The Hydrophilic Isoform of Glutamate Decarboxylase, GAD67, Is Targeted to Membranes and Nerve Terminals Independent of Dimerization with the Hydrophobic Membrane-anchored Isoform, GAD65*

GAD67, the larger isoform of the γ-aminobutyric acid-synthesizing enzyme glutamic acid decarboxylase, is a hydrophilic soluble molecule, postulated to localize at nerve terminals and membrane compartments by heterodimerization with the smaller membrane-anchored isoform GAD65. We here show that the dimerization region in GAD65 is distinct from the NH2-terminal membrane-anchoring region and that a membrane anchoring GAD65 subunit can indeed target a soluble subunit to membrane compartments by dimerization. However, only a fraction of membrane-bound GAD67 is engaged in a heterodimer with GAD65 in rat brain. Furthermore, in GAD65−/− mouse brain, GAD67, which no longer partitions into the Triton X-114 detergent phase, still anchors to membranes at similar levels as in wild-type mice. Similarly, in primary cultures of neurons derived from GAD65−/− mice, GAD67 is targeted to nerve terminals, where it co-localizes with the synaptic vesicle marker SV2. Thus, axonal targeting and membrane anchoring is an intrinsic property of GAD67 and does not require GAD65. The results suggest that three distinct moieties of glutamate decarboxylase localize to membrane compartments, an amphiphilic GAD65 homodimer, an amphiphilic GAD65/67 heterodimer, tethered to membranes via the GAD65 subunit, and a hydrophilic GAD67 homodimer, which associates with membranes by a distinct mechanism.

GAD67, the larger isoform of the ␥-aminobutyric acidsynthesizing enzyme glutamic acid decarboxylase, is a hydrophilic soluble molecule, postulated to localize at nerve terminals and membrane compartments by heterodimerization with the smaller membrane-anchored isoform GAD65. We here show that the dimerization region in GAD65 is distinct from the NH 2 -terminal membrane-anchoring region and that a membrane anchoring GAD65 subunit can indeed target a soluble subunit to membrane compartments by dimerization. However, only a fraction of membrane-bound GAD67 is engaged in a heterodimer with GAD65 in rat brain. Furthermore, in GAD65؊/؊ mouse brain, GAD67, which no longer partitions into the Triton X-114 detergent phase, still anchors to membranes at similar levels as in wild-type mice. Similarly, in primary cultures of neurons derived from GAD65؊/؊ mice, GAD67 is targeted to nerve terminals, where it co-localizes with the synaptic vesicle marker SV2. Thus, axonal targeting and membrane anchoring is an intrinsic property of GAD67 and does not require GAD65. The results suggest that three distinct moieties of glutamate decarboxylase localize to membrane compartments, an amphiphilic GAD65 homodimer, an amphiphilic GAD65/67 heterodimer, tethered to membranes via the GAD65 subunit, and a hydrophilic GAD67 homodimer, which associates with membranes by a distinct mechanism.
Glutamic acid decarboxylase (GAD) 1 (EC 4.1.1.15) is the key enzyme in the synthesis of ␥-aminobutyric acid (GABA) (1). GAD is expressed at comparable levels in GABA-ergic neurons in the central nervous system and in the insulin-producing ␤ cells in the islets of Langerhans (2,3). GABA has been established as the major inhibitory neurotransmitter in the central nervous system, whereas its role in islet cell function remains elusive (4).
Mammalian species express two isoforms of GAD designated GAD65 and GAD67 in accordance with their relative molecular masses in kDa (3,5,6). GAD65 and GAD67 are highly conserved in evolution. Although the two proteins differ substantially in the first 95 NH 2 -terminal amino acids, they share a significant homology in the remaining part of the molecule, which contains the catalytic portion of the enzyme (ϳ78% identity, ϳ95% similarity). GAD65 and GAD67 show significant differences in their levels of expression in different brain regions and steady state saturation with the co-enzyme pyridoxal 5Ј-phosphate (PLP) (1). More than half of GAD in rat brain is present as the PLP-free apoenzyme (7,8), and GAD65 constitutes the majority of this reservoir (9 -11). GAD67 constitutes a small fraction of the apo-GAD reservoir in the brain (11) and is predominantly found as a holoenzyme tightly associated with PLP (9 -11). GAD65, but not GAD67, is a major target of autoimmune responses directed against pancreatic ␤ cells in type 1 diabetes (12,13) and toward GABA-ergic neurons in a rare disease of the central nervous system, stiff-man syndrome (14).
Most biochemical studies on the quaternary structure of purified GAD are consistent with the idea that the basic subunit structure is dimeric (1,15). High resolution mapping of conformational epitopes in GAD65, together with two-and three-dimensional structure prediction and alignment of structural folds with decarboxylases of known structure, resulted in a model of the GAD65 dimer (16). In this model, the two GAD65 subunits are joined in the middle region (amino acids 201-461) and related by a P 21 symmetry. In a separate analysis, several residues in the middle domain were predicted to reach into the cofactor-binding site of the other subunit and interact directly with the phosphate group of PLP and with residues in the other subunit of the enzyme (17). Thus, the middle domains of GAD65 and GAD67 seem to be responsible for the basic homodimeric structure of the enzyme. GAD65 and GAD67 can also associate with each other to form heterodimers (9,18,19), and this association has been attributed to the NH 2 -terminal domain of GAD65 (18). It is estimated that GAD65/GAD67 heterodimers constitute ϳ28% of the total GAD activity in cerebellar membranes (19).
