Chemokine CXCL7 Heterodimers: Structural Insights, CXCR2 Receptor Function, and Glycosaminoglycan Interactions

Chemokines mediate diverse fundamental biological processes, including combating infection. Multiple chemokines are expressed at the site of infection; thus chemokine synergy by heterodimer formation may play a role in determining function. Chemokine function involves interactions with G-protein-coupled receptors and sulfated glycosaminoglycans (GAG). However, very little is known regarding heterodimer structural features and receptor and GAG interactions. Solution nuclear magnetic resonance (NMR) and molecular dynamics characterization of platelet-derived chemokine CXCL7 heterodimerization with chemokines CXCL1, CXCL4, and CXCL8 indicated that packing interactions promote CXCL7-CXCL1 and CXCL7-CXCL4 heterodimers, and electrostatic repulsive interactions disfavor the CXCL7-CXCL8 heterodimer. As characterizing the native heterodimer is challenging due to interference from monomers and homodimers, we engineered a “trapped” disulfide-linked CXCL7-CXCL1 heterodimer. NMR and modeling studies indicated that GAG heparin binding to the heterodimer is distinctly different from the CXCL7 monomer and that the GAG-bound heterodimer is unlikely to bind the receptor. Interestingly, the trapped heterodimer was highly active in a Ca2+ release assay. These data collectively suggest that GAG interactions play a prominent role in determining heterodimer function in vivo. Further, this study provides proof-of-concept that the disulfide trapping strategy can serve as a valuable tool for characterizing the structural and functional features of a chemokine heterodimer.


