The β Subunit Determines the Ion Selectivity of the GABAA Receptor*

The γ-aminobutyric acid, type A (GABAA) receptor is a chloride-conducting receptor composed of α, β, and γ subunits assembled in a pentameric structure forming a central pore. Each subunit has a large extracellular agonist binding domain and four transmembrane domains (M1–M4), with the second transmembrane (M2) domain lining the pore. Mutation of five amino acids in the M1–M2 loop of the β3subunit to the corresponding amino acids of the α7nicotinic acetylcholine subunit rendered the GABAA receptor cation-selective upon co-expression with wild type α2 and γ2 subunits. Similar mutations in the α2 or γ2 subunits did not lead to such a change in ion selectivity. This suggests a unique role for the β3subunit in determining the ion selectivity of the GABAAreceptor. The pharmacology of the mutated GABAA receptor is similar to that of the wild type receptor, with respect to muscimol binding, Zn2+ and bicuculline sensitivity, flumazenil binding, and potentiation of GABA-evoked currents by diazepam. There was, however, an increase in GABA sensitivity (EC50 = 1.3 μm) compared with the wild type receptor (EC50 = 6.4 μm) and a loss of desensitization to GABA of the mutant receptor.

The ␥-aminobutyric acid receptor (GABA A R) 1 is a member of the superfamily of ligand-gated ion channels (LGICs), which also includes the nicotinic acetylcholine receptors (nAChR), the glycine receptors (GlyR), and a subtype of the 5-hydroxytryptamine receptors (5-HT 3 R) (1)(2)(3). The LGICs are involved in mediating fast neurotransmission in the central nervous system but play different roles, which is reflected in their different ion selectivities. GABA A Rs and GlyRs are anion-selective whereas the nAChRs and 5-HT 3 Rs are cation-selective (3)(4)(5)(6). All LGICs are integral membrane proteins, which are formed by assembly of five homologous subunits around a central ion channel (7)(8)(9)(10). Each subunit has a large extracellular N-terminal domain and a C-terminal domain containing four transmembrane segments, designated M1-M4, connected by relatively short loops. The extracellular N-terminal domains are believed to form the agonist binding sites, whereas the transmembrane domains form the channel; with the five M2 domains being the primary lining of the ion-conducting pore of the receptor (11,12).
The structure of the ion-conducting pore of the LGICs have been described thoroughly using the substituted-cysteine-accessibility method (12)(13)(14)(15). All the data support a hypothesis of a predominant ␣-helical structure of the M2 domains with a slight kink in the center. The pore was found to be widest at the extracellular end narrowing toward the cytoplasmic side (12)(13)(14)(15)(16). The selectivity filter and gate was found to be at the intracellular end of the M2 domains and to include a part of the M1-M2 loop (16,17). Differences between the individual LGICs were observed, but in general both anions and cations can enter the extracellular vestibules, charge selection occurs at a more intracellular position than 13Ј, and the gate is constituted by amino acids in positions Ϫ3Ј to 2Ј (see Fig. 1).
Besides the structural homology, the LGICs also show considerable sequence homology, particularly in the pore forming domains. This suggests a common quaternary structure of the channels, irrespective of the charge of the permeating ion, with charge selection being accommodated through minor alterations of the overall design of the channel.
The first example of conversion of the ion selectivity filter of a member of the LGICs was provided by Galzi et al. (18). By conducting a series of mutations of the homomeric nAChR ␣ 7 , it was demonstrated that mutating three residues was sufficient to alter the ion selectivity from cationic to anionic: A proline insertion at position Ϫ2Ј, a mutation of the adjacent glutamate (E-1ЈA) near the intracellular border of the M2 domain, and a mutation in the central part of the M2 domain (V13ЈT) (see Fig. 1). Moreover, the ion selectivity of the homomeric 5-HT 3A R and that of the homomeric GlyR ␣ 1 have been changed by the same or inverse set of mutations, respectively (19,20), thereby proving that charge selection is indeed a function of minor alterations in the M1-M2 loop and in the M2 domains of all LGICs.
This evidence has all been established using members of the LGICs that are capable of assembling into homomeric receptors. However, the majority of the LGICs are known to function as heteromeric complexes consisting of up to four different subunits. The predominant GABA A R subtypes in the central nervous system are believed to consist of two ␣ subunits, two ␤ subunits, and one ␥ subunit (8,9,21). When expressed in vitro, receptors composed of only ␣ and ␤ subunits are believed to consist of two ␣ and three ␤ subunits (9,22). Thus, even though the underlying principle and actual location of the ion selectivity filter have been unequivocally resolved, the question still remains as to how this knowledge extends to the heteromeric receptors. The present study provides the first example of conversion of the ion selectivity of a heteromeric receptor, namely the GABA A R ␣ 2 ␤ 3 ␥ 2 . As for other LGICs, the selectivity filter was shown to be located at the intracellular border of the M2 domain. However, the different subunits were shown to have different roles in defining the ion selectivity. A number of data sets ascribed a unique role of the ␤ 3 subunit in the determination of the ion selectivity. This raises important questions as to the function of the individual subunits in a pentameric receptor.
