Molecular Mechanism of AMD3100 Antagonism in the CXCR4 Receptor

AMD3100 is a symmetric bicyclam, prototype non-peptide antagonist of the CXCR4 chemokine receptor. Mutational substitutions at 16 positions located in TM-III, -IV, -V, -VI, and -VII lining the main ligand-binding pocket of the CXCR4 receptor identified three acid residues: Asp171 (AspIV:20), Asp262 (AspVI:23), and Glu288 (GluVII:06) as the main interaction points for AMD3100. Molecular modeling suggests that one cyclam ring of AMD3100 interacts with Asp171 in TM-IV, whereas the other ring is sandwiched between the carboxylic acid groups of Asp262 and Glu288 from TM-VI and -VII, respectively. Metal ion binding in the cyclam rings of AMD3100 increased its dependence on Asp262 and provided a tighter molecular map of the binding site, where borderline mutational hits became clear hits for the Zn(II)-loaded analog. The proposed binding site for AMD3100 was confirmed by a gradual build-up in the rather distinct CXCR3 receptor, for which the compound normally had no effect. Introduction of only a Glu at position VII:06 and the removal of a neutralizing Lys residue at position VII:02 resulted in a 1000-fold increase in affinity of AMD3100 to within 10-fold of its affinity in CXCR4. We conclude that AMD3100 binds through interactions with essentially only three acidic anchor-point residues, two of which are located at one end and the third at the opposite end of the main ligand-binding pocket of the CXCR4 receptor. We suggest that non-peptide antagonists with, for example, improved oral bioavailability can be designed to mimic this interaction and thereby efficiently and selectively block the CXCR4 receptor.

The proposed binding site for AMD3100 was confirmed by a gradual build-up in the rather distinct CXCR3 receptor, for which the compound normally had no effect. Introduction of only a Glu at position VII:06 and the removal of a neutralizing Lys residue at position VII:02 resulted in a 1000-fold increase in affinity of AMD3100 to within 10-fold of its affinity in CXCR4. We conclude that AMD3100 binds through interactions with essentially only three acidic anchor-point residues, two of which are located at one end and the third at the opposite end of the main ligand-binding pocket of the CXCR4 receptor. We suggest that non-peptide antagonists with, for example, improved oral bioavailability can be designed to mimic this interaction and thereby efficiently and selectively block the CXCR4 receptor.
The CXCR4 receptor is expressed much more broadly than chemokine receptors in general, i.e. not only on a wide variety of leukocytes but also on cells outside the immune system; for example, in the central nervous system (1,2). In contrast to many chemokine receptors, the CXCR4 receptor is only activated by a single chemokine ligand, stromal cell-derived factor (SDF-1␣, 1 also called CXCL12). Like many chemokine receptors, the CXCR4 receptor is involved in the control of migration and tissue targeting, i.e. homing of leukocytes. Importantly, SDF-1␣ and the CXCR4 receptor play a central role for the anchorage of CD34ϩ stem cells in the bone marrow. The importance of the receptor is emphasized by the fact that targeted deletion of either the gene for CXCR4 or for its ligand in both cases leads to embryologic lethality (3)(4)(5). This is only rarely observed in knockouts of genes for chemokine receptors and, in fact, also rarely for 7TM G protein-coupled receptors in general. The CXCR4 receptor is also expressed on many different types of cancer cells, and it seems to function both as a survival factor and to be responsible for the "chemotactic" spread of cancer cells as metastasis, for example, to the bone marrow where the SDF-1␣ ligand for the receptor is produced in large quantities.
