Stromal Cell-derived Factor-1α Associates with Heparan Sulfates through the First β-Strand of the Chemokine*

Biological properties of chemokines are believed to be influenced by their association with glycosaminoglycans. Surface plasmon resonance kinetic analysis shows that the CXC chemokine stromal cell-derived factor-1α (SDF-1α), which binds the CXCR4 receptor, associates with heparin with an affinity constant of 38.4 nm (k on = 2.16 × 106 m −1 s−1 andk off = 0.083 × s−1). A modified SDF-1α (SDF-1 3/6) was generated by combined substitution of the basic cluster of residues Lys24, His25, and Lys27 by Ser. SDF-1 3/6 conserves the global native structure and functional properties of SDF-1α, but it is unable to interact with sensor chip-immobilized heparin. The biological relevance of these in vitro findings was investigated. SDF-1α was unable to bind in a CXCR4-independent manner on epithelial cells that were treated with heparan sulfate (HS)-degrading enzymes or constitutively lack HS expression. The inability of SDF-1 3/6 to bind to cells underlines the importance of the identified basic cluster for the physiological interactions of SDF-1α with HS. Importantly, the amino-terminal domain of SDF-1α which is required for binding to, and activation of, CXCR4 remains exposed after binding to HS and is recognized by a neutralizing monoclonal antibody directed against the first residues of the chemokine. Overall, these findings indicate that the Lys24, His25, and Lys27 cluster of residues forms, or is an essential part of, the HS-binding site which is distinct from that required for binding to, and signaling through, CXCR4.

Based on the relative position of the two first cysteine residues, chemokines are classified in two main subfamilies, CC and CXC chemokines (1). Stromal cell-derived factor-1 (SDF-1) 1 also called pre-B cell-stimulating factor (2) is a CXC che-mokine originally purified from bone marrow cell supernatants (3). Two forms, ␣ and ␤ (68 and 72 amino acids, respectively), generated by alternative splicing from a unique sdf-1 gene, have been identified (4). The ␣ form is the most abundant (4). Human and murine SDF-1␣ proteins differ by a single residue at position 18 (valine to isoleucine in the murine protein) (4).
CXCR4 is the only identified receptor for SDF-1␣. Furthermore, the interaction between SDF-1␣ and CXCR4 appears to be unique and non-promiscuous (5)(6)(7). SDF-1␣ stimulates intracellular calcium flux and chemotaxis in monocytes, T lymphocytes, and neutrophils, a characteristic shared with other CXC chemokines (8,9). However, SDF-1␣ exhibits structural and molecular characteristics that make it a unique chemokine among members of CC and CXC families. SDF-1␣ possesses the peculiar capacity of attracting and promoting bone marrow engraftment of CD34 ϩ CXCR4 hematopoietic stem cells (10). In contrast to most chemokines, which are induced by cytokines or mitogenic stimuli, SDF-1␣ is constitutively expressed in a large number of tissues (4). Importantly, Sdf-1 gene knock-outs induce anomalies in hematopoiesis and the development of cardiovascular system provocating pre-or perinatal death of the embryos (5). Apart from these physiological functions, SDF-1␣ has the selective capacity to inhibit cell entry of CXCR4-dependent human immunodeficiency viruses by occupying and internalizing CXCR4 in T lymphocytes (8,9,11,12). Overall, these findings indicate that SDF-1␣ and its receptor, both of which are expressed widely outside the lympho-hematopoietic system, accomplish important additional functions that are not typical for chemokines.
The biological activities of chemokines are thought to be influenced by their association with cellular or extracellular matrix glycosaminoglycans (GAG). Usually attached to a core protein to form proteoglycans (13), GAG are highly sulfated oligosaccharides characterized by a high degree of structural heterogeneity. The common GAG are heparin, heparan sulfate (HS), dermatan sulfate, chondroitin sulfate, and hyaluronic acid (14,15). The interaction of cytokines with proteoglycans in the extracellular matrix or cell surface have important functional consequences in many biological systems (16,17). Although the in vivo biological roles of chemokine-GAG complexes are not clear, an increasing body of evidence suggests that GAG immobilize and enhance local concentrations of the chemokines, promoting their oligomerization and facilitating their presentation to the receptors (18,19). Thus, it has been proposed that chemokines like macrophage inflammatory protein (MIP)-1␤ or interleukin-8 (IL-8) would be tethered to circulating leukocytes complexed to membrane-bound proteoglycans in endothelial cells (20,21).
Although retention of SDF-1␣ in heparin affinity columns indicates that the chemokine has the capacity to complex with GAG (22), it is not known whether SDF-1␣ is capable to interact with GAG under physiological conditions. Moreover, the nature of GAG is not known, and the structural determinants of the protein that eventually would account for such interactions remain unidentified.