Introduction
Chemokines, a large family of signaling proteins, mediate diverse biological processes, including innate and adaptive immunity, organogenesis, and tissue repair [1][2][3]. Common to these functions is the directed trafficking of various cell types through interactions with seven transmembrane G-protein coupled receptors. Chemokine-chemokine receptor interactions form an intricate network of crosstalk, with a given chemokine binding either a single or multiple receptors and a given receptor binding either a single or multiple chemokines [4,5]. Additional layers of complexity arise from chemokines existing in multiple states, from monomers, dimers, and tetramers to oligomers and polymers, and from their interactions with sulfated glycosaminoglycans (GAG) [6][7][8]. During active inflammation, local chemokine concentrations both in the free and GAG-bound forms could vary by orders of magnitude, which in turn could regulate the steepness and duration of chemotactic and haptotactic gradients [9,10]. Further, several lines of evidence indicate that chemokines also form heterodimers, suggesting yet another layer of complexity in regulating function. Solution nuclear magnetic resonance (NMR) spectroscopy is ideally suited for characterizing CXCL7 heterodimers compared to other techniques. We recently assigned the chemical shifts of the CXCL7 monomer and extensively characterized dimerization propensity as a function of solution conditions such as pH and buffer [39]. The heteronuclear single quantum coherence (HSQC) spectrum of 15 N-labeled CXCL7 shows essentially a monomer along with some weak homodimer peaks. On titrating unlabeled CXCL1 or CXCL4 to 15 N-CXCL7, the monomer and homodimer peaks gradually weaken, and new peaks appear that must correspond to the heterodimer ( Figure 1B,C). These new peaks are in slow exchange with the CXCL7 monomer and homodimer. No changes were observed on titrating CXCL8 to 15 N-CXCL7, indicating the absence of heterodimer formation ( Figure  1D). We also carried out reverse titrations by titrating unlabeled CXCL7 to 15 N-labeled CXCL1 or CXCL8. Titrating CXCL7 to 15 N-CXCL1 resulted in the disappearance of monomer peaks and the appearance of new peaks confirming heterodimer formation. Conversely, titrating CXCL7 to 15 N-CXCL8 resulted in no spectral changes.
The appearance of a new peak during the course of a titration indicates that the environment of the particular residue in the heterodimer is different compared to the monomer or homodimer. On titrating CXCL1 to 15 N-CXCL7, new peaks are observed that correspond to CXCL7 β1-strand residues S21, L22, V24, β2-strand residues V34, E35, V36, and I37, and C-terminal helical residues K56, K62, A64, and G65. These residues are located either at or proximal to the dimer interface. In the reverse titration of adding CXCL7 to 15 N-CXCL1, new peaks corresponding to residues I23 to S30 of the dimer-interface β1-strand, T38 to L44 of the adjacent β2-strand, and C-terminal helical residues I58 to S69 are observed. These data collectively indicate that CXCL7-CXCL1 forms a CXC-type heterodimer and that the residues involved in packing interactions that stabilize the homodimers Solution nuclear magnetic resonance (NMR) spectroscopy is ideally suited for characterizing CXCL7 heterodimers compared to other techniques. We recently assigned the chemical shifts of the CXCL7 monomer and extensively characterized dimerization propensity as a function of solution conditions such as pH and buffer [39]. The heteronuclear single quantum coherence (HSQC) spectrum of 15 N-labeled CXCL7 shows essentially a monomer along with some weak homodimer peaks. On titrating unlabeled CXCL1 or CXCL4 to 15 N-CXCL7, the monomer and homodimer peaks gradually weaken, and new peaks appear that must correspond to the heterodimer ( Figure 1B,C). These new peaks are in slow exchange with the CXCL7 monomer and homodimer. No changes were observed on titrating CXCL8 to 15 N-CXCL7, indicating the absence of heterodimer formation ( Figure 1D). We also carried out reverse titrations by titrating unlabeled CXCL7 to 15 N-labeled CXCL1 or CXCL8. Titrating CXCL7 to 15 N-CXCL1 resulted in the disappearance of monomer peaks and the appearance of new peaks confirming heterodimer formation. Conversely, titrating CXCL7 to 15 N-CXCL8 resulted in no spectral changes.
The appearance of a new peak during the course of a titration indicates that the environment of the particular residue in the heterodimer is different compared to the monomer or homodimer. On titrating CXCL1 to 15 N-CXCL7, new peaks are observed that correspond to CXCL7 β 1 -strand residues S21, L22, V24, β 2 -strand residues V34, E35, V36, and I37, and C-terminal helical residues K56, K62, A64, and G65. These residues are located either at or proximal to the dimer interface. In the reverse titration of adding CXCL7 to 15 N-CXCL1, new peaks corresponding to residues I23 to S30 of the dimer-interface β 1 -strand, T38 to L44 of the adjacent β 2 -strand, and C-terminal helical residues I58 to S69 are observed. These data collectively indicate that CXCL7-CXCL1 forms a CXC-type heterodimer and that the residues involved in packing interactions that stabilize the homodimers also stabilize the heterodimer. We also explored whether side chain chemical shifts of glutamine and asparagine can serve as probes for heterodimer formation. In CXCL1, a glutamine and an asparagine are located at the CXC dimer interface as well as a pair of glutamines in the CC dimer interface (Figure 2A). Upon titrating CXCL7, chemical shift changes were observed for CXC dimer-interface Q24 and N27 but not for CC dimer-interfaces Q10 and Q13 ( Figure 2B), providing further structural evidence for a CXC-type dimer.
( Figure 2A). Upon titrating CXCL7, chemical shift changes were observed for CXC dimer-interface Q24 and N27 but not for CC dimer-interfaces Q10 and Q13 ( Figure 2B), providing further structural evidence for a CXC-type dimer.
Peak intensities can provide valuable information on the relative populations of the monomer, homodimer, and heterodimer. We were able to track intensity changes for a number of residues upon titrating CXCL1 into 15 N-CXCL7 and vice versa. During the course of the titration, populations of both the CXCL7 monomer and homodimer decrease and populations of the heterodimer increase. On adding excess CXCL1, heterodimer and monomer populations become comparable and the homodimer population becomes negligible ( Figure 2C). However, in the case of CXCL1, the heterodimer population continues to increase, but the homodimer levels remain high and the monomer population becomes negligible ( Figure 2D). The relative populations from both titrations indicate that the heterodimer is more favored than the CXCL7 homodimer but less favored than the CXCL1 homodimer. Figure 2. Characterization of the native CXCL7-CXCL1 heterodimer. (A) CXC (green) and CC (red) dimer-interface asparagine and glutamine residues are highlighted in the CXCL1 structure. The CXC and CC dimer interfaces are outlined with a blue arc; (B) Section of the spectra showing the CXCL1 side chain peaks for N27 and N46 in the free form (black) and in the presence of CXCL7 (red). Of the two peaks, only N27 shows reduced intensity and a new peak corresponding to the heterodimer (labeled as N27'); (C,D) Plots showing the relative populations of monomer (M), homodimer (D), and heterodimer (HD) based on NMR peak intensities during the course of the titration. Panel C shows the relative populations on adding CXCL1 to 15 N-CXCL7, and panel D shows the relative populations on adding CXCL7 to 15 N-CXCL1.
We briefly describe our findings for the CXCL7-CXCL4 heterodimer. Considering that the tetramer structures of CXCL7 and CXCL4 reveal both CXC and CC dimer interfaces [16,17], heterodimerization could occur via one or both interfaces. However, similar to the CXCL7-CXCL1 heterodimer, most of the new peaks lie in proximity to the first and second β-strand residues and Figure 2. Characterization of the native CXCL7-CXCL1 heterodimer. (A) CXC (green) and CC (red) dimer-interface asparagine and glutamine residues are highlighted in the CXCL1 structure. The CXC and CC dimer interfaces are outlined with a blue arc; (B) Section of the spectra showing the CXCL1 side chain peaks for N27 and N46 in the free form (black) and in the presence of CXCL7 (red). Of the two peaks, only N27 shows reduced intensity and a new peak corresponding to the heterodimer (labeled as N27'); (C,D) Plots showing the relative populations of monomer (M), homodimer (D), and heterodimer (HD) based on NMR peak intensities during the course of the titration. Panel C shows the relative populations on adding CXCL1 to 15 N-CXCL7, and panel D shows the relative populations on adding CXCL7 to 15 N-CXCL1.
Peak intensities can provide valuable information on the relative populations of the monomer, homodimer, and heterodimer. We were able to track intensity changes for a number of residues upon titrating CXCL1 into 15 N-CXCL7 and vice versa. During the course of the titration, populations of both the CXCL7 monomer and homodimer decrease and populations of the heterodimer increase. On adding excess CXCL1, heterodimer and monomer populations become comparable and the homodimer population becomes negligible ( Figure 2C). However, in the case of CXCL1, the heterodimer population continues to increase, but the homodimer levels remain high and the monomer population becomes negligible ( Figure 2D). The relative populations from both titrations indicate that the heterodimer is more favored than the CXCL7 homodimer but less favored than the CXCL1 homodimer.
We briefly describe our findings for the CXCL7-CXCL4 heterodimer. Considering that the tetramer structures of CXCL7 and CXCL4 reveal both CXC and CC dimer interfaces [16,17], heterodimerization could occur via one or both interfaces. However, similar to the CXCL7-CXCL1 heterodimer, most of the new peaks lie in proximity to the first and second β-strand residues and none were found close to the N-loop residues, indicating a CXC-type dimer interface. Side chain chemical shifts of Asn and Gln residues of the CXC dimer interface were also perturbed, providing further evidence for a CXC-type dimer.

