Characterization of the Stromal Cell-derived Factor-1α-Heparin Complex*

The binding of chemokines to glycosaminoglycans is thought to play a crucial role in chemokine functions. It has recently been shown that stromal cell-derived factor-1α (SDF-1α), a CXC chemokine with potent anti-human immunodeficiency virus activity, binds to heparan sulfate through a typical consensus sequence for heparin recognition (BBXB, where B is a basic residue KHLK, amino acids 24–27). Calculation of the accessible surface, together with the electrostatic potential of the SDF-1α dimer, revealed that other amino acids (Arg-41 and Lys-43) are found in the same surface area and contribute to the creation of a positively charged crevice, located at the dimer interface. GRID calculations confirmed that this binding site will be the most energetically favored area for the interaction with sulfate groups. Site-directed mutagenesis and surface plasmon resonance-based binding assays were used to investigate the structural basis for SDF-1α binding to heparin. Among the residues clustered in this basic surface area, Lys-24 and Lys-27 have dominant roles and are essential for interaction with heparin. Amino acids Arg-41 and Lys-43 participate in the binding but are not strictly required for the interaction to take place. Direct binding assays and competition analysis with monoclonal antibodies also permitted us to show that the N-terminal residue (Lys-1), an amino acid critical for receptor activation, is involved in complex formation. Binding studies with selectively desulfated heparin, heparin oligosaccharides, and heparitinase-resistant heparan sulfate fragments showed that a minimum size of 12–14 monosaccharide units is required for efficient binding and that 2-O- andN-sulfate groups have a dominant role in the interaction. Finally, the heparin-binding site was identified on the crystal structure of SDF-1α, and a docking study was undertaken. During the energy minimization process, heparin lost its perfect ribbon shape and fitted the protein surface perfectly. In the model, Lys-1, Lys-24, Lys-27, and Arg-41 were found to have the major role in binding a polysaccharide fragment consisting of 13 monosaccharide units.

Chemokines are small structurally related chemo-attractant cytokines, characterized by conserved cysteine residues. Almost 40 chemokines have been identified to date which, based on the position of the first N-terminal cysteines, fall into four sub-families. Two of them have been well characterized, the CC group, which includes regulated on activation, normal T-cell expressed, and secreted, monocyte chemoattractant protein-1, and MIP-1 1 (macrophage inflammatory peptides-1), and the CXC group, the prototype of which is interleukin-8. The C chemokine (lymphotactine) and the CX 3 C chemokine (fractalkine) sub-families have been identified more recently (1)(2)(3)(4). These proteins signal through G-protein-coupled seven transmembrane domain receptors (5) and are primarily involved in immunosurveillance, activation, and recruitment of specific cell populations during disease (1, 6 -8). Most, if not all, chemokines bind to heparan sulfate (9, 10), a glycosaminoglycan (GAG) found ubiquitously at the cell surface (11)(12)(13) and in the extracellular matrix (14). This binding is thought to be functionally important, and current models indicate that heparan sulfate (HS) either enhances the local concentration of chemokines in the vicinity of the G-protein-coupled receptor (15) or provides a haptotactic gradient of the protein along cell surfaces. For example, leukocyte migration along the endothelium surface, and migration into the tissues at the site of inflammation, is believed to depend on the local presentation of chemokines by such cell surface-expressed GAGs (16 -18).
Stromal cell-derived factor-1 (SDF-1) is a member of the CXC chemokine family of pro-inflammatory mediators and is a potent chemo-attractant for a variety of cells, including monocytes and T-cells (19,20). However, SDF-1 has many other functions that extend well beyond leukocyte migration, and in many respects, both SDF-1 and its receptor CXCR4 have several unusual features. For example, mice that lack either the SDF-1 or the CXCR4 gene die in utero, with a number of defects including severe developmental abnormalities, indicating that this chemokine clearly has important roles beyond chemotaxis (21). Whereas the production of most chemokines is induced by cytokines or mitogenic stimuli, SDF-1 is constitutively expressed in a large variety of tissues (22)(23)(24). CXCR4, also widely distributed, raises the question of the specificity of this system. HS-protein interaction has selectivity on itself (25) and, although the functional aspects of the SDF/HS interaction have not been reported to date, HS could be of critical importance in orchestrating the cellular response to this chemokine, * This work was supported in part by grants from the Agence National de la Recherche Contre le SIDA and by the CNRS. 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.