AMD3100, which is composed of two 1,4,8,11-tetraazacyclotetradecane (cyclam) moieties connected by a conformationally constraining linker, is a prototype non-peptide antagonist of the CXCR4 receptor. The compound was discovered as an anti-HIV agent long before it was realized that it functioned through specific blockade of the CXCR4 receptor, which is used as a co-receptor for cell entry by so-called X4 strains of the AIDS virus. AMD3100 is a specific CXCR4 antagonist that inhibits the binding and function of SDF-1␣ with high affinity and potency (6,7). AMD3100 has been shown to block the outgrowth of all X4 as well as dual-tropic (T cell-and macrophage-tropic) HIV variants in vitro. During the development of AMD3100, it was discovered that the compound increases white blood cell counts in the blood and, importantly, that the compound is able to mobilize stem cells from the bone marrow. Thus, currently the compound, in combination with G-CSF, is in clinical trials for stem cell mobilization for auto-transplantations in, for example, multiple myeloma patients. In fact, strong evidence has been presented which indicates that many other stem cell mobilizing regimes, such as cyclophosphamide and G-CSF, act through disruption of the SDF-CXCR4 system, thus emphasizing the crucial role of this system in the control of stem cell homing and mobilization (8).
In an initial study with just a few targeted mutations, we identified two acidic residues, Asp 171 (AspIV:20) and Asp 262 (AspVI:23), that are located in the main ligand-binding pocket of the CXCR4 receptor as being important for the binding of AMD3100 (9). That study was based on the knowledge of the strong preference of the cyclam moiety for interactions with carboxylic acid groups (21). Asp 171 and Asp 262 were also found to be essential for the function of the CXCR4 receptor as a co-receptor for HIV (10). The present study was aimed at performing a rather exhaustive mutational analysis ( Fig. 1) of the molecular mechanism of action of AMD3100 using the monoclonal antibody 12G5 and not SDF-1␣ as the radioligand. The reason for this is that the binding of the endogenous chemokine ligand was affected by certain mutations down in the main ligand-binding pocket, which limited the number of residues that could be addressed with that ligand as a probe. The 12G5 antibody is a relevant probe to use because the interactions of bicyclams with CXCR4 monitored by the inhibition of 12G5 binding follows a similar structure-activity relationship for the inhibition of HIV-replication (9,(11)(12)(13). Importantly, the rather simple binding mode of the prototype CXCR4 non-peptide antagonist AMD3100, which was found through mutational disruption of the binding of the non-peptide compound in the CXCR4 receptor, was proven through the transfer of the essential components of this binding site to the binding pocket of the otherwise rather distinct CXCR3 receptor.
Site-directed Mutagenesis-Point mutations were introduced in the receptors by the polymerase chain reaction overlap extension technique (14) using the wild-type CXCR4 or the wild-type CXCR3 receptor as template. All reactions were carried out using the Pfu polymerase (Stratagene) under conditions recommended by the manufacturer. The generated mutations were cloned into the eukaryotic expression vector pcDNA3ϩ. The mutations were verified by restriction endonuclease digestion and DNA sequencing (ABI 310, PerkinElmer).
Iodination of 12G5-The Bolton-Hunter reagent was dried by a gentle stream of nitrogen for 30 -60 min. 250 pmol of 12G5 was incubated on ice with 1 mCi Bolton-Hunter reagent in a total volume of 50 l of 0.1 mM borate buffer, pH 8.5, for 1 h. The reaction was terminated by the addition of 0.25 ml of the borate buffer supplemented with 0.2 M glycine and the Bolton-Hunter-labeled 12G5 separated from free Bolton-Hunter reagent by column chromatography (Econo-Pac DC10, Bio-Rad; Ref. 15).
Binding Experiments-COS-7 cells were transferred to culture plates 1 day after transfection. The number of cells seeded per well was determined by the apparent expression efficiency of the receptors and was aimed at obtaining 5-10% specific binding of the added radioactive ligand (2 ϫ 10 4 to 1 ϫ 10 5 cells/well for the different CXCR4 constructs). Two days after transfection, cells were assayed by competition binding for 3 h at 4°C using 32 pM [ 125 I]12G5 plus unlabeled ligand in 0.5 ml of a 50 mM Hepes buffer, pH 7.4, supplemented with 1 mM CaCl 2 , 5 mM MgCl 2 , and 0.5% (w/v) bovine serum albumin. After incubation, cells were washed quickly 2ϫ in 4°C binding buffer supplemented with 0.5 M NaCl. Nonspecific binding was determined as the binding in the presence of 0.1 M unlabeled 12G5. Determinations were made in duplicates.
