TNF-α Downregulates Inhibitory Neurotransmission through Protein Phosphatase 1-Dependent Trafficking of GABA_A Receptors

Abstract

Inflammation has been implicated in the progression of neurological disease, yet precisely how inflammation affects neuronal function remains unclear. Tumor necrosis factor-α (TNFα) is a proinflammatory cytokine that regulates synapse function by controlling neurotransmitter receptor trafficking and homeostatic synaptic plasticity. Here we characterize the mechanisms through which TNFα regulates inhibitory synapse function in mature rat and mouse hippocampal neurons. Acute application of TNFα induces a rapid and persistent decrease of inhibitory synaptic strength and downregulation of cell-surface levels of GABA_ARs containing α1, α2, β2/3, and γ2 subunits. We show that trafficking of GABA_ARs in response to TNFα is mediated by neuronally expressed TNF receptor 1 and requires activation of p38 MAPK, phosphatidylinositol 3-kinase, protein phosphatase 1 (PP1), and dynamin GTPase. Furthermore, TNFα enhances the association of PP1 with GABA_AR β3 subunits and dephosphorylates a site on β3 known to regulate phospho-dependent interactions with the endocytic machinery. Conversely, we find that calcineurin and PP2A are not essential components of the signaling pathway and that clustering of the scaffolding protein gephyrin is only reduced after the initial receptor endocytosis. Together, these findings demonstrate a distinct mechanism of regulated GABA_AR endocytosis that may contribute to the disruption of circuit homeostasis under neuroinflammatory conditions.

Introduction

Immune molecules expressed in the CNS, such as cytokines, class I major histocompatibility complex, and complement cascade components, are active in a variety of contexts ranging from synapse development to cell survival and synaptic plasticity (Yirmiya and Goshen, 2011). In particular, the proinflammatory cytokine tumor necrosis factor-α (TNFα) is an important regulator of synapse function implicated in neurotransmitter receptor trafficking (Beattie et al., 2002; Ogoshi et al., 2005; Stellwagen et al., 2005), homeostatic synaptic plasticity (Stellwagen and Malenka, 2006; Kaneko et al., 2008; Steinmetz and Turrigiano, 2010), and regulation of gliotransmission (Santello et al., 2011). TNFα is a proteolytically cleaved transmembrane protein that signals through TNF receptor 1 (TNFR1) and TNFR2 (MacEwan, 2002). Expression of TNFα in the CNS is upregulated by glia in response to inflammatory neurological insults, such as stroke, infection, and neurodegenerative diseases (Tonelli and Postolache, 2005; Frankola et al., 2011), and in response to prolonged neuronal activity blockade (Stellwagen and Malenka, 2006). Neurons respond to elevated levels of TNFα by rapidly increasing excitatory synaptic strength (Tancredi et al., 1992; Beattie et al., 2002; Stellwagen et al., 2005) and weakening inhibitory synaptic strength, resulting in a higher excitatory/inhibitory ratio (Stellwagen et al., 2005). Although the increase in excitatory synaptic strength is known to occur through TNFR1, phosphatidylinositol 3-kinase (PI3K) activity and cell-surface delivery of Ca²⁺-permeable AMPA receptors (AMPARs; Ogoshi et al., 2005; Stellwagen et al., 2005; Yin et al., 2012), the detailed molecular mechanisms underlying the weakening of inhibitory synaptic strength induced by TNFα remain essentially unknown.

GABA_A receptors (GABA_ARs) are the primary mediators of fast inhibitory neurotransmission in the brain, exerting tight control over the excitatory/inhibitory balance (Mann and Paulsen, 2007; Jacob et al., 2008). Accordingly, selective disruption of GABA_AR function leads to homeostatic modifications in circuit development (Duveau et al., 2011; Pallotto et al., 2012), epilepsy (DeLorey et al., 1998), and anxiety-like behavior (Crestani et al., 1999). A pivotal point in the regulation of GABAergic synaptic strength is the dynamic trafficking of GABA_ARs (Luscher et al., 2011). In particular, endocytosis of GABA_ARs is regulated by phosphorylation-dependent binding of the clathrin adaptor protein 2 (AP2) adaptor complex to motifs in the intracellular domains of GABA_AR β and γ subunits (Arancibia-Cárcamo and Kittler, 2009). Phosphoresidues within these motifs are subject to regulation by kinases (e.g., PKA and PKC) and phosphatases (e.g., protein phosphatase 1 [PP1]/PP2A and calcineurin; Luscher et al., 2011). The effect of TNFα on phospho-dependent regulation of GABA_AR trafficking remains unexplored.

Here, using multiple experimental approaches, we characterize the mechanisms through which TNFα downregulates GABAergic neurotransmission in hippocampal neurons. TNFα signals through neuronal TNFR1, p38 MAPK, PI3K, and PP1 to rapidly reduce inhibitory synaptic strength and surface levels of synaptic GABA_ARs. These effects persist over time and are followed by a reduction in clustering of the postsynaptic scaffolding protein gephyrin. Moreover, we show that TNFα increases the association of PP1 with GABA_AR β3, dephosphorylates β3 S408/409, and reduces inhibitory synaptic strength via dynamin GTPase activity. Together, these results describe a mechanism of regulated GABA_AR endocytosis via which GABAergic inhibition is downregulated under conditions associated with elevated TNFα levels.

Materials and Methods

Hippocampal neuron culture.

Animal procedures were performed in accordance with the guidelines of the Canadian Council for Animal Care and the Montreal General Hospital Facility Animal Care Committee. All cell culture experiments (except those in Fig. 4C,D) made use of mixed neuron–glia cultures prepared based on established protocols (Kaech and Banker, 2006) from E17–E18 Sprague Dawley rat hippocampus (of either sex). Briefly, dissociated cells were plated at a density of 2.7 × 10⁴/cm² on either 12 mm poly-d-lysine-coated coverslips (Thermo Fisher Scientific) for immunostaining and electrophysiology experiments or on 10 cm poly-d-lysine-coated tissue culture plates for immunoprecipitation and Western blotting experiments. Culture media consisted of Neurobasal medium supplemented with B-27 and Glutamax (Life Technologies). Glial proliferation was inhibited at 12 d in vitro with 8 μm AraC. Experiments were performed at 14–22 d in vitro. For experiments in Figure 4, C and D, glia derived from P1–P2 mouse forebrain (of either sex) were grown to confluence on poly-d-lysine-coated 24-well plates. Neurons were dissociated from P0 mouse hippocampus [of either sex, C57BL/6J wild type (WT; The Jackson Laboratory) or TNFR1^−/− (kindly provided by Dr. P. Brodt, McGill University, Montréal, Québec, Canada)], seeded at a density of 3.7 × 10⁴/cm² on 12 mm poly-d-lysine-coated coverslips for 2 h, transferred to the glial culture plate wells, and grown in the same media used for rat cultures; additional glial proliferation was inhibited with AraC.

Figure 4.

TNFα signals through neuronal TNFR1 to downregulate GABAergic neurotransmission. A, Representative dendritic regions of surface immunostaining for GABA_AR γ2 subunit, from untreated (Ctrl) or TNFα-treated cultured neurons, preincubated with either TNFR1 neutralizing antibody (R1nAb) or TNFR2 neutralizing antibody (R2nAb). Scale bar, 3 μm. B, Quantification data of γ2 cluster area corresponding to conditions in A, showing that blocking TNFR1 function blocks the effects of TNFα (gray bars), whereas blocking TNFR2 does not (n = 135–191 images in each condition; two-way ANOVA, Bonferroni's post hoc test, *p < 0.05, ***p < 0.001; n.s., not significant). C, Representative traces of mIPSC recordings from untreated (Ctrl) or TNFα-treated WT or TNFR1^−/− neurons cocultured with WT glia or TNFR1^−/− glia. D, Group data of average mIPSC amplitude indicating that TNFα-induced downregulation is absent from TNFR1^−/− neurons cocultured with WT glia but present in WT neurons cocultured with WT or TNFR1^−/− glia (n = 12–22 cells in each condition; two-way ANOVA, Bonferroni's post hoc test, *p < 0.05, **p < 0.01; n.s., not significant).

Culture electrophysiology and treatments.