§ Supported by a postdoctoral fellowship from the Agence National de la Recherche Contre le SIDA.
** To whom correspondence should be addressed: IBS/LBM, 41  as has already been reported for other ligands involved in the developmental process (26). In addition to these physiological functions, SDF-1 is a potent inhibitor of the cellular entry of CXCR4-dependent human immunodeficiency virus, a process that relies on the occupation and internalization of CXCR4 (20,27). In view of the general importance of HS in the activity of chemokines, the chemokine-HS interaction represents an attractive target for therapeutic intervention (28). In addition, an obvious application of the knowledge obtained on the SDF-1-HS complex is the engineering of drugs blocking human immunodeficiency virus CXCR4 coreceptor. This study was thus undertaken to characterize further the SDF-1␣-HS complex. It has been previously demonstrated that SDF-1␣ binds selectively to HS and heparin (a chemically related GAG) and that the combined substitution of Lys 24 , His 25 , and Lys 27 by serine residues produces a chemokine unable to bind HS (29). The present study investigates the effect of single mutations in this domain and the relative importance of several additional residues of the protein for HS/heparin recognition. From these results, the heparin-binding site was located on the crystal structure of SDF-1␣, and a docking study was undertaken. The structural features of heparin required for efficient binding were also investigated, and these data, together with those obtained from the directed mutagenesis analysis, were combined with a molecular modeling approach to propose a threedimensional model of the interaction between SDF-1␣ and heparin.

MATERIALS AND METHODS
Equipment and Reagents-An upgraded BIAcore system, pioneer F1 sensor chip, amine coupling kit, and HBS (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, 0.005% surfactant P20, pH 7.4) were from Biacore AB. The Pioneer peptide synthesizer and chemicals were from PE Biosystems. Amino acid analysis was performed on a 6300 Beckman amino acid analyser. 125 I-Labeled SDF-1␣ was from New England Nuclear. The Hitrap heparin column was from Amersham Pharmacia Biotech, and biotin-LC-hydrazide was from Pierce. Streptavidin, extravidin peroxidase conjugate, 3,3Ј,5,5Ј-tetramethylbenzidine, chondroitinase ABC, and papain were from Sigma. Heparinase and heparitinase were from Grampian Enzymes, and Bio-Gel P-10 was obtained from Bio-Rad. [ 3 H]Glucosamine and DEAE-Sephacel were from Amersham Pharmacia Biotech. Synthetic oligosaccharides (disaccharide to dodecasaccharide) were from Sanofi Recherche. These consist of alternating ␣and ␤-linked 3-O-methyl-2,6-di-O-sulfo-D-glucose units, mimicking the regular region of heparin (30). Their general structure is shown here for the hexasaccharide.
The selectively desulfated heparins were a gift from Prof. J. Gallagher, and the anti-SDF-1␣ monoclonal antibodies K15C and 4H4 were from Dr. F. Arenzana-Seisdedos.
SDF-1␣ and SDF-1␣ Mutant Synthesis-Wild type SDF-1␣ and SDF-1␣ mutants were synthesized by the Merrifield solid-phase method on a fully automated peptide synthesizer using fluorenylmethyloxycarbonyl (Fmoc) chemistry, as described (29). Selective biotinylation at the C-terminal position was achieved by incorporating an extra lysine residue, bearing a 1-(4,4-dimethyl-2,6-dioscocyclohex-1-ylidene)ethyl protective group, on the side chain. Coupling of biotin after 1-(4,4dimethyl-2,6-dioscocyclohex-1-ylidene)ethyl hydrazine de-protection was performed on the peptide resin. Concentrations of each chemokine or derivative was determined by amino acid analysis after hydrolysis for 20 h in 6 N HCl, 0.2% phenol in the presence of a known amount of norleucine as internal standard. The molecular weights of the synthesized peptides, as measured by ion spray mass spectrometry did not differ from the expected molecular weights. Final purity of wild type or mutant SDF was analyzed by high pressure liquid chromatography and was found to be, on average, greater than 90%. Finally, receptor binding analysis of wild type and mutant SDF was performed with CXCR4 ϩ CEM cells as described (29).
