Polarized Expression of GABA Transporters in Madin-Darby Canine Kidney Cells and Cultured Hippocampal Neurons*

At least three high affinity Na (cid:49) - and Cl (cid:50) -dependent (cid:103) -aminobutyric acid (GABA) transporters are known to exist in the rat and mouse brain. These transporters share 50–65% amino acid sequence identity with the kidney betaine transporter which also transports GABA but with lower affinity. The betaine transporter (BGT) is expressed on the basolateral surface of polarized Madin- Darby canine kidney (MDCK) cells. Recent evidence suggests that the signals and mechanisms involved in membrane protein sorting share many functional char-acteristics in polarized neurons and epithelial cells. It was previously shown that the rat GABA transporter GAT-1 is located in the presynaptic membrane of axons where it plays a role in terminating GABAergic neuro-transmission. When expressed in MDCK cells by trans- fection, GAT-1 was sorted to the apical membrane. In this report, we have localized the other two GABA trans- porters, GAT-2 and GAT-3, in transfected MDCK cells by GABA uptake, immunofluorescence, and cell surface bi- otinylation. GAT-3, like GAT-1, localized to the apical membrane of MDCK cells while GAT-2, like BGT, local- ized to the basolateral membrane. We have also expressed BGT in low density cultures of hippocampal neurons by microinjection and immunolocalized it to the dendrites. The distribution of GAT-3 in these neu- rons after transfection was axonal as well as somatodendritic. These results indicate that highly homologous

At least three high affinity Na ؉ -and Cl ؊ -dependent ␥-aminobutyric acid (GABA) transporters are known to exist in the rat and mouse brain. These transporters share 50 -65% amino acid sequence identity with the kidney betaine transporter which also transports GABA but with lower affinity. The betaine transporter (BGT) is expressed on the basolateral surface of polarized Madin-Darby canine kidney (MDCK) cells. Recent evidence suggests that the signals and mechanisms involved in membrane protein sorting share many functional characteristics in polarized neurons and epithelial cells. It was previously shown that the rat GABA transporter GAT-1 is located in the presynaptic membrane of axons where it plays a role in terminating GABAergic neurotransmission. When expressed in MDCK cells by transfection, GAT-1 was sorted to the apical membrane. In this report, we have localized the other two GABA transporters, GAT-2 and GAT-3, in transfected MDCK cells by GABA uptake, immunofluorescence, and cell surface biotinylation. GAT-3, like GAT-1, localized to the apical membrane of MDCK cells while GAT-2, like BGT, localized to the basolateral membrane. We have also expressed BGT in low density cultures of hippocampal neurons by microinjection and immunolocalized it to the dendrites. The distribution of GAT-3 in these neurons after transfection was axonal as well as somatodendritic. These results indicate that highly homologous subtypes of GABA transporters are sorted differently when expressed in epithelial cells or neurons and suggest that these two cell types share the capacity to distinguish among these isoforms and target them to distinct destinations.
Synaptic transmission by aminergic neurons is terminated through the uptake of the neurotransmitter by specific sodium and chloride dependent cotransporters located in the presynaptic membranes of neurons (1,2). Following transport into the cell, the neurotransmitters accumulate in synaptic vesicles and are re-released during the next neurotransmission. GABA 1 is an amino acid which serves as the major inhibitory neurotransmitter in the vertebrate central nervous system (3,4). The cell surface transporter responsible for GABA re-uptake was cloned (5) and subsequently proved to be the first member of a large gene family. Other members of this family include the transporters which function in the presynaptic uptake of the biogenic amines norepinephrine (6), serotonin (7,8), and dopamine (9 -11), as well as the transporters for the amino acids taurine (12,13), glycine (14 -16), and proline (17). These transporters are proposed to span the membrane 12 times, with both amino and carboxyl termini in the cytoplasm. Another family member, the Na ϩ -and Cl Ϫ -dependent betaine transporter, has been cloned from MDCK cells (18) and human brain (19). The canine betaine transporter is a basolateral membrane protein which protects cells in the renal medulla from hypertonicity by mediating the uptake of the osmolyte betaine (20). GABA is also a substrate for the betaine transporter. In fact, the transporter's affinity for GABA (K m ϳ 100 M) is higher than its affinity for betaine (K m ϳ 500 M). More recently, the cDNAs for several subtypes of GABA transporters have been isolated from mouse and rat brain, bringing to four the number of GABA transporter isoforms which have been identified (21)(22)(23). Two of the high affinity (K m ϳ 10 M) rat GABA transporters, GAT-2 and GAT-3 (also known as GAT-B), share higher amino acid identity (68% and 65%, respectively) with the betaine transporter than with GAT-1 (52% amino acid identity).
