The Crystal Structure of the Chemokine Domain of Fractalkine Shows a Novel Quaternary Arrangement*

Fractalkine, or neurotactin, is a chemokine that is present in endothelial cells from several tissues, including brain, liver, and kidney. It is the only member of the CX3C class of chemokines. Fractalkine contains a chemokine domain (CDF) attached to a membrane-spanning domain via a mucin-like stalk. However, fractalkine can also be proteolytically cleaved from its membrane-spanning domain to release a freely diffusible form. Fractalkine attracts and immobilizes leukocytes by binding to its receptor, CX3CR1. The x-ray crystal structure of CDF has been solved and refined to 2.0 Å resolution. The CDF monomers form a dimer through an intermolecular β-sheet. This interaction is somewhat similar to that seen in other dimeric CC chemokine crystal structures. However, the displacement of the first disulfide in CDF causes the dimer to assume a more compact quaternary structure relative to CC chemokines, which is unique to CX3C chemokines. Although fractalkine can bind to heparin in vitro, as shown by comparison of electrostatic surface plots with other chemokines and by heparin chromatography, the role of this property in vivois not well understood.

Chemokines, or chemoattractant cytokines, are small (5-20 kDa), basic, heparin-binding proteins that show 20 -70% identity in their amino acid sequences (1). Chemokines provide a chemotactic gradient toward the site of inflammation as they direct the trafficking of a number of leukocytes, accomplishing this by binding to and activating 7-transmembrane-spanning G-protein-coupled receptors on the surface of the leukocytes. Activation of these receptors then leads to an increase in integrin adhesiveness and stable arrest (2)(3)(4). Chemokines are secreted onto the cell surface and immobilized by binding to GAGs 1 (5). This immobilization is accompanied by oligomerization, which seems to enhance the affinity of the chemokines for their receptors (6,7).
To date, more than 40 different human chemokines have been identified (8). They are mainly characterized by the presence and conserved relative position of the first two cysteine residues. The intramolecular disulfide bridges stabilize the fold of the peptide, ensuring its binding to specific receptors and its functional activity. Chemokines are classified into four categories: C, CC, CXC, and CX 3 C. CC chemokines contain two Nterminal adjacent cysteines, whereas the same cysteines in CXC chemokines are separated by one residue. C chemokines contain only one N-terminal cystine residue, and only one member, lymphotactin, has been identified so far (9).
Recently, a new chemokine was discovered that contains the cysteine motif CX 3 C. This 372-residue protein, known as fractalkine (10,11) or neurotactin (12), contains an extracellular 76-residue chemokine domain (CDF or chemokine domain of fractalkine) tethered to a membrane-spanning domain via a mucin-like stalk. Because of this natural immobilization, fractalkine can withstand vascular flow to capture and activate leukocytes (13,14). An extracellular fragment of fractalkine (containing both the chemokine domain and the mucin-like stalk) can be released by proteolysis, and this form of the protein has activities that are different from those of the membrane-bound form (10,12,15). The fractalkine receptor, CX 3 CR1, is expressed in leukocytes and binds tightly (K d ϳ1-4 nM) to CDF (15)(16)(17)(18); the complex seems to play an anti-inflammatory role (19).
A detailed analysis of the structure of CDF would elucidate the differences between CX 3 C and other chemokines. Here, we describe the x-ray crystal structure of CDF, solved at 2.0 Å resolution. The crystal structure shows a novel quaternary structure for chemokines, which is directly dependent on the bulge produced by the introduction of three residues between the N-terminal cysteines. CDF can bind to heparin, as shown by heparin-agarose chromatography, with an affinity similar to MIP-1␣. A comparison of the calculated electrostatic surface maps for CDF with those for other chemokines shows similarities of charge distribution, which is indicative of heparinbinding capability.