Preparation of Heparin Oligosaccharides-Porcine mucosal heparin (1 g) was digested with heparinase I (8 milliunits/ml) in 15 ml of 0.1 mg/ml bovine serum albumin, 2 mM CaCl 2 , 50 mM NaCl, and 5 mM Tris buffer, pH 7.5, for 54 h at 25°C. The enzymatic reaction was stopped by heating the digest at 100°C for 5 min. The digestion products were then size-separated using a Bio-Gel P-10 column (4.4 ϫ 150 cm), equilibrated with 0.25 M NaCl and run at 1 ml/min (31). Eluted material, detected by absorbance at 232 nm, consisted of a graded series of size-uniform oligosaccharides resolved from disaccharide (dp2) to octadecasaccharide (dp18). To ensure homogeneity, only the top fractions of each peak were pooled, and each isolated fraction was re-chromatographed on a gel filtration column to eliminate further possible contamination. Samples were dialyzed against distilled water and quantified by colorimetric assay, using the Bitter and Muir method (32).
Preparation of Heparan Sulfate 3 H-Labeled Oligosaccharides-HS from Chinese hamster ovary cells were prepared essentially as described (33). Briefly, confluent cells were cultured for 48 h in Ham's F-12 medium, without glucose and supplemented with 10 Ci/ml of [ 3 H]glucosamine. The labeled material was purified by DEAE-Sephacel chromatography, and the proteoglycan peak was sequentially digested with chondroitinase ABC (1 unit/ml, 3 h at 37°C) and papain (1 mg/ml, 16 h at 65°C). The digest, consisting of intact HS chains, was repurified on a small DEAE-Sephacel column. Eluted material was desalted and freeze-dried. Low pH nitrous acid treatment of a small aliquot of this material, analyzed by Bio-Gel P-10 chromatography, gave a pattern of depolymerization typical of HS (34). HS was then digested with heparitinase (20 milliunits/ml) for 4 h at 30°C, and heparitinase resistant oligosaccharides were purified on a 1 ϫ 150 cm Bio-Gel P-10 column.
Heparin-Sepharose Chromatography-Each chemokine (5 l of 2 ϫ 10 Ϫ4 M solution) was injected on a 1-ml Hitrap heparin column and submitted to gradient elution from 0.15 to 1 M NaCl in 20 mM Na 2 HPO 4 / NaH 2 PO 4 , pH 7.4, over 20 min at a flow rate of 0.5 ml/min. Detection was carried out at 280 nm.
Biotinylation, Heparin Immobilization, and Kinetic Analysis-Sizedefined heparin (9 kDa) was biotinylated at the reducing end as described and immobilized on a Biacore sensorchip (29). For this purpose, two flow cells of a F1 sensorchip were activated with 50 l of a mixture of 0.2 M 1-ethyl-3-(dimethylaminopropyl) carbodi-imine/N-hydroxy-succinimide, 0.05 M 1-ethyl-3-(dimethylaminopropyl) carbodi-imine/N-hydroxy-succinimide before injection of 50 l of streptavidin (0.2 mg/ml in 10 mM acetate buffer, pH 4.2). Remaining activated groups were blocked with 50 l of 1 M ethanolamine, pH 8.5. Typically, this procedure permitted coupling of ϳ2000 -2500 resonance units of streptavidin. Biotinylated heparin (5 g/ml) in HBS containing 0.3 M NaCl was then injected over the surface to obtain an immobilization level of 50 resonance units. Flow cells were then conditioned with several injections of 1 M NaCl. For binding assays, six different SDF-1␣ concentrations in HBS were injected at 50 l/min onto the heparin surface for 3 min, after which the complexes formed were washed withrunning buffer. The sensorchip surface was regenerated with a 2-minpulse of 1 M NaCl in HBS, and the binding curves were analyzed as described (29).

SDF-1␣/Heparin Filter Binding Assay and Competition
Analysis-To analyze the binding of oligosaccharide to protein in solution, a modified version of the filter binding assay of Maccarana and Lindahl (35) was used. Briefly, SDF-1␣ (0.25 g) and biotinylated heparin (0.1 g/ml) were coincubated for 2 h at room temperature in 200 l of Tris-buffered saline (TBS). Competition assays were performed by adding unlabeled oligosaccharides to the mixture. SDF-1␣ plus any bound biotinylated heparin were then trapped on the surface of a nitrocellulose membrane by drawing the incubation mixture through the membrane with a vacuum-assisted dot-blot apparatus. The nitrocellulose filters were washed twice with 200 l of TBS and blocked with 5% dry milk in TBS, 0.05% Tween 20. The biotinylated heparin bound to SDF-1␣ was revealed by incubating the membrane with extravidin peroxidase (0.5 g/ml) and ECL detection reagents. The membrane was STRUCTURE 1 SDF-1␣-Heparin Complex exposed to film, and chemiluminescent signals were quantified by densitometry. Competitors included size-defined, heparin-derived, or synthetic oligosaccharides and selectively de-sulfated heparin. For direct binding assay, SDF-1␣ (1.5 g) was incubated with 3 H-labeled HS oligosaccharides (10,000 cpm), and bound complexes were trapped with a nitrocellulose filter as described above. The protein-bound HS was dissociated from the filter with 2 M NaCl and measured in a ␤-scintillation counter.
