Identification of amino acid residues responsible for the α5  subunit binding selectivity of L-655,708, a benzodiazepine binding site ligand at the GABAA receptor


Address correspondence and reprint requests to Peter B. Wingrove, Neuroscience Research Centre, Merck Sharp and Dohme Research Laboratories, Terlings Park, Eastwick Road, Harlow, Essex CM20 2QR, UK. E-mail:


L-655,708 is a ligand for the benzodiazepine site of the γ-aminobutyric acid type A (GABAA) receptor that exhibits a 100-fold higher affinity for α5-containing receptors compared with α1-containing receptors. Molecular biology approaches have been used to determine which residues in the α5 subunit are responsible for this selectivity. Two amino acids have been identified, α5Thr208 and α5Ile215, each of which individually confer approximately 10-fold binding selectivity for the ligand and which together account for the 100-fold higher affinity of this ligand at α5-containing receptors. L-655,708 is a partial inverse agonist at the GABAA receptor which exhibited no functional selectivity between α1- and α5-containing receptors and showed no change in efficacy at receptors containing α1 subunits where amino acids at both of the sites had been altered to their α5 counterparts (α1ΔSer205-Thr,Val212-Ile). In addition to determining the binding selectivity of L-655,708, these amino acid residues also influence the binding affinities of a number of other benzodiazepine (BZ) site ligands. They are thus important elements of the BZ site of the GABAA receptor, and further delineate a region just N-terminal to the first transmembrane domain of the receptor α subunit that contributes to this binding site.

Abbreviations used











modified Barth's medium


transmembrane domain.

The mammalian γ-aminobutyric acid type A (GABAA) receptor is a pentameric structure containing different combinations of α1–6, β1–3, γ1–3, δ, ε and θ subunits (for reviews, see Barnard et al. 1998; Mehta and Ticku 1999). Among the many classes of drug which interact with the receptor are the benzodiazepines (BZs) that are used as anxiolytics, anticonvulsants and hypnotics. Immunoprecipitation with subunit specific antibodies, and recombinant receptor studies have demonstrated that high affinity BZ-binding sites are found on receptors of αβγ2 composition. The differential affinity of drugs such as diazepam for such receptors in which the α1 subunit has been substituted by α6, demonstrating that the α subunit is a major contributor to the BZ binding site present on the receptor (Lüddens et al. 1990). Site-directed mutagenesis studies have been shown in a number of cases (e.g. Pritchett and Seeburg 1991; Wieland et al. 1992; Amin et al. 1997; Buhr et al. 1997) to identify residues within the α subunit which are critical for the drug–receptor interaction. Although the diazepam insensitivity of receptors containing α4 and α6 is well documented (e.g. Lüddens et al. 1990; Wisden et al. 1991), few clinically efficacious drugs other than zolpidem (which is 10-fold selective for α1-containing receptors over α2- and α3-containing receptors, with negligible affinity for α4-, α5- and α6-containing receptors) discriminate between receptors containing other α subunits. Recently, however, the identification of a ligand (L-655,708) with selectivity for the α5-containing receptor has been reported (Quirk et al. 1996). In this study we have used chimeric α1/α5 subunits and site-directed mutagenesis of the α1 subunit to identify residues within α5 which are responsible for this selective affinity, and show that these residues are in fact important determinants for the binding of a number of BZ site ligands.

Materials and methods


[3H]Ro15–1788 (87.0 Ci/mmol) was obtained from NEN Life Science Products (Zaventum, Belgium). L-655,708 was synthesized in-house according to the previously published procedure of Watjen et al. (1990). CL218,872 was from Lederle (Belgium), and GABA, flunitrazepam, and chlordiazepoxide (CDZ) were obtained from Sigma-Aldrich (Poole, Dorset, UK). Methyl-6,7-dimethoxy-4-ethyl-β-carboline-3-carboxylate (DMCM), methyl β-carboline-3-carboxylate (β-CCM), Ro15–4513 and zolpidem were from Research Biochemicals International (Sigma-Aldrich). All oligonucleotides were custom synthesised by Life Technologies (Paisley, Scotland). Plasmid vectors pcDNA3, pcDNA1amp and pCDM8 were purchased from Invitrogen (Groningen, the Netherlands). Taq polymerase used in PCR reactions was obtained from Boehringer Mannheim (Roche Molecular Biochemicals, Sussex, England). GF/B filters used for radioligand filtration assays were from Brandel, Gathersburg, MD, USA. All other reagents were of the highest analytical grade available, and all solutions were made from double distilled deionized water. Benzodiazepine ligands were dissolved in 100% (v/v) dimethylsulphoxide (DMSO) which was applied at a final concentration of 0.1% (v/v) DMSO; DMSO had no discernible direct effect at this concentration.

