Mutagenesis and Modeling of the GABAB Receptor Extracellular Domain Support a Venus Flytrap Mechanism for Ligand Binding*

The γ-aminobutyric acid type B (GABAB) receptor is distantly related to the metabotropic glutamate receptor-like family of G-protein-coupled receptors (family 3). Sequence comparison revealed that, like metabotropic glutamate receptors, the extracellular domain of the two GABAB receptor splice variants possesses an identical region homologous to the bacterial periplasmic leucine-binding protein (LBP), but lacks the cysteine-rich region common to all other family 3 receptors. A three-dimensional model of the LBP-like domain of the GABAB receptor was constructed based on the known structure of LBP. This model predicts that four of the five cysteine residues found in this GABAB receptor domain are important for its correct folding. This conclusion is supported by analysis of mutations of these Cys residues and a decrease in the thermostability of the binding site after dithiothreitol treatment. Additionally, Ser-246 was found to be critical for CGP64213 binding. Interestingly, this residue aligns with Ser-79 of LBP, which forms a hydrogen bond with the ligand. The mutation of Ser-269 was found to differently affect the affinity of various ligands, indicating that this residue is involved in the selectivity of recognition of GABAB receptor ligands. Finally, the mutation of two residues, Ser-247 and Gln-312, was found to increase the affinity for agonists and to decrease the affinity for antagonists. Such an effect of point mutations can be explained by the Venus flytrap model for receptor activation. This model proposes that the initial step in the activation of the receptor by agonist results from the closure of the two lobes of the binding domain.

A very large number of membrane receptors modulate the activity of intracellular effectors by activating heterotrimeric GTP-binding proteins (G-proteins). 1 All G-protein-coupled re-ceptors characterized so far possess seven putative transmembrane domains that define intracellular loops critical for the recognition and activation of G-proteins. Sequence comparison revealed the existence of at least three major families of Gprotein-coupled receptors. The members from different families share no sequence similarity. Receptors homologous to rhodopsin (receptors for catecholamines, acetylcholine, certain peptides, glycoproteins, etc.) constitute the first family (family 1), which is to date the best characterized one. Family 2 receptors are those homologous to the vasoactive intestinal peptide and the glucagon receptors. Family 3 receptors comprise the metabotropic glutamate receptors (mGluRs) (1,2), the Ca 2ϩsensing receptor (3), and a recently discovered new family of putative pheromone receptors (4 -6).
Family 3 G-protein-coupled receptors possess several unique features. They all have a surprisingly large extracellular N terminus that shares some sequence similarity with bacterial periplasmic amino acid-binding proteins (PBPs) such as the leucine-binding protein (LBP) and the leucine/isoleucine/valine-binding protein (LIVBP) (7). In agreement, this portion of the receptor has been shown to play a critical role in ligand recognition in mGluRs (7)(8)(9)(10). Recently, the production of the entire extracellular domain of mGluR1 in insect cells revealed that this domain produced as a protein is able to fold correctly, is soluble, and is sufficient to bind glutamate and its analogues in a very similar manner as the wild-type receptor does (11). However, further evidence that this domain shares a similar three-dimensional structure with PBPs is still missing due to the lack of good radioligands. The family 3 receptors also contain 20 conserved cysteine residues; 17 are located in the Nterminal extracellular domain, 9 of which are concentrated among the 100 residues that separate the LBP-like domain and the first transmembrane domain. The functional importance of this Cys-rich region has not yet been elucidated. However, this region has been shown to be necessary for the LBP-like domain of mGluR1 to bind glutamate (11). Other characteristic features of family 3 G-protein-coupled receptors are a highly conserved and short third intracellular loop critical for G-protein activation (12) and a variable second intracellular loop important for G-protein coupling selectivity (13,14).
