Heparin Oligosaccharides Inhibit Chemokine (CXC Motif) Ligand 12 (CXCL12) Cardioprotection by Binding Orthogonal to the Dimerization Interface, Promoting Oligomerization, and Competing with the Chemokine (CXC Motif) Receptor 4 (CXCR4) N Terminus*

Background: GAG/CXCL12 interactions are critical for chemokine function but co-administration may abrogate their individual cardioprotective effects in a clinical setting. Results: Biophysical studies distinguish CXCL12 residues involved in dimerization from those likely to contact heparin directly. Conclusion: CXCL12 dimerization is required for high affinity heparin binding and protects N-terminal degradation. Significance: Knowledge of the GAG-binding site will enable future development of heparin-insensitive CXCL12 therapeutics. The ability to interact with cell surface glycosaminoglycans (GAGs) is essential to the cell migration properties of chemokines, but association with soluble GAGs induces the oligomerization of most chemokines including CXCL12. Monomeric CXCL12, but not dimeric CXCL12, is cardioprotective in a number of experimental models of cardiac ischemia. We found that co-administration of heparin, a common treatment for myocardial infarction, abrogated the protective effect of CXCL12 in an ex vivo rat heart model for myocardial infarction. The interaction between CXCL12 and heparin oligosaccharides has previously been analyzed through mutagenesis, in vitro binding assays, and molecular modeling. However, complications from heparin-induced CXCL12 oligomerization and studies using very short oligosaccharides have led to inconsistent conclusions as to the residues involved, the orientation of the binding site, and whether it overlaps with the CXCR4 N-terminal site. We used a constitutively dimeric variant to simplify the NMR analysis of CXCL12-binding heparin oligosaccharides of varying length. Biophysical and mutagenic analyses reveal a CXCL12/heparin interaction surface that lies perpendicular to the dimer interface, does not involve the chemokine N terminus, and partially overlaps with the CXCR4-binding site. We further demonstrate that heparin-mediated enzymatic protection results from the promotion of dimerization rather than direct heparin binding to the CXCL12 N terminus. These results clarify the structural basis for GAG recognition by CXCL12 and lend insight into the development of CXCL12-based therapeutics.

The ability to interact with cell surface glycosaminoglycans (GAGs) is essential to the cell migration properties of chemokines, but association with soluble GAGs induces the oligomerization of most chemokines including CXCL12. Monomeric CXCL12, but not dimeric CXCL12, is cardioprotective in a number of experimental models of cardiac ischemia. We found that co-administration of heparin, a common treatment for myocardial infarction, abrogated the protective effect of CXCL12 in an ex vivo rat heart model for myocardial infarction. The interaction between CXCL12 and heparin oligosaccharides has previously been analyzed through mutagenesis, in vitro binding assays, and molecular modeling. However, complications from heparin-induced CXCL12 oligomerization and studies using very short oligosaccharides have led to inconsistent conclusions as to the residues involved, the orientation of the binding site, and whether it overlaps with the CXCR4 N-terminal site. We used a constitutively dimeric variant to simplify the NMR analysis of CXCL12-binding heparin oligosaccharides of varying length. Biophysical and mutagenic analyses reveal a CXCL12/ heparin interaction surface that lies perpendicular to the dimer interface, does not involve the chemokine N terminus, and partially overlaps with the CXCR4-binding site. We further demonstrate that heparin-mediated enzymatic protection results from the promotion of dimerization rather than direct heparin binding to the CXCL12 N terminus. These results clarify the structural basis for GAG recognition by CXCL12 and lend insight into the development of CXCL12-based therapeutics.
Cardiovascular disease is the leading cause of death in the United States, and greater than 50% of these fatalities are asymptomatic (1). The lack of warning for acute cardiovascular injuries, such as myocardial infarctions, underscores the need for highly effective clinical therapeutics. To reduce the possibility of thrombosis at the site of infarction, the American College of Cardiology and the American Heart Association recommend the administration of anticoagulant drugs, such as heparin, during medical treatment for myocardial infarctions (2). Heparin actuates antithrombin, which then inactivates proteins in the coagulation pathway (3). Heparin is not a single molecule but rather describes a family of glycosaminoglycans (GAGs) 2 with specific polysaccharide components of varying length and sulfation. The most common formulations in clinical use are unfractionated heparin and low molecular weight heparin. The molecular mass of unfractionated heparin is typically between 5 and 30 kDa, whereas low molecular weight heparin is usually 1-8 kDa (1). Endogenously, heparan sulfate, structurally related to heparin, is intrinsic to the cell migration properties of chemokines, a family of ϳ50 structurally homologous chemotactic cytokines.