Miniature IPSCs (mIPSCs) were recorded at room temperature using the whole-cell patch-clamp method from neurons superfused with ACSF containing the following (in mm): 115 NaCl, 5 KCl, 23 glucose, 26 sucrose, 4.2 HEPES, 2.5 CaCl₂, and 1.3 MgCl₂, pH 7.2 (osmolality, 295–305 mOsm). Recording pipettes were pulled from borosilicate glass (King Precision Glass) to an open-tip resistance of 3–6 MΩ and filled with an internal solution of the following (in mm): 127 CsCl, 8 NaCl, 1 CaCl₂, 10 HEPES, 10 EGTA, 0.3 Na₃-GTP, and 2 Mg-ATP, pH 7.2 (osmolality, 280–290 mOsm). To isolate mIPSCs, 200 nm TTX, 10 μm NBQX, and 25 μm d-APV (Tocris Bioscience) were added to the external solution, and recordings were obtained at V_h = −70 mV. Signals were amplified and filtered (1 kHz) using an Axopatch 200B amplifier, sampled at 5 kHz using a Digidata 1440A, and recorded with Clampex 10.3 (Molecular Devices). Analyses of mIPSC amplitude, frequency, 10–90% rise time, and decay tau were performed using Clampfit 10.3 (Molecular Devices). Only cells with a stable series resistance of <25 MΩ throughout the recording period were included in the analysis. Treatments were performed by adding recombinant mouse TNFα [at the indicated concentrations in Fig. 1 and at 100 ng/ml (6 nm) for all other figures; Kamiya Biomedical] to a defined volume of conditioned growth media originating from the wells being treated and returning the plate to the incubator (37°C, 5% CO₂) for 45 min. Coverslips were then transferred to the recording chamber, and recordings were obtained within 75 min. For treatments with pharmacological inhibitors (Tocris Bioscience) of dynamin GTPase (dynasore, 80 μm), PP1/PP2A (okadaic acid, 0.5 μm), PP1 (tautomycetin, 10 nm), and PP2A (fostriecin, 10 nm), the inhibitor or its corresponding solvent was applied as above but for 55 min (i.e., 10 min before TNFα). For experiments examining multiple treatment conditions, control (Ctrl) and treated cells were recorded on the same day. Data from multiple recording days were then pooled to obtain group data for each experiment.

Figure 1.

TNFα downregulates inhibitory synaptic strength via dynamin GTPase activity. A, Representative traces of mIPSC recordings from untreated (Control) and TNFα-treated cultured neurons (100 ng/ml, 45 min). B, Superimposed average mIPSC traces from an untreated neuron (Ctrl, black trace) and a TNFα-treated neuron (gray trace), showing a reduction in mIPSC amplitude in response to TNFα (100 ng/ml, 45 min). C, Group data showing a reduction of average mIPSC amplitude (left) and mIPSC frequency (right) in response to various concentrations of TNFα applied for 45 min (n = 14–23 cells in each condition; one-way ANOVA, Tukey's post hoc test, *p < 0.05, **p < 0.01, ***p < 0.001). D, Cumulative distribution plots of mIPSC amplitudes (left) and interevent intervals (right) from untreated (black traces) and TNFα-treated (100 ng/ml, 45 min; gray traces) neurons (∼1200 events from each condition; **p < 0.005, ***p < 0.0001, Kolmogorov–Smirnov test). E, Pretreatment with the dynamin GTPase inhibitor dynasore (80 μm) blocks the TNFα-induced (100 ng/ml, 45 min) reduction of mIPSC amplitude (n = 16–18 cells in each condition; two-way ANOVA, Bonferroni's post hoc test, *p < 0.05; n.s., not significant).

Immunocytochemistry and treatments.

For surface GABA_AR immunostaining, cultured cells were live labeled by washing with ice-cold ACSF and incubating with primary antibody in ACSF on ice. The following primary antibodies were used to target the extracellular (N-terminal) domains of GABA_AR subunits: anti-GABA_AR α1 (1:400; catalog #AGA-001; Alomone Labs), anti-GABA_AR α2 (1:500; catalog #224103; Synaptic Systems), anti-GABA_AR α5 (1:1000; kind gift from Dr. J.-M. Fritschy, University of Zürich, Zürich, Switzerland), anti-GABA_AR β2/3 (1:100; catalog #MAB341; Millipore), and anti-GABA_AR γ2 (1:100; catalog #AGA-005; Alomone Labs). Cells were then washed three times with ice-cold ACSF and fixed in 2% paraformaldehyde solution for 15 min. Appropriate fluorescent secondary antibodies (1:1000; Alexa Fluor; Life Technologies) were then applied for 1 h at room temperature in a PBS solution containing 3% bovine serum albumin (BSA) and 2% normal goat serum (NGS). For double immunostaining of gephyrin and microtubule-associated protein 2 (MAP2), or glutamic acid decarboxylase 65 (GAD65) and MAP2, cells were washed with ice-cold ACSF, fixed in 2% paraformaldehyde for 15 min, permeabilized with 0.1% Triton X-100 for 3 min, blocked with PBS/3% BSA/2% NGS solution for 10 min, and incubated with primary antibodies in PBS/3% BSA/2% NGS for 2 h at room temperature. The following primary antibodies were used: anti-gephyrin (1:400; catalog #147011; Synaptic Systems), anti-GAD65/67 (1:500; GAD-6; Developmental Studies Hybridoma Bank, University of Iowa, Iowa City, IA), and anti-MAP2 (1:1000; catalog #188004; Synaptic Systems). For determining the percentage of cells expressing high levels of GABA_AR α1 that are GABAergic interneurons, surface α1 subunit was immunolabeled as above; cells were then fixed, permeabilized, and immunostained with anti GAD65/67 (GAD-6) as above. Treatments with TNFα (100 ng/ml) and pharmacological inhibitors were performed as for electrophysiology experiments. Inhibitors (Tocris Bioscience) were as follows: PKA (KT 5720 [(9S,10R,12R)-2,3,9,10,11,12-hexahydro-10-hydroxy-9-methyl-1-oxo-9,12-epoxy-1H-diindolo[1,2,3-fg:3′,2′,1′-kl]pyrrolo[3,4-i][1,6]benzodiazocine-10-carboxylicacid hexyl ester]; 1 μm), PKC [Gö 6983 (2-[1-(3-dimethylaminopropyl)-5-methoxyindol-3-yl]-3-(1H-indol-3-yl)maleimide); 1 μm], PI3K (wortmannin, 100 nm), p38 MAPK (SB 202190 [4-(4-fluorophenyl)-2-(4-hydroxyphenyl)-5-(4-pyridyl)1H-imidazole]; 10 μm), calcineurin (cyclosporin A, 20 μm), and PP1/PP2A (okadaic acid, 0.05 or 0.5 μm). TNFα receptor function was blocked by starting with a 10 min pretreatment with either anti-TNFR1 neutralizing antibody (5 μg/ml; catalog #MAB430; R&D Systems) or anti-TNFR2 neutralizing antibody (2 μg/ml; catalog #AF726; R&D Systems), before adding TNFα (100 ng/ml) for an additional 45 min.

Image acquisition and analysis.

All images were acquired blind to the experimental conditions, using identical acquisition parameters across all coverslips in each experiment. Images of surface GABA_AR immunostaining were captured using a CCD camera (Hamamatsu Orca-R2) mounted on an epifluorescence microscope [63×, 1.42 numerical aperture (NA); Olympus BX61]. All other images were acquired as single optical sections using a confocal microscope (63×, NA 1.42; Olympus FV1000). Each experimental condition was represented by two to three coverslips per experiment, and ∼10–20 images were acquired per coverslip. Image analysis was performed essentially as described previously (Stellwagen et al., 2005). Briefly, unadjusted whole images were analyzed using NIH ImageJ software, applying identical analysis parameters across all conditions within each experiment. For surface GABA_AR immunostaining, the area selected by a low threshold was used to approximate somatodendritic area, and a high threshold was used to select receptor puncta. High-threshold selections were analyzed using the “Analyze Particles” function, with particle size restricted to 1 μm² to exclude selections of large bright areas that do not represent receptor puncta. The total particle area of each image was divided by its corresponding low-threshold area to obtain values normalized to somatodendritic area. The resulting values were then expressed as percentage control to obtain “cluster area” as reported in Results. The same analysis was used for experiments in which cells were permeabilized and double immunostained for gephyrin plus MAP2 or GAD65 plus MAP2, except that (1) particle size was restricted to 2 μm² for gephyrin images and 4 μm² for GAD65 images and (2) somatodendritic area was determined from corresponding MAP2 images. Cluster density was obtained by normalizing particle count to somatodendritic area for each image and expressing as percentage control. “Average cluster size” is average particle size expressed as percentage control. Because cluster area is a measure of the summated puncta area per image, it represents a combined measure of cluster density and average cluster size.