The aim of this work was to investigate the capacity of SDF-1␣ to form complexes with isolated or cell-bound GAG and to characterize the GAG family accounting for these interactions. On the other hand, we wanted to identify structural determinants of SDF-1␣ involved in the physical contact with GAG. Our findings demonstrate that SDF-1␣ interacts selectively and with relatively high affinity with HS in vitro. HS is also responsible for the binding of SDF-1␣ to CXCR4-negative epithelial or endothelial cells. Finally, we identified a cluster of basic residues in the first ␤-strand of the ␤-sheet of SDF-1␣ which is necessary for interaction with HS both in vitro and in intact cells. Chemokine Synthesis-Wild type SDF-1␣ (SDF-1␣) and SDF-1 3/6 (substitution of Lys 24 , His 25 , and Lys 27 by Ser 24 , Ser 25 , and Ser 27 ) were synthesized by the Merrifield solid-phase method on a fully automated peptide synthesizer (Pioneer, Perspective Biosystems, and Perkin-Elmer) using fluorenylmethyloxycarbonyl (Fmoc) chemistry. All amino acids were double coupled with O-(7-azabenzotriazol-1-yl)-1,1,3,3-tetramethyluroniumhexafluorophosphate/DIEA activation, and piperidine Fmoc deprotection was optimized by elongating the standard deprotection time. After completion of the synthesis, the polypeptide was released from the resin and precipitated in cold diethyl ether. The precipitate was dissolved in aqueous 0.08% trifluoroacetic acid and lyophilized. The crude polypeptide was dissolved in 6 M guanidine hydrochloride, 0.1 M Tris acetate, pH 8.5, and 16% 2-mercaptoethanol, stirred at 37°C for 2 h, and then acidified to pH 4. The reduced chemokine was purified on a preparative medium pressure liquid chromatography column (313 ϫ 26 mm) packed with C18 100-Å and 20-m Nucleoprep packing (Macherey-Nagel GmbH & Co, Dü ren, Germany) using a 20 -80% linear gradient of acetonitrile in 0.08% aqueous trifluoroacetic acid over 120 min at a 25 ml/min flow rate. After lyophilization, the purified, reduced chemokine was solubilized in 6 M guanidine hydrochloride, 0.1 M Tris acetate, pH 8.5, and rapidly diluted into 0.1 M Tris acetate buffer, pH 8.5. Final concentration of the chemokine was 0.4 mg/ml in 1 M guanidine hydrochloride, 0.1 M Tris acetate, pH 8.5. The solution was stirred overnight to allow chemokine folding and gentle air oxidation of the four cysteines. The folded chemokine was purified using the medium pressure liquid chromatography purification procedure described above. Final purity of SDF-1␣ and SDF-1 3/6 was superior to 95% as judged by high pressure liquid chromatography. The average molecular weights determined by ion spray mass spectrometry were 7830.9 Ϯ 0.5 (theoretic molecular weight, 7831.3) for SDF-1␣ and 7699.0 Ϯ 0.5 (theoretic molecular weight, 7699.0) for SDF-1 3/6. SDF 1-34 was synthesized using the same procedure. Cysteine in position 11 was replaced by alanine. Cyclization of the reduced half-chemokine was done by H 2 O 2 at pH 7 (23), and the cyclic peptide was purified under the same conditions as described above. Ion spray mass spectroscopy showed a molecular weight of 3910.8 Ϯ 1 (theoretic molecular weight, 3910.6).

Materials
The concentration of each chemokine or derivative was determined by amino acid analysis on a 6300 Beckman amino acid analyzer after hydrolysis for 20 h in 6 N HCl, 0.2% phenol in the presence of a known amount of norleucine as internal standard. All chemicals for the synthesis were purchased from Perspective Biosystems and Perkin-Elmer, France.
The anti-CXCR4 6H8 (IgG1) mAb used in this work was obtained by immunizing BALB/c mice with a peptide spanning the first 28 aminoterminal amino acids of CXCR4 conjugated via cysteine 28 to ovalbumin (25). 6H8 was screened by flow cytometry analysis of CEM cells and was shown to bind to human peripheral blood lymphocytes and human cell lines (Jurkat, U937, THP-1, HeLa, and SupT1) that constitutively express CXCR4. 6H8 reacts specifically with CXCR4-transfected CHO-K1 cells but not with CHO-K1 cells transfected with CCR5, CCR3, CCR1, or CCR2b (data not shown). The specificity of CXCR4 detection by 6H8 mAb was compared with that obtained using the well characterized 12G5 anti-CXCR4-mAb. Both antibodies exhibit identical patterns of recognition and specificity for a large panel of human or animal cells expressing CXCR4 constitutively or upon transfection.
Flow Cytometric Analysis of SDF-1␣ Binding to Cells-Adherent cells were plated 2 days before binding experiments. Cells were detached with 2 mM EDTA in PBS and washed twice with ice-cold binding buffer (RPMI 1640, 20 mM Hepes, 1% bovine serum albumin). 4 ϫ 10 5 cells were resuspended in the presence of the indicated concentration of chemokines in a total volume of 200 l and incubated for 90 min at 4°C under agitation. Unbound chemokine was removed by washing with binding buffer, and cell-bound SDF-1␣ was detected by incubation with the anti-SDF-1 mAb K15C (15 g/ml, diluted in PBS, 1% bovine serum albumin). After staining with phycoerythrin-conjugated anti-mouse immunoglobulins (Southern Biotechnology), cells were fixed in 1% formaldehyde buffer and analyzed in a FACSscan (Becton Dickinson, CA). To remove cell-surface GAG, CHO-K1 and HeLa cells (10 6 cells) were washed twice and incubated for 90 min at 37°C with 1 milliunit/ml GAG-degrading enzymes. For trypsin treatment, cells were washed once with PBS containing 2 mM EDTA and incubated for 4 min at 37°C with 0.25% trypsin in PBS. After enzyme treatment, cells were washed four times with 5 ml of binding buffer, detached with PBS/EDTA, and then assayed for SDF-1␣ binding as described above.