Both GAD65 and GAD67 are synthesized in the cytosol as hydrophilic and soluble molecules (20). GAD65 undergoes a stepwise post-translational modification in the NH 2 -terminal domain to become hydrophobic and then reversibly membraneanchored to synaptic-like microvesicles in ␤ cells and synaptic vesicles in neurons (21,22). A pool of GAD65 is also detected in the Golgi complex region of neurons, pancreatic ␤ cells, and transfected CHO and COS-7 cells (23,24). The nature and significance of the association of GAD65 with membranes is not well understood. The NH 2 -terminal domain in GAD65 is required for membrane anchoring and is the site of two hydrophobic modifications. One of them involves the thiopalmitoylation of cysteines 30 and 45 (22), whereas the other modification (20) has remained undefined. GAD65 also undergoes phosphorylation of serines 3, 6, 10, and 13 (25). Although both palmitoylation and phosphorylation distinguish the membrane-associated from the soluble forms of GAD65, neither modification is required for membrane anchoring of GAD65 in COS-7 cells (25,26). It was also shown that targeting of GAD65 to the Golgi complex region in transfected CHO cells requires a signal located in the NH 2 -terminal region, which is different from palmitoylation (24). GAD67 has both distinct and similar subcellular localizations. In rat ␤ cells and in transfected rat fibroblasts, GAD67 was found to remain distributed homogeneously throughout the cell cytosol and was recovered only in soluble fractions (20,23). In neurons, GAD67 was detected in cell bodies and proximal dendrites but also appeared concentrated in nerve terminals (9) and in the Golgi complex region (18). Furthermore, a pool of GAD67 was found to be associated with rat brain membrane fractions and distribute into the Triton X-114 detergent phase (18). Because GAD67 was only detected as a soluble hydrophilic protein in rat pancreatic islets (20,22) and in fibroblasts transfected with rat GAD67 (23), the association of GAD65 and GAD67 into heterodimers was suggested to confer association of GAD67 with membranes (18,19).
In the present study, we have addressed the role of GAD65-GAD67 heterodimerization in the subcellular targeting and membrane anchoring of GAD67. We demonstrate that formation of GAD65/67 heterodimers is specific and does not involve nonspecific formation of disulfide-linked bridges between any domains of the proteins. Furthermore, a soluble mutant of GAD65 can be targeted to membranes by dimerization with wild-type protein, suggesting that membrane anchoring of one GAD65 subunit is sufficient for membrane localization of the GAD65 dimer and supporting the notion that membrane anchoring of soluble GAD67 can be mediated by its association with GAD65. Analyses of GAD65Ϫ/Ϫ mice, however, reveal that membrane anchoring and targeting of GAD67 to nerve terminals is not affected by loss of GAD65 and therefore can also be effected independently of the GAD65 isoform.

Antibodies
The GAD65-specific mouse monoclonal antibodies, GAD6 (27), which recognizes a linear epitope localized in the last 41 amino acids of GAD65 (28), and FL-17 (a gift from Dr. Manfred Ziegler), which recognizes a linear epitope in the first 8 amino acids of GAD65, 2 were used in the immunoprecipitation experiments. The 1701 rabbit antiserum was raised against a 19-amino acid peptide in the COOH terminus of rat GAD67 and recognizes GAD65 and GAD67 equally well on Western blots (28). The following antibodies were used in immunofluorescence experiments: (i) the GAD65-specific human monoclonal antibodies MICA 2, 3, and 6, derived from islet cell antibody positive patients, were a gift from Dr. Wiltrud Richter (29); (ii) K2 rabbit antiserum, which recognizes primarily GAD67 on Western blots (9), was purchased from Chemicon (Temecule, CA); (iii) a mouse monoclonal antibody against the SV2 antigen in synaptic vesicle membrane (30) was kindly donated by Drs. Peter Sargent and Regis Kelly (University of California, San Francisco). All secondary antibodies used in immunofluorescence experiments were purchased from Jackson Immunoresearch Laboratories (West Grove, PA).

Expression of Wild-type and Mutant Proteins in COS-7 Cells
Wild-type rat GAD65, a deletion mutant of rGAD65 lacking amino acids 1-38 and having the palmitoylation site, Cys 45 , mutated to Ala (rGAD65/45A⌬1-38), and a deletion mutant of rGAD65 lacking amino acids 1-101 (rGAD65⌬1-101) in the COS expression vector pSV-SPORT were described previously (26). A 2.0-kb cDNA for mouse GAD67 (a gift from Dr. Roland Tisch, University of North Carolina, Chapel Hill) containing the entire coding region and 93 and 181 bp of 5Јand 3Ј-untranslated sequences, respectively, was excised from Bluescript II SKϮ (Stratagene, La Jolla, CA) as an XhoI-XhoI fragment and subcloned into the SalI site of the pSV-SPORT vector. COS-7 cells were cultured and transiently transfected with LipofectAMINE (Life Technologies, Inc.) as described previously (25).

Subcellular Fractionation
Subcellular fractionation of cells was carried out using a modification of methods described previously (26). All procedures were carried out at 4°C. Transfected COS-7 cells, grown to 80 -90% confluency in a 10-cm plate, were washed twice in ice-cold cell harvesting buffer and homogenized on ice using a glass homogenizer with 0.5 ml of HMAP buffer. The homogenate was centrifuged at 800 g for 10 min to remove nuclei and cellular debris. The postnuclear supernatant was centrifuged at 150,000 ϫ g for 1 h in a TLA 100.3 rotor using a Beckman tabletop ultracentrifuge (Beckman, Palo Alto, CA) to separate cytosol and crude membrane fractions. Sedimented membranes were resuspended in 0.5 ml of membrane washing buffer (HMAP buffer containing 0.5 M NaCl) and incubated for 1 h, followed by ultracentrifugation at 150,000 ϫ g for 1 h to pellet the washed membranes. Washed membranes were extracted for 1 h with 0.5 ml of extraction buffer (HMAP buffer containing 150 mM NaCl and 1% Triton X-114) followed by centrifugation for 1 h at 150,000 ϫ g to separate particulate extract from insoluble debris.
For the subcellular fractionation of rat and mouse brain, Harlan Sprague Dawley male rats (250 -300 g) were anesthetized with metofane, and the cerebral cortex and cerebellum were dissected quickly. A 15% (w/v) portion of tissue in HMAP buffer was homogenized with the use of a Tissumizer polytron (Tekmar, Cincinnati, OH). The rest of the procedure was carried out as described above for COS-7 cells.