The ␣ 2 subunit was sub-cloned into the pNS1z vector, the ␤ 3 subunit was sub-cloned into the pNS1n vector, and the ␥ 2L subunit was subcloned into the pZeoSV vector (Invitrogen). The pNS1n and pNS1z vectors are customized vectors derived from pcDNA3 (Invitrogen) with expression under control of the cytomegalovirus promoter, whereas the pZeoSV vector uses the SV40 promoter.

Mutagenesis and Design of Chimeric Receptors
Using site-directed mutagenesis (23), a HindIII site was introduced in the loop between the M1 and M2 domains and a Bsu36I site was introduced in the loop between the M2 and M3 domains in all receptor subunits. HindIII and Bsu36I sites at other positions within the cDNAs were eliminated without affecting the amino acid sequence. Briefly, plasmids were uracilated by means of the Escherichia coli line RZ1032, and uracilated plasmids were used as the template for introducing mutations by means of mutagenic oligonucleotides and T7 DNA polymerase. An aliquot from the mutagenesis reaction was transformed into the E. coli line XL1-Blue, and mutated plasmids were identified by introduction or elimination of restriction sites.
In the second approach mutants were created using site-directed mutagenesis as described above. The oligonucleotides used were: Introduction of the Bsu36I site in the hGABA A R ␤ 3 subunit led to the I25ЈV mutation, which turned out to affect receptor function. This was reversed using the oligonucleotide: ␤ 3 V25ЈI, 5Ј-GAGACCTTGCCTAA-GATCCCCTATGTCAAAGCC-3Ј. All constructs were verified by restriction enzyme analysis and by DNA sequencing

Expression of Mutated Receptors
All constructs were expressed in CHO-K1 cells (ATCC no. CCL61). CHO-K1 cells were maintained in Dulbecco's modified Eagle's medium with 10 mM HEPES and 2 mM Glutamax supplemented with 10% fetal bovine serum and 2 mM L-proline (Invitrogen). The cells were cultured at 37°C in a humidified atmosphere of 5% CO 2 and 95% air and passaged twice a week. CHO-K1 cells were co-transfected with the plasmids described above and a plasmid encoding enhanced green fluorescent protein using the LipofectAMINE Plus kit (Invitrogen) according to manufacturer's protocol. Binding experiments and electrophysiological measurements were performed 24 -48 h after transfection.

Binding Assays
Membranes were prepared from CHO-K1 cells expressing recombinant GABA A R subunits. The cells were washed in phosphate-buffered saline (Invitrogen), trypsinized, washed twice in Tris-citrate buffer (50 mM, pH 7.1), and centrifuged for 10 min at 5000 ϫ g.
[ 3 H]Muscimol Binding-Membranes were resuspended in membrane wash buffer (20 mM KH 2 PO 4 /K 2 HPO 4 , pH 7.5, 50 mM KCl, 0.025% (w/v), NaN 3 , and various protease inhibitors (1 mM EDTA, 2 mM benzamidine chloride, 0.1 mM benzethonium chloride, 50 units/ml bacitracin, 0.3 mM phenylmethylsulfonyl fluoride, 10 mg/liter ovomucoid trypsin inhibitor, 10 mg/liter soybean trypsin inhibitor)) and centrifuged for 30 min at 177,000 ϫ g at 4°C. The pellet was resuspended in binding assay buffer (20 mM KH 2 PO 4 /K 2 HPO 4 , pH 7.5, and 100 mM KCl) to a protein concentration of 1 mg/ml and homogenized just before use. Binding was performed with 1, 3, 10, 30, 100, or 300 nM of [ 3 H]muscimol (20 Ci/mmol, PerkinElmer Life Sciences) in triplicate in a final volume of 250 l containing 200 g of protein, and nonspecific binding was determined in the presence of 1 mM GABA (Sigma). Samples were incubated at 4°C for 30 min, and labeled membranes were harvested on a Brandel cell harvester using GF/B filters (Whatman). The filters were washed with 3 ϫ 4-ml binding assay buffer, and the amount of radioactivity was determined by liquid scintillation counting.
[ 3 H]Flumazenil Binding-Membranes were resuspended in Tris-citrate buffer (50 mM, pH 7.1) and centrifuged for 10 min at 22,000 ϫ g at 4°C. The pellet was resuspended in Tris-citrate buffer to a protein concentration of 100 -200 g/ml. Binding was performed with 0.5, 1, 2, 3, or 6 nM [ 3 H]flumazenil (87 Ci/mmol, PerkinElmer Life Sciences) in triplicate in a final volume of 550 l containing 50 -100 g of protein, and nonspecific binding was determined in the presence of 1 M clonazepam (Roche Molecular Biochemicals). Samples were incubated at 4°C for 40 min, and labeled membranes were harvested using rapid filtration over GF/C filters (Whatman). The filters were washed with 2 ϫ 5-ml Tris-citrate buffer, and the amount of radioactivity was determined by liquid scintillation counting. [

Electrophysiology
All experiments were performed under voltage clamp using conventional whole-cell patch clamp methods (24) at 20 -22°C. The EPC-9 amplifier (HEKA Electronics) was controlled by a Macintosh G3 computer via an ITC-16 interface. The experimental conditions were defined by the Pulse-software accompanying the amplifier, and data were sampled at 1 kHz and low pass-filtered at 333 Hz.