Culture cell-surface biotinylation, immunoprecipitation, and Western blotting.

For surface biotinylation assays, cells were treated with TNFα (100 ng/ml) for 45 min in conditioned growth media, washed three times with ice-cold ACSF, and incubated for 30 min with 0.2 mg/ml Sulfo-NHS-SS-Biotin (Thermo Fisher Scientific) diluted in ice-cold ACSF. Cells were then washed three times with ice-cold ACSF containing 100 mm glycine and lysed in 20 mm HEPES, pH 7.4, 150 mm NaCl, 1% Triton X-100, 0.1% SDS, 1 mm EDTA, 1 mm PMSF, 50 mm NaF, 1 mm Na₃VO₄, and protease inhibitor mixture (BioShop Canada). Biotinylated proteins were isolated by incubating lysates with NeutrAvidin agarose resin (Thermo Fisher Scientific) for 1 h at 4°C. Biotinylated proteins were eluted with 2× SDS sample buffer and analyzed by SDS-PAGE along with corresponding total lysates.

For immunoprecipitation experiments cells were treated with TNFα (100 ng/ml) for 30 min, washed two times with ice-cold ACSF, and lysed in 20 mm HEPES, pH 7.4, 150 mm NaCl, 1% NP-40, 1 mm EDTA, 1 mm PMSF, 50 mm NaF, 1 mm Na₃VO₄, and protease inhibitor mixture (BioShop Canada). Lysates were precleared with protein A/G agarose beads (Thermo Fisher Scientific) and incubated with 2 μg of anti-GABA_AR β3 (catalog #75-149; Neuromab) per milligram of lysate protein. Precipitates were eluted with 2× SDS sample buffer and analyzed by SDS-PAGE together with corresponding unprecipitated lysates.

Gels were transferred onto PVDF membrane and immunoblotted with anti-GABA_AR β3 (1:500; catalog #75-149; Neuromab), anti-GABA_AR β3 (1:400; catalog #NB100-66169; Novus Biologicals), anti-GluA1 (1:1000; catalog #AB1504; Millipore), anti-phospho-p38 MAPK (1:1000; catalog #9215; Cell Signaling Technology), anti-total p38 MAPK (1:1000; catalog #9212; Cell Signaling Technology), anti-phospho Akt (1:2000; catalog #4060; Cell Signaling Technology), anti-total Akt (1:1000; catalog #4685; Cell Signaling Technology), anti-α-tubulin (1:5000; catalog #05-289; Millipore), anti-GABA_AR β3 phospho-S408/409 (1:250; catalog #612-401-D51; Rockland), anti-PP1α (1:1000; catalog #1950-1; Epitomics), and anti-PP2A (1:1000; catalog #1512-1; Epitomics). Quantitative analyses of Western blots were performed by determining band intensities on autoradiography film (Denville Scientific), using NIH ImageJ software. For surface biotinylation assays, “surface” quantifications depict surface/total ratios, and “total” quantifications depict total/α-tubulin ratios. For immunoprecipitation experiments quantifications depict coimmunoprecipitation PP1/β3 or PP2A/β3 ratios.

Acute hippocampal slice cell-surface biotinylation and Western blotting.

Coronal slices (300 μm thick) of the dorsal hippocampus were prepared using a vibratome (Leica VT1200S), from 6- to 9-week-old C57BL/6J WT mice (of either sex) and allowed to recover at ∼32°C for 30 min in ACSF containing 119 mm NaCl, 26.2 mm NaHCO₃, 11 mm glucose, 2.5 mm KCl, 1 mm NaH₂PO₄, 2.5 mm CaCl₂, and 1.3 mm MgCl₂, saturated with 95% O₂/5% CO₂. Slices were then incubated in the same solution, at room temperature, with TNFα (100 ng/ml, 1 h). Next, slices were incubated for 45 min in ice-cold ACSF solution containing 1 mg/ml Sulfo-NHS-SS-Biotin (Thermo Fisher Scientific). Slices were then washed three times with ice-cold ACSF containing 100 mm glycine and lysed in 20 mm HEPES, pH 7.4, 150 mm NaCl, 1% Triton X-100, 0.1% SDS, 1 mm EDTA, 1 mm PMSF, 50 mm NaF, 1 mm Na₃VO₄, and protease inhibitor mixture (BioShop Canada). Biotinylated proteins were isolated, SDS-PAGE was performed, PVDF membranes were immunoblotted, and quantifications were performed as for culture experiments.

Acute slice electrophysiology and treatments.

Coronal slices of the dorsal hippocampus were prepared and treated with TNFα as for cell-surface biotinylation experiments, from 5- to 8-week-old C57BL/6J WT mice (of either sex). DMSO (0.1%) or tautomycetin (10 nm, in DMSO) were added to the treatment chamber 20 min before adding TNFα. Slices were placed in a submersion-type recording chamber and perfused (∼1.5 ml/min) at 30°C with ACSF containing 119 mm NaCl, 26.2 mm NaHCO₃, 11 mm glucose, 2.5 mm KCl, 1 mm NaH₂PO₄, 2.5 mm CaCl₂, and 1.3 mm MgCl₂, saturated with 95% O₂/5% CO₂. Whole-cell patch-clamp recordings were obtained from hippocampal CA1 pyramidal neurons under visual guidance by infrared differential interference microscopy (Nikon Eclipse FN1). mIPSCs were pharmacologically isolated, recorded, and analyzed as for culture experiments.

Statistical analyses.

Statistical analysis and plotting were performed with Prism 5 (GraphPad Software). All data were obtained from a minimum of three independent experiments and presented as mean ± SEM.

Results

TNFα rapidly downregulates inhibitory synaptic strength via dynamin GTPase activity

To study the acute effects of TNFα on inhibitory neurotransmission, we recorded mIPSCs from hippocampal neuron cultures treated with various concentrations of TNFα for 45 min (Fig. 1A–D). Consistent with previous findings (Stellwagen et al., 2005), TNFα treatment significantly reduced average mIPSC amplitude, exerting a gradual effect that reached a plateau by 100 ng/ml (Ctrl, 56.8 ± 2.6 pA; TNFα at 10 ng/ml, 53.7 ± 4.6 pA; TNFα at 50 ng/ml, 44.52 ± 2.5 pA; TNFα at 100 ng/ml, 40.8 ± 3.4 pA; TNFα at 250 ng/ml, 43.3 ± 3.7 pA; Fig. 1C, left). In contrast, mIPSC frequency was significantly reduced starting at 10 ng/ml (Ctrl, 1.86 ± 0.22 Hz; TNFα at 10 ng/ml, 0.97 ± 0.17 Hz; TNFα at 50 ng/ml, 1.02 ± 0.15 Hz; TNFα at 100 ng/ml, 0.64 ± 0.10 Hz; TNFα at 250 ng/ml, 0.98 ± 0.09 Hz; Fig. 1C, right), suggesting that, at least in a mixed population of cultured hippocampal neurons, TNFα may act presynaptically to regulate neurotransmitter release probability at inhibitory synapses. For subsequent experiments, we focused on the 100 ng/ml dose because it exerts the strongest effect on quantal amplitude and it is the most commonly used concentration when studying TNFα-dependent receptor trafficking at hippocampal synapses (Beattie et al., 2002; Ogoshi et al., 2005; Stellwagen et al., 2005; Stellwagen and Malenka, 2006; Leonoudakis et al., 2008). We further determined that mIPSC 10–90% rise time (Ctrl, 2.06 ± 0.08 ms; TNFα, 2.26 ± 0.18 ms; Student's t test, p = 0.30) and decay time (Ctrl, 41.6 ± 0.95 ms; TNFα, 40.3 ± 1.10 ms; Student's t test, p = 0.40) were unaffected by TNFα treatment (100 ng/ml, 45 min), implying no substantial change in the subunit composition of synaptic GABA_ARs. As an additional control, we recorded cells treated with denatured (boiled) TNFα and observed no significant differences compared with untreated control cells (Ctrl, 55.4 ± 3.7 pA, 1.56 ± 0.28 Hz, n = 11 cells; boiled TNFα at 250 ng/ml, 53.8 ± 5.7 pA, 1.48 ± 0.27 Hz, n = 8 cells; Student's t test, p = 0.81 for amplitude and p = 0.86 for frequency).