Biotinylation Procedure and Heparin Immobilization-Fractionated heparin (9 kDa) resuspended in PBS at 1 mM was reacted for 24 h at room temperature with 10 mM biotin/LC-hydrazine. The mixture was then extensively dialyzed against water to remove unreacted biotin and freeze-dried. Two flow cells of an F1 sensorchip were activated with 50 l of a mixture of 0.2 M EDC, 0.05 M NHS 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 allowed coupling of approximately 2000 -2500 resonance units (RU) of streptavidin. Biotinylated heparin (5-10 g/ml in HBS containing 0.3 M NaCl) was then injected on one of the two streptavidin surfaces (the other one being a negative control). Approximately 50 RU of material was immobilized. Both flow cells were then conditioned with several injections of 1.5 M NaCl. The conversion of RU to surface concentration of proteins was performed using a conversion factor of 1000 RU ϭ 1 ng/mm 2 .
SDF-1␣/Heparin Interactions, Binding Assay, and Kinetic Analysis-Test samples were diluted in HBS maintained at 25°C and injected over the heparin surface at a flow rate of 50 l/min. This high flow rate was necessary to eliminate mass transport effect due to the relatively high association rate of the proteins being studied. In a typical analysis, six different SDF-1␣ concentrations (usually ranging between 0 and 200 nM) were injected onto the heparin surface for 3 min (to study the association phase and equilibrium). Thereafter, the formed complexes were washed at 50 l/min with HBS to study the dissociation phase. The sensorchip surface was regenerated with a 2-min pulse of 1.5 M NaCl in HBS. In some experiments, soluble heparin was included in the buffer during the dissociation phase to minimize possible rebinding effects. Kinetic constants were derived from the sensorgrams by fitting the data to different interaction models, using BIAevaluation software, essentially as described (26). Affinities (dissociation equilibrium constants: K d ) were calculated from the ratio of dissociation and association rate constants (K d ϭ k off /k on ).
Receptor Binding of SDF-1␣ and SDF-1 3/6 to CXCR4 -CEM cells (5 ϫ 10 6 cells/ml) were incubated in PBS with 0.25 nM iodinated SDF-1␣ (New England Nuclear, specific activity, 2200 Ci/mmol) and various concentrations of unlabeled SDF-1␣ or SDF-1 3/6 for 1 h at 4°C in a final volume of 300 l. Incubations were terminated by centrifugation at 4°C. The cell pellets were washed twice in ice-cold PBS. Nonspecific binding was determined in the presence of 1 M unlabeled SDF-1␣. Cell pellet-associated radioactivity was counted using a LKB-Wallac microcomputer controlled 1272 CliniGamma counter. The binding data were analyzed using a GraphPrad Prism 2.0 software.
Chemotaxis and Intracellular Ca 2ϩ Mobilization-CEMx174 cells were resuspended in chemotaxis medium (RPMI 1640 containing 1% pasteurized human plasma protein (Swiss Red Cross Laboratory) and buffered with 20 mM Hepes, pH 7.3). Cell migration was performed in 48-well chemotaxis chambers (Neuro Probe Inc., Cabin John, MD) as described previously (27). Chemokines diluted in chemotaxis medium were added to the lower and 10 5 CEMx174 cells in the same medium to the upper wells. Polycarbonate membranes with 5-M pores (PC 5 m PVPF, Costar, Cambridge, MA) were used to measure cell migration for 2 h at 37°C. Membranes were removed, and the upper side was washed with PBS, fixed, and stained. Cell migration was assessed at 1000ϫ magnification in five randomly selected fields. For calcium measurements, CEMx174 cells were washed twice with PBS and resuspended (2 ϫ 10 6 /ml) in 20 mM Hepes, 136 mM NaCl, 4.6 mM KCl, 1 mM CaCl 2 , pH 7.4. Aliquots of 1 ml were loaded with 0.8 M Fura2/AM (Fluka AG, Buchs, Switzerland) for 20 min at 37°C. The cells were sedimented and resuspended in fresh medium (1.25 ϫ 10 6 /ml). Calcium mobilization was determined as described previously (8). Single-cell Ca 2ϩ measurements were performed in HeLa cells loaded with Fura-2 (3 M) as described previously (11).
Structural Studies-NMR experiments were acquired at 35°C on a Varian Unity 500 spectrometer operating at 499.84 MHz for 1 H and equipped with a triple resonance z gradient 5-mm probe. Data were processed on a Sun workstation using the VNMR 5.3 program. All two-dimensional proton NMR experiments were acquired in the phasesensitive mode using the hypercomplex scheme (28). A two-dimensional nuclear Overhauser experiment with pure absorption phase was performed in four quadrants. The SDF-1 3/6 sample was prepared by TOCSY and NOESY experiments were collected with mixing times of 80 and 200 ms, respectively, and with 512 t 1 increments of 32 scans each and 2048 points in the t 2 dimension. In both dimensions, the data were apodized with a shifted sine-bell function before Fourier transformation and zero-filled to a final matrix of 4096 ϫ 2048 points.

