Homology modeling of the transmembrane domain of the human calcium sensing receptor and localization of an allosteric binding site.

A homology model for the human calcium sensing receptor (hCaR) transmembrane domain utilizing bovine rhodopsin (bRho) structural information was derived and tested by docking the allosteric antagonist, NPS 2143, followed by mutagenesis of predicted contact sites. Mutation of residues Phe-668 (helix II), Arg-680, or Phe-684 (helix III) to Ala (or Val or Leu) and Glu-837 (helix VII) to Ile (or Gln) reduced the inhibitory effects of NPS 2143 on [Ca2+]i responses. The calcimimetic NPS R-568 increases the potency of Ca2+ in functional assays of CaR. Mutations at Phe-668, Phe-684, or Glu-837 attenuated the effects of this compound, but mutations at Arg-680 had no effect. In all cases, mutant CaRs responded normally to Ca2+ or phenylalanine, which act at distinct site(s). Discrimination by the Arg-680 mutant is consistent with the structural differences between NPS 2143, which contains an alkyl bridge hydroxyl group, and NPS R-568, which does not. The homology model of the CaR transmembrane domain robustly accounts for binding of both an allosteric antagonist and agonist, which share a common site, and provides a basis for the development of more specific and/or potent allosteric modulators of CaR. These studies suggest that the bRho backbone can be used as a starting point for homology modeling of even distantly related G protein-coupled receptors and provide a rational framework for investigation of the contributions of the transmembrane domain to CaR function.

The calcium sensing receptor (CaR) 1 is a member of family C of the G protein-coupled receptor (GPCR) superfamily, which includes metabotropic glutamate receptors (mGluRs), ␥-aminobutyric acid receptors (GABA B Rs), and a large family of putative pheromone and taste receptors. In addition to the seven transmembrane helices, which are the signature characteristic of all GPCRs, members of this family have large extracellular domains (with structural homology to bacterial periplasmic binding proteins) that contain the agonist binding site(s) (1). CaR signaling includes G q -mediated activation of phosphatidylinositol phospholipase C, production of inositol 1,4,5-trisphosphate and diacylglycerol, followed by increases in intracellular Ca 2ϩ in all cell types examined, activation of G i -mediated pathways in some cell types (reviewed in Ref. 2), and, through an interaction of the CaR carboxyl terminus with filamin A, activation of the mitogen-activated protein kinase cascade (3,4).
CaR not only binds and is activated by Ca 2ϩ at its agonist binding site (localized to the amino-terminal 500 amino acids) (5), but also interacts with allosteric modulators via several sites that have been shown to be distinct from the agonist binding site. CaR activity (in the presence of Ca 2ϩ ) is allosterically modulated by amino acids (6), small peptides (7,8), as well as a family of structurally related phenylalkylamines (9,10). The phenylalkylamines are of particular interest, because both allosteric agonists (calcimimetics) and antagonists (calcilytics) have been identified, typified by NPS R-467 or R-568 (calcimimetics) (9) and NPS 2143 (calcilytic) (11). The putative binding site(s) for amino acids and potentially for peptides (including poly-L-arginine, protamine, ␤-amyloid) have been identified within the extracellular agonist binding domain by homology with metabotropic glutamate receptors, and have been localized to a triple serine motif, Ser-169 -171 (12). Mutant CaRs, which are no longer modulated by amino acids, still exhibit altered potency of Ca 2ϩ in the presence of phenylalkylamines (13), confirming that the phenylalkylamines interact at a distinct site. Chimeras between CaR and mGluRs have shown that the phenylalkylamine binding site is localized to the transmembrane domain of CaR (13)(14)(15), and recent studies have suggested that negatively charged residues within the e2 and e3 loops may contribute to binding of NPS R-568 (15,16). Identification and characterization of the phenylalkylamine binding site on CaR might provide a basis for development of more specific compounds capable of modulating CaR activity on the background of the relatively constant extracellular Ca 2ϩ concentrations observed in vivo.