Construction of site-directed mutants and chimeric α subunit cDNAs

Cloning of human cDNAs for α1, α5, β1, β3 and γ2S has been described previously (Wingrove et al. 1991; Hadingham et al. 1993a; Hadingham et al. 1993b). Oligonucleotide-directed mutagenesis was performed as previously described (Wingrove et al. 1994), using as template single-stranded α1 or α5 cDNAs in pcDNA3 with antisense oligonucleotides, or pcDNA1amp and sense-strand oligonucleotides. All mutations were verified by restriction enzyme digest and DNA sequencing of the 250–300 bp region surrounding the introduced change(s). The oligonucleotides used for mutagenesis, in addition to the desired amino acid changes, introduced restriction sites via silent mutations allowing easy identification of mutated clones.

The human α1 and α5 subunit nucleotide sequences do not contain any conserved restriction enzyme sites which would be useful in the construction of chimeric subunits. Consequently, a combination of PCR and site-directed mutagenesis was used to facilitate the building of chimeric subunits. Site-directed mutagenesis was used on the α1 subunit within pcDNA3 to introduce unique restriction sites via noncoding base changes. PCR was performed using Taq polymerase as described (Whiting et al. 1990) to amplify desired regions from the α5 subunit, with restriction sites introduced into the PCR product to enable final assembly of the α1/α5 chimeric subunit via restriction digest/ligation. PCR products were extracted using phenol/chloroform and ethanol-precipitated before digestion with the appropriate restriction enzyme. The desired product was purified by agarose gel electrophoresis, and ligated as required. The construction of the chimeric subunits described was as follows.

Chimera 1

An approximately 1360 bp α5 PCR product was generated using the oligonucleotide primers ChimHd (sense, pCDM8 vector bp 1375–1400) and ChAas (antisense, α5 bp771–795, introduces unique BsrGI site immediately after α5Leu131). This was digested with HindIII (site in polylinker of pCDM8) and BsrGI, and the 550 bp fragment cloned into similarly digested α1 cDNA in pcDNA3.

Chimera 2

An AflII site was introduced into the α1 subunit in pcDNA3 by oligonucleotide-directed mutagenesis using primer 219Afl (α1ΔAfl219). An α5 fragment generated by PCR with oligos ChimHd and ChBas (antisense, α5 bp 1045–1070, introducing AflII site), cut with HindIII/AflII was ligated into similarly digested α1ΔAfl219.

Chimera 3

Two AflII sites were introduced into α1 by mutagenesis using oligonucleotides 155Afl and 219Afl (α1ΔAfl155,Afl219). An α5 fragment generated by PCR using oligos ChCs (sense, α5 bp 853–877, introducing AflII site) and ChBas was digested with AflII and ligated into similarly cut α1ΔAfl155,Afl219.

Transfection and radioligand binding

The α1, α5, chimeric or mutant α subunits were expressed transiently with β1 and γ2S subunits in human embryonic kidney (HEK) 293 cells using previously described methods (Hadingham et al. 1996. Cell membranes were prepared from cells 48 h post-transfection, using methods described previously (Hadingham et al. 1993b).

All chimeric and mutant subunits were initially assayed (n = 1) for a gross change in affinity for L-655,708 by analysis of the displacement of a sub-Kd concentration of [3H]Ro15–1788 by 50 nmol/L and 0.5 nmol/L L-655,708 (corresponding to the approximate KI values at α1- and α5-containing receptors, respectively). Any chimera or mutant deviating considerably from the displacement seen with the corresponding wild-type subunit was subjected to full binding curve analysis. Saturation binding curves were obtained by incubating membranes with various concentrations of [3H]Ro15–1788, with nonspecific binding being measured in the presence of 10 µmol/L flunitrazepam. All binding assays were performed for 90 min at 4°C in 10 mmol/L potassium phosphate buffer pH 7.4 containing 100 mmol/L KCl (assay buffer). The total assay volume was 0.5 mL, containing approximately 200 µg of membrane protein. Incubations were terminated by filtration through GF/B filters on a Tomtec cell harvester, followed by three washes in ice-cold assay buffer. After drying, filter-retained radioactivity was determined by liquid scintillation counting. Dissociation constants (Kd) were calculated by Scatchard analysis, and KI values determined after experimental data points were fitted to single-site dose–response curves, each using GraFit software (Erithacus Software Ltd, Horley, Surrey, UK). Kd and KI values were determined from three or more independent experiments, with all points on binding curves derived from triplicate assays, and are expressed as mean ± standard error of the mean (SEM).


