PRE patterns demonstrate that CCL5’s interaction with GAGs is highly sulfation position dependent. In particular, a change in a single sulfation position was sufficient to change binding orientation of the GAG fragments. Furthermore, the experiments confirmed that the GAG binding site on CCL5 is not limited to the BBXB motif in the 40s loop. The N-loop and N-terminal residues, especially A16 and R17, are also greatly perturbed, and should constitute a second binding epitope for long GAG fragments. Interestingly, both Y27 and Y29 also showed TEMPO induced relaxation, with the PRE produced by CS644 being particularly strong. The most plausible explanation is that there exists another binding mode in which the non-reducing end of the ligand is bound to the 40s loop and K25 while the reducing end is close to the amide protons of Y27 and Y29. It is also worth noting that residues in the 30s loop, close to Y27 and Y29 were also identified as a possible GAG-binding site in the crystal structure of WT CCL5 with Hp disaccharides (Figure S6). (Shaw et al., 2004) However, those interactions could have been the result of crystal packing that brought the BBXB motif of a neighboring monomer into the vicinity of these residues. The biological significance of these minor GAG-binding motifs has yet to be investigated.

Figure 4.

Quantitative measurement of paramagnetic TEMPO-induced transverse relaxation (R_2,PRE) A. Graph and surface plot of amide proton R_2,PRE produced by TEMPO-labeled CS444. B. Graph and surface plot of amide proton R_2,PRE produced by TEMPO-labeled CS644. Residues Y14, A16 and R21 disappeared upon addition of CS644-TEMPO, but reappeared upon its reduction. They are given the same magnitude of R2,PRE as residue Y3, which showed the biggest change.

Intermolecular contacts between CCL5 and CS444

To further examine the interaction of CS444 with CCL5 at atomic resolution, we carried out several NMR experiments using a number of specifically isotopically labeled samples. Table S1 summarizes the intermolecular NOEs identified using these experiments. Identification of contacts between proteins and GAGs are known to be challenging for solution NMR due to lack of non-exchangeable protons in the interaction interface and the transient nature of many of the interactions. To overcome these difficulties, we utilized a number of approaches to improve the detection sensitivity of the contacts. Our choice of CS444 as the ligand also simplified data analysis since the PRE experiments had already revealed the ligand as having a relatively uniform binding orientation. Traditional ¹³C/¹⁵N filtered/edited NOESY experiments for determining intermolecular contacts showed that K45 from the BBXB cluster and L19 of the N-loop are in the vicinity of the GAG. Similar experiments focused on aromatic protons revealed that Y3 of CCL5 also contacts the GAG protons (Figure 5). Using chemical shift assignments for the free CS444 ligand, we were able to deduce the possible identity of CS444 protons involved in the contacts. Specifically, L19 methyl protons were identified unambiguously as contacting H2 of either GlcA1 or GlcA2 (the naming of the monosaccharides in CS444 follow those indicated in Figure 1). I15 Hδ1 protons are close to H5s of GlcA and GalNAc residues. K45 Hε and Hδ are close to GalNAc H2, H3 and H5 protons; and Y3 aromatic protons are also contacting GalNAc H2 and H3 protons. To improve the detection sensitivity of intermolecular contacts, we produced deuterated proteins containing specifically protonated methyl groups. Using this selectively protonated CCL5, we were able to detect contacts between GAG and CCL5 using conventional NOESY experiments without the need of signal degrading isotopic filters. Figure 5B shows ¹³C-edited HSQC-NOESY spectra of perdeuterated CCL5 containing specifically protonated ILV methyl groups in the presence and absence of CS444. We were able to confirm unambiguously that both I15 Hδ and L19 Hδ methyl groups are involved in binding GAGs. This is consistent with the results of the conventional ¹³C/¹⁵N filtered/edited experiments.

Figure 5.

