Ligand-Gated Ion Channel Subunit Partnerships: GABA_AReceptor α₆ Subunit Gene Inactivation Inhibits δ Subunit Expression

Abstract

Cerebellar granule cells express six GABA_A receptor subunits abundantly (α₁, α₆, β₂, β₃, γ₂, and δ) and assemble various pentameric receptor subtypes with unknown subunit compositions; however, the rules guiding receptor subunit assembly are unclear. Here, removal of intact α₆ protein from cerebellar granule cells allowed perturbations in other subunit levels to be studied. Exon 8 of the mouse α₆ subunit gene was disrupted by homologous recombination. In α₆ −/− granule cells, the δ subunit was selectively degraded as seen by immunoprecipitation, immunocytochemistry, and immunoblot analysis with δ subunit-specific antibodies. The δ subunit mRNA was present at wild-type levels in the mutant granule cells, indicating a post-translational loss of the δ subunit. These results provide genetic evidence for a specific association between the α₆ and δ subunits. Because in α₆ −/− neurons the remaining α₁, β_2/3, and γ₂ subunits cannot rescue the δ subunit, certain potential subunit combinations may not be found in wild-type cells.

In vertebrate brains, GABA_A receptors are the principal mediators of inhibitory synaptic transmission. They are agonist-gated anion channels formed of pentameric assemblies of subunits arranged around an aqueous pore (Seeburg et al., 1990;Sieghart, 1995; Stephenson, 1995; Tyndale et al., 1995; McKernan and Whiting, 1996). The subunit genes (α_1–6, β_1–3, γ_1–3, and δ) are differentially transcribed, and the polypeptides are assembled in many possible combinations depending on cell type (Persohn et al., 1992; Wisden et al., 1992; Fritschy and Möhler, 1995). There are several important unresolved issues. Why is this receptor heterogeneity needed for synaptic function? What is the subunit composition of native subtypes of receptor? Are there rules guiding which subunits assemble with each other?

The cerebellum is an excellent brain area for investigating these questions. Its clearly defined circuitry allows an almost complete account of which cerebellar cell types express which GABA_Areceptor subunit genes (Wisden et al., 1996). For example, cerebellar granule cells express six subunit genes abundantly (α₁, α₆, β₂, β₃, γ₂, and δ), and so they probably have several distinct GABA_Areceptor subtypes of unknown subunit stoichiometry. As for the canonical muscle nicotinic acetylcholine receptor subunits (Verrall and Hall, 1992; Green and Claudio, 1993; Kreienkamp et al., 1995), there is likely to be selective discrimination between GABA_Asubunits; however, assembling a neuronal receptor requires solving an extensive combinatorial element. To form a native receptor, subunits have to recognize and distinguish their neighbors. The assembly pathways used by granule cells to sort the six principal subunits into different receptor subtypes are not known. Granule cells receive a single GABAergic input from Golgi interneurons onto their distal dendrites. At this synapse, the GABA_A receptor subtypes might be colocalized and intermingled (Nusser et al., 1995, 1996), and to date the α₁, α₆, β_2/3, and γ₂ subunits have been demonstrated to be present in the synaptic junction (Nusser et al., 1995, 1996; Somogyi et al., 1996).

Despite the comparative simplicity of the system, the receptor subunit composition of granule cell GABA_A receptors is controversial. Current views accommodate α₁β_2/3γ₂, α₆β_2/3γ₂, α₁α₆β_2/3γ₂, α₁α₆β_2/3γ₂δ, α₁β_2/3γ₂δ, and α₆β_2/3δ combinations (Korpi and Lüddens, 1993; Mertens et al., 1993; Caruncho and Costa, 1994;Khan et al., 1994, 1996; Quirk et al., 1994; Caruncho et al., 1995;Korpi et al., 1995; Pollard et al., 1995). The evidence for such combinations is derived from antibody-based data and correlation of pharmacological fingerprints of native binding sites, with binding profiles of subunits expressed in cell lines. Here we have targeted the α₆ subunit gene by homologous recombination techniques. Removal of α₆ protein from cerebellar granule cells allowed perturbations in other subunit levels to be studied and provided genetic evidence for a specific association between the α₆ and δ subunits. Our results begin to reveal the rules guiding receptor subunit assembly.

MATERIALS AND METHODS

Generation of mutant mice

The replacement vector for homologous recombination, designed for positive–negative selection (Mansour et al., 1988), was generated from a 6 kb mouse 129 strain α₆ subunit gene fragment (Jones et al., 1996), comprising part of exon 4 through to the middle of intron 8. This fragment was subcloned into pBluescript (Stratagene, La Jolla, CA) (see Fig. 1A). Into this plasmid, a BamHI cassette (TAG₃IRESlacZpAMC1neopA) (Nehls et al., 1996) was inserted between the AflII and NcoI sites located in exon 8, after the site ends were modified by adding BamHI adaptors. This cassette, designed to report target gene expression and to provide a dominant marker to select for insertion into the gene, consisted of stop codons (TAG) in all three frames, followed by an internal ribosome entry site (IRES) linked to a lacZ reading frame and SV40 polyadenylation sequences. It also contained a neomycin resistance gene under independent transcriptional control (Nehls et al., 1996). The IRES-lacZ cassette was inserted with the lacZ coding sequence in the same transcriptional orientation as the α₆ gene coding sequences. For negative selection, aXhoI–SalI fragment containing two MC1tk gene head-to-tail repeats (Smith et al., 1995) was placed in the targeting vector polylinker appending the longer homology arm (see Fig.1A). The vector was linearized with SalI and electroporated into 129 strain-derived embryonic stem (ES) cells (ES line “CCB,” kindly supplied by Drs. W. Colledge and M. Evans, Wellcome/CRC, University of Cambridge).

Fig. 1.

GABA_A receptor α₆subunit gene disruption by homologous recombination. A, Wild-type α₆ gene, targeting (replacement) vector, and the disrupted α₆ gene structures. Numbersindicate exons. On the Replacement vector, thebroken line indicates pBluescript (pBS) sequences. On the wild-type allele, theasterisk marks exon 8 where the IRES lacZ/neo cassette was inserted. Only relevant restriction sites are shown. A, AflII;B, BamHI; N,NcoI; S, SphI;X, XhoI. Arrows mark theneo and tk gene promoter sites and direction of transcription. The lacZ coding sequence orientation is the same as the α₆ gene, thus permitting its translation from the IRES sequence (striped box) to be initiated on the mRNA derived from the α₆ promoter. Expected restriction fragment lengths diagnostic for homologous recombination and the probes used to detect these are marked by double-headed arrows andhorizontal bars, respectively. B, Confirmation of α₆ mutant allele germline transmission. Biopsy tail DNA samples were digested with SphI, electrophoresed, and Southern-blotted. The membrane was probed withPROBE A (3′ flanking). Wild-type (+/+) individuals give a 15 kb band, the homozygous null (−/−) animals give a 9 kb band, and heterozygotes (+/−) give both bands.