EXPERIMENTAL PROCEDURES
Sample Preparation and Crystallization-CDF (residues 1-76 of fractalkine, with the N-terminal methionine numbered 0) and its selenomethionine form were overexpressed using the plasmid pLSM103 in Escherichia coli strain BL21/pLysS (20). Both proteins were lyophilized after reverse phase-high pressure liquid chromatography and were dissolved in water just before crystallization. Crystals were grown by vapor diffusion, mixing 4 l of protein (17 mg/ml) with 2 l of well solution (4 M sodium formate, 100 mM disodium citrate, pH 3.0) and 10 l of water. Hexagonal prisms formed quickly (within 1-2 days) and continued to grow for about 1 week. Although very large crystals could be grown (1-2-mm diameter), they diffracted poorly when exposed to a conventional x-ray source (a rotating anode powered by a Rigaku RU200 generator), on average showing a maximum resolution of 2.8 -* This work was supported in part the AIDS Targeted Antiviral Program of the Office of the Director of the National Institutes of Health, and by the NCI, National Institutes of Health HIV Drug Resistance Program (to J. L.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The 3.0 Å for a 1°oscillation of 10-min exposures. The crystals also deteriorated quickly because of radiation damage, as even the largest (2 mm) crystals mounted in capillaries stopped diffracting after several minutes of exposure. Rapid cooling of the crystals in a liquid nitrogen stream (100 K), after a brief (30 s) soaking in cryoprotectant (6 M sodium formate, 100 mM disodium citrate, 10% glycerol, pH 3.0), allowed complete data sets to be collected. In addition, synchrotron radiation allowed for higher resolution data collection (2.0 Å). The space group was determined to be either P6 1  Structure Solution-A data set for the native protein and data sets for the selenomethionine derivative at three wavelengths in the multiple anomalous diffraction experiment were collected (beamline X9B, NSLS) ( Table I). The positions of the selenium atoms were determined using SHELXS (21) and were refined using SHARP (22). The phases were further improved by electron density modification and solvent flattening using DM (23) and PHASES (24). Interestingly, interpretable maps could also be generated using phases derived from data belonging to the pseudocell of length c/2 (l ϭ 2nϩ1 removed). In both cases, the density for the C-terminal helices and internal ␤-strands could be identified only when the symmetry for space group P6 1 22 or P6 2 22 was applied.
Because only four sites were located clearly, only one of the three selenomethionines present in the monomer was ordered. From its location in the electron density as well as by comparison to the NMR model of CDF (20), Met 62 was identified as the ordered selenium site. A relatively good fit to the electron density was achieved for the regions corresponding to the ␤-strands and C-terminal ␣-helix with the sulfur of Met 62 placed at the selenium sites. Because of the strong noncrystallographic symmetry, two monomers could be placed by fitting the NMR model to the density, whereas the remaining two were positioned by translating the first two monomers along the c axis by c/2. Although significant fragments of the main chain of all four monomers were located in the electron density, an extensive remodeling followed by refinement was necessary for most of the model. The model was truncated to residues 7-68, and regions 7-19 and 28 -36 were rebuilt based on initial maps using the program O (25). The model was refined using X-PLOR (26) and SHELXL (21), giving a final R-factor of 0.238 and a free R-factor of 0.321. See Table I for further details.
Heparin-Sepharose Chromatography-The chemokines RANTES (regulated on activation normal T cell expressed), MCP-1, MCP-3, MCP-4, MIP-1␣, and MIP-1␤ were loaded onto a 1-ml HiTrap heparin-Sepharose column (Amersham Pharmacia Biotech) in 20 mM HEPES buffer (pH 7.5) and eluted with a 15-ml linear gradient (0 -2 M NaCl in 20 mM HEPES buffer, pH 7.5) at a flow rate of 1.0 ml/min. The presence of protein was monitored by absorbance at 280 nm, and the concentration of NaCl was determined by using an in-line conductivity meter calibrated using the same gradient without protein. RANTES was a gift from Dr. Amanda Proudfoot (27). MCP-1, MCP-3, and MCP-4 were overexpressed in E. coli and purified using standard protocols. MIP-1␣ and MIP-1␤ were purchased from PeproTech (Princeton, NJ).