A) ¹³C-filtered/edited NOESY of 200 µM ¹³C-labeled E66S-CCL5 in the presence of one molar equivalent of CS444. CCL5 proton chemical shifts correspond to the X-coordinates of the crosspeaks while ligand proton chemical shifts correspond to the Y-coordinates of the crosspeaks. B) HSQC-NOESYs of 200 µM deuterated E66S CCL5 with ILV methyl protonation in the presence (black) and absence (red) of one molar equivalent of CS444. Two new cross peaks can be seen in the spectrum collected in the presence of the ligand. They correspond to contacts between methyl groups on I15 and L19.

To elucidate the identity of other basic amino acids involved in binding CS444, we used perdeuterated CCL5 and unlabeled CS444 to carry out a cross saturation experiment, a technique that has been successfully used to identify carbohydrate binding residues in a previous study (Sakakura et al., 2008). The data revealed that the backbone amide proton of S1, Y3 and R17 are significantly perturbed by saturation of CS444 protons (Figure 6). This is once again consistent with chemical shift mapping and PRE experiments performed with CS444. Unfortunately, we weren’t able to measure the extent of saturation on signals from residues close to the BBXB motif in the 40s loop because of signal broadening in the presence of CS444.

Figure 6.

Reference ¹⁵N-HSQC (left) and saturation transfer difference ¹⁵N-HSQC (right) spectra of perdeuterated E66S-CCL5. Three residues showed the most prominent difference.

Finally, to identify CS444 protons involved in binding CCL5, we used traditional saturation transfer difference (STD) experiments at a GAG-to-protein ratio of 20:1. These experiments clearly indicated that both GalNAc and GlcA protons are involved in binding CCL5 (Figure 7). Specifically, H5 or H6 protons from GlcA1/2, GalNAc1/2 appeared to interact strongly with CCL5. H4, H3 and H2 protons from GlcA1 and GlcA2 also gave significant STD effects, as did H3 from GalNAc1/2 and GalNAc methyl groups. However, moving the saturation frequency from 0.97 to 1.3 ppm significantly increased intensities of the methyl STD signal, indicating STD signals for the GalNAc methyl groups maybe artifacts arising from the closeness of the methyl proton chemical shifts to the saturation frequency of 0.97 ppm. Other than the conventional STD experiments, we also performed an HSQC-NOESY experiment on a sample containing 50 µM ¹³C/¹⁵N-labeled CCL5 and 1 mM CS444 but removing the ¹³C/¹⁵N decoupling during acquisition in order to distinguish between intermolecular and intramolecular NOEs. The experiment benefited from both the larger NOE enhancement of the bound ligand and the fast exchange rate of the complex which amplifyed the effect of intermolecular NOE enhancement involving a much larger fraction of ligands. The spectrum produced NOE cross peaks correlating H5s of GlcA1/2 and GalNAc1/2 with Hδ methyl protons of I15 and Hδ protons of Y3. When considered in combination with the PRE data, which showed the reducing end of CS444 to be close to the 20s loop, the GalNAc1 at the non-reducing end is most likely interacting with I15 of CCL5.

Figure 7.

A. STD 1D of 1 mM CS444 with 50 µM ¹³C labeled E66S-CCL5. Possible assignments of the peaks are indicated. B. sections of F3-coupled HSQC-NOESY collected on the same sample. Cross peaks between ligand protons and protein protons can be seen.