The original structure of the binding domain of the family 3 receptors led to a hypothesis for their mechanism of activation. PBPs are known to be constituted of two lobes that close upon binding of the ligand, like a Venus flytrap when touched by an insect (15,16). It has therefore been proposed that the two lobes of the LBP-like domain of family 3 receptors also close upon binding of the agonist and that this change in conformation is transduced to the transmembrane region to activate the G-protein (7,17 The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. ¶ To whom correspondence should be addressed. Tel.: 33-467-14-2933; Fax: 3-467-54-2432; E-mail: pin@ccipe.montp.inserm.fr. 1 The abbreviations used are: G-proteins, GTP-binding proteins; mGluR, metabotropic glutamate receptor; PBPs, periplasmic amino acid-binding proteins; LBP, leucine-binding protein; LIVBP, leucine/ isoleucine/valine-binding protein; GABA, ␥-amino-n-butyric acid; GABA B , ␥-aminobutyric acid type B; GABA B -R1, GABA B receptor 1; DTT, dithiothreitol; APPA, 3-aminopropylphosphinic acid. glutamate receptor subunits are also homologous to PBPs (7, 18 -21), and the closure of this domain upon agonist binding has also been proposed to be responsible for receptor activation (22).
Recently, the cloning of the GABA B receptor revealed a protein that is distantly related to family 3 receptors. Indeed the GABA B receptor does not share all of the characteristic features of family 3 receptors described above (23). Like other family 3 receptors, the GABA B receptor possesses a large extracellular domain that shares some limited but significant similarity with PBPs such as LBP. However, the intracellular loops of the GABA B receptor are not as well conserved as in the other family 3 receptors. Moreover, the cysteines in the LBPlike domain, highly conserved in the other family 3 receptors, are not present in the GABA B receptor, and no Cys-rich region is found in this receptor. Because high affinity and specific radioligands are available for the GABA B receptor and because this receptor is related to the family 3 G-protein-coupled receptors, it appears to be a good model for a further analysis of the possible structure of the extracellular domain of this new receptor family. This would help identify the possible role of this domain in ligand recognition and receptor activation.
Here, we report the generation of a model for the GABA Bbinding domain and its assessment by site-directed mutagenesis. This study shows that the extracellular domain of the GABA B receptor likely folds into two lobes separated by a hinge region like the PBPs and identifies specific residues within this domain important for ligand recognition.

EXPERIMENTAL PROCEDURES
Materials-␥-Amino-n-butyric acid (GABA) was obtained from Sigma (L'Isle d'Abeau, France). 125 I-CGP64213 was synthesized from ethyl (1,1-diethoxyethyl)phosphinate as described (24) and labeled to a specific radioactivity of Ͼ2000 Ci/mmol (ANAWA AG, Wangen, Switzerland). All other ligands were synthesized in-house (24,25). Dithiothreitol (DTT) was from Roche Molecular Biochemicals (Mannheim, Germany). Serum, culture media, and other solutions used for cell culture were from Life Technologies, Inc. (Cergy Pontoise, France). The polyclonal anti-GABA B -R1 antibody was raised against the C-terminal intracellular part of the receptor as described previously (26). A secondary fluorescein-labeled anti-rabbit antibody (Sigma) was used for immunostaining. The ECL chemiluminescence system (Amersham Pharmacia Biotech ) was used to detect the secondary antibody on Western blots.
Sequence Alignment-A multiple alignment of the metabotropic glutamate receptor (rat mGluR1-8 plus the Drosophila receptor DmGluRA) and Ca 2ϩ -sensing receptor LBP-like domains according to O'Hara et al. (7) and of the bacterial periplasmic proteins that bind leucine/isoleucine/valine (LIVBP) and leucine (LBP) was generated using the ClustalW Version 1.60 program (27). A gap penalty of 30, an extension gap penalty of 0.05, and the Blosum30 matrix were used. The alignment was then manually modified in order not to introduce gaps into the sequence that aligned with the known secondary structure elements of LIVBP and LBP. On top of the multiple alignment, part of the extracellular domain of GABA B -R1a (residues 109 -558) was then aligned using the profile alignment command of the ClustalW Version 1.60 program and a gap penalty ranging from 2 to 50. These alignments indicated that the LBP-like domain of GABA B -R1a maps from residues 164 to 550. This later segment of the GABA B -R1a sequence was thereafter used to perform additional alignments. These were further analyzed and manually modified using the hydrophobic cluster analysis (28). The resulting alignments of the extracellular GABA B -binding site, LIVBP, and LBP were then used to generate three-dimensional models using the program Modeler (see below). The coordinates of these models were then subjected to the Verify3D algorithm (29) using the Verify3D Structure Evaluation Server 2 to identify regions of improper folding. After further manual modifications, the alignment giving rise to models with better Verify3D scores was selected.