Xenopus oocytes were removed from anaesthetized Xenopus laevis and manually defolliculated with fine forceps. After mild collagenase treatment to remove follicle cells (0.5 mg/mL for 6 min), the oocyte nuclei were then directly injected with 10–20 nL injection buffer (88 mmol/L NaCl, 1 mmol/L KCl, 15 mmol/L HEPES, at pH 7.0, nitrocellulose-filtered) containing the human GABAA receptor α subunit cDNA of choice (6 ng/mL). These subunits were expressed with β3 and γ2S GABAA receptor subunits in a 1 : 1 : 1 concentration ratio. The oocytes were stored in an incubator until use; recordings were made 1–4 days after injection. For recording, oocytes were placed in 50 µL bath and perfused with modified Barth's medium (MBS) consisting of (in mmol/L): NaCl 88, KCl 1.0, CaCl2 0.91, MgSO4 0.82, HEPES 10, Ca(NO3)2 0.33, NaHCO3 2.4, pH 7.5 with 10 mol/L NaOH. Cells were impaled with two 1–3 MΩ glass microelectrodes containing KCl (2 mol/L) and voltage clamped at − 60 mV using a Geneclamp amplifier. The oocyte was continuously perfused with MBS at 4–6 mL/min and drugs were applied to the perfusate. For each oocyte the expression of receptors was confirmed by a 30-s application of 300 µmol/L GABA in MBS; GABA currents ranged typically from 0.3 to 3µA, and cells with currents < 0.3µA were rejected. Concentration–response curves (0.1–300 µmol/L) to GABA were calculated using a nonlinear squares fitting program to the equation

inline image

where y is the current at a given concentration of drug, Imax is the maximum current, EC50 is the concentration of drug eliciting a half maximal response, x is the drug concentration, and n is the Hill coefficient. Benzodiazepine ligands were tested for their efficacy on a concentration of GABA that gave a response that was 20% of maximum (EC20), predicted from the maximum current amplitude. At least 3 min wash time was allowed between each GABA application to prevent desensitization. Stable GABA EC20 values were obtained for each individual oocyte and BZ ligands, applied at a concentration approximately 100 times their binding affinity, were allowed to equilibrate for approximately 2 min prior to agonist application in the continuing presence of the BZ ligand. The percentage modulation of the GABA response by the BZ ligand was then calculated. All data are presented as the mean ± SEM and the paired Student's t-test was used to test for significance (p < 0.05 was considered to be significant).


The approximate affinities of L-655,708 for GABAA receptors containing various α subunit chimeras were obtained using a two-concentration displacement assay. As can be seen in Fig. 1, the single point mutations α1ΔSer205Thr and α1ΔVal212Ile each conferred an increased inhibition of [3H]Ro15–1788 binding by L-655,708, as did the double mutation α1ΔSer205Thr, Val212Ile.

Figure 1.

Diagrammatic representation of α1/α5 chimeras and mutants. All α1 sequences are represented by clear areas, and α5 sequences by shaded areas. Numbering is according to the mature α1 polypeptide, and for each chimera the first and last amino acid of the α5 sequence are numbered. Putative transmembrane domains 1–4 (TM1–4) are shown as black boxes. Each construct was expressed transiently, together with β1 and γ2S subunits, in HEK 293 cells. The inhibition of a sub-Kd concentration of [3H]Ro15–1788 by 50 nmol/L and 0.5 nmol/L L-655,708 is indicated (on a scale of – to + + + +) for each construct with respect to the inhibition obtained with α1- and α5-containing subtypes. Mutant 1, α1ΔIle202Asn, Val203Ile,Gln204Ser,Ser205Thr; mutant 2, α1ΔSer205Thr; mutant 3, α1ΔVal212Ile; mutant 4, α1ΔSer205Thr,Val212Ile.