Triton X-114 Phase Separations
Triton X-114 partitioning assays were performed on the subcellular fractions of mouse brain using a modification of the procedure of Bordier (31). For comparative analyses of the amphiphilic properties of GAD65 and GAD67 in subcellular fractions, buffer compositions were adjusted to achieve identical conditions in each fraction. Aliquots of each fraction were subjected to three rounds of temperature-induced phase transitions using 1% Triton X-114. Following the incubation at 37°C for 5 min, the detergent and aqueous phases were separated by centrifugation at 12,000 ϫ g for 2 min at room temperature. The detergent phase from the first and second rounds of phase separations was washed carefully with HMAP buffer containing 150 mM NaCl and resuspended with the same buffer to the original volume. Triton X-114 was added to the aqueous phase obtained after the third round of phase separation to a final concentration of 1%.

Immunoprecipitations
For the immunoprecipitation of GAD65 from cell-free extracts of transfected COS-7 cells, an aliquot of 250 l from each extract was incubated with 30 l of the GAD65-specific monoclonal antibody, FL-17, for 2 h at 4°C with constant shaking. Immunocomplexes were isolated by incubation for 1 h at 4°C with 50 l of 50% solution of protein G-Sepharose (PGS; Amersham Pharmacia Biotech, Uppsala, Sweden) in immunoprecipitation buffer (IP buffer, 10 mM Hepes/NaOH, pH 7.4, 150 mM NaCl, 5 mM EDTA, 0.1 g/liter bovine serum albumin, 10 mM benzamidine, and 0.5% Triton X-114). For the immunoprecipitation of GAD65 from rat brain subcellular fractions, 100-l aliquots from each of the cytosol and membrane extracts were diluted 5-fold in HMAP buffer containing 1% Triton X-114 and 150 mM NaCl and precleared by incubation for 30 min at 4°C with 100 l of preswollen protein A-Sepharose (PAS; Amersham Pharmacia Biotech) in IP buffer. Samples were then centrifuged at 750 ϫ g for 1 min, and the resulting supernatants were incubated with 5 l of GAD6 ascites. After 3 h of incubation at 4°C with constant shaking, immunocomplexes were isolated by incubation with 100 l of preswollen PAS for 1 h at 4°C. The PGS-or PAS-bound immunocomplexes were washed five times with IP buffer, and proteins were eluted from PGS or PAS by boiling for 5 min in SDS-sample buffer followed by centrifugation to remove PAS or PGS.

Electrophoresis and Immunoblotting
To estimate the subunit composition of native GAD65 transiently expressed in COS-7 cells, samples from the cell lysate obtained after removing the cell debris were pretreated in nondenaturing and nonreducing sample buffer (0.3 M Tris-HCl, pH 8.8, 0.003% bromphenol blue, 10% glycerol) and then analyzed by electrophoresis on a native precast 10 or 4 -20% linear gradient of polyacrylamide gels (Novex, San Diego, CA). Standard molecular mass proteins for native gels were from Amersham Pharmacia Biotech and consisted of thyroglobulin (669 kDa), ferritin (440 kDa), catalase (232 kDa), lactate dehydrogenase (140 kDa), and albumin (67 kDa). For the analysis under nonreducing conditions to test the presence of disulfide-linked multimers of GAD65, samples were boiled for 3 min in sample buffer containing 2% SDS without reducing agent and then analyzed by SDS-PAGE on 10% polyacrylamide gels. For quantitative analyses of the ratio of GAD65 and GAD67 in immunoprecipitates and in cytosol and membrane fractions of brain, serial dilutions of each fraction were subjected to SDS-PAGE.
Resolved proteins were transferred from gels by Western blotting to Protran BA nitrocellulose transfer membranes (Schleicher and Schuell). All procedures were carried out at room temperature. The blots were blocked for 1 h in 10% nonfat dry milk, TBST buffer (20 mM Tris buffer, pH 7.4, 150 mM NaCl, and 0.1% Tween 20). After two 5-min washes with TBST, the blots were probed for 1 h with the 1701 antibody (1:5000 dilution in 5% milk, TBST), followed by three 5-min washes with TBST. The blots were then incubated for 30 min with horseradish peroxidase-conjugated donkey anti-rabbit IgG (Amersham Pharmacia Biotech) at a 1:8000 dilution in TBST. After three 5-min washes with TBS, 0.3% Tween 20 and three 5-min washes with TBST, bands were detected using enhanced chemiluminescence reagent (ECL; Amersham Pharmacia Biotech). For quantitative analyses, images made on Kodak X-Omat film were obtained at various times of exposure, and the percentages of GAD65 and GAD67 were determined from the linear range of the standard curve. Scanning of autoradiograms was performed using the DeskScan II program on a Hewlett Packard ScanJet IIcx. Densitometric analysis was performed using NIH Image software.

Cell Cultures of Cortical and Hippocampal Neurons
Cortical Neurons-Primary cultures of dissociated cerebral cortical neurons were prepared from the brains of 18-day-old mouse fetuses, which were recovered by caesarean section under metofane anesthetic. Heterozygous (GAD65ϩ/Ϫ) and homozygous mutant (GAD65Ϫ/Ϫ) mouse fetuses were obtained by breeding homozygous male (GAD65Ϫ/Ϫ) and heterozygous female (GAD65ϩ/Ϫ) mice on the 129 ϫ NOD background as described previously (11). The cortex was dissected from the brains of five fetuses out of seven, and the cultures were prepared from each cortex separately. In parallel, DNA was extracted from tail biopsies of the five fetuses and was used later for genotyping (see below). The cortex was dissected on ice in 2 ml of dissection buffer (160 mM NaCl, 5 mM KCl, 0.5 mM MgSO 4 , 2.9 mM CaCl 2 , 5 mM Hepes, 5.5 mM glucose, 2 mg/liter phenol red, pH 7.4). Tissue fragments were added to 5 ml of dissection buffer containing papain (100 units; Worthington), L-cysteine (1 mg), CaCl 2 (1 mM), and EDTA (0.5 mM) and incubated for 30 min with constant shaking at 37°C. Papain digestion was terminated by resuspension of the tissue fragments in 5 ml of B27/Neurobasal culture medium (Life Technologies, Inc.) supplemented with 1 mM GlutaMAX I, a 50 units/ml concentration of both streptomycin and penicillin (32), 12.5 mg of bovine serum albumin, and 12.5 mg of soybean trypsin inhibitor. After the tissue had settled, the supernatant was replaced with 1 ml of culture medium. Individual cells were generated by trituration of the tissue seven times, each time in 1 ml of culture medium. After the nondispersed tissue had settled, the supernatants were transferred to a 15-ml tube and centrifuged at 200 ϫ g for 3 min. The pellet was gently resuspended in culture medium containing 1% heat-inactivated fetal calf serum to reach a density of 1.7 ϫ 10 6 cells/ml. Cells were plated on sterile glass coverslips (2 ϫ 10 5 cells/ coverslip) precoated with poly-D-lysine (M r Ͼ300,000; Sigma) and collagen (Vitrogen 100, Collagen Corp., Fremont, CA) in 24-well plastic culture dishes (32). Cells were cultured at 37°C in a humidified atmosphere of 5% CO 2 , 95% air. One-half of the growth medium was exchanged 1 day after plating and every 3 days thereafter.
