Research Article

Crystal structure of the chemokine receptor CXCR4 in complex with a viral chemokine

See allHide authors and affiliations

Science  06 Mar 2015:
Vol. 347, Issue 6226, pp. 1117-1122
DOI: 10.1126/science.1261064

Molecular “go” signals reveal their secrets

Chemokines are proteins that direct how cells move within the body. For instance, chemokines help immune cells locate invading pathogens and ensure that cells position themselves correctly within a developing organ. Cells detect chemokines through G protein–coupled receptors on their surface; however, the molecular details of how these proteins interact remain unclear (see the Perspective by Standfuss). Qin et al. solved the crystal structure of the chemokine receptor CXCR4 bound to the viral chemokine vMIP-II. Burg et al. solved the crystal structure of a viral chemokine receptor bound to the chemokine domain of CX3CL1. Given the role of chemokines in a number of diseases, these results may help in future drug design.

Science, this issue p. 1117, p. 1113; see also p. 1071


Chemokines and their receptors control cell migration during development, immune system responses, and in numerous diseases, including inflammation and cancer. The structural basis of receptor:chemokine recognition has been a long-standing unanswered question due to the challenges of structure determination for membrane protein complexes. Here, we report the crystal structure of the chemokine receptor CXCR4 in complex with the viral chemokine antagonist vMIP-II at 3.1 angstrom resolution. The structure revealed a 1:1 stoichiometry and a more extensive binding interface than anticipated from the paradigmatic two-site model. The structure helped rationalize a large body of mutagenesis data and together with modeling provided insights into CXCR4 interactions with its endogenous ligand CXCL12, its ability to recognize diverse ligands, and the specificity of CC and CXC receptors for their respective chemokines.

The chemokine receptor CXCR4 controls cell migration during immune surveillance and development of the cardiovascular, hematopoietic, and central nervous systems (13). Like many other chemokine receptors (CKRs), CXCR4 contributes to inflammatory diseases and cancer (4, 5). It also functions as one of two co-receptors that facilitate entry of HIV into host immune cells (6). Despite the importance of CXCR4 and CKRs in general, structural insights into CKR:chemokine recognition have been limited to nuclear magnetic resonance studies of chemokines with peptides derived from CKR N termini (79). This is partly due to the challenges of structure determination for full-length membrane proteins and their complexes.

Here, we present the structure of CXCR4 in complex with vMIP-II, a CC chemokine encoded by Kaposi’s sarcoma–associated herpesvirus. vMIP-II functions as a broad-spectrum antagonist of many human CKRs (10) and helps the virus to escape the host immune response (11). We chose vMIP-II for structural studies because it is a high-affinity antagonist of CXCR4 [median inhibitory concentration, 6 to 15 nM (10, 12)] and, as a ligand for both CC and CXC chemokine receptors, was expected to provide insight into ligand recognition specificity.

Design of an irreversible CXCR4:vMIP-II complex

Despite high affinity in membranes, the CXCR4:vMIP-II complex was insufficiently stable in detergent to justify crystallization trials. We therefore employed a strategy that uses disulfide trapping to generate an irreversible complex (13, 14). Coexpression of pairs of single cysteine mutants of CXCR4 and vMIP-II was expected to result in spontaneous formation of a disulfide bond if the disulfide was compatible with the native geometry of the CKR:chemokine complex. Guided by three-dimensional models of CXCR4:chemokine complexes (14), 37 cysteine mutant pairs were designed, and , for each pair, the abundance of disulfide-trapped complexes was evaluated (15). These pairs included seven N-terminal cysteine mutants of vMIP-II that were systematically coexpressed with two CXCR4 cysteine mutants, D972.63C or D187ECL2C [superscript denotes the Ballesteros-Weinstein index (16, 17) for helical domain residues; ECL is extracellular loop]. Of all mutant pairs analyzed, CXCR4(D187C) coexpressed with vMIP-II(W5C) formed the highest percentage of trapped complex (Fig. 1A). It also showed an unfolding temperature of 63°C (Fig. 1B), which is 4° to 14°C higher than other mutant combinations, and excellent monodispersity when analyzed by size-exclusion chromatography (fig. S1). By comparison, the mutant pair with the second highest melting temperature, CXCR4(D187C):vMIP-II(H6C) (59°C), was produced in significantly lower yield and showed lower monodispersity, despite the adjacent position of the vMIP-II cysteine (fig. S1). CXCR4(D97C) formed little or no covalent complex with any of the seven vMIP-II mutants tested (Fig. 1, A and B). The observed sensitivity of several biophysical properties of the complex to precise cysteine placement suggests specificity of the disulfide-trapping approach and supports compatibility of the D187C:W5C disulfide bond with the native complex geometry. We therefore selected CXCR4(D187C):vMIP-II(W5C) for crystallization in lipidic cubic phase (LCP) (18) and determined the structure at 3.1 Å resolution. Data collection and refinement statistics are shown in table S1.