Model of the CCL5 with CS444 complex

Using these intermolecular contacts, we attempted to construct a model of the CS444 ligands bound to a CCL5 dimer using the crystal structure of the CCL5 dimer bound to Hp disaccharides as a template. The experimental data suggest that a single GAG ligand interfaces with N-loop residues from one monomer and the N-terminus of the neighboring monomer. This scenario is more likely than the alternative possibility in which the GAG interacts with N-loop and N-terminal residues from the same CCL5 monomer. This is because the alternative scenario would require the dimer of CCL5 to dissociate in order to accommodate the GAG in this manner. However, NMR and other evidence show that the CCL5 dimer is intact and stabilized by the presence of GAGs. Figure 8A shows a representative frame from our ensemble of models constructed using NMR constraints and the MD simulation program AMBER. Details of the modeling protocol are outlined in the experimental methods section. In our model, the non-reducing end GlcA1 is stacked against the aromatic ring of Y3 from the neighboring monomer while the H6 of GalNAc1 interacts with I15. Although no explicit electrostatic constraints were specified between basic residues and acidic functional groups on CS444, the majority of intermolecular hydrogen bonds are formed between these moieties. In particular, R17 was observed to form hydrogen bonds with both the sulfate group of GalNAc1 as well as the carboxyl group of GlcA1. These interactions were observed in 50% of the 50 frames deemed as having the lowest ligand-binding ΔG by the MM/PBSA module of AMBER. R47 of the BBXB motif was observed to form a hydrogen bonding interaction mainly with the carboxyl group of GlcA2 (32% of the frames) and occasionally associates with sulfates of GalNAc2 (16% of the frames). K45 was also observed to interact with the sulfate group of GalNAc2 (15% of the frames), and the side chain of R44 was stabilized by ionic interactions with GlcA3 (20% of the frames). Other hydrogen-bonding interactions seen in the simulation include that between the side chain of T43 and atoms in GalNAc2 (35% of the frames). These interactions are illustrated in Figure 8C. Another binding mode deemed less stable by MMPBSA is also found in a group of coordinates. In this mode, R47 forms strong interactions with the sulfate group of GalNAc2 (65 % of frames), this forced K45 to interact mainly with carboxyl of GlcA3 (25 % of the frames) and R44 was not in position to form hydrogen bond with CS444. It is possible that the two modes may coexist in solution.

Figure 8.

A) A representative frame from the most energetically favored models of the CCL5-CS444 complex. A. The ribbon representation of the complex with side chains of selective amino acids in grey. The CS444 ligand is shown in the stick representation with the non-reducing end and the reducing end sugar labeled GlcA1 and GalNAc3, respectively. B) Surface representation of the same structure with residues known to contact CS444 colored in blue. The protein is in the same orientation as A. C) Schematic illustration of the intermolecular hydrogen bond contacts observed between E66S CCL5 and CS444. Curved arrows indicate contacts observed between CCL5 residues and functional groups on CS444 in the simulation. The percentages indicate the fraction of 50 most energetically favored frames containing the contact.

Contribution of R17 to the affinity of CCL5 for CS

Because residue R17 of CCL5 showed significant chemical shift perturbations by CS444, and was significantly cross-saturated in its presence, we were interested in determining the thermodynamic contribution of R17 to the affinity of CCL5 for CS444 and therefore made a R17G mutant of E66S. Surprisingly, this double mutant (E66S-R17G) possessed a similar affinity for CS444 when compared to the E66S mutant. The direction and magnitude of the chemical shift migrations during the titration of CS444 with E66S-R17G also largely coincided with that of CS444 with E66S. However, even though the affinity of E66S-R17G for CS644 is similar to E66S, the double mutant demonstrated considerably less signs of multiple ligand binding modes at high ligand concentrations. In particular, the movements of the NMR signals representing residues L19 and G17 did not behave in a multimodal manner at high ligand concentrations, indicating the N-loop’s ability to act as a weaker secondary epitope for GAGs may have been reduced in this case (Figure S7).

Discussion

GAG interactions with CCL5 are known to be crucial to the biological activity of the chemokine (Proudfoot et al., 2001; Proudfoot et al., 2003; Sutton et al., 2007). Specifically, binding of CCL5 to GAGs allows a haptotactic gradient of CCL5 to form on the endothelial surface, which is essential to activation of leukocytes (Hoogewerf et al., 1997; Proudfoot et al., 2003). However, the structural details of CCL5 interactions with medium to long chain GAGs are unknown. Previous results from a crystallographic study of CCL5 with Hp disaccharides identified the BBXB motif as a major GAG-binding epitope (Shaw et al., 2004). However, the crystal structure did not show contacts between the Hp disaccharides and the N-loop residues (Shaw et al., 2004). In this study, we probed the interaction of CCL5 with longer CS ligands using a dimeric form of CCL5 (E66S) that also contains an extra N-terminal methionine, although neither modification should perturb CCL5-GAG interactions significantly since both are distant from known GAG-binding epitopes. Our results showed that, besides the BBXB motif at the 40s loop, CCL5 dimers possesses another GAG-binding epitope centered at the N-loop of the protein, specifically residues R17, L19 and I15. The fact that the R17G mutation did not affect the affinity of CCL5 for CS ligands suggests that the BBXB motif is the dominant epitope in the interaction of CCL5 with GAGs. However, opportunities for electrostatic and van der Waals interactions provided by R17, L19 and I15 may be needed to accommodate residues of longer GAGs.