Molecular Modeling-The three-dimensional model of the GABA Bbinding domain was constructed by homology using the coordinates of both LIVBP and LBP from Escherichia coli (Protein Data Bank accession numbers 2LIV and 2LBP, respectively) obtained by x-ray crystallography. The sequence alignment used was obtained as described above. Several models were generated using the program Modeler (30) in the Insight-II environment (Molecular Simulation Inc., San Diego, CA) on a Silicon Graphics R10000 O2 workstation. The hydrophobic cluster analysis suggested that part of the insertion found just after ␣-helix 9 (see Fig. 1) also likely folds into an ␣-helix. Accordingly, new models were generated in which parts of the insertion (residues 398 -406 and 412-418) were folded into an ␣-helix. Accordingly, the side chains of Cys-375 and Cys-409 were in close proximity, making them likely to form a disulfide bond. This was further imposed during the modeling procedure. These modifications improved the scores obtained in this area of the sequence with the Verify3D algorithm (29). The model giving the best scores was selected and subjected to energy minimization using the program Discover Version 2.9.7 (Molecular Simulation Inc.) and the CVFF force field. A steepest descent followed by a conjugate gradient method was applied (without taking into account the electrostatic term) until the maximum energy derivative was Site-directed Mutagenesis and Expression in HEK-293 Cells-Single amino acid replacement was carried out by the Quick Change method (Stratagene, La Jolla, CA) according to the manufacturer's instructions using pBSB5 as a template. This vector was constructed by subcloning the coding sequence of GABA B -R1a into the EcoRI and NcoI restriction sites of pBluescript SK (Stratagene). For each mutagenesis, two complementary 30-mer oligonucleotides (sense and antisense) were designed to contain the desired mutation in their middle. To allow a rapid screening of the mutated clones, the primers carried an additional silent mutation introducing (or removing) a restriction site. The presence of each mutation of interest and the absence of undesired ones were confirmed by DNA sequencing. Subsequently, a short fragment surrounding the mutation has been subcloned in place of the corresponding wild-type fragment of pRKBR1a (an expression vector constructed by inserting the open reading frame of GABA B -R1a downstream of the cytomegalovirus promoter of pRK5).
Wild-type and mutated expression constructs were transfected into HEK-293 cells by electroporation as described previously (31). Electroporation was carried out in a total volume of 300 l with 10 g of carrier DNA, 2 g of plasmid DNA containing the wild-type or mutated GABA B -R1a coding sequences, and 10 ϫ 10 6 cells. The cells were cultured in Dulbecco's modified Eagle's medium (Life Technologies, Inc.) supplemented with 10% fetal calf serum and antibiotics.
Ligand Binding Assay-Ligand competition experiments were performed on membranes of HEK-293 cells prepared as followed. 24 h after transfection, the cells were transferred into serum-free Dulbecco's modified Eagle's medium; 48 h after transfection, the cells were washed and homogenized in Tris/Krebs buffer (20 mM Tris-Cl (pH 7.4), 118 mM NaCl, 5.6 mM glucose, 1.2 mM KH 2 PO 4 , 1.2 mM MgSO 4 , 4.7 mM KCl, and 1.8 mM CaCl 2 ) and centrifuged for 20 min at 40,000 ϫ g. The pellet was resuspended in Tris/Krebs buffer and stored at Ϫ80°C. For ligand competition assays, thawed membranes (10 g of protein) were incubated with 0.1 nM 125 I-CGP64213 in the presence of unlabeled ligands at the indicated concentrations. Nonspecific binding was determined using 10 mM GABA or 100 nM CGP54626A. Incubation was terminated by filtration through Whatman GF/C glass-fiber filters. The curves were fitted with Kaleidagraph software using the following equation: y ϭ ((y max Ϫ y min )/(1 ϩ (x/IC 50 ) nH )) ϩ y min , where IC 50 is the cold drug concentration necessary to displace half of the specifically bound 125 I-CGP64213 and n H is the Hill number.
