Ring Finger Protein 34 (RNF34) Interacts with and Promotes γ-Aminobutyric Acid Type-A Receptor Degradation via Ubiquitination of the γ2 Subunit*

Background: The number of GABA type-A receptors (GABAARs) affects the strength of GABAergic synapses. Results: The E3 ubiquitin ligase RNF34 binds to and ubiquitinates the γ2 GABAAR subunit, tagging GABAARs for degradation. Conclusion: RNF34 regulates the number of postsynaptic GABAARs. Significance: This is the first identification of an E3 ubiquitin ligase involved in GABAAR trafficking. We have found that the large intracellular loop of the γ2 GABAA receptor (R) subunit (γ2IL) interacts with RNF34 (an E3 ubiquitin ligase), as shown by yeast two-hybrid and in vitro pulldown assays. In brain extracts, RNF34 co-immunoprecipitates with assembled GABAARs. In co-transfected HEK293 cells, RNF34 reduces the expression of the γ2 GABAAR subunit by increasing the ratio of ubiquitinated/nonubiquitinated γ2. Mutating several lysines of the γ2IL into arginines makes the γ2 subunit resistant to RNF34-induced degradation. RNF34 also reduces the expression of the γ2 subunit when α1 and β3 subunits are co-assembled with γ2. This effect is partially reversed by leupeptin or MG132, indicating that both the lysosomal and proteasomal degradation pathways are involved. Immunofluorescence of cultured hippocampal neurons shows that RNF34 forms clusters and that a subset of these clusters is associated with GABAergic synapses. This association is also observed in the intact rat brain by electron microscopy immunocytochemistry. RNF34 is not expressed until the 2nd postnatal week of rat brain development, being highly expressed in some interneurons. Overexpression of RNF34 in hippocampal neurons decreases the density of γ2 GABAAR clusters and the number of GABAergic contacts that these neurons receive. Knocking down endogenous RNF34 with shRNA leads to increased γ2 GABAAR cluster density and GABAergic innervation. The results indicate that RNF34 regulates postsynaptic γ2-GABAAR clustering and GABAergic synaptic innervation by interacting with and ubiquitinating the γ2-GABAAR subunit promoting GABAAR degradation.

Regarding GABAergic synapses, the epilepsy-related mutations A322D in the ␣1 and R177G in the ␥2 GABA A R subunits cause misfolding of these subunits, retention in the endoplasmic reticulum (ER) and subsequent ER-associated degradation (ERAD) by the ubiquitin-proteasome system (19,20). Ubiquitination of GABA A Rs is also involved in the regulation of the normal function of GABAergic synapses. Thus the ␤3 GABA A R subunit is ubiquitinated in an activity-dependent manner targeting GABA A Rs for proteasomal degradation (21). Also ubiquitination of the ␥2L GABA A R subunit targets GABA A Rs for lysosomal degradation (22). Ubiquitination of the ␤3 and ␥2 subunits occur in lysine-rich motifs within the large IL of these subunits (␤3IL and ␥2LIL), which is localized between TM3 and TM4. To the best of our knowledge, the E3 UBL(s) responsible for the ubiquitination of GABA A Rs remain to be identified.
In this communication, we show that the E3 UBL RNF34 regulates GABAergic synapse stability and postsynaptic GABA A R clustering by interacting with and ubiquitinating the ␥2 GABA A R subunit.

EXPERIMENTAL PROCEDURES
Animals-All the animal handling procedures are in compliance with the regulations of the Institutional Animal Care and Use Committee of the University of Connecticut, and the guidelines from the National Institutes of Health were followed.
Antibodies-A rabbit (Rb) polyclonal antiserum to RNF34 (anti-RNF34) was generated after immunizing with a C-terminal peptide of the rat RNF34 protein (aa 368 -381, CRQYV-VRAVHVFKS). The antigenic peptide sequence is identical in human, rat, and mouse. The synthetic peptide was covalently coupled to keyhole limpet hemocyanin via an N-terminal cysteine and injected into a New Zealand rabbit in complete Freund's adjuvant for the initial immunization and incomplete Freund's adjuvant for subsequent immunizations. Four months after the first immunization, the antiserum was collected, and the anti-RNF34 antibody was affinity-purified on immobilized antigenic peptide. In all experiments, we used the affinity-purified anti-RNF34. The specificity of the anti-RNF34 antibody was demonstrated by the following. 1) Sequence specificity of the antigenic peptide was determined from Entrez and UniProt protein databases. 2) ELISA showed binding of anti-RNF34 to the antigenic peptide. 3) Immunoblots of homogenates and various subcellular fractions from rat forebrain showed a single 42-kDa protein band corresponding to the molecular weight of RNF34, and the immunoreactivity was displaced by antigenic peptide or purified His-RNF34 (aa 195-381) fusion protein. 4) The anti-RNF34 antibody reacted in immunoblots with purified His-RNF34 (aa 195-381) bacterial fusion protein (40-kDa protein band). 5) The antibody gave a strong RNF34 immunofluorescence signal in HEK293 cells transfected with a pCAGGS-RNF34 construct compared with the low immunofluorescence signal of neighboring nontransfected cells. 6) The anti-RNF34 immunofluorescence of cultured hippocampal neurons was displaceable by antigenic peptide or purified His-RNF34-C (aa 195-381) fusion protein.
For the immunoblots in Fig. 6A only, the secondary goat anti-rabbit IgG antibody and rabbit peroxidase-antiperoxidase complexes were from MP Biomedicals (Aurora, OH). For the other immunoblots, IRDye 800CW-conjugated goat anti-rabbit or anti-mouse IgG and IRDye 680LT-conjugated goat antimouse IgG secondary antibodies were from Li-Cor Biosciences (Lincoln, NE).
For immunofluorescence in cell cultures, Texas red-, Alexa Fluor 594- (Fig. 9, only), FITC-, and aminomethylcoumarinconjugated species-specific anti-IgG secondary antibodies were raised in donkey (Jackson ImmunoResearch, West Grove, PA). For light microscopy immunocytochemistry, biotin-conjugated goat anti-Rb IgG secondary antibody and avidin-biotinhorseradish peroxidase complex were from Vector Laboratories (Burlingame, CA). For confocal microscopy of brain sections, Alexa Fluor 488-and 568 conjugated species-specific anti-IgG secondary antibodies were raised in goat (Invitrogen). For EM, the colloidal gold-labeled goat anti-mouse IgG (10 nm diameter) secondary antibody was from ICN (Irvine, CA), and the colloidal gold-labeled goat anti-Rb IgG (18 nm diameter) secondary antibody was from Jackson ImmunoResearch.
Yeast Two-hybrid (Y2H)-The detailed experimental procedure for Y2H has been described elsewhere (30,31,38). All vectors and yeast strains were from Dr. Roger Brent (University of California at San Francisco) or Origene Technologies (Rockville, MD). For the initial screening of a rat brain cDNA library, we used as bait the large IL between TM3 and TM4 of the ␥2 short GABA A R subunit (␥2sIL, aa 318 -404 of GenBank TM NP_899156). The ␥2sIL with an added stop codon at the C terminus was inserted into pEG202 and expressed as a fusion protein with the DNA-binding protein LexA at its N terminus. The same procedure for bait construction was followed for ␣1IL, ␤1IL, ␤3IL, ␥1IL, ␥2LIL, ␥3IL, and GluA3C, and various deletion constructs of the ␥2sIL. The pSH17-4 plasmid, which contains the LexA DNA binding domain, was used as the positive control. The pRHFM1 plasmid, which contains the bicoid protein bait, or the empty bait plasmid pEG202 was used as negative controls. To map the RNF34-binding site for the ␥2sIL, the full-length RNF34 and various truncated mutants with an added stop codon at the C terminus were subcloned into pJG4-5 vector. All the cloned DNAs were verified by DNA sequencing, and the protein expression for each of the constructs in yeast Saccharomyces cerevisiae EGY48 was confirmed by immunoblotting of the cell lysate with Ms anti-LexA or anti-HA mAbs.