Fig. 1 Design and crystallization of a disulfide-trapped CXCR4:vMIP-II complex.

(A) Nonreducing SDS–polyacrylamide gel electrophoresis and Western blot of CXCR4(D97C) (left) and CXCR4(D187C) (right) coexpressed with cysteine mutants of vMIP-II (residues 1 to 7). Uncomplexed CXCR4 and disulfide-trapped complexes have molecular weights of ~45 and 55 kD, respectively. Band identities were confirmed by Western blot using antibodies against the FLAG and hemagglutinin (HA) tags at the N and C termini of CXCR4 and vMIP-II, respectively (second and third rows). The 55-kD band was labeled by antibodies to FLAG and HA (second to fourth rows); the band at 45 kD was only labeled by the antibody to FLAG (second and fourth rows). (B) Thermal stabilities of the complexes measured by a CPM assay (40) are shown as mean ± SEM measurements performed in triplicate. (C) Overall structure of the CXCR4:vMIP-II complex (gray:magenta ribbon and transparent mesh). (D) Zoomed view of the vMIP-II N terminus in the CXCR4 pocket showing the CXCR4(D187C):vMIP-II(W5C) disulfide bond. The 2mFo – DFc electron density map around the N terminus is contoured at 1.0 σ and colored blue.

Overall complex geometry

In complex with vMIP-II, CXCR4 possesses the typical seven transmembrane (TM) helical topology. Whereas previous dimeric structures of CXCR4 suggested that chemokines might bind receptors in a 2:1 CKR:chemokine stoichiometry (19, 20), the present structure demonstrates that the stoichiometry is 1:1, in agreement with a recent study (14). The chemokine interacts via its globular core with the receptor N terminus [chemokine recognition site 1 (CRS1) (21)] and via its N terminus with the receptor TM pocket (CRS2) (Fig. 1C). Clear electron density is observed for the entire chemokine N terminus, including the CXCR4(D187C):vMIP-II(W5C) disulfide bond, which adopts a favorable geometry (Fig. 1D). Residues 1 to 22 of the receptor are not visible in the density, consistent with the moderate stability of the CRS1 interaction between CXCR4 and vMIP-II, as suggested by disulfide-trapping experiments (fig. S2) and previous mutagenesis studies (12).

