Unnatural Amino Acid Mutagenesis of the GABA_A Receptor Binding Site Residues Reveals a Novel Cation–π Interaction between GABA and β₂Tyr97

Abstract

The binding pockets of Cys-loop receptors are dominated by aromatic amino acids. In the GABA_A receptor α₁Phe65, β₂Tyr97, β₂Tyr157, and β₂Tyr205 are present at the β₂/α₁ interface and have been implicated in forming an important part of the GABA binding site. Here, we have probed interactions of these residues using subtle chemical changes: unnatural amino acid mutagenesis was used to introduce a range of Phe analogs, and mutant receptors expressed in oocytes were studied using voltage-clamp electrophysiology. Serial mutations at β₂97 revealed a ∼20-fold increase in EC₅₀ with the addition of each fluorine atom to a phenylalanine, indicating a cation–π interaction between GABA and this residue. This is the first example of a cation–π interaction in loop A of a Cys-loop receptor. Along with previous studies that identified cation–π interactions in loop B and loop C, the result emphasizes that the location of this interaction is not conserved in the Cys-loop family. The data further show that α₁65 (in loop D) is tolerant to subtle changes. Conversely, mutating either β₂Tyr157 (in loop B) or β₂Tyr205 (in loop C) to Phe substantially disrupts receptor function. Substitution of 4-F-Phe, however, at either position, or 4-MeO-Phe at β₂Tyr157, resulted in receptors with wild-type EC₅₀ values, suggesting a possible hydrogen bond. The molecular scale insights provided by these data allow the construction of a model for GABA docking to the agonist binding site of the GABA_A receptor.

Introduction

The GABA_A receptor is a member of the Cys-loop ligand-gated ion channel family, which includes nicotinic acetylcholine (nACh), glycine, and 5-HT₃ receptors. These proteins have a pentameric structure, with five subunits associating pseudosymmetrically around an ion-selective pore. Each subunit has an extracellular ligand-binding N-terminal domain containing the eponymous Cys-Cys loop, four transmembrane domains, and a large intracellular loop; the C terminus is extracellular.

Agonists bind at subunit interfaces, and of particular importance are a group of aromatic residues associated with six discontinuous loops, termed A to F (Akabas, 2004). These aromatics are clustered together in the crystal structure of the acetylcholine binding protein (AChBP), which is homologous to the extracellular domain of the GABA_A receptor (Brejc et al., 2001). The resulting “aromatic box” is a source of hydrophobicity that occludes water from the binding pocket and provides sites for interactions with ligands, including a cation—π interaction identified in a number of Cys-loop receptors (Table 1). This cation< interaction has been identified between the positive ammonium moiety on the neurotransmitter and an aromatic residue on loop B in the nACh receptor (Zhong et al., 1998), the 5-HT₃ receptor (Beene et al., 2002), and the GABA_C receptor (Lummis et al., 2005), but on loop C in the MOD-1 receptor (Mu et al., 2003). Because the primary amine of GABA has been shown to form a cation N|gp interaction in the GABA_C receptor, it is likely that there is a similar interaction with one of the aromatic residues in the GABA_A receptor binding site. In the α₁β₂ GABA_A receptor, candidate residues for a cation—π interaction include α₁Phe65, α₂Tyr97, α₂Tyr157, and α₂Tyr205, and, based on our knowledge of the other members of the Cys-loop family and the alignment of their aromatic residues (Table 1), both |gb₂Tyr157 (loop B) and α₂Tyr205 (loop C) are good candidates for such an interaction.

Table 1.

Cys-loop aromatic box residues^a

Receptor	Loop A	Loop B	Loop C		Loop D
nAChR	Y93	W149	Y190	Y198	W55
5-HT3	E129b	W183	F226	Y234	W90
MOD-1	C120	Y180	Y221	W226	F83
GABAC	F138	Y198	Y241	Y247	Y102
GABAA	Y97	Y157	F200	Y205	F65

^aResidues previously identified as contributing to cation N|gp bonds are typed in bold.