Phosphatidyl-inositol Assay (PI-turnover)-COS-7 cells were transfected according to the procedure mentioned above. Briefly, 6 ϫ 10 6 cells were transfected with 20 g of receptor cDNA in addition to 30 g of the promiscuous chimeric G-protein, G␣⌬6qi4myr, which turns the G␣i- White letters in black circles indicate mutated residues, substituted with other amino acids to either reduce side chain size or neutralize charge (Ala or Asn) or to increase the size of the side chain (Phe or Trp) as a steric hindrance approach (see text and Table I for details).
coupled signal (the most common pathway for endogenous chemokine receptors) into the G␣q pathway (phospholipase C activation measured as PI-turnover; Ref. 17). One day after transfection, COS-7 cells (2.5 ϫ 10 4 cells/well) were incubated for 24 h with 2 Ci of [ 3 H]myo-inositol in 0.4 ml of growth medium per well. Cells were washed twice in 20 mM Hepes, pH 7.4, supplemented with 140 mM NaCl, 5 mM KCl, 1 mM MgSO 4 , 1 mM CaCl 2 , 10 mM glucose, and 0.05% (w/v) bovine serum albumin, and were incubated in 0.4 ml of buffer supplemented with 10 mM LiCl at 37°C for 90 min in the presence of various concentrations of chemokines or AMD analogs together with a constant concentration of chemokine corresponding to 80% of maximal stimulation. Cells were extracted by the addition of 1 ml of 10 mM formic acid to each well, followed by incubation on ice for 30 -60 min. The generated [ 3 H]inositol phosphates were purified on AG 1-X8 anion-exchange resin (18). Determinations were made in duplicates.
Calculations-IC 50 and EC 50 values were determined by nonlinear regression, and B max values were calculated by using Prism version 3.0 software (GraphPad Software, San Diego).

Mapping of the Binding Site for AMD3100 in the CXCR4
Receptor-Based on knowledge of cyclam-carboxylic acid inter-actions, two acid residues located at each end of the main ligand-binding pocket of the CXCR4 receptor, Asp 171 (AspIV: 20) and Asp 262 (AspVI:23), have previously been identified as key interaction points for AMD3100 (9,19). As shown in Table  I and Fig. 2, these 2 positions plus 15 other positions in TM-III, -IV, -V, -VI, and -VII of the CXCR4 receptor were probed by mutational substitutions using the radio-labeled monoclonal antibody, 12G5, as a radioligand. In most cases, the side chain was substituted with the small methyl group of Ala. However, in some cases, a structurally similar but uncharged Asn residue was introduced instead of a charged Asp; in some cases, steric-hindrance mutagenesis was performed through introduction of larger side chains such as Phe or Trp for Ala, Gly, or Ile (20). None of the substitutions impaired 12G5 binding to the CXCR4 receptor as reflected in the K d and B max values presented in Table I, indicating that the overall structure as well as the cell surface expression of the receptor was unaffected by the mutations. However, five of the substitutions impaired the binding of AMD3100 more than 10-fold. Substitution of As-

) for the wild-type CXCR4 and large amount of mutations located in the main binding pocket
The data were obtained from competition binding with 125I (Bolton-Hunter)-labeled 12G5 antibody as radioligand on transiently transfected COS-7 cells. Values in parentheses indicate the number of experiments. Maximum specific binding for each mutant is given as B max (fmol/10 5 cell). Fmut indicates the effect of the mutant on the affinity of each ligand compared to wildtype affinity.  Gray background indicates Ͻ10-fold decrease in affinity; yellow, 10-to 50-fold decrease; orange, 50-to 100-fold decrease; and red, Ͼ100-fold decrease. The actual binding affinities are shown in Table I. pIV:20 (Asp 171 ) to Asn has previously been shown to affect AMD3100 38-fold in competition against the natural chemokine ligand SDF-1␣ (9). However, when using 12G5 as a radioligand, this substitution only resulted in a 15-fold impairment of AMD3100 binding. In contrast, substitution of AlaIV:24 (Ala 175 , located four residues after AspIV:20) with a Phe residue affected AMD3100 binding 40-fold. In the x-ray crystal structure of rhodopsin, the residue which corresponds to Ala 175 is located at the start of extracellular loop 2 facing down into the main ligand-binding pocket. The relatively large effect of the steric-hindrance substitution of Ala 175 with Phe is in good agreement with the assumption that one of the cyclam rings of AMD3100 should be located in the pocket between TM-III, -IV, and -V, i.e. in front of Asp 171 , which is where the large aromatic side chain of the introduced Phe residue would be expected to be located in the (A175F) mutant.