Western Blot and Immunofluorescence Experiments-For Western blotting, HEK-293 cell membranes prepared as described above were solubilized in Laemmli sample buffer (2% (w/v) SDS, 50 mM Tris-HCl (pH 6.8), 50 mM DTT, 0.1% (v/v) bromphenol blue, and 10% glycerol), and after 5 min of warming at 37°C, 1 g of protein was resolved by SDS-polyacrylamide gel electrophoresis (10% acrylamide) and transferred by electroblotting onto a Hybond C extra membrane (Amersham Life Sciences). The primary polyclonal antibody used for the immunodetection of the GABA B -R1a protein was described previously (26). Protein concentrations were measured with the Bradford method using the Bio-Rad protein assay reagent. The immunofluorescence experiments were performed 18 h after transfection on HEK-293 cells as described previously (32).

RESULTS AND DISCUSSION
Identification and Molecular Modeling of the LBP-like Domain of the GABA B Receptor-Sequence comparison between the extracellular domain of the GABA B receptor, the metabotropic glutamate receptors, and PBPs reveals that the sequence between residues 167 and 550 of GABA B -R1a shares sequence similarity with the LBP-like domain of mGluRs (18.9% with mGluR1 and 15.8% with the Ca 2ϩ -sensing receptor) and LBP and LIVBP (38.1 and 37.1%, respectively) (Fig. 1a). GABA B -R1a and GABA B -R1b are two splice variants that differ in their N-terminal sequence. N-terminal residues 1-163 of GABA B -R1a are replaced by 47 different residues in GABA B -R1b (Fig.  1b) (23). Interestingly, the similarity to the LBP and LIVBP proteins starts just after the splice junction site, suggesting that the LBP-like domain of GABA B receptors starts at the beginning of the first exon common to both GABA B -R1a and GABA B -R1b. In agreement with this proposal, the specific sequence of GABA B -R1a folds in a tandem pair of complement protein modules (also known as sushi repeats) (33)(34)(35), the first corresponding to Thr-25-Arg-97 and the second to Ile-84 -Asn-159. The second sushi-like domain of GABA B -R1a ends almost exactly before the splice site. This reveals that GABA B -R1a and GABA B -R1b share an identical LBP-like domain (Fig. 1b). Accordingly and in agreement with the hypothesis that the LBPlike domain constitutes the ligand recognition domain of family 3 receptors (7-11), the two GABA B receptor variants have identical pharmacological profiles (23). Sequence comparison reveals that the large insertions found in the LBP-like domain of mGluRs (I1, I2, and I3 in Fig. 1a) are not found in the GABA B receptor, further indicating that the GABA B receptor is most distantly related to the other family 3 receptors. Our sequence alignment also confirms that most of the extracellular domain of GABA B -R1b is homologous to LBP (Fig. 1b). A short se- A three-dimensional model was generated using the known three-dimensional structure of the open state of LIVBP and LBP (Fig. 2). This model was assessed using the Verify 3D algorithm according to Luthy et al. (29). As shown in Fig. 3, the three/one-dimensional scores of our model are always positive and are similar to those obtained with the template structures of LBP and LIVBP. The scores obtained are in the range of scores for highly refined correct X-ray structure determinations. This indicates that this domain of the GABA B receptor can fold like the LBP protein. As a further validation of the model, all Asn residues that are part of a consensus glycosylation site (Fig. 1a) were found to be at the surface of the protein (brown residues in Fig. 2b). This model reveals that the LBPlike domain of the GABA B receptor is composed of two lobes linked by three short linkers (Fig. 2a). Both lobes comprise an alternation of ␣-helices and ␤-sheets, one lobe having three additional helices. The latter will be referred to as lobe I, and the other as lobe II (Fig. 2).