^bSee Sullivan et al. (2006).

Cation—π interactions can be identified by incorporation of fluorinated aromatics at prospective cation—π sites. The cation—π binding affinity of aromatics is strongly influenced by electrostatics, and addition of electron-withdrawing fluorines around the aromatic ring systematically diminishes the negative electrostatic potential on the face of the ring, and thus the cation—π binding ability (Mecozzi et al., 1996). Here, we use a series of Phe analogs that contain an increasing number of fluorine atoms on their phenyl rings to probe potential cation—π interactions at the aromatic residues that contribute to the α₁/β₂ GABA_A receptor binding site. These Phe analogs were incorporated into GABA_A receptors expressed in oocytes using nonsense suppression (Nowak et al., 1995), and GABA EC₅₀ values were measured using two electrode voltage-clamp electrophysiology. EC₅₀ values incorporate both binding and gating parameters and thus do not allow us to determine whether binding or gating (or both) are modified (Colquhoun, 1998). However, these residues have already been extensively investigated by other authors, who suggested they are involved in binding and not gating (Sigel et al., 1992; Amin and Weiss, 1993; Boileau et al., 2002). Changes in EC₅₀ are therefore likely to represent changes in binding parameters.

Materials and Methods

Mutagenesis and preparation of cRNA and oocytes.

Mutant GABA_A receptor subunits were developed using pcDNA3.1 (Invitrogen, Abingdon, UK) containing the complete coding sequence for either the human α₁ or human β₂ GABA_A receptor subunit kindly provided by Dr. K. A. Wafford (Merck, Sharp, and Dohme, Harlow, Essex, UK). The codons at positions α₁65, β₂97, β₂157, and β₂205 were replaced by a TAG codon as described previously (Beene et al., 2002). Mutagenesis reactions were performed using the method of Kunkel (1985) and confirmed by DNA sequencing. Harvested stage V–VI Xenopus oocytes were washed in four changes of OR2 buffer (in mm: 82.5 NaCl, 2 KCl, 1 MgCl₂, 5 HEPES, pH 7.5), defolliculated in 1 mg/ml collagenase for ∼1 h, washed again in four changes of OR2, and transferred to 70% Leibovitz media (Invitrogen) buffered with 10 mm HEPES, pH 7.5. The following day, they were injected with 5 ng of mRNA produced by in vitro transcription using the mMESSAGE mMACHINE kit (Ambion, Austin, TX) from DNA subcloned into pGEMHE (Liman et al., 1992) as described previously (Reeves et al., 2001; Beene et al. 2004). Electrophysiological measurements were performed 24–72 h after injection.

Synthesis of tRNA and dCA amino acids.

This was as described previously (Beene et al., 2004). Briefly, unnatural amino acids (Fig. 1) were chemically synthesized as NVOC (nitroveratryloxycarbonyl)-protected cyanomethyl esters and coupled to the dinucleotide dCA, which was then enzymatically ligated to 74-mer THG73 tRNA_CUA as detailed previously (Nowak et al., 1998). Immediately before coinjection with cRNA, aminoacyl tRNA was deprotected by photolysis (Kearney et al., 1996). Typically, 5 ng of total cRNA was injected (1 ng of wild-type α₁ or β₂ subunit and 4 ng of the corresponding α₁ or β₂ mutant subunit) with 25 ng of tRNA-aa in a total volume of 50 nl. For a control, cRNA was injected with THG 74-mer tRNA (no unnatural amino acid attached).

Figure 1.

Structures of the natural and unnatural amino acids used in this study. Carbon numbers for C3, C4, and C5 are highlighted on the phenylalanine structure.