The pharmacology of receptors containing each of the two α1 single point mutants, the α1 double mutant α1ΔSer205Thr,Val212Ile, and the equivalent α5 double mutant α5ΔThr208Ser,Ile215Val (where the corresponding positions in the α5 subunit were mutated to their α1 equivalents) were more fully characterized. The Kd values for the binding of [3H]Ro15–1788 to each mutant, as well as their KI values for a number of commonly studied BZ site ligands are presented in Table 1. All of the mutants were found to be expressed at approximately equal levels to the wild-type α1- and α5-containing receptors (between 0.2 and 0.5 pmol [3H]Ro15–1788 bound/mg membrane protein, data not shown). The critical roles of α5Ser205 and α5Val212 in determining the α5 selectivity of L-655,708 was confirmed. These two residues were also found to be responsible for the α5 selectivity of Ro15–4513. Like L-655,708, zolpidem had an affinity for receptors containing each of the individual mutants α1ΔSer205Thr and α1ΔVal212Ile which was intermediate between those shown for receptors containing wild-type α1 or α5 subunits. However, the affinity of zolpidem at receptors containing the double mutants (α1ΔSer205Thr,Val212Ile or α5ΔThr208Ser,Ile215Val) were also intermediate between the receptors containing wild-type α1 and α5 subunits, indicating that other residues are involved in the binding of zolpidem. Flunitrazepam, which does not discriminate between α1- and α5-containing receptors, had a lower affinity for receptors containing mutant α1ΔVal212Ile than for receptors containing wild-type α1 or α5 subunits. Interestingly, this decrease in affinity conferred by mutant α1ΔVal212Ile was less pronounced in receptors containing both Ser205Thr and Val212Ile mutations. The functional consequences of mutation of these residues was determined at GABAA receptor subunit combinations expressed in Xenopus oocytes. Although the β subunit used in these functional studies differed from that used in ligand binding experiments, it has been previously demonstrated that the β subunit does not significantly influence the affinity or efficacy of BZ site compounds at the GABAA receptor (Hadingham et al. 1993b). Full GABA concentration effect curves were constructed (0.1–300 µmol/L) for the different subunit combinations. No significant difference was observed in EC50, Hill slope or maximum evoked current between oocytes injected with α1β3γ2S, α5β3γ2S or α1ΔSer205Thr,Val212Ileβ3γ2S (Fig. 2). EC20 GABA responses were determined for each oocyte and BZ ligands coapplied to determine their relative efficacy at the three different subunit combinations (Fig. 3). The BZ site ligands studied included the agonists CDZ and zolpidem and the inverse agonists L-655,708 and DMCM. All four ligands modulated the GABA response at α1β3γ2S and α5β3γ2S receptors in a manner consistent with the pharmacology of wild-type α-subunit containing receptors. The efficacy displayed by CDZ and the inverse agonists, L-655,708 and DMCM at receptors containing α1ΔSer205Thr,Val212Ile was not different from the efficacy of these compounds at receptors containing wild-type α subunits. In contrast zolpidem (3 µmol/L), which is a highly efficacious agonist at receptors containing wild-type α1 (+133 ± 31%) and shows no efficacy for receptors containing wild-type α5 (− 1 ± 3%), was profoundly affected by mutagenesis of the α1 subunit (α1ΔSer205Thr,Val212Ileβ3γ2S, efficacy + 36 ± 8%). This reduction in efficacy appears to be due to the reduction in affinity (see Table 1), since at higher concentrations (30 µmol/L), zolpidem shows a efficacy at α1ΔSer205Thr,Val212Ile GABAA receptors (+139 ± 16%) similar to that at receptors containing wild-type α1.

Table 1.  Affinities of selected BZ binding site ligands for human α1- and α5-containing GABAA receptors and their mutants
Val212Ile β1γ2
Ile215Val β1γ2
  1. Wild-type or mutated GABAA receptors were transiently expressed in human embryonic kidney 293 cells. Kd values for the binding of [3H]Ro15–1788 were obtained by Scatchard isotherm analysis of radioligand binding. All other values are KI values obtained by the displacement of sub-Kd quantities of [3H]Ro15–1788 by the various ligands, as described in detail in Materials and methods. Values are mean ± standard error of at least three independent determinations. aKd values. bData from Quirk et al. 1996.