Molecular Dynamics of Chemokine Heterodimers
We utilized a molecular dynamics-based approach to gain insight into the molecular basis for heterodimer formation. Energy minimized heterodimer structures were subjected to~180 ns molecular dynamics (MD) simulations in order to arrive at a stable structure that had minimal fluctuations in backbone root-mean-square deviation (RMSD). To gain insight into the relative stabilities and better understand the structural features that mediate heterodimer formation, we examined several parameters during the course of the simulation; H-bonds and packing interactions of the dimer-interface residues, backbone φ-ψ angles, and charge-charge interactions. The MD simulations collectively indicated that a combination of favorable H-bonding, packing, and electrostatic interactions, similar to what drives any complex formation, dictate heterodimer formation.
In the case of CXCL7-CXCL1, both monomer structures retained their tertiary fold. The H-bond network across the dimer interface β-strands remained intact for the CXCL7 residues L22 and V24 to CXCL1 residues V26 and V28, while peripheral H-bonds between CXCL7 G26 and Q20 to CXCL1 Q24 and S30 are transient throughout the run ( Figure 3B). Backbone φ-ψ angles fall in the allowed region of the Ramachandran plot throughout the simulation. The dimer interface is stabilized by a number of favorable intermolecular packing interactions ? between M66 and L67 of CXCL1 and V24, G26, K56, and V59 of CXCL7 and between K62 and L63 of CXCL7 and V28, S30, V40, and I63 of CXCL1 ( Figure 3C,D). Further, as is the case for the respective homodimer structures, the relative orientation of the helices remained parallel and in register ( Figure 3A). none were found close to the N-loop residues, indicating a CXC-type dimer interface. Side chain chemical shifts of Asn and Gln residues of the CXC dimer interface were also perturbed, providing further evidence for a CXC-type dimer.

Molecular Dynamics of Chemokine Heterodimers
We utilized a molecular dynamics-based approach to gain insight into the molecular basis for heterodimer formation. Energy minimized heterodimer structures were subjected to ~180 ns molecular dynamics (MD) simulations in order to arrive at a stable structure that had minimal fluctuations in backbone root-mean-square deviation (RMSD). To gain insight into the relative stabilities and better understand the structural features that mediate heterodimer formation, we examined several parameters during the course of the simulation; H-bonds and packing interactions of the dimer-interface residues, backbone φ-ψ angles, and charge-charge interactions. The MD simulations collectively indicated that a combination of favorable H-bonding, packing, and electrostatic interactions, similar to what drives any complex formation, dictate heterodimer formation.
In the case of CXCL7-CXCL1, both monomer structures retained their tertiary fold. The H-bond network across the dimer interface β-strands remained intact for the CXCL7 residues L22 and V24 to CXCL1 residues V26 and V28, while peripheral H-bonds between CXCL7 G26 and Q20 to CXCL1 Q24 and S30 are transient throughout the run ( Figure 3B). Backbone φ-ψ angles fall in the allowed region of the Ramachandran plot throughout the simulation. The dimer interface is stabilized by a number of favorable intermolecular packing interactions ⎯ between M66 and L67 of CXCL1 and V24, G26, K56, and V59 of CXCL7 and between K62 and L63 of CXCL7 and V28, S30, V40, and I63 of CXCL1 ( Figure 3C,D). Further, as is the case for the respective homodimer structures, the relative orientation of the helices remained parallel and in register ( Figure 3A). Packing interactions involving CXCL7 helical (cyan) and CXCL1 β-sheet residues (green) and CXCL1 helical (green) and CXCL7 β-sheet (cyan) residues; (G,H) Packing interactions involving CXCL7 helical (cyan) and CXCL4 β-sheet (red) residues and the CXCL4 helical (red) and CXCL7 β-sheet (cyan) residues. The circle Packing interactions involving CXCL7 helical (cyan) and CXCL1 β-sheet residues (green) and CXCL1 helical (green) and CXCL7 β-sheet (cyan) residues; (G,H) Packing interactions involving CXCL7 helical (cyan) and CXCL4 β-sheet (red) residues and the CXCL4 helical (red) and CXCL7 β-sheet (cyan) residues. The circle highlights the potential ionic interaction between CXCL4 E69 and CXCL7 K56. Nitrogen atoms are shown in the conventional dark blue and oxygen in light red.
For the CXCL7-CXCL4 heterodimer, the final MD structure revealed that the monomer structures maintained their tertiary fold ( Figure 3E). The dimer interface H-bonds across the β 1 -strands remain intact for CXCL7 residues L22 and V24 to CXCL4 residues L27 and V29, whereas the edge H-bonds (between CXCL7 Q20 and CXCL4 K31 and between CXCL7 G26 and CXCL4 T25) are transient ( Figure 3F). The dimer is stabilized by favorable packing interactions between K62 and L63 of CXCL7 and V29, L41, Y60, and I64 of CXCL4 and between L67 and L68 of CXCL4 and V24, V34, V36, K56, V59, and L63 of CXCL7 ( Figure 3G,H). Many of these residues are similar in the corresponding homodimers indicating conserved interactions ( Figure 1A). However, there are unique structural differences in the heterodimer. For instance, E69 of CXCL4 (corresponding to A64 in CXCL7) is involved in ionic interactions with K56 from the opposite helix in CXCL7 ( Figure 3H), and CXCL4 L67 and L68 are involved in additional packing interactions with CXCL7 L63 and V59. These new interactions result in the realignment of the helix and partial unwinding of the terminal helical residues.
For the CXCL7-CXCL8 heterodimer, despite favorable H-bonding and packing interactions, there was significant disruption of the tertiary fold due to unfavorable ionic interactions. The structure reveals that CXCL7 K27 and CXCL8 R68 are positioned across the dimer interface, resulting in electrostatic repulsion. In the CXCL8 homodimer, R68 is involved in favorable ionic interactions with E29 across the dimer interface. This swap from favorable to unfavorable interactions provides a molecular basis as to why CXCL7-CXCL8 fails to form a heterodimer.