Pipettes were pulled from borosilicate glass (Modulohm) using a horizontal electrode puller (Zeitz Instrumente). The pipette was filled with an intracellular solution containing 120 mM KCl, 2 mM MgCl 2 , 10 mM EGTA, and 10 mM HEPES adjusted to pH 7.2. The pipette electrode was a chloridized silver wire, and the reference electrode was a silver chloride pellet electrode (In Vivo Metric) fixed to the experimental chamber. The electrodes were zeroed with the open pipette in the bath just prior to sealing, and the pipette resistances were 1.6 -2.6 M⍀.
Coverslips with cells were transferred to a 15-l experimental chamber mounted on the stage of an inverted microscope (Olympus). Transfected cells were identified by the emission of green fluorescence when exposed to UV-light. After gigaseal formation, the whole-cell configuration was attained by suction.
Cells were continuously superfused at a rate of 2.5 ml/min with an extracellular solution (Na-R) containing 140 mM NaCl, 4 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , and 10 mM HEPES adjusted to pH 7.4. Chloride permeability was addressed using an extracellular solution (Gluconate-R) with a low chloride concentration containing 5 mM NaCl, 4 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 135 mM sodium gluconate, and 10 mM HEPES adjusted to pH 7.4. When Gluconate-R was used, the electrodes were zeroed in this solution. Cation permeability was addressed using an extracellular solution (NMDG-R) with a low sodium concentration containing 14 mM NaCl, 4 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 126 mM NMDG, and 10 mM HEPES adjusted to pH 7.4 with HCl. Liquid junction potentials were estimated using the Henderson liquid junction potential equation.
The cells were voltage-clamped at Ϫ60 mV and the holding current was monitored for 30 s at the start of each experiment to ensure a stable baseline. The I-V experiments were performed by holding the cells at potentials of Ϫ60, Ϫ50, Ϫ40, Ϫ30, Ϫ20, Ϫ10, 0, 10, 20, 30, 40, 50, and 60 mV and recording the currents activated by 3 or 30 M GABA at each membrane potential. GABA-containing solutions were delivered to the chamber through a custom-made gravity-driven U-tube, the tip of which was placed ϳ50 m from the cell. In general, GABA was applied for 0.5-1 s every 30 -40 s. Currents were measured at the peak of the response, and I-V plots were fitted to a polynomial function using GraphPad Prism software. Reversal potentials (V rev ) were estimated from I-V plots covering two to three membrane potentials on each side of the reversal potentials for each cell. I-V curves and reversal potentials were corrected for the estimated liquid junction potentials.
When evaluating the effect of modulators, cells were preincubated with the modulator for at least 10 s before GABA was applied. The effects of the modulators Zn 2ϩ (ZnCl 2 , Sigma), bicuculline (Sigma), and diazepam (Roche Molecular Biochemicals) were quantified against the GABA response evoked prior to addition of the modulator. Results are presented as mean Ϯ S.E.M., and comparisons were made using a two-tailed t test.

RESULTS
Wild type and mutated GABA A Rs were characterized with respect to ligand binding properties and by electrophysiological experiments. I-V curves were recorded in different extracellular solutions and the reversal potentials (V rev ) were estimated from the curves. Anion permeability was assessed by measuring the change in reversal potential when switching from a high (Na-R) to a low (Gluconate-R) extracellular chloride concentration, and cation permeability was assessed by measuring the change in reversal potential when switching from a high (Na-R) to a low (NMDG-R) extracellular sodium concentration. Data from the electrophysiological measurements for the wild type (wt) GABA A R values ␣ 2 ␤ 3 ␥ 2 and ␣ 2 ␤ 3 are presented in Table I. As expected for chloride-selective channels, the reversal potentials were sensitive to changes in the concentration of extracellular anions. Replacement of 90% of the extracellular chloride ions by the less permeable gluconate ion resulted in a shift in reversal potentials in the range of 39 -43 mV. The permeability of the gluconate ion has been shown to be ϳ10% of the chloride permeability in GABA A Rs (4); using this value, the expected shift in reversal potential is ϩ43 mV according to the Nernst equation. Reduction of the extracellular sodium concentration by replacement of 90% of the extracellular sodium by the less permeable NMDG ion did not change the reversal potentials significantly, indicating that wt GABA A Rs are cation-impermeable.