[3H]Ro15–1788a0.81 ± 0.070.55 ± 0.040.62 ± 0.053.1 ± 1.30.42 ± 0.053.2 ± 0.97
L-655,70849 ± 6b0.45 ± 0.15b9.5 ± 3.022 ± 4.71.0 ± 0.17160 ± 44
Ro15–45133.8 ± 0.640.31 ± 0.061.9 ± 0.359.3 ± 4.60.34 ± 0.045.3 ± 1.5
β-CCM0.76 ± 0.0423 ± 1.42.7 ± 1.55.3 ± 0.559.2 ± 2.911 ± 3.5
Zolpidem58 ± 18> 10,000950 ± 410790 ± 2203400 ± 3504300 ± 1500
CL218,872125 ± 7.8400 ± 1721 ± 5.4340 ± 9088 ± 221060 ± 165
Flunitrazepam4.6 ± 0.771.8 ± 0.083.8 ± 1.0160 ± 413.1 ± 1.724 ± 8.0
Figure 2.

Full concentration effect curves to GABA (0.1–300 µmol/L) on α1β3γ2S, α5β3γ2S and α1ΔSer205Thr,Val212Ileβ3γ2S GABAA receptors expressed in Xenopus oocytes. No significant difference was observed in EC50, Hill slope or maximum evoked current between oocytes injected with wild-type α1 (squares, EC50 of 7.1 ± 0.6 µmol/L, Hill slope of 1.15 ± 0.09, maximum of 1865 ± 247nA, n = 20), wild-type α5 (circles, EC50 of 7.0 ± 0.5 µmol/L, Hill slope of 1.01 ± 0.07, maximum of 1535 ± 144nA, n = 50) α1ΔSer205Thr,Val212Ile (triangles, EC50 of 7.1 ± 0.7 µmol/L, Hill slope of 1.19 ± 0.11, maximum of 1846 ± 393nA, n = 10) subunits.

Figure 3.

Efficacy of benzodiazepine ligands at α1β3γ2S, α5β3γ2S and α1ΔSer205Thr,Val212Ileβ3γ2S GABAA receptors expressed in Xenopus oocytes. Each column is the mean ± standard error of at least three oocytes. The efficacy displayed by chlordiazepoxide (CDZ), L-655,708 and DMCM at the α1ΔSer205Thr,Val212Ileβ3γ2S receptor was not different from the efficacy of these BZ site ligands for the wild-type receptors. The efficacy displayed by zolpidem at the α1ΔSer205Thr,Val212Ileβ3γ2S receptor was intermediate between the efficacy for the wild-type receptors. Since the affinity of zolpidem at the α1ΔSer205Thr,Val212Ileβ3γ2S GABAA receptor was reduced compared with the α1 wild-type, this compound was also tested at a concentration of 30 µmol/L, where the efficacy was similar to that found for the α1 wild-type (+139 ± 16%; data not shown).


This study investigated the amino acids responsible for the 100-fold selectivity in binding affinity of the BZ site inverse agonist L-655,708 for α5-over α1-containing GABAA receptors. A ‘gain of affinity/function’ approach was used which is more likely to identify residues directly involved in ligand binding by avoiding the complications of conformational modifications which can confound studies which have ‘loss of affinity/function’ as a readout. Two amino acids, α5Thr208 and α5Ile215, were found to be responsible for the higher affinity of L-655,708 for α5-containing receptors. As L-655,708 shows no functional selectivity between receptors containing α1 or α5 subunits, it was not surprising to find that its efficacy was unaffected by introduction of the double mutation (α1ΔSer205Thr,Val212Ile) into the α1 subunit.

Both of the amino acid changes identified that determine the α5 selectivity of L-655,708 are very conservative. The serine or threonine at position 205 (α1 equivalent) differ only in the presence of a methyl group on the threonine side chain. If one speculates that the primary interaction of serine or threonine with BZ site ligands is hydrogen bonding through the hydroxyl group, then the additional methyl group of threonine may either position the hydroxyl group so as to lead to a more favourable hydrogen bonding interaction, or directly provide some binding energy. The valine or isoleucine at position 212 (α1 equivalent) differs only in the length of the side chain and consequently the position of the two methyl groups. Since the larger amino acid confers the higher affinity for L-655,708 this apparently minor change may improve a hydrophobic interaction between them. The decrease in affinity for flunitrazepam may indicate a less favourable hydrophobic interaction or steric hindrance by the larger isoleucine residue. In support of this, Wieland and Lüddens (1994) reported that the binding of diazepam and Ro15–4513 is differentially influenced by mutation of α6Ile211 (homologous to α1Val212 and α5Ile215 suggesting a specific hydrophobic interaction of this domain of α6-containing GABAA receptors with BZ site ligands.