Transfected ES cells were grown on G418^r primary embryonic fibroblast feeder cells, in medium supplemented with leukemia inhibitory factor (Life Technologies, Paisley, UK) and selected in G418 (Life Technologies) and FIAU (Bristol-Myers, Hounslow, UK) (Mansour et al., 1988; Smith et al., 1995). Genomic DNA was isolated from individual colonies, digested with SphI, Southern-blotted, and probed with an intron 8-derived SacI–XbaI restriction fragment (see PROBE A in Fig.1A). The addition of the neomycin gene creates an additional SphI site in the α₆ gene locus, thus enabling the discrimination between wild-type (15 kb) and null (9 kb) alleles (see Fig. 1A,B). Confirmation of correct targeting events was established with restriction fragment probes from the lacZ and neo genes (marked on Fig. 1A) (data not shown) and probe B (the complete sequence of probe B, which comprises the promoter and 5′ untranslated region, is deposited in the EMBL database, accession number X97475; Jones et al., 1996) (see Fig.1A). Additional diagnostic restriction digests of targeted genomic DNA used BamHI (see Fig.1A).

Male chimeras derived from targeted cells were mated with wild-type C57BL/6. Mutation germ-line transmission was determined by Southern blot analysis of agouti progeny tail DNA (see Fig.1B). Heterozygotes were intercrossed to generate a homozygous α₆ −/− line.

β-Galactosidase staining

Mice were transcardially perfused with 4% paraformaldehyde (PFA) in PBS. Brains were removed, post-fixed for 1 hr in 4% PFA, and then equilibrated at 4°C in PBS containing 30% sucrose. Sections (40 μm) were cut on a sliding microtome and incubated free-floating in 5-bromo-4-chloro-3-indolyl-β-galactoside (X-Gal) (Bonnerot and Nicolas, 1993). After X-Gal staining, some sections were counterstained with neutral red (Sigma, Poole, UK), allowing non-lacZ-expressing cells to be visualized. Alternatively, whole brains (see Fig. 3C) were immersed in the X-Gal staining reagent.

Fig. 3.

LacZ expression driven from the α₆gene locus in α₆ −/− adult mouse brains illustrated in horizontal (A), coronal (B), and whole-mount views (C); blue colorationindicates lacZ activity. A and B show the confined expression of the α₆ gene to the cerebellar granule cell layer; C shows the expression in the dorsal regions of the inferior colliculi; D shows higher-power view of α₆ gene expression in the molecular layer of the cerebellum. The arrow indicates an example of the numerous lacZ positive cells in the molecular layer. These are probably nonmigrated granule cells. The arrowheads mark putative parallel fiber staining; E, α₆ gene expression in the dorsal cochlear nucleus granule cells.BS, Brainstem; Cb, cerebellum;CbGr, cerebellar granule cell layer; Ctx, neocortex; Gr, cerebellar granule cells;DCGr, dorsal cochlear nucleus granule cells;H, hippocampus; IC, inferior colliculi;Mol, cerebellar molecular layer. Scale bars:A, 1.3 mm; B, C, 1 mm;D, 150 μm; E, 300 μm.

Staining of cultured granule cells. Cells on coverslips (see Granule Cell Culture and Electrophysiological Analysis) were washed in PBS and fixed in ice-cold 2% PFA/0.2% glutaraldehyde in PBS for 5 min. The coverslips were washed in PBS, incubated with X-Gal solution at 37°C overnight, and counterstained with neutral red.

Antibodies

α₁-specific antibodies.α₁(1–9), an N-terminal-specific antibody (Zezula and Sieghart, 1991); α₁-N, affinity-purified, raised against rat N terminus residues 1–14 (S. Pollard and F. A. Stephenson, unpublished data). α₁(328–382)/α₁L was prepared as described (Mossier et al., 1994). Rabbits were immunized with an MBP-α₁(328–382)-7His fusion protein, and the antibodies were purified with a GST-α₁(328–382)-7His fusion protein. This antibody precipitates GABA_A receptors and is selective for the α₁ subunit (R. Pelz and W. Sieghart, unpublished data).

α₆-specific antibodies. α₆-N (Batch R54XV), affinity-purified polyclonal, was raised to bovine α₆ subunit N-terminal residues 1–16 (Thompson et al., 1992); α₆(429–434) batch P24, affinity-purified rabbit polyclonal antibody, was raised to rat α₆ subunit residues 429–434 (Tögel et al., 1994); α₆-C, affinity-purified rabbit polyclonal, was directed against the C-terminus sequence CSKDTMEVSSTVE (S. Pollard and F. A. Stephenson, unpublished data).

δ-specific antibodies. δ(318–400), rabbit polyclonal was raised against the rat cytoplasmic loop sequence between TM3 and TM4 (Quirk et al., 1995); δ(1–44) (rabbit R7) polyclonal was prepared by immunizing with an MBP-δ(1–44)-7His fusion protein and purifying by affinity chromatography, as described (Mossier et al., 1994; R. Pelz and W. Sieghart, unpublished data). This antibody is specific for the δ subunit and does not precipitate α1β3γ2 receptors (R. Pelz and W. Sieghart, unpublished data).

Immunocytochemistry

Five α₆ −/− and five +/+ mice were transcardially perfused with 4% PFA, 0.05% glutaraldehyde, and ∼0.2% picric acid for 7–17 min. After perfusion the brains were washed in 0.1 m phosphate buffer. Preembedding immunocytochemistry was carried out on 70-μm-thick vibratome sections (Somogyi et al., 1989). Floating sections were incubated in 20% normal goat serum (NGS) diluted in Tris-buffered saline (TBS), pH 7.4, for 1 hr. The purified antibodies were diluted in TBS containing 1% NGS. After they were washed, the sections were incubated for 2 hr in biotinylated goat anti-rabbit IgG (diluted 1:50 in 1% NGS containing TBS), followed by incubation in avidin–biotinylated horseradish peroxidase complex (diluted 1:100; Vector Laboratories, Peterborough, UK) for 90 min. Peroxidase enzyme reaction was with 3,3′-diaminobenzidine tetrahydrochloride as chromogen and H₂O₂ as oxidant. In some cases, Triton X-100 (0.1–0.3%) was added to the TBS throughout the experiment. The antibody concentrations used for immunocytochemistry were δ(1–44)R7, 0.7–2.2 μg/ml; α₆-N, 1.5–3.0 μg/ml.

For controls, selective labeling could not be detected when the primary antibodies were either omitted or replaced by 5% normal rabbit serum. No immunoreactivity was obtained when the antibodies were preincubated with the appropriate peptides used for immunization (Nusser et al., 1996).

