Identification of Edg1 Receptor Residues That Recognize Sphingosine 1-Phosphate*

Originating from its DNA sequence, a computational model of the Edg1 receptor has been developed that predicts critical interactions with its ligand, sphingosine 1-phosphate. The basic amino acids Arg120 and Arg292 ion pair with the phosphate, whereas the acidic Glu121 residue ion pairs with the ammonium moiety of sphingosine 1-phosphate. The requirement of these interactions for specific ligand recognition has been confirmed through examination of site-directed mutants by radioligand binding, ligand-induced [35S]GTPγS binding, and receptor internalization assays. These ion-pairing interactions explain the ligand specificity of the Edg1 receptor and provide insight into ligand specificity differences within the Edg receptor family. This computational map of the ligand binding pocket provides information necessary for understanding the molecular pharmacology of this receptor, thus underlining the potential of the computational method in predicting ligand-receptor interactions.

The G protein-coupled receptor (GPCR) 1 superfamily includes more than 2000 genes encoding receptors for bioactive molecules ranging from hormones through neurotransmitters (1). The physiological significance of these ligands makes GPCRs the target of many drugs and the subject of drug development studies. GPCRs are integral membrane proteins with physical properties that vastly increase the difficulty of standard methods of structure analysis. A precise understanding of their ligand-receptor interactions is essential for the rational design of ligands. Therefore, an atomic resolution map of the binding pocket including the locations of the interactions necessary for ligand binding would provide crucial drug development information. The endothelial differentiation gene (Edg) receptors are GPCRs that are activated by lysophospholipids ( Fig. 1). Five members of the Edg family (Edg1, Edg3, Edg5, Edg6, and Edg8) show a preference for sphingosine 1-phosphate (SPP) (2)(3)(4)(5)(6)(7). The remaining three Edg family members (Edg2, Edg4, and Edg7) are activated by lysophosphatidic acid (LPA) (8,9). The receptor-mediated effects of SPP include stimulation of cell proliferation, prevention of apoptosis, regulation of cell shape, adhesion, motility, vascular differentiation (10 -12), and cancer cell invasiveness (11,13). Therefore, receptors for SPP are important targets for the design of receptor-specific ligands both as potential therapeutic agents and to assist in the elucidation of the physiological function of the receptor.
Membership of the Edg receptors in the GPCR family confers significant homology with other GPCRs in regard to sequence, topology, and function (14). These shared features make the Edg receptors good candidates for homology modeling. Homology modeling uses a known template protein structure to build an analogous structure for a protein sequence having unknown structure. Homology modeling thus assumes that homologous function and amino acid sequences confer three-dimensional structural similarity. GPCRs have extracellular amino termini, intracellular carboxyl termini, and seven alpha-helical domains that span the cell membrane. These shared structural features provide strong constraints for molecular model development. The GPCRs sharing the greatest homology with the Edg receptors are the cannabinoid receptors, which have been previously modeled based on a low resolution structure of rhodopsin (15,16). The success of these post hoc studies in aiding the interpretation of pharmacological data and in predicting the impact of site mutations has encouraged us to undertake a model-driven approach to study the interaction between Edg1 and its natural ligand, SPP. Edg1 was selected for our initial work because it shares significant homology with both the SPPand LPA-specific members of the family and will serve as a good template for modeling the other Edg receptors.

EXPERIMENTAL PROCEDURES
Materials-SPP and LPA were from Avanti Polar Lipids (Alabaster, AL). Anti-flag M2 antibody and horseradish peroxidase-or fluorescein isothiocyanate-labeled anti-mouse were purchased from Sigma.
Edg1 Homology Model Development-The Edg1 model was developed using homology modeling in the MOE program (Chemical Computing Group, Montreal, Canada). The corrected Edg1 sequence from Gen-Bank TM (accession number AFF43420) was aligned against the transmembrane helices (THs) of a rhodopsin model (Protein Data Bank entry 1boj (17)). The amino-terminal 30 and carboxyl-terminal 50 residues were deleted, because no template structure was available for these regions. Manual realignments were made to remove gaps in the THs. The structure of the seventh TH of the dopamine D2 receptor was obtained from Dr. H. Weinstein (18) and mutated to produce the corresponding TH for Edg1. This modified structure was substituted into the template structure for homology modeling. The preliminary model was refined by converting cis-amide bonds in the loops to trans-amide bonds and by manually rotating side chains at polarity-conserved positions (19) to optimize hydrogen bonding between the THs. Geometry optimization was performed with the AMBER94 force field (20) to a 0.1 root mean square gradient. Interhelical hydrogen bonds formed that were retained upon minimization connect helices 1 and 2 (Asn 63 to Asp 91 ), 2 and 3 (Asn 86 to Ser 134 ), 2 and 4 (Asn 86 to Trp 168 ), 2 and 7 (Asp 91 to Ser 304 ), 3 and 7 (Ser 131 to Ser 304 ), and 4 and 5 (Trp 182 to His 2O1 ).