With one high resolution crystal structure of a GPCR available (bovine rhodopsin bRho (17)), any homology modeling of GPCRs based on experimental structural information remains speculative and is only useful in combination with reliable experimental data. Implications of the structure of bRho on generating models for other GPCRs are addressed by Ballesteros and Palczweski (Ref. 18). Various attempts to build and refine atomic-level GPCR models from structural information (for example, see reviews in Refs. 19 -22) and/or first principles (e.g. 23,24) have been reported. In this work, we have opted for the pragmatic assumption that the backbone of bRho is a reasonable starting point for building a model of the 7TM region of human CaR, despite the fact that CaR is not a member of the rhodopsin family of GPCRs. Furthermore, distinctions regarding active or inactive morphologies in the structure were not incorporated into the model. We have assumed that the ligands of interest interact primarily with the helical transmembrane (7TM) parts of the receptor and refrained from building the extra-and intracellular loops (except for extracellular loop 1, connecting helices II and III, see below and in "Experimental Procedures"). Here we report the general features of the model for the transmembrane domain of human CaR, as well as the identification of residues predicted to contact the phenylalkylamines NPS 2143 and NPS R-568. Mutations at the predicted residues eliminate or severely attenuate the efficacy of the allosteric compounds, whereas the abilities of Ca 2ϩ or the amino acid phenylalanine to activate the receptor are largely unaffected. In addition to providing a basis for the development of more specific and/or potent allosteric modulators of CaR, these studies provide an example of the generalizability of the bRho backbone as a starting point for homology modeling of even distantly related GPCRs.

EXPERIMENTAL PROCEDURES
Generation of the Homology Model-The helical parts of bRho were isolated from the x-ray coordinates (protein data base file 1F88.pdb) and used as a template. The selection was moderately extended on both ends of the helices. The final sequence fragments of bRho chosen to serve as a template for model building are those listed in Table I. A specific routine was used to perform the sequence alignment of the 7TM domain of human CaR to the selected residues of the corresponding regions in bRho, taking into account that no insertions or deletions can be applied for the superposition of helices. In addition to the bRho sequence, the consensus sequence of the entire rhodopsin family was used in this procedure. The alignment was optimized with regard to various physical properties of the amino acids such as their formal charge, hydrophobicity, and size of side chains. Once a given sequence alignment was established, the side chains in bRho were replaced by the corresponding ones of CaR, using version 2.9 of the SCWRL program (25,26) with the side-chain conformation library bbdep01. Jul.sortlib. The resulting structure was refined by molecular mechanics, using the parm94 parameter set of AMBER (27) with the conjugate gradient minimizer as implemented in our in-house software Wit!P. The energy minimization was carried out with all C␣ atoms being fixed at this stage. Electrostatics were evaluated with the original force field charges, using a distance-dependent dielectric function with ⑀ ϭ 1 ϫ r. Helices II and III (presumed shorter than in bRho, according to the sequence alignment in Table I) were then connected to form the extracellular loop 1, and the final structure was again refined, this time allowing all the residues connecting helices II and III to move freely without any constraints.
The ligand NPS 2143 was assigned standard AMBER atom types resulting in acceptable force field parameters for this type of structure. Partial charges were computed with the MPEOE method in the Wit!P software. These charges are close to the RESP charges to be expected for such connectivities and were considered as suitable for the current work. Docking of the ligand into the receptor was carried out manually. The starting point for docking was the presumed close interaction between Glu-837 and the charged nitrogen on the ligand. The rest of the ligand was then positioned by avoiding severe steric overlap with the receptor, trying to embed the aromatic (i.e. hydrophobic) groups as deeply as possible into the receptor while respecting the internal strain of the ligand. The eventual refinement of the entire complex was again carried out by conjugated gradient minimization with the same settings as for the receptor alone, but this time allowing all C␣ in the receptor to move freely within 0.5 Å, applying a harmonic force of 10 kcal/mol/Å 2 beyond that distance ("tethering"). The resulting geometry is essentially free of strain in the framework of the applied force field. The docking of NPS R-568 into CaR was performed in the same way as for NPS 2143, however, the docking mode for the latter was used as the starting point.
