Identification of the Binding Sites of the SR49059 Nonpeptide Antagonist into the V1a Vasopressin Receptor Using Sulfydryl-reactive Ligands and Cysteine Mutants as Chemical Sensors*

To identify the binding site of the human V1a vasopressin receptor for the selective nonpeptide antagonist SR49059, we have developed a site-directed irreversible labeling strategy that combines mutagenesis of the receptor and use of sulfydryl-reactive ligands. Based on a three-dimensional model of the antagonist docked into the receptor, hypothetical ligand-receptor interactions were investigated by replacing the residues potentially involved in the binding of the antagonist into cysteines and designing analogues of SR49059 derivatized with isothiocyanate or α-chloroacetamide moieties. The F225C, F308C, and K128C mutants of the V1a receptor were expressed in COS-7 or Chinese hamster ovary cells, and their pharmacological properties toward SR49059 and its sulfydryl-reactive analogues were analyzed. We demonstrated that treatment of the F225C mutant with the isothiocyanate-derivative compound led to dose-dependent inhibition of the residual binding of the radio-labeled antagonist [125I]HO-LVA. This inhibition is probably the consequence of a covalent irreversible chemical modification, which is only possible when close contacts and optimal orientations exist between reactive groups created both on the ligand and the receptor. This result validated the three-dimensional model hypothesis. Thus, we propose that residue Phe225, located in transmembrane domain V, directly participates in the binding of the V1a-selective nonpeptide antagonist SR49059. This conclusion is in complete agreement with all our previous data on the definition of the agonist/antagonist binding to members of the oxytocin/vasopressin receptor family.

The neurohypophysial antidiuretic hormone arginine vasopressin (AVP) 1 is involved in the regulation of body fluid osmo-lality, blood volume, and blood pressure via the stimulation of specific receptors currently classified into V 1a vascular (V 1a R) and V 2 renal (V 2 R) receptors. In addition, AVP modulates the adrenocorticotropic hormone secretion through V 1b pituitary (V 1b R) receptors. These different receptor subtypes along with the oxytocin receptor (OTR), which is classified in the same subfamily, possess distinct pharmacological profiles and intracellular second messengers (1,2). Moreover, AVP belongs to the family of vasoactive and mitogenic peptides involved in physiological and pathological cell growth and differentiation (3). AVP has been shown to be one of the most powerful in vitro vasoconstrictor substances, and its vasoconstrictor and mitogenic actions may contribute to the pathogenesis of arterial hypertension, heart failure, and atherosclerosis (4,5). AVP plays a role in the maintenance of blood pressure in several conditions, including upright posture, dehydration, hemorrhage, adrenal insufficiency, cardiac failure, and during surgery (6,7). An abnormal vascular reactivity specific for AVP has been noted in models of genetic and experimental hypertension, and AVP is instrumental in the genesis and maintenance of several models of experimental hypertension (4,5,7). AVP implication in the development or maintenance of hypertension, or both, is based on measurements of plasma and urinary AVP levels and responses to specific AVP antisera or peptide or nonpeptide antagonists (8 -10). The first potent and selective V 1a R antagonist to be synthesized is a cyclic peptide, d(CH 2 ) 5 [Tyr(Me) 2 ]AVP (11). In addition to cyclic peptide antagonists, linear peptide antagonists have been developed, such as HO-phenylacetyl 1 -D-Tyr(Me) 2 -Phe 3 -Gln 4 -Asn 5 -Arg 6 -Pro 7 -Arg-8 NH 2 (HO-LVA) (12). However, the lack of oral bioavailability and short half-life of these peptide compounds have limited their use in clinical medicine. During the last decade, nonpeptide AVP antagonists were discovered through random screening of chemical libraries (13,14). The availability of these orally active compounds now facilitates the assessment to the potential therapeutic applications of AVP receptor blockade in human diseases. One of these compounds, SR49059 ((2S)1-{(2R, 3S)-5-chloro-3-(2-chloro-phenyl)-1-(3,4-dimethoxybenzene-sulfonyl)-3-hydroxy-2,3-dihydro-1H-indole-2-carbonyl}pyrrolidine-2-carboxamide), is presently the most potent and selective orally active V 1a R antagonist described so far (14). It has a marked affinity, selectivity, and efficacy toward both animal and human V 1a R and is devoid of partial agonist activity. In healthy human volunteers, SR49059 inhibits exogenous AVPinduced platelet aggregation and vasoconstriction (15,16).