Ligand autoradiography

The procedures were slightly modified from Olsen et al. (1990)and Wong et al. (1996). Cryostat sections (14 μm) from frozen nonfixed adult mouse brains were preincubated in 50 mmTris-HCl, pH 7.4, and 120 mm NaCl for 15 min at 0°C, except for the GABA site assays when 0.31 m Tris-citrate solution, pH 7.1, was used. Incubations with ligands used fresh buffers of composition identical to those used for preincubation. For the benzodiazepine (BZ) site, [³H]Ro 15-4513 (5 nm, Du Pont de Nemours, NEN Division, Dreieich, Germany) was used with and without 100 μm diazepam (Orion, Espoo, Finland) for a 60 min incubation at 0°C, followed by three 30 sec washes, a dip in distilled water, and rapid drying. The same conditions and washes were used for the GABA site, with [³H]muscimol (20 nm, Amersham, Buckinghamshire, UK) and [³H]SR 95531 (20 nm, Du Pont), except that the incubation time was 30 min. The sections were washed three times for 15 sec in 10 mm Tris-HCl, pH 7.4, followed by dipping in distilled water and air drying. Sections were exposed to Hyperfilm-³H (Amersham) for 1–6 weeks. The images were produced by scanning the films. The nonspecific binding components to BZ and GABA sites were defined in the presence of 10 μmRo 15-1788 (Hoffmann-La Roche, Basel, Switzerland) and 100 μm GABA, respectively.

Radioligand binding

Radioligand binding on membranes prepared from individual mouse cerebella was as described previously (Quirk et al., 1994). Membranes prepared from each animal were used for saturation binding with [³H]Ro 15-1788 (0.1–17.0 nm), [³H]zolpidem (1–30 nm), and [³H]Ro 15-4513 (0.8–60.0 nm) displaced with Ro 15-4513 (10 μm) to define the total number of BZ binding sites, or with flunitrazepam (1 μm) to define binding to diazepam-sensitive sites only. Saturation binding with [³H]muscimol (2–45 nm) used 1 mmGABA to determine nonspecific levels. All assays used eight concentrations of ligand, with total and nonspecific binding measured in duplicate with 30–80 μg of protein/assay tube.B_max and K_d values were determined by nonlinear least-squares fit of the saturation curves using the data analysis software RS1 (Bolt, Beranek and Newman, Cambridge MA).

Immunoprecipitation analysis

Immunoprecipitation of GABA_A receptors solubilized from individual mouse cerebella used antibody δ(318–400) bound to protein A-Sepharose as described previously (Quirk et al., 1994, 1995). [³H]muscimol binding to the solubilized receptor was measured after gel filtration through Sephadex G-25 to remove any remaining endogenous GABA.

PAGE and immunoblotting

Membranes from individual +/+ and −/− cerebella were prepared, and equal amounts of protein per slot were subjected to SDS-PAGE in 10% polyacrylamide gels and immunoblotted. For the α₁(1–14), α₆(1–16), and α₆-C antibodies, the ECL Western blotting system (Amersham) was used for detection (Pollard et al., 1995). ECL blots were quantitated by normalizing with an anti-neuron specific enolase (NSE) antibody (Sigma) and then probing with α₁(1–14) (S. Pollard and F. A. Stephenson, unpublished data). Multiple exposures were taken for both anti-NSE and α₁(1–14) immunoreactivity and quantitated using a Molecular Dynamics Personal Quantitator. For the δ(1–44)R7 and α₁(328–382) antibodies, membranes were incubated with digoxygenin-labeled antibodies and were then treated with anti-digoxygenin-alkaline phosphatase Fab fragments (Boehringer Mannheim, Mannheim, Germany). Proteins were detected by fluorescence using the CSPD substrate (Tropix). Blots were quantitated by densitometry of Kodak X-Omat S films with the DocuGel 2000i gel documentation system using the RFLPscan software (MWG-biotech).

Granule cell culture and electrophysiological analysis

Cell culture. Cerebellar granule cells, attached to matrigel-coated coverslips, were cultured from postnatal day 5 (P5) mice as described for rat cells (Randall and Tsien, 1995). Minimal essential medium was supplemented with glucose (5 mg/ml), transferrin (100 μg/ml), insulin (5 mg/ml), glutamine (0.3 mg/ml), and 10% fetal calf serum. After 2 d, the cells were fed with media that was supplemented further with 4 μm cytosine arabinoside, and they were then fed every 5 d by a 50% replacement of the culture media. Electrophysiological measurements were made after 14–17 din vitro.

Electrophysiology

Recordings were from single, visually identified, granule cells using both outside-out patches and whole cells pulled away from the underlying cell-attachment substrate. No differences were observed between data from patches and whole cells, and results from both data sets were therefore pooled. A piezoelectrically driven theta tube-based application system delivered 120 msec pulses of GABA. Concentration jumps from control to agonist and vice versa took place within ∼1 msec. Five 120 msec 20 μm GABA pulses were applied at 0.1 Hz before and during the application of 1 μmflunitrazepam (Sigma). Recovery from the actions of flunitrazepam were studied with 20 additional GABA applications. Data were filtered at 2 kHz and sampled directly to computer at 10 kHz under control of the pClamp software suite. Because of the presence of some application-to-application variability in the current peak amplitude generated by GABA, an arbitrary threshold was set, with a 15% increase in the GABA response considered to be a potentiation above the baseline variability.

In situ hybridization

In situ hybridization with ³⁵S-labeled oligonucleotide probes was as described (Wisden and Morris, 1994). The oligonucleotide sequences used were α₄: 5′-TTCTGGACAGAAACCATCTTCGCCACATGCCATACTTTAAGCCTGT-3′ (EMBL accession number L08493) and δ: 5′-AGCAGCTGAGAGGGAGAAAAGGACGATGGCGTTCCTCACATCCAT-3′ (EMBL accession number M60596)

Behavioral observations

The animals (n = 23 for both +/+ and −/− lines, from two generations) were observed in their normal activities. Open field explorative activity was determined, under artificial lighting, in a round area (diameter 83 cm) divided into 19 segments. The mice were in the area for the first time. Their behavior was recorded for 5 min with a video recorder, and the behavioral parameters (number of segment crossings with all four feet and number of rearings) were scored blindly afterward. The number of fecal boli was counted before the area was cleaned for the next animal. The ability of the mice to learn to climb up onto a thin horizontal wire while initially hanging from their forepaws was tested in three trials during 1 d. Their ability to learn to stay on an accelerating rotating rod (Rotamex, Columbus Instruments, Columbus, OH) for 180 sec was tested during daily sessions. The initial session was 3 min on a nonmoving rod. On subsequent sessions, the mice were placed on the stationary rod, and the rotation speed was then set at 5 rpm and increased to 15 rpm over a 180 sec interval.