Mutagenesis and Cell Transfections-Point mutations were produced in the CaR-EGFP background (fusion protein previously described in Ref. 28) by site-directed PCR mutagenesis and confirmed by restriction endonuclease digestion and direct sequencing. Sequences of primer pairs used for mutagenesis are available upon request. HEK-293 cells (American Type Culture Collection, Rockville, MD) were grown in high glucose Dulbecco's modified Eagle's medium (Invitrogen, Grand Island, NY), supplemented with 10% fetal bovine serum (Invitrogen), penicillin (50 units/ml), streptomycin (50 g/ml), and amphotericin B (2 g/ml) at 37°C, 5% CO 2 . For transient transfections, 1 g of each CaR construct, 2 l of NovaFECTOR (Venn Nova LLC, Pompano, FL), and 100 l of medium were premixed, then added to HEK-293 cells in a 24-well plate, supplemented with regular medium as previously described (28). Experiments were carried out 68 -72 h after transfection.
IP Formation Assay-Transiently transfected HEK-293 cells were seeded to confluence in 24-well plates and maintained for 1 day in full medium before labeling with myo[ 3 H]inositol (100 MBq/ml; ART/Anawa Trading, Wangen, Switzerland) for 24 h in serum-free Dulbecco's modified Eagle's medium. Cells were then washed and incubated at 37°C in a phosphate-free physiologic salt solution, and experiments were carried out exactly as previously described (45).
Measurement of Intracellular Ca 2ϩ -Single cell fluorescence measurements of [Ca 2ϩ ] i were made as previously described (28,32,33). Solutions containing the calcilytic NPS 2143 N-[(R)-2-hydroxy-3-(2cyano-3-chlorophenoxy)propyl]-1,1-dimethyl-3-(2-naphthyl)ethylamine ( Fig. 1A) or the calcimimetic NPS R-568 (R)-N-(3-methoxy-␣-phenylethyl)-3-(2Ј-chlorophenyl)-1-propylamine hydrochloride (Fig. 1B) were prepared from stock solutions in Me 2 SO and added to the extracellular solutions immediately prior to the experiment. NPS R-568 was synthesized as described by Nemeth et al. (9); NPS 2143 was prepared as in a previous study (46). All other chemicals were obtained from Sigma Chemical Co. Me 2 SO had no effect on the parameters measured when tested under identical experimental conditions. All experiments were carried out at room temperature (22-24°C). 10 -20 cells were analyzed during each experiment; experiments were repeated in at least three independent transfections unless otherwise noted. Selected regions of cells were excited at 340/380 nm and emitted light collected at 510 nm at intervals of 10 s using an imaging system (Universal Imaging Corp., West Chester, PA) based on the MetaFluor software package. Background images were obtained at the beginning of each experiment from an area devoid of cells. Calibration of [Ca 2ϩ ] i was done with a series of buffered calcium standards (Molecular Probes Inc., Eugene, OR) assuming a K d of 145 nM determined in vitro at 22°C (30).