Recently, we have examined the sorting behaviors of members of this family expressed by transfection in polarized epithelial cells (24). The plasma membrane of an epithelial cell is divided into an apical membrane, which frequently faces a lumen, and a basolateral membrane which is in contact with the extracellular fluid space. The protein and lipid compositions of these two membrane domains are quite different (25,26). Neurons are also polarized cells whose plasma membranes can be divided into two domains: axons (including termini) and the somatodendritic surfaces (composed of the cell body and the dendrites). Axons are thin and usually extend great distances from the cell body while dendrites are thick at the base, tapered and shorter in length. The axonal and dendritic processes must be biochemically distinct in order for them to receive and transmit information in a highly regulated manner. Very little is known about the mechanism of generating and maintaining polarity in neurons. Dotti and Simons (27) proposed that the mechanisms involved in sorting proteins to the axonal and somatodendritic membranes are similar to those required to sort proteins to the apical and basolateral membranes, respectively, of polarized epithelial cells. When cultured hippocampal neurons were infected with the influenza fowl plague virus or vesicular stomatitus virus, these virus' major glycoproteins were targeted to distinct membrane domains just as they are in infected epithelial cells. The influenza hemagglutinin (HA), which is targeted to the apical membrane in MDCK cells, was predominantly sorted to the axon. In contrast, the vesicular stomatitus virus G protein, which accumulates in the MDCK basolateral membrane, was sorted to the dendrites. Furthermore, Thy-1, a glycophosphorylinositol-linked protein, was sorted to the axon in hippocampal neurons (28) whereas glycophosphorylinositol-linked proteins are generally present in the apical membranes of epithelial cells. The small GTP-binding protein rab8 (29) and the transferrin receptor (30) are endogenously expressed in the basolateral compartment of MDCK cells and in the somatodendritic domain of neurons.
In order to examine this model further, Pietrini et al. (24) studied the sorting behavior of the neuronal GABA transporter (GAT-1) in both neuronal and epithelial cells. GAT-1 was localized exclusively to the axons of hippocampal neurons in culture (24) and in situ (31) and to the apical membranes of transfected MDCK cells. The betaine transporter, expressed by transfection, was basolateral in MDCK cells, in keeping with previous studies on the polarity of the MDCK cell's endogenous complement of betaine transporters (20). To characterize further the nature and position of the sorting information which mediates the differential targeting of these proteins, we have expressed cDNAs encoding the highly homologous GAT-2 and GAT-3 transporters in MDCK cells. We find that in stably transfected MDCK cells, GAT-2 is expressed predominantly in the basolateral membrane while GAT-3 is expressed in the apical membrane. We have also expressed the basolateral betaine transporter protein in hippocampal neurons by microinjection and found that it accumulates in the dendrites. GAT-3, on the other hand, was found in both axons and the somatodendritic membrane. These results indicate that highly homologous subtypes of GABA transporters are distributed differently when expressed in epithelial cells and suggest that these transporters require distinct subcellular distributions in order to subserve their physiologic functions. Moreover, the apical/ axonal and basolateral/dendritic sorting model proposed by Dotti and Simons (27) is consistent with the behaviors manifest by this group of membrane proteins.

EXPERIMENTAL PROCEDURES
Materials-Rabbit antisera R22 and R23, raised against amino acids 189 -206 (VVEFWERNMHQMTDGLDK) and amino acids 270 -288 (FYITPNFRKLSDSEVWLDA), respectively, of the rat GABA transporter (GAT-1) were a generous gift from L. Edelmann and R. Jahn, Yale University, New Haven, CT. A polyclonal antiserum (anti-670) directed against the carboxyl terminus of GAT-3 (amino acids 592-608) was a generous gift from B. Kanner, The Hebrew University, Jerusalem. Anti-MAP2 polyclonal antibody was a generous gift from R. Vallee, Worcester Foundation for Experimental Biology, Worcester, MA. Antibodies to hemagglutinin and synapsin I were from L. Roman and P. De Camilli, respectively, both of Yale University. Production of the monoclonal antibody (6H) against the ␣ 1 subunit of the Na,K-ATPase has been described (32). pCB6 and pCB6-HA were kindly provided by M. Roth, University of Texas Southwestern Medical Center, Dallas, TX. Anti-MAP2 monoclonal antibody and enzymes were purchased from Boehringer Mannheim. All chemicals, unless noted otherwise, were purchased from J. T. Baker, Phillipsburg, NJ.