Design and Characterization of a Trapped Heterodimer
Characterizing the structural and functional features of the native heterodimer is challenging due to contributions from two native homodimers and two native monomers. In principle, the solution contains as many as ten species; two monomers in the free and bound form, two dimers in the free and bound form, and heterodimers in the free and bound form. NMR experiments reduce this complexity by selectively labeling one of the monomers of the heterodimer, which simplifies the spectra to six species. In reality, we observe three sets of peaks due to fast exchange between the free and the bound form. Nevertheless, interpretation of such spectra is still challenging due to challenges in unambiguously assigning the chemical shifts of the newly formed heterodimer and tracking chemical shift perturbations (CSPs) of multiple species. This was evident when we initially attempted to characterize GAG binding to a wild type (WT) heterodimer mixture of 15 N-CXCL7/CXCL1 or CXCL7/ 15 N-CXCL1 at a 1:1 molar ratio. In order to overcome these limitations, we designed a disulfide-linked "trapped" CXCL7-CXCL1 heterodimer.
We used our heterodimer structural models from MD simulations to examine potential mutation sites in the CXCL7-CXCL1 heterodimer. To ensure formation of only the disulfide-linked heterodimer and no disulfide-linked homodimers, we looked for residues that are away from the two-fold symmetry axis. Other criteria that we considered were that these residues should minimally contribute to dimerization and/or influence the native fold. Our analysis pinpointed the solvent exposed β 1 -strand residues as likely candidates ( Figure 4A,B). From this group, we chose the pair S21 from CXCL7 and K29 from CXCL1. The individual cysteine mutants (CXCL7 S21C and CXCL1 K29C) were recombinantly expressed and purified, and the trapped heterodimer was allowed to form by simple mixing of the proteins. We confirmed trapped heterodimer formation using SDS-PAGE, mass spectrometry, and NMR spectroscopy. Bands corresponding to the heterodimer were observed only under non-reducing conditions, indicating a disulfide-linked heterodimer ( Figure 4C). The NMR spectra of the trapped heterodimer showed well-dispersed peaks characteristic of a single folded protein ( Figure 5A,B). We also compared NMR spectra of the trapped heterodimer to the WT heterodimer ( Figure 5C). The spectra were essentially similar except for residues in and around the mutation, indicating that the introduction of the disulfide does not perturb the native fold and that the trapped heterodimer retains the structural characteristics of the native heterodimer.   , and a mixture of CXCL7 and CXCL1 in which both native heterodimer and native homodimer are present (black). Trapped heterodimer alone exists as a single species, free CXCL7 exists as monomers and homodimers, and native heterodimer is present along with native monomer and homodimer ( Figure 2C). The trapped and native heterodimers have similar chemical shifts, as is evident from superimposed peaks. Please note the absence of a green peak superimposed on the homodimer peak. The peak corresponding to the monomer is not shown as it resonates outside of the displayed spectral window.   , and a mixture of CXCL7 and CXCL1 in which both native heterodimer and native homodimer are present (black). Trapped heterodimer alone exists as a single species, free CXCL7 exists as monomers and homodimers, and native heterodimer is present along with native monomer and homodimer ( Figure 2C). The trapped and native heterodimers have similar chemical shifts, as is evident from superimposed peaks. Please note the absence of a green peak superimposed on the homodimer peak. The peak corresponding to the monomer is not shown as it resonates outside of the displayed spectral window. , and a mixture of CXCL7 and CXCL1 in which both native heterodimer and native homodimer are present (black). Trapped heterodimer alone exists as a single species, free CXCL7 exists as monomers and homodimers, and native heterodimer is present along with native monomer and homodimer ( Figure 2C). The trapped and native heterodimers have similar chemical shifts, as is evident from superimposed peaks. Please note the absence of a green peak superimposed on the homodimer peak. The peak corresponding to the monomer is not shown as it resonates outside of the displayed spectral window.
Knowledge of the chemical shifts is essential for NMR characterization of trapped heterodimer GAG interactions. Towards this, we carried out 15 N-edited NOESY and TOCSY experiments on 15 N-CXCL7-CXCL1 and 15 N-CXCL1-CXCL7 trapped heterodimer samples. We were able to assign the backbone 1 H and 15 N chemical shifts of all CXCL1 residues and~80% of CXCL7 residues. Some of the CXCL7 residues could not be assigned due to overlap or lack of sequential nuclear Overhauser effects (NOEs), but this was not limiting as most of the unassigned residues play no role in GAG interactions.

Heterodimer-GAG Interactions
We characterized the binding interactions of GAG heparin octasaccharide (dp8) by individual titrations to 15 N-CXCL7-CXCL1 and 15 N-CXCL1-CXCL7 trapped heterodimer samples. In the 15 N-CXCL7-CXCL1 trapped heterodimer, significant perturbations were observed for N-loop, β 3 -strand, and α-helical residues. Of particular interest are the basic residues H15 and K17 of the N-loop, R44 and K45 from the β 3 -strand, and K56 and K57 from the helix ( Figure 6A). CSPs for hydrophobic or acidic residues located proximal to these basic residues are likely due to indirect interactions. In the case of the 15 N-CXCL1-CXCL7 trapped heterodimer, significant perturbations were observed for residues in the N-loop, β 3 -strand, and α-helix. These include the basic residues H19 and K21 of the N-loop, K45 and R48 of the 40s loop and β 3 -strand, and K61, K65, and K71 of the α-helix ( Figure 6B). Knowledge of the chemical shifts is essential for NMR characterization of trapped heterodimer GAG interactions. Towards this, we carried out 15 N-edited NOESY and TOCSY experiments on 15 N-CXCL7-CXCL1 and 15 N-CXCL1-CXCL7 trapped heterodimer samples. We were able to assign the backbone 1 H and 15 N chemical shifts of all CXCL1 residues and ~80% of CXCL7 residues. Some of the CXCL7 residues could not be assigned due to overlap or lack of sequential nuclear Overhauser effects (NOEs), but this was not limiting as most of the unassigned residues play no role in GAG interactions.