The rapid reduction of mIPSC amplitude in response to TNFα could be attributed to reduced GABA_AR delivery to the synapse, reduced receptor stability at the synapse, or increased receptor endocytosis (Luscher et al., 2011). Because treatment with TNFα is known to cause a nearly twofold increase in the internalization of antibody-labeled GABA_ARs (Stellwagen et al., 2005), we suspected that the majority of the reduction in inhibitory synaptic strength would be endocytosis dependent. Endocytosis of GABA_ARs occurs primarily via an AP2/clathrin/dynamin-dependent mechanism (Kittler et al., 2000, 2005, 2008; Kanematsu et al., 2007); therefore, in a separate set of experiments, we pretreated cultures with the dynamin GTPase inhibitor dynasore (Macia et al., 2006) and measured mIPSC amplitudes in response to TNFα (Fig. 1E). The TNFα-induced reduction of mIPSC amplitude was blocked by dynasore (DMSO pretreatment: Ctrl, 50.4 ± 3.2 pA, 0.84 ± 0.15 Hz; TNFα, 39.1 ± 2.5 pA, 0.75 ± 0.10 Hz; dynasore pretreatment: Ctrl, 57.5 ± 1.9 pA, 3.51 ± 0.68 Hz; TNFα, 53.3 ± 3.9 pA, 4.05 ± 0.73 Hz), suggesting that TNFα reduces the strength of hippocampal inhibitory synapses primarily through GABA_AR endocytosis.

TNFα rapidly downregulates cell-surface levels of GABA_ARs

GABA_ARs consist of heteropentameric assemblies of diverse subunit composition, resulting in receptor subtypes that differ in expression, cell-surface localization, pharmacology, and function (Pirker et al., 2000; Möhler et al., 2004; Farrant and Nusser, 2005; Fritschy et al., 2012). At hippocampal synapses, GABA_ARs are composed of combinations of two α1–α3 subunits, two β1–β3 subunits, and one γ2 subunit (Brünig et al., 2002; Michels and Moss, 2007; Olsen and Sieghart, 2009; Kasugai et al., 2010). We used cell-surface immunostaining to determine whether TNFα preferentially endocytoses certain receptor subtypes (Fig. 2A). TNFα treatment (100 ng/ml, 45 min) reduced the surface cluster area of α1, α2, β2/3, and γ2 subunits by 25–30% compared with untreated Ctrl cells (α1, 75.9 ± 4.7% of Ctrl; α2, 71.4 ± 1.5% of Ctrl; β2/3, 71.7 ± 2.3% of Ctrl; γ2, 69.7 ± 2.1% of Ctrl; Fig. 2B). There was no obvious difference in the specificity of the effect among these subunits, likely because they are trafficked in tandem as common components of synaptically localized receptors. Consistent with previous reports (Fritschy et al., 1994; Brünig et al., 2002), we observed a differential basal expression of the α1 subunit, with ∼20% of cells expressing markedly higher levels of surface α1 subunit in our cultures. Consequently, we independently assessed the effects of TNFα on cells expressing high levels of the α1 subunit. We found that levels of surface α1 on high expressors were unaltered after TNFα treatment (102.1 ± 3.5% of Ctrl), suggesting that α1 high expressors define a subset of neurons unresponsive to TNFα. Previous reports have suggested that, in the hippocampus, higher expression of the α1 subunit is observed in interneurons (Fritschy et al., 1994; Brünig et al., 2002; Yu et al., 2006); however, based on colabeling of α1 and GAD, we observed that only ∼13% of high α1-expressing cells also expressed GAD, indicating that high α1-expressing cells in our cultures consisted of a mixed population of GABAergic interneurons and other cell types.

Figure 2.

TNFα downregulates cell-surface levels of GABA_ARs. A, Representative images of surface immunostaining for α1, α2, α5, β2/3, and γ2 GABA_AR subunits in untreated (Control) and TNFα-treated (100 ng/ml, 45 min) cultures. Scale bars: low magnification, 20 μm; high magnification, 3 μm. B, Quantification data of subunit cluster area corresponding to subunits represented in A, showing a significant reduction in surface GABA_ARs in response to TNFα treatment (gray bars), for all subunits examined (α1, n = 119–120, **p < 0.01; α2, n = 189–261, ***p < 0.001; α5, n = 110–116, *p < 0.05; β2/3, n = 210–258, ***p < 0.001; γ2, n = 316–420, ***p < 0.001; all comparisons by two-tailed Student's t test). C, Representative Western blots of surface biotinylation assays comparing control and TNFα-treated (100 ng/ml, 45 min) cultures. D, Quantification of surface biotinylation assays showing a decrease in surface GABA_AR β3 after TNFα treatment but no change in total β3 levels (n = 11 independent experiments; Mann–Whitney U test, *p < 0.05).

We also assessed the effects of TNFα treatment on surface levels of α5 subunit-containing receptor clusters, which are mainly located at extrasynaptic sites and primarily mediate tonic inhibition (Brünig et al., 2002; Caraiscos et al., 2004; Prenosil et al., 2006). We observed a much smaller but significant decrease in surface α5 cluster area (92.4 ± 1.98% of Ctrl; Fig. 2B), suggesting that tonic inhibition mediated by α5-containing receptors may not be reduced as robustly as phasic inhibition.

To further determine whether TNFα exerts its effects by increasing receptor endocytosis rather than by solely reducing receptor clustering, we used cell-surface biotinylation assays to measure cell-surface levels of GABA_AR β3 (Fig. 2C,D). TNFα treatment (100 ng/ml, 45 min) decreased surface GABA_AR β3 without affecting total levels (surface, 70.8 ± 12.9% of Ctrl; total, 99.3 ± 15.5% of Ctrl), providing additional evidence for increased receptor endocytosis.

A reduction of gephyrin clustering follows the reduction of surface GABA_ARs in response to TNFα

Next, we determined whether the effects of TNFα on GABA_AR trafficking persisted over a short time course and whether other functionally important components of the GABAergic synapse were also altered over this time course. Using cell-surface immunostaining, we observed that surface GABA_AR γ2 cluster area gradually decreased in the presence of TNFα, reaching a minimum at 3 h (62.0 ± 3.3% of Ctrl), followed by a partial recovery at 6 h (79.5 ± 4.4% of Ctrl; Fig. 3A,B). This indicated that the effects of TNFα were rapid and persistent but that a recovery may eventually occur, possibly as a result of ligand-induced TNFR internalization (Mosselmans et al., 1988; Higuchi and Aggarwal, 1994) or proteolytic degradation of the exogenously supplied TNFα (Vey et al., 1996). To determine whether over this time course TNFα reduces surface levels of GABA_ARs by mechanisms other than receptor trafficking (e.g., via reduced receptor protein synthesis or increased receptor degradation), we measured total levels of GABA_AR β3 after 3 and 6 h of TNFα treatment. We observed no significant differences compared with untreated control cells (TNFα at 100 ng/ml for 3 h, 97.0 ± 0.8% of Ctrl; TNFα at 100 ng/ml for 6 h, 96.8 ± 3.5% of Ctrl; n = 3 independent experiments; Kruskal–Wallis one-way ANOVA, p = 0.24).

Figure 3.