Cloning of Full-length Rat RNF34 cDNA and Preparation of HA-tagged RNF34 Constructs-The Marathon-ready rat brain cDNA library (Clontech), containing full-length cDNA clones, was used as template in a PCR to clone the cDNA of rat RNF34 (GenBank TM NM_001004517). Two 26-base antisense primers recognizing two sequences in the 3Ј-UTR of RNF34 were used in two sequential PCRs: a 5Ј-rapid amplification of cDNA ends PCR using the 5Ј library adaptor primer AP1 and a 3Ј-UTR RNF34 antisense primer (5Ј-GGCGAGTGGCCTCCACAC-CCACACTG-3Ј) followed by a nested rapid amplification of cDNA ends PCR using the 5Ј library adaptor primer AP2 and a second 3Ј-UTR RNF34 antisense primer (5Ј-AGCCATGT-GTCCACGGTTGACGGGCC-3Ј). A 1.5-kb DNA fragment containing the 5Ј-UTR, complete coding sequence, and partial 3Ј-UTR of RNF34 was generated, purified, cloned into pCR-XL-TOPO vector (Invitrogen), and sequenced. The RNF34 coding region (encoding aa 1-381) was amplified by PCR and subcloned into pCAGGS vector and pCAGGS-HA containing the HA tag at the N terminus. A similar procedure was followed for generating pCAGGS-HA-RNF34 ⌬C. The pCAGGS-HA-RNF34 H351A (histidine to alanine) point mutant was generated by using pCAGGS-HA-RNF34 as template with the GeneTailor site-directed mutagenesis kit (Invitrogen), following the instructions of the manufacturer.
shRNAs for RNF34 and Rescue mRNA-The procedure for making the shRNAs has been described elsewhere (34,36,38,40). Two shRNA constructs targeting RNF34 were generated (Fig. 9A). The Sh1 shRNA targeted a sequence in the coding region (nucleotides 1049 -1067, GenBank TM NM_001004075), whereas the Sh2 shRNA targeted the 3Ј-UTR (nucleotides 1623-1643) of rat RNF34. Each shRNA has an antisense strand that perfectly matches the target mRNA, a loop, and a sense strand that contains a mismatch in the middle to facilitate DNA sequencing (Fig. 9A). A pair of complementary DNA oligonucleotides encoding both arms of the shRNA were synthesized, annealed, and inserted into the mU6pro vector at BbsI and XbaI restriction sites. The corresponding DNA oligonucleotides that carry three point mutations were also made as control shRNAs (Fig. 9A).
To make the rescue mRNA for Sh2, the HA tag and the complete protein coding region of RNF34 corresponding to aa 1-381 were amplified by PCR using pCAGGS-HA-RNF34 as the template, and subcloned into pRK5 vector (rescue). This construct does not have the 3Ј-UTR, and therefore, the mRNA was not targeted by the Sh2 shRNA. We also subcloned it into pRK5, the HA-RNF34 that contains both the protein coding region and the 3Ј-UTR, with a similar procedure. The RNF34 constructs that were subcloned in the pRK5 vector and the shRNAs were used for the knockdown experiments in HEK293 cells or hippocampal neurons as shown in Fig. 9.
In Vitro Pulldown Assay-Equal moles of purified His (8.6 g), His-RNF34-C (aa 195-381, 19.3 g), or His-RNF34-C* (aa 195-381, H351A, 19.3 g) fusion proteins, were adsorbed to a 50-l bed volume of His60Ni beads and incubated with purified GST (13 g) or GST-␥2sIL (18 g) in equilibration buffer (140 mM NaCl, 10 mM Na 2 HPO 4 , 1.8 mM KH 2 PO 4 , pH 7.5) with a protease inhibitor mixture (Roche Applied Science; catalogue no. 1697498) overnight at 4°C. For the pulldown with His-RNF34 FL, due to the very low yield of the expression and purification of the bacterial His-RNF34 FL fusion protein, 2 ml of the His-RNF34 FL bacterial lysate were adsorbed to a 50-l bed volume of His60Ni beads, followed by six washes and incubation with purified GST (13 g) or GST-␥2sIL (18 g). After six washes by centrifugation, the bound proteins were eluted from the beads at 4°C with 250 mM imidazole in equilibration buffer. Eluates were analyzed by SDS-PAGE followed by immunoblotting with Rb anti-GST, Ms anti-␥2IL, and Ms anti-His antibodies.
In Vitro Ubiquitination Assay-The assay was carried out in a 50-l reaction mixture containing 1ϫ ubiquitination buffer and 1ϫ Mg-ATP (catalogue no. SK-10), 315 ng of UBE1 (catalogue no. E-305), 650 ng of UbcH5a (catalogue no. E2-616), 10 g of HA-ubiquitin (Ub, catalogue no. U-110) from Boston Biochem (Cambridge, MA), 2 mM DTT, 820 ng of purified GST-␥2IL, and 7.5 ng of purified His-RNF34 FL. In control reactions, purified GST-␥2IL or His-RNF34 FL was omitted, or 600 ng of GST was used in place of GST-␥2IL. The reaction mixture was incubated in a water bath at 30°C for 2 h and then analyzed by SDS-PAGE followed by immunoblotting with Ms anti-Ub, Ms anti-␥2IL, or Rb anti-GST antibodies.
Preparation of Rat Forebrain Homogenates and Subcellular Fractionation-The preparation of rat forebrain homogenates and various subcellular fractions, including crude synaptosomal (P2), synaptosomal (P2B), microsomal (P3), synaptic plasma membrane (SPM), and "one Triton" postsynaptic den-sity (PSD) fractions, has been described elsewhere (35,37). For membrane preparation, the procedure has been described previously (37). Briefly, the forebrain (telencephalon) tissue of adult female Sprague-Dawley rats was homogenized with 10% sucrose in 50 mM Tris-HCl, pH 7.4 buffer, and centrifuged for 5 min at 1,000 ϫ g. The supernatant was centrifuged at 100,000 ϫ g for 1 h, and the pellet was suspended in 5 mM Tris-HCl, pH 7.4 buffer, and then subjected to homogenization in an all-glass Dounce homogenizer. The lysate was centrifuged at 12,000 ϫ g for 30 min, and the pellet was suspended in 50 mM Tris-HCl, pH 7.4 buffer. All steps for preparation of brain homogenates and fractions were carried out at 4°C. The samples were analyzed by SDS-PAGE, followed by immunoblotting with Rb anti-RNF34 antibody and corresponding secondary antibody.
Immunoblotting-For Fig. 6A only, immunoblots were done according to the method described elsewhere (44). For the other figures and panels, the immunoblot images were collected with a LI-COR Odyssey Infrared Imaging System (Li-Cor Biosciences), analyzed, and quantified with Odyssey software, version 3.0.
Cell Culture, Transfections, and Immunoprecipitation of Transfected Cell Extracts-Hippocampal (HP) neuronal cultures were prepared according to Higgins and Banker (45), as described elsewhere (26,29,46). Briefly, dissociated neurons from embryonic day 18 (E18) Sprague-Dawley rat hippocampi were plated at low density (3,000 -8,000 cells per 18-mm diameter coverslip) and maintained in rat glial cell conditioned medium for up to 3 weeks in vitro, followed by immunofluorescence.