Heterodimer-GAG Interactions
We characterized the binding interactions of GAG heparin octasaccharide (dp8) by individual titrations to 15 N-CXCL7-CXCL1 and 15 N-CXCL1-CXCL7 trapped heterodimer samples. In the 15 N-CXCL7-CXCL1 trapped heterodimer, significant perturbations were observed for N-loop, β3-strand, and α-helical residues. Of particular interest are the basic residues H15 and K17 of the N-loop, R44 and K45 from the β3-strand, and K56 and K57 from the helix ( Figure 6A). CSPs for hydrophobic or acidic residues located proximal to these basic residues are likely due to indirect interactions. In the case of the 15 N-CXCL1-CXCL7 trapped heterodimer, significant perturbations were observed for residues in the N-loop, β3-strand, and α-helix. These include the basic residues H19 and K21 of the N-loop, K45 and R48 of the 40s loop and β3-strand, and K61, K65, and K71 of the α-helix ( Figure 6B). Interestingly, the CSP profiles of CXCL1 versus CXCL7 residues were strikingly different ( Figure 7A,B). Whereas all CXCL1 residues showed similar hyperbolic profiles, CXCL7 showed three distinctly different profiles. A subset of residues showed hyperbolic profiles ( Figure 7C), a subset showed an initial delay in perturbation followed by a hyperbolic profile (Figure 7D), and a subset showed sigmoidal like profiles ( Figure 7E). We define these residues as belonging to Set-I, Set-II, and Set-III, respectively. Interestingly, the CSP profiles of CXCL1 versus CXCL7 residues were strikingly different ( Figure 7A,B). Whereas all CXCL1 residues showed similar hyperbolic profiles, CXCL7 showed three distinctly different profiles. A subset of residues showed hyperbolic profiles ( Figure 7C), a subset showed an initial delay in perturbation followed by a hyperbolic profile (Figure 7D), and a subset showed sigmoidal like profiles ( Figure 7E). We define these residues as belonging to Set-I, Set-II, and Set-III, respectively. These residues lie along the dimer interface across from the CXCL1 β-sheet and the helical residues. Considering that these residues show hyperbolic perturbation profiles similar to CXCL1 residues, it is likely that their CSPs are due to indirect interactions of dp8 binding to CXCL1. For example, our structural model reveals that the CXCL7 K27 side chain is oriented towards the CXCL1 helix, likely making it sensitive to any structural changes in the CXCL1 helix, such as those often associated with dp8 binding. Set-II residues include G13 to I19 of the N-loop, D42 to I46 of the β3-strand, and V59 to A64 of the helix ( Figure 6A). These residues are located away from the dimer interface and are not influenced by CXCL1 binding. These perturbations can thus be attributed to direct dp8 binding to CXCL7.
Set-III residues include Q20 to I25 of the β1-strand, K57, I58, and G65 to A69 of the helix, and L48 to A52 that precede the helix ( Figure 6A). In addition to sigmoidal binding profiles, these peaks showed non-linear chemical shift perturbations ( Figure 7B). These residues are located at the crossroad between the CXCL7-GAG binding interface and the dimer interface. Therefore, their perturbations are likely a composite of both CXCL1 and CXCL7 dp8-binding. Residues K56, K57, and I58 are prominent examples. The K56 side chain is pointed towards the dimer interface, while K57 points out towards the N-loop. K56 shows a linear perturbation similar to CXCL1 residues, suggesting that its perturbation is due to direct or indirect interactions from dp8 binding to CXCL1. The initial perturbation of residues K57 and I58 can thus be attributed to a proximity effect of K56. However, the perturbation profile of K57 and I58 is altered upon further addition of dp8, indicating that these changes must be due to direct dp8 binding to CXCL7. Thus the sigmoidal profiles are a composite of CXCL1 and CXCL7 binding ( Figure 7B,E). These data collectively indicate two independent binding sites, with one heparin binding one monomer and the second heparin binding Set-I residues include K27 and C31 to V34 of the 30s-loop and K56 of the helix ( Figure 6A). These residues lie along the dimer interface across from the CXCL1 β-sheet and the helical residues. Considering that these residues show hyperbolic perturbation profiles similar to CXCL1 residues, it is likely that their CSPs are due to indirect interactions of dp8 binding to CXCL1. For example, our structural model reveals that the CXCL7 K27 side chain is oriented towards the CXCL1 helix, likely making it sensitive to any structural changes in the CXCL1 helix, such as those often associated with dp8 binding.
Set-II residues include G13 to I19 of the N-loop, D42 to I46 of the β 3 -strand, and V59 to A64 of the helix ( Figure 6A). These residues are located away from the dimer interface and are not influenced by CXCL1 binding. These perturbations can thus be attributed to direct dp8 binding to CXCL7.
Set-III residues include Q20 to I25 of the β 1 -strand, K57, I58, and G65 to A69 of the helix, and L48 to A52 that precede the helix ( Figure 6A). In addition to sigmoidal binding profiles, these peaks showed non-linear chemical shift perturbations ( Figure 7B). These residues are located at the crossroad between the CXCL7-GAG binding interface and the dimer interface. Therefore, their perturbations are likely a composite of both CXCL1 and CXCL7 dp8-binding. Residues K56, K57, and I58 are prominent examples. The K56 side chain is pointed towards the dimer interface, while K57 points out towards the N-loop. K56 shows a linear perturbation similar to CXCL1 residues, suggesting that its perturbation is due to direct or indirect interactions from dp8 binding to CXCL1. The initial perturbation of residues K57 and I58 can thus be attributed to a proximity effect of K56. However, the perturbation profile of K57 and I58 is altered upon further addition of dp8, indicating that these changes must be due to direct dp8 binding to CXCL7. Thus the sigmoidal profiles are a composite of CXCL1 and CXCL7 binding ( Figure 7B,E). These data collectively indicate two independent binding sites, with one heparin binding one monomer and the second heparin binding the other monomer of the heterodimer, and that heparin first binds to CXCL1, due to higher affinity, and then to CXCL7.
As discussed above, characterizing GAG binding to the WT heterodimer is challenging. However, using the trapped heterodimer titration spectrum as a template, we explored whether we could characterize heparin binding to the native heterodimer. Indeed, we were able to track heparin binding to a few well-dispersed heterodimer peaks. For instance, upon titration, we observed a heterodimer peak, which showed significant CSP, a non-linear sigmoidal profile, and similar chemical shifts as K57 and I58 in the trapped heterodimer. Additionally, heterodimer peaks that could be assigned to Q20, L48, and G65 showed sigmoidal profiles similar to what was observed in the trapped heterodimer. These observations provide compelling evidence that binding interactions of the trapped heterodimer capture the complexity of the native heterodimer.
To gain insight into the binding geometries, we generated models of the GAG heparin dp8 bound CXCL1-CXCL7 heterodimer complex using HADDOCK-based docking. We performed two independent runs. In run-I, restraints were given between one dp8 and CXCL7 and between another dp8 and CXCL1. In run-II, restraints were given between two GAGs and both monomers of the heterodimer. Both runs showed essentially the same binding geometry, with one GAG binding to each monomer of the heterodimer ( Figure 8A). In CXCL7, the GAG-binding interface spans the β 3 -strand, the N-loop, and the helix and is mediated by H15 and K17 of the N-loop, R44 of the β 3 -strand, and R54, K57, and K61 of the helix ( Figure 8B). In CXCL1, the GAG-binding interface also spans the β 3 -strand, the N-loop, and the helix and is mediated by H19 and K21 of the N-loop, R48 of the β 3 -strand, and K61 and K65 of the helix ( Figure 8C). CXCL1 K45 and CXCL7 K27 were not involved in binding, though both showed significant CSP, indicating that their CSP is most likely due to indirect interactions. We also carried out modeling of one GAG to either CXCL1 or CXCL7 and observed the same binding interactions as observed for two GAGs. Our models provide the structural basis for stepwise and non-overlapping binding geometry, which is consistent with the NMR titrations. Further, the GAG-binding geometry is distinct from that observed in the CXCL1 dimer, wherein GAG binds across the β-sheet dimer interface [40]. Considering that previous studies have established that the N-loop residues in CXCL7 and CXCL1 are involved in receptor binding, the models also suggest that GAG-bound heterodimer cannot bind the receptor [39,41]. Mol. Sci. 2017, 18, 748 10 of 17 the other monomer of the heterodimer, and that heparin first binds to CXCL1, due to higher affinity, and then to CXCL7. As discussed above, characterizing GAG binding to the WT heterodimer is challenging. However, using the trapped heterodimer titration spectrum as a template, we explored whether we could characterize heparin binding to the native heterodimer. Indeed, we were able to track heparin binding to a few well-dispersed heterodimer peaks. For instance, upon titration, we observed a heterodimer peak, which showed significant CSP, a non-linear sigmoidal profile, and similar chemical shifts as K57 and I58 in the trapped heterodimer. Additionally, heterodimer peaks that could be assigned to Q20, L48, and G65 showed sigmoidal profiles similar to what was observed in the trapped heterodimer. These observations provide compelling evidence that binding interactions of the trapped heterodimer capture the complexity of the native heterodimer.
To gain insight into the binding geometries, we generated models of the GAG heparin dp8 bound CXCL1-CXCL7 heterodimer complex using HADDOCK-based docking. We performed two independent runs. In run-I, restraints were given between one dp8 and CXCL7 and between another dp8 and CXCL1. In run-II, restraints were given between two GAGs and both monomers of the heterodimer. Both runs showed essentially the same binding geometry, with one GAG binding to each monomer of the heterodimer ( Figure 8A). In CXCL7, the GAG-binding interface spans the β3-strand, the N-loop, and the helix and is mediated by H15 and K17 of the N-loop, R44 of the β3-strand, and R54, K57, and K61 of the helix ( Figure 8B). In CXCL1, the GAG-binding interface also spans the β3-strand, the N-loop, and the helix and is mediated by H19 and K21 of the N-loop, R48 of the β3-strand, and K61 and K65 of the helix ( Figure 8C). CXCL1 K45 and CXCL7 K27 were not involved in binding, though both showed significant CSP, indicating that their CSP is most likely due to indirect interactions. We also carried out modeling of one GAG to either CXCL1 or CXCL7 and observed the same binding interactions as observed for two GAGs. Our models provide the structural basis for stepwise and non-overlapping binding geometry, which is consistent with the NMR titrations. Further, the GAG-binding geometry is distinct from that observed in the CXCL1 dimer, wherein GAG binds across the β-sheet dimer interface [40]. Considering that previous studies have established that the N-loop residues in CXCL7 and CXCL1 are involved in receptor binding, the models also suggest that GAG-bound heterodimer cannot bind the receptor [39,41].