CXCR4-independent
Binding of SDF-1␣ to Cells-The capacity of SDF-1␣ to bind cell membrane molecules other than CXCR4 was investigated using a novel mAb. The K15C mAb was obtained by immunizing mice with a linear peptide derived from the amino terminus of SDF-1␣. In enzyme-linked immunosorbent assay experiments, the antibody showed specific reactivity with immobilized SDF-1␣ (Fig. 1a) as competition occurred when the antibodies were preincubated with free SDF-1␣ but not with the non-related chemokines RANTES, MIP-1␣, or MIP-1␤. Apart from SDF-1␣ or SDF-1␤ (Fig. 1b) which differ exclusively by four amino acids at the carboxyl terminus domain, the antibody failed to recognize any other known CC or CXC chemokine (data not shown). To characterize further the specificity of epitope recognition by K15C mAb, we performed Western blot experiments using SDF-1␣ derivatives with progressive deletion of amino acids at the amino terminus of SDF-1␣. SDF-1␣ derivatives lacking the two first residues (Lys 1 and Pro 2 ) (29) were not recognized by the antibody. This proves that Lys 1 and Pro 2 are essential residues in the epitope recognized by K15C mAb (Fig. 1b). Importantly, incubation of K15C mAb fully prevented the biological activity of SDF-1␣ as assessed by its capacity to block SDF-1␣-mediated CXCR4 endocytosis in Jurkat cells (Fig. 1c) or chemotaxis of human T lymphocytes (data not shown).
CXCR4-negative CHO-K1 cells (Fig. 2b) were incubated with 10 -1000 nM SDF-1␣ and labeled with the K15C mAb. The cytofluorographic analysis showed that SDF-1␣ bound to the cell surface in a concentration-dependent manner (Fig. 2a). The recognition of the SDF-1␣⅐K15C complex by the secondary antibody was specific since no fluorescence was observed when SDF-1␣ was omitted or replaced by MIP-1␤ (1000 nM) (Fig. 2a).
The capacity of SDF-1␣ to bind to cells in a CXCR4-independent manner is not restricted to a particular cell type. Indeed, SDF-1␣ also bound to CXCR4-negative, endothelial cells (ECV304 cell line, Fig. 2b) with a comparable efficiency as to CHO-K1. Similarly, cell membrane-bound SDF-1␣ was found when the chemokine was incubated with HeLa cells that constitutively express CXCR4 (Fig. 2b). SDF-1␣ binding detected by FACS analysis on HeLa cells was not due to interaction of the chemokine with CXCR4 because the K15C mAb exclusively recognizes the critical amino-terminal residues that are engaged in binding to CXCR4 and are required for SDF-1␣-induced signal transduction (29). Accordingly, we failed to detect SDF-1␣ bound to CXCR4-positive, CEM T lymphoblastoid cells (Fig. 2b). CEM cells express high levels of functional CXCR4 receptors, and saturable binding of SDF-1␣ in these cells has been shown (29). This suggests that in CEM cells, CXCR4 accounts for most of the cellular binding of SDF-1␣. Collectively, our results indicate that other structures apart from CXCR4 have the capacity to attach SDF-1␣ to the cell surface. These interactions apparently do not involve the amino terminus of SDF-1␣ that is recognized by the K15C mAb and is masked after interaction with CXCR4.
SDF-1␣ Binds to Cell Surface GAG-To ascertain whether proteoglycans can account for the CXCR4-independent binding of SDF-1␣ on CHO-K1 or HeLa cells (Fig. 3a), cells were treated for 90 min with enzymes that selective degrade either chondroitin sulfate or HS. The cells were then incubated with 300 nM SDF-1␣ at 4°C for 90 min. Treatment of both cell types with heparinase or heparitinase, which specifically remove HS, dramatically reduced the binding of SDF-1␣ detected by the K15C mAb (Fig. 3a). In contrast, exposure of CHO-K1 or HeLa cells to chondroitinase ABC failed to modify the binding of SDF-1␣. Treatment with trypsin, which cleaves the protein core of proteoglycans, also prevented detection of CHO-K1-or HeLa-bound SDF-1␣ by the antibody (Fig. 3a). Preincubation of SDF-1␣ with soluble HS markedly competed binding of the chemokine to CXCR4-negative parental CHO-K1 cells (data not shown).
The capacity of SDF-1␣ to interact with proteoglycans was further investigated using CHO-derived pgsB618 and pgsD677 cells which lack galactosyltransferase I (pgsB618) or N-acetylglucosaminyltransferase and glucuronyltransferase (pgsD677) activities and consequently are deficient for the synthesis of any GAG or HS, respectively (Fig. 3b). In sharp contrast to the parental CHO-K1 cells, no SDF-1␣ was detected to bind to these cells even when high concentrations of the chemokine were used (Fig. 3b).
In conclusion, CXCR4-independent binding of SDF-1␣ to cell surfaces can be mediated by proteoglycan molecules containing HS as polyosidic components.