Chemokines induce the migration of cells by activating G protein-coupled receptors. The direction of chemotaxis is encoded in a concentration gradient; secreted chemokines are immobilized, by the heparan sulfate GAG chains of proteoglycans, along the endothelial surface. This establishes a gradient with the highest concentrations closest to the origin of secretion (4 -6). Originally recognized as the major promoter of leukocyte trafficking, chemokines have been found to participate in a variety of developmental and housekeeping roles, including lymphopoiesis, myelopoeisis, embryogenesis (7), angiogenesis (8), cardiogenesis (9), neuron migration, and cerebral development (10,11).
Although there is ambiguity regarding the cardioprotective mechanism, researchers are aggressively pursuing therapies as evidenced by numerous methods for chemokine administration to ischemic tissue. These include: CXCL12 attached to poly(ethylene glycol)-fibrin patches (18), intravenously introducing self-assembling nanofibers with a timed chemokine release mechanism (15), and injection of viruses engineered to produce protein (21). In each case, CXCL12 treatment is effective, but variations in protection may be related to stability, dosage, or drug interactions. We recently demonstrated that CXCL12 is cardioprotective over a narrow concentration range, whereas a constitutively dimeric form of the chemokine exerts little to no effect (19). Because heparin is commonly administered in response to myocardial infarctions (2) and induces CXCL12 dimerization (25,26), we hypothesized that the protective properties of CXCL12 may be attenuated by interactions with soluble heparin in a modern clinical setting.
The vital role of GAG binding in CXCL12-directed migration has motivated a series of studies, but its impact on chemokine activity, stability, and receptor interactions remains poorly defined. CXCL12 recognition by heparin oligosaccharides has been characterized using mutagenesis, in vitro binding assays, NMR, crystallography, and molecular modeling (27)(28)(29)(30)(31). We showed previously that heparin disaccharides bind preferentially to the CXCL12 dimer (26), and others have proposed a binding mode localized to the dimer interface (27,28). Consequently, we speculated that mutations previously used to define the binding site might reduce heparin affinity indirectly by limiting dimerization.
Using the previously described preferentially monomeric (CXCL12 H25R ) and constitutively dimeric (CXCL12 2 ) variants, we examined the affinity of distinct CXCL12 oligomers for heparin oligosaccharides of varying length. To deconvolute chemical shift perturbations caused by heparin binding from those reporting on dimerization, we monitored oligosaccharide binding to the CXCL12 2 dimer by two-dimensional NMR. Heparininduced chemical shift changes and mutagenesis of basic side chains was used to further refine the GAG binding interface. Binding studies with a CXCR4 N-terminal peptide demonstrate partial overlap and competition with the heparin interface. In contrast to previous reports that heparin binding to Lys-1 protects CXCL12 from cleavage by dipeptidyl peptidase-IV (DPPIV/CD26), NMR, surface plasmon resonance (SPR), and enzymatic assays demonstrate that reduced cleavage rates are a result of heparin-induced dimerization rather than direct association with the N terminus.
Our results suggest that heparin blocks the cardioprotective effect of CXCL12 by promoting dimerization. Heparin binds a site orthogonal to the dimer interface that partially overlaps the CXCR4 N terminus binding site and does not involve Lys-1. In principle, CXCL12 cardioprotection may be enhanced by modifications that alter its ability to self-associate or interact with heparin.

EXPERIMENTAL PROCEDURES
Protein Engineering, Expression, and Purification-CXCL12, CXCL12 H25R , and CXCL12 2 were expressed and purified as previously described (26,32). All other chemokine variants were produced via mutagenesis of CXCL12 and CXCL12 2 constructs with complementary primers and the QuikChange sitedirected mutagenesis kit (Stratagene) per the manufacturer's instructions. All expression vector inserts were confirmed by DNA sequencing.
Ex Vivo Cardioprotection Assay-Isolated hearts were perfused for 15 min, followed by aerobic perfusion 15 min prior to 30 min of global, no-flow ischemia, and 180 min of reperfusion. Aerobic perfusion including either 50 nM CXCL12, 50 units ml Ϫ1 heparin, or 50 nM CXCL12 plus 50 units ml Ϫ1 heparin. Hearts subjected to the above mentioned perfusion sequence in the absence of both CXCL12 and heparin served as ischemic controls. Hearts perfused continuously for 245 min served as nonischemic controls. Administration of the perfusion sequence, monitoring of cardiac function, and analysis of infarcted tissue was performed as previously described (19,33,34). Each dosage group consisted of four replicates. Resistance to injury from myocardial infarction/reperfusion was determined by a reduction in infarct size and/or an increase in recovery of developed pressure.