RESULTS

Creation of a mouse line with no GABA_A receptor α₆ subunit protein: mapping α₆ expression with a dicistronic RNA encoding lacZ

Homologous recombination in embryonic stem cells was used to create a 129/Sv × C57BL/6 mouse line in which the α₆subunit gene was disrupted at exon 8 (Fig.1A,B) (see Materials and Methods). The mutation, located just after the TM2 (channel lining) region, consisted of an insertion of stop codons in all three reading frames, an IRES linked to a β-galactosidase (lacZ) open reading frame and SV40 polyadenylation site, and finally a neomycin resistance gene expressed from its own promoter (Mountford et al., 1994; Nehls et al., 1996) (Fig. 1A). Translation of the α₆subunit mRNA from the mutant allele should terminate just after the TM2 region, resulting in a 300 amino acid protein designated α₆M2. The stop codon-IRES insertion generates a dicistronic mRNA in which β-galactosidase protein translation is then linked to an α₆ gene expression, i.e., lacZ expression is under α₆ transcriptional regulatory element control.

Homozygous mutant mice had no overt defects and could breed normally (see Behavioral Characterization of α₆ −/− Mice). By the criteria of Nissl staining, the size, folding of folia, and histological appearance of the cerebellum in α₆ −/− animals were completely normal. Immunocytochemistry with the α₆ N-terminal-specific antibody α₆-N (Thompson et al., 1992), however, demonstrated a total loss of α₆-specific immunoreactivity from the cerebellar granule cell layer of −/− mice (Fig. 2A,B). Using both N- and C-terminal α₆ subunit-specific antibodies (Thompson et al., 1992; S. Pollard and F. A. Stephenson, unpublished data), a complete loss of the 57 kDa immunoreactive α₆ band was also seen on Western blots of cerebellar protein extracts isolated from −/− animals (Fig. 2C). Identical results were found with an additional C-terminal anti-peptide antibody α₆(429–434) (R. Pelz and W. Sieghart, data not shown). Long exposure times of blots probed with the N-terminal antibody failed to show any α₆-specific degradation products in the −/− samples (data not shown), indicating that α₆M2 is not a stable entity. In a γ₂subunit gene knockout study, which similarly used an exon 8 disruption, a truncated γ₂ form was also not detectable (Günther et al., 1995).

Fig. 2.

Immunodetection of the α₆ subunit of the GABA_A receptor in α₆ +/+ (A, C) or α₆ −/− (B, C) cerebella as visualized with either light microscopic immunoperoxidase reactions (A, B) or immunoblotting (C). A, B, An intense immunoreactivity for the α₆subunits in the granule cell layer (gr) disappeared in α₆ −/− mice (B). The sections and one immunoblot were reacted with the same N-terminal domain-specific antibody. Scale bars: A, B, 500 μm.C, The 57 kDa α₆ protein is absent in α₆ −/− cerebella, as shown with either α₆-N, an N-terminal-specific antibody, or α₆-C, a C-terminal-specific antibody to the α₆ subunit.

We mapped α₆ gene expression in adult −/− animals using β-galactosidase staining (Fig. 3). An intense blue coloration was seen in the cerebellar granule cell layer (Fig.3A,B). The reaction product started to appear within the first 5 min of incubating the sections at room temperature in X-Gal and was fully developed within 30 min, thus demonstrating the extremely high level of α₆ locus expression. There were numerous small blue cells in the cerebellar molecular layer. These are probably ectopic granule cells (Fig. 3D) (cf. Thompson et al., 1992;Gao and Fritschy, 1995; Gutiérrez et al., 1996). Strong, blue coloration was also seen along the molecular layer outer edge, probably corresponding to β-galactosidase enzyme transported into granule cell axons, the parallel fibers. As expected, there were many blue granule cells in the dorsal cochlear nuclei (Fig. 3E) (Varecka et al., 1994). With little exception, the rest of the brain showed no detectable staining (Fig. 3A). Unexpectedly, however, many cells in the inferior colliculi dorsal regions stained blue (Fig.3C), and there were other minor cell populations in the substantia nigra and thalamus (geniculate nuclei) with faint but consistent blue staining (data not shown). These populations of stained cells were not seen in wild-type animals, and thus were not attributable to endogenous β-galactosidase-like activity.

α₆ subunit expression has not been noted previously in the inferior colliculi, substantia nigra, or thalamus by in situ hybridization or immunoreactivity, although a rat α₆ gene proximal promoter fragment consistently drives lacZ expression in the inferior colliculi of transgenic mice (Jones et al., 1996). The reasons for the lack of detection in previous studies could include low α₆ mRNA and protein levels. Alternatively, α₆ may be part of presynaptic receptors transported to distant axon terminals outside the inferior colliculi and so may escape detection in the inferior colliculi nucleus itself. The long half-life of β-galactosidase in mammalian tissue contributes to the extreme sensitivity of the lacZ reporter method. Over time, low levels of transcription from the α₆-lacZ hybrid gene will lead to accumulating amounts of β-galactosidase. These results make clear the usefulness of tracking gene expression using dicistronic-based reporters (Mountford et al., 1994; Nehls et al., 1996).

Pharmacological characterization of α₆ −/− cerebellar granule cells: BZ sensitivity

The α₆ protein absence was established further by pharmacological analysis. GABA_A receptors containing the α₆ subunit are insensitive to most types of BZs, such as diazepam or flunitrazepam (Lüddens et al., 1990; Hadingham et al., 1996). Ro 15-4513, however, is a BZ that binds to all subtypes of GABA_A receptor with α_xβγ₂subunit combinations, including those containing the α₆subunit (Lüddens et al., 1990; Sieghart, 1995). Thus, a diagnostic assay for α₆ in cerebellar granule cells is the high level of [³H]Ro 15-4513 binding on granule cell membranes that is insensitive to full BZ agonists such as diazepam (Sieghart et al., 1987; Malminiemi and Korpi, 1989; Lüddens et al., 1990; Turner et al., 1991). Normally, more than half of the [³H]Ro 15-4513 binding in the granule cell layer is diazepam insensitive (DIS) but can be displaced by micromolar concentrations of the BZ antagonist flumazenil (also known as Ro 15-1788). This profile is thought to be attributable to the abundant expression of α₆β_2/3γ₂receptors on granule cells (Lüddens et al., 1990; Korpi and Lüddens, 1993; Korpi et al., 1993; for review, see Wisden et al., 1996). In the α₆ −/− animals, DIS binding over the cerebellar granule cell layer is completely absent (Fig.4A), whereas in wild-type brains [³H]Ro 15-4513 still binds over the granule cell layer even in the presence of 100 μm diazepam (Fig.4A). From binding studies using isolated cerebellar membranes, the contribution that the α₆ subunit makes to the number of total cerebellar Ro 15-4513 binding sites was estimated to be ∼40% (see Evaluation of α₁ Subunit Levels; also see Table 1).

Fig. 4.