Identification of the N-loop as a secondary GAG-binding epitope is highly significant. Numerous mutagenesis studies of CCL5 and other chemokines have shown that the N-loop, and R17 in particular, is an important epitope for receptor binding and activation (Pakianathan et al., 1997). The recently published structure of CXCR4 bound to vMIP-II (Qin et al., 2015) allowed the interactions of the N-terminus of the receptors with chemokines to be modeled at atomic resolution. Figure 9A shows the crystal structure of vMIP-II bound to the N-terminus of CXCR4. In the structure, the N-loop basic amino acids can be seen to participate in electrostatic interactions with residue D22 in the CXCR4 N-terminus while sulfotyrosine sY21 interacted with the basic residues in the 40s loop. Homology model of vMIP-II bound to the CCR5 N-terminus constructed using the crystal structure of the CXCR4-vMIP-II complex reveals similar interactions between the chemokine and the receptor (Figure 9B). Specifically, sulfated tyrosines in the CCR5 N-terminus are in favorable positions to interact with basic amino acids in both the N-loop and the 40s loop through electrostatic forces. This result is consistent with previous in vitro studies that utilized sulfated peptides derived from the N-terminus of CCR5, which showed residues in both the N-loop (including R17) and the BBXB motif were perturbed by the sulfated peptide. In fact, the chemical shift perturbations that were observed for the N-loop in these studies were larger than those observed for the residues in the 40s loop (Schnur et al., 2013). Comparisons of the vMIP-II-CXCR4 complex structure and the CCR5 N-terminus-bound vMIP-II model with the current CS-bound CCL5 show the models share a good deal of similarity (Figure 9). In particular, both GAG and receptor N-termini utilize electrostatic interactions with the same basic amino acid epitopes in the N-loop and the 40s loop to enhance binding affinity. While it is known that the GAG and receptor binding sites of chemokines overlap, at least partly, explaining the ability of GAGs to inhibit receptor binding and activation in both CC and CXC chemokines (Jen and Leary, 2010; Proudfoot et al., 2003; Ziarek et al., 2013), our data provides additional structural insight for the reason. They also provide further confirmation for the postulate that GAG’s role in promoting leukocyte activation may lie more in their creation of the chemokine gradient necessary to guide the leukocytes than providing direct interactions needed for receptor activation. The knowledge that GAG-binding epitopes of CCL5 overlap that of the receptor-binding epitope also creates new opportunities for designing GAG-based inhibitor of inflammation. GAG hexasaccharides’ affinity for CCL5 is on par with that of the isolated sulfated CCR5 N-terminus peptide (Kd ~ 1 to 50 µM) (Duma et al., 2007; Schnur et al., 2013). But, at physiological concentrations of chemokines, these hexasaccharides do not induce CCL5 oligomerization at all. This implies they have the potential of inhibiting the initial CCL5-CCR5 interaction without inducing oligomerization. This property makes them good candidates as inhibitors of the CCL5-CCR5 interaction. Recent work by Hsieh and co-workers demonstrated the potential of GAG mimetics in preventing CCL5-mediated inflammation.(Sheng et al., 2013) Results in this study can potentially improve specificity and potency of such a strategy.

Figure 9.