SPP Docking to Edg1-Docking of SPP was performed using the docking module in MOE with a docking box encompassing the majority of Edg1. Partial charges on protein atoms were applied from AMBER94 force field, and those on SPP atoms were generated using the semiempirical AM1 method (21) in the Spartan program (Wavefunction Inc., Irvine, CA). 25 different SPP-Edg1 complexes based on computer-generated random starting positions were generated with full ligand flexibility. These geometries were evaluated using the criteria from our previous modeling studies using the incorrect Edg1 sequence (Gen-Bank TM accession number M31210; Refs. 22 and 23), namely the presence of favorable interactions involving the phosphate group. Geometry optimization of the best complex obtained with the correct Edg-1 sequence (accession number AFF43420) was performed to a 0.01 root mean square gradient after manual optimization of ion-pairing interactions.
Site-directed Mutagenesis-The amino-terminal FLAG epitopetagged Edg1 receptor construct (GenBank TM accession number AF233365) was obtained from Dr. T. Hla. This receptor construct has been shown to behave the same as the wild type Edg1 receptor (24). Eight mutations of the EDG1 receptor were generated using the Ex-Site TM (Stratagene) mutagenesis kit. Positive clones were verified by complete sequencing of the receptor insert.
Cell Culture and Transfection-SPP non-responsive RH7777 cells (12) (ATCC, Manassas, VA) were maintained in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum (Summit Biotechnology, Ft. Collins, CO). Cells (10 6 ) were transfected with 1.5 g of plasmid DNA and 12 l of Cytofectene reagent (Bio-Rad) for 16 h. The following day, the transfection medium was replaced with 3 ml of fresh medium. The cells were used the next day for ligand binding and receptor activation assays.
Radioligand Binding Assays-The human embryonic kidney (HEK) 293 cell line (ATCC) does not endogenously express the Edg1 receptor and has been used for the heterologous expression of SPP receptors (6,25). The Edg1 mutants were tested for SPP binding after transient transfection into HEK293 cells as described previously (6).
Receptor Activation Assays-Functional assays were performed in RH7777 cells by measuring SPP-activated [ 35 S]GTP␥S binding. Transfected cells were homogenized in 20 mM HEPES, 50 mM NaCl, 2 mM EDTA (pH 7.5). Nuclei and cell debris were removed by centrifugation at 2,000 ϫ g for 5 min at 4°C. The supernatant was centrifuged at 40,000 ϫ g for 30 min at 4°C. Membranes were washed and resuspended in 50 mM HEPES (pH 7.5), 100 mM NaCl, 2 mM EDTA, 1 mM MgCl 2 and stored at Ϫ80°C. 5 g of membrane protein from Edg1 receptor expressing RH7777 cells was incubated in 1.0 ml of binding buffer (50 mM HEPES (pH 7.5), 100 mM NaCl, 1 mM MgCl 2 , 10 M GDP, 2 mM dithiothreitol, 0.1 nM [ 35 S]GTP␥S (1191 Ci/mmol; Amersham Pharmacia Biotech)). After incubating for 30 min at 30°C, bound radioactivity was separated by rapid vacuum filtration through a Whatman GF/B glass filter, and determined by liquid scintillation counting of triplicate samples.
Receptor Expression, Localization, and Internalization-For Western blot analysis, lysates from 24-h transfected cells were collected after an 8-h serum starvation, separated on 12.5% SDS-polyacrylamide gel elec-trophoresis, and transferred to polyvinylidene difluoride membranes (Bio-Rad). The primary antibody was the FLAG epitope tag-specific M2 monoclonal antibody (Sigma, 1:500 dilution), and the secondary antibody was a horseradish peroxidase-labeled goat anti-mouse antibody (Sigma, 1:4000 dilution). The bound antibody was visualized using the SuperSignal chemiluminescent substrate kit (Pierce).