Autoradiographic analysis of GABA_Areceptor binding sites in wild-type (+/+) and α₆ −/− mice. A, Benzodiazepine sites labeled by 5 nm [³H]R0 15-4513 showing total and diazepam-insensitive binding in the presence of 100 μmdiazepam. B, GABA_A receptor sites labeled by 20 nm [³H]muscimol, showing total binding. The nonspecific signal in the presence of 100 μm GABA was at the film background level. C, GABA_Areceptor sites labeled by 20 nm [³H]SR 95531, showing total binding. The nonspecific binding signal in the presence of 100 μm GABA was similar in wild-type and −/− brains (data not shown). Similar distinct pharmacological profiles were observed between the wild-type and α₆ −/− brains in each of seven pairs of adult mice studied. D, E,In situ hybridization x-ray film autoradiographs of adult mouse brains hybridized with α₄ (D) and δ-specific (E) ³⁵S-labeled oligonucleotide probes. Wild-type (+/+) brains are on theleft; α₆ −/− brains are on theright. No differences can be seen in subunit mRNA levels between +/+ and −/− brains. Note also the very similar pattern of α₄ and δ gene expression in the forebrain/thalamus regions, and the correlation with the distribution of [³H]muscimol (B). Cbgr, Cerebellar granule cells; CP, caudate-putamen;Ctx, cerebral cortex; Gr, cerebellar granule cell layer; H, hippocampus; IC, inferior colliculus; OB, olfactory bulb;T, thalamus.

Table 1.

Determination of Bz binding in α₆ −/− mice

Ligand	+/+		−/−
	Kd	Bmax	Kd	Bmax
Ro 15-1788	0.91  ± 0.16	1222  ± 144	0.84  ± 0.17	1021  ± 141
Ro 15-4513 total sites	9.4  ± 2.0	2106  ± 203	6.3  ± 1.5*	1159  ± 1071-160
DS	5.9  ± 0.8	1199  ± 114	5.7  ± 0.5	1088  ± 174
Zolpidem	17.9  ± 4	1155  ± 152	19.6  ± 4.8	993  ± 161
Muscimol	6.3  ± 0.8	2568  ± 326	6.2  ± 1.7	661  ± 234

Data shown are the mean ± SEM of cerebellar membranes prepared independently from six animals. Saturation analysis used eight concentrations of ligand in duplicate for each animal.B_max (fmol/mg protein) andK_d (nm) values were determined by nonlinear least-squares fit of the saturation curves using the data analysis software RS1.

^*Significantly different from +/+ (p < 0.05);

F1-160: significantly different from +/+ (p < 0.005). DS, Diazepam sensitive.

The α₆ subunit has a closely related homolog, the α₄ subunit, which is expressed in certain forebrain areas such as the thalamus (Wisden et al., 1991, 1992). The recombinant α₄ subunit in an α₄β_xγ₂ configuration displays a pharmacological profile identical to that of α₆β_xγ₂ receptors, and α₄ mRNA is found at low levels in cerebellar granule cells of adult rats (Wisden et al., 1991; Laurie et al., 1992). Thus we looked to see whether there had been a compensatory change in α₄ expression in the cerebellum of α₆ −/− mice (Fig. 4D); however, consistent with the absence of DIS binding in α₆ −/− animals, there was no upregulation of α₄ mRNA in −/− cerebella (Fig.4D).

The BZ sensitivity of GABA_A receptors in α₆ −/− cerebellar granule cells was investigated directly using electrophysiology on cultured granule cells isolated from P5 animals. After 14–17 d in vitro, we tested the effects of BZ agonist flunitrazepam (1 μm) coapplication on the current amplitude generated by 20 μm GABA. Results of a typical culture are shown in Figure 5. Examination of wild-type cells revealed a heterogeneous response: flunitrazepam-induced potentiation of the GABA response took place in approximately half (20 of 36) of the cells tested (Fig. 5C,top row). In those cells with flunitrazepam-potentiated GABA responses, the average potentiation was 58 ± 7%. In contrast, in age-matched cultures derived from α₆ −/− cerebella, 16 of 18 cells tested had flunitrazepam-induced potentiations of their GABA responses. The average potentiation was 62 ± 7% (Fig.5C).

Fig. 5.

Electrophysiological characterization of GABA_A receptors in cerebellar granule cells from wild-type and α₆ −/− cells. Photomicrographs in Aand B show typical examples of lacZ-expressing cerebellar granule cells, isolated from P5 α₆ −/− mouse cerebella, and cultured for 3 weeks. A is a low-magnification view, showing the mosaic of blue(lacZ-positive) cells scattered throughout the culture. Both isolated and clustered blue cells can be seen. Within any given cluster, not all the cells are blue and therefore are not expressing the α₆ gene. The cells have been counterstained with neutral red. All electrophysiological recordings were from isolated cells. Arrows inB show examples of non-lacZ-expressing cells. Scale bars: A, 200 μm; B, 30 μm.C, Example responses to 120 msec applications of 20 μm GABA alone, and 20 μm GABA with 1 μm flunitrazepam (open arrows). Thetop row shows an example of wild-type cells (+/+) with GABA_A receptors that responded to flunitrazepam (left trace) or were insensitive to flunitrazepam (right trace). The bottom row shows a typical GABA_A response from an α₆ −/− cell and the associated flunitrazepam potentiation. From left to right and top to bottom, the valuex on the scale bar corresponds to 200, 230, and 170 pA, respectively. The traces were averages of three to five consecutive records.

To examine a possible reason for the heterogeneity of the GABA_A receptor response, the extent of gene expression from the α₆ locus in cultured α₆ −/−granule cells was analyzed with β-galactosidase histochemistry. At 3 weeks in culture, numerous cells strongly stained dark blue after incubation with X-Gal (Fig. 5A,B); however, there were many adjacent “granule-like” cells that either contained just a few blue particles or were completely unstained (Fig. 5B). This applied both to isolated cells and to cells in large clusters. There was no obvious correlation between the location of cells (isolated or in clusters) and lacZ expression. This mosaic of blue cells is evidence that at least in culture, not all granule cells or granule-like cells express the α₆ gene (cf. Santi et al., 1994), and may explain the heterogeneous nature of the BZ potentiation seen in our cultures.

Selective δ subunit protein loss from cerebellar granule cells of α₆ −/− mice

A key and controversial question for cerebellar granule cell GABA_A receptors has been which subunits co-assemblein vivo (Wisden et al., 1996). To examine one aspect of this, we immunoprecipitated deoxycholate-solubilized cerebellar GABA_A receptors from α₆ −/− mice with a δ-specific polyclonal antiserum, δ(318–400), raised against the putative intracellular loop domain between TM3 and TM4 (Quirk et al., 1994, 1995). In wild-type and α₆ +/− cerebella, the δ(318–400) antiserum precipitated the same number of muscimol binding sites (Fig. 6A). By this assay, the δ protein was also present in both pure wild-type 129/Sv and pure C57BL/6 cerebella (data not shown). In contrast, immunoprecipitation of δ-containing receptors from α₆−/− cerebella was greatly reduced (Fig. 6A). These data were extended by Western blot analysis of membrane protein samples isolated from individual +/+ and α₆ −/− cerebella. With use of a δ-subunit-specific antibody, δ(1–44)R7, raised against the N terminus (R. Pelz and W. Sieghart, unpublished data), α₆ −/− samples showed a dramatic reduction in the 53 ± 1 kDa δ subunit band intensity to 25 ± 8% of +/+ levels (Fig. 6B). The residual δ protein in the α6 −/− tissue had the same molecular weight as that in wild-type tissue (Fig. 6B).