Up to now, identification of the binding site of the SR49059 into the V 1a R has not been investigated at a molecular level. Several studies based on a combination of receptor three-dimensional modeling and site-directed mutagenesis experi-ments have suggested that an AVP-binding pocket is buried into a 15-20-Å-deep central cavity of the V 1a receptor, defined by the transmembrane helices and surrounded by the extracellular loops (17,18). Extracellular residues also play a role in the binding of the hormone (19,20). Binding domains for synthetic peptide antagonists overlap those for peptide agonists into the AVP/OT receptors; however, discrimination of agonist versus antagonist ligands is achieved by conserved aromatic residues (Trp 304 , Phe 307 , and Phe 308 ) located at the bottom of the binding pocket in the transmembrane helix VI (21)(22)(23)(24). Although the binding mode of SR49059 into the V 1a R is not yet defined, nonconserved residues Thr 333 and Ala 334 located in transmembrane region VII have been shown to control the V 1a R/V 2 R receptor binding selectivity for SR49059 and for cyclic peptide antagonists as well (21). To precisely investigate the SR49059-V 1a R binding interactions at a molecular level, we first constructed a three-dimensional model of the antagonist docked into the receptor based on the x-ray crystal structure of bovine rhodopsin (25). To validate or invalidate the hypotheses of potential ligand-receptor interactions, we developed a sitedirected irreversible labeling strategy that combines mutagenesis of the receptor and binding of chemically reactive probes derived from SR49059. This approach generates a chemical bond between the nucleophilic moiety of a cysteine residue incorporated into the receptor and the electrophilic moiety of a sulfydryl-reactive group (isothiocyanate or an ␣-chloroacetamide) in the SR49059 analogue. Because such a chemical bond formation is only possible when a close contact exists between reactive elements created both on the ligand and on the protein, the covalent link can be taken as a proof of direct interaction versus a long range effect (26). This strategy combines the advantages of three-dimensional modeling and sitedirected mutagenesis techniques with those of a direct unambiguous identification of contact regions between a receptor and a specific ligand, like in photoaffinity labeling studies. This is an elegant complement to receptor photolabeling; however, it is much easier to develop than photolabeling, because it does not require the use of a radiolabeled ligand and the overproduction of the target receptor. However, this strategy necessitates introduction of a mutation in the receptor, which could be by itself detrimental to the overall structure of the protein. In this study, we demonstrated that the use of cysteine mutants of V 1a R as chemical sensors for sulfydryl-reactive ligands allows unambiguous localization of the SR49059 nonpeptide-binding site and offers a reliable molecular model of the V 1a receptor.

EXPERIMENTAL PROCEDURES
Materials-SR49059 and its affinity probe derivatives were synthesized according to a SANOFI Recherche Montpellier patent (27). The synthesis will be published in a paper to come. 2 [ 3 H]AVP (60 -80 Ci⅐mmol Ϫ1 ) was purchased from PerkinElmer Life Sciences. HO-LVA was kindly provided by Dr. M. Manning (Toledo, OH) and was radioiodinated in our laboratory as previously described (12).
Alignment of Amino Acid Sequences-The amino acid sequences of the three human AVP receptor subtypes and that of the OT receptor were retrieved from the Swiss-Prot data base (accession numbers were as follows: V 1a receptor, P37288; V 1b receptor, P47901; V 2 receptor, P30518; and OTR, P30559) and aligned to the sequence of bovine rhodopsin (accession number P02699) using the ClustalW multiple alignment program (28). A slow pairwise alignment using BLOSUM matrix series (29) and a gap opening penalty of 15.0 were chosen for aligning the amino acid sequences to the sequence of bovine rhodopsin. Because the disulfide bridge occurring between the third transmembrane segment (TM III) and the second extracellular loop (E II) in the structure of bovine rhodopsin is conserved in all AVP/OT receptors, we manually adjusted the alignment of E II to align the corresponding cysteines.