Hippocampal Neurons-Hippocampi of P0 Harlan Sprague Dawley rat pups were removed, and the dentate gyri were grossly dissected. Cells derived from the remaining tissue were plated as described by Lester et al. (33) except that papain was not used. B27-supplemented primary hippocampal cultures were prepared as described by Brewer et al. (32), and a 50 units/ml concentration of both streptomycin and penicillin were added. One-half of the growth medium was exchanged 1 day after plating and weekly thereafter.

Immunofluorescence Analyses
After 2 weeks of growth, cultures were fixed with 4% paraformaldehyde in phosphate-buffered saline for 15 min at room temperature. All of the subsequent steps were carried out at room temperature. Permeabilization of the cells and blocking of nonspecific binding were performed by incubation for 20 min with 1% bovine serum albumin, 0.1% Triton X-100 in phosphate-buffered saline (blocking buffer). For colocalization of both GAD65 and GAD67 in rat hippocampal neurons, the human monoclonal antibodies to GAD65 (MICA 2, 3, and 6, diluted 1:250) along with rabbit antibodies to GAD67 (K2, diluted 1:250) were applied in blocking buffer. After 1 h of incubation, immunoreactivity was visualized by secondary antibody staining using donkey anti-human IgG conjugated to Cy3 and goat anti-rabbit IgG conjugated to fluorescein isothiocyanate. For co-localization of GAD65 and GAD67 proteins with the synaptic marker SV2, triple immunostaining was performed using mouse monoclonal anti-SV2 (diluted 1:200) in addition to the antibodies to GAD65 and GAD67. SV2 was visualized using donkey anti-mouse IgG conjugated to 7-amino-4-methylcoumarin-3acetic acid. For the localization of GAD67 and SV2 in mouse cortical neurons, the rabbit antibodies to GAD67 along with the mouse antibodies to SV2 were applied in blocking buffer, followed by the donkey anti-rabbit IgG conjugated to Cy3 and goat anti-mouse IgG conjugated to fluorescein isothiocyanate. Identification of GAD clusters and their co-localization with synaptic markers was accomplished with dual color microscopy using a Nikon ϫ 100 objective (NA 1.4), mercury arc lamp illumination and standard fluorescein and Cy3 filter sets (Omega). Fluorescent images were acquired using a cooled digital CCD camera (Princeton Instruments, Inc.). Minimal bleed-through between channels was confirmed by imaging single-labeled specimens. 12-bit images were linearly written to 8-bit data set after normalizing images to maximize usable contrast range using IPLAP Spectrum software (Sygnal Analytics). For display in figures, monochrome and merged colors were processed using Adobe Photoshop software (Adobe Systems, San Jose, CA). In unprocessed raw 8-bit data, synaptic structures (identified as SV2-positive puncta) were considered co-localized with GAD if the intensity of immunoreactivity was Ͼ2-fold higher than the background level. Most GAD-positive synaptic structures had fluorescence intensities Ͼ4-fold higher than background.

Genotyping of mGAD65 Mice
DNA extracted from tail biopsies was used as a template for PCR. The following PCR primers were used in one reaction to look simultaneously at the endogenous and targeted alleles:

Specific and Nonspecific
Oligomerization of GAD65 and GAD67-Each of the GAD isoforms contains a large number of cysteine residues, suggesting a propensity for forming nonspecific disulfide bridges in nonreducing in vitro environment. It was therefore important to establish conditions for maintaining the native structure of the proteins during analysis.
In analyses of the oligomerization state of GAD65 expressed in COS-7 cells using SDS-gel electrophoresis in the presence or absence of ␤-mercaptoethanol (␤-ME), the protein was detected as a monomer, dimer, and trimer under nonreducing but denaturing conditions, whereas under reducing and denaturing conditions all of the protein was detected as a monomer (Fig.  1A), suggesting the presence of disulfide-linked bridges. The formation of disulfide-linked dimers and sometimes trimers was also observed for GAD65 and GAD67 expressed in mouse and rat brain ( Fig. 1C and data not shown). The formation of disulfide-linked oligomers was, however, significantly diminished or prevented when lysis of COS-7 cells and homogenization of rat brain was performed in the presence of alkylating reagents (Fig. 1, B and C). These results strongly suggest that the formation of disulfide-linked multimers of GAD65 and GAD67 shown here and described earlier (34) is an artifact generated in vitro.
Since the presence of alkylating reagents during the preparation of samples inhibited the formation of nonspecific disulfide bridges, all subsequent studies of GAD oligomerization were performed in the presence of 5 mM NEM. In these conditions, the native form of wild-type rGAD65 is a noncovalently linked dimer in both the cytosol and membrane fractions of COS-7 cells, as revealed by nondenaturing gel electrophoresis under nonreducing (Fig. 1D) and reducing conditions (results not shown). The native subunit structure of rGAD65 expressed in COS-7 cells did not change when lysis of cells was performed in the presence or absence of 200 M of the coenzyme PLP or in the presence or absence of 10 mM L-glutamate (data not shown). This is in contrast to results obtained for Escherichia coli GAD65, in which the dimer assembly to form a hexamer is enhanced by the addition of PLP (35).