The amino acids substituted are shown in Figure 1. Although probing tryptophan or phenylalanine residues via fluorination is straightforward, tyrosine introduces a complication. Fluorination of the aromatic ring of tyrosine substantially lowers the pK_a of the ring OH, such that highly fluorinated tyrosines are expected to be ionized at physiological pH. To avoid this complication, we first studied the Tyr-to-Phe mutant, and then introduced fluorinated phenylalanines. Previous studies have validated this strategy (Lummis et al., 2005).

Characterization of mutant receptors.

Peak GABA-induced currents were recorded at 22−25°C from individual oocytes using the Opus-Xpress system (Molecular Devices, Union City, CA). GABA (Sigma, St. Louis, MO) was stored as 100 mm aliquots at −80°C, diluted in ND96 buffer (in mm: 96 NaCl, 2 KCl, 1 MgCl_2, 1.8 CaCl₂, 5 HEPES, pH 7.5) and delivered to cells via the automated perfusion system of the OpusXpress. Glass microelectrodes were backfilled with 3 m KCl and had a resistance of ∼1 MΩ. The holding potential was −60 mV. To determine EC₅₀ values, GABA concentration–response data were fitted using PRISM software (GraphPad, San Diego, CA) to the four-parameter logistic equation, Y = I_min + (I_max − I_min)/1 + 10^{(logEC₅₀ − [A])nH}, where I_max is the maximal peak current and n_H is the Hill coefficient, and with the concentration values entered as the molar concentration in scientific notation. One-way ANOVA was performed with a Dunnett's post test to determine statistical significance using the same PRISM software.

Modeling the GABA _A receptor and docking of GABA.

The GABA_A receptor α₁ and β₂ subunit sequences were aligned to the AChBP subunit sequence using FUGUE (Shi et al., 2001), a program that matches predicted secondary structures as well as amino acid characteristics for improved accuracy of alignment (Forrest et al., 2006). The loop A portion of this FUGUE alignment can be seen in Figure 2, along with an alternative version in which a 2 aa space has been imposed manually to align the β₂97 residue with the nACh receptor αTyr93. Manipulating the A loop improved the alignment of both 5-HT₃ and GABA_A receptors (Cromer et al., 2002; Sullivan et al., 2006). A three-dimensional model of the extracellular region of the GABA_A α₁/β₂ receptor was then built using MODELLER (Sali and Blundell, 1993) based on the crystal structure of AChBP in the HEPES-bound state (Protein Data Bank ID 1i9b). GABA was docked into the GABA binding site using GOLD, version 2.2 (Jones et al., 1995). Twenty GOLDScore algorithm runs were performed to dock GABA with flexibility or rotation of the carboxyl tail permitted. The program was run with a population size of 100 and the number of generations set to 100,000. The active site cavity was defined as 10 Å radius from the α carbon of β₂Tyr97, a hydrogen bond was defined with a limit of 2.5 Å and a van der Waals interaction at 4.5 Å. The docking of GABA was performed in two ways: (1) with no constraints and (2) by orientating the primary amine of GABA toward C4 carbon of β₂Tyr97 by limiting the distance between the two atoms to 3.5 Å.

Figure 2.

Alignment of the A loops in the Cys-loop family. Residues underlined are Tyr89 of AChBP, αTyr93 of the nACh receptor, Glu129 of the 5-HT₃ receptor, and β₂Tyr97 of the GABA_A receptor. The GABA_A β₂ (alternative) sequence has 2 aa spaces added, aligning β₂Tyr97 with αTyr93 of the nACh receptor.

Results

Functional characterization of α₁β₂ GABA_A receptors

All mutant mRNAs, when coinjected into Xenopus oocytes with tRNA-Tyr (β₂97, β₂157, and β₂205) or tRNA-Phe (α₁65) molecules, produced functional receptors that responded to application of GABA with EC₅₀ values and Hill coefficients similar to each other, and to previously published values (Boileau et al., 2002). This indicates that the wild-type phenotype was successfully “rescued.” Typical responses are shown in Figure 3. No currents in response to application of high concentrations of GABA (100 mm) were detected from oocytes injected with mRNA alone or with mRNA and tRNA not ligated to dCA-aa (THG73 74-mer tRNA).