Importance of Cysteine Residues for 125 I-CGP64213 Binding-Among the five cysteine residues found in the extracellular domain, the model predicts that Cys-219 and Cys-245 as well as Cys-375 and Cys-409 are in close proximity and likely form disulfide bonds (Fig. 2a) (the residue numbers correspond to those of the GABA B -R1a sequence, with the first Met being 1; this numbering will be used throughout this work). Cys-187 is buried inside lobe I and cannot be involved in a disulfide bond (Fig. 2a). Cys-219 and Cys-245 correspond to Cys-53 and Cys-78 of LIVBP and LBP, which are disulfide-bonded (36,37). This putative disulfide bridge found in the GABA B receptor is not found in the other family 3 receptors since only one of the Cys residues is conserved (see Fig. 1a for the mGluR1 sequence).
To examine if these Cys residues are important for the correct folding of the protein and/or for the binding of GABA B receptor ligands, they were changed into either Ala or Ser by in vitro mutagenesis. The mutation of Cys-187 did not prevent the binding of the GABA B receptor antagonist 125 I-CGP64213 (Ta- ble I), and displacement curves indicated that the affinities for the agonists GABA and APPA and for the antagonists CGP64213 and CGP54626A were not affected by these mutations (Table I and data not shown). Both mutants C187A and C187S were expressed and had the expected molecular mass, although C187S expression levels were lower than those of the wild-type receptor (Fig. 4). In contrast, mutation of any of the other four Cys residues into either Ala or Ser or the double mutation of Cys-219 and Cys-245 or of Cys-375 and Cys-409 totally prevented the binding of 125 I-CGP64213 to the receptor (Table I). In any case, the mutant proteins with a correct molecular mass could be detected on Western blots, although some (C245S and C219A/C245A) were expressed at very low levels (Fig. 4). These data indicate that the loss of binding is not due to a loss of expression or a rapid degradation of the protein.
These data are in agreement with our model predicting that these residues are involved in two disulfide bonds. We were therefore expecting that the binding of 125 I-CGP64213 would be sensitive to reducing conditions. However, as shown in Fig.  5, membrane treatment with 10 -100 mM DTT did not suppress 125 I-CGP64213 binding. This indicates that, if there are disulfide bonds, they are not required for proper binding after the protein has been synthesized and correctly folded. To explain the lack of binding observed with the Cys mutants, one may propose that these disulfide bonds are required for the correct folding of the protein during synthesis. After the protein is fully matured, these covalent links may only increase the stability of this domain of the receptor. We therefore studied the thermostability of the binding site in the presence or absence of 10 mM DTT (Fig. 5). Exposure of the wild-type receptor for a period of 30 min at temperatures ranging from 25 to 65°C revealed that 50% of the 125 I-CGP64213 binding disappeared at 50°C. This temperature was found to be 5.2 Ϯ 0.1°C (n ϭ 3) lower after DTT treatment (Fig. 5). These results are consistent with the hypothesis that disulfide bonds are involved in the stability of the binding domain of the GABA B receptor. However, we cannot exclude the possibility that the DTT effect results from its action on another protein stabilizing the GABA B receptor-binding site.