TNFα-induced downregulation of surface GABA_ARs precedes a downregulation of gephyrin clustering, with no change in GAD65 puncta. A, Representative images from time course experiments with cultured neurons undergoing TNFα treatment ranging from 10 min to 6 h in duration. Top row shows a rapid downregulation in surface levels of the γ2 subunit in response to TNFα. Middle row represents a matching set of experiments in which cells were permeabilized and immunostained for gephyrin (white) and MAP2 (blue), showing a slower downregulation of cluster area than for γ2. Bottom row represents another matching set of experiments in which cells were permeabilized and immunostained for GAD65 (white) and MAP2 (blue), showing no change in puncta area. Scale bars: low magnification, 20 μm; high magnification, 1.5 μm. B, Quantification data of surface γ2 cluster area, over the 6 h time course of TNFα treatment (n = 120–158 images per time point; one-way ANOVA, Tukey's post hoc test, **p < 0.01, ***p < 0.001). C, Quantification data of gephyrin cluster area normalized to MAP2 area (white boxes). MAP2 area (dark circles) was unaltered (n = 65–82 images per time point; one-way ANOVA, Tukey's post hoc test, *p < 0.05, ***p < 0.001). D, Quantification data of GAD65 puncta area normalized to MAP2 area (white boxes). MAP2 area (dark circles) was unaltered (n = 80–112 images per time point; one-way ANOVA, Tukey's post hoc test, no significant differences).

We hypothesized that exposure to TNFα may also affect clustering of gephyrin—a postsynaptic scaffolding protein present at most inhibitory synapses (Tretter et al., 2012). Multiple lines of evidence indicate that gephyrin contributes significantly to the stabilization of GABA_ARs at postsynaptic sites (Kneussel et al., 1999; Lévi et al., 2004; Yu et al., 2007; Tretter et al., 2008; Mukherjee et al., 2011). We observed a significant downregulation of gephyrin cluster area at 3 h (84.5 ± 4.1% of Ctrl) and 6 h (66.6 ± 3.2% of Ctrl) of TNFα treatment, whereas shorter treatments of 10 and 45 min elicited only small, nonsignificant decreases (96.4 ± 3.6% of Ctrl at 10 min; 93.0 ± 4.2% of Ctrl at 45 min; Fig. 3A,C). This effect was primarily attributable to reduced cluster density (100.3 ± 3.5% of Ctrl at 10 min; 95.7 ± 4.2% of Ctrl at 45 min; 85.2 ± 3.9% of Ctrl at 3 h; 70.1 ± 1.2% of Ctrl at 6 h) rather than reduced average cluster size (96.4 ± 1.2% of Ctrl at 10 min; 97.5 ± 1.5% of Ctrl at 45 min; 99.5 ± 3.3% of Ctrl at 3 h; 95.7 ± 2.4% of Ctrl at 6 h). Together, these findings suggest that GABA_AR clusters are initially lost mainly as a result of endocytosis but that, after longer exposures to TNFα (>3 h), a reduction in gephyrin clustering may also destabilize synaptic GABA_ARs.

Because we observed a decrease in mIPSC frequency (Fig. 1C), we asked whether correlates of GABA synthesis or the number of GABA release sites could be altered by TNFα. To address this possibility, we immunostained for GAD65—the rate-limiting enzyme in the production of GABA from glutamate at nerve terminals (Buddhala et al., 2009). GAD65 is aggregated in puncta at inhibitory presynaptic terminals and its expression is subject to activity-dependent homeostatic regulation (Rannals and Kapur, 2011; Lau and Murthy, 2012). Over the same 6 h time course of TNFα treatment as for surface GABA_AR γ2 and gephyrin, we observed no significant changes in GAD65 puncta area (95.1 ± 3.8% of Ctrl at 10 min; 99.4 ± 3.3% of Ctrl at 45 min; 105.3 ± 4.4% of Ctrl at 3 h; 94.6 ± 4.1% of Ctrl at 6 h; Fig. 3A,D), nor any significant changes in puncta density (95.9 ± 3.6% of Ctrl at 10 min; 106.2 ± 3.7% of Ctrl at 45 min; 114.1 ± 4.6% of Ctrl at 3 h; 105.7 ± 4.7% of Ctrl at 6 h) or average puncta size (102.0 ± 2.5% of Ctrl at 10 min; 97.9 ± 2.3% of Ctrl at 45 min; 95.0 ± 1.9% of Ctrl at 3 h; 94.1 ± 2.4% of Ctrl at 6 h). Therefore, the mIPSC frequency decrease caused by TNFα is unlikely to be due to a decrease in the number of GABA release sites. As well, because average puncta size was not affected, TNFα does not appear to affect the production of GABA at synaptic sites.

We costained for MAP2 to normalize gephyrin and GAD65 measurements to somatodendritic area, as well as to assess somatodendritic integrity over a prolonged exposure to TNFα. We observed no significant differences in MAP2 area in either set of experiments (Fig. 3C,D, plotted as “MAP2 area only”), indicating that TNFα does not exert any overt deleterious effects on neuronal structure and integrity under our conditions.

TNFα signals through neuronal TNFR1, p38 MAPK, PI3K, and PP1 to downregulate surface GABA_AR levels and GABAergic neurotransmission

To identify signal transduction mechanisms, we first examined the relative contribution of TNFR1 versus TNFR2 signaling to TNFα-dependent GABA_AR trafficking. Both TNFRs are constitutively expressed throughout the CNS by neurons and glia (Dopp et al., 1997; Zhang et al., 2010; Brambilla et al., 2011). Knowing that AMPAR trafficking in response to TNFα occurs via neuronal TNFR1 (Stellwagen et al., 2005; He et al., 2012) and that soluble TNFα preferentially signals through TNFR1 (McCoy and Tansey, 2008), we predicted that TNFR1 would be required in this case. Indeed, when selectively blocking either TNFR1 or TNFR2 using neutralizing antibodies, we observed that blockade of TNFR1 signaling prevented the TNFα-induced downregulation of surface GABA_AR γ2 cluster area (109.2 ± 4.7% of TNFR1 control), whereas blockade of TNFR2 did not (81.4 ± 4.0% of TNFR2 control; Fig. 4A,B). Because TNFα can also control the release of glutamate from astrocytes (Domercq et al., 2006; Santello et al., 2011), we assessed whether the effect occurred via direct activation of neuronally expressed TNFR1 or via glial TNFR1 activation followed by release of a glial factor that would control GABA_AR trafficking. To this end, we prepared Banker cocultures (Kaech and Banker, 2006) using WT or TNFR1^−/− glial feeder layers in combination with either WT or TNFR1^−/− neurons. When using a WT glial feeder layer, mIPSC amplitude was reduced in WT neurons after TNFα treatment (Ctrl, 59.5 ± 5.2 pA; TNFα, 31.4.0 ± 3.6 pA) but was unaffected in TNFR1^−/− neurons (Ctrl, 59.3 ± 7.1 pA; TNFα, 58.2 ± 4.0 pA; Fig. 4C,D). When using a TNFR1^−/− glial feeder layer, we still observed a reduction in mIPSCs recorded from WT neurons (Ctrl, 56.7 ± 3.6 pA; TNFα, 36.8 ± 3.2 pA; Fig. 4C,D). Therefore, we conclude that neuronal TNFR1 controls GABA_AR trafficking.