Fig. 6.

Immunoprecipitation and immunoblot analysis of GABA_A receptor δ subunit levels in wild-type and α₆ −/− cerebella. A, After cerebellar GABA_A receptors were solubilized in Triton X-100/deoxycholate, the number of [³H]muscimol binding sites immunoprecipitated by the δ(318–400) antiserum from +/+, +/−, and −/− cerebella was determined (n = 10–14).B, Immunoblot analysis: the marked δ subunit reduction in α₆ −/− cerebella detected with the δ(1–44)R7 antiserum. The identity of the low molecular weight doublet (33 and 31 kDa) seen in all samples is unknown. C, A 51 kDa α₁ immunoreactive band is present in both −/− and +/+ cerebellar samples as detected with the α1(328–382)/α1L antibody.

Immunocytochemistry with the δ(1–44)R7 antibody clearly supported the Western blot and immunoprecipitation data (Fig. 7). Light microscopic immunocytochemistry with this antibody revealed a very intense cerebellar granule cell layer staining in wild-type animals (Fig. 7A), similar to that reported earlier using a different δ-specific antibody (Benke et al., 1991; Gao and Fritschy, 1995). The immunoreactivity originated mainly from staining of the glomeruli, and granule cell bodies were only weakly outlined (Fig.7C). The glomeruli appeared as dark rings of labeled granule cell dendrites surrounding pale centers representing the unstained mossy fiber terminals (Fig. 7C). In contrast to the wild-type animals, in α₆ −/− mice the granule cell layer immunostaining for the δ subunit was virtually absent (Fig.7B). In particular, no immunoreactivity could be detected in the glomeruli (Fig. 7D), suggesting that α₆−/− granule cell dendrites contain either no δ subunit protein or an undetectably low level. Electron microscopic examination of the immunoreactivity for the δ subunit in the granule cell layer further confirmed the lack of δ subunit immunoreaction in α₆−/− granule cells (not shown). Therefore, the residual δ subunit immunoreactivity seen on Western blots may represent a level of protein undetectable by immunocytochemistry under our conditions, or it could come from cell types other than granule cells, because whole cerebella were used to prepare the protein extracts.

Fig. 7.

Immunodetection of the δ subunit of the GABA_A receptor in α₆ +/+ (A,C) or α₆ −/− (B, D) cerebella, using a polyclonal antibody δ R7 and immunoperoxidase reaction. The granule cell layer showed intense immunoreactivity in α₆ +/+ animals but almost no staining was observed in the α₆ −/− mouse. C, At higher magnification, it is evident that the δ subunit is localized mainly in the glomeruli, granule cell bodies (gc) being only weakly outlined. The glomeruli appear as dark rings of granule cell dendrites surrounding a pale center (arrows) representing the unstained mossy fiber terminal. D, In the α₆ −/− mice, both the granule cell bodies (gc) and the glomeruli (asterisks) are immunonegative for the δ subunit. C andD were photographed using DIC optics. Scale bars:A, B, 500 μm; C, D, 10 μm.

The δ subunit loss occurs post-translationally

One possibility to explain the loss of δ protein from α₆ −/− granule cells is through a change in regulation at the mRNA level; however, the δ subunit mRNA level in the cerebellar granule cells was at normal levels when examined by in situ hybridization (Fig. 4E). High levels of δ mRNA were seen in both wild-type and −/− granule cells. δ mRNA expression was also examined in both pure 129/Sv and C57BL/6 strain wild-type animals and found not to differ (not shown). This result suggests that the loss of δ subunit from the α₆ −/− cerebellar granule cells occurs post-translationally.

[³H]Muscimol and [³H]SR95531, two ligands that mark out α₆-and δ-containing GABA_A receptors

[³H]Muscimol and [³H]SR95531 are ligands that highlight restricted GABA molecule conformations (Sieghart, 1995). In particular, [³H]muscimol is the classic GABA_A ligand and has been used extensively for mapping GABA_A receptors in the brain (Palacios et al., 1980; Olsen et al., 1990). We used [³H]muscimol and [³H]SR95531 to autoradiographically probe the remaining GABA_A receptors in the δ subunit-deficient/α₆ −/− cerebellar granule cell layer. A clear cut, but completely unanticipated, pharmacological feature was revealed: the selective and extensive loss of high-affinity [³H]muscimol (Fig. 4B) and [³H]SR95531 (Fig. 4C) binding from the granule cells. Binding over the cerebellar molecular layer with these ligands remained unchanged, as did the levels of binding in the forebrain, e.g., normal levels of [³H]muscimol binding remain over the thalamus of −/− animals (Fig. 4B). The decrease in [³H]muscimol binding seen by autoradiography in α₆ −/− individuals was further quantified by studying [³H]muscimol binding to membranes from whole cerebella (Table 1). The level of high-affinity [³H]muscimol sites was reduced to ∼25% of that found in control animals (Table 1). No significant reduction in [³H]muscimol binding was seen in +/− animals (data not shown). Saturation analysis revealed no change in the observed K_d values for [³H]muscimol in −/− animals (K_dis ∼6 nm; Table 1). The residual binding is likely to come from sites within the molecular layer, the granule cell layer, and the deep cerebellar nuclei, all of which contain an α₁β_2/3γ₂ component. Under autoradiographic conditions, however, [³H]muscimol does not highlight these α1-containing receptors in the cerebellum. Rather, [³H]muscimol and [³H]SR95531 seem to selectively highlight α₆δ-containing receptors.

Evaluation of α₁ subunit levels in α₆−/− and δ-deficient cerebella

The α₁ protein is expected to account for the majority of the remaining α subunits in the cerebellum of α₆ −/− mice (Sieghart, 1995; McKernan and Whiting, 1996; Wisden et al., 1996). To examine whether there was any concomitant change in the α₁-receptor population in the α₆ −/− cerebella, the portion of α₁subunits complexed with the β_2/3 and γ₂subunits was measured using three different ligands targeting the BZ binding site. These assays are not likely to measure any α₁ subunits that are complexed with the δ but not the γ₂ subunits, e.g., α₁β_2/3δ, because the GABA responses of these complexes cannot be modulated by BZs (Saxena and Macdonald, 1994).