Computational Methods-The coordinates of the crystal structure of the N33A variant of the SDF-1␣ dimer (36) were taken from the Protein Data Bank (code 1A15). The amino acids missing on one of the monomers (Lys-1 to Tyr-7 on the N-terminal moiety and Ala-65 to Asn-67 on the C-terminal moiety) were built by homology with the other monomer. The missing side chains of two amino acids, i.e. Val-3 and Arg-8, were modeled. The geometry of the modeled amino acids was optimized using the Tripos force field (37) in the Sybyl molecular modeling package. 2 Hydrogen atoms were added and optimized, and partial atomic charges were derived using the Pullman procedure. Connolly surfaces of both monomer and dimer were calculated using the MOLCAD program (39) from the Sybyl package. The GRID program (40) was used to predict the most favorable anchoring position for a sulfated oligosaccharide. The probe used in the calculation was the charged oxygen of a sulfate or phosphate group. The grid spacing was set to 1 Å.
The coordinates of heparin were taken from the NMR-derived structures (41) deposited in the Protein Data Bank (code 1HPN). Both structures, differing by the ring shape of the L-iduronic acid residues that can be either 2 S 0 or 1 C 4 , were considered in the present study. They are referred to in this paper as 2 S 0 heparin and 1 C 4 heparin. Atom types and partial charges were defined according to the PIM energy parameters for carbohydrates (42) to be used within the Tripos force field (37). The torsional angles of the glycosidic linkages between atom C-1 of residue i and atom C-4 of residue j were defined as ⌽ ϭ ⌰(O-5iC-1iO-1iC-4j) and ⌿ ϭ ⌰(C-1iO-1iC-4jC-5j). The signs of the torsion angles were in agreement with the IUPAC-IUPAB conventions (43).
Docking of Heparin in the Binding Site of SDF-1␣-The heparin chain was extended to 7 repeating units (i.e. 14 monosaccharides) to obtain a length longer than the dimension of the SDF-1␣ dimer positively charged cleft. The heparin chain was located exactly parallel to the cleft. Since the heparin dimer included a pseudo axis of symmetry perpendicular to the basic cleft, it was not necessary to consider the two possible directions (i.e. location of the nonreducing end) for the heparin chain. Different starting orientations were generated by rotating the chain along its long axis by 4 steps of 90°, while keeping it at a van der Waals contact distance from the protein surface. Due to the 2-fold symmetry of the heparin ribbons, this was equivalent to combining a 90°rotation and translation of half a repeating unit. This procedure was applied to both 2 S 0 heparin and 1 C 4 heparin, thereby generating 8 start docking modes.
The geometry of each of the complexes was optimized by several cycles of energy minimization. The hydrogen atoms and pendent groups were first optimized. Finally the whole heparin moiety together with the side chain of the amino acids in the positively charged cleft were fully optimized. All energy calculations were performed with the Tripos force field (37) together with energy parameters specially derived for carbohydrates (44) and sulfated derivatives (42). To evaluate the energy of interaction, the complex was also optimized with the heparin at a distance of 50 Å from the protein, therefore with no intermolecular interaction.

