Abstract
The stromal cell-derived factor 1α (CXCL12) belongs to the CXC chemokine family and plays an important role in tissue regeneration and the recruitment of stem cells. Here, a stable chemotactic gradient is essential that is formed by the interaction of CXCL12 with the extracellular matrix. Binding properties of CXCL12 to naturally occurring glycosaminoglycans (GAGs) as well as to the artificial highly sulfated hyaluronic acid (HA) are investigated by using a combination of NMR spectroscopy, molecular modeling and molecular dynamics simulations. Our results demonstrate a preferred protein binding for the sulfated GAGs heparin (HE) and highly sulfated HA. Furthermore, we could demonstrate that the orientation of the sulfate is crucial for binding. All sulfated GAGs interact with the CXCL12 GAG-binding motif (K24-H25-L26-K27-R41-K43-R47), where K27 and R41 represent the anchor points. Furthermore, differences could be observed in the second interaction interface of CXCL12: both HE and highly sulfated HA interfere with the receptor-binding motif, while chondroitin sulfate binds different amino acids in close proximity to this motif. CXCL12 does not interact with HA, which was directly demonstrated by NMR spectroscopy and molecular modeling and explained by the lack of sulfate groups of the HA molecule.
Introduction
Chemokines are small proteins with a molecular weight between 8 and 12 kDa. About 50 members of this family have been identified (Proudfoot et al. 2003; Laguri et al. 2008). So far, the classification of chemokines is defined by the position of the cysteine(s) at the N-terminus (Prieschl et al. 1995). Thus, chemokines can be divided into CC-, CXC-, CX3C- and C-class, where X represents any given amino acid (Murphy et al. 2000; Zlotnik and Yoshie 2000). Commonly, chemokines have a highly conserved tertiary structure consisting of an unstructured N-terminus, followed by a three-stranded antiparallel β-sheet and a C-terminal amphiphilic α-helix (Salanga and Handel 2011). Within the structure, three binding interfaces have been described, for oligomerization, glycosaminoglycan (GAG) and receptor binding. Those regions are either spatially separated or (partially) overlapping (Graham et al. 1996; Chakravarty et al. 1998; Kuschert et al. 1998; Amara et al. 1999; Laurence et al. 2001; Proudfoot et al. 2001; Ziarek and Volkman 2012).
The activity of chemokines is mediated by G-protein-coupled chemokine receptors, of which 20 are known until now (Murphy et al. 2000). Previous studies have displayed a distinct promiscuity between chemokines and their respective receptors, as one receptor can be activated by different chemokines and vice versa (Proudfoot 2002; Nomiyama et al. 2011). Receptor activation leads to various additional signaling pathways like JAK/STAT signaling, MAPK signal transduction, Ca2+-flux and activation of phospholipase Cγ (Borst et al. 2012; Fang et al. 2012; Stadtmann and Zarbock 2012). Thus, chemokines exhibit distinct functions and display important roles in the regulation of the immune response, wound healing and inflammation (Gillitzer and Goebeler 2001; Laguri et al. 2008; Borst et al. 2012). Additionally, they influence angiogenesis, hematopoiesis and promote tissue-specific metastasis (Gerard and Rollins 2001; Fang et al. 2012; Lira and Furtado 2012).
To act specifically, chemokines form a stable haptotactic gradient by interacting with GAGs. GAGs can be found on the cell surface and are important components of the extracellular matrix (ECM) (Jackson et al. 1991; Conrad 1998; Gandhi and Mancera, 2008). They are classified by their monosaccharide composition, their glycosidic linkage and sulfation pattern (Gandhi and Mancera 2008). Several types of GAGs can be distinguished, e.g. heparin (HE)/heparan sulfate (HS), chondroitin/dermatan sulfate, keratan sulfate and hyaluronic acid (HA). The binding of GAGs to proteins is dominated by the interaction of the negatively charged sulfate and carboxyl groups in the GAG and the positively charged side chains of lysines and arginines in the protein. This interaction leads to various functions of GAGs. First, they stabilize protein gradients, protect proteins from degradation and present them to cooperative binding molecules (Sommer and Rifkin, 1989; Arai et al. 1994; Middleton et al. 1997; Middleton et al. 2002). In addition, GAGs are able to promote oligomerization and accumulation of chemokines, which leads to their active states (Cripps et al. 2005; Yu et al. 2005; Proudfoot 2006; Salanga and Handel 2011).
Here, we focus on the interaction of various GAGs with the stromal cell-derived factor 1α (SDF-1α/CXCL12). CXCL12 is composed of 68 amino acids and belongs to the CXC class of chemokines (Ray et al. 2012). Previous studies verified the important role of CXCL12 in embryogenesis and organogenesis (Nagasawa et al. 1996; Tachibana et al. 1998), while in adult organisms, the main functions of the protein include the promotion of angiogenesis, homing of hematopoietic stem cells, neural regeneration and inhibition of infection of HIV-CXCR4+-strain (Juarez et al. 2004; Tiveron and Cremer 2008; Patrussi and Baldari 2011).
CXCL12 binds and activates the chemokine receptors CXCR4 and CXCR7, which both are G-protein coupled and able to induce signaling pathways like protein kinase C activation, NFκB-signaling and MAPK-activation by G-protein coupled signaling or arrestin recruitment (Ratajczak et al. 2006; Rajagopal et al. 2010; Sun et al. 2010; Decaillot et al. 2011).
Previous studies demonstrated a GAG-binding region containing the tetrapeptide K24-H25-L26-K27 of CXCL12 (Sadir et al. 2001). This linear-binding region is flanked by R41, K43 and R47 in the 3D structure (Sadir et al. 2001; Juarez et al. 2004; Murphy et al. 2007; Laguri et al. 2008; Sapay et al. 2011) and described as high-affinity heparin-binding region (HAHBR). Solution 2D heteronuclear single quantum coherence nuclear magnetic resonance spectroscopy (HSQC NMR) experiments in combination with chemical shift perturbation (CSP) with CXCL12 in the presence of HE and HS led to the detailed characterization of the GAG-binding region (Veldkamp et al. 2005; Murphy et al. 2007). However, interaction studies using other GAGs are missing up to now.
Structure of the GAGs used for the experimental characterization of the interaction with CXCL12. GAG binding by CXCL12 was investigated using the sulfated GAGs (A) HE hexasaccharide (dp6), (B) pHA tetrasaccharide (dp4), (C) CS dp6, (D) unsulfated HA dp4 and (E) desulfated heparin (deHE). The titration with CS was carried out using a mixture of CS–GAGs with different sulfation pattern at position C4 and C6.