A) Structure of vMIP-II (surface) bound to N-terminus of CXCR4 (ball-and-stick). B) Model of vMIP-II bound to the CCR5 N-terminus constructed using the crystal structure of the CXCR4-vMIP-II. Basic residues involved in contacting the CCR5 N-terminus are colored in blue and labeled. B) Model of CS444-bound CCL5 determined in this study. Basic residues involved in binding GAGs are colored in blue. Panels A & B are adapted from Qin et al. (Qin et al., 2015)

Furthermore, Y3 at the N-terminus is also involved in binding CS GAGs; this interaction may explain the ability of short GAGs to stabilize the CCL5 dimer. Stabilization of the dimer structure also suggests that native dimers are the building block of GAG-induced CCL5 oligomers. This is consistent with mass spectrometry data we previously produced with WT CCL5, which showed that oligomers of CCL5 found in the absence of GAGs are composed of multiples of dimer units (Wang et al., 2011). In addition, studies using covalently linked CC chemokine dimers have shown that N-terminal cross-linked dimers are incapable of activating the receptor while dimeric CXC chemokines are known to retain their ability to bind and activate the receptor (Jin et al., 2007; Ravindran et al., 2013; Tan et al., 2012) in some cases, but inhibit chemotaxis in others, which indicate some CXC chemokines only retain the ability to bind, but not activate, the receptor (Veldkamp et al., 2008). Our data indicate that CS GAGs can stabilize CCL5 dimers by acting as an interface between the N-loops and N-termini of the monomers, providing an additional mechanism to inhibit receptor activation by further reducing the concentration of monomeric chemokines.

It should be noted that although GAG-bound dimeric or monomeric CCL5 does not appear to be capable of binding the receptor, GAG’s interactions with CCL5 oligomers may be different. These differences in GAG interactions may leave the N-loop epitope free to interact with the receptor. Given the large number of receptor binding sites in the oligomer, their affinity for the receptor may be substantial. In addition, GAG does not sequester the N-loop and N-terminus of all chemokines. In particular, a recent study by Ziarek et al. showed that although Hp can block CXCL12’s interaction with the receptor, the chemokine’s N-terminus did not bind soluble GAGs (Ziarek et al., 2013).

The CCL5 dimer-GAG interaction model proposed here appears to be distinct from the model proposed by Vives et al. using SPR-derived kinetic data (Vivès et al., 2002). In particular, the involvement of N-loop and N-terminus in GAG-binding was not predicted by the SPR model. However, the two models are not mutually exclusive. It is conceivable that a long piece of GAG maybe of sufficient length to cover the N-loops and the 40s loops in both monomers, thus creating a binding model that combines the feature of both the SPR model and model proposed here. Although the monomeric forms of CCL5 appears to be capable of binding GAGs in vitro in this and other studies (Vivès et al., 2002), in vivo conditions maybe sufficiently different to render this irrelevant. Specifically, monomeric forms of CCL5 may not exist in large quantities in vivo since the dimerization constant of CCL5 at physiological pH is in the nM range (Rek et al., 2009), far lower than the projected concentration of CCL5 near the endothelial surface during inflammation. Because the high stability of the CCL5 dimer under these conditions and the protein’s tendency to form even higher oligomers, the interactions of the dimer and higher order oligomers of CCL5 with GAGs should be more important to GAG-induced formation of the CCL5 haptotactic gradient.