NMR Spectroscopy-NMR experiments were performed on a Bruker DRX 600 instrument equipped with a 1 H, 15 (35). HCCH-TOCSY spectra were collected on 0.4 mM [U-15 N, 13 C]CXCL12 2 , 25 mM D-MES buffer (pH 6.8), 200 mM NaCl, 10% (v/v) D 2 O, and 0.02% (w/v) NaN 3 in the presence of 0.0, 0.3, and 0.6 mM heparin decasaccharide of degree of polymerization (dp)10. Partial CXCL12 2 side chain assignments were transferred from previously published chemical shift table (Biological Magnetic Resonance Bank identification code 15633), and complete assignments were achieved with three-dimensional HCCH-TOCSY and three-dimensional 15 N-edited NOESY-HSQC. All of the experiments were performed at 298 K.
Binding site competition between CXCR4 1-38 and tetrasaccharide dp4 was assessed using 15  Preparation of Heparin Biochip and SPR Analysis-The biotinylated heparin was immobilized to a streptavidin chip based on the manufacturer's protocol. The successful immobilization of biotinylated heparin was confirmed by the observation of a 250-resonance unit increase in the sensor chip. SPR measurements were performed on a BIAcore 3000 (GE Healthcare). The buffers were filtered and degassed prior to analysis. The protein samples were diluted in HBS-EP buffer (0.01 M HEPES, 150 mM NaCl, 3 mM EDTA, 0.005% surfactant P20, pH 7.4). Different dilutions of protein samples were injected at a flow rate of 30 l min Ϫ1 . At the end of the sample injection, the same buffer was flowed over the sensor surface to facilitate dissociation. After a 3-min dissociation time, the sensor surface was regenerated by injecting 30 l of 2 M NaCl. For competition with immobilized heparin, protein (50 nM) was mixed with certain concentrations of low molecular mass heparin (dp avg 20), heparin oligosaccharides (di (dp2), tetra (dp4), hexa (dp6), etc.) in HBS-EP buffer and injected over the heparin chip at a flow rate of 30 l min Ϫ1 . For receptor competition studies, CXCL12 2 (50 nM) was premixed with CXCR4 1-38 (500 nM) and injected over the heparin chip. The response was monitored as a function of time at 298 K.
Fluorescence Polarization Assay-Fluorescence polarization assays were performed on a fluorometer (PTI) equipped with automated polarizers at 298 K. Titrations were conducted by monitoring the intrinsic tryptophan fluorescence polarization as a function of CXCL12 variant concentration in 100 mM NaPO 4 (pH 7.4), using emission and excitation wavelengths of 283 and 325 nm, respectively. All of the samples were filtered and degassed prior to analysis. The dimerization equilibrium dissociation constants (K d ) were determined by fitting polarization data as previously described (26).
Calcium Response Assay-The cells were washed twice and resuspended in 96-well format at 2 ϫ 10 5 cells well Ϫ1 in assay buffer: Hanks' buffered saline solution, 20 mM HEPES (pH 7.4), 0.1% (w/v) BSA, and FLIPR Calcium4 dye (Molecular Devices) and then incubated for 1 h at 310 K with 5% CO 2 . Fluorescence was measured at 310 K using a FlexStation3 Microplate Reader (Molecular Devices) with excitation and emission wavelengths at 485 and 515 nm, respectively. After a 20-s base-line measurement, 30 nM CXCL12 variants were added, and the resulting calcium response was measured for an additional 50 s. Calcium flux was normalized to the response of CXCL12 or CXCL12 2 . The samples were prepared in quadruplicate (n ϭ 2). DPPIV

Heparin Attenuates the Cardioprotective Effect of CXCL12-
CXCL12 is a potential anti-ischemic drug that has been observed to decrease the amount of infarcted tissue and improve cardiac function when administered either pre-or post-myocardial infarction (12-14, 16, 18). We previously showed that CXCL12 cardioprotection is mediated by the monomeric species (19). The antithrombotic properties of heparin are well established and highlighted by its presence as a clinical mainstay for the treatment of myocardial infarction. Because soluble heparin binds to and promotes dimerization of the chemokine (25,26) and inhibits cell migration in vitro (30), we hypothesized that this interaction would also attenuate CXCL12-mediated cardioprotection.