Preparation of Starting Protein Coordinates-The three-dimensional model of the human V 1a receptor was constructed by mutating the side chains of the amino acids in rhodopsin. Standard geometries for the mutated side chains were given by the BIOPOLYMER module of SYBYL (SYBYL 6.8, Tripos Inc., St. Louis, MO). Whenever possible, the side chain torsional angles were kept to the values occurring in bovine rhodopsin. Otherwise, a short scanning of side chain angles was performed to remove steric clashes between the mutated side chain and the other amino acids. The third intracellular loop between helices V and VI, which shows a high degree of variability, was not included in any of the three models. This loop is not involved in direct interactions with the ligand (31). We therefore assume that omitting this loop should not influence our docking results. Insertions/deletions occurred only in loops but not in secondary structure elements (␣-helix and ␤-sheet). The insertions/deletions in the loops were achieved through a simple knowledge-based loop search procedure using the LOOPSEARCH module of the SYBYL package. In this procedure, a set of 1478 high resolution x-ray structures was searched for loops of similar length and similar distance between the C␣ atoms of the residues delimiting the loop window. The loop showing the highest sequence identity and the lowest root mean square deviation was then selected for insertion in the model. Special caution had to be given with E II, which has been described in bovine rhodopsin to fold back over transmembrane helices (25) and thereby limit the size of the active site. Hence, amino acids of this loop could be involved in direct interactions with the ligands. A driving force to this peculiar fold of the E II loop might be the presence of a disulfide bridge between cysteines in TM III and E II. Because this covalent link is conserved in all receptors modeled in the current study, the E II loop was modeled using a rhodopsin-like constrained geometry around the E II-TM III disulfide bridge. After the heavy atoms were modeled, all of the hydrogen atoms were added, and the protein coordinates were then minimized with AMBER (32) using the AMBER95 force field (33). Topology and charge parameters for the nonpeptide V 1a antagonist were computed using a previously described procedure (34). The minimizations were carried out by 1,000 steps of steepest descent followed by conjugate gradient minimization until the root mean square gradient of the potential energy was less than 0.05 kcal/mol⅐Å. A twin cut-off (10.0 and 15.0 Å) was used to calculate nonbonded electrostatic interactions at every minimization step, and the nonbonded pair list was updated every 25 steps. A distance-dependent (⑀ ϭ 4r) dielectric function was used.
Modeling the Antagonist-bound V 1a Receptor Model-To obtain an "antagonist-bound" model of the V 1a receptor, the above-described coordinates were refined by the following procedure. Conivaptan, a known V 1a antagonist (35) was first manually docked into the putative active site (21) so that its shape optimally fit in that of the binding pocket. The resulting protein-ligand complex was then refined by minimization using the above-described AMBER parameters. Removing the ligand atoms from the minimized complex finally yielded one set of coordinates for an antagonist-bound receptor model.
Automated Docking of a Selective V 1a Antagonist (SR49059)-The Gold 1.2 docking program (36) was used to automatically dock the selective V 1a antagonist SR49059. For each of the 10 independent genetic algorithm (GA) runs, a maximum number of 1,000 GA operations was performed on a single population of 50 individuals. Operator weights for cross-over, mutation, and migration were set to 100, 100, and 0, respectively. To allow poor nonbonded contacts at the start of each GA run, the maximum distance between hydrogen donors and fitting points was set to 5 Å, and nonbonded Van der Waals energies were cut off at a value equal to the k ij well depth of the van der Waals energy for the atom pair i and j). To further speed up the calculation, the GA docking was stopped when the top three solutions were within 1.5 Å root mean square deviation. If this criterion is met, we can assume that these top solutions represent a reproducible pose for the ligand.
Site-directed Mutagenesis of the Human V 1a R-Point mutations F225C, Y300C, F308C, and K128C introduced in the cDNA sequence of the human V 1a R vasopressin receptor were directly achieved on the eukaryotic expression vector pCMV (37) using the QuikChange TM sitedirected mutagenesis kit (Stratagene). All of the mutations were verified by direct dideoxynucleotide sequencing (T7 TM sequencing kit; Amersham Biosciences).
Cell Culture, Receptor Expression, and Membrane Preparations-The wild-type human V 1a R and its mutants were transiently expressed in COS-7 cells or Chinese hamster ovary cells by electroporation as previously described (17,22). Briefly, the cells were suspended in electroporation buffer (10 7 cells/0.3 ml) and incubated with plasmid DNA (20 g of carrier DNA and 1-4 g of expression vector containing the mutant cDNA insert) for 10 min at room temperature before being pulsed (280 V, 950 microfarads; GeneZapper system, Kodak Scientific Imaging). After electroporation, the cells were plated in Petri dishes and grown in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10% fetal calf serum, 4 mM L-glutamine, 500 units/ml penicillin and streptomycin each, and 0.25 g/ml amphotericin B in an environment containing 95% air and 5% CO 2 at 37°C. To increase the level of expression of some mutants, the cells were treated overnight with 5 mM sodium butyrate (38,39). As already published, this treatment does not modify the pharmacological properties of the receptors (22,23). Transfected cells were harvested 48 -72 h after electroporation, and the membranes were prepared as already described (17). In short, the cells were washed twice in phosphate-buffered saline without Ca 2ϩ and Mg 2ϩ , Polytron-homogenized in lysis buffer (15 mM Tris-HCl, pH 7.4, 2 mM MgCl 2 , 0.3 mM EDTA), and centrifuged at 800 ϫ g for 5 min at 4°C. The supernatants were recovered and centrifuged at 44,000 ϫ g for 20 min at 4°C. The pellets were washed in Buffer A (50 mM Tris-HCl, pH 7.4, 5 mM MgCl 2 ) and centrifuged at 44,000 ϫ g for 20 min at 4°C. The membranes were suspended in a small volume of Buffer A, and the protein content was determined by the Bradford method (Bio-Rad) using bovine serum albumin as the standard. Aliquots of membranes were used immediately for binding assays or stored at Ϫ80°C.