Finally, by demonstrating unambiguously that the reducing end of CS444 resides close to the BBXB motif, specific PRE perturbation experiments played a crucial role in determining our model structure of the CS444-bound CCL5 dimer. The fact that changes in the sulfation position of one GalNAc changes the binding orientation of the GAG fragment so dramatically is testament to the specificity of CCL5-GAG interactions for certain sulfation patterns. The main reason for the sensitivity of CCL5 to sulfation position may be the preference of the BBXB motif for 6-O-sulfated GalNAcs compared to 4-O-sulfated GalNAcs. This hypothesis is based on the observation that CSC, which contains mostly 6-O-sulfated GalNAc, is a much better inhibitor of CCL5 binding to Hp than 4-O-sulfated CS, which contains mostly 4-O-sulfated GalNAc.(Martin et al., 2001) Because of the bias of the 40's loop for the 6-O-sulfated GalNAc in CS644, the non-reducing end (where the 6-O-sulfation is located) becomes associated with the 40s loop, thus the reducing end is more likely to be located at the N-loop. Because heterogeneous protein-ligand interactions are known to produce small chemical shift perturbations (Guan et al., 2014), the small chemical shift changes of the N-loop and N-terminal residues observed in the titration of CS644 may also be a manifestation of the heterogeneity in binding orientations of CS644. The change in chemical shift migration direction as the ratio of CS644-to-CCL5 goes beyond one maybe a sign of multiple ligand binding involving the N-loop. Another possibility is that the heterogeneity in binding orientations seen with CS644 may produce CS644-CCL5 complexes with two CS644 ligands bound to the same CCL5 dimer, but adopting different orientations when CS644 concentrations are high.

Experimental Procedures

Expression and purification of CCL5

Expression and purification of E66S CCL5 was conducted according to established protocols (Czaplewski et al., 1999). Specifically, the ORF mature CCL5 with the E66S mutation was cloned into pET23a and transformed into E. coli BL21(DE3)pLysS. The protein produced differs from the mature form in the presence of an N-terminal methionine as well as the mutation of E66S. The transformed cells were grown in M9 minimal media at 37 °C until an OD₆₀₀ of 0.5 was reached. The incubation temperature was then lowered to 28 °C and the culture was induced with 0.5 mM IPTG for 18 hours. The expressed protein was found in inclusion bodies after cells were harvested and these were lysed with sonication. To refold the protein, the inclusion bodies were solubilized with 0.1 M Tris, pH 8 and 6 M GnHCl. The solubilized inclusion body material was further purified using GE life sciences superdex 200 size exclusion chromatography column to separate the large molecular weight contaminants from medium molecular weight CCL5. The fractions containing the medium molecular weight CCL5 were then refolded through a fast dilution protocol in which CCL5 in 6 M GnHCl was added dropwise to a refolding buffer consisting of 0.1 M Tris pH 8.0, 100 µM reduced glutathione and 10 µM oxidized glutathione. To ensure proper refolding, the volume ratio of refolding buffer to solubilized CCL5 was kept above 10. The mixture was then stirred over night at 4 °C and dialyzed against 20 mM acetic acid for 24 hours. The dialysate was filtered to remove insoluble precipitate and the filtrate was then injected onto a SP strong cation exchange column equilibrated with 20 mM acetate at pH 4.0. The protein eluted from the SP column with a gradient of salt from 0 M to 1.5 M NaCl. The resulting protein was dialyzed against the appropriate buffer and concentrated.

Purification of CS hexasaccharides

CS hexasaccharides (dp6, degree of polymerization 6) were obtained by partially depolymerizing native CS (Sigma Aldrich) using type V Sheep testes hyaluronidase in 0.05 M sodium phosphate, pH 6.0 containing 150 mM NaCl. The ratio of enzyme to substrate was approximately 0.12 grams of enzyme for each gram of CS. The digestion was carried out for two days before the mixture was purified using a 2.5 cm by 150 cm Biorad Biogel P-10 size exclusion column with aqueous 1 M NaCl, 10 % ethanol as the eluent. The fractions containing hexasaccharides were pooled, dialyzed against water to remove the salt and lyophilized. The mixture was then purified using SAX-HPLC to separate them according to sulfation position in a procedure similar to Yu et al. (Yu et al., 2007) Finally, the concentrations of all CS ligand solutions were confirmed using carbazole assays.

TEMPO functionalization of CS hexasaccharides

Homogeneous CS dp6 was functionalized with a TEMPO tag using reductive amination. Specifically, 2 mg of CS dp6 was incubated with 10 mg/ml of 4-amino-TEMPO and 15 mM NaCNBH₃ in 200 µL of water at 65 °C for two to three days. The mixture was then desalted and purified with SAX HPLC. TEMPO tagged CS dp6 was confirmed using ESI-MS.