None of the substitutions in TM-V affected AMD3100 more than 5-fold, whereas in TM-VI, the mutational analysis (as expected) pointed to AspVI:23 (Asp 262 ) but also to TyrVI:16 (Tyr 255 ) as being interaction points for AMD3100. Surprisingly, substitution of IleVI:20 (Ile 259 , located in between these two residues and facing right into the middle of the main ligandbinding pocket) had only minimal effect on AMD3100 binding (Table I and Fig. 2). Even the introduction of a large Trp residue in position VI:20 only gave a 5-fold effect on AMD3100 binding.
In TM-VII, substitution of IleVII:02 (Ile 284 ) had no effect on AMD3100 binding, despite the fact that this residue (like IleVI: 20) is facing directly into the main pocket and at a location close to AspVI:23. In contrast, substitution of GluVII:06 (Glu 288 ) to Ala affected AMD3100 around 70-fold.
Thus, besides the previously described Asp 171 and Asp 262 , the major result of the mutational analysis using 12G5 as a radioligand was the identification of an additional acid residue Glu 288 as a potential interaction point for AMD3100.
Effect of Zn(II) on the Mutational Mapping of AMD3100 Binding-Binding of Zn(II) in the cyclam rings of AMD3100 is known to increase its affinity for the CXCR4 receptor around 35-fold (Table I; Refs. 21 and 22). Previously, we have found that this increased affinity in fact is determined by only one of the Zn(II) ions in one of the cyclam rings and that the effect is achieved through interaction with the carboxylic group of As-pVI:23 (Asp 262 ) (22). When the Zn(II)-loaded AMD3100 (also called AMD3479) was probed in the library of mutated CXCR4 receptors, an interesting picture emerged. The four to five residues that were identified as interaction points for AMD3100 itself were all positive, but the effects of the mutations were in all cases (except for Asp171) larger for the Zn(II)loaded version than for AMD3100 alone (Table I and Fig. 2). Moreover, a number of substitutions which had given small, single-digit effects for AMD3100 now gave clear, i.e. Ͼ 10-fold, effects for the Zn(II)-loaded version: 28-  (Table I and Fig. 2).
Important Acidic Residue in TM-VII for the Binding of AMD3100 -GluVII:06 (Glu 288 ) is clearly the most interesting among the novel potential interaction sites for AMD3100, because the only other mutation which has a relatively large effect on the binding of the antagonist (Ala 175 to Phe) is con- sidered to be a reflection of steric hindrance. As shown in Fig.  3, by using [ 125 I]12G5 as a radioligand, the Glu 288 to Ala substitution shifted the competition binding curve for AMD3100 with or without Zn(II) 70-to 90-fold to the right. We were unable to use [ 125 I]SDF-1␣ as radioligand in binding experiments in this construct because of a lack of specific binding (data not shown). This is very likely a reflection of a decreased affinity of SDF-1␣ on the Glu 288 to Ala mutant receptor, as the potency of SDF-1␣ in signaling was impaired 87-fold in this construct (Fig. 4A). Thus, Glu 288 seems to be a residue that is critical for the function of SDF-1␣ as an agonist on the CXCR4 receptor. By using appropriate sub-maximal doses of SDF-1␣ on the wild-type and on the Glu 288 to Ala mutant form of CXCR4, respectively, it was possible to demonstrate that, not only the affinity (Table I and Fig. 3), but also the potency of AMD3100 (with or without Zn(II)) were highly dependent on the presence of a Glu in position VII:06. Thus, in the wild-type CXCR4, receptor potencies of 79 nM and 17 nM were observed for AMD3100 and the Zn(II)-loaded version, respectively, whereas right shifts of 23-to 116-fold of the bicyclam inhibition curves were observed in the Glu 288 to Ala mutant form of CXCR4 (Fig. 4B).