HEK293 cells were cultured on either the 100 ϫ 20-mm plastic dishes or poly-L-lysine-coated coverslips of 18 mm in diameter in DMEM (Invitrogen) with 10% FBS (Invitrogen). For immunoblotting analysis, the ␥2s (WT or Lys-Arg mutants) alone (4 g of plasmid) or a combination of ␣1␤3␥2s (4 g of each) GABA A R subunits together with 1 g of Ub (47) and different RNF34 constructs (13-17 g in a receptor subunit/ RNF34 ϭ 1:5 molar ratio) were transfected into HEK293 cells cultured on 100 ϫ 20-mm plastic dishes. Three days after transfection, cells were harvested and extracted with RIPA containing protease inhibitors. After centrifugation, the supernatants were either subjected to SDS-PAGE and immunoblotting or used for immunoprecipitation experiments. For the latter, the RIPA extracts were precipitated with protein A-Sepharose beads that had been previously incubated with and cross-linked to the Rb anti-␥2(1-29) antiserum (or preimmune serum as control). After four washes with RIPA, proteins were eluted from the beads by incubation with 100 l of 2ϫ sample dissociation buffer for 20 min at room temperature, followed by centrifugation. The supernatants were then subjected to SDS-PAGE in sister gels and immunoblotting with Ms anti-Ub and Ms anti-␥2IL.
For immunofluorescence experiments in Fig. 4, D and E, HEK293 cells were co-transfected with 2 g of total plasmids (pCAGGS-RNF34 ϩ EGFP-Rab5 or EGFP-Rab7) per coverslip for protein expression. Twenty four hours after transfection, HEK293 cells were subjected to immunofluorescence procedures.
HP neurons and HEK293 cells (see below) were transfected with various plasmids using the CalPhos mammalian transfection kit (Clontech), following the instructions of the manufacturer.
Immunofluorescence of Cell Cultures-Immunofluorescence of HP neurons and transfected HEK293 cells was done as described elsewhere (26, 36 -38, 43). Briefly, cells on glass coverslips were fixed in 4% paraformaldehyde, 4% sucrose in PBS for 15 min, followed by incubation with 50 mM NH 4 Cl in PBS for 10 -15 min to quench the free aldehyde groups. Permeabilization was done by incubation with 0.25% Triton X-100 in PBS for 5 min, followed by incubation with 5% donkey normal serum in PBS for 30 min. The coverslips were then incubated overnight at 4°C with a mixture of primary antibodies from different species in 0.25% Triton X-100/PBS, followed by incubation with a mixture of Texas red or Alexa Fluor 594 (in Fig. 9, only)-, FITC-, or aminomethylcoumarin fluorophore-conjugated species-specific anti-IgG secondary antibodies raised in donkey in 0.25% Triton X-100/PBS at room temperature for 1 h. The coverslips were then washed with PBS twice and mounted on glass slides with Prolong Gold antifade mounting solution (Invitrogen).
Light Microscopy Immunocytochemistry of Rat Brain Sections-This procedure has been described elsewhere (30 -32, 36, 37, 39, 48). Briefly, a postnatal day 83 (P83) adult Sprague-Dawley rat was anesthetized with 60 mg/kg ketamine hydrochloride, 8 mg/kg xylazine, and 2 mg/kg acepromazine maleate. The anesthetized rats were perfused through the ascending aorta with 0.1 M phosphate buffer (PB), pH 7.4, and paraformaldehyde/ lysine/periodate (PLP) fixative (4% w/v paraformaldehyde, 1.37% w/v lysine and 0.21% w/v sodium periodate in 0.1 M PB, pH 7.4). Brains were cryoprotected, frozen, and sectioned into 25-m thick sagittal sections with a freezing microtome. After incubating with Rb anti-RNF34 antibody in 0.3% Triton X-100 in 0.1 M PB at 4°C overnight, the free-floating brain sections were incubated with biotin-labeled anti-Rb IgG secondary antibodies followed by the avidin-biotin-horseradish peroxidase complex (ABC procedure, Vectastain Elite ABC kit, Vector Laboratories). The reaction product was visualized by incubating the sections with 3,3Ј-diaminobenzidine tetrahydrochloride in the presence of cobalt chloride, nickel ammonium sulfate, and H 2 O 2 . Sections were then washed and mounted onto gelatin-coated glass slides.
Immunofluorescence of Rat Brain Sections-The procedure has been described elsewhere (36,37). Briefly, 25-m-thick brain slices (prepared as described above) from a P90 Sprague-Dawley rat were subjected to an antigen retrieval procedure by incubation with pepsin (0.15 mg/ml in 0.2 N HCl) at 37°C for 5 min to facilitate the exposure of the tissue antigens to the antibodies. The pepsin treatment did not affect the localization of RNF34 or the other protein markers tested in the brain when compared with nontreated sections. Brain sections were incubated with 5% normal goat serum, 0.3% Triton X-100 in 0.1 M PB for 1 h at room temperature, followed by incubation with a mixture of Rb anti-RNF34 and GP anti-GABA in 2% normal goat serum, 0.3% Triton X-100 in 0.1 M PB at 4°C for 48 h. The brain sections were then incubated with a mixture of Alexa Fluor 488-conjugated anti-Rb and Alexa Fluor 568-conjugated anti-GP IgG secondary antibodies raised in goat dissolved in 2% normal goat serum, 0.3% Triton X-100 in 0.1 M PB at room temperature for 1 h. After washes with 0.1 M PB, the sections were mounted on gelatin-coated glass slides with Prolong Gold anti-fade mounting solution.
Postembedding EM Immunogold-The Lowicryl-embedded tissue blocks were kindly provided by Drs. Peter Somogyi and Zoltan Nusser, and the preparation of the blocks was described elsewhere (49). The tissue blocks were sectioned in our laboratory, and the ultrathin sections were subjected to the immunogold labeling procedure as described elsewhere (28,33,36,37,39). The tissue sections were incubated first with a mixture of affinity-purified Rb anti-RNF34 and Ms mAb to ␤2/3 GABA A R subunits, followed by incubation with goat anti-rabbit and goat anti-mouse IgG secondary antibodies labeled with colloidal gold particles of 18 and 10 nm in diameter, respectively. The tissue sections were counterstained with 2% uranyl acetate and then with 2% lead citrate. No immunolabeling was observed when the primary antibody was omitted. All sections were visual-ized with a Tecnai Biotwin 12-kV transmission electron microscope (FEI Corp., Eindhoven, Holland). EM images were stored in Adobe Photoshop 7.0, and contrast/brightness was adjusted.
Image Acquisition, Analysis, and Quantification-Fluorescence images of cultured HP neurons and HEK293 cells were collected using a Nikon Plan Apo 60ϫ/1.40 objective on a Nikon Eclipse T300 microscope with a Photometrics CoolSnap HQ2 CCD camera, driven by IPLab 4.0 acquisition software (Scanalytics, Rockville, MD). For rat brain sections, fluorescence images were acquired on a Leica TCS SP2 laser confocal microscope using a HCX PL Apo 40ϫ/1.25 oil CS objective lens and a pinhole set at 1 Airy unit. Images of single optical sections (0.5 m thick) were collected.
For qualitative analysis, images were processed with Photoshop version 7.0 (Adobe, San Jose, CA) by adjusting brightness and contrast. For the quantification of the percentage of GABAergic synapses that have associated endogenous RNF34 clusters, 30 cultured HP pyramidal neurons from three independent experiments (15 synapses per neuron, 10 neurons per experiment) were randomly selected for analysis. Thus, a total of 450 GABAergic synapses defined as GADϩ and ␥2ϩ were analyzed. To quantify the association of nonsynaptic ␥2 clusters (GADϪ and ␥2ϩ) with RNF34 clusters, a total of 635 nonsynaptic ␥2 clusters were analyzed from 15 neurons (50 m 2 area per dendritic field, three dendritic fields per neuron, five neurons per experiment, and three independent experiments).