Figure 3.

GABA responses. Example responses from α₁β₂ wild-type receptors (A) and α₁β₂ Phe157 mutant receptors (B) at EC₅₀ and maximal [GABA].

Functional characterization of α₁Phe65 mutations

Incorporation of phenylalanine at position α₁65 (loop D) through nonsense suppression reproduced the wild-type GABA EC₅₀ (Table 2). The addition of a fluorine to the phenyl ring produced an increase in EC₅₀ from 3.0 to 11 μm, but three fluorines (3,4,5-F₃-Phe) did not further increase the EC₅₀ (11.5 μm). There were no significant differences in Hill coefficients across the group.

Table 2.

Functional characterization of α₁65 mutant GABA_A receptors

Mutant	LogEC50 ± SEM	EC50 (μm)	nH	n
α1β2	−5.457 ± 0.04	3.5	1.2	4
α165-Phe β2	−5.520 ± 0.09	3.0	1.2	4
α165-4-F-Phe β2	−4.962 ± 0.06a	11	1.4	3
α165-3,4,5-F3-Pheβ2	−4.941 ± 0.14a	11.5	1.5	3

^aDenotes logEC₅₀ values statistically significantly different from wild type by one-way ANOVA; p < 0.01.

Functional characterization of β₂Tyr97 mutations

Phe functions well at position β₂97, giving an EC₅₀ that is actually sixfold lower than wild type (Table 3). When referenced to the Phe variant, introduction of a single F resulted in a 33-fold increase in EC₅₀ to 20 μm for β₂97-4-F-Phe. An additional 22-fold increase in EC₅₀ to 420 μm (700-fold from Phe) was observed for β₂97-3,5-F₂-Phe and a final 24-fold increase for β₂97-3,4,5-F₃-Phe to EC₅₀ 9900 μm (Fig. 4). Hill coefficients were not significantly different from wild type.

Table 3.

Functional characterization of β₂97 mutant GABA_A receptors

Mutant	LogEC50 ± SEM	EC50 (μm)	EC50mut/EC50phe	nH	n
α1β2	−5.457 ± 0.04	3.5		1.2	4
α1β297-Phe	−6.190 ± 0.16a	0.6	1	1.4	4
α1β297-4-F-Phe	−4.706 ± 0.07a	20	33	1.2	5
α1β297-3,5-F2-Phe	−3.372 ± 0.05a	420	700	1.1	4
α1β297-3,4,5-F3-Phe	−2.002 ± 0.04a	9900	16,500	1.9	5

^aDenotes logEC₅₀ values statistically significantly different from wild type by one-way ANOVA; p < 0.01.

Figure 4.

Concentration–response curves for β₂97 mutant receptors. Normalized responses of individual oocytes expressing receptors incorporating fluorinated phenylalanine analogs at β₂Tyr97. The curves were generated using the four-parameter logistic equation as described in Materials and Methods. Data are mean ± SEM; n = 3–6 oocytes. WT, Wild type.

A cation–π interaction between GABA and β₂Tyr97

The systematic increase in EC₅₀ with the addition of fluorines to the phenyl ring of receptors modified at position β₂97 indicates a cation–π interaction. A “fluorination” plot (Zhong et al., 1998) of the cation–π binding energy [determined as described by Mecozzi et al. (1996)], and the EC₅₀ values compared with the parent molecule is shown in Figure 5; the direct relationship between these parameters shows there is a cation–π interaction here.

Figure 5.