Results
Production of 15N-labeled CXCL12 and assignment of the 1H-15N HSQC NMR signals
For 1H-15N HSQC NMR experiments, the 15N-labeled CXCL12 was recombinantly expressed as fusion protein in minimal medium, by the intein-mediated purification with an affinity chitin-binding tag (IMPACT™)-system purified and refolded, as described previously (Pichert, Samsonov, et al. 2012; Pichert, Schlorke, et al. 2012). The protein was investigated by 1H-15N HSQC NMR spectroscopy in solution under the conditions published by Veldkamp et al. (2009). The resonance lines are narrow and indicative of a well-established structure of CXCL12 enabling the full resonance assignment of the 1H-15N HSQC spectrum based on previous data (Veldkamp et al. 2009). In addition, nuclear Overhauser effect spectroscopy (NOESY) and total correlation spectroscopy (TOCSY) were performed to confirm the assignment. Altogether, 90% of the backbone amide signals could be assigned unambiguously (Supplementary Data Figure S1).
Interaction of CXCL12 with GAGs investigated by 1H-15N HSQC NMR
To understand the functionality of CXCL12 and its ability to specifically recruit various cell types, the binding properties of the chemokine to naturally occurring GAGs were investigated using HE, CS and HA, which are components of the ECM, and the artificial pHA, a synthetic GAG that was developed for the coating of functional biomaterials for regeneration (Pichert, Samsonov, et al. 2012; Pichert, Schlorke, et al. 2012; Schnabelrauch et al. 2013; Köhling et al. 2016). The investigation of the interaction of CXCL12 with GAGs was performed by using HE and CS hexasaccharide (dp6) as well as pHA and HA tetrasaccharide (dp4). A detailed monitoring of the amino acids involved in GAG binding was performed using HSQC NMR titration experiments with CXCL12 at varying GAG concentrations. GAG binding induces CSP, which were obtained from the 1H-15N HSQC spectra. After each titration step, a 1H-15N HSQC NMR spectrum was recorded and the chemical shift changes induced by GAG binding were measured for all assigned amino acid signals.
Interaction of CXCL12 with HE in the HAHBR
Sections of 1H-15N HSQC NMR spectra enlarged for specific amino acids of 15N-labeled CXCL12 (500 µM, pH 6.8) in the presence of varying concentrations of different natural and artificial GAGs. Enlargement of the GAG-binding amino acids of CXCL12 analyzed for HE and corresponding chemical shift changes in the course of the titration with either of HE dp6 as positive control, CS dp6, pHA dp4 or HA dp4. Each color corresponds to a titration step. This figure is available in black and white in print and in color at Glycobiology online.
The NMR signals of the protein backbone displayed distinct chemical shift values that partially changed with increasing HE concentration indicative of the interaction. Additionally, some signals were split suggesting that the interaction of HE with the respective amino acids represents a slow exchange process regarding the relevant NMR time regime (e.g. R41 in Figure 2, Supplementary Data Table SI). Therefore, two different states of the affected amino acid can be detected: in one state, the amino acid is bound to HE, while the other state correlates to the free protein. The exchange between the two states is slow, i.e. it occurs in milliseconds. At higher concentration of HE, the peak shifts to the HE-bonded state. This observation is seen for different amino acids, e.g. S16 (data not shown). Using a low HE concentration (52 µM), a narrow and highly resolved peak of S16 could be detected, but by increasing the concentration to 251 µM the peak broadened and was less well resolved. This may also indicate the transition to exchange that occurs in the intermediate time regime.
(A) Crystal structure of the complex of CXCL12 dimer (in cartoon) with HE dp2 (in sticks) (PDB ID: 2NWG, 2.07 Å). (B) In the same structure, high and low affinity HE-binding sites are shown in blue and green, respectively. (C) The residues with the weighted CSP of CXCL12 induced by HE and pHA dp6 superior than 2 standard deviations are shown in blue. (D) The residues with weighted CSPs of CXCL12 induced by CS dp6 larger than 2 standard deviations are shown in blue and in green for HAHBR and LAHBR, respectively.This figure is available in black and white in print and in color at Glycobiology online.
pHA binds to CXCL12 via the HAHBR
The artificial pHA was investigated to study the interaction of synthetic and modified GAGs with CXCL12. Furthermore, the general mechanism of the binding of highly sulfated GAGs with four sulfate groups on average per dp2 unit was of interest. Titration with artificial pHA was performed in five titration steps. HSQC NMR spectra of titration with 201 and 488 µM were overlaid with amino acids from the HAHBR of CXCL12 (Figure 2, Supplementary Data Figure S2B). Again, precipitation occurred after adding 960 µM pHA as described for HE. Similar to titration with HE, peak broadening and peak splitting were observed (Supplementary Data Table SI).
Plot of the weighted CSPs of CXCL12 induced by GAG binding. The bar graphs (A) to (D) show the chemical shift changes after the titration of (A) 345 µM HE, (B) 488 µM pHA, (C) 750 µM CS and (D) 2.2 mM HA. The region HAHBR is framed in box I and the sequence LAHBR is shown in box II. For HE and pHA, the LAHBR corresponds to the receptor interaction interface, where CS has a different LAHBR in the C-terminal amino acids, marked with an arrow. Hatched columns display signals of amino acids of CXCL12, which split up after titration and thus the CSP is shown for only one signal. The thick broken line represents the cut-off of the CSPs after calculation of the two times the standard deviation (σ): σ(HE) = 0.049 ppm; σ(pHA) = 0.026 ppm; σ(CS) = 0.01 ppm; σ(HA) = 0.004 ppm.
Similar to HE, CXCL12 interacts with pHA mainly through the basic amino acids K24, K27, K54 and K64 (Figures 2, 4B). Again, two times the standard deviation was chosen as a cut-off for indicative CSPs and similar amino acids for pHA interaction were observed as described above (Figure 3C).
CS interacts weakly with CXCL12 via the HAHBR
Next, CS was chosen as lowly sulfated GAG containing only one sulfate group per dp2 unit. After titration with CS, remarkable chemical shifts could be observed solely at high GAG concentration. Precipitation of the protein–GAG complex did not occur. During the titration experiments, the NMR peaks remained well resolved and narrow, except for the last titration step (Figure 2, Supplementary Data Figure S2C).