Identification of SDF-1␣ Residues Allowing Interaction with GAG-CXC chemokines have been shown to interact with GAGs namely through positively charged sites located in the ␣-helix at the carboxyl terminus (30 -33). We searched for a cluster of residues in SDF-1␣ with a net positive charge that fulfills criteria to form a GAG-binding site. A putative site is formed by the BBXB (B for basic and X any amino acid) sequence Lys 24 -His 25 -Leu 26 -Lys 27 that resembles the proposed consensus for GAG binding in other chemokines (34). According to the three-dimensional structure of SDF-1␣, this motif is located in the first ␤-strand of the ␤-sheet and is exposed on the surface of the molecule (29,35). To investigate the capacity of this site to interact with GAG, we synthesized a SDF-1␣ derivative in which the basic residues Lys 24 , His 25 , and Lys 27 were substituted by Ser (SDF-1 3/6). When applied onto a heparin-Sepharose column, SDF-1 3/6 was eluted at low ionic strength of a linear NaCl gradient (Fig. 4a). In contrast, SDF-1␣, like RANTES which also associates with GAGs (19), requires considerably higher NaCl concentrations to disrupt the interaction with the heparin-Sepharose. To determine whether the reduced affinity of SDF-1 3/6 for heparin correlates with its lower capacity to bind on cells in a CXCR4-independent manner, we incubated CHO-K1 cells with SDF-1␣ or SDF-1 3/6 and determined cell-associated chemokine with the K15C mAb. On Western blots, both SDF-1␣ and SDF-1 3/6 are recognized with similar efficiency by K15C mAb (Fig. 4c). By using increasing concentrations of SDF-1 3/6 up to 1000 nM, we could show that the modified chemokine retained less than 10% of the capacity of SDF-1␣ to bind to cell surface proteoglycans (Fig. 4b). A peptide containing SDF-1␣ amino acids 1-34 (SDF 1-34) had the lowest affinity for heparin, although it included the proposed sequence for interaction with GAG (Fig. 4a). The finding suggests that proper folding of the ␤-strand containing the BBXB motif is critical for binding to HS. However, at this point, we cannot exclude that an additional domain that is not contained in SDF 1-34 may contribute to the overall binding of SDF-1␣.
By comparing the TOCSY and NOESY spectra of SDF-1␣ and SDF-1 3/6 very similar proton chemical shift values are revealed, suggesting that both molecules must fold with a comparable three-dimensional structure. In particular the three-stranded anti-parallel ␤-sheet, the carboxyl-terminal ␣-helix, and the rigid turns are similar for both molecules (see Ref. 29, Protein Data Base access code 1sdf). Of particular interest is the structure of the first ␤-strand of the ␤-sheet where the modifications were made. Strong H␣-H␣ interactions are observed between Ala 40 and Leu 26 and between Gln 48 and Arg 41 (Fig. 5) as well as weak NH-NH interaction between Ala 40 and Val 49 indicating that the three-stranded anti-parallel ␤-sheet is preserved in SDF-1 3/6. The strong NH-NH interaction observed between Lys 24 and His 25 in SDF-1␣ is maintained in SDF-1 3/6 between Ser 24 and Ser 25 (Fig. 5). In addition, the numerous long range dipolar interactions between Trp 57 side chain protons and the first strand of the ␤-sheet are also present (not shown), and all the chemical shift values for this residue are identical for both molecules indicating a similar environment. Although, minor local rearrangements can-not be excluded, it is apparent that the global fold of SDF-1␣ is essentially maintained in the SDF-1 3/6. The similar affinity of SDF-1␣ and SDF-1 3/6 for CXCR4 supports this conclusion (Fig. 8a). These findings further strengthen our view that the cluster of basic residues formed by Lys 24 , His 25 , Leu 26 , and Lys 27 is likely involved in physical contacts with the polyanionic chains of HS and is an essential part of the SDF-1␣ HS-binding site. Characterization of GAG/SDF-1␣ Interactions-The BiAcore technology was used to confirm and to extend the analysis of the SDF-1␣/GAGs interactions in vitro. Surface plasmon resonance (SPR) was used to measure changes in refractive index caused by the binding of SDF-1␣ (analyte) to the immobilized biotinylated heparin. Injection of SDF-1␣ (200 nM, 150 l) over an activated sensor chip containing 120 RU of heparin gave a signal of 350 RU, whereas injection of the chemokine over a control surface (containing streptavidin only) did not lead to any significant signal (Fig. 6a). By using this binding assay, a competition analysis was performed to identify GAG with the capacity to interact with SDF-1␣. For this purpose, SDF-1␣ was preincubated with different GAG and then injected over the heparinized sensorchip. Pretreatment of SDF-1␣ with heparin or HS significantly attenuated the binding. Fifty percent inhibition of binding was obtained at concentrations of approximately 0.5 and 40 g/ml for heparin and HS, respectively (data not shown). Dermatan sulfate showed much milder activity and chondroitin sulfates (A and C) were inactive (Fig. 6b).