In the RNF34 overexpression or knockdown experiments, ␥2 cluster density was analyzed from 16 randomly selected nontransfected or transfected neurons (50 m 2 area per dendritic field, six dendritic fields per neuron, four neurons per experiment, and four independent experiments). The number of GADϩ boutons per cell was counted in 16 randomly selected nontransfected or transfected neurons (four neurons per experiment and four independent experiments). The RNF34 fluorescence intensity in dendrites was analyzed from 16 neurons using the ImageJ software (National Institutes of Health, Bethesda, MD). For each neuron, the dendritic fluorescence intensity was calculated by averaging the fluorescence intensity of 5 dendritic fields (50 m 2 area per dendritic field). The background fluorescence (to set the zero immunofluorescence intensity value) was calculated by determining the average fluorescence intensity in five randomly chosen areas of the culture devoid of neurons and dendrites. The average RNF34 fluorescence intensity of neurons co-transfected with shRNA constructs (four neurons per experiment and four independent experiments) was normalized to that of nontransfected sister neurons from the same cultures (100% ϭ 1.0). For RNF34 overexpression experiment, the number of counted ␥2 clusters for each plasmid combination or nontransfected neurons was 312-580, and the number of counted GADϩ boutons was 616 -987. In the RNF34 knockdown experiment, for each plasmid combination or nontransfected neurons, the total number of ␥2 clusters was 1724 -2189 and GADϩ boutons was 1393-1892. All values are given as means Ϯ S.E. One-way ANOVA Tukey-Kramer multiple comparison test was used for statistical analysis.

Y2H and in Vitro Pulldown Show a Direct Interaction between RNF34 and the Large Intracellular Loop of the ␥2s and
␥2L GABA A R Subunits-Clone GS29 was isolated from a rat brain cDNA library in a Y2H assay using the large IL of the ␥2 GABA A R subunit short isoform (␥2sIL) as bait. Clone GS29 encoded approximately the C-terminal half of the E3 UBL RNF34, including the following: 1) a 561-bp open reading frame encoding the 187-aa C-terminal polypeptide of the RNF34 (aa 195-381); 2) the stop codon; 3) a 688-bp 3Ј-untranslated region (UTR); and 4) a poly(A) tail. Fig. 1A shows that in addition to ␥2sIL, RNF34 (aa 195-381) also interacted with the ILs of the ␥2L (␥2LIL), ␥1 (␥1IL), and ␥3 (␥3IL) GABA A R subunits. The amino acid sequences of the ILs of the ␥ subunits are highly conserved. However, RNF34 (aa 195-381) did not interact with the ILs of ␣1 (␣1IL), ␤1 (␤1IL), or ␤3 (␤3IL) GABA A R subunits or the cytoplasmic 50-amino acid C terminus of the GluA3 AMPA receptor subunit (GluA3C) or the empty vector pEG202 (Fig. 1A).
Further analysis showed that the RING domain of RNF34 (aa 332-370) was both necessary and largely sufficient for the interaction with ␥2sIL (Fig. 1B), although the interaction was not as strong as that of RNF34 (aa 195-381). The H351A point mutation in the RING domain, which is known to inactivate the RNF34 ubiquitin ligase, did not affect the interaction of RNF34 (aa 195-381) with ␥2sIL. However, the full-length RNF34 did not induce reporter expression. Presumably, the two zinc finger domains in the N terminus, which bind to membrane phospholipids, and the possible palmitoylation of the N terminus of RNF34 (8) keep the full-length RNF34 associated with membranes, preventing its entry to the nucleus and the activation of the reporter. This has also been the case for another ␥2-subunit interacting protein, GABARAP, which has a tubulin binding domain in the N terminus, preventing the translocation of the full-length GABARAP into the nucleus and the interaction with ␥2sIL in the Y2H assay (50).
We have also identified a domain in the ␥2sIL (Fig. 1C) that strongly interacts with RNF34, corresponding to aa 362-404 of ␥2s. A short 14-aa domain in the ␥2sIL (ECLDGKDCASFFCC) was sufficient for interaction with RNF34, although the interaction intensity was lower than that of ␥2sIL. This 14-aa domain is highly conserved among the ILs of the various GABA A R ␥ subunits.
We tested whether the interaction between RNF34 (aa 195-381) and ␥2IL observed in the Y2H assay could also be observed  OCTOBER 17, 2014 • VOLUME 289 • NUMBER 42 in vitro in a bimolecular interaction assay. Immobilized bacterial His-RNF34-C (aa 195-381) and His-RNF34-C* (aa 195-381, H351A) fusion proteins (but not immobilized His) pulled down GST-␥2sIL (Fig. 1D), as revealed by immunoblotting with Ms anti-␥2IL and Rb anti-GST antibodies. Fig. 1E shows that the GST control was not pulled down by immobilized His-RNF34-C or His, indicating that the His or GST tags did not interfere with the interaction between ␥2sIL and RNF34. The results showed that RNF34-C and RNF34-C* directly interact with ␥2sIL.

RNF34 and GABA A Receptors
We continued testing whether His-RNF34 FL interacts with GST-␥2sIL. For this purpose, His-RNF34 FL fusion protein was expressed in bacteria and purified. It is worth mentioning that the yield of purified His-RNF34 FL was much lower than that of His-RNF34-C. Nevertheless, the results (Fig. 1F) clearly showed that His-RNF34 FL pulled GST-␥2sIL down, but not GST, indicating that the full-length RNF34 interact with ␥2sIL. These results and the co-precipitation of RNF34 with assembled GABA A Rs from brain extracts shown below, indicate that the full-length RNF34 interacts with assembled GABA A Rs.
RNF34 Ubiquitinates ␥2sIL in Vitro-RING E3 UBLs recognize and bind to the substrates that they ubiquitinate. To test whether ␥2sIL is a substrate for ubiquitination by RNF34, we performed an in vitro ubiquitination assay using the bacterial fusion proteins His-RNF34 FL (as E3 UBL) and GST-␥2sIL or GST (as substrates). The E3 UBL activity of the bacterial His-RNF34 FL fusion protein is very low. Nevertheless, a smear corresponding to polyubiquitinated GST-␥2sIL (Ub n ) was observed only when RNF34 and all the other components of the ubiquitination machinery were present (Fig. 1G, left IB, lane 3). Under the same conditions, no polyubiquitination smear was observed when the protein substrate was GST instead of GST-␥2sIL (Fig. 1G, left IB, lane 4). Although the smear was observed above ϳ55 kDa, it was particularly strong above ϳ100 kDa.
Immunoblotting with anti-␥2IL or anti-GST antibodies revealed an ϳ55-kDa protein band (marked with a *) only when all the components of the ubiquitination machinery and the GST-␥2IL were present (Fig. 1G, middle and right IB, lane 3). A faint protein band in the same position was also revealed with the anti-Ub Ab (Fig. 1G, left IB, lane 3). This ϳ55-kDa protein band corresponds to ubiquitinated GST-␥2sIL (calculated having two Ub molecules, because the molecular mass of GST-␥2sIL is 35 kDa and that of HA-Ub monomer is 9.8 kDa).