Thus, the overall picture for the AMD3100 binding mode is that it is critically dependent on two acid residues at the extracellular ends of transmembrane segments VI and VII, AspVI:23 and GluVII:06, respectively, plus one Asp residue located at the opposite end of the main ligand-binding pocket, i.e. in position IV: 20. Surprisingly, few other residues that are known to be lining each end of the pocket as well as the space in between seem to be directly involved in the recognition of the relatively large bicyclam compound. For the Zn(II)-loaded version of AMD3100, a similar picture is observed; however, the dependence on Asp 262 is increased 10-fold, and a number of other residues on the inner face of TM-IV, -V, and -VI are picked up in the mutational analysis.
Transfer of the Binding Site for AMD3100 to the CXCR3 Receptor-A survey of the main ligand-binding pocket of all human 7TM receptors revealed that the combination of AspIV: 20, AspVI:23, and GluVII:06 is unique to the CXCR4 receptor, which is in agreement with the fact that AMD3100 is known to be a highly selective antagonist for the CXCR4 receptor (23). However, among the chemokine receptors, we verified that the CXCR3 receptor had two of the three residues, i.e. AspIV:20 and AspVI:23, but that it had a Ser residue in position VII:06 instead of a Glu as in CXCR4 (Fig. 5). Interestingly, at position VII:02, a Lys residue is located in the CXCR3 receptor (Lys 300 ), which very likely will form a neutralizing salt bridge with AspVI:23 (Asp 278 ) because of the close proximity of the extracellular ends of TM-VI and VII (Fig. 5). Thus, in the CXCR3 receptor, it is very likely that of the three proposed key residues, only one, AspIV:20 (Asp 186 ), is available for interaction with AMD3100. To try to verify the proposed binding site for AMD3100, which in the CXCR4 receptor is based on mutational "destruction" of the binding and function of the nonpeptide antagonist, we attempted to use the binding pocket of the CXCR3 receptor as a scaffold to gradually build up the binding site for the antagonist.
As expected, AMD3100 had no effect on the inositol phosphate accumulation induced by the agonist chemokines ITAC or IP10 in the CXCR3 receptor. Moreover, neither introduction of Glu at position VII:06 nor substitution of LysVII:02 with a non-neutralizing Ala residue changed this (Fig. 6). However, in the CXCR3 construct, where these two substitutions were combined to ensure the presence and exposure of all three proposed key interacting residues for AMD3100, the bicyclam antagonist inhibited in a dose-dependent manner both the ITAC and the IP10-induced signaling, with IC 50 values of 1.1 and 9.3 M, respectively (Fig. 6). However, only a partial inhibition down to 35-40% of the maximal stimulation was observed with AMD3100. As shown in Fig. 7, this was not the case for the Zn(II)-or Ni(II)-loaded analogs of AMD3100 which acted as full antagonists.
Metal Ion-loaded Versions of AMD3100 Are More Potent and Efficacious Antagonists-Besides the fact that the metal ionloaded versions of AMD3100 completely inhibited the agonist activity in the "CXCR4-mutated" CXCR3 receptor (Fig. 7), a surprisingly high potency of these compounds was observed. Thus, for the Zn(II)-loaded AMD3100, left-shifts of 19-to 69fold were observed in the inhibition of the ITAC-and IP10induced activity, whereas for the Ni(II)-loaded AMD3100, leftshifts of 4-to 70-fold were observed (Fig. 7).