Heterodimer Receptor Binding Activity
We characterized receptor activity by measuring Ca 2+ release using HL60 cells stably transfected with the CXCR2 receptor [41]. We compared the receptor activities of WT CXCL1, WT CXCL7, a mixture of both chemokines (CXCL7 and CXCL1), and our trapped heterodimer (CXCL7-CXCL1). The trapped heterodimer was as potent as the WT chemokines, and the activity of the mixture of CXCL1 and CXCL7 (that corresponds to the native heterodimer) was no different from the trapped heterodimer or WT proteins (Figure 9). These data indicate that there is no synergy and that essentially one of the monomers of the heterodimer binds and activates the receptor. Previous studies using a trapped homodimer for CXCL1 and CXCL8 have also shown that the activity of the homodimer was no different from the monomer [41][42][43].

Heterodimer Receptor Binding Activity
We characterized receptor activity by measuring Ca 2+ release using HL60 cells stably transfected with the CXCR2 receptor [41]. We compared the receptor activities of WT CXCL1, WT CXCL7, a mixture of both chemokines (CXCL7 and CXCL1), and our trapped heterodimer (CXCL7-CXCL1). The trapped heterodimer was as potent as the WT chemokines, and the activity of the mixture of CXCL1 and CXCL7 (that corresponds to the native heterodimer) was no different from the trapped heterodimer or WT proteins ( Figure 9). These data indicate that there is no synergy and that essentially one of the monomers of the heterodimer binds and activates the receptor. Previous studies using a trapped homodimer for CXCL1 and CXCL8 have also shown that the activity of the homodimer was no different from the monomer [41][42][43].