In order to obtain more information on the interaction with CS, CSP WAS calculated using the last titration step with 750 µM CS (Figure 4). Regions with strong CSPs were identified including amino acids V18, V23, K24, H25 and K27 in the N-terminus and the first β-strand, N45 in the second β-strand and K64 in the α-helix. Additionally, R12, V18, A19, R20, R41 and L62 showed higher but less intense CSPs than the before-mentioned amino acids.
Again, mainly basic amino acids in the N-terminus and first β-strand are involved in the interaction of CS with CXCL12 (Figure 4). Within the GAG-binding region, strong chemical shifts were observed for K24 and K27 and increased CSP for R41, while K43 and R47 showed only slight changes and did not seem to be in direct contact with CS (Figures 2, 4C). For comparison of the calculation method to determine the cut-off, two times the standard deviation was applied as a second cut-off for indicative CSPs. Significant amino acids like V18, 19 A, 24 K and 27 K were identified, which are labeled in Figure 3D to visualize the CS-binding region.
HA does not interact with CXCL12 in HAHBR
In addition to naturally and artificially sulfated GAGs, also HA is known to be important for the functionality of CXCL12 and is the only GAG without sulfate groups. To this date, no details about the direct interaction of CXCL12 with HA have been reported.
A NMR titration experiment involving four titration steps did not lead to broadening of peaks or strong peak shifting. All peaks remained narrow and well resolved including the amino acids of the GAG-binding region (Figure 2, full spectrum shown in Supplementary Data Figure S2D).
Accordingly, calculation of CSPs using the results of the last titration step with 2.2 mM HA generated only insignificant CSPs for the amino acids H17, V23, L29, I38, V49 and A65 with a weighted chemical shift change of 0.01 ppm. No basic amino acids except H17 were detected to respond to the presence of HA. In contrast to HE, pHA and CS interaction, signals were only observed for polar amino acids.
Taken together, varying magnitudes of CSPs were found for each GAG interacting with CXCL12. The strongest CSPs could be observed for the interaction with HE, while a 5-fold reduced magnitude was observed for CS-CXCL12 binding. Titration with HA showed almost no CSPs in comparison to HE titration. Based on these observations, sulfated GAGs led to significantly stronger alterations in the protein backbone structure of CXCL12, whereas an interaction of HA with CXCL12 could not be observed after titration with a 4-fold excess of the GAG.
Molecular modeling of CXCL12 interactions with GAGs
Representation of CXCL12 (in cartoon) highlighting the residues with the weighted CSP of the protein induced by HE dp6 superior than 2 standard deviations (in blue) and the 12 most favorably contributing residues in terms of the calculated free-binding energy (in yellow sticks). This figure is available in black and white in print and in color at Glycobiology online.
Blind docking with Autodock identified the same HAHBR for GAGs of different type and length as observed by HSQC NMR titration experiments. In particular, K24, H25, K27 and R41 were found to be the important residues for GAG binding as previously identified by X-ray crystallography for HE dp2 (PDB: 2NWG, Murphy et al. 2007). If the binding site is already occupied by a GAG, blind docking led to a second binding site overlapping with the previously characterized LAHBR (Figure 3). MD-based MM-GBSA free energy calculations for these two sites support their differences in terms of affinity (for example, for HE dp2, MM-GBSA free energies of binding are −55.9 and −25.1 kcal/mol, respectively). Binding poses for longer GAGs obtained by docking appeared to be nucleated in the HAHBR experimentally observed for HE dp2 (Figure 5, Supplementary Data Figure S3, Murphy et al. 2007). Interestingly, the LAHBR, consisting mainly of receptor-binding amino acids, can be seen as a further elongation of a GAG bound to the HAHBR. The observed binding pose in HAHBR agrees well with the NMR-based structural models proposed for HE and HS oligosaccharide interaction with CXCL12 (Ziarek et al. 2013). Ziarek et al. (2013) demonstrated an interaction of the amino acids of HAHBR using HE decasaccharide, which is in agreement with the modeling performed in this study. In terms of binding free energies, the observed interactions are electrostatics-driven, and a clear dependence on GAG sulfation and GAG length is observed. Thus, binding is more favorable for longer and higher sulfated GAGs (Table I, Supplementary Data Figure S4). While no significant differences for highly sulfated HE and pHA were found, the calculations suggest that for unsulfated GAGs deHE could be a better binder than HA. By applying per residue free energy decomposition, CXCL12 residues, which are the most crucial for binding of GAGs, were identified (Table II, Figure 5, Supplementary Data Figure S4). These include also the amino acids R8 and R12 in the N-terminal part of CXCL12. The residues overlap with the residues proposed to be key for HE oligosaccharide binding based on NMR, mutagenesis and SPR data and do not differ significantly for different GAG types (Ziarek et al. 2013).