These data indicate that the bacterial GST-␥2sIL fusion protein can indeed be ubiquitinated by the bacterial His-RNF34 fusion protein. Note that the polyubiquitination of GST-␥2sIL produces an amplification effect in the immunoblot signal with the anti-Ub Ab (Ub n ). Such amplification of the signal does not occur when using immunoblots with the anti-␥2IL or anti-GST antibodies. These antibodies, however, show better single protein bands than when the protein signal is diluted in a smear. Also note that in the reported in vitro ubiquitination of other synaptic proteins with other bacterial E3 UBL fusion proteins, only a small fraction of the substrates becomes ubiquitinated (14,51). Also note that a very small amount of His-RNF34 FL was used in the reaction mixture due to the very low purification yield. Because of the limitations of the in vitro ubiquitina-tion assay, we assessed the ubiquitination of the ␥2 GABA A R subunit by RNF34 in living cells.
RNF34 Ubiquitinates ␥2s GABA A R Subunits in Co-transfected HEK293 Cells-We have investigated whether RNF34 is involved in the ubiquitination of the ␥2 GABA A R subunit in cultured cells. For this purpose, we made several HA-tagged RNF34 constructs ( Fig. 2A). In the HA-RNF34 ⌬C construct, the RING finger domain and the C terminus (aa 332-381) were deleted. In the HA-RNF34 H351A construct, histidine (His-351), which is critical for the E3 UBL activity of RNF34 and other RNF proteins, was mutated to alanine (9,52,53).
We examined whether the reduced protein expression level of ␥2s induced by RNF34 was accompanied by increased ubiquitination of ␥2s. Immunoprecipitation of HEK293 cell extracts with the Rb anti-␥2(1-29) antiserum (Fig. 2E) immunoprecipitated both nonubiquitinated 45-kDa and ubiquitinated forms of ␥2s. The latter was revealed as a smear of ubiquitin immunoreactivity (Ͼ55 kDa) in the precipitates, corresponding to the conjugation of one or more ubiquitin molecules to ␥2s. The smear was strongest at Ͼ75 kDa. No immunoprecipitation of ubiquitinated or nonubiquitinated ␥2s was obtained with the preimmune serum (IgG, Fig. 2E). Increased intensity of the ubiquitinated ␥2s and decreased intensity of the nonubiquitinated 45-kDa ␥2s protein in the immunoprecipitates was observed when RNF34 or HA-RNF34 was co-transfected, compared with the enzymatically inactive HA-RNF34 ⌬C or HA-RNF34 H351A forms. The ubiquitination activity of the two RNF34 constructs was determined by calculating the ratio of ubiquitinated ␥2s to nonubiquitinated ␥2s in the immunoprecipitates (22). Fig. 2F shows that RNF34 or HA-RNF34 (lanes 1 and 2) significantly increased the ratio (23.6 Ϯ 3.1, p Ͻ 0.001 and 19.6 Ϯ 3.4, p Ͻ 0.001, respectively) compared with either HA-RNF34 ⌬C (6.8 Ϯ 2.5, lane 3) or HA-RNF34 H351A (7.8 Ϯ

3.2, lane 4).
The increase in the ratio of ubiquitinated to nonubiquitinated ␥2s in the immunoprecipitates induced by RNF34 or HA-RNF34 compared with the E3 UBL inactive controls resulted from both an increase in ␥2s ubiquitination (Fig. 2G, 97.1 Ϯ 7.6 and 92.8 Ϯ 18.8 in lanes 1 and 2 compared with 57.0 Ϯ 20.9 and 51.9 Ϯ 18.5 in lanes 3 and 4) and a decrease in nonubiquitinated ␥2s in the immunoprecipitates (Fig. 2H 3 and 4). The ␥2s ubiquitination observed in the HA-RNF34 ⌬C and HA-RNF34 H351A controls (Fig. 2E) corresponds to the ubiquitination of ␥2s by endogenous E3 UBLs. Similar experiments using ␥2L (data not shown) indicated that RNF34 is also able to ubiquitinate ␥2L. These experiments show that RNF34 increases the pool of ubiquitinated ␥2 (␥2s and ␥2L) and that this effect requires the E3 UBL activity of RNF34, because deletion of the RING finger domain or the H351A mutation abolish the RNF34-induced increased ubiquitination of ␥2.
Thus, neither the 7KR nor K373R mutations could prevent the RNF34-induced degradation of the mutated ␥2s. However, the combination of both in the 8KR mutant prevented the deg-radation of this mutant. The combined results from Figs. 2 and 3 show that RNF34 ubiquitinates ␥2s promoting its degradation and that the ubiquitination occurs in several lysines in the ␥2sIL.

RNF34 Promotes the Degradation of Assembled GABA A Rs via the Lysosome-and Proteasome-dependent Pathways-When
HEK293 cells are transfected only with the ␥2s subunit (without ␣1 and ␤3) some of the expressed ␥2s goes to the cell surface, but much of the ␥2s remain in the ER (54,55). When ␥2s is co-transfected with ␣1 and ␤3, much of the ␥2s assembles with ␣1 and ␤3 forming functional heteromeric ␣1␤3␥2s receptors that exit the ER and are translocated to the cell surface (38, 54 -57). When HEK293 cells were transfected with ␥2s only, the ␥2s protein migrate significantly faster (ϳ45 kDa) than when cells were co-transfected with ␥2s, ␣1, and ␤3 (ϳ50 kDa, Fig. 4A, lanes 2 and 3, respectively). After treating the cell extracts with the glycosidase peptide-N-glycosidase F, which cleaves N-linked oligosaccharides from glycoproteins, the ␥2s  subunits showed the same mobility (ϳ38 kDa) in the presence or absence of ␣1 and ␤3 (Fig. 4A, lanes 5 and 6). These results indicate that ␥2s is differentially N-glycosylated when it is assembled with ␣1 and ␤3 than when ␣1 and ␤3 are absent. This is likely due to the differential glycosylation of ␥2s during translocation from the ER to the cell surface, which is favored by the assembly into ␣1␤3␥2s heteropentamers.
We also tested whether RNF34 reduces the level of assembled GABA A Rs by monitoring the mature glycosylated ␥2s in ␣1␤3␥2s recombinant receptors. Fig. 4B shows that HA-RNF34 significantly reduces the expression levels of the ϳ50 kDa ␥2s subunit in the ␣1␤3␥2s recombinant receptors (51.3 Ϯ 6.1%, p Ͻ 0.01), compared with cells co-transfected with HA (100 Ϯ 2.4%) or the UBL inactive mutant HA-RNF34 H351A (126.9 Ϯ 23.2%). Upon treatment of the transfected HEK293 cells for 12 h with leupeptin or MG132 before cell harvesting, the reduction of the expression level of ␥2s subunit induced by HA-RNF34 (46.1 Ϯ 8.2%, p Ͻ 0.01) was partially reversed (76 Ϯ 13.6%, p Ͻ 0.05 for leupeptin treatment; 89.2 Ϯ 12.3%, p Ͻ 0.05 for MG132 treatment; Fig. 4C). Note that HEK293 cells were treated only for 12 h with leupeptin or MG132 to prevent the toxicity of these drugs. The results indicate that RNF34 induced the degradation of assembled ␥2s-containing GABA A Rs via both the lysosome and proteasome pathways.
The effect of leupeptin suggests that part of the assembled GABA A Rs present at the surface, after endocytosis, are degraded via the endosome-lysosome pathway. Moreover, endosomal GABA A Rs can also go into the proteasome degradation pathway. We investigated whether RNF34 is indeed localized in endosomes by co-transfecting HEK293 cells with RNF34 and EGFP-Rab5 or EGFP-Rab7. It has been shown that EGFP-Rab5 and EGFP-Rab7 target to early and late endosomes, respectively (58). RNF34 forms clusters (Fig. 4, D and E), which are frequently co-localizing with EGFP-Rab5 (Fig. 4, D1-D3, arrowheads) or EGFP-Rab7 (Fig. 4, E1-E3, arrows) clusters. The results show that RNF34 is in part localized at endosomes (early and late endosomes), which is consistent with RNF34 being involved in ubiquitination and degradation of membrane GABA A Rs.