NMR spectroscopy

NMR data was collected either on a Bruker Avance 850 MHz spectrometer equipped with a triple resonance cryoprobe, or on an Agilent Inova 800 MHz spectrometer equipped with a triple resonance coldprobe. All experiments were collected at a temperature of 25 °C except the E66S CCL5 dimer-monomer equilibrium data which were collected at 40 °C. ¹³C-filtered/edited experiments were performed on samples containing 200 µM of ¹³C/¹⁵N-labeled E66S CCL5 or perdeuterated E66S CCL5 with ¹³c/¹H ILV methyl groups and 1 molar equivalent of reduced 4,4,4 sulfated CS dp6 (CS444) in 20 mM sodium acetate buffer, pH 4.5. Standard saturation transfer difference (STD) experiments were performed on samples containing either 0.5 mM or 1.0 mM CS444 with 1/20 equivalence of unlabeled E66S CCL5. ¹H saturation of protein protons were performed for three seconds with the saturation frequency set at 0.97 ppm or −20 ppm. The titrations were performed by preparing a series of samples containing either 50 or 40 µM of protein and in the presence of eith 0, 0.5, 1, 2, 3 or 4 equivalents of CS ligands. A ¹⁵N-edited HSQC was collected for each sample and the interaction Kd values of individual residues were extracted using xcrvfit (www.bionmr.ualberta.ca/bds/software/xcrvfit/). The global Kd values were calculated using the nlinmultifit macro in MATLAB using information for all residues.

ITC analysis of E66S CCL5’s interactions with CS444

ITC titration of E66S CCL5 with heparin dp6 was performed on a Microcal ITC-200 calorimeter. In particular, samples consisting of 300 µL of 50 µM DBPA were titrated with aliquots of 5 mM CS444 stock solution at 25 C. Buffer containing no protein was used as a reference. The titration was performed twice times and the average value of the dissociation constant.

Computational modeling of the CCL5-GAG complex

The molecular simulation package AMBER was used to create a model of the CCL5-GAG complex. The starting model of the CCL5 dimer was derived from the crystal structures of Hp disaccharide bound to wild type CCL5 (PDB code 1U4L) by adding two missing N-terminal residues to monomer B (Shaw et al., 2004). The CS444 ligand was constructed using parameters in the GLYCAM06 parameter set (Kirschner et al., 2008). The sulfate groups were added using the sulfate geometry specified in GLYCAM06 parameters and charges of the sulfate atoms were derived from ab initio calculation based on methyl sulfates (Huige and Altona, 1995). The complexes were constructed by first placing the ligand away from the CCL5 at different positions and orientations. Simulation of the system was first conducted using NMR distance constraints and CS444 dihedral angle constraints derived from a study of free CS ligands (Yu et al., 2007) for 2 ns. Details of distance and torsion angle constraints used are listed in Table S2. Most atoms in CCL5 were held fixed except those from residues near GAG binding epitopes. These include residues P2 to P9, C11 to H23 and V42 to C50. NMR constraints were then removed and the simulation was conducted for another 10 ns. Trajectories from the last 4 ns of simulation were characterized using the MM/PBSA program in AMBER. 50 frames that possessed the lowest ΔG of protein-ligand interaction as determined by the MM/PBSA calculations were further analyzed using the Hydrogen-bond analysis module in the software VMD (Humphrey et al., 1996) to determine the existence of favorable intermolecular hydrogen bonds.

Highlights.

• CCL5 possesses multiple GAG-binding epitopes.

• N-loop and N-terminus of CCL5 interact with GAGs.

• Binding orientation of GAG ligands is sulfation pattern-dependent.

eTOC Blurb.

Deshauer et al. study interactions of CCL5 with longer glycosaminoglycan fragments. They discovered that CCL5 possesses multiple GAG-binding epitopes that overlap with the receptor-binding sites, and the orientation of the ligand binding is very sensitive to sulfation pattern of the ligand.