Molecular interactions between CXCR4 and vMIP-II

The CXCR4:vMIP-II interaction is mediated by an extensive (1330 Å2) contiguous interface, with every residue in the chemokine N terminus and N loop (1-LGASCHRPDKCCLGYQ-16) contacting the receptor (Fig. 2 and table S2). Although parts of the interface can be classified as CRS1 or CRS2, the absence of a distinct boundary prompted introduction of an intermediate region, CRS1.5 (Fig. 2, A and B). The CRS1 interaction involves CXCR4 N-terminal residues 23-SMKEP-27 packing against the chemokine N loop (residues 13-LGYQ-16) and its third β strand (β3, residues 49-QVC-51) (Fig. 2, C and D, and table S2). This interaction continues toward CRS1.5, where receptor residues 27-PCFRE-31 bind to chemokine residues 8-PDKCC-12 (Fig. 2, C and D) and form an antiparallel β sheet. In CRS2, the chemokine N terminus makes hydrogen bonds to receptor residues D972.63, D2626.58, and E2887.39 and numerous van der Waals packing interactions (Fig. 2, C and D, and table S2). Most of the interacting CXCR4 residues are known determinants of either vMIP-II binding (table S3) or CXCL12 binding and activation (2226). The dominant role of the vMIP-II N terminus is supported by the fact that an isolated vMIP-II(1-21) peptide binds CXCR4 with appreciable affinity [190 nM (12) versus 6 to 15 nM for wild-type (WT) vMIP-II (10, 12)], which is dramatically reduced by mutations L1A, R7A, and K10A (27) (table S3). Notably, a W5A mutation has only a moderate effect (27). Disulfide-trapping studies also support the role of the chemokine N loop (fig. S2).

Fig. 2 Interactions between CXCR4 and vMIP-II.

(A and B) The interaction is mediated by a contiguous interface containing CRS1 (green), CRS2 (red), and CRS1.5 (blue). (A) The receptor is shown as a cut-open surface, the chemokine is shown as a ribbon, and chemokine residues making substantial contacts with the receptor are shown as sticks. (B) The receptor is shown as a ribbon, receptor residues making substantial contacts with chemokine are shown as sticks, and vMIP-II is shown as a surface mesh. (C) Key residues (gray sticks) from CXCR4 (ribbon) that bind vMIP-II (surface representation). (D) Key residues (magenta sticks) from vMIP-II (white ribbon) that bind CXCR4 (cut-open surface). Noncarbon atoms are red (O), blue (N), and yellow (S); carbon stick color intensity is indicative of residue contact strength (table S2).

Comparison of CXCR4:vMIP-II with previous structures

The conformation of the observed part of the receptor N terminus differs significantly from previous small-molecule and peptide-bound structures (19) in that it adopts an orientation almost perpendicular to the membrane to form a β-sheet interaction in CRS1.5 with chemokine residues C11 and C12 (Fig. 3, A and B). To accommodate this change as well as binding of the chemokine N terminus in the TM pocket, the extracellular half of helix I is laterally shifted outward by ~2.4 Å, forming an extra α-helical turn and bending at the top (Fig. 3A). ECL2 forms a β hairpin as in other CXCR4 structures but is more closed onto the binding pocket (Fig. 3A), bringing D181 and D182 of CXCR4 in close proximity with K10 of vMIP-II (Fig. 2, C and D).

Fig. 3 Comparison between CXCR4:vMIP-II and earlier CXCR4 structures.

(A) Overlay of CXCR4 in the vMIP-II complex (gray), the IT1t complex (PDB ID 3ODU; cyan), and the CVX15 complex (PDB ID 3OE0; pale green). vMIP-II is shown as a gray transparent mesh. (B) CRS1 interaction between CXCR4 (gray) and vMIP-II (magenta), in comparison with IT1t-bound (cyan) and CVX15-bound (green) structures. Key residues mediating the CXCR4:vMIP-II interactions are shown as sticks. (C) Binding modes of vMIP-II, IT1t, and CVX15 to CXCR4. CXCR4 is shown as a cut-open surface, colored by electrostatic potential; the bound ligands are shown as spheres. The white dotted line represents the boundary between the major and minor subpockets. (D and E) Comparison of CRS2 interactions of vMIP-II (magenta) with IT1t (cyan) and CVX15 (green).

The binding pocket of CXCR4 is open and negatively charged (Fig. 3C) and can be separated into a major and minor subpocket (28). Similar to the small-molecule antagonist, IT1t, the chemokine N terminus makes the majority of contacts in the minor subpocket and makes polar interactions with D972.63 and E2887.39 (Fig. 3, C and D). By contrast, the spatial overlap between the vMIP-II N terminus and CVX15 is moderate, with common recognition determinants including D187ECL2 and D2626.58 (Fig. 3, C and E). The limited overlap between CVX15 and the chemokine N terminus may enable the design of modulators that simultaneously occupy the minor and major subpockets; in fact, a series of CXCR4 ligands obtained by grafting the N terminus of CXCL12 onto a peptide analog of CVX15 (29) may bind CXCR4 in this manner.