Tissue Culture-MDCK cells were maintained in complete medium consisting of minimal essential medium (MEM), 10% fetal bovine serum (Sigma), 2 mM L-glutamine, 50 units/ml penicillin, 50 g/ml streptomycin (Life Technologies, Inc., Grand Island, NY). Stably transfected MDCK cells were maintained in the above medium supplemented with 0.9 g/liter Geneticin (Life Technologies, Inc.). COS-1 cells were grown in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum, 2 mM L-glutamine, 50 units/ml penicillin, 50 g/ml streptomycin. Cells were grown in a 37°C humidified incubator with 5% CO 2 and passaged twice weekly through exposure to 0.05% trypsin, 0.5 mM EDTA (Life Technologies, Inc.). Hippocampal neurons were isolated from 17 day rat embryos, plated on coverslips at 3000 cells/cm 2 , and co-cultured with glia as described previously (33). One-half of the conditioned medium was replaced with fresh neuronal medium (0.1% bovine albumin (Sigma), 2 mM L-glutamine, 1% HL-1 supplement (Ventrex Laboratories)) every 4 -5 days. When neurons were to be maintained for longer than 1 week, cytosine arabinoside was added to inhibit proliferation of non-neuronal cells.
Construction of c-Myc-tagged BGT-1-The c-Myc epitope EQKLI-SEEDL was added to the carboxyl terminus of the canine betaine transporter cDNA (BGT-1) by the polymerase chain reaction method. BGT-1 was first subcloned into the vector pACYC184 (New England BioLabs) which lacks BsaI and StuI restriction sites. This plasmid pACYC184-BGT1 was used as a template to synthesize a 0.8-kilobase fragment with the primers GTTGGGGAGAAGGAGACCCACTTGGAG-CAGAAGCTGATCTCTGAGGAAGACCTCTAGGATGTGGCCAGC and TTGGAAGGCCTGTGA. The polymerase chain reaction product was cloned into the TA cloning vector pCRII (Invitrogen, San Diego, CA), digested with BsaI and StuI, and ligated to the BsaI-StuI sites of pACYC184-BGT1. The modified BGT-1 (BGT * ) was finally cloned into the MluI-BamHI sites of the expression vector pCB6. Plasmid DNA was amplified in competent Escherichia coli DH5␣ cells (gift of L. Roman) and purified using Qiagen columns (Qiagen, Chatsworth, CA). The modified region of pCB6-BGT * was confirmed by dideoxy DNA sequencing.
Stable Expression of cDNAs in MDCK Cells-GAT-2 cDNA (kindly provided by L. A. Borden and K. E. Smith, Synaptic Pharmaceutical, Paramus, NJ) was subcloned into the HindIII and XbaI sites of the mammalian expression vector pCB6. GAT-3 was subcloned into the ClaI site of pCB6. MDCK cells were transfected in the presence of 0.1 mM chloroquine using the calcium phosphate method as described elsewhere (34 -36). After about 2 weeks of selection in 0.9 g/liter Geneticin, colonies of cells were transferred to 24-well plates using 8-mm cloning cylinders (Bellco Glass, Vineland, NJ). Subconfluent cultures were screened by their ability to take up [ 3 H]GABA (see assay below).
Transient Expression of cDNAs in COS-1 Cells-Cells were transfected using DEAE-dextran as described (37). Cells (2 ϫ 10 5 per 35-mm dish) were plated on glass coverslips the night before transfection and washed twice in TBS (137 mM NaCl, 5 mM KCl, 0.7 mM CaCl 2 , 0.5 mM MgCl 2 , 0.6 mM Na 2 HPO 4 , 25 mM Tris-HCl, pH 7.4) at 37°C. To each dish, 0.5 ml of TBS containing 5 g of DNA and 0.5 mg/ml DEAEdextran (Pharmacia Biotech Inc.) were added for 40 min at 37°C. The solution was discarded and the cells were incubated in 1 ml of DMEM containing 10% fetal bovine serum and 0.1 mM chloroquine for 3 h at 37°C. The cells were washed twice in TBS (37°C) and incubated in complete DMEM for 48 h before measuring protein expression.