Data Analysis-Average [Ca 2ϩ ] i in a given condition was calculated as the average of at least 10 consecutive time points for each cell analyzed. For studies of intracellular Ca 2ϩ oscillations, a cell was characterized as exhibiting oscillations if three or more consecutive peaks could be clearly differentiated, and the amplitude of the peaks was Ն 10 nM [Ca 2ϩ ] i . Data were fitted to the Hill equation by least squares minimization using the Marquardt-Levenberg algorithm (NFIT, Island Products, Galveston, TX). All statistical analyses were done with SPSS 10.0 (SPSS Inc., Chicago, IL) and SigmaPlot 2000 for Windows version 6.0 (SPSS Inc.). To compare the fractions of cells having intracellular Ca 2ϩ oscillations under distinct conditions, Chi-squared tests or Fisher's exact tests, for group frequencies of less than 5 cells, were applied. A p value of Ͻ0.05 was considered significant. Table I is the result of an iterative process following several rounds of model building and experimental verification or falsification. For all helices, the global location in the sequence is consistent with earlier estimates based on hydrophobicity analysis. Possible alignments with a good overall score (see "Experimental Procedures") were taken as a starting point; the final alignment emanates from additional information provided by the mutational studies. The very low homology of the 7TM domain of human CaR to that of bRho (the template) or to the rhodopsin family consensus, and the absence of various key residues usually conserved in the rhodopsin family, did not allow the straightforward use of standard sequence alignment tools. Nevertheless, the alignment algorithm that was used worked well for helices I, III, V, VI, and VII. Difficulties were encountered particularly for helix IV, and the alignment in Table I was ultimately chosen based on the proline doublet toward the C-terminal of the helix. Note that helix IV is not involved in any direct interactions with the ligands considered here, and experimental data challenging this alignment are therefore not available. Similar problems were encountered for helix II. In this case, however, the mutational studies demonstrating the importance of Phe-668 over Phe-667 in ligand binding provided critical evidence in favor of the alignment shown.

Sequence Alignment of hCaR with Bovine Rhodopsin-The alignment shown in
Building the Model and Docking the Ligands-Selection of residues to be mutated was originally based on a model for human CaR, which was built using the Baldwin C␣ trace of rhodopsin (31). Glu-837 at helix VII was selected as the main anchor point for binding of NPS 2143, and the ligand was embedded as described under "Experimental Procedures"; mutational studies concentrated on amino acids that were within a distance of 6 Å from the ligand in the preliminary 3-dimensional model, preferentially considering residues within the helices rather than within the loops. The primary results from the mutations were then incorporated into the refinement of the model of the CaR⅐NPS 2143 complex, which also resulted in refinement of the sequence alignment. Fig. 2A illustrates the predicted location of the allosteric modulator site at the extracellular transmembrane interface. Fig. 2B shows the results of docking trials using the calcilytic NPS 2143 and indicates the residues predicted to participate in bonding interactions with the ligand. The final model indicates that if the carboxylic group of Glu-837 forms a salt bridge with the protonated amino group in NPS 2143, then Phe-668 and Phe-684 may form hydrophobic interactions with the planar rings of the ligand. Additional interactions are suggested with Arg-680. As mentioned above, in earlier models based on different possible sequence alignments, it was also possible that Phe-667 might interact with the ligand. The final results from mutation studies (see below) indicate that this residue is not involved in the binding of the ligand, supporting the model presented here.