RESULTS
The Dimeric Association of SDF-1␣ Creates a Basic Cleft Suitable for Heparin Binding-As with most chemokines, SDF-1␣ is monomeric in solution (45) but a dimeric association is observed upon crystallization (36). The possibility that this dimer could result from a packing artifact in the crystalline state is unlikely because first, the same dimer has been observed in another crystal form of SDF-1␣ (46), and second, the observed dimer interface involves the first ␤-strand of the chemokine as found in all CXC chemokines analyzed to date. The Connolly surfaces of both the SDF-1␣ monomer and dimer have been calculated here. When looking for a possible site of binding for the negatively charged heparin fragment, the most positively charged area in the monomer takes the shape of a crest, formed mainly by amino acids Lys-27, Arg-41, Lys-24, and Lys-43 (Fig. 1a). Amino acids Lys-24, His-25, and Lys-27, present in the first ␤-strand of the chemokine, constitute a typical BBXB heparin-binding motif (47) and have already been shown to be collectively involved in heparin recognition. lographic dimer, the two crests formed by residues 24, 27, 41, and 43 are on the sides of the protein-protein interface, each of them forming the edge of a deep, straight crevice (Fig. 1b). Arg-41 and Lys-43, present in the second ␤-strand of the chemokine, protrude within the BBXB cluster and appear to be ideally located to participate in the interaction. On the opposite side, His-25 is buried at the dimer interface and is more likely to play a role in protein-protein association rather than in the interaction with heparin. It is worth noting that the two Nterminal peptides running parallel to the crevice in both directions extend the dimension of the cleft. In particular, the side chain of the two Lys-1s are responsible for another positively charged area on the accessible surface of the protein and could also be involved in heparin binding. The binding region was further characterized using the GRID program. In this protocol, the interaction of a probe group with a protein of known structure is computed at sampled positions throughout and around the macromolecule, giving an array of energy values. The analysis of the GRID calculations, performed with a probe consisting of a charged oxygen from a sulfate or phosphate group, is graphically displayed in Fig. 1c. Indeed, the most favorable region for binding sulfate is at the protein-protein interface, in the cleft bordered by four basic amino acids, i.e. Lys-27 and Arg-41 of each protein chain. Lys-24 and Lys-43 also represent a favorable binding site on each chain, therefore extending the binding area along the crevice. It is worth noting that a single 9-kDa heparin chain can accommodate up to six SDF-1␣ molecules, indicating that each SDF-1␣ occupies, on average, 6 monosaccharide units (29). Such dense packing of the chemokine along the chain implies protein-protein contact, further suggesting that SDF-1␣ has a dimeric association when bound to heparin. In addition, using the "zero-length crosslinking" strategy described by Pye and Gallagher (48) dimerization of SDF-1␣ has been observed in the presence of heparinderived oligosaccharides. 3 Experimental Identification of SDF-1␣ Amino Acids Involved in Heparin Binding-Based on the charge distribution on the Connolly surface of the proteins, we produced a set of mutant SDF-1␣s, in which residues 24, 27, 41, or 43 were substituted with a Ser residue, or in which the first Lys was omitted. To analyze the relative importance of these amino acids in heparin binding, we used heparin chromatography and a surface plasmon resonance binding assay that we have already described for wild type SDF-1 (29). Typical sensorgrams were obtained for the wild type (Fig. 2a) and mutants SDF-1␣ (Fig. 2, b-g), and sensorgram evaluations are reported in Table I. A double mutant K24S/K27S was unable to bind heparin (Fig. 2b), demonstrating the importance of these two residues for heparin recognition. Single substitution at these positions (mutant K24S or K27S, Fig. 2, c and d) also strongly impaired the binding, indicating that both Lys-24 and Lys-27 are critically involved in the complex formation. However, significant binding of these mutants was observed at higher concentrations (Fig. 2h), indicating that other residues contributed to form the binding site. Substitution of Lys-43 (Fig. 2f) and Arg-41 (Fig.  2g) produced more subtle changes, mainly reducing the association rate constants (Table I)  amino acid outside the charged crevice (see Fig. 1), also led to a significantly reduced affinity (Fig. 2e) demonstrating the involvement of this amino acid in heparin binding. Chemokines were also applied to a heparin chromatography column and eluted with a linear salt gradient. The retention time on the column for each mutant is also reported in Table I and correlates well with the measured changes in affinity. It is worth noting that a mutant SDF, in which Lys-24, Lys-27, and His-25 were mutated to Ser residues fold with a structure comparable with the structure of the wild type chemokine (29). Thus, misfolding of the SDFs mutated at positions 24 or 27 could not be the cause of the absence of heparin binding activity. In addition, the similar affinity of SDF-1␣ and mutant SDFs for CXCR4 ( Fig. 3 and see Ref. 45 for SDF-(2-67)), further support the correct folding of these peptides and allowed us to conclude that the residues analyzed here all take part in the heparinbinding reaction.