Fluorination plots. A, Plot of the β₂Tyr97 mutant logEC₅₀ values from Table 3 normalized to the wild-type logEC₅₀ and fit by a line of best fit. B, Comparison of the cation–π interactions of the nACh, 5-HT₃, GABA_C, and GABA_A receptors. The x-axis is offset so that the parent, nonfluorinated amino acid (Trp for nACh and 5-HT₃ and Phe for GABA_C and GABA_A) is given a value of zero. The actual values are 32.6 kcal/mol for Trp and 26.9 kcal/mol for Phe. The values are fitted by a line of best fit to aid comparison. The gradient for the MOD-1 receptor is almost identical with that for 5-HT₃ and so is removed here for clarity.

In principle, the fluorination plot of Figure 5A could signal a cation–π interaction between β₂Tyr97 and some cationic moiety other than GABA. There are four positively charged amino acids within 10 Å of β₂Tyr97 that could perform this role, and without detailed structural information it is difficult to exclude any of them. However, two of these, β₂Lys102 and β₂Lys103, face away from the binding site, suggesting they are unlikely partners for β₂Tyr97, whereas interaction with α₁Arg132 would change the structure of the binding site such that GABA could not access the aromatic box. The remaining positively charged amino acid is β₂Arg207, which has previously been proposed to interact with the carboxylate of GABA (Wagner et al., 2004), and would therefore exclude it from contributing to a cation–π interaction at the other end of the molecule. Because all full GABA agonists have a charged amine, it is not possible to use an alternative agonist to provide definitive proof, but given the arguments described above, combined with the precedence of the ligand as the source of the cation in other Cys-loop receptors, we believe that GABA is the source of the cation.

Figure 5B presents fluorination plots for the current study and for some other members of the Cys-loop family. Small, focused ions provide the strongest electrostatic interactions, and this is illustrated by the fluorination plots; the nACh receptor has the weakest interaction as a result of the cation of ACh being a quarternary ammonium. 5-HT and GABA both have primary ammonium ions and therefore are expected to have stronger cation–π interactions. β₂Tyr97 has the steepest gradient indicating the strongest cation–π interaction found to date.

Functional characterization of β₂Tyr157 mutations

Removal of the hydroxyl group from β₂Tyr157 (producing Phe) caused a 400-fold increase in the GABA EC₅₀ from 3.5 to 1400 μm (Table 4). Introduction of a fluorine or methoxy group at position C4, however, resulted in near wild-type EC₅₀ values (3 and 5 μm, respectively), whereas a methyl group at this position resulted in an intermediate EC₅₀ (150 μm). Introducing 3,5-F₂-Phe increased the EC₅₀ 120-fold to 420 μm, but β₂157-3,4,5-F₃-Phe increased the EC₅₀ by only 30-fold. Hill coefficients were not significantly different from wild type.

Table 4.

Functional characterization of β₂157 mutant GABA_A receptors

Mutant	LogEC50 ± SEM	EC50(μm)	EC50mut/EC50wt	nH	n
α1β2	−5.457 ± 0.04	3.5		1.2	4
α1β2157-Phe	−2.846 ± 0.10a	1400	400	0.9	6
α1β2157-4-F-Phe	−5.514 ± 0.15	3.0	0.86	1.2	6
α1β2157-3,5-F2-Phe	−3.371 ± 0.04a	420	120	2.0	4
α1β2157-3,4,5-F3-Phe	−3.950 ± 0.08a	110	32	1.7	4
α1β2157-4-MeO-Phe	−5.294 ± 0.10	5	1.4	1.2	3
α1β2157-4-Me-Phe	−3.817 ± 0.15a	150	43	0.8	3

^aDenotes logEC₅₀ values statistically significantly different from wild type by one-way ANOVA; p < 0.01.

Functional characterization of β₂Tyr205 mutations

Substitutions at position β₂Tyr205 (Table 5) showed some similarity to those at β₂Tyr157. There was an increase of 24-fold in EC₅₀ when this tyrosine was replaced by phenylalanine, but a wild-type EC₅₀ when 4-F-Phe was substituted. Replacement with 3,5-F₂-Phe yielded a similar EC₅₀ to phenylalanine, as did introduction of β₂205-3,4,5-F₃-Phe. There were no significant differences in Hill coefficients.