As in five earlier structures (19), CXCR4 forms a dimer in the vMIP-II–bound form (Fig. 4A). The preservation of similar dimerization patterns in all CXCR4 structures (Fig. 4B) suggests possible physiological relevance and is consistent with numerous reports of CXCR4 homo- and heterodimerization in cells (30). The structure also suggests that a receptor dimer can accommodate two monomeric chemokine ligands.

Fig. 4 Crystallographic dimer of CXCR4:vMIP-II.

(A) The overall dimer geometry is similar between CXCR4:vMIP-II (gray:magenta), CXCR4:IT1t (PDB ID 3ODU, light cyan:dark cyan), and CXCR4:CVX15 (PDB ID 3OE0, light green:green). (B) In all CXCR4 complexes solved thus far, the interaction between two CXCR4 molecules (ribbon) is mediated by the top halves of helix V, with additional contacts provided by either intracellular parts of helices III and V or the top halves of helix VI (spheres). Views of the dimer interface on one monomer are shown for CXCR4:vMIP-II, CXCR4:IT1t, and CXCR4:CVX15 complexes. Residues that contribute to dimer formation are shown as spheres; color intensities represent contact strength.

Structure comparisons, bioinformatics, and homology modeling insights into the specificity of CC and CXC chemokine recognition by CKRs

With the exception of atypical CKRs, human CC and CXC chemokines generally pair exclusively with CKRs from the same subfamily. To gain insight into this specificity, as well as the noncanonical pairing of a human CXC receptor (CXCR4) with a viral CC chemokine (vMIP-II), structural and sequence analyses (fig. S4) were complemented by molecular modeling (15). A complex between CXCR4 and its endogenous CXC chemokine, CXCL12, as well as a complex between vMIP-II and another human CKR, CCR5, were chosen for analysis due to available structural and mutagenesis information.

An initial systematic analysis of chemokine structures revealed conformational differences between CC and CXC motifs of the respective chemokines: Whereas in CC chemokines, this region is straight and forms β-sheet interactions within chemokine dimers, it is bent in CXC chemokines and forms no substantial protein-protein interface contacts (fig. S4A). This difference is reflected in the CRS1.5 interactions of the structure and the modeled complexes (Fig. 5, A to C). In the CXCR4:CXCL12 model (Fig. 5A), the bend directs the chemokine N terminus toward receptor helices V/VI and enables hydrogen bonding between chemokine R8 (highly conserved as a base in CXC but not CC chemokines) (fig. S4B) and receptor D2626.58 (highly conserved as an acid in CXC but not CC CKRs) (fig. S4C). By contrast, in the CCR5:vMIP-II model (Fig. 5C), the straightened conformation of the chemokine CC motif directs the chemokine N terminus along the receptor N terminus toward helix I, aided by interactions with receptor K22 in position C+2 (where C is the conserved N-terminal cysteine) and with D2767.32. Notably a base in position C+2 and an acid in position 7.32 are both highly conserved in CC but not CXC CKRs (fig. S4C). Furthermore, mutation of K22 or D2767.32 in CCR5 abrogates binding to vMIP-II, CCL3, and CCL5 (31). Interestingly, both vMIP-II and CXCR4 possess features that are atypical for their respective classes; vMIP-II has three basic residues (H6, R7, and K10) in its proximal N terminus (fig. S4B), and CXCR4 has a base (R30) at C+2 (fig. S4C), which may partially explain the unusual coupling between CXCR4 and vMIP-II.

Fig. 5 Molecular models of homologous receptor:chemokine complexes.