Oligosaccharide Length and Sulfate Dependence of the SDF-1␣-Heparin Interaction-Oligosaccharides of defined length that have been enzymatically produced from heparin (dp2, disaccharide to dp18, octadecasaccharide), or chemically synthesized heparin mimics (dp2 to dp12), were investigated for their ability to compete with biotinylated heparin for binding to SDF-1␣. As shown in Fig. 4a, no efficient competition was observed until a fragment length of 8 monosaccharide units (dp8) was included as a competitor. However, for heparinderived oligosaccharides, the inhibition was very low, and significant competition was only observed using oligosaccharides of at least dp10. To investigate further the length dependence, we also produced 3 H-labeled HS oligosaccharides from Chinese hamster ovary cells. SDF-1␣ was coincubated with 10,000 cpm of size-defined heparitinase-resistant oligosaccharides, and the nitrocellulose filter assay was used again to quantitate the bound 3 H-labeled oligosaccharides. The smallest HS fragment able to bind to SDF-1 was an octasaccharide, albeit at a very low level. A first increase in binding was observed with dp10 and dp12, and a second increase was observed for oligosaccharides containing 14 or more monosaccharide units (Fig. 4b). To determine which sulfate groups were important in the binding with SDF-1␣, a competition study, using a range of specifically desulfated heparins, was carried out (Fig. 4c). A similar decrease in competition effectiveness was observed with N-and 2-O-desulfated heparin, indicating that both N-sulfate and 2-O-sulfate groups were involved in SDF-1␣ binding. However, since 2,6-O-desulfated material was ineffective as a competitor, N-sulfate by itself did not allow heparin to bind efficiently to SDF-1␣. 6-O-Desulfated heparin had slightly reduced efficiency compared with native heparin, demonstrating that this position was less important.
The N-terminal Lys Residue of SDF-1␣ Is Involved in Heparin Complex Formation-In view of the reduced affinity of SDF-1␣ lacking the first N-terminal lysine residue, we further investigated the involvement of this amino acid in the SDF-1␣heparin complex. For this purpose, we analyzed the ability of heparin-derived oligosaccharides to prevent the binding of SDF-1␣ to mAb K15C, a mAb that recognizes the two first amino acids of SDF-1␣ (29). Biotinylated SDF-1␣ was coincubated with a range of concentrations of either dp14 or dp8, and the complexes were captured with mAb K15C. The amount of captured biotinylated SDF-1␣ was quantified with extravidin peroxidase. The results showed that dp14 bound to SDF-1␣ prevents the binding to mAb K15C (Fig. 5), suggesting that, as a result of its involvement in the complex with the oligosaccharide, Lys-1 became inaccessible to the mAb. It is worth noting that dp8, an oligosaccharide too short to bridge the two Lys-1s in the SDF-1␣ dimer (see below), failed to compete with the interaction between mAb K15C and the chemokine. As a control we also show here that the binding of mAb 4H4 (which recognizes the C-terminal part of SDF-1␣) was only slightly inhibited by dp14.
Modeling of the Interactions between Heparin and SDF-1␣-To define further the SDF-1␣-heparin complex, we combined the above experimental data with a molecular modeling approach. For this purpose, a heparin chain was docked in the positively charged cleft formed at the SDF-1␣ dimer interface. The starting geometry of heparin was the elongated conformation with a 2-fold axis that has been demonstrated to exist in solution (41). A combination of translation and rotation along the polysaccharide chain was investigated, together with the two possible conformations of the iduronic acid ring. This resulted in the investigation of eight different docking modes. In all cases, the polysaccharide extended above the cleft, and 13 to   Fig. 2 were analyzed as described (29) The association rate constant (k on ), the dissociation rate constant (k off ), and the affinity constant (K d ) are reported, as are the retention times on a heparin affinity column of the wild type (wt) and mutant SDF-1␣. ND, not determined. 14 monosaccharides were needed to cover the distance between the two Lys-1 residues. Due to the large number of variables already included in the docking study, it was not possible to consider polysaccharides containing various amounts of 1 C 4 and 2 S O conformations, the case that is more likely to occur in solution. However, this variation in ring shape does not alter the general ribbon shape of the polysaccharide. The geometry of the eight binding modes was optimized, and the resulting potential energies are listed in Table II. All the complexes displayed a very favorable energy of interaction, mainly due to the electrostatic contact between the heparin negative charges and the positively charged amino acids. In the two energetically preferred binding modes, the heparin ribbon-like chain sat "flat" on the cleft and the L-iduronic acid residues displayed a 1 C 4 ring shape. The calculated energy difference between the two best modes was not significant since changing the orientation of one sulfate group could result in the creation or destruction of a salt bridge. However, the two best modes displayed a potential energy significantly lower than the others, and it could be hypothesized that this particular binding mode, with the negative charges interacting with the edges of the crevice, was the preferred one.