Full saturation analysis was carried out on cerebellar membranes to determine the K_d and B_maxvalues for [³H]Ro 15-1788 binding, diazepam-sensitive [³H]Ro 15-4513 binding, and [³H]zolpidem binding (Table 1). All three ligands identified approximately the same number of binding sites in the cerebellar membranes (∼990–1160 fmol/mg protein). There was no statistically significant difference between the number of binding sites for any ligand between the α₆ +/+ and δ-deficient/α₆ null groups, although the trend was always toward a reduced number of sites in the −/− animals. Total [³H]Ro 15-4513 binding sites (both diazepam-sensitive and -insensitive components, with the nonspecific binding being defined in the presence of 10 μm Ro 15-4513), however, were decreased by 44% in the −/− cerebella (Table1), with a minor change in the affinity K_dconstant. This figure is in line with the 30–40% contribution that the α₆β_2/3γ₂ component has been reported to make to the total Ro 15-4513 binding sites in the cerebellum (Sieghart et al., 1987; Turner et al., 1991; Korpi et al., 1993; Quirk et al., 1994).

These binding results suggest that the amount of total α₁ protein complexed with β_2/3 and γ₂ in the cerebellum is essentially unchanged between δ-deficient/α₆ null and wild-type animals. Furthermore, immunocytochemistry with a polyclonal α₁-specific antibody (P16) showed no overt change in granule cell layer immunostaining at the light microscopic level in α₆ −/− cerebella compared with wild-type tissue (data not shown); however, this method may not pick out small changes in subunit levels. In fact, immunoblotting with α₁-specific antibodies did show a downward trend in α₁ protein levels between δ-deficient/α₆ null and +/+ cerebellar samples (Fig.6C). In −/− animals, a small reduction with high variability was seen in the α₁ 51 kDa band intensity, as determined by densitometric measurements. This was observed independently with three different α₁-specific antibodies: α₁(1–14) (S. Pollard and F. A. Stephenson, unpublished data), α₁(1–9) (Zezula and Sieghart, 1991), and α₁(328–382) (R. Pelz and W. Sieghart, unpublished data) (see Materials and Methods).

Behavioral characterization of α₆null/δ-deficient mice

The total loss of α₆ from the cerebellum and the associated severe reduction of δ subunit levels might be expected to have consequences for nervous system function in the mutant mice. The cerebellum integrates sensory input needed for maintaining balance and orientation, has a prominent role in the refinement of motor action, and may also participate in motor memory storage (Raymond et al., 1996). We looked for evidence of cerebellar-associated motor deficits in the −/− mice. In terms of simple observable behavior, mutant mice seem indistinguishable from wild-type littermates. The adult α₆ null/δ-deficient mice are active and agile, whether they be in the cage or roaming freely, and exhibit spontaneous activity, such as walking upside down on the ceiling of their cages. In an open field test, mutant mice showed as much exploratory activity as individuals with normal levels of α₆ and δ proteins (Table 2). Mutant as well as wild-type mice learned the horizontal wire task (data not shown; see Materials and Methods). Both the wild-type and α₆ null/δ-deficient mice reached the rotating rod test learning criterion (Table 2). With these tests, we found no evidence to suggest any form of ataxia associated with cerebellar dysfunction. Additionally, a detailed behavioral analysis on an independently generated α₆ −/− mouse line, where the same exon was disrupted by insertion of a neo gene (exon 8,NcoI site) in a 129xC57BL/6 background, showed no abnormalities in motor behaviors (Gregg E. Homanics, Department of Anesthesiology, University of Pittsburgh, personal communication).

Table 2.

Open field activity and rotating rod learning of wild-type and α₆ −/− mice

	n	Open field activity/5 min
		Number of crossings		Number of fecal boli		Number of rearings
+/+	10	96  ± 5		3.3  ± 0.7		6.7  ± 1.5
−/−	10	84  ± 18		3.2  ± 0.5		5.2  ± 1.1

	n	Rotarod performance: seconds on the rod per 180 sec/(percent of animals staying for 180 sec)
		Trial 1	Trial 2	Trial 3	Trial 4	Trial 5	Trial 6
+/+	15	114  ± 11 (20%)	147  ± 9 (40%)	160  ± 11 (67%)	179  ± 1 (87%)	179  ± 1 (87%)	180  ± 0 (100%)
−/−	14	144  ± 11 (43%)	148  ± 12 (57%)	167  ± 8 (79%)	176  ± 3 (86%)	176  ± 4 (93%)	180  ± 0 (100%)

Results are given as means ± SEM. No statistically significant differences (p < 0.05; Student’s ttest and repeated measures ANOVA). The Rotarod learning test was repeated with another batch of mice with similar results.

DISCUSSION

A mouse line lacking functional GABA_A receptor α₆ subunit protein has been generated. Because of the restricted α₆ gene expression profile, this mutation was expected to principally affect the cerebellum. Furthermore, in the cerebellum, the granule cell δ subunit protein level was markedly reduced relative to wild-type levels. Thus these mice effectively harbor a region-specific double subunit knock-out, and the GABA_A receptor complexity on granule cells is reduced to receptors largely containing just α₁, β_2/3, and γ₂ subunits. Two issues are discussed: the significance of multiple α subunits and defined assembly pathways for receptor subunits.

GABA_A receptor α subunit heterogeneity

Surprisingly, in spite of a large loss of granule cell GABA_A receptors, the α₆ null/δ-deficient mice are not grossly impaired in motor skills. This lack of phenotype under laboratory conditions was not anticipated from α₆gene comparative studies. Both the conservation of peptide sequence in the N-terminal domain and a granule cell-specific expression pattern in the cerebellum of fish, birds, rodents, and humans imply that there has been a continual selection for the α₆ protein (Bahn et al., 1996; Hadingham et al., 1996).

In the α₆ null/δ-deficient mice, physiological changes in granule cell GABA_A receptors are expected, but these have not obviously impaired cerebellar function. Removal of two of the six subunits from granule cells will still leave functional receptors with α₁β_2/3γ₂ subunit combinations. Nevertheless, substitution of different α subunits in an α_xβ_xγ₂ complex may influence the inhibitory postsynaptic current kinetics (Gingrich et al., 1995; Tia et al., 1996). Synaptic transmission at GABAergic synapses is generated by millisecond pulses of 0.5–1 mmGABA (reviewed by Mody et al., 1994). Under these conditions, recombinant α₁β_2/3γ₂ and α₆β_2/3γ₂ receptors do behave differently, with the α₆-containing receptors having a slower deactivation rate (Tia et al., 1996). The physiological role of the δ subunit remains obscure (Shivers et al., 1989). During long applications of micromolar GABA, δ-subunits slow the acute macroscopic desensitization rate of recombinant GABA_Areceptors (Saxena and Macdonald, 1994); however, this property has not been studied using fast, brief GABA application.