A Membrane-anchoring GAD65 Subunit Mediates a Robust Membrane Association of a Soluble Subunit by Dimerization-If soluble GAD67 associates with membranes by heterodimerization with a membrane-anchored GAD65 subunit as suggested by Dirkx et al. (18), this would imply that membrane anchoring of only one GAD65 subunit suffices to achieve membrane association of a homo-or heterodimer. To test this possibility, we first addressed the question whether soluble mutants of GAD65, lacking the membrane-anchoring NH 2terminal region, can form dimers. As shown in Fig. 1D (lanes 4  and 5), the deletion mutants rGAD65/45A⌬1-38 and rGAD65⌬1-101, which are soluble when expressed in COS-7 cells (22,26), are detected exclusively as noncovalently linked , and membranes were prepared as described under "Experimental Procedures." Aliquots of a TX-114 membrane extract were pretreated in sample buffer containing 1% SDS without reducing agent and analyzed by SDS-PAGE. Immunoblotting was performed with the 1701 antibody that recognizes both GAD65 and GAD67. Similar results were obtained for the cytosol fractions. D, wild-type rGAD65, rGAD65/45A⌬1-38, and rGAD65⌬1-101 form noncovalently linked dimers in native conditions. Lanes 1 and 2, cytosol and membrane fractions of COS-7 cells expressing wild-type rGAD65 prepared in the presence of 5 mM NEM. Lanes 3-5, TX-114 extract of COS-7 cells expressing wild-type rGAD65, rGAD65/ 45A⌬1-38, or rGAD65⌬1-101. Aliquots of cytosol fraction (lane 1), membrane fraction (lane 2), or TX-114 extract (lanes 3-5) were boiled for 3 min in nonreducing and nondenaturing (native) sample buffer and analyzed by electrophoresis on a native 10% polyacrylamide gel followed by immunoblotting with the 1701 antibody. dimers in native conditions. Thus, the NH 2 -terminal 101 amino acids are not involved in the dimerization of GAD65.
We next addressed the question whether wild-type rGAD65 and the rGAD65/45A⌬1-38 mutant can form heterodimers when co-expressed in COS-7 cells. As shown in Fig. 2A, a fraction of rGAD65/45A⌬1-38 protein was co-immunoprecipitated with wild-type rGAD65 from lysates of COS-7 cells expressing both proteins using an antibody that only recognizes wild-type rGAD65, demonstrating that the two proteins can form a heterodimer. To assess whether the heterodimer can associate with membranes, subcellular fractions of COS-7 cells expressing one or both proteins were prepared in the presence of alkylating reagents to prevent formation of nonspecific disulfide bridges following lyses of cells. Densitometric analyses of immunoblots (a representative immunoblot is shown in Fig.  2B) revealed that while the rGAD65/45A⌬1-38 protein expressed alone is exclusively localized to the cytosol and membrane wash fractions, approximately 26% of the mutant protein becomes localized to membrane fractions in the presence of wild-type rGAD65. Thus, the heterodimer between the rGAD65/45A⌬1-38 protein and wild-type rGAD65 can become membrane-anchored.
These results show that a firm membrane anchoring of a dimer only requires tethering of one GAD65 subunit to membranes. We have previously shown that the two subunits in a GAD65 dimer undergo distinct post-translational modifications and that only one is phosphorylated in membranes (25). Thus, the two subunits in the GAD65 dimer are not identical and probably serve distinct roles in regulation of membrane trafficking of the protein.
A Pool of Rat Brain GAD67 Forms a Noncovalently Linked Dimer with GAD65 in the Cytosol and Membrane Fractions, and Dimerization Does Not Involve the NH 2 -terminal Region of GAD65-The formation of heterodimers between GAD65 and GAD67 (18,19) was suggested to involve the NH 2 -terminal region of GAD65 (18). Thus, dimerization with and membrane tethering of soluble GAD67 was suggested to be mediated by the same region of GAD65 (18).
Dirkx et al. (18) showed that heterodimerization with GAD67 does not involve formation of nonspecific disulfide bridges through cysteines in the NH 2 -terminal region of GAD65. However, the possibility that a population of GAD65/GAD67 heterodimers detected in vitro is formed by nonspecific formation of disulfide bridges involving other regions of GAD65 was not addressed. We therefore studied the heterodimerization of GAD65 and GAD67 in the presence of 5 mM NEM during lysis and subcellular fractionation of cells and subsequent immunoprecipitation to prevent the formation of nonspecific disulfide bridges between any parts of the two proteins.
GAD67 was detected in both cytosol and membrane fractions of rat brain (Fig. 3A, lanes 1 and 5), confirming that in neurons both GAD65 and GAD67 are associated with membranes (18,19). Furthermore, GAD67 was recovered in immunoprecipitates with a GAD65-specific antibody, GAD6, in both cytosol and membrane fractions, demonstrating that the two proteins specifically associate into dimers (Fig. 3A, lanes 2-4 and 6 -8). Similar results were obtained using the anti-GAD65-specific monoclonal antibody FL-17, which is directed against the first 8 amino acids of GAD65 (data not shown). The identity of the two isoforms in the Western blot (Fig. 3A) was confirmed by stripping bound antibodies and reprobing the blot first with the anti-GAD67-specific antibody, K2, and then with the anti-GAD65-specific antibody, FL-27 (data not shown).