In the optimized complex, corresponding to the lowest energy binding mode (Fig. 6), the heparin fragment, consisting of 14 monomers, appeared to have the appropriate size for binding the dimer. All the monosaccharides but two interacted directly with the protein surface. The positively charged amino acids involved in salt bridges with the sulfate and carboxyl groups of heparin were Lys-1, Lys-24, Lys-27, and Arg-41 of both monomers and Lys-8 of monomer B and Lys-43 of monomer A. Table  III lists details of the electrostatic interactions. Contacts were not identical for chains A and B of the dimer, since the directionality of the polysaccharide chain prevented it from adapting fully to the pseudo-symmetry axis of the protein. The side chains of two neutral amino acids, Asn-46 and Gln-48, were also involved in hydrogen bonding with the heparin molecule. A detailed study of the interaction indicated that Lys-24, Lys-27, and to a lesser extent Lys-1 and Arg-41 had the major role in binding the polysaccharide fragment since all of them established at least five salt bridges or hydrogen bonds with the sugar residues. As displayed in Fig. 6, the heparin conformation fitted perfectly with the shape of the crevice, whereas the two extremities established contact with the two N-terminal regions of the SDF-1␣ dimer. During the energy minimization process, heparin lost the perfect ribbon shape with a 2-fold axis that had been demonstrated to exist in solution (41). The final shape was still extended but less regular and matched perfectly the protein surface. Analysis of the conformation of the polysaccharide indicated that all of the L-iduronic acid remained in its starting 1 C 4 ring shape, but with small variations in the puckering parameters. Further variations were observed at the glycosidic linkages. The values of the ⌽ and ⌿ torsion angles after optimization differed by up to 60°from the starting conformation. However, when compared with the previously calculated energy maps of each linkage (41,49), all 13 glycosidic linkages remained in the conformational region that contained the global minimum. When compared with MM3 energy maps (49), the energy cost of the variation ranged between 0 and 4 kcal/mol depending on the glycosidic linkage. The deformation involving the higher energy was restricted to the center of the binding site, where interaction with the protein was strongest.

DISCUSSION
Molecular modeling, mutant SDF-1␣, and structurally defined heparin oligosaccharides were used to characterize the SDF-1␣-heparin complex. Based on the heparin-binding consensus sequence proposed by Cardin and Weintraub (47), a first group of basic amino acids (KHLK, residues 24 -27) has been shown to be collectively involved in heparin binding. The present study shows that within this cluster both Lys-24 and Lys-27 are essential, since the substitution of either of these two residues strongly impairs the heparin binding activity of SDF-1␣. We also identified three other amino acids that play a secondary role in heparin binding. These new residues (Lys-1, Arg-41, and Lys-43) are located in a different peptide sequence, but interestingly, two of them (Arg-41  basic surface (i.e. Lys-24/Arg-41/Lys-27/Lys-43) only two of them (Lys-24 and Lys-27) are essential for binding, whereas the other two are only slightly involved. Such differences in the importance of spatially colocalized basic residues for heparin binding have been reported for MIP-1␤, in which, of the six basic residues that are clustered into a basic domain, one was found to be critical (Arg-46), two played a secondary role (Arg-18 and Lys-45), and the remaining three (Lys-19, Arg-22, and Lys-48) appeared not to participate in the interaction (51). It is worth noting that in our model of the SDF-1␣-heparin complex, Lys-24 and Lys-27, the two critical residues, create more salt bridges with heparin than Arg-41 and Lys-43, but more structural data will be required to understand these differences. Interestingly, the heparin-binding amino acids are located close to the protein-protein interface in the SDF-1␣ dimer. Analysis with the GRID software showed that the most energetically favorable domain for binding sulfated groups is located at the protein-protein interface, and thus the functional binding site appears to be created by the dimerization of the protein.