Using an isolated rat heart model of ischemia/reperfusion injury, the cardioprotective effect of CXCL12 was measured in the presence of medicinal heparin (average molecular mass ϭ 14 kDa; dp avg 42) (36). Increased resistance to injury from myocardial ischemia/reperfusion was determined by a reduction in infarct size and/or increase in recovery of left ventricle diastolic pressure. Whereas CXCL12 (50 nM) significantly reduced both measures of ischemia/reperfusion injury in isolated buffer-perfused rat heart relative to untreated controls, inclusion of heparin sodium (50 units ml Ϫ1 ) completely eliminated the cardioprotective effect of CXCL12 (Fig. 1). This suggests that in a clinical setting, co-administration of heparin could attenuate the therapeutic benefit of CXCL12, underscoring the value of the molecular details for this chemokine/GAG interaction.
Heparin Dodecasaccharide Produces Substantial Line Broadening in CXCL12-Heparin-insensitive CXCL12 variants were identified by mutagenesis of the BBXB motif (residues 24 -27) more than 10 years ago. Nonetheless, structural data on the binding interface is limited to a crystal structure of a heparin disaccharide bound to CXCL12 and NMR chemical shift data for a disaccharide and octasaccharide (30,31). We monitored the binding of longer chain heparin oligosaccharides using twodimensional 1 H-15 N HSQC NMR spectroscopy. Residues that exhibit large relative changes in chemical shift correlate with significant involvement in heparin binding-identifying the interface. Titration of 50 M CXCL12 with increasing concentrations of heparin dodecasaccharide (0 -150 M) produced chemical shift changes and line broadening over a large area of the molecule including substantial perturbations clustered at the dimer interface (Fig. 2). Because heparin binding promotes CXCL12 dimerization (25,26), we concluded that heparin-induced shifts arose from a combination of GAG-protein and protein-protein contacts.
High Affinity GAG Binding Requires CXCL12 Dimerization and Is Chain Length-dependent-To simplify analysis of the binding interaction by removing the monomer-dimer equilibrium, we tested whether the covalent dimer CXCL12 2 or preferential monomer CXCL12 H25R are appropriate models for CXCL12-GAG binding (26,32). SPR was used to measure the interaction of immobilized porcine intestinal heparin (average molecular mass ϭ 14 kDa; dp avg 42) with CXCL12, CXCL12 H25R , and CXCL12 2 ( Table 1). Our results indicate that both monomeric and dimeric CXCL12 bind heparin but exhibit different rates of association and dissociation. Not surprisingly, dimeric CXCL12 2 binding is 200 times stronger than monomeric CXCL12 H25R . Wild type CXCL12 binds with an intermediate affinity that presumably reflects the relative contributions of monomers and dimers in equilibrium.
Next, we used SPR to measure the ability of immobilized heparin to displace CXCL12 from soluble heparin fragments of varying length. CXCL12 variants (50 nM) and various oligosaccharides (dp2 to dp20) were premixed and presented to immobilized heparin (average molecular mass ϭ 14 kDa; dp avg 42). Supplemental Fig. S1 illustrates the changes in response units as a function of heparin concentrations that were used to calculate the IC 50 and K i values ( Table 2). For CXCL12 and CXCL12 H25R , little or no displacement occurred for complexes with dp2, dp4, dp6, and dp8 heparin, but dp10 and longer oligosaccharides reduced binding to the immobilized heparin. The dp20 produced a sharp change in CXCL12 and CXCL12 H25R response units characteristic of a change in complex stoichiometry. This is consistent with longer heparin chains increasingly promoting dimerization (25). It is also possible that dp20 may induce higher order oligomerization. The similar response of CXCL12 H25R to wild type is consistent with its 10-fold weaker dimerization affinity (26). In contrast to CXCL12 and CXCL12 H25R , the affinity of CXCL12 2 for heparin increases in direct proportion to oligosaccharide length. CXCL12 2 had the tightest binding affinity and its association with the immobilized heparin was reduced the greatest by soluble heparin, regardless of size.    Fig. 3B, the heparin tetrasaccharide (dp4) produced perturbations confined mainly to residues 23-30 (␤1 strand), 35-43 (␤2 strand), and 47-50 (␤3 strand). Titration of heparin octasaccharide (dp8) induced corresponding shifts and additionally included residues 17-19 in the N-loop (Fig. 3C). Increasing the oligosaccharide length resulted in line broadening across the ␤-sheet and further extension of shift perturbations toward the helices of CXCL12 2 (Fig. 3D). The addition of 0.25 molar equivalents of octadecasaccharide (dp18) resulted in substantial sample precipitation, which precluded chemical shift mapping. However, the similar SPR response of dp16 and dp20 in the presence of CXCL12 2 suggests there are no substantial rearrangements of the binding interface at oligosaccharides greater than hexadecassacharide (dp16) (supplemental Fig. S1).