Selective subunit partnerships

The α₆ −/− mouse has provided insight into GABA_A receptor subunit assembly pathways in neurons. The α₆ protein derived from the targeted gene should terminate just after TM2. An analogous example has been studied for the mouse muscle nicotinic receptor δ subunit. When truncated just after TM2 (δM2) and co-expressed with wild-type nicotinic α, β, γ, and δ subunits in COS cells, δM2 interferes with receptor assembly (Verrall and Hall, 1992). Similarly, the truncated GABA_Aα₆ protein (α₆M2) may prevent α₆δ-containing receptors from reaching the granule cell surface. The association of α₆M2 and δ may inhibit mature receptor expression by forming specific complexes in the endoplasmic reticulum that are not permissive for further receptor assembly and/or trafficking. These will be retained and degraded (Verrall and Hall, 1992; Connolly et al., 1996). As for the nicotinic acetylcholine and glycine receptors (Verrall and Hall, 1992; Kuhse et al., 1993; Sumikawa and Nishizaki, 1994; Kreienkamp et al., 1995), the information needed for specific assembly of the GABA_Areceptor α₆ and δ proteins is likely to be in their N-terminal domains, because the N-terminal domain of α₆is sufficient to block δ expression. Because they interact as an assembly intermediate, α₆ and δ probably occur adjacent to each other in the mature receptor subunit ring.

There are several other scenarios. The α₆M2 polypeptide could be degraded before pairing with the δ subunit. Because the δ subunit is not efficiently incorporated with other subunits, this might in turn be degraded. Alternatively, if no α₆ protein is present, the δ mRNA might be translated inefficiently, implying that α₆ protein levels feed back to regulate the translation of δ mRNA. Although this is an interesting possibility, there is no known mechanism.

Our results seem to confirm the antibody-based data suggesting that in vivo, δ predominantly assembles with α₆ and not α₁ (Caruncho and Costa, 1994;Quirk et al., 1994; Caruncho et al., 1995). Nevertheless, from the genetic results alone, an α₆ and δ interaction may be simply the first step allowing other subunits such as α₁to subsequently join the complex. Thus, both α₁α₆βδ or even α₁α₆βγ₂δ might existin vivo (Mertens et al., 1993; Pollard et al., 1995; R. Pelz and W. Sieghart, unpublished observations); however, because α₁, β_2/3, and γ₂ subunits cannot rescue the δ subunit in an α₆ −/− background, we predict that the α₁β_xδ and α₁β_xγ₂δ combinations will not be found to any great extent in vivo.

In recombinant systems [Xenopus oocytes, human embryonic kidney (HEK) 293 cells, and mouse L929 fibroblast cells], the situation is different. The δ subunit can assemble to form functional receptors with either α₁ or α₆ as α₁β_xδ, α₆β_xδ, and possibly α₁β_xγ₂δ complexes, with the exact β subunit used having little influence (Saxena and Macdonald, 1994, 1996; Ducic et al., 1995; Krishek et al., 1996). Thus there may be unique architectural editing or chaperone mechanisms present in granule neurons that are not found in Xenopus oocytes or HEK cells. Alternatively, the subunits may differ slightly in affinity for each other. In a recombinant system, the large amounts of protein present may allow many combinations to assemble, even if they have nonoptimal association parameters. The results presented here demonstrate the importance of studying subunit assembly pathways in the brain.

[³H]Muscimol as a selective autoradiographic probe for α₄, α₆, and δ subunit associations

It has been suggested that under autoradiographic binding conditions the GABA_A site ligands [³H]muscimol and [³H]SR 95531 highlight a subpopulation of receptors in native membranes (Olsen et al., 1990). In a wide range of vertebrates, a hallmark of GABA_A sites in the CNS is the high levels of [³H]muscimol binding over the cerebellar granule cell layer (Palacios et al., 1980; Schmitz et al., 1988; Olsen et al., 1990; for review, see Wisden et al., 1996). A striking feature of our study was the almost total loss of high-affinity [³H]muscimol and [³H]SR95531 binding from the granule cell layer of α₆null/δ-deficient cerebella (Fig. 4B,C), suggesting that the α₆ and/or δ subunits are responsible for these ligand profiles. From recombinant data, the α₆β_xδ subunit combination is insensitive to BZs (Saxena and Macdonald, 1996) and sensitive to GABA (EC₅₀ in the low micromolar range) but gives small currents (Saxena and Macdonald, 1994, 1996; Ducic et al., 1995). These properties would be consistent with the pharmacology of the cerebellar δ-containing receptors immunoprecipitated with a δ-specific antibody: high muscimol affinity and no BZ binding (Quirk et al., 1994).

Despite the absence of autoradiographic signal in −/− cerebella, muscimol is still an effective agonist of GABA_A receptors on cultured α₆ −/− granule cells (J. Mellor and A. D. Randall, unpublished observations), although electrophysiological assays most likely use the low-affinity site. The remaining α₁β_2/3γ₂ receptors in α₆ −/− granule cells should have aK_d for [³H]muscimol of ∼5 nm (Lüddens et al., 1990) and could be expected to bind [³H]muscimol, but this is not the case under autoradiographic assay conditions. Similarly, in the inferior colliculi of wild-type and −/− animals, given the high concentration of α₁β₂γ₂ receptors present (Persohn et al., 1992; Wisden et al., 1992; Fritschy and Möhler, 1995), it is difficult to explain the virtual absence of [³H]muscimol binding sites in autoradiography (Fig.4B). There may be some factor specifically associated with α₁β_2/3γ₂ receptors on native membranes that prevents high-affinity [³H]muscimol binding.

As pointed out previously (Shivers et al., 1989; Laurie et al., 1992), both δ subunit mRNA and protein closely parallel the distribution and abundance of [³H] muscimol binding (Fig.4B,E), e.g., highest in cerebellar granule cells, followed by the thalamus, and then caudate-putamen, neocortex, and dentate gyrus (Shivers et al., 1989; Olsen et al., 1990; Benke et al., 1991). In the forebrain, α₄ subunit mRNA also largely follows δ subunit distribution (compare Fig. 4, B andD; and see Wisden et al., 1992). The α₄ and α₆ subunits are closely related and form a subgroup set apart from the other α subunits (Seeburg et al., 1990; Ortells and Lunt, 1995; Tyndale et al., 1995). The total evidence suggests strongly that α₄δ forms part of a GABA_A receptor that is an α₆δ combination homolog. A phylogenetic clock comparison calculated that α₄ and α₆are the oldest α subunits and that the δ subunit is the oldest of all GABA_A receptor subunit genes (Ortells and Lunt, 1995). Therefore, a selective interaction of α₄ and α₆ with δ might represent an early vertebrate GABA_A receptor subtype.

In conclusion, we have provided evidence for a specific association between the α₆ and δ subunits in granule cell GABA_A receptors. It seems likely that similar assembly rules exist for other brain heteromeric ligand-gated channels, e.g., the neuronal nicotinic acetylcholine receptor (Vernallis et al., 1993) and ionotropic glutamate receptors.