Binding of the GABA agonist [ 3 H]muscimol and of the GABA A pore blocker [ 35 S]TBPS were observed for both receptors (Table II). Binding of the benzodiazepine binding site antagonist [ 3 H]flumazenil was observed for the ␣ 2 ␤ 3 ␥ 2 receptor but not for the ␣ 2 ␤ 3 receptor, which is consistent with the location of the benzodiazepine binding site at the interface between the ␣ and ␥ subunits (25).   Chimeras between GABA A R Subunits-By aligning the M2 domains of several members of the LGIC family, we observed ( Fig. 1), that the GABA A R M2 domains are not identical for the subunits ␣ 2 , ␤ 3 , or ␥ 2 . To investigate whether a GABA A R consisting of subunits with identical M2 domains would be functional, a set of "internal" chimeric GABA A R subunits was designed. The chimeras were expressed in combination with either wt or chimeric GABA A R subunits.
As shown in Table III, substituting the M2 domains of the ␣ 2 or ␥ 2 subunits with that of either the ␣ 2 , ␤ 3 , or ␥ 2 subunits did not affect the receptors with respect to GABA-evoked wholecell currents and reversal potentials in either Na-R or Gluconate-R. In contrast, substituting the M2 domain of the ␤ 3 subunit with that of the ␣ 2 or ␥ 2 subunits invariably led to non-functional receptors. Interestingly, when the M2 domains of both the ␣ 2 and ␥ 2 subunits were substituted with that of the ␤ 3 subunit, a GABA-evoked current could be measured, but it was reduced by an order of magnitude with respect to amplitude. Interchanging the M2 domains did not affect binding affinity of the benzodiazepine antagonist [ 3 H]flumazenil for any of the chimeras (Table III). These data show that it is possible to make functional receptors with identical M2 domains.
Chimeras of GABA A R Subunits and the nAChR ␣ 7 Subunit-In a first attempt to change the ion selectivity of the heteromeric GABA A Rs, chimeric GABA A R/nAChR ␣ 7 subunits were generated. The design of chimeras containing the M2 domain of the nAChR subunit ␣ 7 (␣ 2 M2␣ 7 , ␤ 2 M2␣ 7 , and ␥ 2 M2␣ 7 ) is outlined in Fig. 1. The chimeras were expressed in combination with either wt or chimeric GABA A R subunits.
As seen in Table III, no GABA-evoked current (30 M GABA) could be detected when the chimeric subunits ␣ 2 M2␣ 7 or ␤ 3 M2␣ 7 were expressed. The only receptor responding to GABA was that of wt ␣ 2 and ␤ 3 subunits co-expressed with the chimeric ␥ 2 M2␣ 7 subunit. However, because co-expression of the subunits ␣ 2 and ␤ 3 can form functional receptors without the ␥ 2 subunit, it cannot be ruled out that this functionality can be attributed to wt ␣ 2 ␤ 3 receptors. Moreover, the shift in reversal potential from high to low extracellular chloride concentration for this complex was indistinguishable from that of the wt receptor (Table I), suggesting chloride selectivity. Because the chimeras containing the nAChR ␣ 7 M2 domain were all non-

FIG. 1. Sequence alignments of anion-or cation-conducting
LGICs. Amino acid sequences of the pore lining region (M2) of selected subunits from the anion-conducting receptors GlyR (␣ 1 ) and GABA A R (␣ 2 , ␤ 3 , and ␥ 2 ) and selected subunits from the cation conducting receptors nAChR (␣ 7 , ␤ 2 , and ␣ 4 ) and 5-HT 3 R (3A) numbered according to Miller (28). The intra-and extracellular borders of the M2 domain are outlined, and the amino acids that are shaded were mutated in previous studies (18 -20). The stars above the amino acid sequences mark the amino acids that are exposed to the channel lumen of the GABA A R ␣ 1 subunit (12), and the open arrow indicates that the selectivity filter is located at a more intracellular position than 6Ј. The stars below the amino acid sequence show the amino acids that are exposed to the channel lumen of the nAChR ␣ 1 subunit (14), and the filled arrow indicates that the selectivity filter is located at a more intracellular position than 13Ј. The region of the nAChR ␣ 7 subunit inserted into the GABA A R subunits (named ␣ 2 M2␣ 7 , ␤ 2 M2␣ 7 , and ␥ 2 M2␣ 7 ) is indicated by a box (dotted line). The amino acids used to design GABA A Rs with homologous M2 domains are also shown as a box (hatched line) as are the amino acids mutated at the intracellular border (solid line).
functional or chloride conducting, two other sets of mutated constructs were designed in which smaller portions of the GABA A R M2 domains were mutated to the corresponding nAChR ␣ 7 residues.
Chimeras of GABA A R Subunits and the nAChR ␣ 7 M1-M2 Loop-The first set of mutations was constructed on the basis of the work presented in Refs. 18 -20. The Ϫ2Ј proline (Ϫ2Ј alanine for the GABA A R subunit ␤ 3 ) was deleted, and the Ϫ1Ј alanine was mutated to a glutamate for the three GABA A R subunits as shown in Fig. 1 (designated ␣ 2 -E, ␤ 3 -E, and ␥ 2 -E). Because the T13ЈV mutation in the ␤ 1 subunit has previously been shown to abolish [ 3 H]muscimol binding and activation by GABA (26), this mutation was not included. Again, the chimeras were expressed in combination with either wt or chimeric GABA A R subunits.