To identify specific interactions with positively charged side chains, we monitored changes in the HCCH-TOCSY of CXCL12 2 induced by heparin decasaccharide. The effects of heparin binding on each NMR signal were categorized as unaffected, perturbed, or exchange-broadened. Perturbed or exchange-broadened signals were assumed to identify residues proximal to the heparin-binding site. Fig. 3 illustrates the difference in HCCH-TOCSY signals for arginine (Fig. 3E) and lysine (Fig. 3F) residues in the absence and presence of 1.5 equivalents of dp10 heparin. The side chains of Arg-20, Arg-41, and Arg-47 display complete line broadening upon decasaccharide addition as expected from previous mutagenesis results (27,30). Lys-24, Lys-27, and Lys-43 were also previously implicated in GAG binding (27,28) and are similarly affected by dp10 binding.
The side chains of Arg-8 and Arg-12 display a mixture of perturbed and broadened resonances; the broadened H ␣ , H ␤ , and H ␥ resonances and perturbed H ␦ suggest a nominal contribution by the guanidinyl groups. Lys-54, Lys-56, and Lys-64 had varying degrees of incomplete broadening in H ␣ , H ␤ , H ␥ , and H ␦ with no perturbation in H ⑀ , suggesting no direct role in heparin binding through the primary amine. Consistent with the Lys-68 H N resonance in the presence of all oligosaccharides tested, the side chain nuclei do not respond to decasaccharide and appear to be uninvolved in heparin binding.
Dissecting the Contribution of CXCL12 Residues to Dimerization and Heparin Binding-Several CXCL12 variants have been produced with mutations in Lys-24, Lys-27, or both that have a diminished affinity for heparin while retaining the ability to bind the CXCR4 receptor, promote calcium flux, and induce cell migration (27,28). Because these residues are located in the ␤1-strand along the dimer interface, we hypothesized that either the mutations directly affect heparin binding or they inhibit dimerization and, indirectly, heparin binding. Fluorescence polarization was used to measure the dimerization dissociation constant of CXCL12(K24S), CXCL12(K27S), and CXCL12(K24S/K27S) (supplemental Fig. S5). The CXCL12(K27S) variant had a K d ϭ 131 Ϯ 28 M similar to measurements for CXCL12 (K d ϭ 140 Ϯ 19 M (26)), whereas CXCL12(K24S) and CXCL12(K24S/K27S) exhibited a reduced ability to dimerize with K d ϭ 432 Ϯ 92 and 327 Ϯ 76 M, respectively. To directly assess the effect of each substitution on heparin binding, we introduced the K24S, K27S, and K24S/ K27S mutations into CXCL12 2 and measured the binding affinity to immobilized heparin (average molecular mass ϭ 14 kDa; dp avg 42) using SPR (Table 3). Neither the single mutants nor the K24S/K27S variant reduced the CXCL12 2 binding affinity by more than 2-fold. This stands in stark contrast to the Ͼ1,000-fold change in affinity the mutations produce in the wild type CXCL12 background and underscores the importance of dimerization in heparin binding.