4). [ 3 H]GABA (␥-[2,3-3 H]aminobutyric acid, 30
Ci/mmol, DuPont NEN) in PBS ϩ was added to either the apical (0.2 ml) or basolateral (0.5 ml) side of the filter at a final concentration of 50 nM (0.33 Ci/ml). After 10 min at room temperature, the cells were washed three times in ice-cold PBS ϩ . Filters were excised and the cells were lysed in 200 l of 1% SDS. A 50-l aliquot of cell extract was removed for protein determination using the Pierce BCA protein assay reagent (Pierce) and bovine albumin as the standard. The remainder was counted in 3 ml of Optifluor (Packard, Downers Grove, IL). Nonspecific uptake (Ͻ10%) was determined in the presence of 0.5 mM unlabeled GABA.
Cell Surface Biotinylation-Cells were grown on 24-mm Transwell inserts (0.4 m pore, Costar) for 1 week and the medium was replaced with complete MEM (without Geneticin) containing 10 mM sodium butyrate 18 -24 h before the experiment. Steady state biotinylation was carried out at pH 9.0 as described by Gottardi et al. (38) except that the cells were incubated with NHS-biotin for 2 ϫ 20 min. Biotinylation at pH 9.0 was chosen because our laboratory has previously shown that many membrane proteins are more efficiently biotinylated at pH 9.0. Performing the biotinylation at pH 7.5 instead of at pH 9.0 did not alter the results. The biotinylated proteins were isolated with 100 l of packed immobilized streptavidin and analyzed by SDS-polyacrylamide gel electrophoresis (39) and Western blotting (40) as described previously (34,38). The blot was incubated with R22 (1:200) or 6H (1:500) primary antibodies followed by peroxidase-conjugated secondary antibodies (Sigma) and developed with Enhanced Chemiluminescence reagent (Amersham).
Microinjection-DNA solutions (0.5-1 g/l H 2 O) were passed through a 0.22-m Spin-X membrane (Costar) and verified on an ethidium bromide containing agarose gel. Lysine-fixable fluorescein/ dextran (M r 10,000; Molecular Probes) was added to the DNA at a final concentration of 10 mg/ml, 2.5 mM Tris-HCl, pH 7.5. Injection pipettes were pulled from Narishige model GD-1 microcapillary tubes to an approximate diameter of 0.5 m. The pipette was filled through the larger opening with 2.5 l of solution using Seque/pro pipette tips (Bio-Rad). Each coverslip with neurons attached was transferred, cell side up, to a 35-mm dish containing 1.5 ml of fresh neuronal medium supplemented with 20 mM NaHepes, pH 7.4. The dish was placed on the stage of an inverted microscope (Olympus) and a filled pipette was positioned vertically over the cells. The cells were injected manually using a micromanipulator (Narishige). Injection was indicated by swelling of the nucleus and change in light reflection. Fifty to one hundred cells were injected per coverslip over a period of 15 min at room temperature. Dishes were kept in a humidified 37°C incubator until the injections were completed (up to 3 h). The coverslips were returned, cell side down, to glial cultures in conditioned medium for 48 h.
Immunofluorescence-MDCK cells were grown on 9-mm Falcon cell culture inserts (0.45 m pore size Cyclopore membrane, Becton Dickinson Labware, Lincoln Park, NJ) or 24-mm Transwell inserts (0.4 m pore, Costar) and processed for indirect immunofluorescence (34). All primary and secondary antibodies and Texas Red avidin were diluted in GSDB (16% goat serum (Sigma), 0.3% Triton X-100, 0.45 M NaCl, 20 mM NaP i , pH 7.4; Ref. 30). Incubations were carried out at room temperature in a humidified, light-protected box. MDCK cells were rinsed in PBS ϩ and fixed in methanol (Ϫ20°C) for 10 min or in 3% paraformaldehyde in PBS for 25 min. After washing in PBS ϩ , the cells were incubated for 15 min in 0.3% Triton X-100, 0.1% bovine serum albumin (Sigma) in PBS ϩ (Buffer A), and blocked for 30 min in GSDB. Next, the cells were incubated with anti-GABA transporter (R22 or anti-670, diluted 1:50) and anti-Na,K-ATPase ␣ subunit (6H, 1:100) antibodies for 1 h, followed by three washes (5 min each) in Buffer A. The cells were then incubated with fluorescein isothiocyanate-conjugated anti-rabbit IgG (Sigma) and rhodamine-conjugated anti-mouse IgG (Boehringer Mannheim), both diluted 1:100, for 1 h and washed again in Buffer A. Hippocampal neurons were rinsed twice briefly in PBS ϩ (37°C), fixed in freshly prepared 4% paraformaldehyde in PBS, washed three times (5 min each) in 10 mM glycine in PBS ϩ , and once more in PBS ϩ . Cells were permeabilized in Buffer A for 15 min, blocked in GSDB for 30 min, blocked in avidin solution (Vector Laboratories, diluted 