MM-GBSA free energy of binding for the CXCL12 dimer with GAGs docked to the high-affinity GAG-binding site
| GAG | GAG length | ∆GMM-GBSA (kcal/mol) |
|---|---|---|
| HA | dp2 | −20.2 ± 6.5 |
| HA | dp4 | −25.1 ± 5.5 |
| HA | dp6 | −31.0 ± 8.8 |
| HA | dp8 | −47.0 ± 8.4 |
| deHE | dp2 | −22.7 ± 6.2 |
| deHE | dp4 | −38.4 ± 8.3 |
| deHE | dp6 | −52.8 ± 7.0 |
| deHE | dp8 | −63.3 ± 8.1 |
| HE | dp2 | −55.9 ± 13.6 |
| HE | dp4 | −119.4 ± 8.4 |
| HE | dp6 | −132.0 ± 19.8 |
| HE | dp8 | −176.0 ± 12.1 |
| pHA | dp2 | −84.0 ± 9.7 |
| pHA | dp4 | −131.0 ± 18.1 |
| pHA | dp6 | −141.7 ± 20.7 |
| pHA | dp8 | −172.9 ± 12.9 |
| GAG | GAG length | ∆GMM-GBSA (kcal/mol) |
|---|---|---|
| HA | dp2 | −20.2 ± 6.5 |
| HA | dp4 | −25.1 ± 5.5 |
| HA | dp6 | −31.0 ± 8.8 |
| HA | dp8 | −47.0 ± 8.4 |
| deHE | dp2 | −22.7 ± 6.2 |
| deHE | dp4 | −38.4 ± 8.3 |
| deHE | dp6 | −52.8 ± 7.0 |
| deHE | dp8 | −63.3 ± 8.1 |
| HE | dp2 | −55.9 ± 13.6 |
| HE | dp4 | −119.4 ± 8.4 |
| HE | dp6 | −132.0 ± 19.8 |
| HE | dp8 | −176.0 ± 12.1 |
| pHA | dp2 | −84.0 ± 9.7 |
| pHA | dp4 | −131.0 ± 18.1 |
| pHA | dp6 | −141.7 ± 20.7 |
| pHA | dp8 | −172.9 ± 12.9 |
MM-GBSA free energy of binding for the CXCL12 dimer with GAGs docked to the high-affinity GAG-binding site
| GAG | GAG length | ∆GMM-GBSA (kcal/mol) |
|---|---|---|
| HA | dp2 | −20.2 ± 6.5 |
| HA | dp4 | −25.1 ± 5.5 |
| HA | dp6 | −31.0 ± 8.8 |
| HA | dp8 | −47.0 ± 8.4 |
| deHE | dp2 | −22.7 ± 6.2 |
| deHE | dp4 | −38.4 ± 8.3 |
| deHE | dp6 | −52.8 ± 7.0 |
| deHE | dp8 | −63.3 ± 8.1 |
| HE | dp2 | −55.9 ± 13.6 |
| HE | dp4 | −119.4 ± 8.4 |
| HE | dp6 | −132.0 ± 19.8 |
| HE | dp8 | −176.0 ± 12.1 |
| pHA | dp2 | −84.0 ± 9.7 |
| pHA | dp4 | −131.0 ± 18.1 |
| pHA | dp6 | −141.7 ± 20.7 |
| pHA | dp8 | −172.9 ± 12.9 |
| GAG | GAG length | ∆GMM-GBSA (kcal/mol) |
|---|---|---|
| HA | dp2 | −20.2 ± 6.5 |
| HA | dp4 | −25.1 ± 5.5 |
| HA | dp6 | −31.0 ± 8.8 |
| HA | dp8 | −47.0 ± 8.4 |
| deHE | dp2 | −22.7 ± 6.2 |
| deHE | dp4 | −38.4 ± 8.3 |
| deHE | dp6 | −52.8 ± 7.0 |
| deHE | dp8 | −63.3 ± 8.1 |
| HE | dp2 | −55.9 ± 13.6 |
| HE | dp4 | −119.4 ± 8.4 |
| HE | dp6 | −132.0 ± 19.8 |
| HE | dp8 | −176.0 ± 12.1 |
| pHA | dp2 | −84.0 ± 9.7 |
| pHA | dp4 | −131.0 ± 18.1 |
| pHA | dp6 | −141.7 ± 20.7 |
| pHA | dp8 | −172.9 ± 12.9 |
MM-GBSA free energy per residue contribution of CXCL12 to GAG binding
| GAG length | dp2 | dp4 | dp6 | dp8 |
|---|---|---|---|---|
| Residue | ||||
| K24 | −2.6 ± 2.3 | −3.0 ± 2.7 | −2.7 ± 2.1 | −5.4 ± 6.0 |
| K27 | −4.3 ± 3.4 | −6.0 ± 5.2 | −6.4 ± 5.0 | −5.7 ± 4.0 |
| R41 | −4.5 ± 4.1 | −7.8 ± 4.9 | −6.0 ± 3.7 | −7.2 ± 5.4 |
| R47 | −0.5 ± 0.5 | −2.2 ± 2.4 | −2.9 ± 3.6 | −2.8 ± 3.2 |
| GAG length | dp2 | dp4 | dp6 | dp8 |
|---|---|---|---|---|
| Residue | ||||
| K24 | −2.6 ± 2.3 | −3.0 ± 2.7 | −2.7 ± 2.1 | −5.4 ± 6.0 |
| K27 | −4.3 ± 3.4 | −6.0 ± 5.2 | −6.4 ± 5.0 | −5.7 ± 4.0 |
| R41 | −4.5 ± 4.1 | −7.8 ± 4.9 | −6.0 ± 3.7 | −7.2 ± 5.4 |
| R47 | −0.5 ± 0.5 | −2.2 ± 2.4 | −2.9 ± 3.6 | −2.8 ± 3.2 |
The energies for four residues with the most favorable contributions are averaged for CXCL12 monomers and for the analyzed GAG molecules (HA, deHE, HE, pHA) of dp2, dp4, dp6 and dp8.
MM-GBSA free energy per residue contribution of CXCL12 to GAG binding
| GAG length | dp2 | dp4 | dp6 | dp8 |
|---|---|---|---|---|
| Residue | ||||
| K24 | −2.6 ± 2.3 | −3.0 ± 2.7 | −2.7 ± 2.1 | −5.4 ± 6.0 |
| K27 | −4.3 ± 3.4 | −6.0 ± 5.2 | −6.4 ± 5.0 | −5.7 ± 4.0 |
| R41 | −4.5 ± 4.1 | −7.8 ± 4.9 | −6.0 ± 3.7 | −7.2 ± 5.4 |
| R47 | −0.5 ± 0.5 | −2.2 ± 2.4 | −2.9 ± 3.6 | −2.8 ± 3.2 |
| GAG length | dp2 | dp4 | dp6 | dp8 |
|---|---|---|---|---|
| Residue | ||||
| K24 | −2.6 ± 2.3 | −3.0 ± 2.7 | −2.7 ± 2.1 | −5.4 ± 6.0 |
| K27 | −4.3 ± 3.4 | −6.0 ± 5.2 | −6.4 ± 5.0 | −5.7 ± 4.0 |
| R41 | −4.5 ± 4.1 | −7.8 ± 4.9 | −6.0 ± 3.7 | −7.2 ± 5.4 |
| R47 | −0.5 ± 0.5 | −2.2 ± 2.4 | −2.9 ± 3.6 | −2.8 ± 3.2 |
The energies for four residues with the most favorable contributions are averaged for CXCL12 monomers and for the analyzed GAG molecules (HA, deHE, HE, pHA) of dp2, dp4, dp6 and dp8.
Top six most favorably contributing residues of CXCL12 for HE dp6 binding, ascendingly sorted by ensemble-averaged binding energy. The data were obtained via SRED based on a DMD study. Each data point corresponds to 40 samples; the error bars show the standard error of the mean. The “a” and “b” suffixes indicate whether a residue is part of either monomer in dimer structure of CXCL12.