RH7777 cells transfected with empty vector or plasmids containing Edg1 construct inserts were cultured on coverslips in Dulbecco's modified Eagle's medium with 10% fetal bovine serum. After an 18-h serum starvation, cells were fixed with 4% paraformaldehyde for 20 min and blocked with 10% fetal bovine serum in phosphate-buffered saline (pH 7.4), 0.2% Triton X-100, 0.05% NaN 3 at 4°C. The anti-Flag M2 monoclonal antibody (Sigma) was used at a 1:3000 dilution at 4°C for 12 h. Following three washes with phosphate-buffered saline, the cells were incubated with a 1:750 dilution of fluorescein isothiocyanate-conjugated anti-mouse IgG secondary antibody (Sigma), washed with phosphatebuffered saline, and mounted with Aqua-Mount (Lerner Laboratories, Pittsburgh, PA). For ligand-induced receptor internalization experiments, following 18 h of serum starvation, the cells were treated with 100 nM SPP at 37°C for 15 min. The cells were fixed, and the receptor protein was localized with indirect immunofluorescence using the anti-FLAG antibody as described above. Immunofluorescence staining was viewed by laser-scanning confocal fluorescence microscopy (Carl Zeiss laser scanning system LSM 510.

RESULTS
Three-dimensional Structure of Edg1 in Complex with SPP-A preliminary homology model of Edg1 was developed from a model of bovine rhodopsin (16) that is consistent with the 9-Å resolution structural data on the rhodopsin molecule (the best structural information available at the time model development began; Ref. 26). The preliminary model was refined by manually optimizing the interhelical network of hydrogen bonds at polarity-conserved positions followed by energy minimization (19). Hydrogen bonds formed that were retained upon minimization connect helices 1 and 2 (Asn 63 to Asp 91 ), 2 and 3 (Asn 86 to Ser 134 ), 2 and 4 (Asn 86 to Trp 168 ), 2 and 7 (Asp 91 to Ser 304 ), 3 and 7 (Ser 131 to Ser 304 ), and 4 and 5 (Trp 182 to His 2O1 ). The refined model was used in automated docking studies with SPP to generate a model of their complex. The model complex most consistent with the available data on the ligand selectivity of the Edg1 receptor shows three different ion-pairing interactions with the SPP head-group (Fig. 2C). Two of these involve cationic amino acids, Arg 120 and Arg 292 , predicted to be within 2 Å of the anionic phosphate group of SPP. The third involves an anionic amino acid, Glu 121 , positioned within 2.5 Å of the positively charged ammonium group of the sphingosine backbone. These ion-pairing interactions can provide an explanation of why the Edg1 receptor shows a strong preference for SPP over LPA, a glycerophospholipid that lacks an ammonium substituent. LPA is unable to ion-pair with Glu 121 , leaving the carboxylate of Glu 121 within 5 Å of the anionic phosphate of LPA with no counterion to mitigate the repulsive interaction between the two.
The theoretical model of the complex between Edg1 and SPP has been compared with the recently published 2.8-Å resolu- tion crystal structure of the inactive form of bovine rhodopsin (27). The root mean square deviation is 3.4 Å for the alpha carbon positions in the helical domains of these structures. A notably similar feature shared by the experimental structure of rhodopsin and the theoretical model of Edg1 is the network of interhelical hydrogen bonds formed by amino acids at corresponding positions. Table I shows these common interhelical hydrogen bonding networks using helical reference numbers (28) to point out the correspondence between the different sequences. The first part of each label is the number of the helix in which the residue appears. The second part of each label refers to its position in that helix relative to the most conserved residue in that helix, which is given the number 50. Notably, both structures involve the same residues, 1.50, 2.50, and 7.46, in a hydrogen bonding network. The rhodopsin structure (1F88 in the Protein Data Bank) has the backbone carbonyl of residue 7.46 occupying a central role, hydrogen bonding with both an asparagine (1.50) and a glutamate (2.50) side chain. This arrangement would require a protonated glutamate as shown in Fig. 3A. The Edg1 model has glutamate 2.50 occupying the central role, hydrogen bonding with the asparagine (1.50) and serine (7.46) side chains. The arrangement in the Edg1 model uses glutamate in its anionic form (Fig. 3B), which would not be possible in the rhodopsin sequence because the side chain of alanine 7.46 cannot donate a hydrogen bond. A second trio of residues in both structures mediates hydrogen bonds among helices 2, 3, and 4. In both structures asparagine 2.45 occupies a central role, hydrogen bonding with side chains of residues serine 3.42 and tryptophan 4.50. The rhodopsin structure includes an additional hydrogen bond between asparagine 2.45 and the side chain of threonine 4.49 that does not appear in the Edg1 model. Although the Edg1 model shares several features with the crystal structure of rhodopsin, there are also several notable differences between the structures. These may reflect deficiencies in the model, structural differences that arise because of the sequence differences between the two proteins, or differences between the inactive and active forms of the GPCR structures examined. The most notable difference is in the SPP binding pocket predicted by the Edg1 model. The position of the SPP phosphate in the Edg1 model is occupied by the extracellular beta sheet (␤4) that forms the top of the retinal binding site in the rhodopsin structure. The residues in rhodopsin (glutamine 3.26, glutamine 3.29, and phenylalanine 7.34) corresponding to the residues predicted by the Edg1 model to ion-pair with the phosphate of SPP (arginine 3.26, glutamine 3.29, and arginine 7.34) are thus oriented more toward neighboring helices or even outside the helical bundle to make room for ␤4.