Selection and Mutagenesis of Additional Residues Possibly Involved in 125 I-CGP64213 Binding-The amino acid residues of LIVBP (open state) involved in the binding of leucine have been identified by x-ray crystallography and were found in lobe I (36). The side chain of Ser-79 and Thr-102 have been shown to form hydrogen bonds with the ␣-amino and ␣-carboxylic groups of leucine. Base on this information, it has been shown that the homologous residues in the mGluR1-binding domain, Ser-165 and Thr-188, are involved in the binding of glutamate. Mutation into alanine decreases the glutamate potency by 100-and 10,000-fold, respectively (7). In LIVBP, the side chain of leucine interacts in a hydrophobic pocket formed by Tyr-18, Ala-100, Ala-101, and Phe-276 (36). To try to identify the amino acid residues possibly involved in the binding of GABA B receptor ligands, a set of point mutations around residues homologous to those that constitute the leucine-binding pocket in LIVBP have been generated (Figs. 1a and Fig. 2). Some residues located on lobe II and facing the putative binding pocket were also mutated (residues corresponding to the ␤6 -␣6 loop; see Figs. 1a and 2b).   4. Expression of the Cys mutants of GABA B -R1a. Membranes (1 g of total protein) prepared from mock-transfected HEK-293 cells or cells expressing wild-type (Wt) GABA B -R1a or the indicated mutated receptors were subjected to SDS-polyacrylamide gel electrophoresis and blotted as described under "Experimental Procedures." GABA B -R1a proteins were detected using a polyclonal antibody directed against the carboxyl-terminal region of the protein. The specific binding of 125 I-CGP64213 was determined on the same membranes and is indicated at the top. 0, no specific binding was detected.
Among these 20 different mutations, only six were found to change the binding properties of the GABA B receptor (Fig. 2 and Table II). These include S246A, S247A, S269A, S270A, Q312A, and Y470A. All these mutated receptors were detected on Western blots as proteins with the correct molecular mass (Fig. 6) and had the same subcellular distribution as the wildtype receptor (Fig. 7), indicating that the changes in their binding properties were not due to a loss of correct expression or trafficking.
Mutations That Suppress Binding-Two of these mutations, S246A and Y470A, resulted in a total loss of the antagonist 125 I-CGP64213 binding, even when a 10-fold higher concentration of the radioligand (1 nM) was used (data not shown). Ser-246 aligns with Ser-79 of LIVBP and Ser-165 of mGluR1 ( Fig. 1), which have been shown and proposed to form a hydrogen bond with the ␣-carboxylic group of leucine and glutamate in LIVBP and mGluR1, respectively (7,36). It is therefore possible that Ser-246 is directly involved in the binding of agonists and antagonists. Tyr-470 is conserved in all family 3 receptors, LIVBP, and LBP (Fig. 1a). As the side chain of Tyr-470 is located inside lobe I (Fig. 2a), it is unlikely directly interacting with the ligand. Instead, the removal of the phenyl ring may simply affect the conformation of lobe I. Assuming that this lobe contains the binding site, this would affect ligand binding. In agreement with this proposal, mutation of Tyr-470 into phenylalanine restored binding and affinity for 125 I-CGP64213 and GABA, although binding was difficult to detect due to a very low expression of the mutant protein (data not shown).
Mutations That Differently Affect Binding Affinities of Various Ligands-The mutations S269A and S270A affected the affinities of some but not all ligands (residues shown in violet in Fig. 2b). In mutant S269A, the affinity of the antagonist CGP54626A was decreased by a factor of 50, whereas that of the antagonists CGP64213 and CGP56999A was affected by a factor of only 5 ( Fig. 8 and Table II). Similarly, the affinities of the three agonists GABA, APPA, and CGP47656 were decreased by 11-, 30-, and 10-fold, respectively, whereas the affinity of baclofen possessing an additional chlorophenyl group was not affected ( Fig. 8 and Table II). In mutant S270A, the affinities of the agonists GABA, APPA, and baclofen were decreased by a factor of Ͼ10, whereas those of the antagonists CGP54626A, CGP64213, and CGP56999A were not affected significantly ( Fig. 8 and Table II). Due to the relatively small changes in affinity observed, it is unlikely that these two residues are directly involved in an interaction with the ligands. Ser-269 aligns with Thr-102 of LIVBP and Thr-188 of mGluR1 (Fig. 1a). These residues have been shown and proposed to form a hydrogen bond with the ␣-amino group of leucine and glutamate in LIVBP and mGluR1, respectively (7,36). Accordingly, mutation of Thr-188 into Ala in mGluR1 resulted in a decrease in the quisqualate potency by a factor of 10,000. GABA lacks the typical amino acid moiety found on the ␣-carbon of both Binding experiments were performed using 0.1 nM 125 I-CGP64213 as described under "Experimental Procedures." Displacement curves performed with 11 different concentrations of GABA or CGP54626A were fitted as described under "Experimental Procedures." Concentrations giving rise to 50% inhibition of specific binding (IC 50 ) and Hill coefficients (n H ) were determined. Values are means Ϯ S.E. of at least three experiments performed in triplicates. wt, mutants in which the IC 50 values of GABA and CGP54626A were similar to those of the wild-type receptor; Ϫ, mutants in which the affinity of some ligands was decreased (see Fig. 8); ϩ, mutants in which the affinity of agonists was increased.  leucine and glutamate. It is therefore not surprising that Ser-269 of the GABA B receptor does not play a role similar to that of Thr-102 and Thr-188 of LIVBP and mGluR1, respectively.