(A and D) CXCR4:CXCL12. (B) CXCR4:vMIP-II. (C) CCR5:vMIP-II. (E) CXCR4:CXCL12(P2G). Panels (A) to (C) focus on CRS1 and CRS1.5 interactions, whereas panels (D) and (E) show predicted CRS2 interactions. The dotted line indicates the approximate extracellular membrane-solvent boundary. (B) A model of WT CXCR4 sY21-F304 (gray) is built in complex with WT vMIP-II (magenta). Despite the absence of the D187C-W5C disulfide bond, the predicted interactions coincide precisely with those observed in the x-ray structure. vMIP-II W5 provides additional packing interaction with ECL2 of CXCR4. [(A) and (C)] Models of CXCR4:CXCL12 (gray:orange) and CCR5:vMIP-II (navy:magenta). [(D) and (E)] Models of CXCR4:CXCL12 (gray:orange) and CXCR4:CXCL12(P2G) (gray:green).

Relevant differences between CC and CXC families are also observed in the predicted CRS1 interactions. The presence of sulfotyrosines sY14 and sY15 (32) in proximity of the conserved N-terminal cysteine in CCR5 (fig. S4C) facilitates interactions with basic residues in the vMIP-II N loop (K17 and R18) and β23 loop (R46 and R48) (Fig. 5C). When evaluated family-wide, high acidity and sulfotyrosine content of the proximal N terminus are characteristic of CC but not CXC receptors (fig. S4C), whereas the basic nature of N and β23 loops distinguishes CC from CXC chemokines (fig. S4B). It appears therefore, that even when sulfotyrosines in the N terminus of CXC receptors contribute to chemokine affinity, they do not engage the N or β23 loops of CXC chemokines. Consistent with this notion, CXCR4 sY21 is predicted to interact with the CXCL12 N-loop–β1-strand junction (Fig. 5A) instead of the neutral N and β23 loops, similar to positions of sulfate groups in multiple CXCL12 structures (33, 34). The cleft defined by the N and β23 loops of CXCL12 is occupied by the backbone of CXCR4 residues S23 to M24, which closely mimic the interaction of a small-molecule CXCR4:CXCL12 inhibitor (34). CXCR4 is a rare CXC receptor that possesses a sulfotyrosine in the proximal N terminus (position C-7) (fig. S4C), which may explain its ability to engage a CC chemokine (vMIP-II) via its basic N or β23 loops. This engagement is further assisted by a four-residue epitope in the chemokine β3-strand that is strictly conserved between vMIP-II (48-RQVC-51) and CXCL12 (47-RQVC-50) and that interacts with receptor D22 and E26 (Fig. 5B), both of which are important for vMIP-II and CXCL12 recognition (23, 26).

The CXCR4:vMIP-II structure can also explain why CXC (35) but not CC (36) chemokines bind and activate their receptors as dimers. CC chemokines dimerize by β-sheet interactions between the straight CC motifs and N-terminal residues (Fig. 6A). This largely coincides with the CRS1.5 interaction in the CXCR4:vMIP-II structure, making it sterically impossible for a CC chemokine to simultaneously bind its dimer partner and a receptor (Fig. 6B). By contrast, CXC chemokines dimerize by their β1 strands (Fig. 6C), which are not involved in receptor interactions and therefore are compatible with the geometry of the CKR:chemokine complexes (Fig. 6D). This model also suggests that CXC chemokine dimers likely bind to single receptor subunits (Fig. 6E) and not to both subunits in a dimer as previously hypothesized (37).

Fig. 6 Implications of the CXCR4:vMIP-II structure for understanding the stoichiometry of receptor:chemokine recognition.

(A) vMIP-II (shown, PDB ID 2FHT) and other CC chemokine dimers are stabilized by β-sheet formation between the CC region and neighboring residues. (B) Superposition of the vMIP-II dimer onto the CXCR4:vMIP-II structure shows that binding of a CC dimer to the receptor is sterically impossible. (C) CXCL12 (shown, PDB ID 3GV3) and other CXC chemokine dimers are stabilized by β-sheet formation between their β1 strands. (D) Superposition of the CXCL12 dimer onto the CXCR4:vMIP-II structure shows that binding of a dimer is feasible. (E) If the receptor dimer geometry is relevant, and the vMIP-II orientation is predictive of CXC chemokine binding, CXC dimers do not simultaneously bind both receptors in a dimer.