Similarly, it has been demonstrated that monomeric interleukin-8 has affinity too weak for heparin to bind the GAG at physiological ionic strength, whereas the dimeric form of this chemokine mediates binding to heparin or to HS (52). Our study strongly suggests that indeed, when bound to heparin, SDF-1␣ behaves as a dimer. In particular, while the stoichiometry of the binding indicated that each SDF-1␣ occupies an average of six monosaccharide units on heparin (29), competition assay and direct binding analysis with HS-derived oligosaccharides clearly indicate that 12-14 monosaccharides are required to interact efficiently with the chemokine. These data are also consistent with our model in which 13 monosaccharide units interact with the dimeric form of the chemokine. As SDF-1␣ has been reported to be monomeric in solution (45), it is likely that the chemokine is dimerized by heparin. It has recently been shown that CXCR4 undergoes dimerization after being bound by SDF-1␣ (53), and it is possible that this process could be promoted by HS at the cell surface. SDF-1␣-induced dimerization of CXCR4 has been proposed to be an important step in triggering the biological response via activation of the JAK/signal transducers and activators of transcription pathway (53). Thus, in addition to providing a chemokine gradient along cell surfaces, HS could help in the constitution of a signaling complex through the dimerization process. However, SDF-1␣ and a mutant SDF (called SDF-1␣ 3/6) that does not bind to heparin are almost indistinguishable with regard to CXCR4 binding, chemotaxis, and intracellular calcium mobilization (29). It is thus possible that different activities may be differentially dependent on HS binding.
An additional finding of the present study is the involvement of the N-terminal lysine residue (Lys-1) in heparin binding. The N-terminal region of the chemokine (peptide 1-9) is critical for its biological activity (54). Although it is disordered in solution, and adopts multiple conformations, a recent study suggests a link between structuring of the N-terminal SDF-1 peptide and its ability to activate its receptor (55). Whether or not the interaction of Lys-1 with heparin stabilizes the Nterminal structure of the chemokine is not known, but this could represent a way by which heparin acts on SDF-1. Nterminal truncations showed that SDF-1␣, lacking the first residue (SDF-1␣-(2-67)), still had an affinity for the receptor but was unable to trigger CXCR4 signaling (45). Of interest is the fact that CD26, a leukocyte-activating antigen, which possesses dipeptidyl peptidase IV activity, cleaves the first two residues (KP) of SDF-1 and thus inactivates the chemokine. It is tempting to imagine that heparin protects the chemokine from such cleavage, and this will be investigated in a future study. The conformation of the N-terminal region proposed in our model is, of course, hypothetical since both NMR and crystallographic studies indicate that this peptide is flexible. However, the proposed conformation not only permits rationalization of the biochemical data but also agrees with a recent crystal structure of SDF-1 in which the N-terminal tail can be located and is shown to extend from the globular region, parallel to the ␤-trands (46). Finally, our model also suggests that Asn-46 and Gln-48 are involved in hydrogen bonding with FIG. 5. Biotinylated SDF-1␣ (1 g/ ml) was preincubated with a range of concentrations of size-defined, heparin-derived oligosaccharides (octasaccharide, dp8 or tetradecasaccharide, dp14), and captured with coated monoclonal antibodies (K15C,  closed bars; 4H4, open bars). The amount of captured biotinylated SDF-1␣ was quantified with extravidin peroxidase and 3,3Ј,5,5Ј-tetramethylbenzidine development.

TABLE II
Potential energies for the optimized complexes following the 8 modes of binding Energies (kcal/mol) are given relative to the "non-interacting" state where the two molecules are separated by 50 Å and optimized following the same procedure as the "interacting" complexes. The total relative energy ⌬E tot is partitioned into contributions from the protein and the ligand (⌬E prot and ⌬E heparin ) and the interaction energy ⌬E inter . heparin and, as such, could contribute to the interaction. Although this has not been experimentally analyzed here, it is worth noting that such bonds have been found in other chemokine-GAG complexes, including PF4, in which Thr-25 and Asn-47 are implicated in the interaction (56). One of our aims was also to obtain some information on the heparin structure required for binding to SDF-1␣. Our data show that a dodeca-to tetradecasaccharide was required to efficiently interact with the chemokine, and in our model this corresponds exactly to the length necessary to span the heparin-binding site on the SDF-1␣ dimer, including the two Nterminal lysine residues. Removal of 2-O-or N-sulfate groups in heparin impedes the binding to SDF, indicating that these particular groups contribute to the binding, whereas removal of 6-O-sulfate has less effect. In contrast, the binding of heparin to interleukin-8 required all types of sulfate substituents and a minimal sequence of 18 -20 monosaccharide units (52). The exact organization of various sulfated saccharide units optimal for SDF-1␣ binding remains to be defined. We are currently  analyzing HS-derived oligosaccharides with high affinity for SDF-1␣ with the aim of obtaining sequence information. This information together with the present study should help us design some HS mimics able to modulate some functions of the chemokine.