1:1 with 2ϫ GSDB) for 15 min, rinsed in PBS ϩ for 5 min, and blocked in biotin solution (diluted 1:1 with 2ϫ GSDB) for 15 min. The cells were incubated with anti-c-Myc (Oncogene Sciences, 1:50) or anti-HA (1:200) monoclonal antibodies for 1 h. Double labeling was carried out by incubation with anti-MAP2 (1:100) or synapsin (1:50) polyclonal antibodies in the same primary antibody solution. The cells were washed three times (5 min each) in Buffer A, then incubated with 10 g/ml biotinylated anti-mouse IgG (Vector) and Cy5-conjugated anti-rabbit IgG (Jackson Immunochemicals, 1:100) for 1 h. The washed cells (3 ϫ 5 min in Buffer A) were incubated in Texas Red-conjugated avidin (Molecular Probes, 1:200) for 30 min and washed again. All specimens were washed for 10 min in 5 mM NaP i , pH 7.5, before mounting. Coverslips were mounted cell side down in a freshly prepared solution of 0.1% p-phenylenediamine (Sigma), 70% glycerol, 150 mM NaCl, 10 mM NaP i , pH 7.4, or in Vectashield mounting solution (Vector Laboratories, Burlingame, CA) and sealed with nail polish. Filters were mounted cell side up and coverslipped. The cells were initially examined using a Zeiss Axiophot microscope. Immunofluorescence and phase-contrast images were photographed using TMX100 or 400 film (Eastman Kodak Co., Rochester, NY). Confocal images were generated on a Zeiss laser scanning microscope using the following excitation/emission wavelengths: 488 nm/515-560 nm (fluorescein), 568 nm/590 -640 nm (Texas Red), 647 nm/650 -710 nm (Cy5). Confocal xz sections were generated using a 0.2-m motor step. All images are the product of 8-fold line averaging. There was no significant bleed-through from the fluorescein to rhodamine/Texas Red channels since many of the cells which were strongly labeled with fluorescein had no rhodamine or Texas Red labeling.

GAT-2 and GAT-3 Are Expressed on Opposite Membrane
Domains in MDCK Cells-The GABA transport activities catalyzed by the different MDCK cell lines expressing GAT-2 or GAT-3 are shown in Fig. 1. Five positive clones were isolated from each transfection. Cells transfected with GAT-2 transported GABA 3-10 times faster when the labeled substrate was added to their basolateral surfaces as opposed to their apical surfaces. Cells transfected with GAT-3 transported GABA rapidly only when GABA was added to their apical surfaces. Therefore, these two GABA transporters are functionally expressed on opposite membranes in MDCK cells. Cell lines GAT2#70 and GAT3#3 were used in subsequent experiments.
We wished to confirm the apparent polarized distributions of these transporters using a method to localize the expressed proteins which is independent of their transport capacities since functional assays only report the distribution of the active populations. We chose, therefore, to complement the transport assay with antibody based detection methods. Two polyclonal antibodies (R22 and R23) raised against peptides from GAT-1 were tested for their ability to cross-react with GAT-2 and GAT-3. In order to express high levels of transporter for antibody characterization studies, we transiently transfected COS-1 cells with GAT-2 or GAT-3 cDNAs and measured antibody binding by indirect immunofluorescence. R22 recognized GAT-2 and GAT-3 in COS-1 cells (Fig. 2, a and b); however, the intensity of fluorescence was considerably lower for the GAT-3 transfected COS-1 cells (Fig. 2b). Antibody R23 recognized neither GAT-2 nor GAT-3 (not shown). Although cell surface labeling was clearly visible, most of the transporters were localized intracellularly, probably in the endoplasmic reticulum and Golgi apparatus. This pattern of staining is often seen in COS-1 cells induced to overexpress membrane proteins. The GAT-3 specific antibody, anti-670, labeled the surface as well as intracellular membranes of COS-1 cells expressing GAT-3 (Fig. 2c). This antibody did not react with cells expressing GAT-2 (not shown).
When stably transfected MDCK cells expressing GAT-2 were labeled with R22, a lateral staining pattern was observed (Fig. 3, A and C), similar to the pattern obtained with an antibody against the Na,K-ATPase, a known basolateral protein (Fig. 3, B and D). We discovered that MDCK cells transfected with GAT-3 did not stain with R22, so anti-670 had to be used for immunofluorescence. With the latter antibody, we observed bright staining of the apical membrane of MDCK cells transfected with GAT-3 (Fig. 3, E and G) while the Na,K-ATPase was distributed basolaterally in the same cells (Fig. 3, F  and H).