Radioligand Binding Assays-The binding assays were performed at 30°C using [ 125 I]HO-LVA or [ 3 H]AVP as the radioligands and 1-3 g (for assays with [ 125 I]HO-LVA) or 10 -15 g (for assays with [ 3 H]AVP) of membrane proteins in standard radioligand saturation and competition binding assays as previously described (17). Briefly, the membranes were incubated in Buffer A supplemented with 1 mg/ml bovine serum albumin (binding buffer) and with radiolabeled and displacing ligands for 30 min (with . Indeed, because several washes of the membranes are necessary for eliminating the irreversible ligand after preincubation of the receptors, more than half of the protein content is lost during this protocol. Thus, 25-40 g of proteins are preincubated with the ligand when using the F225C mutant. Only 5-10 g of proteins are preincubated with the irreversible ligand when using the wild-type V 1a receptor because of a receptor expression level much higher than that of the mutant. After preincubating the membranes with the irreversible ligand and then extensive washing, the protein content used for measuring [ 125 I]HO-LVA residual binding is much lower, approximately 10 -15 g/assay for the F225C mutant and 2-4 g/assay for the wild-type V 1a receptor. For instance, in the residual binding experiments illustrated in Fig. 6, protein contents are as follows: 3.7, 2.3, and 2.3 g/assay for wild-type V 1a -expressing membranes that have been pretreated with vehicle, compound 2, and compound 4, respectively; 12.3, 14.6, and 13.4 g/assay for mutant F225C-expressing membranes that have been pretreated with vehicle, compound 2, and compound 4, respectively.

RESULTS
AVP and OT antagonists have been widely used as pharmacological tools to study the physiological and pathophysiological roles of AVP and OT, to determine tissue localization, and to define the precise pharmacological binding profiles of the different receptor subtypes. Currently, much effort is focused on the development of AVP and OT antagonists for a potential therapeutic use. As illustrated in Fig. 1, the different V 1a R antagonists fall into three major classes: 1) cyclic peptide antagonists related to the parent AVP structure, like d(CH 2 ) 5 [Tyr(Me) 2 ]AVP; 2) linear peptide antagonists that retain little resemblance to the natural hormone, such as HO-LVA; and 3) nonpeptide antagonists, like SR49059, derived from screening and subsequent optimization studies. As stated in the introduction, the SR49059 nonpeptide ligand is the most potent and selective orally active V 1a R antagonist described so far. Understanding how SR49059 binds to the V 1a R should facilitate the rational design of new potent and selective therapeutic agents useful in the treatment of arterial hypertension and congestive heart failure. To investigate the SR49059-V 1a R binding interactions at a molecular level, we generated a three- The sequence of the three human AVP receptors and that of the OT receptor were first aligned to that of bovine rhodopsin in a consistent manner. As seen in Fig. 2, class I GPCR-specific fingerprints (41) are all aligned at key positions of each of the seven transmembrane domains. Because the two cysteine residues forming a disulfide bridge between TM III and E II in the rhodopsin x-ray crystal structure (25) are conserved in the V 1a R, we postulated the existence of the same disulfide linkage in the target receptor, although no experimental data yet support this hypothesis. Taking into account the alignment described on Fig. 2 and the x-ray crystal structure of bovine rhodopsin, a three-dimensional model of the V 1a R was constructed as detailed under "Experimental Procedures." The model is suitable for an automated docking of the selective antagonist SR49059 and is compatible with most experimental data derived from studies undertaken to map the V 1a antagonist-binding sites (21-24, 42, 43).