Table 5.

Functional characterization of β₂205 mutant GABA_A receptors

Mutant	LogEC50 ± SEM	EC50(μm)	EC50mut/EC50wt	nH	n
α1β2	−5.457 ± 0.04	3.5	-	1.2	4
α1β2205-Phe	−4.082 ± 0.04a	80	24	1.4	5
α1β2205-4-F-Phe	−5.810 ± 0.08	1.5	0.43	1.9	3
α1β2205-3,5-F2-Phe	−4.174 ± 0.06a	70	20	1.9	4
α1β2205-3,4,5-F3-Phe	−4.039 ± 0.05a	90	26	2.5	4

^aDenotes logEC₅₀ values statistically significantly different from wild type by one-way ANOVA; p < 0.01.

A GABA_A receptor model

Images from the GABA_A receptor model developed as described in Materials and Methods are shown in Figure 6. The four aromatic residues studied here have been displayed; their faces are oriented together forming the aromatic box (Fig. 6A). The model was created using the alternative loop A alignment (Fig. 2), which exposed the face of β₂Tyr97 to the binding pocket. Figure 6B shows a possible hydrogen bond formed between the β₂Tyr157 hydroxyl and the backbone amide of α₁Thr130.

Figure 6.

Model of the GABA binding site aromatic box. A, Model showing the positions of the aromatic box residues found at the GABA binding site between β₂α₁. B, Hydrogen bond formation between β₂Tyr157 and α₁Thr130. C, A GABA molecule docked into the aromatic pocket. Those residues discussed are shown in bold.

In docking simulations, the small size of GABA allows it to adopt many energetically favorable orientations in the binding pocket. This apparently promiscuous behavior arises primarily because current docking programs are relatively crude; they do not recognize some interactions (e.g., cation–π interactions) and most do not allow complete flexibility of side chains. It is therefore essential that some constraints, based preferably on experimental data, are used to compensate for these inadequacies. Initial studies docking GABA with no constraints revealed four distinct GABA orientations: 10% of the results located the NH₄⁺ close to β₂Tyr97, 10% located the NH₄⁺ between β₂Tyr157 and β₂Tyr205, 10% located the NH₄⁺ between β₂Tyr157 and β₂Tyr205, with the COO⁻ over β₂Tyr97, and 70% located the NH₄⁺ toward the base of β₂Tyr157 and the COO⁻ toward loop C (see supplemental Fig. 1, available at www.jneurosci.org as supplemental material). Thus, when unrestricted, GABA predominantly faces β₂Tyr157 and/or β₂Tyr205, which is not supported by the data. We therefore allowed the molecule to find an energetically favorable orientation when its ammonium was close to β₂Tyr97, as implied by the fluorination data. These data (Fig. 6C) indicate that the ammonium head lies over the aromatic ring of β₂Tyr97, whereas the carboxyl tail is directed upwards between β₂Tyr157 and β₂Tyr205 toward α₁Arg67.

Discussion

The GABA_A receptor binding site has been intensively investigated using a range of techniques, including mutagenesis, radioligand binding assays, and photoaffinity labeling (Sigel et al., 1992; Amin and Weiss, 1993; Smith and Olsen, 1994; Boileau et al., 1999, 2002). These techniques have implicated many amino acids that may be important in the binding site, but they cannot identify the chemical-scale interactions of each amino acid with the neurotransmitter. Here, we used unnatural amino acid mutagenesis combined with functional studies to probe the effects of subtle chemical modifications to aromatic amino acids that form a critical part of this binding site. The results indicate that the aromatic at position β₂97 contributes a cation–π interaction, the aromatics at positions β₂157 and β₂205 require electronegative groups at their C4 positions to function efficiently, and the aromatic at position α₁65 is relatively insensitive to subtle chemical changes. The roles of these residues are discussed further below, and also considered in relation to a model of the GABA binding pocket.