RNF34 and GABA A Rs Co-immunoprecipitate from Brain Membrane Extracts-We studied whether the interaction between RNF34 and GABA A Rs that we have observed in Y2H and in vitro also occurs in the brain. We found that GABA A Rs and RNF34 co-immunoprecipitated from rat forebrain membrane extracts. The antisera to the ␥2 GABA A R subunit, such as the two anti-␥2 Abs raised in rabbit (Rb ␥2 and Rb ␥2(1-29), Fig. 5A) or an anti-␥2 Ab raised in guinea pig (GP ␥2, Fig. 5B) co-precipitated the 42-kDa RNF34 protein together with the GABA A Rs ␥2 (45 kDa) and ␣1 (51 kDa) subunits. The co-precipitation of ␣1 with ␥2 indicates that heteromeric assemblies of GABA A R subunits were co-precipitated with RNF34. The corresponding control with nonimmune rabbit or guinea pig serum (IgG) did not precipitate RNF34 or GABA A Rs.
Moreover, the antisera to the ␣1 GABA A R subunits that were raised in rabbit (Rb ␣1, Fig. 5C) or guinea pig (GP ␣1, Fig. 5D) co-precipitated RNF34 together with GABA A Rs ␥2 and ␣1 subunits, confirming that RNF34 co-precipitates with assembled GABA A Rs. No precipitation was obtained with nonimmune rabbit or guinea pig serum (IgG). The data indicate that in the brain, RNF34 is associated with GABA A Rs.
For revealing RNF34 in the immunoprecipitation experiments described above and for the immunocytochemistry experiments described below, we raised a novel rabbit polyclonal antibody (Rb anti-RNF34) to the C terminus of rat RNF34 (aa 368 -381) and prepared affinity-purified antibody on immobilized antigen peptide. Immunoblots of a rat forebrain crude synaptosomal (P2) fraction showed that the affinity-purified Rb anti-RNF34 antibody specifically recognized a 42-kDa protein (Fig. 6A), which corresponds to the molecular weight of the rat RNF34 protein calculated by its amino acid sequence (42,680 Da). The immunoreactivity was blocked by preincubation of Rb anti-RNF34 antibody with the antigenic peptide (Fig. 6A).
Developmental Expression and Cellular and Subcellular Localization of RNF34 in the Rat Brain-Some GABAergic interneurons show high expression levels of RNF34. We have used the affinity-purified Rb anti-RNF34 antibody to study the expression of RNF34 during brain development. Immunoblots of homogenates of the forebrains from rats of various ages show that the RNF34 protein is not expressed until the 2nd postnatal week coinciding with the peak of synaptogenesis, reaching the maximum expression at P30 (Fig. 6B). The immunoreactivity of Rb anti-RNF34 antibody was displaced by the purified fusion protein His-RNF34 (aa 195-381, Fig. 6B) or antigenic peptide (data not shown).
RNF34 is highly enriched in the microsomal (P3) fraction, although it is also present in other fractions, including crude synaptosomal (P2), synaptosomal (P2B), total membrane, SPM, and postsynaptic density (PSD) fractions (Fig. 6C). RNF34 carries the N-terminal putative signal sites for palmitoylation (8), which is consistent with its presence in membrane fractions (membrane and SPM). Moreover, RNF34 contains a FYVE domain that can serve as a membrane-targeting or endosomelocalizing signal (9,59), which is consistent with the enrichment of RNF34 in the microsomal fraction.
Light microscopy immunocytochemistry of adult rat brain sagittal sections showed that RNF34 is ubiquitously expressed in most regions of the brain, including cerebral cortex (Fig. 6D) and corpus striatum (Fig. 6E). At the cellular level, the immunoreactivity concentrated on the perikaryon of neurons such as the large pyramidal neurons of layer V of the cerebral cortex (Fig. 6D). Some interneurons showed very strong immunoreactivity in cerebral cortex (Fig. 6, D, arrowheads, and F) and corpus striatum (Fig. 6, E, arrowheads, and G). Double labeling of rat brain sections with Rb anti-RNF34 and GP anti-GABA antibodies confirmed the high expression of RNF34 in some GABAergic interneurons (GABA ϩ ) in corpus striatum (Fig. 6, H and I, arrows) and cerebral cortex (data not shown). The RNF34 immunoreactivity was blocked by purified fusion protein His-RNF34 (aa 195-381, data not shown).
GABAergic Synapses Often Have RNF34 Associated with Them-We have studied the expression and localization of RNF34 in 21 DIV cultured HP neurons by immunofluorescence. At this culture age, most pyramidal neurons and interneurons show RNF34 clusters both in dendrites (Fig. 7A) and in soma (data not shown). The RNF34 cluster immunofluorescence was blocked by the purified fusion protein His-RNF34 (aa 195-381) or antigenic peptide (data not shown). Although the majority of the RNF34 clusters were not associated with synapses, a significant proportion of GABAergic synapses had associated RNF34 clusters. Triple-label immunofluorescence with anti-RNF34, anti-GAD, and anti-␥2 GABA A R subunit antibodies showed that 75 Ϯ 2% of the GABAergic synapses (GADϩ and ␥2ϩ) had associated RNF34 clusters (Fig. 7, A, and A1-A4, arrowheads). Many nonsynaptic (GADϪ) ␥2 clusters also showed colocalizing RNF34 clusters, although the proportion of nonsynaptic ␥2 clusters that colocalized with RNF34 was significantly smaller (51 Ϯ 2%, p Ͻ 0.001 by Student's t test) than when presynaptic GADϩ terminals were juxtaposed to ␥2 clusters. This result shows that although the presence of a presynaptic GAD terminal is not necessary for the colocalization of RNF34 with ␥2, the colocalization of RNF34 with ␥2 clusters significantly increased at GABAergic synapses.
The EM immunogold data are consistent with the notion that RNF34 is localized at synapses (pre-and post-synaptically), nonsynaptic membranes, and cytoplasm, probably at cytoplasmic organelles such as endosomes. Thus, the immunofluores-FIGURE 6. Expression and distribution of RNF34 in the rat brain. A, immunoblot with the Rb anti-RNF34 antibody of the rat forebrain crude synaptosomal (P2) fraction, shows a major protein band ϳ42 kDa, which is displaced by 100 g/ml of the antigenic peptide (ϩ Pep). 20 g of the total proteins were loaded into each lane. B, immunoblots of rat forebrain homogenates of various ages from embryonic day 18 (E18) to postnatal day 90 (P90) with Rb anti-RNF34 and Ms anti-actin antibodies. 75 g of total protein were loaded into each lane. The RNF34 antibody recognizes a 42-kDa protein band (arrowhead), whose immunoreactivity is displaced by incubating the antibody with 100 g/ml of the purified bacterial His-RNF34 (aa 195-381) fusion protein (ϩ FP). The bar graph shows the quantification of the relative RNF34 protein expression levels by densitometry of the RNF34 protein band normalized to actin protein band density. C, immunoblot of various rat forebrain subcellular fractions with Rb anti-RNF34 antibody. The RNF34 protein band is present in all the fractions tested, including crude synaptosomal (P2), microsomal (P3), synaptosomal (P2B), total membrane (mem), SPM, and one Triton PSD fractions. 25 g of the total proteins were loaded into each lane. D and E, RNF34 immunocytochemistry in layers IV and V of the cerebral cortex (Ctx) and corpus striatum (St). Arrowheads indicate neurons with high RNF34 immunoreactivity. cence and immunogold data show that although RNF34 is not exclusively localized at GABAergic synapses, the latter frequently show the presence of RNF34. GABA A R clusters frequently have RNF34 associated with them. Moreover, the high expression of RNF34 in some GABAergic interneurons is consistent with the presynaptic localization of some RNF34 (in the axon of these GABAergic interneurons).