Analysis of our NMR titration data suggests a key role for Lys-27, Arg-41, and Arg-47 in heparin binding. Mutagenesis studies were undertaken to measure the contributions of these residues in direct heparin binding by producing alanine mutants in both the CXCL12 and CXCL12 2 background. As illustrated in supplemental Fig. S6, the function of each variant was confirmed via measurement of the calcium flux response. Mutations in the CXCL12 2 background report only on heparin binding, whereas the affinity of CXCL12 background mutants reflects both direct and indirect effects. The binding affinity of each mutant for immobilized heparin (average molecular mass ϭ 14 kDa; dp avg 42) was determined by SPR (Table 3). In the CXCL12 background, single or multiple amino acid substitutions produce substantial changes in affinity. In contrast, single and double residue substitutions have little effect on CXCL12 2 binding: mutagenesis of all three basic residues is necessary to significantly reduce heparin binding. These results  are consistent with a model in which clusters of positively charged side chains in two CXCL12 monomers join to form an elongated heparin-binding site that crosses the six-stranded ␤ sheet of a CXCL12 dimer. The Heparin Interface Partially Overlaps the CXCR4 N-terminal Binding Site-CXCR4-mediated cell migration is presumed to require that GAG-immobilized CXCL12 oligomers in the extracellular matrix be presented to the receptor, which transmits the external migratory signal to the internal cellular machinery. The initial interaction with CXCR4 is mediated through the receptor N terminus; however, inconsistencies exist as to whether this binding site overlaps with that of heparin (28,30). The heparin-binding surface defined by chemical shift mapping is shared with the receptor-binding surface defined in structural studies with CXCR4 N-terminal peptides, including key sulfotyrosine binding residues Arg-47 and Lys-27 (32,37). We tested the ability of an unsulfated CXCR4 N-terminal peptide comprising the first 38 amino acids, CXCR4 1-38 , to either displace dp4 or form a ternary complex by binding simultaneously to a complementary site. Addition of CXCR4 1-38 had essentially no effect on the HSQC spectrum of dp4-bound CXCL12 2 (Fig. 4, A and B), demonstrating that heparin competes effectively with CXCR4 for CXCL12 binding. Using SPR, we tested whether the large chain immobilized heparin oligosaccharides (average molecular mass ϭ 14 kDa; dp avg 42) could compete CXCL12 2 from CXCR4 1-38 . As illustrated in Fig. 4C, CXCL12 2 (50 nM) preincubated with CXCR4 1-38 (500 nM) was unable to bind the immobilized heparin despite a 1000-fold higher affinity for the heparin.
Heparin Indirectly Protects the N Terminus from CD26/DP-PIV Degradation-DPPIV/CD26 is a well known exopeptidase found in both a membrane-bound and a soluble form. It recognizes CXCL12 and cleaves between residues 2 and 3 to produce a truncated variant with a reduced affinity for, and an inability to activate, CXCR4 (38). Sadir et al. (29) used Western blot analysis to demonstrate that heparin molecules of increasing chain length protected CXCL12 from DPPIV/CD26-mediated degradation and attributed this effect to the direct association of heparin with Lys-1. To detect the participation of Lys-1 in the CXCL12 2 /heparin interaction, we monitored dp10-induced chemical shift perturbations by HCCH-TOCSY. Although   dodecasaccharides or larger were previously reported necessary for binding Lys-1 (29), our HSQC analysis attests the CXCL12 N-terminal residues 1-8 are similarly perturbed by dp10 and dp12; therefore, the decasaccharide was chosen as a compromise between sufficient oligosaccharide length and improved solubility. As illustrated in Fig. 5A, the sharp resonances of Lys-1 nuclei and residues 2-7 (data not shown) indicate little role for the primary amine or N terminus in oligosaccharide binding. Similarly, deletion of residues 1 and 2 (⌬1-2) from either CXCL12 or CXCL12 2 had no effect on binding long chain heparin by SPR (Fig. 5, B and C, and Table 3). We recently demonstrated that the half-life of the CXCL12 2 N terminus is 25-fold greater than CXCL12 in the presence of DPPIV/CD26 (39) and therefore hypothesized that heparin may indirectly protect Lys-1 by promoting CXCL12 dimerization. We incubated 10 M CXCL12 and 0.2 ng l Ϫ1 DPPIV/ CD26 alone or in the presence of dp2 or dp4 heparin. The reaction was monitored by following CXCL12(3-68) product formation using MALDI-TOF MS (supplemental Fig. S7). Relative quantification was achieved by normalizing product intensities to a 2 M [U-15 N]CXCL12(3-68) internal standard. Similar to our previous CXCL12 half-life of 26.5 Ϯ 8.7 min (39), herein the buffer control exhibited a half-life of 41.5 Ϯ 10 min. As shown in Fig. 6, protection of CXCL12 N-terminal cleavage is dependent on the size of heparin. Whereas dp2 does not provide protection (1 ⁄ 2 ϭ 17.1 Ϯ 2.4 min), dp4 slows the half-life to 1257 Ϯ 75 min. The long term stability of DPPIV/CD26 was confirmed by measuring the half-life of CXCL12 degradation after 4 days (data not shown). The drastic difference in heparinpromoted dimerization between dp2 and dp4 is consistent with previous mass spectrometry studies at similar protein and oligosaccharide concentrations (25). We collected two-dimensional HSQC spectra at CXCL12 and heparin concentrations (dp2 and dp4) identical to those used in the DPPIV assay to look for signs of self-association (supplemental Fig. S8). Whereas dp2 induces little change in CXCL12, dp4 promotes chemical shift perturbations and line broadening consistent with dimerization.