Finally, we performed cell surface biotinylation experiments to confirm the polarity of GAT-2 and GAT-3 in MDCK cells. Fig. 4A shows that GAT-2, an 85-kDa protein, was accessible to biotinylation predominantly from the basolateral side of intact MDCK cells. In contrast, GAT-3 was preferentially biotinylated from the apical side. The Na,K-ATPase was biotinylated from the basolateral surface in all three cell lines (Fig. 4B), indicating that these cells are properly polarized. These data are consistent with the transport and microscopy results. We conclude, therefore, that GAT-2 behaves as a basolateral protein when expressed by transfection in MDCK cells, whereas its homologue GAT-3 accumulates in the apical plasma membrane.
Epitope-tagged Betaine Transporter Is Basolateral in MDCK Cells-The data presented by Pietrini et al. (24) and the work presented here describe the sorting behavior of all four GABA transporters in MDCK cells. So far, however, only GAT-1 has been localized to a specific subcellular domain in neurons. We wondered whether neurons would recognize the basolateral sorting information associated with the betaine transporter (BGT) and target this protein to the dendritic membrane. In order to detect the expression of BGT by immunofluorescence after transfection, we added a c-Myc epitope to its carboxyl terminus and used an anti-c-Myc antibody to stain the cells. Expression of the Betaine Transporter in Hippocampal Neurons-We wished to determine if members of the GABA transporter family would occupy different membrane domains when synthesized in polarized neurons. The betaine transporter is not endogenously expressed in rat neurons (41). Thus, we transfected cultured hippocampal neurons by microinjecting the cDNA directly into their nuclei and visualized the expressed protein by immunofluorescence microscopy. Table I summarizes the data on the success of the injection protocol. There was considerable day to day variation in the transfection efficiency. In one experiment, 10 out of 300 injected cells were positively labeled, while in another experiment only one cell (out of 300 injected) expressed the transporter. Fig. 6 shows a neuron that has been double-labeled with antibodies to c-Myc (BGT*) and the dendritic marker, MAP2 (42). Nearly all the processes labeled with anti-c-Myc (BGT*) were also positive for MAP2. It would appear, therefore, that the betaine transporter was selectively expressed in the dendrites of hippocampal neurons. Several MAP2 negative processes (arrows in the phasecontrast micrograph) were devoid of betaine transporter, but these axons may belong to neighboring uninjected cells. To avoid this ambiguity, we co-injected BGT * cDNA with lysinefixable fluorescein/dextran. The fluorescein/dextran was small enough to pass through the nuclear membrane and label all the processes of the injected cell (Fig. 7, DEX) without affecting the biosynthesis or targeting of the proteins. In the neuron shown in Fig. 7, it is clear that the fluorescein-labeled axon (arrow in DEX) originates from the injected cell and does not express the betaine transporter (BGT*).
Some of the injected cells were double-labeled with anti-c-Myc and an antibody against the synaptic vesicle protein, synapsin (43). Synapsin was found in the axons (of both injected  GAT-3 (b and c) were fixed in methanol, permeabilized, and stained with R22 (a and b) or anti-670 (c) followed by a fluorescein isothiocyanate-coupled anti-rabbit IgG secondary antibody. Left panels, immunofluorescence images. Right panels, corresponding phase-contrast micrographs. The distribution of the GABA transporter is mainly intracellular (endoplasmic reticulum and Golgi) as indicated by the bright staining around the cell nucleus. Cell surface labeling, which is marked by uniform staining extending to the cell boundary, is seen in a and c. Bar, 25 m. and uninjected cells) and appeared as bright spots at sites of synaptic contact with the cell body and dendrites (Fig. 7, SYN). The arrowheads point to processes which express BGT * . Three of the six short BGT * -labeled processes are most likely dendrites since they do not contain synapsin. The other three processes are labeled by synapsin but they could be dendrites in contact with neighboring axons.
Expression of HA in Hippocampal Neurons by Microinjection-To demonstrate that the somatodendritic distribution of the betaine transporter is not an artifact of the microinjection method, we also microinjected neurons with the cDNA for influenza hemagglutinin (HA), a membrane protein which has previously been shown to localize to the axons of cultured hippocampal neurons infected with the influenza virus (27). Cells injected with HA cDNA had a staining pattern very different from those injected with BGT * (Fig. 8). Although the cell body and some dendrites were labeled with HA antibody, the axon was also labeled. The labeled processes often extended more than 200 m from the cell body and formed multiple loops, as is characteristic for axons of cultured neurons. Such axonal labeling was not observed in neurons microinjected with BGT * . GAT

FIG. 4. Steady state biotinylation of MDCK cells expressing GAT-2 or GAT-3.