Crystal Packing and Space Group Dependences-
The interactions between monomers of CDF in the crystal are limited, and as a result the average B-factor of the molecule is relatively high (46 Å 2 ) for a frozen crystal. This is not the first such example for chemokines; however, for instance, the crystal structure of viral MIP-2 solved using multiple anomalous diffraction also shows a high average B-factor (49 Å 2 ) (PDB accession code 1cm9). The loose packing of the molecules reflects the high solvent content of 65-70%. Four monomers of CDF are found within the asymmetric unit arranged as layers along the c-parameter (Fig. 1). This arrangement gives rise to large (70 Å) channels along this direction.
The four monomers in the asymmetric unit are arranged as two asymmetric dimers related by a noncrystallographic 2-fold rotation. This 2-fold axis lies along the (1 -1 0) vector, translated to approximately c ϭ 1/6. Because of slight translational misalignments and packing anomalies between the two dimers (see below), the exact symmetry for P6 2 22, with c ϭ 62 Å (the pseudo half-cell), is broken, forcing four monomers rather than two into the asymmetric unit of P6 1 22, with c ϭ 124 Å. When the data were scaled as the pseudo half-cell (l ϭ 2nϩ1 removed) and the model was refined against these data, the free R-factor did not drop below 0.40, confirming the correct space group as P6 1 22. There is no direct contact seen between these two created dimers, but rather through bound solvent molecules. The closest contact (ϳ4 Å) between the dimers is at the C terminus of monomer D, the ␣-helix of which is ordered for an extra 1 1 ⁄2 turns (residues 68 -74), and of the symmetry-related monomer B. There is a presumed contact between the disordered C termini of monomers A and C, whose helices, if ordered past residue 68, would overlap at the site of the noncrystallographic 2-fold axis in the crystal.
Dimerization-The two monomers that create the dimers (monomers A and D, and B and C) make close contact with each other through an asymmetric interaction of residues 8 -14. The dimer is stabilized by a ␤-sheet formed by residues Cys 8 to Thr 11 (monomers A and C) and residues Thr 11 to Lys 14 (monomers B and D) and additional hydrogen bonds between loops 26 -28 and 46 -48. Note that CDF has not been found to dimerize in solution at any concentration (20,28). However, this does not rule out dimer formation in vivo, in the presence of the receptor and other participating molecules.
It has been observed that at high concentrations, as well as where T represents a test set of reflections (10% of total, chosen at random) not used in the refinement. d RMSD, root mean square deviation.
in crystalline environments, chemokines frequently form well defined dimers (or tetramers (29, 30)) but remain monomeric at lower concentrations in solution (31). The modes of dimerization of chemokines can be classified into two general categories, one observed for most of the CXC chemokines and the other found for CC chemokines. The dimer formed by CDF does not resemble either of these modes, except that the dimeric interaction is formed via an intermolecular ␤-sheet, similar to CC chemokines. However, the monomers in CDF form a more compact dimer than that of CC chemokine dimers. This motion is facilitated by the 3-residue insertion between the disulfides. Although the second disulfide of CDF (Cys 12 -Cys 50 ) can be superimposed onto the second disulfide of two CC chemokines (MCP-1 (30) and RANTES (PDB accession code 1b3a)), the first disulfide of CDF (Cys 8 -Cys 34 ) forces the N terminus to remain close to the core of the monomer. This arrangement causes the second monomer to twist around the peptide Ser 13 -Lys 14 to form the intramolecular ␤-sheet. The first disulfide also causes a bulge in the ␤-strands formed between the two CDF monomers, further distorting the dimeric interface and possibly weakening the dimer formation. The asymmetry of the dimerization is evident when the monomers are superimposed onto each other (Fig. 2). Description of the Structure-Overall, the structure of the CDF monomer (Fig. 3) is similar to that of other chemokine monomers, particularly MCP-1 and RANTES. This similarity is not surprising, because CDF has the highest identities with these two chemokines (35% and 21%, respectively). The Cterminal ␣-helix on monomer D, however, extends out to Ala 71 , with the remaining residues in an extended conformation (Leu 72 -Arg 74 ). The extension of the helical conformation is probably weakly stable, because there are no direct contacts between this extension and other protein atoms. Also, the extension of C termini for the other three monomers is blocked by the presence of protein crystal contacts. Almost all residues are clearly defined by electron density. Residues 42-47, which form a loop and are completely solvent accessible, are the most disordered within the chain, except for the extreme N and C termini. Several solvent-accessible side chains also appear to be disordered, such as Lys 14 and Arg 44 . Met 15 , although clearly visible in the electron density, has a higher B-factor than Met 62 , which might be the reason why only one selenium site was determined by the multiple anomalous diffraction experiment.