Signal transduction via TNFR1 involves assembly of adaptor proteins in a signaling complex around the intracellular “death domain” of the receptor, resulting in activation of various downstream targets and various cellular responses, depending on cell type and cell state (Sriram and O'Callaghan, 2007; McCoy and Tansey, 2008; Zhang et al., 2009). However, little is known about the intracellular substrates of TNFR1 that control receptor trafficking in neurons; only PI3K is known to be required for increasing surface levels of AMPARs (Stellwagen et al., 2005). Therefore, using pharmacological inhibitors, we sought to identify such signaling components, aiming to find a link between TNFα signaling and mechanisms mediating GABA_AR endocytosis. Given the rapid effects of TNFα on surface levels of GABA_ARs, we focused on inhibitors of kinases and phosphatases known to regulate GABA_AR trafficking (Luscher et al., 2011) or known to be downstream of TNFR1 (Grivennikov et al., 2006; Zhang et al., 2010). We found that blocking PKA (with KT 5720) or PKC (with Gö 6983)—kinases that phosphorylate GABA_ARs (McDonald et al., 1998; Brandon et al., 2000, 2003)—did not alter the effects of TNFα (KT 5720 alone, 100.0 ± 3.9% of DMSO alone; KT 5720 + TNFα, 59.1 ± 2.8% of DMSO alone; Gö 6983 alone, 100.0 ± 6.6% of DMSO alone; Gö 6983 + TNFα, 73.7 ± 3.3% of DMSO alone; Fig. 5A,B). However, preincubation with inhibitors of PI3K (wortmannin) or p38 MAPK (SB 202190) blocked the TNFα-induced reduction in surface γ2 subunit cluster area (wortmannin alone, 73.5 ± 5.4% of DMSO alone; wortmannin + TNFα, 83.1 ± 3.9% of DMSO alone; SB 202190 alone, 68.3 ± 6.6% of DMSO alone; SB 202190 + TNFα, 71.7 ± 4.1% of DMSO alone; Fig. 5A,B). Because wortmannin alone and SB 202190 alone reduced γ2 cluster area, it is not clear whether an additional decrease in response to TNFα was occluded in these cases. We targeted PP1 and PP2A using two different concentrations of okadaic acid (0.05 vs 0.5 μm), because lower concentrations are known to preferentially block PP2A, whereas higher concentrations also block PP1 (Cohen et al., 1989; Surmeier et al., 1995; Suh et al., 2013). Interestingly, we found that we could block the effect of TNFα with 0.5 μm okadaic acid but not with 0.05 μm, suggesting that PP1 but not PP2A is required (0.05 μm okadaic acid alone, 103.2 ± 5.9% of DMSO alone; 0.05 μm okadaic acid + TNFα, 77.5 ± 5.5% of DMSO alone; 0.5 μm okadaic acid alone, 96 ± 5.8% of DMSO alone; 0.5 μm okadaic acid + TNFα, 98.3 ± 4.7% of DMSO alone; Fig. 5A,B). Lastly, we found that calcineurin activity (blocked with cyclosporin A) was not required in this signaling pathway (cyclosporin A alone, 90.2 ± 9.1% of DMSO alone; cyclosporin A + TNFα, 60.2 ± 3.3% of DMSO alone). Because PP1/PP2A activity has been implicated previously in GABA_AR endocytosis (Jovanovic et al., 2004; Kanematsu et al., 2006, 2007; Lin et al., 2011) whereas calcineurin activity has been implicated in GABA_AR cluster dispersal (Bannai et al., 2009; Muir et al., 2010), our findings are again consistent with an endocytosis-dependent effect.

Figure 5.

Downregulation of surface GABA_AR γ2 levels in response to TNFα is blocked by inhibitors of PI3K, p38 MAPK, and PP1/PP2A. A, Representative dendritic regions of surface immunostaining for GABA_AR γ2 subunit, from control (Ctrl) versus TNFα-treated (100 ng/ml, 45 min) cultured neurons, preincubated with DMSO (solvent, 0.1%) or inhibitors of PKA [KT 5720 (KT), 1 μm], PKC [Gö 6983 (Gö), 1 μm], PI3K [wortmannin (Wortm), 100 nm], p38 MAPK [SB 202190 (SB), 10 μm], calcineurin [cyclosporin A (CsA), 20 μm], and PP1/PP2A (okadaic acid (OKA), 0.05 or 0.5 μm). Scale bar, 3 μm. B, Quantification data corresponding to conditions in A, indicating that TNFα-induced downregulation of surface γ2 cluster area is blocked by preincubation with wortmannin, SB 2022190, and 0.5 μm okadaic acid but not KT 5720, Gö 6983, cyclosporin A, or 0.05 μm okadaic acid (n = 60–228 images in each condition; two-way ANOVA, Bonferroni's post hoc test, *p < 0.05, **p < 0.01, ***p < 0.001; n.s., not significant).

To define which phosphatase is required for TNFα-dependent trafficking of synaptic GABA_ARs, we assessed the effects of TNFα on mIPSC amplitude after simultaneous blockade of PP1 and PP2A (with 0.5 μm okadaic acid) versus selective blockade of PP1 (with tautomycetin; Mitsuhashi et al., 2001) or PP2A (with fostriecin; Walsh et al., 1997). Consistent with immunostaining experiments, inhibition of PP1/PP2A blocked the TNFα-induced decrease in mIPSC amplitude (DMSO alone, 56.8 ± 3.9 pA; DMSO + TNFα, 43.3 ± 3.8 pA; okadaic acid alone, 62.3 ± 4.7 pA; okadaic acid + TNFα, 58.1 ± 4.4 pA; Fig. 6A,B). Selective inhibition of PP1 also blocked the effects of TNFα (tautomycetin alone, 44.4 ± 4.8 pA; tautomycetin + TNFα, 55.4 ± 4.0 pA), whereas selective inhibition of PP2A did not [dimethyl formamide (DMF) alone, 50.5 ± 3.2 pA; DMF + TNFα, 35.2 ± 2.2 pA; fostriecin alone, 51.3 ± 5.1 pA; fostriecin + TNFα, 32.5 ± 2.8 pA]. Interestingly, the effects of TNFα on mIPSC frequency were reversed by okadaic acid and tautomycetin but unaffected by fostriecin (DMSO alone, 0.80 ± 0.10 Hz; DMSO + TNFα, 0.59 ± 0.07 Hz; okadaic acid alone, 1.45 ± 0.18 Hz; okadaic acid + TNFα, 1.79 ± 0.23 Hz; tautomycetin alone, 0.84 ± 0.14 Hz; tautomycetin + TNFα, 1.42 ± 0.22 Hz; DMF alone, 1.35 ± 0.24 Hz; DMF + TNFα, 0.84 ± 0.19 Hz; fostriecin alone, 1.39 ± 0.21 Hz; fostriecin + TNFα, 0.78 ± 0.13 Hz). We conclude that TNFα downregulates GABAergic neurotransmission through PP1 but not PP2A activity.

Figure 6.

TNFα downregulates GABAergic neurotransmission through PP1 but not PP2A. A, mIPSCs were recorded from control or TNFα-treated (100 ng/ml, 45 min) cultured neurons (preincubated with solvent alone or inhibitor). DMSO (0.1%) was used as a solvent for okadaic acid (OKA) and tautomycetin (TMT); DMF (0.1%) was used as a solvent for fostriecin (FOST). Shown are representative traces for control with DMSO preincubation, TNFα-treated with DMSO or DMF preincubation, TNFα-treated with okadaic acid preincubation (PP1/PP2A inhibitor, 0.5 μm), TNFα-treated with tautomycetin preincubation (PP1 inhibitor, 10 nm), and TNFα-treated with fostriecin preincubation (PP2A inhibitor, 10 nm). B, Group data of average mIPSC amplitudes for all conditions, showing that preincubation with okadaic acid or tautomycetin blocks the TNFα-induced downregulation of mIPSC amplitude, whereas preincubation with fostriecin does not (n = 16–25 cells in each condition; two-way ANOVA, Bonferroni's post hoc test, *p < 0.05, **p < 0.01, ***p < 0.001; n.s., not significant).

TNFα dephosphorylates GABA_AR β3 S408/409 by sequentially activating p38 MAPK and PI3K while enhancing the association of PP1α with β3

To further characterize the signaling pathway linking neuronal TNFR1 to GABA_AR trafficking, we focused on the three molecular components we had identified using pharmacological inhibition: p38 MAPK, PI3K, and PP1. We used Western blots to assess activation of p38 and PI3K using antibodies specific for phospho-p38 MAPK and phospho-Akt—a downstream target of PI3K activation. Phosphorylation of both p38 MAPK and Akt was significantly upregulated in response to TNFα (100 ng/ml, 45 min; phospho-p38: DMSO + TNFα, 150.0 ± 9.8% of DMSO alone; phospho-Akt: DMSO + TNFα, 182.5 ± 34.6% of DMSO alone; Fig. 7A,B), indicating that both signaling components are activated. In addition, using pharmacological blockade of p38 MAPK (using SB 202190) or PI3K (using wortmannin), we determined that TNFα-induced phosphorylation of p38 MAPK is unaffected by wortmannin (wortmannin alone, 108.0 ± 2.6% of DMSO alone; wortmannin + TNFα, 160.5 ± 8.5% of DMSO alone; Fig. 7A,B), whereas TNFα-induced phosphorylation of Akt is blocked by SB 202190 (SB alone, 83.5 ± 12.5% of DMSO alone; SB + TNFα, 59.6 ± 21.9% of DMSO alone; Fig. 7A,B), indicating that p38 MAPK activation occurs upstream of PI3K activation in this signaling pathway.