Gradual Construction of the AMD3100-binding Pocket in CXCR3-When the Zn(II)-loaded AMD3100 was probed, it became clear that its binding site could gradually be built up in the CXCR3 receptor to as close as 4-to 8-fold from the affinity of the CXCR4 receptor (Fig. 8). Thus, just the Ala substitution of LysVII:02 (K300A), which is proposed to "neutralize" As-pVI:23 (Fig. 5), shifted the dose-response curve for AMD3100 43-fold to the left in inhibition of IP10-induced signaling (Fig.  8B). Introduction of Glu in position VII:06 (S304E) had an even larger effect, 113-fold, and the combination of the two substitutions (K300A,S304E) gave a potency for AMD3100 of ϳ136 nM, which is an 820-fold improvement from the wild-type CXCR3 receptor and within 10-fold of that observed on the CXCR4 receptor. For the inhibition of ITAC-induced signaling, an even larger effect was observed for the combined CXCR4mutated CXCR3 receptor, K300A,S304E-CXCR3, because a to-tal of 1420-fold improvement was observed for this construct. Also, in respect of the two single mutations K300A-CXCR3 and S304E-CXCR3, a gradual build-up of the binding pocket was observed when looking at the ITAC-induced signaling (Fig. 8A).
Probing Additional Presumed Interaction Points for AMD3100 in the CXCR3 Receptor-It was assumed above that in the CXCR3 receptor, the two naturally occurring Asp residues in positions IV:20 (Asp 186 ) and VI:23 (Asp 278 ) formed the basis for the high affinity binding of the AMD3100 bicyclam when they were combined with the Glu, which was introduced at position VII:06 (and when the disturbing LysVII:02 had been removed). This assumption was supported by the observation that the dose-response curve for the Zn(II)-loaded AMD3100 inhibition of ITAC was shifted respectively 51-fold and 38-fold to the right when either AspIV:20 (D186N) or AspVI:23 (D278N) were mutated to Asn in the K300A,S304E-CXCR3 receptor background (Fig. 9). DISCUSSION In the present study, a rather exhaustive mutational analysis of essentially all residues facing the main ligand-binding pocket of the CXCR4 receptor identified only three residues, all acidic, located in TM-IV, -VI, and -VII as being critical for the binding of the bicyclam non-peptide antagonist AMD3100 (Fig.  10). The assumption that AMD3100 acts through binding to essentially only these three anchor-point residues was confirmed through the construction of the tri-acidic motif in the binding pocket of the CXCR3 receptor, which is structurally rather distinct in the remaining part of the main ligand-binding pocket. Nevertheless, it bound AMD3100 with an affinity within 10-fold of that of the CXCR4 receptor after the three acidic residues were all made accessible in the correct CXCR4like positions in the CXCR3 pocket.
The Binding Mode of AMD3100 with and without Zn(II) in the CXCR4 Receptor-From a chemical point-of-view, AMD3100 is a rather atypical non-peptide antagonist, which probably relates to the way it was originally discovered (24). A characteristic feature of the vast majority of non-peptide ligands for 7TM receptors is that they are composed of two or usually more aromatic, hydrophobic moieties connected through some conformationally constraining bonds. However, the only aromatic part of AMD3100 is the phenyl moiety in the 1,4-dimethylene(phenylene) linker, which connects the two cyclam rings. Previous structure activity relationship analysis demonstrated that the function of this linker is not based on its aromaticity but rather on the fact that it simply constrains the mobility and distance between the two macrocycles (12,13). In agreement with this, our mutational analysis of the CXCR4 receptor did not point to a dependence of AMD3100 upon side chains of any residue located in the part of the main ligand-binding pocket that would be expected to be where the linker was located (Table I), i.e. on the inner face of TM-III, -V, or -VI, for example. This point is emphasized by the observation that an affinity of almost the same magnitude could be obtained in the mutated CXCR3 receptor, despite the fact that this receptor has, for example, two Args on the inner face of TM-V where CXCR4 has a Val and a Gln, and that CXCR3 has a Gly and a Phe on the inner face of TM-III where CXCR4 has a His and a Leu, respectively (Figs. 1 and 5).