To study the role of the NH 2 -terminal region of GAD65 in the formation of heterodimers, we tested whether the NH 2 -terminal deletion mutants of this isoform can associate with GAD67. The rGAD65/45A⌬1-38 and rGAD65⌬1-101 mutants were each coexpressed with mGAD67 in COS-7 cells and immunoprecipitated with the GAD65-specific monoclonal antibody GAD6. As shown in Fig. 3B, mGAD67 was co-precipitated with both rGAD65/45A⌬1-38 (lane 6) and rGAD65⌬1-101 (lane 10), demonstrating that the NH 2 -terminal 101 amino acids of GAD65 are not involved in the heterodimerization with GAD67. Thus, hetero-/homodimerization and membrane anchoring are mediated by distinct GAD65 regions.
GAD67 Is Associated with Membrane Fractions Independent of GAD65-The results shown above are consistent with an ability of GAD65 to mediate membrane anchoring of GAD67 in a heterodimer. Densitometric analyses of total GAD67 and GAD67 in a complex with GAD65 (GAD6 immunoprecipitate) obtained from three independent rat brain extracts (a representative immunoblot is shown in Fig. 3A), however, indicate that while the ratio between GAD65 and GAD67 is about 1:1 in membrane fractions prepared from rat cerebellum/cerebral cortex (Fig. 3A, lanes 1 and 5), the ratio in immunoprecipitates of GAD65 from those fractions (Fig. 3A, lanes 4 and 8) is approximately 3:1. Similar ratios between GAD65 and GAD67 were obtained for membrane fractions prepared from three extracts

FIG. 2. A fraction of the soluble rGAD65/45A⌬1-38 mutant is localized to membranes by heterodimerization with wild-type rGAD65 in COS-7 cells.
A, association of rGAD65/45A⌬1-38 with wild-type rGAD65. Immunoblotting analysis of wild-type rGAD65 and rGAD65/45A⌬1-38 in COS-7 cell extracts (lanes 1, 3, and 5) and in immunoprecipitates (IP, lanes 2, 4, and 6) with the monoclonal antibody FL-17, which recognizes wild-type rGAD65 but not the deletion mutant. Proteins were resolved by SDS-PAGE and then subjected to Western blotting and immunostaining using the 1701 antibody. A fraction of rGAD65/45A⌬1-38 is detected in immunoprecipitates of wild-type rGAD65. B, analyses of subcellular distribution of rGAD65/45A⌬1-38 in the presence and absence of wild-type rGAD65. The cytosol proteins, proteins washed from membranes, and proteins extracted from membranes of COS-7 cells expressing the wild-type rGAD65, the mutant rGAD65/45A⌬1-38, or both proteins were boiled for 3 min with SDSsample buffer and then analyzed by 10% SDS-PAGE followed by immunoblotting with the 1701 antibody. rGAD65/45A⌬1-38 is only detected in the cytosol and membrane wash fractions in the absence of wild-type rGAD65, but it is membrane-associated in the presence of wild-type rGAD65. of mouse cerebellum/cerebral cortex (a representative blot is shown in Fig. 4, lanes 1 and 3). These results suggest that approximately 33% of GAD67 in membrane fractions of rat and mouse brain is present as a heterodimer with GAD65, consistent with results obtained with rat cerebellar membranes in the absence of alkylating reagents (19). Thus, the majority of GAD67 appears to reside in membranes as a homodimer independent of dimerization with GAD65, suggesting that GAD67 can become membrane-anchored by a mechanism distinct from dimerization with GAD65.
To clarify whether GAD67 is strictly dependent on GAD65 for its localization to membranes, the subcellular distribution of GAD67 was analyzed in the brains of GAD65 knock-out mice (GAD65Ϫ/Ϫ). As shown in Fig. 4, GAD67 was recovered in the extensively washed membrane fractions prepared from both GAD65-deficient mice (lane 4) and their wild-type littermates (lane 3). A similar recovery of GAD67 was observed in membranes after three successive washes with either hypotonic Hepes buffer (not shown), isotonic Hepes buffer containing 150 mM NaCl (not shown), or hypertonic Hepes buffer containing 500 mM NaCl (Fig. 4). The distribution of GAD67 between cytosol and membrane fractions from wild-type mice (2.3:1) was not significantly different for GAD65-deficient mice in all conditions of membrane washing. Thus, GAD67 associates with high avidity to membranes independent of GAD65.
GAD67 is hydrophilic in rat islets and transfected fibroblasts (20,22,23). To test whether the membrane association of GAD67 is mediated by a specific hydrophobic modification of the protein in neurons, we assessed the ability of GAD67 in cytosol and membrane fractions of mouse brain to partition into the detergent phase following a temperature-induced Triton X-114 phase separation. Whereas a portion of GAD67 in both cytosolic and membrane fractions prepared from wild-type (GAD65ϩ/ϩ) mouse brains partitioned into the detergent phase (Fig. 5, lanes 2 and 6), all GAD67 remained in the aqueous phase in the absence of GAD65 (Fig. 5, lanes 3, 7, and  11), suggesting (i) that the partition of GAD67 into the detergent phase is mediated by its association with GAD65 in a heterodimer and (ii) that GAD67 is not by itself hydrophobic. Densitometric analysis of the immunoblot (Fig. 5, lanes 2 and  6) demonstrates that the ratio between GAD65 and GAD67 is approximately 3:1 and 2.6:1 in the detergent phase of the cytosolic and membrane fractions, respectively. Since these ratios are similar to that obtained for immunoprecipitates of FIG. 4. GAD67 association with membranes in mouse brain is independent of GAD65. Immunoblotting analysis of cytosol and membrane fractions prepared from the brain (cerebrum and cerebellum) of wild-type (GAD65ϩ/ϩ) and GAD65 knockout (GAD65Ϫ/Ϫ) mice. Before extraction with 1% TX-114, membranes were washed three times by incubation on ice for 1 h with HMAP buffer containing 0.5 M NaCl. Proteins were resolved by SDS-PAGE. Immunoblotting was performed using the 1701 antibody.