Figure 7.

TNFα dephosphorylates GABA_AR β3 S408/409 by sequentially activating p38 MAPK and PI3K and enhancing the association of PP1α with β3 subunits. A, Representative Western blots of phospho- p38 MAPK, total p38 MAPK, phospho-Akt, and total Akt in lysates from control or TNFα-treated (100 ng/ml, 45 min) cultures, preincubated with DMSO (solvent, 0.1%) or inhibitors of PI3K [wortmannin (Wm), 100 nm] and p38 MAPK [SB 202190 (SB), 10 μm]. B, Quantification data showing an increase in phospho-p38 MAPK with TNFα treatment, unaffected by PI3K inhibition (left), and an increase in phosphorylation of the PI3K downstream effector Akt, blocked by p38 MAPK inhibition (right) (n = 4–5 independent experiments; Mann–Whitney U test, *p < 0.05, **p < 0.01; n.s., not significant). C, Representative Western blots of phospho-S408/409 GABA_AR β3, total GABA_AR β3, and α-tubulin in lysates from control or TNFα-treated (100 ng/ml, 45 min) cultures, preincubated with DMSO (solvent, 0.1%) or inhibitors of PI3K [wortmannin (Wm), 100 nm], p38 MAPK [SB 202190 (SB), 10 μm], PP1/PP2A [okadaic acid (OKA), 0.5 μm], PP1 [tautomycetin (TMT), 10 nm], and PP2A [fostriecin (FOST), 10 nm]. D, Quantification data showing a decrease in GABA_AR β3 phospho-S408/409, blocked by inhibition of PI3K, p38 MAPK, and PP1, but not by inhibition of PP2A (n = 5 independent experiments; Mann–Whitney U test, **p < 0.01, ***p < 0.001; n.s., not significant). E, Representative Western blots of coimmunoprecipitation experiments in which the GABA_AR β3 subunit was immunoprecipitated (IP) from untreated (Ctrl) or TNFα-treated (100 ng/ml, 30 min) cultures; blots were then probed for GABA_AR β3 and for the catalytic subunits of PP1 (PP1α) and PP2A (PP2Ac). F, Quantification showing that TNFα treatment enhances the association of PP1α but not PP2Ac with β3 subunits (n = 5–6 independent experiments; Mann–Whitney U test, **p < 0.01; n.s., not significant).

Knowing that PP1 can localize to inhibitory synapses (Bausen et al., 2010) and that dephosphorylation of GABA_AR β3 should favor receptor endocytosis (Kittler et al., 2005; Smith et al., 2012), we hypothesized that PP1 acts downstream of p38 MAPK and PI3K to directly dephosphorylate GABA_AR β3. We assessed the phosphorylation state of β3 subunits using a phospho-specific antibody detecting phosphorylated β3 S408/409 (Fig. 7C). We focused on this site for two reasons. First, whereas other serine sites on GABA_ARs (e.g., γ2 S327) are known to be regulated by calcineurin (Bannai et al., 2009; Muir et al., 2010), β3 S408/409 is the only site known to be regulated by PP1 (for review, see Luscher et al., 2011). Second, this site is located in the β1–β3 intracellular domains, in a basic patch sorting motif that directly binds the μ2 subunit of the clathrin adaptor AP2 to mediate dynamin-dependent endocytosis (Kittler et al., 2005; Smith et al., 2012). Phosphorylation of this site disrupts the interaction with μ2–AP2 and increases synaptic GABA_AR content (Kittler et al., 2005). We observed a significant reduction in phospho-S408/409 in response to TNFα treatment (100 ng/ml, 45 min; DMSO + TNFα, 62.3 ± 3.7% of DMSO alone; Fig. 7C,D), consistent with an increased rate of GABA_AR endocytosis and a loss of surface GABA_AR β3. However, pretreatment with either 0.5 μm okadaic acid or tautomycetin, but not with fostriecin, blocked TNFα-induced β3 S408/409 dephosphorylation (Fig. 7C,D), indicating that this is a PP1-mediated dephosphorylation (okadaic acid + TNFα, 91.0 ± 6.7% of okadaic acid alone; tautomycetin + TNFα, 113.6 ± 14.5% of tautomycetin alone; fostriecin + TNFα, 55.4 ± 16.1% of fostriecin alone; Fig. 7D). Furthermore, inhibition of p38 MAPK or PI3K also blocked TNFα-induced β3 S408/409 dephosphorylation (Fig. 7C,D), indicating that activation of p38 MAPK and PI3K is required for receptor dephosphorylation (wortmannin + TNFα, 112.5 ± 7.2% of wortmannin alone; SB 202190 + TNFα, 101.6 ± 5.8% of SB 202190 alone; Fig. 7D).

Previous studies have suggested that an increased abundance of protein phosphatase associated with the GABA_AR protein complex correlates with increased dephosphorylation of β3 and subsequent endocytosis of GABA_ARs (Jovanovic et al., 2004; Kanematsu et al., 2006). The association of PP1 and PP2A with the β3 subunit is facilitated by the adaptor/regulatory proteins PRIP-1 and PRIP-2, which selectively bind to the intracellular domains of β1–β3 subunits but not α1–α3 or γ2 subunits (Terunuma et al., 2004; Kanematsu et al., 2006, 2007; Yanagihori et al., 2006). We used coimmunoprecipitation to determine whether the association of β3 with the catalytic subunits of PP1 (PP1α) or PP2A (PP2Ac) was regulated by TNFα (Fig. 7E,F). Consistent with mIPSC recordings showing a requirement of PP1 but not PP2A activity, we observed an increase in the association of PP1α with β3 but no change in the association of PP2Ac with β3 (PP1α, 180.7 ± 28.3% of Ctrl; PP2Ac, 94.6 ± 13.1% of Ctrl; Fig. 7F). This suggests that, in response to TNFα, PP1 is targeted to the intracellular domain of the β3 subunit, which may enable it to dephosphorylate key residues that regulate interactions with the endocytic machinery.

A similar TNFα signaling pathway regulates GABAergic synapses in acute hippocampal slices

To extend our findings from cultured hippocampal neurons to a more intact hippocampal preparation, we treated acute hippocampal slices from young adult mice with TNFα (100 ng/ml, 1 h) and assessed activation of the p38 MAPK–PI3K–β3 S408/409 signaling pathway. In agreement with our results from culture experiments, we observed a significant increase in phospho-p38 MAPK (179.8 ± 20.2% of Ctrl), a trend for increased phospho-Akt (144.3 ± 20.7% of Ctrl) and a significant decrease in β3 S408/409 phosphorylation (68.6 ± 8.9% of Ctrl; Fig. 8A,B). Similarly, surface GABA_AR β3 was decreased (67.5 ± 12.9% of Ctrl), whereas, in accordance with previous findings (Stellwagen et al., 2005), surface GluA1 was increased (150.1 ± 20.53% of Ctrl; Fig. 8A,B). We observed no changes in total p38 MAPK, total Akt, total GABA_AR β3, or total GluA1 (p38 MAPK, 102.7 ± 6.1% of Ctrl; Akt, 87.5 ± 9.7% of Ctrl; β3, 103.7 ± 6.2% of Ctrl; GluA1, 108.9 ± 9.4% of Ctrl; n = 8–9 animals in each condition; Mann–Whitney U test, p > 0.05 for each). To determine whether in acute hippocampal slices TNFα (100 ng/ml, >1 h) decreases synaptic GABA_AR content in a PP1-dependent manner, we recorded mIPSCs from CA1 pyramidal neurons, in slices pretreated with either DMSO or the PP1-specific inhibitor tautomycetin (Fig. 8C). Consistent with results from culture recordings (Fig. 6), TNFα treatment caused a significant decrease in mIPSC amplitude that was blocked by pretreatment with tautomycetin (DMSO alone, 51.8 ± 3.2 pA; DMSO + TNFα, 39.8 ± 2.7 pA; tautomycetin alone, 48.4 ± 2.0 pA; tautomycetin + TNFα, 47.1 ± 2.5 pA; Fig. 8C,D). However, we did not observe any differences in mIPSC frequency (DMSO alone, 10.6 ± 0.9 Hz; DMSO + TNFα, 10.8 ± 1.4 Hz; tautomycetin alone, 8.1 ± 0.7 Hz; tautomycetin + TNFα, 8.5 ± 0.7 Hz; n = 10–13 cells; two-way ANOVA, Bonferroni's post hoc test, p > 0.05), suggesting that the effect observed in culture (Fig. 1) is either culture specific or does not occur at GABAergic synapses onto CA1 pyramidal neurons.