Cells transfected with pCB6 alone, GAT-2, or GAT-3 were grown on 24-mm Transwell filters (0.4 m) and treated with 10 mM sodium butyrate 24 h before the experiment (to boost the expression of the exogenous proteins). NHS-biotin was added to either the apical (ap) or basolateral (bl) media compartments. After biotinylation, cells were solubilized and streptavidin-agarose was used to isolate biotinylated proteins which were then separated on an 8.5% polyacrylamide gel, transferred to nitrocellulose, and probed with anti-GABA transporter antibody R22 (A) or anti-Na,K-ATPase monoclonal antibody 6H (B). The positions of prestained molecular weight standards (M r ϫ 10 Ϫ3 ) are shown on the left. GAT-2 was mainly biotinylated from the basolateral surface while GAT-3 was biotinylated from the apical surface. The Na,K-ATPase was basolaterally localized in all of these cell lines (B), demonstrating that transfection does not alter these cells' characteristic membrane polarity.

FIG. 5. The c-Myc-tagged betaine transporter is expressed in the basolateral membrane of transfected MDCK cells.
En face (top) and xz section (bottom) confocal images show that the betaine transporter is restricted to the basolateral membrane. Cells were processed for immunofluorescence as for Fig. 3 except that the antibody to c-Myc was used to detect the transporter. Betaine transport activity has previously been measured at the basolateral surface of MDCK cells (20,24); this distribution is confirmed by immunolocalization of the c-Myctagged transporter. ap, apical surface; bl, basolateral surface.

TABLE I Summary of microinjection data
The following DNA constructs were introduced into hippocampal neurons by microinjection after 6 to 17 days in culture as described under "Experimental Procedures." The total number of cells injected, the number of cells containing fluorescein (FL)/dextran, and the number of cells expressing protein by immunofluorescence are listed. N/A ϭ not applicable; n ϭ number of experiments, each performed on a different day. stained with an antibody specific for GAT-1, a subset (ϳ10%) of the cells were labeled (24). To determine if GAT-3 is also endogenously expressed, hippocampal neurons were stained with the GAT-3 specific antibody, anti-670. We were unable to see any positive staining above the background. Therefore, GAT-3 cDNA was microinjected into these neurons to determine if the apical membrane sorting information in GAT-3 specifies an axonal localization. Fig. 9 shows a typical neuron expressing GAT-3 in both the axon and dendrites. This expression pattern resembles that of HA and is quite distinct from that of the somatodendritically localized BGT * . These results suggest that certain proteins (e.g. betaine transporter) are restricted to the somatodendritic membrane while others (HA and GAT-3) are expressed in both somato-dendritic and axonal membranes. Moreover, apical membrane proteins appear to be transported to axons while basolateral membrane proteins remain in the somatodendritic membrane when expressed in neurons. DISCUSSION We have previously shown that two members of the neurotransmitter transporter gene family, GAT-1 and the betaine transporter, are sorted to two different membrane domains in MDCK cells (24). It should be noted that the betaine transporter can also transport GABA but with a lower affinity. In this report, we show that the two most recently cloned GABA transporters, GAT-2 and GAT-3, are also differentially localized in MDCK cells according to both functional and biochemical criteria. GAT-3, like GAT-1, was expressed on the apical surface of transfected MDCK cells. GAT-2, like the betaine transporter, was expressed mainly on the basolateral surface of MDCK cells. The construction of chimeric proteins consisting of sequences from GAT-3 and GAT-2, which share ϳ65% sequence identity, should help us to determine the sorting signals which mediate the subcellular targeting of members of this family of proteins.
In the context of our data on sorting, it is interesting to consider the tissue specific expression of the GABA transporter isoform mRNAs by Northern blot analysis. GAT-1 and GAT-3 are expressed in brain and retina, while GAT-2 mRNA is found in liver and kidney as well as the brain and retina (21). It would appear, therefore, that both of the exclusively neuronal forms of the GABA transporter behave as apical proteins, while the isoforms endogenously expressed in epithelia are primarily basolateral. The functional significance of these non-neuronal GABA transporters is not known. Presumably, however, epithelial GAT-2 may function like the betaine transporter to import a solute from the serum. Thus, we expect that this protein is basolaterally disposed in the tissues which express it endogenously.