Superpositions of CC chemokine monomers MCP-1 and RANTES onto the CDF monomer show significant differences where the sequence alignment deviates (Fig. 4). The crystal structures of MCP-1 and RANTES superimpose on the crystal structure of CDF with fewer deviations than the superposition of the CDF NMR model (20) on the crystal structure of CDF. This is seen clearly within the 40s loop (Thr 43 -Arg 47 ) and in the bulge between Cys 8 and Cys 12 . The N terminus, from Val 5 -Cys 12 , and the loop encompassing residues 28 -38 are shifted unidirectionally relative to the crystal structure of CDF. This shift is considerable, because the superpositions show a maximal difference of 7 Å between the NMR model and crystal structure of CDF. The N terminus, which is attached via the disulfide between Cys 8 and Cys 34 , is pulled along with this loop into its new position. This unidirectional shift between the NMR model and the crystal structure is partly because of crystal contacts at the edge of the 30s loop (Ser 33 ), stabilizing this region of the molecule and perhaps twisting the chain fragments away. However, the crystal structures of MCP-1 and RANTES, although determined in completely different crystallographic conditions, are closer in structural alignment to CDF.
Charge Distribution and Heparin Binding-Most chemokines are highly positively charged. This property facilitates the binding to GAGs on the surface of cells, which immobilizes and presents the chemokines for interaction with their receptors on mobile lymphocytes (5). The free, proteolytically cleaved fractalkine form can chemoattract neutrophils and T lymphocytes, whereas the tethered fractalkine form can only chemoattract neutrophils after posttranslational processing in vivo (12). This finding raises the question of whether free or tethered CDF can interact and bind to GAGs similar to other freely diffusible chemokines. There are several highly conserved, positively charged residues present in most chemokines, which can be grouped into two sites based on sequence proximity, mu- tagenesis, and structural proximity. Site 1 comprises residues within the C-terminal helix (which usually begins after a conserved proline, Pro 54 , in CDF). Site 2 comprises residues from the loop between the second and third ␤-strands of the main ␤-sheet (residues 44 -47 in CDF). These two sites have been found to be important in binding chemokines to GAGs (32)(33)(34).
The electric charge distribution on the molecular surface of CDF shows clustered positive patches in three regions (Fig. 5). Two of the patches overlap with sites 1 and 2, as described above. It is not known whether CDF can bind to GAGs or whether binding to GAGs is required for CDF to chemoattract lymphocytes or neutrophils, but the free form of fractalkine might require some other type of immobilization to create a haptotactic gradient. To investigate whether CDF binds to GAGs with an affinity similar to other chemokines, we compared CDF with an assortment of chemokines using heparin affinity chromatography (Table II). The binding of CDF to heparin-Sepharose is closest in affinity to MIP-1␣. Although the theoretical pI, number of charged residues, and protein sequence of CDF are most similar to those of RANTES and MCP-1, the binding of CDF to heparin-Sepharose is lower. MIP-1␣ has fewer positive charges on its surface than CDF, RANTES, or MCP-1. Thus, the affinity of CDF for binding to GAGs is not dependent strictly on total net charge but on charge localization on the surface of the protein. DISCUSSION Although it is known CDF does not oligomerize in solution, it does in the crystal. This phenomenon has been seen for the chemokines stromal cell-derived factor-1␣ (35) and viral MIP-2 (36). The driving forces in forming oligomers within a crystal lattice probably include the high concentrations of protein available in the crystal, as well as the unusual electrostatic environment that would be present under such conditions. Similar conditions might occur in vivo, because chemokines are not fully diffusible in simple buffered solutions but are immobilized within a two-dimensional area on the cell surface. It is unknown what local concentrations of the chemokine could be achieved under such conditions.