Neurons transfected with HA-RNF34 also exhibited decreased number of endogenous GAD boutons contacting the transfected cells (38.5 Ϯ 3.1 GAD boutons/cell, Fig. 8 These results show that overexpression of RNF34 significantly decreases the number of ␥2-GABA A R clusters in the transfected neurons and the number of GABAergic contacts that the transfected neurons receive, and these effects depend on the E3 UBL activity of RNF34. Knocking Down Endogenous RNF34 in Hippocampal Neurons Increases Both ␥2-GABA A R Cluster Density and GABAergic Innervation-We made short hairpin RNAs (Sh1 and Sh2) that specifically target RNF34 mRNA in the coding region and the 3Ј-UTR, respectively. The corresponding shRNAs with three point mutations (Sh1 3m and Sh2 3m) were used as controls (Fig. 9A).
The knockdown effect of Sh1 and Sh2 on RNF34 expression was determined in HEK293 cells co-transfected with HA-RNF34 and shRNAs (Fig. 9, B and C). The HA-RNF34 protein expression was significantly knocked down by Sh1 (51.3 Ϯ 4.3%, p Ͻ 0.05) and even more by Sh2 (29.2 Ϯ 3.9%, p Ͻ 0.001) but not by Sh1 3m (104.7 Ϯ 13%, p Ͼ 0.05) or Sh2 3m (88.5 Ϯ 8.7%, p Ͼ 0.05) controls, compared with HEK293 cells co-transfected with HA-RNF34 and mU6 vector (100 Ϯ 4.8%). Because of the stronger knockdown effect of Sh2, we decided to use Sh2 in subsequent experiments and designed a rescue mRNA for Sh2. Because Sh2 targets the 3Ј-UTR of RNF34 mRNA, we used for rescue an mRNA that contains the coding region of RNF34 but not the 3Ј-UTR (rescue). In another experiment, we tested the effect of Sh2 on the rescue construct. Sh2 did not affect rescue expression (108 Ϯ 17.2%, p Ͼ 0.05), compared with HEK293 cells co-transfected with rescue and mU6 vector (107.1 Ϯ 15%) as shown in Fig. 9C. The internal positive control showed that Sh2 significantly knocked down HA-RNF34 (30.1 Ϯ 2.7%, p Ͻ 0.001) compared with HEK293 cells co-transfected with HA-RNF34 and mU6 vector (100 Ϯ 8.6%). Actin was used as the loading control, and no effect on actin expression was observed by Sh1 or Sh2.
Endogenous RNF34 is fully expressed in 21 DIV HP neurons (Fig. 7A). In the rat brain, RNF34 was expressed postnatally by the 2nd postnatal week (Fig. 6A). In HP cultures, RNF34 immu- FIGURE 7. GABAergic synapses frequently have RNF34 associated with them. A, triple-labeled immunofluorescence of 21 DIV cultured hippocampal neurons with GP anti-␥2 GABA A R subunit (red, A1), Rb anti-RNF34 (green, A2), and sheep anti-GAD (blue) antibodies. A3 shows the overlay of ␥2 and RNF34 fluorescence, whereas A4 shows the overlay of the three fluorescence channels. Arrowheads indicate GABAergic synapses (␥2 ϩ and GAD ϩ ) that have co-localizing RNF34 clusters. Scale bar, 2.5 m (A1-A4). B-G, postembedding EM immunogold double-labeling of rat brain cerebellum (B, C, and E) and cerebral cortex (D, F, and G) with Rb anti-RNF34 and mouse anti-␤2/3 GABA A R subunits. The ␤2/3 immunolabeling (smaller gold particles) is indicated by arrowheads. GABAergic synapses are identified by their morphology and the concentration of ␤2/3 GABA A R subunits. The RNF34 immunolabeling (larger gold particles, arrows) is localized at or near GABAergic synapses (B-G), GABAergic presynaptic terminals (C, E, and F), and GABAergic postsynapses (B, D, and G). The goat anti-rabbit IgG and goat anti-mouse IgG secondary antibodies were conjugated to 18 and 10 nm diameter colloidal gold particles, respectively. Scale bar, 100 nm (B-G). OCTOBER 17, 2014 • VOLUME 289 • NUMBER 42 nofluorescence signal was detected after 17 DIV. Therefore, for the following knockdown experiments, neuronal cultures were transfected at 10 DIV, and the effects were studied at 21 DIV, when nontransfected cells show strong endogenous RNF34 expression. Fig. 9, D and F, shows that neurons co-transfected with Sh2 exhibited a significant reduction in the immunofluorescence intensity of endogenous RNF34 (59.2 Ϯ 3.6%, p Ͻ 0.001), compared with nontransfected neurons (100 Ϯ 4.4%) and other controls shown below. Neurons transfected with control plasmids, mU6 vector (105.3 Ϯ 5.4%) or Sh2 3m (103.4 Ϯ 3.9%) showed no reduction in RNF34 immunofluorescence intensity, compared with nontransfected neurons (p Ͻ 0.05). The knockdown effect of Sh2 on endogenous RNF34 expression was prevented by the rescue mRNA (104.9 Ϯ 4.1%, p Ͻ 0.001 compared with Sh2). Note that the rescue RNF34 was constructed in pRK5 plasmid to achieve a moderate level of expression in neurons. Under the transfection conditions, this construct acted as rescue restoring control levels of RNF34 expression. In all neuronal transfections, EGFP was co-transfected as indicator of the transfected neurons.

RNF34 and GABA A Receptors
Knocking down endogenous RNF34 expression in cultured hippocampal neurons with Sh2 significantly increased the density of ␥2-GABA A R clusters (38.9 Ϯ 1.6 clusters/100 m 2 , p Ͻ 0.001, Fig. 9, E and G), compared with nontransfected neurons

DISCUSSION
The interaction between the E3 UBL RNF34 and the ␥2 GABA A R subunit was initially revealed by Y2H screening of a rat brain cDNA library. We found that a C-terminal half (aa 195-381) of RNF34 strongly interacts with the large IL of ␥2 GABA A Rs. That the two proteins directly interacted with each other was confirmed by an in vitro pulldown assay with bacterially expressed fusion proteins. The notion that the two proteins interact in the rat brain was supported by the co-precipitation of RNF34 and assembled GABA A Rs from rat brain membrane extracts.
There are several lines of evidence supporting the notion that RNF34 ubiquitinates the ␥2 subunit of GABA A Rs. In HEK293 cells, RNF34 significantly reduced the expression level of the ␥2s GABA A R subunit. This effect was accompanied by an increased ratio of ubiquitinated to nonubiquitinated ␥2s in the immunoprecipitates and was prevented either by deleting the C terminus containing the RING domain (RNF34 ⌬C) or by the point-mutation H351A in the RING domain. Deletion of the RING domain both inactivates the E3 UBL and prevents the binding of RNF34 ⌬C to the ␥2IL; however, the H351A mutation inactivates the E3 UBL (RNF34 H351A (9)) without affecting the binding of RNF34 to ␥2IL (Fig. 1). Additional support for the notion that RNF34 ubiquitinates the ␥2 subunit was derived from mutating several lysines of the ␥2sIL into arginines (␥2s 8KR, 9KR, or 10KR) which prevented RNF34-induced degradation of ␥2. Also, bacterially expressed His-RNF34 was able to in vitro ubiquitinate GST-␥2sIL but not GST. The E3 UBL activity of the bacterial fusion protein was very low but measurable. Other reports using bacterially expressed E3 UBL for in vitro ubiquitinating synaptic proteins also show low activ-ity, and only a very small fraction of the synaptic substrates become ubiquitinated (14,51).