Discussion
Animal model and in vitro studies have shown enhanced or altered activity for a wide variety of CXC, CC, and CXC/CC chemokine pairs [22][23][24][25][44][45][46]. For instance, high levels of CXCL1 (KC) and CXCL2 (MIP-2) have been observed in a number of murine disease models: virus-infected epithelial cells release multiple chemokines that direct neutrophil chemotaxis; peptides that inhibit CCL5/CXCL4 heterodimer formation alleviate atherosclerosis in a mouse model; and the CXCL7/CXCL4 pair compared to CXCL7 alone shows differential activity for neutrophil adhesion and transendothelial migration [29,[47][48][49]. However, whether altered activity is due to non-additive receptor activity of two chemokines or to distinct heterodimer receptor activity is unknown.
Knowledge of the structural basis and molecular mechanisms by which chemokines form heterodimers is essential to understanding how heterodimers mediate function. In this study, using solution NMR spectroscopy, we were able to describe the structural features and molecular basis by which CXCL7 is able to form heterodimers with some chemokines but not with others. Further, using NMR spectroscopy, we were able to describe the molecular basis of heparin GAG binding to the CXCL7-CXCL1 heterodimer. NMR detects direct binding and does not require exogenous tagging, as do the fluorescence-based FRET/BRET methods, and so does not suffer from potential artifacts. Popular techniques for distinguishing between monomers and dimers such as gel filtration and native gel electrophoresis cannot distinguish between heterodimers and homodimers due to

Discussion
Animal model and in vitro studies have shown enhanced or altered activity for a wide variety of CXC, CC, and CXC/CC chemokine pairs [22][23][24][25][44][45][46]. For instance, high levels of CXCL1 (KC) and CXCL2 (MIP-2) have been observed in a number of murine disease models: virus-infected epithelial cells release multiple chemokines that direct neutrophil chemotaxis; peptides that inhibit CCL5/CXCL4 heterodimer formation alleviate atherosclerosis in a mouse model; and the CXCL7/CXCL4 pair compared to CXCL7 alone shows differential activity for neutrophil adhesion and transendothelial migration [29,[47][48][49]. However, whether altered activity is due to non-additive receptor activity of two chemokines or to distinct heterodimer receptor activity is unknown.
Knowledge of the structural basis and molecular mechanisms by which chemokines form heterodimers is essential to understanding how heterodimers mediate function. In this study, using solution NMR spectroscopy, we were able to describe the structural features and molecular basis by which CXCL7 is able to form heterodimers with some chemokines but not with others. Further, using NMR spectroscopy, we were able to describe the molecular basis of heparin GAG binding to the CXCL7-CXCL1 heterodimer. NMR detects direct binding and does not require exogenous tagging, as do the fluorescence-based FRET/BRET methods, and so does not suffer from potential artifacts. Popular techniques for distinguishing between monomers and dimers such as gel filtration and native gel electrophoresis cannot distinguish between heterodimers and homodimers due to their similar size and molecular weight. Mass spectrometry and co-immunoprecipitation techniques have been used to detect chemokine heterodimers [30,50,51], but these techniques do not provide any insight into the molecular basis of heterodimer formation. NMR chemical shifts of the backbone amide ( 1 H and 15 N) are sensitive to secondary, tertiary, and quaternary structures. Therefore, under ideal conditions, NMR could distinguish heterodimers from homodimers and monomers. Previous NMR studies have shown heterodimer formation between CXCL4 and CXCL8 [32] and that the CCL2-CCL8 heterodimer is favored compared to the CCL2 homodimer [50] but did not describe the structural features of the heterodimer. This is challenging and requires chemical shift assignments not only of the monomer but also of the heterodimer.
The role of in vivo heterodimer function is dependent on receptor and GAG interactions. GAG interactions play multiple roles that include determining the makeup of the chemotactic/haptotactic gradients, influencing whether it is the free or GAG-bound chemokine that activates the receptor, and regulating the levels of the free monomer and homodimer. Further, free and GAG-bound heterodimer levels depend on the GAG affinities, the equilibrium constants (K d ) of the heterodimer and of the two homodimers, and the relative amounts of the two chemokines. Using trapped dimers, it has been shown that the dimer could be as active as the monomer for CXCR2 function in cellular assays [41]. However, the in vivo recruitment activity of the monomers and dimers is distinctly different, indicating that the monomer-dimer equilibrium and GAG binding are coupled and regulate in vivo recruitment [8,52,53]. Therefore, any novel activity of the heterodimer can be inferred only under conditions in which the heterodimer dominates and in which its activity is different from monomers and dimers, and this becomes challenging if its levels are not high and/or its activity is not very different from monomers and homodimers.
In this study, using a disulfide-trapping strategy, we characterized heparin dp8 binding and CXCR2 activity of the CXCL7-CXCL1 heterodimer. We observed that the calcium release activity of the trapped CXCL7-CXCL1 heterodimer, which functions as a readout for the G-protein signaling pathway, was no different compared to the WT proteins. However, chemokine engagement of the CXCR2 receptor activates G-protein and β-arrestin signaling pathways and β-arrestin mediated endocytosis. Several studies have shown that a given chemokine-receptor pair, or multiple chemokines that target a single receptor, can have large differences in G-protein or β-arrestin mediated signaling or receptor internalization activities. Future functional studies of both G-protein and β-arrestin signaling and internalization activities are required to completely understand how heterodimers differ from monomers and homodimers in eliciting receptor function.
To our knowledge, this is the very first study that describes the GAG interactions and receptor activity of a heterodimer without interference from the monomers or homodimers. The GAG interactions of the heterodimer were strikingly different from the CXCL7 monomer and CXCL1 homodimer [39,40]. Further, a number of residues implicated in GAG binding also mediate receptor interactions, suggesting that a GAG-bound heterodimer is unlikely to activate the receptor. We conclude that differences in heterodimer-GAG interactions may play a role in fine-tuning chemotactic/haptotactic gradients and also control the amount of free chemokine available to activate the receptor. Finally, our strategy of engineering a disulfide-linked trapped chemokine heterodimer opens up new avenues to characterize in vivo heterodimer function and the role of differential receptor signaling pathways and to elucidate the heterodimer's role for a variety of chemokine pairs in health and disease.