To investigate further the nature of SDF-1␣/HS interactions, we performed a kinetic analysis of SDF-1␣ binding to heparin. SDF-1␣ was injected over the Biacore heparin surface in a range of concentrations (usually 0 -200 nM) to produce a set of sensorgrams from which association and dissociation phases could be analyzed (Fig. 6, c and d). Biotinylated heparin was immobilized to levels less than 50 pg/mm 2 (50 RU), and the flow rate was maintained at 50 l/min so that mass transport problems were minimized (not shown). The sensorgrams could be fitted to an A ϩ B ϭ AB model and analyzed by linear transformation. A plot of k s versus SDF-1␣ concentration from the association phase yielded an association rate constant (k on ) of 2.16 ϫ 10 6 M Ϫ1 s Ϫ1 . A dissociation rate constant (k off ) of 0.083 s Ϫ1 was obtained from the direct analysis of the dissociation phase. Thus, binding of SDF-1␣ to heparin is characterized by an affinity, K d (K d ϭ k off /k on ) of 38.4 nM. By contrast, injection of SDF-1 3/6 in the same concentration range did not produced any binding signal (Fig. 6d). Binding of SDF-1 3/6 was observed at high concentrations (up to 6.5 M), but the corresponding sensorgrams could not be fitted to any model (data not shown). This finding may be explained by a nonspecific interaction of the modified chemokine with heparin or, alternatively, by the existence of other basic residues apart from the Lys 24 , His 25 , and Lys 27 cluster that may contribute to form the HS-binding site. Altogether, our findings indicate that the Lys 24 , His 25 , and Lys 27 cluster is critical for physical interaction of SDF-1␣ with HS.
Since 50 RU of immobilized heparin permitted a maximum binding of 300 RU of SDF-1␣, we calculated that each heparin molecule bound to 5 to 6 SDF-1␣ molecules. Conversely, we estimate that each SDF-1␣ molecule should occupy on average 6 monosaccharide units along the GAG chain. This finding strongly suggests that the SDF-1␣/HS interactions could cause the oligomerization of the chemokine. In keeping with this assumption, incubation of SDF-1␣ with heparin results in the formation of SDF-1␣ complexes that can be detected by gel filtration or chemical cross-linking. The calculated apparent molecular weight corresponds to the association of 5-6 molecules of SDF-1␣ per complex (data not shown).
Functional Properties of SDF-1 3/6, CXCR4 Binding, and Activation-To test if the loss of HS-binding capacity affects the functional properties of SDF-1 3/6, both its ability to bind to, and signal through, CXCR4 were compared with that of SDF-1␣. The affinity of SDF-1 3/6 for CXCR4 was not affected (K d SDF-1␣, 1.67 nM; K d SDF-1 3/6, 2.12 nM) (Fig. 7a). Moreover, SDF-1 3/6 induced in CEMx174 both chemotaxis (Fig. 7b) and calcium release from intracellular stores (Fig. 7d) as efficiently as SDF-1␣ (Fig. 7, b and c). Similarly SDF-1␣ and SDF-1 3/6 did not differ significantly on their capacities to stimulate intracellular calcium mobilization via CXCR4 in HeLa cells (Fig. 8, a and b). In agreement with this observation, enzymatic removal of GAG from the surface of HeLa cells did not affect signal transduction through CXCR4 as assessed by SDF-1␣-stimulated calcium mobilization (Fig. 8c). Soluble SDF-1␣⅐HS complexes activate CXCR4 as efficiently as monomeric SDF-1␣, confirming that the signaling domain of the protein remains available for interaction with the receptor in the complexes (Fig. 8d). Overall, those findings indicate that SDF-1␣-CXCR4-and HS-binding domains are distinct and do not overlap. Moreover, they suggest that the primary consequence of the attachment of SDF-1␣ to HS on cell surfaces could be the localization and concentration of the chemokine in the surrounding of CXCR4 rather than increasing signaling capacity through the receptor. DISCUSSION It is conceivable that the interaction with cell surface proteoglycans enhances the physiological activities of SDF-1␣. The reported binding of the majority of blood T lymphocytes to the vascular endothelium induced by SDF-1 (36) could be the consequence of haptotactical recruitment and the arrest of circulating cells by the chemokine attached to GAG on the luminal surface of endothelial cells. Like the hematopoietic growth factor GM-CSF (16), SDF-1␣ could be selectively immobilized by the GAG on the surrounding extracellular matrix or by the membrane of SDF-1␣-producing, bone marrow stromal cells. Concentration of SDF-1␣ in the bone marrow microenvironment could be critical for the pro-hematopoietic activity of the chemokine in embryonic and post-natal life. Similarly, immobilization of SDF-1␣ by attachment to cell surface or extracellular matrix GAG may be essential for the induction of adhesive cell interactions necessary for tissue modeling (5). The specific interaction of SDF-1␣ with HS on the plasma membranes of CXCR4-negative epithelial or endothelial cell lines supports the assumption. The inability of SDF-1␣ to bind CHO mutant cells lacking either any GAGs or, more selectively, HS provides direct evidence that cell membrane-bound HS have the capacity to interact with the chemokine. The remarkable ability of the K15C mAb to recognize the amino-terminal domain of SDF-1␣ enables us to predict that SDF-1␣ bound to cell proteoglycans is in a biologically active form. Indeed, the epitope of SDF-1␣ recognized by K15C mAb is essential for activation of CXCR4 but is not involved in SDF-1␣/GAG interaction. SDF-1␣ variants that lack the two first amino acids (Lys 1 and Pro 2 ) fail to induce chemotaxis, intracellular calcium mobilization, or prevent human immunodeficiency virus entry (29). Accordingly, the K15C mAb which recognizes these residues neutralizes the biological activity of SDF-1␣. Thus, the specific interaction of the antibody with the HS-complexed chemokine implies that the amino terminus domain of SDF-1␣ remains exposed and is available for engagement with the interhelical signaling pocket of CXCR4. Moreover, these findings suggest that the SDF-1␣ domains that interact with CXCR4 or HS are distinct.