Discussion
The chemokine CXCL12 is an attractive target in medicinal research because of its ability to regulate the immune system and the recruitment of tissue-specific stem cells. Important aspects to understand the character of the protein and its functions include its ability to interact with components of the ECM. Here, we focused on the characterization of CXCL12 interaction with natural GAGs (HE, CS, HA) and the artificially pHA, a synthetic GAG, which could be used for coating of biomaterials and formation of specific hydrogels (Schnabelrauch et al. 2013).
The binding properties of GAGs to CXCL12 were investigated by observing CSPs in 1H-15N HSQC NMR experiments. In order to complete the description of CXCL12–GAG interactions in terms of glycosidic linkages and sulfation pattern, we additionally analyzed computationally the interaction of CXCL12 with deHE.
The correct fold CXCL12 was confirmed by 1H-15N HSQC NMR. All amino acids of the protein were detectable, except of proline residues, K1, L26 and N44 in agreement with the literature (Veldkamp et al. 2005). Thus, more than 90% of CXCL12 backbone amide signals could be assigned, and all of them showed distinct and spatially limited peaks in the 1H-15N HSQC spectrum.
On the basis of a correct conformation of CXCL12, GAG binding was investigated by following chemical shift changes in 1H-15N HSQC spectra in response to GAG titration. This method is based on the assumption that most intensive shifts are likely to correlate with amino acids involved in the interaction. Thus, CSPs are used to determine the strength of interaction between each amino acid of the protein and the ligand, where binding regions can be proposed. The advantage of this method is the ability to observe the absence of chemical shift changes if a ligand does not bind to the protein leading to no false positive results (Williamson 2013). The titration with HE revealed an elevated CSP by increasing the HE concentration. Here, basic amino acids K24, K27, R41, K43, K54 and K64 and their adjacent amino acids A21, V23, H25, N33 and A65 showed the largest CSPs. Basically, the observed changes confirm previous results of CXCL12 interaction with HE and HS (Sadir et al. 2001; Murphy et al. 2007; Veldkamp et al. 2009). Inside the protein, the amino acids K24 and K27 in the first and R41 and K43 in the second β-strand were identified as essential binding partners for HE (Sadir et al. 2001). This region was completed by the participation of H25 in the HE interaction leading to the sequential GAG-binding motif BBXB (K24-H25-L26-K27), where B is any basic and X is any given amino acid (Hileman et al. 1998; Amara et al. 1999; Sadir et al. 2001; Murphy et al. 2007). Furthermore, previous NMR studies and mutagenesis experiments demonstrated a stabilizing effect of R41 and K43 in HS and HE interaction (Sadir et al. 2001; Juarez et al. 2004). Additionally, the binding of HE in the HAHBR was supported by computational studies. MM-GBSA free energy calculation demonstrated that R41 shows the highest average energy followed by K27 (Table II). This indicates an important role of R41 and K27 for the interaction of HE as suggested earlier (Sadir et al. 2001).
To date, no structural analysis for CXCL12 binding to CS by NMR has been reported. The closest structural studies were performed by Zhou et al. using circular dichroism titration studies with CS A and oversulfated CS. They observed weaker binding of CS A to CXCL12 in comparison to HE and a better interaction of the protein with the oversulfated CS (Zhou et al. 2014). Additionally, Zhou et al. hypothesized the amino acids H25, K27 and R41 to be important for the interaction, which is supported by our NMR data (Figures 2, 4C). In our studies, K24 and K27 play predominant roles for interaction, detected by a strong CSP.
Similar to HE, CXCL12 interacts through the specific GAG-binding region K24-H25-L26-K27 with pHA. This binding is also stabilized by R41, K43 and R47. pHA has not been investigated so far, but a similar binding region of CXCL12 with pHA is suggested, since the CSP intensity was only 3-fold less than for HE. This observation was supported by the MD studies, where the calculated MM-GBSA free energy of pHA is comparable to the free energy with HE in the HAHBR of CXCL12. No interaction studies were performed before using artificial HA derivatives, but we could demonstrate that artificially pHA displays the same GAG-binding region as HE.
Finally, we analyzed the interaction of CXCL12 with HA. Previous studies focused on the interaction of HA in combination with CXCL12 and different cell types and also in combination with hydrogels (Sbaa-Ketata et al. 2002; Avigdor et al. 2004; Purcell et al. 2012; Fuchs et al. 2013). Taken together, Sbaa-Ketata et al. showed that the chemotactic potential of the protein was improved by adding HA or HA fragments, whereas Fuchs et al. found opposite effects by investigating the influences of high- and low-molecular weight HA. Although some studies have found increased bone marrow cell homing during regeneration of the myocardium after injection of CXCL12-HA-hydrogels (Purcell et al. 2012), a detailed analysis regarding the interaction of CXCL12 with HA was not performed so far. Our approach of characterizing the HA interaction with CXCL12 using HSQC NMR revealed a different binding behavior compared to previously investigated GAGs. Very low CSPs could only be detected for nonpolar amino acids V23, L29, I38 and V49 and positively charged H17 using a 4-fold excess of HA. No significant effects could be observed for basic amino acids in either HAHBR or LAHBR. Additionally, the molecular docking led to similar results by calculation of an adverse low free energy in binding of CXCL12 with HA, suggesting no direct interaction (Table I). Similarly, to sulfated GAGs, the binding energy increases with the growing GAG length, but in comparison to the other GAGs the free energy remained the lowest and unfavorable. Furthermore, the differences of the MM-GBSA free energy observed for HA and deHE revealed that the interaction is influenced by other parameters like e.g. glycosidic linkage (Table I).
Detailed comparative analysis of the results obtained from the 1H-15N HSQC CSP data and the computational calculations suggest that all sulfated GAGs use the GAG-binding region K24-H25-L26-K27 in the first β-strand in combination with R41 and K43 in the second β-strand (Figures 3, 5, Supplementary Data Figure S3). The binding strength to CXCL12 is nearly similar for HE and pHA, while CXCL12 showed less binding ability for CS.