Modeling-based insights into agonist versus antagonist chemokine binding to CXCR4

CXCL12 can be converted into a potent antagonist of CXCR4 by as little as a single N-terminal amino-acid substitution (P2G) (38). To investigate the basis for this notable change in pharmacology, modeling of CRS2 interactions for both CXCL12 and CXCL12(P2G) with CXCR4 was performed. With both chemokine variants, the four distal N-terminal residues were predicted to bind in the minor subpocket of CXCR4 in a manner similar to vMIP-II (Fig. 5, D and E). The N-terminal and side-chain amines of chemokine K1 were predicted to form hydrogen bonds to receptor residues D972.63 and E2887.39, respectively, whereas chemokine residues S4 (in CXCL12) and Y7 [in both CXCL12 and CXCL12(P2G)] hydrogen-bond to D187ECL2. Notably, K1 in CXCL12 and D972.63, D187ECL2, and E2887.39 in CXCR4 are all critical for receptor interaction and activation (22, 25, 26, 38). In CXCL12, the side chain of P2 was found in proximity of receptor residue Y1163.32 (Fig. 5D), whose direct interaction with agonists is frequently involved in activation of GPCRs (39). By contrast, due to its greater flexibility and smaller steric volume, the G2 to S4 region of CXCL12(P2G) packed differently (Fig. 5E), avoiding interaction with Y1163.32 and potentially explaining the inability of CXCL12(P2G) to activate CXCR4. However, because docking was performed with an inactive receptor conformation, further structural studies will be necessary to fully understand activation mechanisms.

Chemokine receptor plasticity, promiscuity, and implications for drug design

CXCR4 is remarkable in its ability to recognize multiple unrelated small molecules, peptides, and proteins. While engaging a conserved set of binding determinants, the ligands occupy different regions of the binding pocket due to receptor conformational plasticity involving receptor side-chain and backbone adjustments. Such versatility may allow the receptor to accommodate ligands of different classes, including both CC- and CXC-type chemokines as well as allosteric inhibitors. The growing number of chemokine receptor structures with different ligands opens possibilities for rational design of ligands that have improved inhibition profiles and mechanisms of action.

Supplementary Materials

Materials and Methods

Figs. S1 to S4

Tables S1 to S3

References (4156)

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. In Ballesteros-Weinstein numbering, the most conserved residue in class A GPCRs is designated x.50, where x is the transmembrane helix number. All other residues on a given helix are numbered relative to this position, e.g., x.49, x.51 are the flanking residues.
  3. Acknowledgments: The authors thank H. Zhang and W. Liu for help with x-ray data collection; A. Walker for assistance with manuscript preparation; M. Gustavsson and V. Katritch for valuable discussions and critical reading of the manuscript; M. Chu for assistance with insect cell expression; E. Lolis and J. Murphy for suggesting the use of vMIP-II; and T. Kawamura, D. Hamel, and K. Wang for assistance with chemokine production. The data in this manuscript are tabulated in the main paper and in the supplementary materials. Atomic coordinates and structure factors have been deposited in the Protein Data Bank (PDB) with identification code 4RWS. This work was funded by National Institutes of Health grants R01 GM071872 (R.A.); U01 GM094612, R01 GM081763, and R21 AI101687 (T.M.H.); and U54 GM094618 (R.C.S.). L.G.H. is supported by a 2012 Postdoctoral Fellowship in Pharmacology/Toxicology from the Pharmaceutical Research and Manufacturers of America (PhRMA) Foundation. The General Medical Sciences and Cancer Institutes Structural Biology Facility at the Advanced Photon Source (GM/CA@APS) is supported by the National Cancer Institute (ACB-12002) and the National Institute of General Medical Sciences (AGM-12006).
View Abstract

Stay Connected to Science

Navigate This Article