DISCUSSION
Heparin is a well established antithrombotic agent and a clinical mainstay for the treatment of myocardial infarction. More recently, the anti-ischemic effects of CXCL12 have led to several novel treatment strategies. CXCL12 binds to soluble heparin and extracellular matrix GAGs. In the present study, we monitored the interaction of CXCL12 with soluble heparin by several methods and defined a binding site that may also be relevant to immobilization in the extracellular matrix. Although interactions with cell surface GAGs are essential for   chemotactic cell migration, soluble GAGs can block chemokine activity. We hypothesized that simultaneous administration of chemokine and heparin would diminish the protective properties of CXCL12 rather than synergizing their individual beneficial effects. As predicted, co-administration inhibited the cardioprotective effect of CXCL12 in an isolated rat heart model. This model reports directly on the function of cardiac tissue. If recruitment of progenitor cells is critical to the cardioprotective mechanism, the effect of soluble heparin administration on chemotaxis is unclear. In vitro studies suggest that heparin would inhibit the migration of cells (30). Further, the effect of CXCL12 on heparin activity is completely unknown and warrants future exploration.
The interaction between CXCL12 and heparin has been the focus of five papers over the last 13 years. The original mutagenic characterization by Amara et al. (28) identified residues Lys-24, His-25, and Lys-27 at the dimer interface as a canonical BBXB heparin-binding motif that is distinct from the CXCR4 N-terminal binding site. In 2001, Arg-41 and Arg-43 were proposed as additional contributors to the heparin binding interface. In the proposed structural model, heparin bound along the CXCL12 dimer interface, primarily making contacts with the ␤1-strand of each protomer (27). Later, a crystal structure of CXCL12 in complex with a heparin disaccharide analog identified two distinct binding sites: one high affinity site at the dimer interface and a second site between the N-loop region and the C-terminal helix (30). Finally, Sadir et al. (29) further expanded the proposed binding site to include the N termini; they concluded that the direct interaction of Lys-1 with heparin was responsible for limited proteolysis by DPPIV/CD26. Despite the wealth of data published on this crucial interaction, there are several inconsistencies in the predicted binding mode.
CXCL12/Heparin Interface-The monomer-dimer equilibrium and lack of data using long chain heparin molecules has significantly hindered the structural characterization of CXCL12-heparin complexes. We used the CXCL12 2 dimer to eliminate complications from CXCL12 self-association and monitor interactions with isolated heparin molecules ranging in size from dp2 to dp20. Whereas titration of heparin dodecasaccharide into wild type CXCL12 produced substantial line broadening throughout the dimer interface, CXCL12 2 enabled visualization of heparin-specific contacts. Our results differ from previous structural models. Instead of an interface mostly comprising the ␤1 strands and N terminus of each protomer in a dimer (Fig. 7A) (27,29,31), our data suggest that heparin drives dimer formation by interacting nearly orthogonal to the dimerization interface. We propose that the interface between CXCL12 and heparin crosses the entire six-stranded ␤-sheet, does not involve the N terminus, and is primarily mediated by Arg-20, Lys-24, Lys-27, Arg-41, Lys-43, and Arg-47 with the potential peripheral involvement of Arg-8 and Arg-12 (Fig. 7B).
A recently published model illustrated the octasaccharidebinding site more or less along the dimer interface; the terminal sugars interact with Arg-8 and Arg-12 on each protomer, whereas the core saccharides contact Lys-24, His-25, Lys-27, and Arg-41 (31). Although both Arg-8 and Arg-12 exhibit modest 1 H-15 N chemical shift perturbations and broadening of H ␣ , H ␤ , and H ␥ resonances, the H ␦ spins are minimally perturbed and indicate little role for the guanidinyl groups in heparin binding. These results are consistent with previous R8E/R12E mutations that actually increased CXCL12 affinity for low molecular weight heparin (30). Perhaps removal of Arg-8 and Arg-12 enhances the interaction between heparin and basic residues in the chemokine cleft formed by the N-loop and ␤3 strand.