β₂Tyr97 forms a cation–π interaction with GABA

Cysteine accessibility studies previously showed that β₂Tyr157 lies in the binding pocket: Y97C was protected from covalent modification by the presence of GABA or the GABA_A receptor antagonist gabazine (SR95531) (Boileau et al., 2002). These studies further suggested that this residue is involved specifically in GABA binding, because the Y97C mutation had the same relative effect (∼100-fold increase) on the GABA EC₅₀, which incorporates both binding and functional parameters, as on the SR95531 IC₅₀, which involves a purely binding phenomenon.

Our data show a strong correlation between the cation–π binding ability of tyrosine derivatives incorporated at β₂97 and logEC₅₀ (Fig. 5A). The size of this effect and the systematic dependence on the number of fluorines unambiguously establish a cation–π interaction at this site. Given the precedent from other Cys-loop receptors, we propose that, when GABA binds to the GABA_A receptor, the primary ammonium ionic moiety of the agonist interacts with the π-orbital of β₂Tyr97. This cation–π interaction is the strongest of these relationships found in the Cys-loop family so far (Fig. 5B). The differences in the 5-HT₃, GABA_C, and GABA_A receptor fluorination plot slopes may reflect the significance of this interaction within each of the binding sites.

Thus, the GABA_A receptor becomes the first instance of a Cys-loop receptor that displays a cation–π interaction with a loop A residue. Loop A does contain an aromatic box residue in the nACh receptor, αTyr93, and in AChBP crystal structures this Tyr aligns with Tyr89, which is clearly located at the “bottom” of the aromatic box. Presumably, β₂Tyr97 of the GABA_A receptor plays the role of αTyr93 and Tyr89, but the alignment is problematic; a gap equivalent to 2 aa is required to bring the β₂Tyr97 into alignment with the nACh receptor and AChBP Tyrs (Fig. 2). Loop A contains a highly conserved WxPDxxxxN motif, which plays a crucial role in positioning the rest of the binding site, particularly loop B (Lee and Sine, 2004; Cashin et al., 2005). It is therefore surprising that a gap insertion is necessary in a conserved region with such an important structural role, but the mutagenesis results are compelling.

The location of a cation–π interaction on loop A, contrasting to loops B or C for other members of the Cys-loop family, supports previous suggestions that the “lock-and-key” metaphor is not appropriate for such receptors (Mu et al., 2003). If they required a precise protein–ligand lock-and-key interaction, we would expect related receptors to use cation–π interactions at a conserved location in the three-dimensional structure. This is not the case in the 5-HT-gated 5-HT₃ and MOD-1 receptors, in which the cation–π interactions occur on loop B and loop C, respectively (Beene et al., 2002; Mu et al., 2003). We now find similar variability in the ionotropic GABA receptor family; the cation–π interaction moves from loop B in GABA_C (Lummis et al., 2005) to loop A in the GABA_A receptor (the present study). These are very similar receptors, showing 39% sequence identity and 64% homology between GABA_A(β₂) and GABA_C(ρ₁) in the extracellular domain. Nevertheless, GABA binds with different orientations in the two receptors. Apparently Cys-loop receptors require only that a ligand occupies the general binding region defined approximately by the aromatic box.

Contributions of α₁Phe65

There is no cation–π interaction at α₁65: addition of 4-F to the phenyl ring increased GABA EC₅₀ slightly, but addition of more fluorines around the ring resulted in no additional increase. This residue has been previously reported to play a role in GABA binding: mutation of α₁Phe64 (α₁Phe65 by human numbering) to Leu increased the GABA EC₅₀ from 6 to 1260 μm, with the IC₅₀ values of bicuculline and SR95531 increasing by similar amounts (Sigel et al., 1992). This suggests that the effects of mutations at this position are attributable to disruption of binding rather than gating, and that an aromatic is preferred here, although our data indicate that α₁Phe65 is tolerant to small chemical changes. Our model indicates that α₁Phe65 is located on the “right-hand” face of the aromatic box (see below), and does not form significant chemical interactions with any neighboring residues or the GABA molecule. In fact, it is partially obscured by β₂Tyr157and thus may solely contribute to the general hydrophobicity of the region.