The LAHBR could be observed for the sulfated GAGs by HSQC NMR experiments. For HE, the second binding site consists of the receptor interface RFFESH in CXCL12, which could be detected by a higher CSP of R12, F13, F14, E15, S16 and H17 as observed earlier (Laguri et al. 2008). The idea that HE also binds to the receptor interaction site arose through previous modeling studies by Sapay et al. (2011), who showed that HE binds to two interfaces in the CXCL12 dimer, where interface II comprises the amino acids S4, L5, R8 and R12 of chain 1 and R8, R12, N45 and R47 of chain 2. This is supported by our CSP, which revealed that a strong shift was observed inside the C-terminal α-helix and the receptor-binding site in the N-terminus for the interaction of CXCL12 with pHA. Binding in the C-terminal part is possibly induced by interaction of pHA with K64 and K68. This also explains the intense CSP of the amino acids L62, A65, L66 and N67. The strong CSP in the receptor-binding site may arise from the involvement of the receptor-binding amino acids as a second interface. Concluding, the NMR data obtained for pHA indicate that CXCL12 binds GAGs by two regions, one comprising the BBXB region and the second interface consisting of the receptor-binding amino acids. Additionally, the C-terminal α-helix seems to have a small interaction region for the binding of pHA, which is in agreement with earlier studies for CXCL8 (David et al. 2008; Nordsieck et al. 2012; Pichert, Samsonov, et al. 2012; Pichert, Schlorke, et al. 2012; Möbius et al. 2013; Hofmann et al. 2015), but has to be investigated in more detail in the future.
A second binding site for the interaction of CXCL12 consisting of R20, A21, N30 and K64 with CS was hypothesized by Zhou et al. (2014). Our CSP data also confirm this second interaction interface of CXCL12 with CS (Figures 2, 4C). Additionally, the interaction is supported by R12 and H17 of the receptor-binding site as well as V18, A19 and L62. In the case of CS, the receptor-binding site of CXCL12 is not directly involved in interaction to the same extent as for HE and pHA. The region consisting of H17-V18-A19-R20-A21 seems to be preferable for binding (Figure 4C).
Summarizing previous and current data, HE and pHA bind to the LAHBR, while CS has a higher affinity to the amino acids N-terminally of this region. Besides the binding regions, distinct from the HAHBR for GAG interaction we suggest that the orientation of the GAG sulfate groups plays an important role for binding. We observed that HE and pHA show different strengths in CSP, even though pHA has a higher degree of sulfation than HE. Additionally, the obviously highest change in chemical shift of CXCL12 was observed for K24 in case of HE and CS and for K27 in case of pHA. This suggestion is supported by previous studies, which point to the importance of the sulfation pattern of GAGs, in particular demonstrated for HS (Sadir et al. 2001; Zhang et al. 2012). Sadir et al. (2001) documented that the N- and 2-O-sulfate groups are essential for CXCL12 binding. Additionally, Zhang et al. (2012) showed an importance of the 6-O-sulfation of HE to induce leukocytosis by CXCL12. In addition to HE, Laguri et al. performed structural analysis with HS by using 13C-labeled GAG and 15N-labeled CXCL12. They demonstrated that both N-sulfated and 6-O-sulfated glucosamine residues are relevant for the interaction (Laguri et al. 2011).
Similar interaction studies were also performed for CS with different degrees of sulfation. Zhou et al. (2014) demonstrated that oversulfated CS was more effective to bind CXCL12 than HE, which has been known so far to be the best protein-binding GAG. The comparison of all three GAGs showed that CS A binding to CXCL12 is rather low. The highest CSP was calculated for HE followed by pHA with a 5-fold less intensity compared to HE, HE contains three and pHA contains four sulfate groups per dp2. CSPs continue to decline 50-fold for CS and are almost undetectable for HA in comparison to HE. It has to be noted that the use of artificial highly sulfated GAGs is only possible in in vitro studies so far. Previous investigations from Guerrini et al. (2008) and Li et al. (2009) demonstrated a toxic effect through an increase of the vasodilator bradykinin in blood plasma after treatment with oversulfated CS. A detailed in vitro characterization in cells and tissue of pHA is necessary in further investigations to exclude toxic effects.
The interaction of sulfated GAGs could also be influenced by the glycosidic linkage conformational space in the dp2 units. This was shown previously for CXCL8–GAG interactions (Pichert, Samsonov, et al. 2012; Pichert, Schlorke, et al. 2012). Here, glycosidic linkages of GAGs could influence the sulfate group orientation and thereby the interaction with the basic amino acids of CXCL12 with HE/pHA and deHE/HA.
The length of GAGs has to be considered as well. HE length, for example, was shown to influence dimerization of CXCL12 and protection against proteases (Sadir et al. 2004; Fermas et al. 2008). Our calculations confirm this for the studied GAGs, demonstrating that an increasing length of the GAG led to an enhanced free-binding energy of the HAHBR in CXCL12.
The interaction studies of CXCL12 with HE and pHA also demonstrate a strong CSP of F14. This observation could result from interaction with the GAG in adjacent regions containing charged amino acids as well as from dimerization of CXCL12. Previous studies demonstrated a chemical shift of F14 after dimerization of CXCL12 (Baryshnikova and Sykes 2006). Additionally, they could show that CXCL12 dimers develop at a protein concentration of 300 µM. The HSQC NMR studies were performed with CXCL12 concentrations of approximately 500 µM, where dimer formation is likely. This effect could also explain the CSPs of C50 and S16 because of conformational changes inside the protein (Baryshnikova and Sykes 2006). Furthermore, the dimerization could also be a reason for the weighted chemical changes in the C-terminal part of the protein. Previous studies by Fermas et al. demonstrated that in the absence of sulfated GAGs CXCL12 exists as a monomer (15 µM protein). Additionally, these authors found that dimerization is promoted by HE, and this finding proved to be length dependent. The investigation showed that HE dp4 and dp6 promote dimerization, while dp4 is the minimal required unit (Fermas et al. 2008).
Gangavarapu et al. (2012) demonstrated the dependence of the monomer-dimer equilibrium of CXCL8 by specific GAG interactions. In their experiments, they analyzed various effects for the recruitment of neutrophils in lung and peritoneum. For both organs, a different distribution of heparan and dermatan sulfate could be detected by which the dimerization of CXCL8 is influenced (Goger et al. 2002; Pichert, Samsonov, et al. 2012; Pichert, Schlorke, et al. 2012). Due to the fact that CXCL12 has LAHBR localized similarly to the GAG-binding region in CXCL8, an experiment can be proposed with the aim to establish suitable determinants for CXCL12–GAG interaction. Nevertheless, factors like pH value, salt and temperature should be considered for the interaction in the future. Veldkamp et al. (2005) demonstrated crucial influence of pH value, phosphate and sulfate concentration on the interaction with CXCL12. These authors could observe dimerization of CXCL12 via H25 at nonacidic pH value and in the presence of counterions. In our NMR measurements, we used a pH value of 6.8, where natural dimerization of CXCL12 is possible (Veldkamp et al. 2005).