As seen in Fig. 3, the dimethoxyphenyl group of SR49059 could be extensively buried among hydrophobic and aromatic amino acids (Phe 225 , Tyr 300 , Trp 304 , Phe 307 , and Phe 308 ) of transmembrane segments V and VI. Mutation of some of these residues is detrimental to antagonist binding, particularly those of Phe 307 and Phe 308 (21). The nearby indoline moiety could also develop hydrophobic contacts to TM III, V and VI, but the corresponding subsite is partly hydrophilic. Two glutamine residues (Gln 185 and Gln 311 ) known to map the SR49059 binding site (17,21) are in close proximity to the antagonist. Interestingly, the Lys 128 side chain that has previously been shown to play a key role in SR49059 binding to the V 1a receptor does not contribute by the polar nitrogen atom but by its apolar carbon atoms (Fig. 3A). Similar interactions have also been evidenced by x-ray diffraction analysis of major histocompatibility complex-peptide complexes (44). The present model also accounts for the selectivity of this compound toward the human V 1a receptor. Hence, the two residues (Thr 333 and Ala 334 ) located on top of TM VII and previously reported to control V 1a /V 2 binding selectivity (21) directly interact with SR49059 in the present model To validate the three-dimensional model described above, different analogues of SR49059 derivatized with sulfydryl-reactive moieties were synthesized. These compounds containing an isothiocyanate moiety or an ␣-chloroacetamide as chemical  (25). E1, E2, and E3 and I1, I2, and I3 indicate the position of extracellular and intracellular loops, respectively. Specific fingerprints of the rhodopsin-like GPCR family are displayed in dark blue. The residues displayed in red are proposed to delimit the ligand binding cavity of vasopressin/oxytocin receptors. Some of these residues have been experimentally demonstrated to play a role in the binding of peptide agonists and antagonists (17)(18)(19)(20)(21)(22)(23)(24). Amino acids in cyan boxes are proposed to control the binding selectivity for the vasopressin/oxytocin receptors toward the nonpeptide SR49059. The residue numbering of each receptor is indicated on the left of the alignment. Because the N-and C-terminal ends show a high degree of variability, the alignment starts 5 residues upstream the TM I and stops 5 residues downstream from the TM VII. reactive entities as well as their precursors are shown in Fig. 4. All of the chemical substitutions were introduced on the dimethoxyphenyl ring. 2 Assuming a conserved binding mode for close analogues of SR49059, the isothiocyanate moiety of compound 4 (or the ␣-chloroacetamide of compound 5) should be located close enough to three residues of the V 1a receptor (Phe 225 , Tyr 300 , and Phe 308 ) to afford covalent coupling with the respective cysteine mutants (F225C, Y300C, or F308C). These three residues were chosen to maximize chances of covalent linkage for three main reasons: 1) they are close enough to the dimethoxyphenyl moiety of SR49059, 2) their side chains are almost coplanar to the dimethoxyphenyl ring of the reference antagonist, and 3) their cysteine mutants could react with synthetically accessible reactive analogues of SR49059. We thus decided to construct these receptor mutants and to verify this hypothesis. Lys 128 , which is not supposed to directly interact with the isothiocyanate moiety of compound 4 (or the ␣-chloroacetamide of compound 5), was substituted with a cysteine residue as well as a negative control.
Affinities of the nonpeptide antagonist compounds were first FIG. 4. Chemical structure of the analogues derived from SR49059. Compounds 1-3 are inactive and cannot create a covalent bond with a sulfydryl moiety of a cysteine residue. Compounds 4 (isothiocyanate moiety) and 5 (␣-chloroacetamide moiety) are sulfydrylreactive. Based on the three-dimensional model of the SR49059 docked into the V 1a R, these two analogues could bind covalently to cysteine residues introduced in the receptor-binding pocket, assuming that close contact and optimal orientation can occur. determined for the V 1a receptor and compared with those of the parent ligand SR49059. As seen in Table I, all of the chemical substitutions led to decreases in affinity, with the SR49059 being the most potent. Interestingly, the two sulfydryl-modifying compounds 4 and 5 still displayed a relative high affinity for the V 1a receptor (6.3 and 16.3 nM, respectively), which is a good prerequisite for investigating their capacity to establish a covalent binding to the cysteine mutants incorporated into the receptor.
In a second step, the affinities of the SR49059 and its nonpeptide analogues to the cysteine mutants of the V 1a receptors were calculated. The results are also indicated in Table I. We could not express the Y300C V 1a mutant. Thus, only binding properties of mutants F225C, F308C, and K128C are displayed in Table I ϭ 3), respectively. These values are in the range of that measured for the wild-type V 1a receptor, K d ϭ 50 Ϯ 4 pM (21,22). Because compounds 4 and 5 are supposed to potentially create a covalent bond with the mutant receptors and because affinity of an irreversible ligand cannot be determined reliably in a conventional competition binding assay, only apparent K i values instead of true K i values were measured for these two nonpeptide ligands. The apparent affinity determined in this way may actually represent both reversible and irreversible binding. Thus, simultaneous covalent and competitive binding should overestimate the apparent K i for the mutant receptors.