The binding of chemokines to GAG is mediated through ionic forces generated by the interactions of negatively charged side chains on GAG with clusters of basic residues in the chemokines (34,37). Heparin and HS are chemically related molecules, both composed of glucosamines and uronic acids (38). Whereas the majority of glucosamines are N-sulfated and the predominant uronate residues are iduronic acids in heparin, HS contains a variable amount of N-acetylated and N-sulfated glucosamines, and the uronic acids occur as iduronates or glucuronates. Within HS, N-sulfated glucosamines and iduronic acids are usually assembled in short heparin-like sequences separated by extended N-acetylated and glucuronate-rich domains (39). These heparin-like sequences could account for the binding to SDF-1␣ through direct interaction with clusters of basic, positively charged residues.
The residues responsible for the binding of the CC chemokines (MCP-1 and MIP-1␣) to heparin are located in distinct domains outside the carboxyl-terminal ␣-helix (34,40). In contrast, mutagenic and biochemical studies of the chemokines IL-8 and PF-4 led to the conclusion that CXC chemokines may bind heparin through a cluster of positively charged residues (mainly Lys) located in the carboxyl-terminal amphipathic ␣-helix (30 -33). Similar conclusions were raised from analysis FIG. 7. Comparative analysis of functional characteristics of SDF-1␣ and SDF-1 3/6. a, binding of 125 I-labeled SDF-1␣ to CXCR4 ϩ CEM cells was competed with either cold SDF-1␣ or SDF-1 3/6. Results are representative of six different experiments (mean Ϯ S.E.). Total specific binding was 164,762 cpm/mg protein for SDF-1␣ and 195,417 cpm/mg protein for SDF-1 3/6. Scatchard analysis of the competition curves yielded linear plots with similar K d values as follows: K d SDF-1␣ ϭ 1.67 Ϯ 0.33 nM and K d SDF-1 3/6 ϭ 2.12 Ϯ 0.089 nM. b, SDF-1␣ and SDF-1 3/6 stimulated chemotaxis of CEMx174 cells. Cell migration was measured following stimulation with the indicated concentrations of chemokines for 2 h. Intracellular Ca 2ϩ mobilization in CEMx174 by SDF-1␣ (c) or SDF-1 3/6 (d) is shown. of the carboxyl-terminal ␣-helix of the chemokines GRO␣ or NAP-2 (41). However, recent findings indicate that the interaction of CXC chemokines with heparin may be mediated by other domains than the carboxyl-terminal ␣-helix. In PF-4, a loop containing 5 positively charged residues (Arg 20 , Arg 22 , His 23 , Lys 31 , Lys 46 , and Lys 49 ) contributes more than the previously identified carboxyl-terminal end Lys residues to the interaction of the chemokine with heparin (37). This situation is reminiscent of what we find for SDF-1␣. Structural differences between SDF-1␣ and other CXC chemokines were reported showing a distinct packaging of the hydrophobic core, orientation of the ␣-helix, and distribution and clustering of electric charges (29,35). Studies of the electrostatic potential map revealed that the carboxyl-terminal ␣-helix of SDF-1␣ possesses a predominant negative surface charge (29) that makes its relevance as a potential HS-binding site very unlikely. In contrast, the basic residues Lys 24 , His 25 , Lys 27 , and Arg 41 clustered along the two first ␤-strands exhibit a high positive potential and represent a suitable region for interaction with HS. The combined substitution of Lys 24 , His 25 , and Lys 27 by Ser confirmed that this cluster of residues represents, or is part of, an exposed region with the required charge complementarity necessary for interaction with heparin and HS. The fact that a truncated, cyclic SDF-1␣ derivative (amino acids 1-34) encompassing the Lys 24 -His 25 -Lys 27 sequence failed to associate with heparin strongly indicates that the HS-binding site formed by Lys 24 -His 25 -Leu 26 -Lys 27 relies on proper folding of the chemokine. The impaired capacity of SDF-1 3/6 to associate with HS is unlikely to be a consequence of misfolding. Indeed, NMR studies confirm that the global folding of SDF-1 3/6, which occupies and activates CXCR4 as efficiently as SDF-1␣, is preserved.