Conclusion
By using a combination of HSQC NMR, molecular modeling and simulation, we analyzed the interaction between CXCL12 and GAGs of varying sulfation pattern and glycosidic linkages. The results demonstrate that HE and pHA bind strongest to the HAHBR site of CXCL12, followed by CS. A specific interaction of CXCL12 with HA could not be detected. We analyzed the impact of individual residues to CXCL12–GAG interactions in both sites: HAHBR and LAHBR. Here, we identified a predominant role of K24 and K27 in the HAHBR for sulfated GAG interaction, which is supported by the amino acids R41 and K43 in the first β-strand. For HE and CS, the strongest chemical shift was observed for K24 and the interaction of CXCL12 with pHA led to the highest change in chemical shift of K27. This observation supports the importance of sulfation pattern for GAG interaction. A different situation was found in the LAHBR. Here, interaction depended on the sulfated GAG. HE and pHA used amino acids of the receptor-binding site, whereas CS interacts with CXCL12 through N-terminally adjacent residues next to the receptor-binding site.
Using this knowledge could help to rationally design biomaterials with immobilized CXCL12 and thus lead to a controlled release and formation of prolonged stable haptotactic gradient to efficiently recruit tissue-specific stem cells and hematopoietic progenitor cells.
Materials and methods
Materials
The pTXB1 vector encoding hCXCL12 was prepared as described earlier (David et al. 2003). CS dp6 and HE dp6 were purchased from Iduron Ltd (Paterson Institute, Manchester, UK). 15N-labeled ammonium salts were purchased from Eurisotop (Saarbrücken, Germany) and all other chemicals from Sigma-Aldrich (Taufkirchen, Germany). HA dp4 and pHA dp4 were obtained as described (Toida et al. 1999; Suzuki et al. 2001; Chaidedgumjorn et al. 2002; Köhling et al. 2016).
Protein expression
Escherichia coli strain ER2566 was transformed with pTXB1 vector including the DNA for human CXCL12 variant α (1–68 amino acids) C-terminally linked to Mxe intein-chitin-binding domain. The expression of the protein was performed in a high-density feed-batch fermentation. The cells were cultivated in 2 L minimal medium containing 15N-NH4Cl (1 g/L), 15N-(NH4)2SO4 (4.92 g/L) and glucose (15 g/L) at 37°C and pH 7.0. The CXCL12 fusion protein expression was induced with 1 mM isopropyl-β-d-thiogalactopyranoside at OD600 of 14. The expression was performed under the same conditions until an OD600 of 40 was reached. During the entire expression process, pH value, temperature and glucose concentration were tightly controlled, foam formation was kept to a minimum by using antifoam.
The cells were harvested and the cell pellet was resuspended in column buffer (20 mM HEPES, 150 mM NaCl, 1 mM ethylenediaminetetraacetic acid, pH 8.0). The cells were broken by passing them three times through French Press (SLM AMINCO®, SLM Instruments, Urbana, USA) at 700 bar cold on ice. Subsequently, DNase I and 4 mM MgCl2 were added and incubated for 1 h at 37°C. After centrifugation of the suspension (1 h, 5000 x g, 4°C), the inclusion bodies containing the CXCL12 fusion protein were washed two times with column buffer containing 0.2% Tween20 and two times with column buffer containing 1 M NaCl. The supernatant was discarded after each washing step. The inclusion bodies were solubilized in column buffer containing 8 M urea (1 mL/g pellet) over night at 4°C. After centrifugation (35 min, 45,000 x g, 4°C), the supernatant containing solubilized CXCL12 fusion protein was separated from pellet and solubilization was repeated one time with the remaining pellet.
Protein purification and refolding
The purification of CXCL12 was performed by using the IMPACT™ system. A column containing chitin beads (New England BioLabs, Frankfurt am Main, Germany) was prepared and equilibrated in 3 M urea column buffer. After loading the protein onto the column, the chitin beads were washed with 10 bed volumes of 2 M urea column buffer containing 0.2% Tween20. The intein cleavage was induced by one bed volume of 0.1 M dithiothreitol in 2 M urea column buffer including 0.2% Tween20 over night at 4°C. After elution, the CXCL12 was hydrolyzed by increasing the pH value from 8.0 to 10.0 for 2 h at 4°C. The reaction was stopped by decreasing the pH value to 8.0 and increasing the urea concentration to 8 M. The CXCL12 was purified by using preparative RP-HPLC [acetonitrile (ACN) gradient from 10% to 60% in 40 min, C18 column]. Finally, the protein was diluted in 6 M guanidine hydrochloride with 100 mM NaHPO4, pH 6.0 and refolded by rapid dilution in refolding buffer (20 mM TRIS, 0.3 M sucrose, pH 7.4) containing 0.3 mM reduced and 0.3 mM oxidized glutathione (GSH and GSSG) over night at 4°C. On the following day, the pH value of the solution was adjusted to 3.5. The folded protein was separated from the unfolded and misfolded ones by semi-preparative RP-HPLC (ACN gradient from 10% to 50% in 40 min, C18 column). After lyophilization of CXCL12, the protein was stored at −20°C. The characterization of CXCL12 was performed by SDS-PAGE and Fourier transform ion cyclotron resonance electrospray ionization mass spectrometry. The protein concentration was photometrically determined using the extinction coefficient of 8730 per M cm.
NMR spectroscopy
For NMR measurements, the 15N-labeled CXCL12 was dissolved in NMR buffer (20 mM MES buffer, pH 6.8) containing 5% D2O. Solution NMR experiments were performed on an Avance III 600 MHz NMR spectrometer (Bruker Biospin GmbH, Rheinstetten, Germany) using the TOPSPIN 2.1 software. Standard 1H-15N HSQC experiments were carried out at a temperature of 25°C using a recycle delay of 1 s and the standard Bruker pulse sequence fhsqcf3gpph (Mori et al. 1995).