Initial Characterization of Predicted Ligand Interaction Sites-The initial model predicted potential interaction(s) of NPS 2143 with Phe-667 or Phe-668 in helix II, Arg-680 and Phe-684 in helix III, and Glu-837 in helix VII at the extracellular interface. Point mutations at each site were generated, with substitutions of Ala, Leu, and Val for each phenylalanine, and Ile and Gln for Glu-837. To screen out mutants that were inefficiently trafficked to the plasma membrane, we performed ELISA assays utilizing paraformaldehyde-fixed, non-permeabilized cells. The data, presented in Fig. 3, indicate that most mutants were well expressed at the plasma membrane; exceptions were the F668A, F668L, F668V and R680L, R680V mutants, which were present at Յ40% of wt CaR levels. As a preliminary screen of mutants for functional activity, HEK-293 cells were transiently transfected with wild type CaR-EGFP or a mutated CaR and IP formation was measured. A measure of the general functionality of the mutants was obtained from the responses to a range of extracellular Ca 2ϩ concentrations (2, 5, and 8 mM) (Fig. 4A), the relative potency of the allosteric antagonist NPS 2143 (at 0.1, 1, and 10 M) was determined in the presence of 8 mM Ca 2ϩ (Fig. 4B), and the potency of the allosteric agonist NPS R-568 (at 1 and 10 M) was determined in the presence of 2 mM Ca 2ϩ (Fig. 4C). Wild type CaR is activated in a concentration-dependent manner by extracellu- . Framed residues are those contained in the model. For bRho and the consensus sequence, only the sequence corresponding to these parts is shown (additional residues being represented by "ϳ"). The unframed hCaR residues are assigned as belonging to intracellular and extracellular loops. Loops are not included in the model, with the exception of the extracellular loop connecting helices II and III, this loop had to be modeled by unwinding parts of the original helical domains in the bRho template structure. White on black residues denote identity between the hCaR sequence and bRho or the consensus (or both); gray-shaded residues indicate close matches: L/I/V, N/Q, D/E, K/R. lar Ca 2ϩ ; NPS 2143 causes a progressive decrease in IP formation at 8 mM Ca 2ϩ , whereas NPS R-568 causes a dramatic increase in IP formation at 2 mM Ca 2ϩ . Mutations at position Phe-667 had activation characteristics comparable to wild type CaR (Table II), whereas the adjacent position Phe-668 exhibits normal activation by Ca 2ϩ but attenuated responses to either NPS 2143 or NPS R-568 (Fig. 4, A-C). Mutations at positions Phe-684 and Glu-837 also exhibit attenuated responses to both allosteric ligands. Finally, mutations at position Arg-680 attenuate responses to NPS 2143 but have no effect on activation by extracellular Ca 2ϩ and NPS R-568 relative to wt CaR (Fig. 4).
[  intracellular Ca 2ϩ (32,33), in particular with regard to interaction(s) of the receptor with allosteric activators (32). Cells expressing comparable levels of wt or mutant CaR can be selected based upon GFP fluorescence intensity (all constructs are fused to EGFP at the carboxyl terminus), and receptor responses can be assessed with high time resolution. Therefore, to characterize the differences in sensitivity of the mutants identified in the preliminary screen (IP formation, Fig. 4) to the allosteric modulators, all mutants were analyzed for their ability to modulate intracellular Ca 2ϩ . Fig. 5 illustrates the doseresponse relations for extracellular Ca 2ϩ -mediated activation of CaR, monitored by changes in intracellular Ca 2ϩ . All receptor mutants were activated in a cooperative manner by increases in extracellular Ca 2ϩ , although several of the mutants exhibited modest changes in the EC 50 for Ca 2ϩ and/or the Hill coefficient ( Fig. 5 for mutations of key residues to alanine, and Table II for all substitutions). The reduced maximal responses observed for several of the mutants (F668A and R680A) correlated with the reduced plasma membrane localization determined by ELISA assay, whereas for others (F684A and E837I) the reduction in maximal response was observed despite enhanced surface localization (Fig. 3).
To determine whether point mutations of the amino acid residues proposed for the allosteric binding site (Phe-668, Phe-684, Arg-680, and Glu-837) decreased the potency of NPS 2143 for inhibition of [Ca 2ϩ ] i responses, transiently transfected HEK-293 cells were briefly exposed to 5 mM [Ca 2ϩ ] e to activate CaR, then increasing concentrations of NPS 2143 (0.1, 1, and 10 M) were added for 2-4 min each (Fig. 6A) Fig. 6B). The mutants R680A, F684A, and E837I showed no significant decrease in average [Ca 2ϩ ] i nor a decline in the percentage of cells exhibiting intracellular Ca 2ϩ oscillations in the presence of even the highest concentration of NPS 2143 (Fig. 6, A and B) Fig. 7 and Table II). As seen with NPS 2143, the NPS R-568 potency for the F667A mutant was comparable to wt CaR-EGFP (Fig. 7, analysis in Table II), confirming the importance of the neighboring Phe-  Table II (Fig. 7 and Table II). Mutations at Phe-684 (Ala or