As seen in Table IV, GABA-evoked whole-cell currents from GABA A Rs containing either the ␣ 2 -E or the ␤ 3 -E subunit were reduced by two orders of magnitude, relative to the wt GABA A Rs, whereas no whole-cell currents could be detected from receptors containing both of these mutated subunits. To facilitate detection of cation permeability, reversal potentials were measured in Na-R and NMDG-R rather than in Gluconate-R. Expression of the ␣ 2 -E subunit did not lead to any changes in the reversal potentials of the receptor. However, a small change in the reversal potential measured in NMDG-R of ϳ13 mV was observed for receptors containing the ␤ 3 -E subunit. Receptors containing the ␥ 2 -E subunit were indistinguishable from counterparts containing the wt ␥ 2 subunit with respect to both amplitude and reversal potentials. In the second set of mutations, a block of six amino acids at the M2 border (Ϫ5Ј-0Ј) of the GABA A R subunits was replaced by the five amino acids (DSGEK) of the corresponding region of the nAChR ␣ 7 as shown in Fig. 1 (designated ␣ 2 DSG-EK, ␤ 3 SG-EK, and ␥ 2 SG-EK, because the aspartate is conserved for the ␤ 3 and ␥ 2 subunits). Table IV, GABA-evoked whole-cell current amplitudes from GABA A Rs containing ␣ 2 DSG-EK or ␤ 3 SG-EK subunits were reduced by a factor of 10 -50, and, as for the first set of mutations, no whole-cell current could be measured when both the ␣ 2 and the ␤ 3 subunits were mutated. Expression of the ␣ 2 DSG-EK subunit did not change the reversal potentials; on the other hand, expression of the ␤ 3 SG-EK subunit changed the reversal potential by Ϫ31.3 Ϯ 1.3 mV for the ␣ 2 ␤ 3 SG-EK␥ 2 receptor and by Ϫ33.9 Ϯ 1.8 mV for the ␣ 2 ␤ 3 SG-EK␥ 2 SG-EK receptor. This indicates that GABA A Rs containing the ␤ 3 SG-EK subunit are cation-permeable. Again, it was not possible to detect any difference between receptors containing the ␥ 2 SG-EK or the wt ␥ 2 subunit, with respect to either current amplitude or reversal potentials. For reasons of simplicity, the ␥ 2 subunit was therefore omitted in a number of the following evaluations.

As seen in
Significance of Serine, Glycine, and Lysine in the SG-EK Motif-As seen in Table IV the ␤ 3 -E subunit gave rise to a small shift in reversal potential when expressed with wt ␣ 2 and ␥ 2 subunits, whereas the ␤ 3 SG-EK gave rise to a larger shift. To investigate in more detail the role of the amino acids Ϫ4Ј serine, Ϫ3Ј glycine, and 0Ј lysine in the determination of the ion selectivity, a set of ␤ 3 mutations were designed and co-expressed with the wt GABA A R ␣ 2 subunit. As seen in Table V, all the mutated ␤ 3 subunits resulted in receptors, which displayed a negative shift in reversal potentials (⌬V rev ) measured in NMDG-R compared with Na-R. It is also evident that the mutated receptors fall into three categories with respect to shifts in reversal potentials. The shift in reversal potentials for receptors containing ␤ 3 SG-EK is significantly different from that of receptors containing ␤ 3 G-EK (p ϭ 0.02), which again is significantly different from that of receptors containing ␤ 3 -E (p ϭ 0.04). However; whether the 0Ј position holds a lysine or arginine makes no difference, because the ␣ 2 ␤ 3 SG-EK and the ␣ 2 ␤ 3 SG-E receptors show identical shifts in reversal potentials.

TABLE V Characterization of the significance of the amino acids in the SG-EK-motif
Reversal potentials estimated from I-V curves (30 M GABA) recorded in Na-R and NMDG-R to identify changes in cation conductance.   (n ϭ 3). A positive shift in reversal potential is observed for the ␣ 2 ␤ 3 receptor when extracellular chloride is reduced from 150 to 15 mM, whereas this shift is not observed for the ␣ 2 ␤ 3 SG-EK receptor. Contrary, a negative shift in reversal potential is observed for the ␣ 2 ␤ 3 SG-EK receptor when extracellular sodium is reduced from 140 to 14 mM, whereas no shift is observed for the ␣ 2 ␤ 3 receptor.
This demonstrates that substitution of four amino acids (ASAA) with three (SG-E) in the ␤ 3 subunit resulted in conversion to cationic selectivity. No [ 35 S]TBPS binding was observed for any of the mutated receptor combinations, indicating substantial structural changes of the pore domain associated with just a few amino acid mutations (data not shown).
Characterization of the ␣ 2 ␤ 3 SG-EK Receptor-To further address the ion selectivity of the GABA A R ␣ 2 ␤ 3 SG-EK compared with the wt ␣ 2 ␤ 3 receptor, full I-V curves from Ϫ60 to 40 mV were recorded. The I-V curves for the wt ␣ 2 ␤ 3 receptor shown in Fig. 2 exhibits shift in reversal potentials of ϳ39 mV for Gluconate-R compared with Na-R and a minor shift of ϳ3 mV for NMDG-R compared with Na-R. This demonstrates anion but no significant cation permeability. Furthermore, outward rectification was observed for the wt ␣ 2 ␤ 3 receptor in all extracellular solutions, although it was less pronounced in NMDG-R.