Experimental Validation of the Edg1 Complex with SPP-Validation of the theoretical model of the Edg1 complex with SPP involved the generation of eight site mutations and evaluation of the ability of SPP to bind and activate these mutant receptors. Five of the site mutations, R120A, R292A, R292V, E121A, and E121Q, were designed specifically to evaluate the importance of the residues shown by the model to ion-pair with SPP. Two of these mutations, R292V and E121Q, introduced residues that are found in the corresponding positions in the LPA receptors of the Edg family. The other three site mutations, K111A, N101K, and N101I, were designed as controls to verify that changes in charge and other properties near to, but outside, the predicted binding pocket do not have as dramatic an impact on the ability of SPP to bind to, and activate the mutant receptors. In the model, Lys 111 and Asn 101 are 20.2 and 7.1 Å, respectively, from the docked phosphate and are not predicted to interact with SPP.
The level of expression and proper targeting of each mutant to the plasma membrane was verified by Western blots and confocal immunofluorescence microscopy using monoclonal antibodies recognizing an amino-terminal FLAG epitope (Fig. 4, A  and B). The mutants showed expression levels and localization patterns similar to the wild type Edg1. Fig. 5 shows the results of the radioligand binding assays for the wild type Edg1 receptor and the site mutants. Each mutation of an ion-pairing residue showed specific binding dramatically less than the wild type Edg1 receptor and similar to that of the vector-transfected controls. The binding results to mutants R120A, R292A, R292V, E121A, and E121Q showed no concentration dependence in the nanomolar range (data not shown), preventing Scatchard analysis to calculate K d values.
Site mutations in a GPCR can have an impact on the activation state of the receptor as well as its ability to interact with its natural ligand. To verify that the mutations have similar  a Helical reference numbers include a two-part label. The first part of the label indicates the helix in which the residue appears. The second part of the label refers to its position in that helix relative to the most conserved residue in that helix, which is given the number 50. effects on ligand binding and subsequent receptor activation and have not caused constitutive activation, receptor activation was assessed using a ligand-induced [ 35 S]GTP␥S binding assay. Fig. 6A demonstrates that four of the mutations to the proposed ion-pairing residues, R120A, R292A, R292V, and E121A, all produce a receptor that SPP is unable to activate at concentrations up to 100 nM. Calculated EC 50 concentrations for each mutant are shown in Fig. 6B. These results confirm and extend those from the radioligand binding assays that demonstrated that the receptor was unable to specifically bind SPP at a concentration of 1 nM. The fifth mutation involving an ion-pairing residue, E121Q, is a mutation that changes a residue conserved in the SPP binding Edg receptor subfamily (Edg1, -3, -5, -6, and -8) to a residue conserved in the LPA binding subfamily (Edg2, -4, and -7). This mutation, only at the maximal 1 M ligand concentration, showed a slight but detectable 17% activation relative to the response of the wild type receptor, indicating that the glutamine residue is able to interact somewhat with the protonated 2-amino group of SPP, perhaps by hydrogen bonding. It should be noted that this result is not in conflict with the radioligand binding assay, which was performed at an 8-fold lower concentration than the 8.1 nM K d for SPP on the Edg1 receptor (6) and gives a higher background in the higher concentration range due to the hydrophobic nature of the ligand. The control mutations K111A, N101K, and N101I showed activation results that were expected based on the radioligand binding results and were similar to that of the wild type Edg1.