Leu) showed an increased EC 50 for Ca 2ϩ as well as a substantial reduction in the maximal receptor response (Fig. 7 and Table II), despite surface localization comparable to or greater than wt CaR-EGFP (Fig. 3). NPS R-568 was able to modestly decrease the EC 50 for Ca 2ϩ for both the F684A (from 3.8 Ϯ 0.18 to 2.99 Ϯ 0.24 mM) and F684L (from 6.91 Ϯ 1.93 to 5.06 Ϯ 0.66 mM) mutants ( Fig. 7 and Table II (33) and aromatic amino acids (34). Intracellular Ca 2ϩ oscillations elicited by Ca 2ϩ plus NPS R-568 are a very sensitive measure of CaR function, i.e. NPS R-568 plus elevated Ca 2ϩ is able to elicit intracellular Ca 2ϩ oscillations in mutated receptors more efficiently than Ca 2ϩ alone (32). Cells expressing mutations that attenuated or abolished the ability of NPS R-568 to decrease the EC 50 for Ca 2ϩ (Fig. 7 and Table III) were exposed to several experimental paradigms known to elicit robust intracellular Ca 2ϩ oscillations in wt CaR. Experiments were begun at 2 mM [Ca 2ϩ ] e , and cells were then sequentially exposed to 3.5 mM Ca 2ϩ , 5 mM phenylalanine, and 1 M NPS R-568 (with a return to 2 mM [Ca 2ϩ ] e between each test activator). At the end of the experiments, cells were exposed to 20 mM Ca 2ϩ to elicit maximal receptor activation (Fig.  8A), which serves as a measure of CaR expression. For any mutant CaR that did not exhibit [Ca 2ϩ ] i oscillations under these conditions, further experiments utilizing higher [Ca 2ϩ ] e and NPS R-568 concentrations (5 mM and 10 M, respectively) were performed (results in Table III). The percentage of cells having [Ca 2ϩ ] i oscillations when exposed to NPS R-568 was similar or only slightly reduced compared with wild type CaR in cells expressing F667A or R680A (Fig. 8B and Table III). Replacement of residues Phe-668 or Phe-684 by Ala or Leu or Glu-837 by Ile abolished or significantly reduced the [Ca 2ϩ ] i oscillatory behavior compared with wt CaR upon exposure to 1 or 10 M NPS R-568. Despite the loss of responsiveness of these mutants to NPS R-568, 3.5 mM Ca 2ϩ or 5 mM phenylalanine were still able to elicit [Ca 2ϩ ] i oscillations in cells expressing F668A or F668L, F684A or F684L, and E837I ( Fig. 8A and Table III). As demonstrated previously, NPS R-568 was more potent at inducing [Ca 2ϩ ] i oscillations than an increment of [Ca 2ϩ ] e in cells expressing wild type CaR (32). In contrast, [Ca 2ϩ ] e or phenylalanine were more potent at inducing [Ca 2ϩ ] i oscillations than NPS R-568, for cells expressing F668A, F684A, and E837I. Taken together, the results in Fig. 8 and Table III indicate that single point mutations at Phe-668, Phe-684, or Glu-837 eliminate or significantly reduce the potency of NPS R-568 for eliciting intracellular Ca 2ϩ oscillations, implicating these residues in the stabilization of NPS R-568 binding to CaR.

DISCUSSION
Obtaining high resolution structures of membrane proteins remains a difficult and low yield process, but homology modeling based on available structures has met with considerable success (35,36). In this report, we have used the high resolution x-ray structure for bRho as the basis for homology modeling of the transmembrane domain of human CaR, a member of a distinct family of GPCRs. To further refine the preliminary model, docking of the allosteric antagonist NPS 2143, known to interact with CaR via its transmembrane domain, was performed. Finally, the critical residues predicted to be in contact with NPS 2143 were mutated and the resultant receptors studied for their ability to be inhibited by NPS 2143 and to be activated by a related allosteric agonist, NPS R-568. Both the allosteric antagonist and agonist are phenylalkylamines, containing an amino group within the alkyl bridge between phenyl groups. The primary interaction of both ligands was proposed to be through a salt bridge interaction with Glu-837 at the extracellular end of helix 7. Subsequent to development of the preliminary model, Glu-837 was identified by Hu et al. (16) as a potential interaction site for NPS R-568 from a mutagenesis screen of extracellular loop acidic residues, confirming the importance of this residue in the model. Hydropathy plot predictions for the beginning and end of helices in CaR were insufficient for alignment purposes; alignments were refined utilizing both the bRho sequence as well as a consensus sequence for the entire rhodopsin family. A recent report utilized a similar approach based on the bRho structure to model the transmembrane domain of mGluR1 and identify an allosteric antagonist binding site (37).