There are other factors known to drive oligomerization. A key structural determinant for aggregation of CC chemokines has been identified as two key acidic residues, one within the 20s loop (Glu 26 in RANTES), the other within the C-terminal helix (Asp 67 in RANTES) (29). Although there is no acidic residue in CDF corresponding to Glu 26 , CDF residue Asp 67 does correspond to the second acidic residue in RANTES. Because the aggregation of MIP-1␣, MIP-1␤, and RANTES is pH-dependent, the oligomers are likely to be held primarily by electrostatic forces. Also, it was found that only one mutation is required to abolish all oligomerization in MIP-1␣, whereas single mutations lower the amount of oligomerization in RAN-TES from large, nonspecific aggregates to dimers and tetramers (29). Because CDF has a more disperse positive charge on its surface than RANTES, similar to MIP-1␣, as well as only a single acidic residue, it seems to be in agreement with data for MIP-1␣ and RANTES that CDF does not oligomerize in solution (20,28).
Because the CDF dimer formed within the crystal is both asymmetric and distinct from CC chemokines in form, a comparison may not totally explain the forces behind oligomerization. Although it is known that CDF does not oligomerize in solution, it cannot be ruled out that fractalkine might dimerize while tethered to a membrane. An analysis of the electrostatic surface maps and structure shows that this dimerization is possible. The dimeric interface is relatively devoid of charges, whereas the side opposite the interface is highly charged (data not shown). Also, both C termini for the two monomers within the dimer are on the same side (Fig. 1), unlike the dimers seen in CC chemokine structures (37). The C-terminal helix is likely to extend farther from the main ␤-sheet, as seen in monomer D. Thus, while bound to the membrane by a flexible mucin-like tether, CDFs might form dimers.
CDF binds to heparin with roughly the same affinity as MIP-1␣ (20,32). The electrostatic surface of CDF is most similar to MIP-1␣. We expect that the binding of CDF to heparin in vivo would be similar to that of MIP-1␣. Mutation studies have shown that a cationic cleft composed of residues Arg 18 , Arg 46 , and Arg 48 (Lys 19 , Arg 45 , and Arg 48 in CDF) is required for binding MIP-1␣ to heparin; the neutralization of any one of  these positive charges eliminates binding to heparin (32). The same is probably true for CDF. Although disruption of MIP-1␣-GAG interactions has little or no effect on binding to and signaling through CCR1 on Chinese hamster ovary cells, it impairs the ability of MIP-1␣ to induce both a change in the shape of monocytes and chemotaxis, presumably through disruption of interactions with CCR5 (32,38). MIP-1␣, as opposed to interleukin-8, MCP-1, or RANTES, does not oligomerize on the surface of human umbilical vein endothelial cells, and its activation of CXCR1-, CCR1-, and CCR2-transfected Chinese hamster ovary cells is only slightly weakened by deglycosylation (6). Thus, although CDF might bind to GAGs, this binding probably only affects the chemoattractant and activation activities of CDF with specific receptors. In summary, we have shown that CDF forms a novel quaternary structure relative to other chemokines. CDF binds to heparin and is most similar to MIP-1␣ in terms of electrostatic properties, although it is most similar to MCP-1 in terms of sequence homology. The coordinates for CDF were deposited at the Protein Data Bank, accession code 1f2l.