Activation of a GPCR is terminated by ligand-induced internalization of the receptor, which can be followed by the translocation of the epitope-tagged receptor from the plasma membrane to the cytoplasm. In complete agreement with the results of the [ 35 S]GTP␥S binding assay, mutants of the predicted ion-pairing residues did not show internalization (Fig. 7), indicating that they were not activated during the prolonged 15min incubation with the 100 nM physiological concentration of the ligand. This is in contrast to the control mutations, which produced receptors that were activated and internalized by SPP treatment under the same conditions.

DISCUSSION
Starting from the DNA-derived amino acid sequence, a theoretical model of the Edg1 receptor and its complex with SPP has been developed. The Edg1 receptor model demonstrates an interhelical hydrogen bonding network that is very similar to that found in the recently reported 2.8-Å resolution crystal structure of bovine rhodopsin (27). Docking studies with this model have predicted that SPP best fits into a pocket formed within the helical bundle, a region that has been experimentally identified as the ligand binding pocket for several GPCRs   (14). The model predicts three ion-pairing interactions critical to the recognition of SPP by Edg1. These include two ion pairs between cationic amino acids, Arg 120 and Arg 292 , and the anionic phosphate of SPP, as well as a single ion pair between an anionic amino acid, Glu 121 , and the protonated amino group of SPP. The importance of these ion-pairing interactions identified by the model has been confirmed by site-directed mutagenesis and subsequent radioligand binding assays for both specific binding of SPP by the receptor and receptor activation. There are 42 basic amino acids (19 of them are arginine) and 20 acidic amino acids (10 of them are glutamate) in the cDNA-derived amino acid sequence of Edg1. Thus identification of the critical residues involved in ion pairing by either random or alanine/cysteine-scanning mutagenesis would have been prohibitive.
Several studies have delineated the specificity of ligand recognition by the Edg1 receptor. Non-phosphorylated sphingosine derivatives (sphingosine, sphinganine, ceramide) have been demonstrated not to compete with the binding of SPP and/or not to activate the receptor. Phospholipids lacking a basic amine (sphingomyelin, LPA, lysophosphatidyl inositol) also fail either to compete with SPP binding or to activate the receptor at physiologically relevant concentrations. Additionally, SPP-phosphonate, having one less atom between the cationic and anionic moieties, was unable to compete with SPP (29). Of the many compounds evaluated for Edg1 interaction, only dihydro-SPP (29), sphingosylphosphorylcholine (30), and a phosphonate (SPP-homophosphonate; Ref. 31) that maintains the appropriate chain length between the cationic and anionic moieties have shown the ability to displace SPP from Edg1. Dihydro-SPP and SPP-homophosphonate were almost as effective as SPP itself at displacing radiolabeled SPP, whereas sphingosylphosphorylcholine was 1-2 orders of magnitude FIG. 7. Agonist-induced internalization of the Edg1 receptor mutants. RH7777 cells expressing the wild type and mutant receptors were serum-starved for 18 h and prepared for immunocytochemistry. Panels on the left show non-treated cells, whereas panels on the right depict cells treated with 100 nM SPP for 15 min before fixation. Note that receptors unable to bind and be activated by SPP remain localized to the plasma membrane; in contrast, those receptors that are activated and bind SPP are internalized into the cytoplasm. The calibration bar is 25 m. weaker in binding affinity. These studies collectively demonstrate that the anionic phosphate and cationic amino groups, as well as the distance between them, are critical components of the interaction between SPP and Edg1.
The predicted interactions also provide insight into the ligand selectivity differences within the Edg receptor family. Relevant portions of the amino acid sequences of Edg1 through Edg8 have been aligned and are shown in Fig. 8. The SPPspecific receptors, Edg1, -3, -5, -6, and -8, all share an anionic residue that corresponds to the Glu 121 residue in the Edg1 receptor predicted by the model to interact with the ammonium of SPP. The LPA-specific receptors, Edg2, -4, and -7, instead have a neutral glutamine residue at that position that could interact with the neutral hydroxyl group in LPA.
Our work demonstrates the applicability of computational modeling to correctly predict the critical ionic interactions defining the ligand binding pocket of a GPCR, thus supporting the utility of a model-driven approach to gene function. The model also lays the foundation for the rational design of therapeutic agents targeted at Edg1.