Early sequence alignments with the bRho sequence (plus family consensus sequence) could not distinguish between res- idues Phe-667 and Phe-668 as a contributor to the NPS 2143 binding site, but initial mutagenesis screens utilizing the IP formation assay rapidly focused attention on Phe-668 (Phe-667 behaved like wt CaR in all assays). Experimental discrimination between two adjacent phenylalanine residues validates the orientation of helix II in the final model. The binding site of the allosteric agonist NPS R-568 is expected to overlap with that of NPS 2143. Because the active and inactive conformations of CaR, which are promoted by the two compounds, respectively, likely differ within the trans-   Table II. membrane domain, it was of interest to test the effects of NPS 2143 site mutations on the efficacy of NPS R-568. The characteristic increase in potency of extracellular Ca 2ϩ produced by NPS R-568 in wt CaR persisted in the F667A mutant but was significantly reduced in the F668A, F684A, and E837I mutants. Interestingly, R680A exhibited an NPS R-568-mediated increase in potency of [Ca 2ϩ ] e , indicating structure-dependent discrimination at this position between NPS 2143 (which contains a hydroxyl group coordinated with Arg-680 and is affected by mutations at this position) and NPS R-568 (which does not contain a hydroxyl group and is not predicted to interact with Arg-680). Further confirmation of the involvement of residues Phe-668, Phe-684, and Glu-837 in forming the NPS R-568 binding site comes from studies of intracellular Ca 2ϩ oscillations. For each residue, mutations to Ala (Val or Leu) eliminated or severely impaired the ability of NPS R-568 to elicit intracellular Ca 2ϩ oscillations, despite the continued ability of increments of [Ca 2ϩ ] e or the amino acid phenylalanine to elicit robust oscillations. Again, mutations at Arg-680 did not alter NPS R-568 responses. Fig. 9 illustrates the likely docking mode of NPS R-568 into the allosteric ligand site on CaR. Because NPS R-568 is somewhat smaller, offers less hydrophobic contacts, and is less sterically constrained than NPS 2143, its position within the binding site is not as well defined. Nevertheless, the reduction of NPS R-568 potency in receptors bearing mutations at Phe-668, Phe-684, and Glu-837 strongly argues that NPS R-568 is indeed interacting at this site.
While the manuscript for this report was under revision, Petrel et al. (47) published a model of the calcium sensing receptor that is based on a sequence alignment to rhodopsin that deviates considerably in several helices from the alignment presented in Table I. However, docking of the Calhex 231 structure (Fig. 1 in Ref. 47) into the receptor model presented here can easily be achieved at the site proposed for NPS 2143, and the effects of the mutations reported in their paper (47) can be accounted for (data not shown). Also, the importance of Arg-680 demonstrated in our studies fully supports the model presented in this report. It is further of concern that the alignment proposed by Petrel et al. (47) would force Arg-680 to form a tight contact with Glu-837 (assuming the rhodopsin template backbone remains fixed) and thus prevent it from interacting with the compounds studied, which appears inconsistent with  Table III. FIG. 9. Model for NPS R-568 binding to human CaR. Docking of NPS R-568 to human CaR, starting from a binding mode similar to that suggested for NPS-2143 (see Fig. 2B). The salt bridge with Glu-837 remains the predominant interaction, Phe-668 and Phe-684 make hydrophobic contacts, whereas there are no strong electrostatic interactions between the ligand and Arg-680 (which binds to an -OH group in NPS-2143, Fig. 2B). Roman numerals indicate helix number. the experimental data. Given these considerations, the alignment presented in Table I and the resulting model can account  for binding of NPS 2143, NPS R-568, as well as Calhex 231, and  thus represents a solid starting point for further experimental  validation. Family C of the GPCR superfamily has several unique features, the