The two large, symmetrical cyclam macrocyclic rings are also very atypical chemical moieties in 7TM small molecule ligands. However, although these rings are complex structures, the fact that they have a strong preference for binding to carboxylic acid groups and that they can do so in a couple of different but strong manners we believe offers a unique key to understanding the over-all binding mode and molecular mechanism of action of AMD3100 in the CXCR4 receptor. At physiological pH, the cyclam ring has an overall charge of ϩ2 and can adopt the most stable so-called trans-III R,R,S,S type of conformation at the four nitrogen atoms (25,26). X-ray and neutron diffraction analysis of a protonated cyclam complex with 4-tert-butylbenzoic acid demonstrated a direct complex of the ring system with a carboxylic acid group (27). Three non-equivalent hydrogen bonds (one strong, one intermediate and one weak) could be formed between the oxygens of the carboxylic acid and the amines of the cyclam ring (27). With Zn(II) or another divalent transition metal ion bound in the middle of the ring, cyclam adopts the thermodynamically favored trans-III configuration to also give a complex with an overall charge of ϩ2 (6,28,29). In complexation with carboxylic acids, the metal ion coordinates the uni-dentate C-O Ϫ , whereas the non-coordinated oxygen of the carboxylate group forms a hydrogen-bond with a secondary amine proton of the ring (7,28). Recently, x-ray and NMR analysis of the bis-zinc complex of AMD3100 demon-  strated that the cyclam rings were found in two major conformations, trans-I and trans-III (30). Importantly, acetate induced a conformational change in the cyclam ring to the unusual cis-V R,R,R,R configuration, with a bidentate coordination of one molecule of acetate to the metal ion on one side of the ring and a second acetate bond through hydrogen-bonds to the opposite side of the ring.
In the present study, we find that AMD3100 binds to the CXCR4 receptor and to the mutated CXCR3 receptor in what seems to be a rather simple way through interaction with three acidic residues located in TM-IV, -VI, and -VII, of which Glu 288 in TM-VII is an important novel interaction site. In previous analysis of this system, we used SDF-1␣ for binding as the main assay, which, however, prevented us from identifying Glu 288 as an interaction site for AMD3100 because substitution of this residue eliminated binding of the radiolabeled ligand. Consequently, at that time, we suggested that the two cyclam rings of AMD3100 bound to Asp 171 in TM-IV and Asp 262 in TM-VI (9), respectively, and that chelation of a divalent metal ion such as Zn(II) in one of the rings of AMD3100 increased its affinity to CXCR4 through a specific interaction with the carboxylate of Asp 262 (22). In the present study, we can confirm that the metal effect is mediated apparently solely through interaction with Asp 262 , and whereas Asp 171 and Glu 288 both are highly important for the binding of the compound, the metal effect is confined to Asp 262 . In principle, the bicyclam could bind in three different ways between the three acidic residues, with each of the two cyclam rings interacting with two of the acidic residues, i.e. between Asp 171 and Asp 262 , between Asp 171 and Glu 288 , and between Asp 262 and Glu 288 . However, molecular modeling suggests that the carboxylates from all three acidic residues could, in fact, bind to AMD3100 at the same time. Thus, as previously suggested, the bicyclam could bind with its two cyclam rings interacting with Asp 171 (AspIV: 20) and Asp 262 (AspVI:23). This binding mode is one that can be obtained in a model of the CXCR4 receptor built over the inactive form of rhodopsin without moving the helical backbone but by basically only adjusting side chain conformations (9). In this configuration, the carboxylate from Glu 288 (GluVII:06) can in fact reach the opposite side of the cyclam ring which interacts with Asp 262 (Fig. 10). In other words, one cyclam ring could be "sandwiched" between Asp 262 and Glu 288 , whereas the other one is bound to Asp 171 at the other end of the main ligandbinding pocket.