Previous studies by us and others have shown that ␣, ␤, and ␥2 subunits could assemble into functional GABA A Rs and target to the cell membrane when they were co-expressed in HEK293 cells (38, 54 -57). Our data indicate that RNF34 reduces the protein expression level of the ␥2 subunit, regardless of whether ␥2 is expressed alone or expressed together with ␣1 and ␤3 subunits forming assembled ␣1␤3␥2 pentameric GABA A Rs at the cell surface. The treatment of HEK293 cells co-expressing ␣1␤3␥2 subunits with leupeptin or MG132 partially reversed the effect on GABA A R expression caused by RNF34, which strongly suggests the involvement of both lysosomal and proteasomal degradation of the ␥2 subunits that had been ubiquitinated by RNF34.
Our results also showed that RNF34, although not exclusively, is associated with GABAergic synapses. Immunofluorescence of cultured hippocampal neurons showed that RNF34 is associated with 75% of GABAergic synapses (␥2ϩ and GADϩ) and with 51% nonsynaptic ␥2 clusters. Postembedding EM immunogold also revealed the presence of RNF34 in GABAergic synapses, both pre-and postsynaptically. Consistent with a postsynaptic localization, RNF34 is present in the SPM fraction and the classical one-Triton PSD fraction, which contains postsynaptic densities from both GABAergic and glutamatergic synapses (35). The presence of RNF34 in GABAergic presynaptic terminals observed by EM is consistent with the high expression of RNF34 in GABAergic interneurons (Fig. 7). A synaptic role of RNF34 is also supported by our observation that during development the RNF34 expression appears and increases during the 2nd and 3rd postnatal week coinciding with the time course of synaptogenesis (60). The presence of RNF34 in membrane and synaptic fractions as well as in endosomes (discussed below) suggests that RNF34 plays an important role in the ubiquitination of synaptic proteins regulating their endocytosis, recycling, and degradation.
Additional evidence consistent with the notion that RNF34 plays a functional role in regulating ␥2 GABA A R expression and GABAergic synaptic function and stability was derived from overexpression and knockdown experiments in neurons. It has been shown that the ␥2 GABA A R subunit is essential for the postsynaptic localization of GABA A Rs (34,61,62). Therefore, the ubiquitination and degradation of the ␥2 GABA A Rs play an important role in controlling the number of postsynaptic GABA A Rs. Thus, overexpression of RNF34 significantly reduced the postsynaptic clustering of ␥2 GABA A Rs in cultured hippocampal neurons. Moreover, it also reduced the presynaptic GABAergic innervation that these neurons received. We and others have previously shown that reduced postsynaptic clustering of ␥2 GABA A Rs in neurons is accompanied by reduced presynaptic innervation that these neurons receive (34,63). However, knocking down the endogenous RNF34 expression by shRNAs increased the density of postsynaptic ␥2-GABA A R clusters as well as the number of presynaptic GADϩ boutons. In our experiments, the transfected neurons are postsynaptic, and the observed effects resulted from altering RNF34 expression levels in the postsynaptic neurons. Taken together, our data indicate that in hippocampal neurons RNF34 regulates the expression and clustering of GABA A Rs and also the stability of GABAergic synapses. This is not to imply that RNF34 is the only E3 UBL involved in the ubiquitination and degradation of GABA A Rs. It has been shown that GABA A R subunits other than ␥2 are subjected to ubiquitination (21) by still unidentified E3 UBL(s). Moreover, we cannot exclude the involvement of other E3 UBL in the ubiquitination of ␥2.
The ubiquitination of ␥2 by RNF34 and subsequent degradation of ␥2 could explain by themselves the effects observed after overexpressing and knocking down RNF34 in neurons. Nevertheless, E3 UBLs could ubiquitinate multiple protein targets. It is likely that, in addition to ␥2, RNF34 ubiquitinates other synaptic and nonsynaptic proteins. As indicated above, much of the RNF34 is not associated with GABAergic synapses.
RNF34 (also named hRFI, momo, and CARP1) was initially isolated from human esophageal cancer (64). RNF34 has E3 UBL activity and RNF34 mRNA is highly expressed in brain and testis compared with other tissues (8,9). In studies done on non-neural tissues, RNF34 had anti-apoptotic activity, against both death receptor-and mitochondrion-mediated apoptosis, by ubiquitinating protein substrates such as caspases 8/10 and p53, targeting them for proteasome-mediated degradation (9,52,53,59,(65)(66)(67)(68). To the best of our knowledge, this is the first report of a synaptic role for RNF34.
We have found that RNF34 is enriched in the microsomal fraction, which contains endosomes and ER among other micro-organelles. Support for a role of RNF34 in endosomes is derived from the presence in RNF34 of an N-terminal FYVE (Fab1, YOTB, Vac1, and EEA1) domain (9,59,69). The FYVE domain of RNF34 preferentially binds in vitro to phosphatidylinositol 3-phosphate and phosphatidylinositol 5-phosphate and with lower binding affinity to phosphatidylinositol 4-phosphate (9). The binding of RNF34 to phosphatidylinositol 3-phosphate, which is mainly localized in endosomes (70), is consistent with an endosomal function of RNF34. Moreover, our immunofluorescence data show an association of some RNF34 with the early and late endosome markers Rab5 and Rab7, respectively. Others have previously shown that a lysinerich motif (Lys-325, Lys-328, Lys-330, Lys-332, Lys-333, Lys-334, and Lys-335) within the ␥2IL is involved in the ubiquitination and targeting of ␥2 GABA A Rs for lysosomal degradation (22). The presence of RNF34 in endosomes and the ubiquitination of these and other lysines of ␥2IL by RNF34 is consistent with this E3 UBL involvement in the reported ubiquitination and lysosomal degradation of ␥2 GABA A Rs. It is also consistent with our observed partial reversal of the RNF34 effect on ␥2 expression by leupeptin. We have also observed a partial reversal by MG132, indicating that ubiquitination of ␥2 by RNF34 can also be derived in proteasomal degradation.
Further mapping data by Y2H shows that RNF34 specifically binds to the C-terminal region (aa 362-404) of ␥2sIL. This protein region contains two motifs involved in the interaction of the ␥2 subunit with other proteins as follows: the GODZbinding motif (aa 368 -381 (71)) and the GABARAP binding motif (aa 386 -403 (50)). Further analysis of the RNF34-binding motif reveals that there is no overlap between the GABARAP and RNF34-binding motifs in ␥2sIL. However, the 14-aa GODZ-binding motif of ␥2sIL (aa 368 -381) is also involved in the interaction with RNF34. This raises the possibility that these two types of post-translational modifications of the ␥2 subunit, involving palmitoylation by GODZ, which promotes the translocation of GABA A Rs to the cell surface and clustering (63,71), and ubiquitination by RNF34, which promotes GABA A R degradation, might work in concert regulating the stability and clustering of GABA A Rs.
In summary, our results, consistent with the notion that RNF34, a RING domain E3 ubiquitin ligase, are as follows: 1) interact with and ubiquitinate the ␥2 GABA A R subunit; 2) are present, although not exclusively, in GABAergic synapses; 3) regulate the expression and clustering of GABA A Rs; and 4) regulate GABAergic synapse stability.