SDF-1␣ shows an overall positive surface charge (ϩ8) that may contribute to an electrostatic interaction with the highly negatively charged extracellular domains of CXCR4 (net charge Ϫ9). On this basis, others (35) have proposed that in addition to the unfolded amino terminus of the chemokine, the cluster formed by residues Lys 24 , His 25 , Lys 27 , and Arg 41 could participate in the binding of SDF-1␣ to CXCR4. The unmodified affinity of SDF-1 3/6 for CXCR4 as compared with SDF-1␣ does not support this hypothesis and suggests that model should be reconsidered. Moreover, the comparable capacity of SDF-1␣ and SDF-1 3/6 to activate CXCR4 provided direct evidence that HS-and CXCR4-binding sites of SDF-1␣ are on opposite faces and do not overlap. According to the three-dimensional structure of SDF-1␣, residues Lys 24 , His 25 , and Lys 27 in the first ␤-strand fold together in a surface that is distant from the CXCR4 binding and signaling domain. This finding further strengthens our viewpoint that the binding and signaling regions of SDF-1␣ will remain free for interaction with CXCR4 in SDF-1␣⅐HS complexes.
Although we propose that the cluster formed by Lys 24 , His 25 , and Lys 27 represents an essential part of the HS-binding site of SDF-1␣, we cannot exclude the contribution of other positively charged amino acids. Thus, basic residues (Arg 20 and Arg 41 ) located in close proximity in the three-dimensional structure of SDF-1␣ could, in addition to the putative site Lys 24 -His 25 -Leu 26 -Lys 27 , form a site with the highest capacity to associate with HS. In this regard, our preliminary evidence indicates that substitution of Arg 20 by Ser further decreases the residual low affinity for heparin of SDF-1 3/6. The contribution of Arg 41 located in the second ␤-strand is being investigated.
The relationships between SDF-1␣/HS interactions and CXCR4 signaling were addressed. Our results indicate that neither binding nor signaling of CXCR4 are affected by the association SDF-1␣ with HS. Studies on the consequences of GAG/chemokine interactions for receptor-mediated signal transduction have provided conflicting results. Substitution of basic residues in the carboxyl-terminal ␣-helix of either PF-4 (31) and MCP-1 (40) prevented binding to GAGs without affecting receptor binding and signal transduction. Similarly, mutation of Lys 45 to Ala in MIP-1␣ which prevents binding to heparin does not affect binding and activation of CCR1 expressed either in parental CHO-K1 or HS-deficient CHO cell mutants (34). In contrast, it was reported that an IL-8 analog with a truncated carboxyl-terminal ␣-helix does not associate with heparin and lacks receptor binding capacity and signaling activity (30). Furthermore, it was shown that enzymatic removal of cell membrane GAG reduces substantially the capacity of IL-8, RANTES, MCP-1, and MIP-1␣ to bind CXCR1, CCR1, or CCR2 expressed on CHO cells (19). A more recent report states that degradation of cell surface-bound GAG affects receptor binding and Ca 2ϩ influx induced by RANTES on peripheral blood leukocytes (42). However, we were unable to confirm these data (data not shown).
The apparent disparity of those findings could be accounted for by the different experimental protocols used. Single amino acid mutations or large deletions of putative GAG-binding domains were made to prevent chemokine/GAG interactions. Enzymatic degradation by glycanases or GAGs-defective cell mu- FIG. 8. Activation of CXCR4 by SDF-1␣ complexed to HS. Comparative analysis of the intracellular mobilization of Ca 2ϩ induced in HeLa cells by 30 nM SDF-1 3/6 (b) or SDF-1␣ either alone (a) or complexed with soluble HS (50 g/ml) (d). a, b, and d, HeLa cells were loaded for 30 min with Fura-2 before stimulation (arrow) as indicated. c, HeLa cells were treated with 1 milliunit/ml glycanases (heparitinase, heparinase, and chondroitinase ABC) for 1 h, loaded for 30 min with Fura-2, and then stimulated with 30 nM SDF-1␣ (arrow). tants were employed to analyze the capacity of cell-surface proteoglycans to associate with chemokines. Further studies will be required to elucidate fully the role of GAG on the activation of the receptors by the chemokines. We believe that chemokines carrying single amino acids substitutions, like SDF-1 3/6 that prevent association with GAG, can be very useful to assess their biological activity in vivo.
As defined by SPR analysis, each SDF-1␣ molecule occupied a hexasaccharide on heparin. Naturally occurring HS chains typically contain multiple heparin-like domains that integrate up to 16 -18 saccharidic residues. It is conceivable that for cell-bound HS, each of the domain repeats can account for SDF-1␣ oligomerization. The high affinity of SDF-1␣ for heparin (K d ϭ 38.4 nM) suggests that in vivo SDF-1␣ is essentially bound to HS. This hypothesis is further supported by the SPR kinetic analysis which demonstrates that the HS⅐SDF-1␣ complex formation is characterized by a high on-rate constant (k on ϭ 2.16 ϫ 10 6 M Ϫ1 s Ϫ1 ).
Although SDF-1␣ and HS interact with relative high affinity, the stronger affinity for CXCR4 permits the preferential interaction of the chemokine with CXCR4 through its free, exposed amino terminus carrying the signaling domain of SDF-1␣. Binding to membrane-associated HS causes the presentation of functional, oligomerized SDF-1␣ at the surface of cells either producing or passively expressing the chemokine. As a consequence, SDF-1␣ becomes available at enhanced concentrations in the close proximity of CXCR4 receptors expressed on cells attracted by cell membrane-bound HS⅐SDF-1␣ complexes. This, rather than the modification of the intrinsic capacities of the chemokine to bind and signal through CXCR4, could be the major biological role of the interaction of SDF-1␣ with cellsurface HS.