As indicated above, there are several possible conformations of the cyclam ring with and without metal ions. With a bound metal ion, one possible configuration of the ring that is bound to the two acidic residues (one on each side) would be the cis-V conformation suggested by Sadler and coworkers, where the Asp 262 should be the one coordinating the metal ion and Glu 288 the one binding through hydrogen-bonds to the opposite face of the ring (30). Conceivably, Asp 171 would interact mainly through hydrogen-bond formation to the other cyclam ring or at least in a manner in which the metal ion does not provide any further advantage (22). Although the carboxylates of each of the acidic residues should, in principle, be able to interact optimally with a metal-loaded free cyclam ring, it is only Asp 262 which, in fact, is able to do so in the bicyclam AMD3100. Conceivably, steric clashes of the ring or conformational constraints induced by the binding of both of the rings at the same time in the binding pocket apparently prevent an optimal interaction of both of the rings at the same time.
It should be noted that, although the combination of AspIV: 20, AspVI:23, and GluVII:06 is unique to the CXCR4 receptor, GluVI:06 (i.e. Glu 288 in the CXCR4 receptor) is a residue which is conserved among the majority of chemokine receptors. Interestingly, it is not only in the CXCR4 receptor that GluVII:06 is an essential interaction point for non-peptide ligands. Thus, it has been shown that GluVII:06 is the most important interaction point for non-peptide antagonists in the CCR2 receptor (31,32). Similarly, in the homologous CCR5 receptor, it has been found that several classes of non-peptide antagonists are highly dependent upon GluVII:06 (33)(34)(35). Thus, it seems that GluVII:06 is a very useful anchor-point for non-peptide antagonists in chemokine receptors and possibly can be used in basically all receptors that express this residue. In this context, it is very interesting that an acidic residue in this position (VII:06) is rather specific for chemokine receptors and, thus, only rarely found in 7TM receptors in general (36). In addition to the CCR2, CCR5, and the CXCR4 receptors, non-peptide antagonists have been discovered and developed for several other chemokine receptors, as for instance the CCR1, CCR3, and the CXCR2 receptor (37)(38)(39).
Development of Novel CXCR4 Antagonists-AMD3100 is an efficient CXCR4 antagonist, which currently is in clinical trials as a stem cell mobilizing agent as an adjunct to G-CSF. The compound has a total of four positive charges at neutral pH and is not bioavailable orally. However, stem cell mobilization is an indication where parenteral administration is well accepted. Nevertheless, the CXCR4 receptor is potentially a very interesting target for several other indications such as blockade of HIV cell entry and chronic inflammation (1,3,40); for these indications, an orally active compound would be preferred. AMD3100 has some excellent properties as a CXCR4 antagonist, such as being a very broad spectrum blocker of nearly all types of gp120 envelope proteins from a large number of HIV viral isolates (41,42). Thus, it would be preferred that these properties, which very likely are inherent to the molecular mechanism of action of AMD3100 on the CXCR4 receptor, could be conserved in a non-peptide compound with improved oral bioavailability, for example. We believe that detailed knowledge of the molecular mechanism of action of AMD3100 as presented in this paper could serve as the basis for the design of non-macrocyclic (i.e. chemically distinct) compounds which nevertheless could mimic the action of AMD3100 on the CXCR4 receptor.
Acknowledgment-We thank Lisbet Elbak for excellent technical assistance.  (43) and shows the interaction of one of the cyclam rings of AMD3100(Zn 2 ) with AspIV:20 (Asp 171 in CXCR4 and Asp 186 in CXCR3), whereas the other cyclam ring is "sandwiched" between AspVI:23 (Asp 262 in CXCR4 and Asp 278 in CXCR3) and Glu-VII:06 (Glu 288 in CXCR4, which is a Ser residue in CXCR3 and part of the reason why AMD3100 does not bind to the wild-type CXCR3 receptor).