In contrast, the I-V curves for the ␣ 2 ␤ 3 SG-EK receptor revealed a shift in reversal potential of ϳϪ29 mV for NMDG-R compared with Na-R, whereas a shift of ϳ5 mV was seen for Gluconate-R compared with Na-R. This demonstrates that the ␣ 2 ␤ 3 SG-EK receptor is a cation-selective channel almost insensitive to changes in extracellular chloride. Besides the change in ion selectivity, the outward rectification observed for the wt ␣ 2 ␤ 3 receptor was lost for the cation-selective ␣ 2 ␤ 3 SG-EK receptor.
As seen in Fig. 3, binding affinity of [ 3 H]muscimol was not significantly different for the ␣ 2 ␤ 3 SG-EK receptor (K d 58 Ϯ 8.0 nM, n ϭ 3) compared with the wt ␣ 2 ␤ 3 receptor (K d 45 Ϯ 0.7 nM, n ϭ 3) (p ϭ 0.23). Moreover, the B max value for the ␣ 2 ␤ 3 SG-EK (0.20 Ϯ 0.02 pmol/mg, n ϭ 3) compared with that of the ␣ 2 ␤ 3 receptor (0.44 Ϯ 0.03 pmol/mg, n ϭ 3) indicate uniform expression levels. As seen in Fig. 3, the GABA concentration-response curves for the two receptors were only slightly different, with an EC 50 of 1.3 M (n H of 0.9) for the ␣ 2 ␤ 3 SG-EK receptor and an EC 50 of 6.4 M (n H of 1.3) for the wt ␣ 2 ␤ 3 receptor. Zn 2ϩ (3 M) was found to block GABA responses evoked by 10 M GABA by 63 Ϯ 6% for the mutated and by 95 Ϯ 1% for the wt receptor. The GABA A R antagonist bicuculline (10 M) was found to inhibit responses evoked by 30 M GABA by 38 Ϯ 1% for the mutated ␣ 2 ␤ 3 SG-EK receptor and by 59 Ϯ 7% for the wt ␣ 2 ␤ 3 receptor; representative traces are shown in Fig. 4. These observations indicate similar pharmacology. Besides the obvious difference in current amplitudes, it is notable that desensitization is reduced for the mutated receptor. Although binding of the GABA A R channel blocker [ 35 S]TBPS could be observed for the wt ␣ 2 ␤ 3 receptor in the range of 0.15-0.25 pmol/mg of protein at 5 nM [ 35 S]TBPS, no binding could be demonstrated for the ␣ 2 ␤ 3 SG-EK receptor, indicating a fundamental functional change in pore structure of the ␣ 2 ␤ 3 SG-EK receptor.
Assembly of the ␥ 2 Subunits into the ␣ 2 ␤ 3 SG-EK Receptor-To investigate whether the ␥ 2 subunit integrates functionally in the ␣ 2 ␤ 3 SG-EK receptor, binding studies using the benzodiazepine antagonist [ 3 H]flumazenil were performed. As seen in Fig. 5A, the GABA A R ␣ 2 ␤ 3 SG-EK␥ 2 binds [ 3 H]flumazenil with a K d of 1.8 Ϯ 0.6 nM, which is not significantly different from the wt ␣ 2 ␤ 3 ␥ 2 receptor (2.4 Ϯ 0.7 nM). In contrast, no binding of [ 3 H]flumazenil could be detected for cells transfected with the GABA A R subunits ␣ 2 and ␥ 2 only (up to 6 nM) indicating assembly of the ␥ 2 subunit with the ␣ 2 and ␤ 3 SG-EK subunits. To investigate whether the modulatory effect of classic benzodiazepines was retained for a receptor containing the ␤ 3 SG-EK subunit, the ability of diazepam (1 M) to potentiate responses evoked by 1 or 3 M GABA was investigated, and representative traces are shown in Fig. 5B. As seen from the traces, diazepam was capable of potentiating the response evoked by 1 M GABA for both the wt ␣ 2 ␤ 3 ␥ 2 receptor and the ␣ 2 ␤ 3 SG-EK␥ 2 receptor, although the potentiation appeared reduced for the ␤ 3 SG-EK containing receptor. As seen from the bar graph in Fig. 5B, diazepam potentiation was more pronounced at 1 M GABA than at 3 M GABA, consistent with a benzodiazepine agonist induced leftward shift of the concentration-response curve of GABA. In line with this, the reduced potentiation seen for ␣ 2 ␤ 3 SG-EK␥ 2 receptor may in part be due to a slightly lower EC 50 value for GABA. Collectively, these experiments strongly suggest that the ␥ 2 subunit can be incorporated into the complex and that the ␣ 2 ␤ 3 SG-EK␥ 2 receptor has retained its ability to be modulated by benzodiazepines.