In the case of each receptor mutant, affinities for the SR49059 and for its analogues were all diminished (Table I). For the F225C V 1a R mutant, the K i values for SR49059 and compounds 2-5 are decreased three to six times when compared with that obtained with the wild-type receptor, leading to concentrations still in the 10 -100 nM range. The moderate decrease in the binding affinity of SR49059 for the F225C V1a receptor mutant can be easily explained from the current molecular model. Hence, Phe 225 is only one of the eight amino acids delimiting a strong hydrophobic pocket interacting with the dimethoxyphenyl moiety of SR49059 (Fig. 3B). The single point mutation does not significantly modify the polarity and the physicochemical property of that subsite. Docking SR49059 into the three-dimensional model of F225C mutant alters neither the binding mode (root mean square deviation of 0.25 Å from the wild-type-bound SR49059 coordinates) nor the GOLD docking score (63.52 for wild type versus 60.49 for the F225C mutant). For F308C V 1a R and K128C V 1a R mutants, the affinities are much lower, being in the 100 -1000 nM range. Like for photoactivatable ligands (22,23), which can be covalently bound to their receptors upon UV irradiation, the affinity for sulfydryl-modifying affinity ligands is expected to be high to favor such covalent chemical reactions. Taking into account the previous results, using compound 4 (or compound 5) in combination with the F225C V 1a R mutant receptor and at a much lesser extent with F308C and K128C V 1a R mutants, thus represented a particularly favorable situation for applying the site-directed irreversible labeling approach.
In the next step, potential irreversible binding of the sulfydryl-reactive analogues of SR49059 was verified on the different cysteine mutants of the human V 1a receptor. To facilitate the covalent chemical modification of the receptor mutants, compound 4 was first used at a very high concentration (10 Ϫ5 M) during a 30-min incubation with COS cell membranes. As  I Affinities of SR49059 and its analogues for the human V 1a receptors Chinese hamster ovary cells or COS-7 cells were electroporated with a pRK5 vector containing a cDNA coding either for wild-type V 1a , mutant F225C, mutant F308C, or mutant K128C, and membranes were prepared as described under "Experimental Procedures." Depending on the mutant, the cells were treated overnight with 5 mM sodium butyrate before harvesting. For wild-type V 1a and F308C mutant, affinities (apparent or true K i ) for the different compounds were determined in competition binding assays by displacement of [ 3 H]AVP used at 1-2 nM and with 10 -20 g of membrane proteins/assay. The K d value of [ 3 H]AVP for the V 1a receptor expressed in CHO cells has been previously published and is 0.67 Ϯ 0.17 nM (22). For F225C and K128C mutants, because affinity for [ 3 H]AVP cannot be measured, affinities (apparent or true K i ) for the different compounds were determined in competition binding assays by displacement of [ 125 I]HO-LVA used at 100 -500 pM and with 1-3 g (for the wild-type V 1a ) or 10-20 g of membrane proteins/assay. Depending on the radioligand, nonspecific binding was determined in the presence of an excess of HO-LVA (400 nM) or AVP (10 M). The data were analyzed as described under "Experimental Procedures." All of the values in this table are expressed as the means Ϯ S.E. calculated from three independent determinations. ND, not determined. the K d and B max values were 0.15 Ϯ 0.02 nM and 0.15 Ϯ 0.016 pmol/mg of membrane proteins, respectively, some residual binding was measured (10 -30%), but we were unable to calculate the K d and B max parameters in this condition (Table II). This result suggested an irreversible and almost complete binding of analogue 4. The results obtained with F225C mutant V 1a receptor were very encouraging and led us to check whether the effect of the sulfydryl-reactive analogue was dose-dependent. The membranes expressing the F225C mutant were treated with different concentrations of analogue 4 and tested for [ 125 I]HO-LVA residual binding. As illustrated in Fig. 5, a dose-dependent effect of the nonpeptide antagonist was unambiguously measured. At 100 nM, the SR49059 analogue was almost without effect because residual binding of [ 125 I]HO-LVA represented 94.2 Ϯ 2.9% of that of the control condition (membranes treated with vehicle only). At 1 M, the residual binding was decreased, being 44.7 Ϯ 6.9%, whereas at 10 M, binding of the radioligand only represented 26.6 Ϯ 3.4%. Concentrations of compound 4 required to irreversibly inactivate more than 50% of receptors are higher than measured K i affinity values, suggesting that the covalent chemical reaction between the two partners is slower than their association/dissociation. One should also remember that, as explained above, because affinity of an irreversible ligand cannot be determined reliably in a conventional competition binding assay, only an apparent K i value (25 nM here for the F225C mutant) instead of a true K i value was measured for this nonpeptide ligand. The apparent affinity determined in this way may actually represent both reversible and irreversible binding. Thus, simultaneous covalent and competitive binding should overestimate the apparent K i for the mutant receptors. Consequently, the true K i value for this compound could be much higher, in the 100 nM range or even worse.