It should also be noted that the biological role for chemokine/GAG interactions is varied and context-dependent. Although there is a possibility that heparin and heparan sulfate recognize distinct binding sites, the similarity of their oligosaccharide structures and the inhibitory effect of heparin in vivo suggest that they share a single, overlapped binding site. At the same time, there may be important differences in the contributions of individual amino acid or glycan residues to the interaction. Our studies have identified residues that are likely to be involved in binding endogenous GAGs, but the unique structural features of heparin (i.e., greater charge density compared with heparan sulfate and other extracellular matrix GAGs) may lead to slight differences in their binding modes.
Heparin and CXCR4 Sites Overlap-Previous analyses disagree as to whether the heparin and CXCR4 binding sites overlap (28,30). Using a N-terminal peptide of CXCR4, we directly demonstrate competition between the receptor and heparin. The inability of the CXCR4 receptor peptide to dissociate dp4 does not prove that a native N terminus would also be insufficient. The CXCR4 N terminus can be O-sulfated at three tyrosine residues that together increase complex affinity from 4.5 Ϯ 2.2 to 0.2 Ϯ 0.2 M (40). However, our results suggest that some of the surfaces occupied by CXCR4 sulfotyrosines may also be utilized for recognition of the sulfate moieties decorating GAG molecules. Considering the structural similarity between heparin and endogenous heparan sulfate, we speculate that similar competition occurs in vivo.
CXCL12 N Terminus Does Not Bind Heparin-The initial characterization of GAG binding excluded a role for Lys-1, as an antibody specific to the Lys-1-Pro-2 dipeptide was able to recognize CXCL12 bound to cell surface heparin (28). However, the same antibody was later used to show that heparin FIGURE 7. Schematic representation of the heparin-CXCL12 binding mode. A, the previously published binding mode proposed that heparin (orange) associates with CXL12 (gray) along the dimerization interface, primarily contacting the ␤1 strands and the Lys-1 residue of each protomer. Residues highlighted on the CXCL12 surface were identified in previous studies by Lortat-Jacob and co-workers (27-29) (green), Lolis and co-workers (30) (magenta), and Laguri et al. (31) (cyan). B, our combined biophysical analyses support a new model in which heparin promotes CXCL12 dimerization by contacting residues along the entire six-stranded sheet. The highlighted CXCL12 residues associate with heparin as determined by two-dimensional NMR, mutagenesis, SPR (blue), and three-dimensional NMR (purple). molecules equal to and larger than dp12 compete for the N terminus (29). Our measurements of side chain chemical shift perturbations and SPR analysis of CXCL12(3-68) and CXCL12 2 (3-68) clearly demonstrate that Lys-1 plays no role in binding to heparin oligosaccharides. Consequently, we concluded that heparin-mediated protection of the CXCL12 N terminus from proteolysis must be an indirect effect related to the GAG-induced oligomerization.
Dimerization Inhibits CXCL12 Degradation-DPPIV/CD26 is a multidomain enzyme that possesses distinct recognition and catalytic regions (41). To study the enhanced serum stability of CXCL12 2 compared with CXCL12, we recently monitored in vitro degradation by DPPIV/CD26 (39). The half-life of dimeric CXCL12 is 25-fold greater than wild type (39). Because oligosaccharide-induced dimerization is chain length-dependent, we hypothesized that heparin may possess a similar sizemediated influence on protection of DPPIV/CD26 degradation. Whereas incubation of CXCL12 with dp2 had no effect on degradation, dp4 delayed N-terminal cleavage nearly 30-fold. The lack of structural data for DPPIV/CD26 in complex with protein substrates makes the mechanism of enhanced stability unclear. The simplest explanation is the potential for overlap between the DPPIV/CD26 binding site and the CXCL12 dimerization interface. Most DPPIV/CD26 substrates are compact peptides with molecular masses less than 10 kDa (42). The large, dimeric CXCL12 may therefore also prevent the dimerization of DPPIV/CD26, which has been shown to reduce k cat 30-fold (43). Matrix metalloprotease-2 also degrades CXCL12 2 slower than CXCL12 (39). Together, this suggests that heparin may indirectly protect CXCL12 from numerous other enzymes such as matrix metalloproteases 1, 3, 9, 13, and 14 (44), cathepsin G (45), and elastase (46). Because of growing clinical attention, it will be of interest to identify CXCL12 variants that maximize its cardioprotective properties while minimizing secondary interactions.