DISCUSSION
The majority of GABA A Rs are believed to be expressed as heteromeric complexes of two ␣, two ␤, and one ␥ subunit (8,9,21). An alignment of the subunit members ␣ 1-6 , ␤ 1-3 , and ␥ 2-3 shows that, with a few exceptions, the M2 domains as well as the important loops between M1 and M2 are identical within each subclass (data not shown). To simplify the work on locating/changing the ion selectivity filter, it was first investigated whether a GABA A R in which the M1-M2 loop and the M2 domain were identical in all subunits could be functional. The results were somewhat surprising with a number of chimeric receptors being indistinguishable from wt counterparts, some quite different, and others again non-functional. As a general feature it was not possible to distinguish receptors containing only a chimeric ␣ 2 or a chimeric ␥ 2 subunit from those with wt subunits. By contrast, a receptor giving reduced current amplitudes in response to GABA was obtained when chimeric ␣ 2 M2␤ 3 and ␥ 2 M2␤ 3 subunits were co-expressed with a wt ␤ 3 subunit. Non-functional receptors were obtained whenever the ␤ 3 subunit was the chimera (Table III). These data hint at a unique function of the ␤ 3 subunit in pore formation but also imply the importance of features absent in the ␤ 3 M1-M2 loop or M2 domain but found in both the ␣ 2 and ␥ 2 counterparts. The fact that it was possible to express a GABA A R in which all subunits had identical M2 regions, albeit with low current amplitudes, prompted us to make chimeric receptors with the nAChR ␣ 7 M1-M2 loop and M2 domain. No GABA-evoked currents could be recorded from receptors containing either the ␣ 2 M2␣ 7 or the ␤ 3 M2␣ 7 chimera. Receptors with only a chimeric ␥ 2 M2␣ 7 subunit were indistinguishable from the wt ␣ 2 ␤ 3 receptor, questioning the incorporation of the ␥ 2 M2␣ 7 subunit into the receptor (Table III).
Because the chimeras with the nAChR ␣ 7 M1-M2 loop and M2 domain resulted in non-functional receptors, it was decided to make chimeric receptors with smaller changes. The first attempt was to make the reverse set of mutations presented in Ref. 18 (P-2Ј⌬ (A-2Ј⌬ for the ␤ 3 subunit) and A-1ЈE) but without the T13ЈV mutation, because other work had indicated the importance of this threonine for wt GABA A R function (26). Receptors containing either the ␣ 2 -E or ␤ 3 -E mutation suffered a current amplitude loss of almost two orders of magnitude compared with the wt receptors. Moreover, no whole-cell currents could be recorded when the ␣ 2 -E and ␤ 3 -E subunits were co-expressed. No discernible effects of the ␥ 2 -E versus the ␥ 2 FIG. 5. Assembly of the ␣ 2 and ␤ 3 SG-EK subunits with the ␥ 2 subunits. A, saturation curve from a [ 3 H]flumazenil binding experiment for the ␣ 2 ␤ 3 SG-EK␥ 2 receptor with the K d value (n ϭ 3). B, the ability of 1 M diazepam to potentiate the whole-cell current evoked by 1 M GABA is presented as traces for the ␣ 2 ␤ 3 ␥ 2 and the ␣ 2 ␤ 3 SG-EK␥ 2 receptors. The bar graph shows the potentiation of the GABA response by 1 M diazepam at 1 and 3 M GABA for the GABA A Rs ␣ 2 ␤ 3 ␥ 2 (f) and ␣ 2 ␤ 3 SG-EK␥ 2 (Ⅺ). subunit were noticeable. Interestingly, a small shift in reversal potential, indicative of a change in ion selectivity, could be observed in receptors containing the ␤ 3 -E subunit but not in receptors with the ␣ 2 -E subunit.
When looking at an alignment of cation-selective receptors (Fig. 1) it is clear that the M1-M2 loop has a conserved motif in the region shown to influence selectivity: DSG-EK. In some receptors aspartate, serine, and lysine are replaced by glutamate, cysteine, and arginine, respectively, but these are the most conservative substitutions possible. It was therefore decided to make chimeras of GABA A R subunits with this motif in the second attempt. The results were actually similar to the results for the P-2Ј⌬ (A-2Ј⌬), A-1ЈE mutations with respect to functionality of the receptors and current amplitudes. Most importantly, however, ␣ 2 ␤ 3 SG-EK␥ 2 and ␣ 2 ␤ 3 SG-EK␥ 2 SG-EK receptors displayed large shifts in reversal potentials of approximately Ϫ30 mV when extracellular sodium was replaced with NMDG, demonstrating cation selectivity. Apart from a further loss in current amplitudes, no discernible effect of the ␥ 2 SG-EK versus the wt subunit was evident. Furthermore, the shifts in reversal potentials were independent of the ␥ 2 subunit presence.