FIG. 3. GAD67 forms noncovalently linked heterodimers with GAD65, and the association does not involve the NH 2 -terminal region of GAD65. A, GAD67 is associated with GAD65 in cytosol and membrane fractions of rat brain. Immunoblotting analysis of immunoprecipitates of cytosol and membrane fractions of rat brain using the GAD65specific monoclonal antibody GAD6 is shown. Proteins were resolved by SDS-PAGE. Immunoblotting was performed with the 1701 antibody, which recognizes both GAD65 and GAD67. IP, immunoprecipitation. GAD67 is co-precipitated with GAD65 in both cytosol and membrane fractions. B, heterodimerization does not involve the NH 2 -terminal region of GAD65. Immunoblotting analysis of immunoprecipitates of TX-114 extracts of COS-7 cells expressing the deletion mutants rGAD65/45A⌬1-38 or rGAD65⌬1-101 using the GAD6 antibody is shown. Proteins were resolved by SDS-PAGE. Immunoblotting was performed with the 1701 antibody. GAD67 is co-precipitated with both rGAD65/45A⌬1-38 (lane 6) and rGAD65⌬1-101 (lane 10).
GAD65 from cytosolic and membrane fractions of rat brain (Fig. 3A, lanes 4 and 8), we conclude that association with GAD65 accounts for approximately all GAD67 found in the detergent phase of the cytosol and membrane fractions prepared from wild-type mouse brain. These results suggest that the association of GAD67 to membranes in the GAD65 knockout mouse brain is not mediated by its own hydrophobicity but rather by interaction with membrane components that are distinct from GAD65 and do not confer hydrophobicity upon GAD67.
We next addressed the question whether GAD67 can mediate membrane association of GAD65 mutants (rGAD65/ 45A⌬1-38 and rGAD65⌬1-101) that cannot by themselves localize to membranes. Membrane and cytosol fractions of COS-7 cells coexpressing these mutants together with mGAD67 were subjected to immunoprecipitation with the GAD65-specific antibody GAD6. In these experiments, the heterodimers between GAD67 and the soluble mutant subunits were only detected in the cytosol fraction (results not shown). Therefore, although GAD67 can associate with membranes, it cannot mediate membrane association of a soluble GAD65 subunit.
GAD67 Localization to Nerve Terminals in Primary Cultures of Neurons Is Independent of GAD65-It has been suggested that the association of GAD67 with synaptic vesicles from rat brain and its localization to nerve terminals of GABA-ergic neurons, where synaptic vesicles are clustered, is mediated by its interaction with GAD65 (18,36). However, GAD67 was also found to localize to the terminals of dentate granule cells of the hippocampus, which contain little or no GAD65 (37). To examine the role of GAD65 in targeting GAD67 to nerve terminals, we studied the localization of GAD67 in cultured rat hippocampal neurons and mouse cortical neurons prepared from wildtype embryos, where GAD67 is expressed together with GAD65, and in cultured cortical neurons prepared from GAD65 deficient mouse embryos, where GAD67 is expressed alone.
Double immunostaining of 2-week-old primary cultured rat hippocampal neurons with anti-GAD67 and anti-GAD65 antibodies revealed that in addition to its diffuse distribution in the cell body and proximal dendrites, a pool of GAD67 co-localized with GAD65 in the perinuclear region and in punctate structures (Fig. 6). The punctate structure localization of GAD65 and GAD67 was demonstrated to correspond to axon terminals by triple labeling with the antibodies directed against the synaptic vesicle protein SV2 (Fig. 7, white color). Immunostaining of 2-week-old primary cultured cortical neurons prepared from homozygous mutant mouse embryos (GAD65Ϫ/Ϫ) and their heterozygous littermates (GAD65ϩ/Ϫ), demonstrated that GAD67 is localized to nerve terminals both in the presence (Fig. 8A) and absence of GAD65 (Fig. 8B).
We also performed immunostaining of primary cultures of cortical neurons obtained from homozygous mutant GAD67Ϫ/Ϫ mouse embryos and their wild-type GAD67ϩ/ϩ littermates. In those cultures, GAD65 was localized to nerve terminals in the presence and absence of GAD67 (data not shown). Taken together, these results indicate that although the nerve terminals of neurons obtained from wild-type mouse embryos were always positive for both GAD67 and GAD65, the two isoforms localize independently to nerve terminals. FIG. 5. Membrane-associated GAD67 is hydrophilic, but heterodimerization with GAD65 can mediate its partition into the TX-114 detergent phase. Immunoblotting analysis of TX-114 detergent and aqueous phases of cytosol, membrane extract, and membrane wash fractions prepared from the brain of wild-type (ϩ/ϩ) and GAD65 knock-out (Ϫ/Ϫ) mice. Proteins were resolved by SDS-PAGE. Immunoblotting was performed using the 1701 antibody. GAD67 is hydrophilic but partitions into the TX-114 phase by association with GAD65.
FIG. 6. Co-localization of GAD67 and GAD65 in rat hippocampal neurons in culture. Double immunofluorescence analysis of GAD65 and GAD67 in rat hippocampal neurons using a mixture of GAD65-specific human monoclonal antibodies MICA 2, 3, and 6 (red fluorescence) and the GAD67-specific rabbit polyclonal antibody K2 (green fluorescence). GAD67 is evenly distributed in the cell body but also co-localizes with GAD65 in the perinuclear region and in punctate structures (orange color), representing boutons in nerve terminals.

DISCUSSION
Targeting to Nerve Terminals and Membrane Anchoring Is an Intrinsic Property of GAD67-The evolution of two different isoforms of glutamic acid decarboxylase in mammals provides a fundamental aspect of regulation of GABA-ergic neurotransmission. This is best evidenced by the distinct phenotypes of GAD65Ϫ/Ϫ and GAD67Ϫ/Ϫ mice. GAD67 provides Ͼ90% of brain GABA, and GAD67 deficiency results in neonatal death (38,39). In contrast, GAD65Ϫ/Ϫ mice are viable and have near normal GABA levels in brain but develop spontaneous seizures and are deficient in the fine tuning of inhibitory neurotransmission involved in responses to a number of stimuli (11,40,41). Both forms are synthesized as soluble hydrophilic molecules (20). Based on experiments in pancreatic ␤ cells, hydrophobic modifications and anchoring to the membrane of synaptic vesicles in neurons and synaptic-like microvesicles in pancreatic ␤ cells was described to be an exclusive characteristic of GAD65 (20 -22). Thus, targeting of GAD67 to membrane compartments was attributed to heterodimerization with GAD65 (18,19). We and others have proposed that the anchoring of GAD65 to the membrane of synaptic vesicles distinguishes it from GAD67 and provides a mechanism by which GAD65-generated GABA is more rapidly available for secretion (26). The results presented here, however, show that targeting to nerve terminals and anchoring to membranes are also intrinsic properties of GAD67 and occur independent of an association with GAD65. Thus, the two isoforms share properties previously attributed to GAD65. It will be important to establish whether the dynamics, location, and control of membrane anchoring differ for the two isoforms and how these properties translate to the differences in their function.