The specificity of the irreversible binding reaction using the wild-type V 1a receptor and the nonreactive analogue 2 (which is also the precursor of analogue 4) as two negative controls was then investigated. To be able to quantify residual binding (K d and B max values) for [ 125 I]HO-LVA after treatment of the membranes, we used analogue 4 at 1 M. This concentration has been demonstrated to inhibit approximately 50% of the radiolabeled HO-LVA ligand binding (Fig. 5). The membranes expressing either the wild type or the F225C mutant were treated with vehicle, analogue 2 (1 M), or analogue 4 (1 M). Representative Scatchard plots of the residual binding of [ 125 I]HO-LVA are illustrated in Fig. 6. As expected for the wild-type receptor, analogues 2 and 4 were without effect on the binding of [ 125 I]HO-LVA, and Scatchard representations were almost superimposable with that of the vehicle condition (K d were FIG. 5. Residual binding of [ 125 I]HO-LVA to mutant F225C receptor; dose-dependent inhibition effect of the isothiocyanate compound. COS cells were transfected with a pRK5 vector containing the cDNA for the mutant F225C V 1a R (4 g of DNA for 10 7 cells). 36 h after transfection, 5 mM sodium butyrate was added to the culture for increasing the receptor expression level. The cells were harvested 16 h later, and the membranes were prepared as described under "Experimental Procedures." The protein content of the membranes was calculated. Then the membranes were treated with vehicle (Me 2 SO) or different concentrations (10 Ϫ7 , 10 Ϫ6 , or 10 Ϫ5 M) of analogue 4 (isothiocyanate moiety) for 30 min in binding buffer. The membranes were washed with binding buffer and centrifuged at 21,000 ϫ g at 4°C. This step was repeated four times for eliminating the free ligand. Protein content was again measured, and the different batches were used for radiolabeled binding studies with [ 125 I]HO-LVA used at different concentrations (50 -800 pM). Scatchard representations of the saturation studies were analyzed to determine the K d and B max values. Residual binding of [ 125 I]HO-LVA corresponds to the B max value determined in each assay and is shown as a percentage of the B max value determined for the vehicle (Me 2 SO) condition. The actual B max value (Table II) Table II are expressed as the means Ϯ S.E. calculated from at least three independent determinations from three different batches of electroporated COS cells. COS cells were transfected with a pRK5 vector containing the cDNA for wild-type V 1a (20 g of carrier DNA and 0.075 g of specific DNA for 10 7 cells), mutant F225C (20 g of carrier DNA and 4 g of specific DNA for 10 7 cells), mutant F308C (20 g of carrier DNA and 2 g of specific DNA for 10 7 cells), or mutant K128C (20 g of carrier DNA and 2 g of specific DNA for 10 7 cells). The cells were cultured and, depending on the mutant, were treated overnight with 5 mM sodium butyrate before harvesting. The membranes were prepared as described under "Experimental Procedures." After calculating the protein content of the samples, the membranes were treated with vehicle (DMSO) or 10 Ϫ5 M of analogue 4 (isothiocyanate moiety) for 30 min in binding buffer (10 Ϫ6 M for the wild-type V 1a receptor). The membranes were washed, and the protein content was again measured. The different batches were used for radiolabeled binding studies with [ 125 I]HO-LVA (50 -800 pM for the mutants; 20 -400 pM for the wild-type V 1a ) or  In conclusion, analogue 4 (R ϭ NCS) is able to bind irreversibly and specifically to mutant F225C in a dose-dependent manner. Unfortunately, these results could not be reproduced with a high concentration (10 M) of analogue 5, which carries an ␣-chloroacetamide as a sulfydryl-reactive group, whereas analogue 4 carries an isothiocyanate moiety (data not shown).

DISCUSSION
Conventional loss-of-function mutagenesis has not always produced definitive answers for the identification of ligandbinding sites of G protein-coupled receptors. Indeed, site-directed mutagenesis experiments must be questioned regarding whether the observed effect is due to a disruption of a direct interaction of the mutated amino acid with the ligand or to an allosterically induced effect. In the present study, we have used an extension of the substituted cysteine accessibility method, which is based on the irreversible chemical modification of a receptor cysteine mutant with an affinity-selective sulfydrylreacting ligand. This method provides unambiguous results regarding ligand-receptor interactions after identification of the labeled amino acid residue. The application of this strategy to the definition of the nonpeptide antagonist-binding site of the human V 1a receptor allowed us to create an irreversible covalent bond between the isothiocyanate analogue of the SR49059 compound and the F225C mutant receptor. Because such a chemical bond formation is only possible if a close contact (less than 3 Å) exists between reactive elements created both on the ligand and the receptor, our results suggest that residue Phe 225 , which is located in transmembrane domain V of the receptor, probably directly interacts with the nonpeptide antagonist compound. The three-dimensional model of the SR49059 docked onto the human V 1a receptor proposed in this work was thus supported and validated by the results obtained through construction of site-directed cysteine mutants and subsequent irreversible binding with a sulfydryl-reactive analogue of the nonpeptide antagonist. A close structural analogue having an ␣-chloroacetamide instead of an isothiocyanate moiety as a sulfydryl-reactive group could not irreversibly label the F225C human V 1a receptor mutant, suggesting that an optimal orientation of the sulfydryl group of the cysteine residue and the electrophilic moiety of the reactive ligand is necessary to afford such a covalent bond. Moreover, the ␣-chloroacetamide group is much more bulky than an isothiocyanate moiety, a steric parameter that could be crucial as well.