To study the sorting mechanism in neurons, we have developed a method of transfecting primary cultures of embryonic hippocampal neurons. Primary cultures of neurons are difficult to transfect with DEAE-dextran, calcium phosphate, or cationic lipids, methods commonly used to transfect other mammalian cells in culture. High efficiency transfection of neurons has been accomplished using Herpes Simplex or Semliki Forest viral vectors (44,45); however, there are problems associated with introducing viral components. Hippocampal neurons do not survive for more than 8 h following viral infection (45) and may be subjected to alterations of cellular physiology which could affect the polarized sorting of neuronal proteins. For example, in cultured dorsal root ganglion and spinal cord neurons, the normally somatodendritic MAP2 appears in axons after rotavirus infection (46). The introduction of foreign DNA into the nucleus of neurons by microinjection should cause little damage to the cells since the plasma membrane is expected to immediately re-seal by a calcium-dependent mechanism following puncture (47).
We expressed the renal betaine transporter by microinjection of cultured hippocampal neurons and found that this basolateral membrane protein is sorted exclusively to the somatodendritic membrane. Next, we expressed influenza hemagglutinin by microinjection and found that, unlike the betaine transporter, this protein was also delivered to axons. Finally, we expressed the GABA transporter GAT-3 in hippocampal neurons and noted that, like HA, it was transported to the axon as well as being present in the cell body and dendrites. GAT-3 would be expected to reside in presynaptic membranes if its role in the brain, like that of GAT-1, involves neurotransmitter uptake from the synaptic cleft following neu- rotransmission. The subcellular distribution of GAT-2 in neurons has yet to be determined. However, in light of its preference for the basolateral membrane when expressed in MDCK cells, we can predict that GAT-2 will be sorted like BGT, i.e. to the dendrites.
Our results are consistent with the hypothesis that apical and basolateral sorting signals are interpreted as axonal and dendritic targeting signals, respectively, by hippocampal neurons. While the betaine transporter was expressed exclusively in the somatodendritic membrane, influenza HA and GAT-3 were less polarized. The latter two proteins were found in dendrites as well as axons. Dotti and Simons (27) also found some HA in the dendrites of influenza-infected neurons. The  presence of axonal proteins in processes resembling dendrites may be explained in part by the close proximity of axons and dendrites, since axons often run along the surfaces of dendrites and may be superimposed upon these MAP2-positive processes (Fig. 8, insets). It is also possible that there may actually be expression of the HA and GAT-3 proteins in the membranes of dendrites, especially when these proteins are overexpressed in transfected neurons. It is possible, for example, that if the axonal delivery system becomes saturated, proteins with axonal sorting information would be shunted to the somatodendritic surface.
Since the Dotti and Simons (27) model has been proposed, the sorting of several epithelial and neuronal proteins have been examined in both cell types. At least two proteins do not abide by this model. The amyloid protein precursor is axonal in cultured hippocampal neurons (48) but basolateral in MDCK cells (49). The amyloid protein precursor sequence includes a tyrosine-based internalization signal. Recent evidence suggests that proteins containing the YXRF motif may be targeted to the cell's most endocytically active domain (34). In the case of MDCK cells, the preferred domain would be basolateral. Therefore, when expressed in MDCK cells, the internalization signal of amyloid protein precursor may overide the axonal/apical sorting signal and target the protein to the basolateral surface. In neurons, both axons and dendrites participate in endocytosis (50) and the internalization signal might not be a dominant sorting signal. In keeping with this hypothesis the transferrin receptor, which manifests both an autonomous basolateral sorting signal as well as an endocytosis motif, is only present in the somatodendritic membrane of cultured neurons (30). The ␣1 and ␣3 isoforms of the Na,K-ATPase are found in both the axons and dendrites of cultured hippocampal neurons (32) although these proteins are polarized to the basolateral membranes of MDCK cells. Further study of membrane protein dynamics in neurons and epithelia will be required to understand fully the similarities and differences relating sorting in these cell types.
Our data, taken together with that of Pietrini et al. (24), demonstrate that the four members of a highly homologous family of transport proteins are differentially distributed in epithelial cells and neurons. Attainment of the proper subcellular distribution almost certainly plays an important role in ensuring the correct function of these transporters in situ. It will be important, therefore, to identify the sorting information embedded in these proteins and to determine its role in their function. Once these sorting signals are determined, the techniques presented here can be employed to ascertain whether the same signals are recognized by the neuronal and epithelial sorting machinery.