One of the residues identified in this study has also been implicated elsewhere to contribute to the binding of BZ site ligands. Renard et al. (1999) demonstrated that α1Ser205 is one of three residues responsible for the selectivity of zolpidem for α1- over α5-containing receptors (the other two residues identified being α1Gly201 and α1Thr163). It is interesting to note that this same residue is implicated in the binding of these chemically distinct BZ site ligands that have opposite α-selective binding profiles. Other investigators, using similar approaches, have also mapped BZ binding site determinants for other ligands to the same short region of the α subunit (Fig. 4). For example, Buhr et al. (1997) described the marked effect of mutation of two residues conserved in all α subunits, α1Thr206 and α1Tyr209 (corresponding to Thr207 and Tyr210 in the amino acid numbering system used here), on the binding of a number of BZ site ligands, whilst Pritchett and Seeburg (1991) found that the mutation α3Glu225Gly (the homologue to α1Gly201) increased the affinity for CL218,872 10-fold.

Figure 4.

Alignment of amino acids 200–215 (human GABAA receptor α1 subunit numbering) of the six human GABAA receptor α subunits and the inhibitory glycine receptor α1 subunit, showing residues implicated in ligand-receptor binding (BZ site ligands to GABAA receptors and strychnine to the glycine receptor). Residues found to play a key role in ligand–receptor interaction are boxed, with the publication indicated by italicized superscript numbering: 1, this study; 2, Renard et al. 1999; 3, Pritchett and Seeburg 1991; 4, Wieland and Lüddens 1994; 5, Buhr et al. 1997; 6, Vandenberg et al. 1992b.

The homologous region of the α subunits of the inhibitory glycine receptor forms a second Cys-loop of 10 amino acids, part of which has been suggested to be a β-strand due to the importance of alternating residues for ligand binding (Vandenberg et al. 1992a). Hence glycine receptor α1Tyr202 (homologous to GABAA receptor α1Ser205) is critical for binding the antagonist strychnine (Vandenberg et al. 1992b). The homologous domain of the α subunits of the GABAA receptor may also form a β-strand. Indeed, in support of this Schaerer et al. (1998) found no effect on modulation by BZs upon mutation of the intervening position 204 of the α1 subunit (human sequence equivalent numbering). Whether or not there is a β-sheet structure, it is clear that this domain, just N-terminal to transmembrane domain 1, is a critical structural feature of the BZ site of the GABAA receptor.

A pharmacophore model recently developed by Huang et al. (2000) allows speculation as to the nature of the interaction of some of the BZ site ligands, specifically the structurally similar imidazobenzodiazepines L-665,708, Ro15–1788 and Ro15–4513, with the GABAA receptor. The benzodiazepine C8 substituent (O-methyl in L-665, 708, azo-in Ro15–4513 and fluorine in Ro15–1788) is proposed to occupy a lipophilic region L2 that is deeper in α5-containing receptors. Indeed, the C8 position is the only difference between Ro15–4513 and Ro15–1788 and thus is presumably responsible for the 10-fold higher affinity of Ro15–4513 for α5-containing receptors that is determined by α5Thr208 and α5Ile215. It is important to also note that Ro15–1788 (but not Ro15–4513) has a reduced affinity for α4- and α6-containing receptors (Wieland et al. 1992; Davies et al. 1998) and this is determined by α1His101 (α5His105). Thus it may be speculated that α5His105, Thr208 and Ile215 may be closely associated in space so as to contribute to the tertiary structure of the lipophilic pocket L2 occupied by the C8 constituent.


The authors wish to thank Peter Blurton for his synthesis of the batch of L-655,708 used throughout this study.


  1. 1Current address: University of Oxford, Department of Pharmacology, Mansfield Road, Oxford OX1 3QT, UK.