Taking into account the result of affinity-directed irreversible binding, the model is in agreement with previous experimental results. First, we have already demonstrated that the SR49059 could be in contact with residues Thr 333 and Ala 334 located at the top of transmembrane domain VII (21). Based on a gain-of-function mutagenesis strategy, these two amino acids were shown to control the V 1a /V 2 receptor subtype binding selectivity of the nonpeptide antagonist. This result and the one described in this study make the three-dimensional model of the SR49059-V 1a receptor complex much more reliable. In addition, based on the three-dimensional model, Lys 128 (TM III), Gln 185 (TM IV), Trp 304 , Phe 307 , and Phe 308 (TM VI) could also participate in the binding site of the nonpeptide ligand, an hypothesis compatible with previous site-directed mutagenesis results. Indeed, mutation of these residues was detrimental to the antagonist binding (17)(18)(19)(21)(22)(23)(24). In conclusion, SR49059 seems to bind deeply in the transmembrane region of the human V 1a receptor, surrounded by a hydrophobic/aromatic cage. The involvement of Phe 225 confirmed the importance of aromatic residues in the binding of all classes of V 1a receptorspecific antagonists, both cyclic and linear peptides as well as nonpeptides (21-24). The binding sites for SR49059 and AVP only partially overlap. This overlap is constituted with polar residues such as Lys 128 and Gln 185 (21,24) and to a certain degree Gln 131 and Gln 311 (17). On the contrary, although AVP and agonists interact with extracellular residues of the receptor such as Tyr 115 in the extracellular loop I, which is responsible for receptor subtype selectivity (19), or residue Arg 46 in the N-terminal part (20), SR49059 does not. In addition, Phe 225 does not play any role in the binding of agonists (21), but the homologous residue in the OTR (Tyr 209 ) modulates the response of this receptor to the partial agonist AVP (45). AVP was demonstrated to become a full agonist on OTR when this aromatic residue was replaced by the residue present at the equivalent position in the V 1a receptor (F225). Moreover, transmembrane V, in which Phe 225 is located, seems to play an important role for other nonpeptide V 1a antagonists. It has been experimentally demonstrated that Ile 224 , adjacent to Phe 225 , controls the rat/human V 1a receptor selectivity toward the OPC21268 nonpeptide antagonist binding (42). Based on molecular modeling hypotheses, transmembrane domain V could also interact with the nonpeptide YM087, a potent V 1a /V 2 antagonist (43). Finally, it is now commonly accepted that Phe 225 , like other conserved hydrophobic/aromatic residues in the transmembrane domains of all GPCRs, participates into an intramolecular network of interactions that maintains the receptors as a constrained/inactive conformation (46). SR49059, while interacting with Phe 225 , could reinforce this network of intramolecular interactions, thus behaving as an antagonist.
Cysteine scanning has been widely used to allow the incorporation of chemical probes into membrane receptors (26). However, incorporating such a cysteine mutant to be labeled with a sulfydryl-reactive affinity ligand has been applied very few times, for example in the case of the ␥-aminobutyric acid, type A receptor with noncompetitive channel blockers (47) or in the case of the human ␣ 2A -adrenergic receptor (30). For this GPCR, chloroethylclonidine, an alkylating derivative of the ␣ 2A -adrenergic agonist clonidine, bound irreversibly to the receptor by forming a covalent bond with the sulfydryl side chain of a cysteine residue exposed in the binding cavity, leading to inactivation of the receptor. Using cysteine mutants as chemical sensors for interacting with sulfydryl-reactive affinity ligands is a powerful strategy to investigate ligand-receptor interactions. This approach is complementary to molecular modeling, site-directed mutagenesis and photoaffinity labeling. Like photoaffinity labeling, this approach results in unambiguous identification of residues directly involved in the binding of the ligands and in the precise definition of the binding pockets of GPCRs. Interestingly, this approach does not require the use of a radiolabeled ligand, the overproduction of the target protein, and its tedious sequencing to locate the labeled residue. However, this strategy necessitates a reliable threedimensional model of the hypothetical ligand-receptor interactions and the introduction of a mutation in the receptor, which could itself be detrimental to the overall structure of the protein.
AVP and OT receptors represent a logical target for drug development, and the strategy we described in the present paper would be of high interest for defining the binding sites of nonpeptide antagonists and agonists selective to the vasopressin V 2 and V 1b receptors or to the OTR. Such characterization should facilitate the rational design of potent and selective therapeutic molecules useful in many physiopathological conditions.