β₂Tyr157 and β₂Tyr205

Previous data suggest that these two tyrosine residues specifically participate in GABA binding: phenylalanine mutations at both these sites significantly increased EC₅₀ values, whereas activation of the receptors using pentobarbital, which binds at a location distinct from the GABA binding site, resulted in no change in functional response (Amin and Weiss, 1993). Similarly, we observed that removal of the OH (introducing Phe) is highly deleterious at both β₂Tyr157 and β₂Tyr205, resulting in ∼400- or 24-fold increases in EC₅₀, respectively. Wild-type behavior can be rescued at β₂Tyr157 by incorporation of 4-F or 4-MeO substituents, whereas 4-Me-Phe or multiply fluorinated phenylalanines increased EC₅₀ values 35- to 120-fold. These data suggest that the OH of β₂Tyr157 acts as a hydrogen bond acceptor. The strong penalty for removing the OH and the near wild-type behavior of 4-MeO-Phe support this analysis. In this light, the near wild-type behavior for 4-F-Phe is perhaps surprising. Fluorine is the most electronegative element, and as such it is reluctant to donate a lone pair of electrons to a hydrogen bond donor. As a result, organic fluorine (fluorine bonded to a carbon) hardly ever accepts a hydrogen bond, especially if an alternative, better acceptor can be accessed (Dunitz, 2004). Inspection of the region around β₂Tyr157 in the receptor model we have developed (see below) suggests that there are no alternative hydrogen bond acceptors close to the hydroxyl group within the protein. In such a case, even the very poor acceptor of 4-F-Phe may be better than nothing at all (Phe). The model suggests β₂Tyr157 could form a hydrogen bond with the backbone NH of α₁Thr130 (Fig. 6B), and thereby support the architecture of this region of the binding pocket and help to expose β₂Tyr97 for the cation–π interaction. Support for this hypothesis comes from previous modeling studies that also propose that β₂Tyr157 forms a hydrogen bond with α₁Thr130 (Cromer et al., 2002). The data for β₂Tyr205 show no clear pattern, although there is an indication that an electronegative atom at C4 is favored.

A GABA_A receptor model

The identification of a cation–π interaction between the ammonium ionic moiety of GABA and the side chain of β₂Tyr97 provides an excellent starting point for locating GABA accurately in the binding site. Orientating the amine of GABA close to β₂Tyr97 fits well with the probable location of the carboxyl tail of GABA close to α₁Arg67 and β₂Arg207. Wagner et al. (2004) showed that β₂Arg207 affects GABA binding and unbinding without changing channel gating, and hypothesized that the arginine side chain directly contacts the carboxyl group of the GABA molecule in conjunction with α₁Arg66 (α₁Arg67 by human numbering) (Holden and Czajkowski, 2002). If the amine group of GABA is associated with β₂Tyr97, the carboxyl tail is located appropriately to form such a contact with β₂Arg207 and α₁Arg66. This association is supported when we dock GABA into the model of the GABA_A receptor: the carboxyl tail of GABA lies between α₁Arg67 and β₂Arg207 (Fig. 6C).

In conclusion, a cation–π interaction between GABA and a tyrosine on loop A has been identified by unnatural amino acid mutagenesis. This is the first example of a cation–π interaction with a loop A residue in a Cys-loop receptor, and along with previous studies that identified cation–π interactions with loop B and loop C residues, emphasizes the importance of this type of interaction in the binding of neurotransmitter. These results also further establish that the exact location of the cation–π interaction in the binding site is not critical, because it is not conserved among closely related members of the Cys-loop family.