Molecular Dynamics Simulations
Initial structures were prepared using NMR or X-ray coordinates available from the protein data bank (PDB). The PDB IDs used were 1NAP (CXCL7) [16], 1MSG (CXCL1) [37], 1PFM (CXCL4) [35], and 1IL8 (CXCL8) [38]. Structures were generated by alignment of homodimer backbones and then removal of one of the monomers of each homodimer using PyMol [54]. In the heterodimer, the monomer structures were adjusted by translational and rotational motions about the two fold symmetry axis to align the hydrogen bond network across the β-strands of the dimer interface. The modelled heterodimer structures were then subjected to constrained energy minimization to eliminate any steric clashes, followed by free minimization using the AMBER 12 suite software and the ff03 force field [55,56]. The energy-minimized structures were subjected to an equilibration protocol in explicit solvent [57], followed by~180 ns of MD production runs carried out using the PMEMD (Particle mesh Ewald molecular dynamics) module of the AMBER 12 software suite on the Lonestar Dell Linux Cluster at the Texas Advanced Computing Center (Texas Advanced Computing Center, The University of Texas, Austin, TX, USA). The trajectories were analyzed using AMBERtools 12, VMD, and PyMol [54,56,58].

Expression and Purification of Chemokines
Chemokines were expressed in Escherichia coli cultured in either LB or 15 N-enriched minimal medium and purified using a combination of nickel column and reverse phase high-performance liquid chromatography, as previously described [59]. The CXCL7-CXCL1 trapped heterodimer was prepared by introducing a disulfide across the dimer interface. CXCL7 S21C and CXCL1 K29C mutants were purified using a Ni-NTA column, cleaved using Factor Xa, and were combined without further purification and left overnight at 35 • C. Heterodimer was purified using high performance liquid chromatography, lyophilized, and stored at −20 • C until further use.

NMR Spectroscopy
The samples were prepared in a 50 mM sodium phosphate buffer pH 7.4 at 25 • C containing 1 mM 2,2-dimethyl-2-silapentansesulfonic acid (DSS), 1 mM sodium azide, and 10% D 2 O. Heterodimer formation between two chemokines can be inferred from changes in the HSQC spectra on titrating an unlabeled chemokine to a 15 N-labeled chemokine prepared in the same buffer. Initial 15 N-labeled chemokine concentrations varied between 30 and 150 µM. The final molar ratios of labeled to unlabeled chemokine varied from 1:2 to 1:4. For these experiments, titrations were carried out until essentially no change in the spectra was observed. NMR experiments were performed on a Bruker Avance III 600 (with a QCI cryoprobe) or 800 MHz (with a TXI cryoprobe) spectrometer. All spectra were processed and analyzed using Bruker Topspin 3.2 or Sparky software [60].
The 1 H and 15 N chemical shifts of the trapped CXCL7-CXCL1 heterodimer were assigned using 15 N-CXCL1-CXCL7 and 15 N-CXCL7-CXCL1 samples prepared in 50 mM phosphate pH 6.0 and 35 • C.
The concentrations of 15 CXCL7-CXCL1 and CXCL7-15 CXCL1 were 300 and 670 µM, respectively, and the assignments were obtained from analysis of 1 H-15 N heteronuclear NOESY and TOCSY experiments with mixing times of 150 and 80 ms, respectively.

Heparin-Heterodimer Interactions
The binding of heparin dp8 to the CXCL7-CXCL1 heterodimer was characterized using solution NMR spectroscopy in 50 mM phosphate buffer at pH 6.0 and 30 • C. The protein concentration for the titrations varied between 50 and 70 µM. Heparin dp8 was purchased from Iduron (Manchester, UK) and prepared in the same buffer (10 mM stock), and a series of 1 H-15 N HSQC spectra were collected upon titrating GAG until no changes in the spectra were observed. The final molar ratio of heterodimer to GAG was 1:4. For the trapped heterodimer, both 15 N-CXCL7-CXCL1 and 15 N-CXCL1-CXCL7 samples were used. For native heterodimer interactions, a mixture of CXCL7 and CXCL1 at 1:1 molar ratio was used. The final molar ratio of heterodimer to GAG was~1:3 to 1:4. For all titrations, chemical shift perturbations were calculated as a weighted average of changes in the 1 H and 15 N chemical shifts, as described previously [61].

Heterodimer-GAG Docking
Molecular docking of heparin to the CXCL7-CXCL1 heterodimer was carried out using the High Ambiguity Driven biomolecular DOCKing (HADDOCK) approach, as described previously [62][63][64]. The CXCL7-CXCL1 heterodimer structure from MD studies and the NMR structure of heparin (PDB ID: 1HPN) [65] were used for docking. Ambiguous interaction restraints (AIRs) were selected based on NMR chemical shift perturbation data. The pair-wise "ligand interface RMSD matrix" over all structures was calculated and final structures were clustered using an RMSD cut-off value of 4 Å. The clusters were then prioritized using RMSD and a "HADDOCK score" (weighted sum of a combination of energy terms).

Receptor Activity of the Heterodimer
The CXCR2 receptor activity of the heterodimer was determined using a Ca 2+ release assay, as described previously [41]. Ca 2+ levels were measured using a FlexStation III microplate reader using the Calcium 5 assay kit (FLIPR, Molecular Devices). Differentiated HL60 cells expressing CXCR2 were incubated with varying concentrations of either WT CXCL1, WT CXCL7, a mixture of both WTs, or the trapped CXCL7-CXCL1 heterodimer. Changes in fluorescence of the Calcium 5 dye upon addition of chemokine were measured every 5 s for up to 500 s, and the agonist response was determined from the maximum change in fluorescence. EC 50 values were calculated based on the response over a range of concentrations.