Additional characterization showed that the ␣ 2 ␤ 3 SG-EK receptor was by far more sensitive to changes in the extracellular sodium concentration, relative to changes in the chloride concentration, which demonstrates predominant cation conductance. Evaluation of the contribution of the individual amino acids in the SG-EK motif to the change of ion selectivity showed that only the 0Ј lysine does not contribute significantly to cation selectivity. [ 3 H]Muscimol binding experiments revealed that the ␣ 2 ␤ 3 SG-EK receptor has the same K d and same B max as wt ␣ 2 ␤ 3 receptors. From GABA concentration-response curves, the ␣ 2 ␤ 3 SG-EK receptor is seen to be slightly more sensitive to GABA but with a similar Hill slope, and a pharmacological characterization showed that Zn 2ϩ and bicuculline still block the ␣ 2 ␤ 3 SG-EK receptor. Although no evident effects of the ␥ 2 subunit were observed when studying reversal potentials, the ␣ 2 ␤ 3 SG-EK␥ 2 receptor still binds [ 3 H]flumazenil, and very importantly GABA-evoked currents can be potentiated by diazepam. These data demonstrate that the ␥ 2 subunit is able to integrate in the receptor and form a functional benzodiazepine site.
Besides the fundamental change in ion selectivity the ␣ 2 ␤ 3 SG-EK channel shows distinct changes in gating behavior, similar to the "gain-of-function" characteristics of mutated nAChR ␣ 7 receptors (18,27). Thus, the ␣ 2 ␤ 3 SG-EK receptor displayed an increased GABA sensitivity and a loss of desensitization. This stands in contrast to the "loss-of-function" characteristics that were reported for the GlyR-STM mutation (19), where it was speculated that the T13ЈV mutation could be responsible for the loss-of-function characteristics. Interestingly, these features were attributed to the V13ЈT mutation (27), which fit well with the present data. As observed for the mutated nAChR ␣ 7 (27), a 10-fold loss of current amplitudes was observed for the ␣ 2 ␤ 3 SG-EK receptor, whereas binding data indicate that the total number of receptors is similar. Possible explanations for this could be either a decrease in transport to the surface of the cells, a decrease in unitary conductance, or altered gating efficiency.
Corringer et al. (27) proposed the term "exclusive" for the P-2Ј insertion (P236Ј), based on the findings that a proline in the selectivity filter is incompatible with a cation channel for the homomeric nAChR ␣ 7 and necessary for a conversion to an anion channel. The data presented here indicate that other features in the heteromeric GABA A R substitute for this "exclusivity" making the role of the proline less obvious. The wt ␤ 3 subunit has an alanine instead of a proline in position Ϫ2Ј for which reason a wt ␣ 2 ␤ 3 ␥ 2 receptor only holds a total of three proline residues. However, the data for the "internal" chimeras demonstrate that it is possible to make functional anion permeable heteromeric GABA A Rs lacking the proline in all subunits (␣ 2 M2␤ 3 ␤ 3 ␥ 2 M2␤ 3 ). Furthermore, the data for ␣ 2 ␤ 3 SG-EK␥ 2 receptors show that it is possible to make heteromeric cation-permeable GABA A R channels containing a total of three prolines in the ␣ 2 and ␥ 2 subunits.
Corringer et al. (27) showed that the exact nature of the amino acid side chains in the 234 -238 (Ϫ5Ј to 0Ј) loop in the anionic version of the nAChR ␣ 7 , besides the proline insertion, is less important. Interestingly, the present study demonstrates that the identity of the side chains in this same region is highly important for the mutated ␤ 3 subunit generating a cationic GABA A R. This seems to support the hypothesis that the M1-M2 loop has differential contribution in ion selectivity in anion and cation channels (27).
Because all previous reports on the determinants of ion selectivity of LGICs have been concerned exclusively with homomeric ion channels, it has not yet been possible to address the issue of the symmetry of the selectivity filter. If the selectivity filter were assumed to be symmetrical, all five subunits in a receptor would contribute equally to ion selectivity. It would then be expected that all subunits would have to be mutated for full conversion of charge selectivity, and if only some subunits were mutated, the phenotype of the resulting receptors in terms of ion selectivity would be independent of which subunits were mutated. It is evident from the present data that the results are essentially the opposite, i.e. mutations in all subunits abolish function and ␣ 2 DSG-EK␤ 3 conducts anions, whereas ␣ 2 ␤ 3 SG-EK conducts cations. One explanation for these findings could be that not all subunits contribute equally or at all to the ion selectivity filter; indeed, the present data strongly suggest that the ␤ 3 subunit has a dominant role in determining the ion selectivity of the heteromeric GABA A R. The present data demonstrate that integration of the ␥ 2 subunit into the ␣ 2 ␤ 3 SG-EK␥ 2 receptor has no effect on the shift in reversal potentials, which means that only two ␤ 3 SG-EK subunits are sufficient for conversion of the ion selectivity. It seems possible that, although all five subunits are involved in the overall design of the pore, only the two ␤ 3 subunits are responsible for ion selectivity. Future studies will show whether other features such as Ca 2ϩ permeability and unitary conductance are also determined by the ␤ 3 subunit exclusively or by all subunits in unison.