GAD65 and GAD67 Form Specific Noncovalently Linked Heterodimers-The native forms of both GAD65 and GAD67 are noncovalently linked homodimers (Ref. 19 and results presented here). However, because of the large number of cysteines, each isoform has a tendency to form nonspecific disul-fide-linked dimers and multimers in vitro (Ref. 34 and results presented here). In view of this phenomenon, it was important to establish whether the formation of heterodimers of the two forms (18,19) was specific or due to an in vitro formation of disulfide bridges. Although Dirkx et al. (18) presented evidence that -SH groups in the NH 2 -terminal region of GAD65 were not involved in the dimerization with GAD67, a role of other sulfhydryl groups was not assessed. In the present study, the tendency of recombinant GAD65 and endogenous GAD65 and GAD67 to form disulfide-linked multimers was almost completely prevented by blocking the reactive cysteines in both isoforms with the alkylating reagents NEM or iodoacetamide. Under alkylating conditions, a significant fraction of GAD65 and GAD67 was still detected as heterodimers, demonstrating that the association between the two isoforms is specific and does not require nonspecific formation of disulfides. We therefore conclude that heterodimers of GAD65/GAD67 are formed in vivo.
Homo-and Heterodimerization of GAD65 Does Not Involve the NH 2 -terminal Membrane-anchoring Region-Molecular modeling studies have suggested that the middle domain of each subunit is involved in GAD65 dimerization (16,17). Since there are no proteins of known structure that share sequence homology with GAD65 in the first 100 amino acids at the NH 2 terminus of GAD65 (16), a structure prediction is not available for this region. The results presented here show that the GAD65⌬1-101 mutant is exclusively detected as a dimer in nondenaturing gels and therefore provide the first evidence that the NH 2 -terminal region of GAD65 is not important for homodimerization. Thus, the predicted interactions of several residues in the middle domain of one subunit with PLP in the active site of the second subunit may represent the main bonds between the subunits (17).
Furthermore, the GAD65⌬1-101 mutant forms a heterodimer with GAD67, demonstrating that the NH 2 -terminal region of GAD65 is not required for heterodimerization. Dirkx et al. (18) linked the first 83 amino acids of GAD65 to ␤-galactosidase and suggested based on immunofluorescence studies that this protein could mediate localization of GAD67 to the Golgi complex region when the two proteins were co-expressed in CHO cells. Based on this indirect evidence, it was concluded that the NH 2 -terminal region of GAD65 mediates heterodimerization with GAD67. However, as demonstrated in this study, GAD67 can associate with membranes independent of GAD65. It is, therefore, possible that the localization of GAD67 to the Golgi region observed in CHO cells (18) occurred independent of the GAD65-(1-83)/␤-gal protein.
The first 101 amino acids of GAD65 are demonstrably not necessary for heterodimerization. Therefore, in light of the high homology between GAD65 and GAD67 in the remainder of the molecule, we propose that the structural interactions involved in heterodimerization may be similar to those for homodimerization (19).
What Is the Role of GAD65/GAD67 Heterodimerization?-The results presented here establish that the two enzymes GAD65 and GAD67 specifically form noncovalently linked heterodimers in addition to homodimers and that each homodimer associates with membranes. Firm membrane association of a dimer can be achieved by only one membrane anchoring-competent GAD65 subunit. This infers that a heterodimer between a hydrophilic GAD67 subunit and a membrane-anchoring GAD65 subunit can localize to membranes. In contrast, a heterodimer between a soluble GAD65 subunit and a membrane anchoring-competent GAD67 subunit does not anchor to membranes. The realization that GAD67 can mediate its own association to membranes but cannot do so for soluble GAD65 FIG. 7. Co-localization of GAD65 and GAD67 with SV2 in nerve terminals of rat hippocampal neurons in culture. Triple immunofluorescence analyses are shown of neurons prepared from a rat hippocampus and stained for GAD65 (MICA 2, 3, and 6, red fluorescence), GAD67 (K2 antibody, green fluorescence), and the synaptic vesicle marker SV2 (anti-SV2 mouse monoclonal antibody, blue fluorescence). GAD65 and GAD67 immunoreactivities precisely coincide with that of SV2 (white color), consistent with their accumulation at nerve terminals of GABA-ergic neurons.
implicates a homodimer-dependent association that is distinct from that of GAD65. We suggest that the GAD67/GAD67 homodimer has a distinct microlocalization within membrane compartments of nerve terminals to the GAD65/GAD65 and GAD65/GAD67 dimers and that this form can therefore impart different GABA secretion responses. FIG. 8. GAD67 localization to axon terminals of mouse cortical neurons in culture is independent of GAD65 expression. A, immunofluorescence analysis of primary cultures of cortical neurons prepared from the brains of 18day embryonic heterozygous (GAD65ϩ/Ϫ) mice, which express similar levels of GAD65 as wild-type mice (11). After 2 weeks in culture, the neurons were double immunostained for SV2 (green fluorescence) and GAD67 (red fluorescence). B, immunofluorescence analysis of cortical neurons prepared from the brains of 18day embryonic GAD65Ϫ/Ϫ mice and double immunostained for SV2 (green fluorescence) and GAD67 (red fluorescence). No staining for GAD65 can be observed in those mice (Ref. 11 and data not shown). GAD67 is targeted to nerve terminals (axons) in the absence of GAD65.