Mutations That Increase Agonist Affinities-Finally, both mutations S247A (lobe I) and Q312A (lobe II) (blue residues in Fig. 2b) resulted in an increase in agonist affinity (GABA, APPA, and baclofen) by a factor of Ͼ3 and in a decrease in antagonist affinity (CGP54626A, CGP64213, and CGP56999A) by a factor of 2 ( Fig. 8 and Table II). The changes in affinity were too small to assume that Ser-247 and Gln-312 directly interact with these ligands. How can one explain the opposite effect of these two mutations on the binding affinities of agonists and antagonists? One possible explanation, as detailed below, is that these mutations affect the closure of the LBP-like domain of the GABA B receptor. As mentioned in the Introduction, PBPs adopt two conformational states, an open and a closed state. Binding of the ligand is known to favor the closed state according to Model I (38 -42), , and is therefore equal to K d ϭ K 1 (1 ϩ K 2 )/(1 ϩ ␣K 2 ). According to the similarity between the GABA B receptorbinding domain and PBPs, the same model can be proposed for the GABA B receptor. If we consider that the closure of the LBP-like domain is a necessary step for receptor activation, the factor ␣ must be higher than 1 for agonists (the agonist stabilizes the closed state). In contrast, ␣ may be lower or equal to 1 in the case of antagonists (they do not favor or even prevent the formation of the closed state). According to the above equation, an increase in K 2 due to a mutation in the receptor will result in a decrease in the K d for an agonist and an increase in K d for an antagonist, as observed for the S247A and Q312A mutations.
As shown in the three-dimensional model of the open state of the GABA B -binding site (Fig. 2, a and b), Ser-247 and Gln-312 are located in such a position that they can be in close proximity to the other lobe in the closed state (43). Accordingly, mutation of these residues may affect K 2 and therefore have opposite effects on the affinities of agonists and antagonists. In agreement with this proposal, amino acids at comparable positions in PBPs such as the histidine-binding protein have been shown to affect the equilibrium constant between the closed and open states (40,41,44). It is interesting to note here that Ser-247 aligns with Ser-73 in LBP and Ser-166 in mGluR1. This latter residue in mGluR1 is responsible for the calciuminduced activation of mGluR1 (38).
Conclusion-Taken together, our study shows that the structure of the GABA B receptor-binding site is likely to be similar to that of PBPs. Our three-dimensional model predicts that four Cys residues are important for the structural stability of this domain in the GABA B receptor and is confirmed by mutations of these residues and by the decrease in the thermostability of the binding site resulting from DTT treatment. In agreement with the three-dimensional model, 14 different mutations do not affect the binding properties of the receptor, and one residue (Ser-246) that is likely to be directly involved in the binding of GABA B receptor ligands has been identified. These data support a Venus flytrap model for the GABA B receptor activation, as previously proposed for the other family 3 receptors.