most striking of which is an extracellular domain where the agonist/competitive antagonist binding site is localized. In addition, many members of this family (CaR, mGluRs, and GABA B Rs) have transmembrane domain-localized allosteric sites for agonists and antagonists, which modulate receptor activation in conjunction with the primary agonist. The allosteric sites represent primary drug targets, providing enhanced subtype-specific effects due to sequence diversity within the transmembrane domains. For example, the Group I metabotropic glutamate receptors (types 1 and 5) discriminate poorly between drugs that bind at the agonist binding site, but several selective antagonists for mGluR1 receptors have been found (37,38). 7-(hydroxyimino)cyclopropan[b]chromen-1a-carboxylic acid ethyl ester (CPCCOEt) binds to mGluR1 near residues Thr-815 and Ala-818 of transmembrane helix VII (39), whereas 1-ethyl-2-methyl-6-oxo-4-(1,2,4,5-tetrahydro-benzo[d]azepin-3-yl)-1,6-dihydropyrimidine-5-carbonitrile (EM-TBPC) interacts with the mGluR1 transmembrane domain at a cluster of aromatic residues within helix VI (Trp-798, Phe-801, and Tyr-805), as well as at a residue in helix V (Val-757) (37). The allosteric antagonist MPEP (40) selectively inhibited mGluR5, with residues Pro-655, Ser-658, and Ala-810 contributing to the binding site. The compound did not affect mGluR1, but substituting the corresponding mGluR5 residues at the appropriate locations in mGluR1 conferred block by 2-methyl-6-(phenylethylnyl)pyrimidine (MPEP) on mGluR1. Furthermore, binding of MPEP was competitively antagonized by CPC-COEt, suggesting that the binding sites for the two drugs overlap (41). Following localization of the binding site(s) for a number of allosteric antagonists of mGluRs, hypotheses regarding their mechanism of action were formulated, i.e. how they prevent receptor activation. Activation of family A GPCR members has been localized to rotation and changes in pitch with respect to the membrane for two key helices (helices III and VI) (42), and it has been suggested that EM-TBPC antagonizes receptor activation by stabilizing the position of helix VI (37). It remains to be seen how allosteric antagonists with distinct or partially overlapping binding sites stabilize the inactive state of the receptor, although the development of distinct families of antagonists and a robust transmembrane model for this family of GPCRs should greatly facilitate these studies.
Specific allosteric agonists that enhance activation of mGluR1 in the presence of glutamate have been identified (RO 01-6128, RO 67-4853, and RO 67-7476), although their specific site(s) of interaction with mGluR1 have not been localized (43). Allosteric agonists have also been identified for GABA B Rs; CGP7930 and CGP13501 potentiate the effects of GABA both in vitro and in vivo (44). Although less information regarding the binding site(s) and contribution(s) to receptor activation is available for allosteric agonists for mGluRs and GABA B Rs, the present study suggests that sites for such agonists may overlap significantly with those of allosteric antagonists. Careful comparisons of residue contributions to binding sites for allosteric antagonists and agonists may provide critical insights into the conformational changes that occur upon receptor activation in this class of GPCRs.
In summary, we have utilized a homology modeling approach based upon the structure of bovine rhodopsin to derive a model for the human CaR transmembrane domain. Mutations at crit-ical residues attenuate the effects of both an allosteric antagonist and agonist, confirming key residues contributing to the ligand binding site. Given the importance of CaR in the regulation of parathyroid hormone secretion by the parathyroid as well as regulation of diverse cellular functions in many other tissues (2), this model can serve as the basis for the development of more selective and potent drugs. In addition, this model provides a rational framework for investigation of the contributions of the transmembrane domain to CaR function.