For assignment of CXCL12, 1.2 mM protein was diluted in 500 µL NMR buffer and measured under denoted conditions. Additionally, the assignment was verified by 3D-TOCSY-HSQC and 3D-NOESY-HSQC. The parameters for these measurements originated from different publications (Bax and Davis 1985; Palmer et al. 1991; Davis et al. 1992; Kay et al. 1992; Schleucher et al. 1993; Schleucher et al. 1994). The analysis of the NMR data took place with the publication of Veldkamp et al. (2009) (PDB: 2KEE).
Titration experiments were performed in NMR buffer including 150 mM NaCl under the same conditions. For titration, the amount of GAG was added to protein solution, mixed in the NMR tube and the 1H-15N- HSQC NMR spectrum was recorded. For each titration step, the weighted CSP for each assigned residue was calculated corresponding to (Seo et al. 2010), where ΔδH and ΔδN represent the change of chemical shift in 1H and 15N dimension. The determination of the cut-off was performed by the calculation of the two times the standard deviation. The standard deviation (σ) was calculated using all CSPs from all titration steps and 2σ was used as a cut-off to find CSPs, which might indicate the binding site. The standard deviation was calculated using GraphPad Prism 5.01 and “Column Statistics” with 95% confidence intervals.
Blind docking
HA, deHE (with IdoA rings in 1C4 conformation), HE (with IdoA2S rings in 1C4 conformation) and pHA of dp2, dp4, dp6 and dp8 (where dp is the degree of polymerization defined as the number of GAG monosaccharide units) were modeled in AMBER11 to be used as ligands (Case et al. 2010). Charges were taken from the GLYCAM06 force field (Kirschner et al. 2008) and from literature for sulfates (Huige and Altona 1995). CXCL12 dimer structure was extracted from the structure of CXCL12/HE dp2 complex (PDB ID: 2NWG, 2.07 Å) (Murphy et al. 2007), minimized in MOE (2013.8) prior to docking using default parameters and was used as a receptor. Docking calculations were performed with Autodock 3 (Morris et al. 1998) with a spacing grid of 0.5 Å. GAGs were treated completely flexible and the receptor rigid. The whole receptor surface was used for docking. The Lamarckian genetic algorithm with an initial population size of 300 and a termination condition of 105 generations or 9995 × 105 energy evaluations was used. A total of 103 independent runs were performed. Spatial clustering of the 50 top docking solutions was done using the DBSCAN algorithm (Ester et al. 1996).
Molecular dynamics
The complexes obtained by docking were further used for MD simulations carried out in AMBER11 using ff99SB force field parameters for the protein and GLYCAM06 for the GAGs. The complexes were solvated in TIP3P octahedral periodic box with a minimal distance to the periodic box border of 6 Å, counterions were used. Two energy-minimization steps were carried out: 0.5 × 103 steepest descent cycles and 103 conjugate gradient cycles with harmonic force restraints on solute atoms, then 3 × 103 steepest descent cycles and 3 × 103 conjugate gradient cycles without constraints. Then, the system was heated up to 300 K for 10 ps, equilibrated for 50 ps at 300 K, 106 Pa in isothermal isobaric ensemble (NPT) and, finally, a 10 ns productive MD run was carried out in NTP. The SHAKE algorithm, 2 fs time integration step, 8 Å cut-off for nonbonded interactions and the Particle Mesh Ewald method were used as previously (Pichert, Samsonov, et al. 2012; Pichert, Schlorke, et al. 2012). For GAGs, pyranose rings were harmonically restrained. Free energy calculations and per residue decomposition were performed using MM-GBSA implemented in AMBER for 100 frames evenly distributed in the productive MD run.
Dynamic molecular docking
The interaction of CXCL12 with HE dp4 and HE dp6 was investigated via DMD. DMD is a targeted molecular dynamics (tMD)-based docking method with less radical approximations than usually applied in classical docking, especially accounting for the effects of receptor flexibility and explicit solvation (Samsonov et al. 2014). As shown previously, DMD yields reliable data, in particular about the importance of single protein residues in protein-GAG systems (Samsonov et al. 2014). The geometrical DMD parametrization used here was based on the CXCL12/HE-binding region as found in PDB entry 2NWG. The Cα atom of CXCL12’s residue E63 was chosen as core atom. Selection of the focus point yielded a tMD target distance of 25 Å. Other parameters were chosen as published earlier (Samsonov et al. 2014). Two DMD studies were performed, differing only in the type of the ligand molecule (one for HE dp4, one for HE dp6). Both studies involved 100 independent run. In total, about 4 μs worth of MD trajectory data was created and analyzed. For each free MD trajectory within both DMD studies, a SRED was performed as described earlier (Samsonov et al. 2014). Subsequently, the resulting data were ensemble-averaged.
Supplementary data
Supplementary data for this article is available online at http://glycob.oxfordjournals.org/.
Funding
This work was financially supported by the German Science Foundation (DFG, SFB-TRR67; A4, A6, A7, A8).
Acknowledgements
The authors thank Dr. Christian Berger and Mario Keller for the help during fermentation and Aleksandra Margetić for assignment of the CXCL12 amino acids.
Conflict of interest statement
None declared.
Abbreviations
ACN, acetonitrile; CS, chondroitin sulfate; CSP, chemical shift perturbation; CXCL12, stromal cell-derived factor 1α; deHE, desulfated heparin; DMD, Dynamic Molecular Docking; dp2, disaccharide; dp4, tetrasaccharide; dp6, hexasaccharide; dp8, octasaccharide; ECM, extracellular matrix; GAG, glycosaminoglycans; HA, hyaluronic acid; HAHBR, high-affinity heparin-binding region; HE, heparin; HS, heparan sulfate; HSQC NMR, heteronuclear single quantum coherence nuclear magnetic resonance spectroscopy; IMPACT™, intein-mediated purification with an affinity chitin-binding tag; LAHBR, low-affinity heparin-binding region; MD, molecular dynamics; MM-GBSA, molecular mechanics with generalised born and surface area solvation; NOESY, nuclear Overhauser effect spectroscopy; pHA, persulfated hyaluronic acid; RFFESH, receptor binding region Arg-Phe-Phe-Glu-Ser-His of CXCL12; SDF-1, stromal cell-derived factor 1α; SRED, single-residue energy decomposition; tMD, targeted molecular dynamics; TOCSY, total correlation spectroscopy
References
Author notes
Present address: Leiden Universiteit, Leiden 2311 EZ, The Netherlands.
Present address: Department of Chemistry, University of British Columbia, Vancouver, BC, Canada BC V6T 1Z1.
These authors contributed equally to this work.