Figure 8.

A similar TNFα signaling pathway regulates inhibitory synapses in acute hippocampal slices. A, Representative Western blots of phospho-p38 MAPK, total p38 MAPK, phospho-Akt, total Akt, phospho-S408/409 GABA_AR β3, surface (biotinylated) GABA_AR β3, total lysate GABA_AR β3, surface (biotinylated) GluA1, total lysate GluA1, and α-tubulin, from acute slices of mouse dorsal hippocampus, either untreated (Ctrl) or treated with TNFα (100 ng/ml, 1 h). B, Quantification data corresponding to experiments represented in A, showing a TNFα-induced increase in phospho-p38 MAPK, an increase in phospho-Akt, a decrease in phospho-S408/409 β3, a decrease in surface levels of GABA_AR β3, and an increase in surface levels of GluA1 (n = 8–9 animals in each condition; Mann–Whitney U test, **p < 0.01). C, Sample traces of mIPSCs recorded from acute hippocampal slice CA1 pyramidal neurons, either control or TNFα-treated (100 ng/ml, >1 h), preincubated with either DMSO (0.1%) or tautomycetin (TMT, 10 nm in DMSO). D, Group data of average mIPSC amplitudes, showing that preincubation with tautomycetin blocks the TNFα-induced mIPSC amplitude reduction (n = 10–13 cells in each condition; two-way ANOVA, Bonferroni's post hoc test, **p < 0.01; n.s., not significant).

Discussion

In this study, we characterized the effects of TNFα at GABAergic hippocampal synapses, defining the signaling pathway through which it controls GABA_AR endocytosis. We demonstrated that TNFα rapidly downregulates GABAergic synaptic strength and surface GABA_ARs via neuronal TNFR1 activation, p38 MAPK, PI3K, PP1, and dynamin GTPase activity. This is accompanied by an enhanced association of PP1α with GABA_AR β3 and dephosphorylation of β3 S408/409. Furthermore, on a longer timescale, gephyrin cluster density is also reduced. Together, these findings support a model whereby TNFα rapidly and persistently tunes down inhibition by enhancing GABA_AR endocytosis through PP1-dependent dephosphorylation of β3 subunits, which is then compounded by a loss of synaptic scaffolding support for GABA_ARs (Fig. 9).

Figure 9.

Proposed model for TNFα-induced downregulation of GABAergic neurotransmission at hippocampal synapses. Compared with normal conditions (left), conditions associated with elevated levels of TNFα (right) lead to diminished postsynaptic responses to GABA as a result of lower levels of synaptic GABA_ARs. Activation of TNFR1 signals through p38 MAPK, PI3K, and PP1, which in turn dephosphorylates β3 S408/409 to enhance GABA_AR endocytosis. A reduction in synaptic GABA_ARs is followed by a reduction in gephyrin clustering. TNFα may also affect presynaptic function by reducing synaptic vesicle release probability at some synapses, without affecting GABA synthesis.

A large number of functionally distinct GABA_AR subtypes are assembled from various subunit combinations (Olsen and Sieghart, 2009). Our results indicate that receptors containing subunits α1, α2, β2/3, and γ2—which are preferentially localized to synapses (Mangan et al., 2005; Michels and Moss, 2007)—are all reduced by 25–30% on the cell surface. Therefore, despite the differential subcellular distribution of α1- and α2-containing receptors in vivo (Prenosil et al., 2006; Thomson and Jovanovic, 2010; Panzanelli et al., 2011), TNFα has similar effects on both receptor subtypes. In agreement with this, we observed no significant difference in mIPSC decay kinetics, which would be expected to change with an altered ratio of α1- to α2-containing receptors remaining at the synapse (Okada et al., 2000; Bosman et al., 2005; Schneider Gasser et al., 2007). Furthermore, we observed a full shift in the cumulative distribution of mIPSC amplitudes, suggesting that all synapses were affected to a similar extent. Such a broad effect on synaptic receptor subtypes is probably achieved by controlling endocytosis through common components, such as the β subunits. However, because we observed that surface levels of clustered α5-containing receptors—which also coassemble with β subunits (Glykys and Mody, 2007)—are only modestly decreased, it is possible that some level of specificity is achieved at the subcellular level by differential activation of signaling molecules present at synaptic versus extrasynaptic compartments. Furthermore, some cell-type specificity of the effect is also likely because TNFα did not reduce surface levels of α1 subunits in a subpopulation of cells expressing high levels of α1. Although we are not certain what neuron subtype(s) this population represents, it probably consists of a mixture of hippocampal interneurons and a small fraction of subicular neurons introduced during the dissection procedure; both of these neuron types are known to express higher levels of the α1 subunit (Brünig et al., 2002; Yu et al., 2006; Schneider Gasser et al., 2007). It is possible that such neuron subtypes do not express some necessary component of the signaling pathway linking TNFα to GABA_AR trafficking.

In addition to GABA_AR content at the synapse, TNFα also regulated clustering of gephyrin, a scaffolding protein that helps maintain a high density of GABA_ARs at the inhibitory synapse by directly interacting with GABA_AR subunits (Tretter et al., 2012; Kowalczyk et al., 2013). We determined that, whereas surface γ2 clusters were rapidly reduced within 45 min, gephyrin clustering was reduced more slowly, reaching a minimum after 3 h. Because loss of gephyrin is known to disrupt GABA_AR clustering by increasing receptor lateral mobility (Jacob et al., 2005; Yu et al., 2007), the TNFα-induced slow reduction in gephyrin clustering is likely to destabilize synaptic GABA_AR clusters, thus serving as a second mechanism via which TNFα weakens inhibitory synapses. We suspect that the decrease in gephyrin clustering occurs in response to the initial reduction in surface levels of GABA_ARs. This would be consistent with reports documenting a loss of gephyrin clusters in α1–α3 and γ2 knock-out mice (Essrich et al., 1998; Schweizer et al., 2003; Studer et al., 2006; Patrizi et al., 2008; Panzanelli et al., 2011). However, we cannot exclude the possibility that a separate TNFα signaling pathway directly regulates gephyrin cluster density, perhaps by regulating phosphorylation of gephyrin (Tyagarajan et al., 2011).

TNFα is known to traffic AMPARs to the surface via neuronal TNFR1 (Stellwagen et al., 2005), whereas astrocytic TNFR1 can control the release of glutamate from astrocytes (Domercq et al., 2006). We now show that activation of neuronally expressed TNFR1 is necessary for the effects of TNFα on GABA_AR trafficking, whereas astrocytic TNFR1 does not play a role in this pathway. Therefore, it is apparent that TNFα controls neurotransmitter receptor trafficking by acting directly on neurons rather than by acting on glial TNFR1 to induce the release of a glial factor that would subsequently signal to neurons. Furthermore, the divergence in signaling pathways controlling AMPARs exocytosis versus GABA_AR endocytosis must occur downstream of neuronal TNFR1. We provide evidence that p38 MAPK and PI3K are sequentially activated by TNFα and are required for TNFα-dependent GABA_AR endocytosis. Through activation of Akt, PI3K is known to regulate insulin-dependent translocation of GABA_ARs to the neuronal cell surface (Wang et al., 2003). However, PI3K can also control PP1 activity (Ragolia and Begum, 1998; De Luca et al., 1999), suggesting that it can indirectly lead to receptor dephosphorylation.

Using several experimental approaches, we identified PP1 as a mediator of TNFα-dependent GABA_AR trafficking in cultured hippocampal neurons and acute hippocampal slice CA1 pyramidal neurons. In addition, we have shown that, in response to TNFα treatment, GABA_AR β3 is dephosphorylated at S408/409 via a signaling pathway that requires activation of p38 MAPK, PI3K, and PP1 but not PP2A. These data support a model of regulated endocytosis occurring via PP1-dependent dephosphorylation of β3 subunits, which enhances the proportion of β3 subunits bound to AP2 and consequently enhances dynamin-dependent endocytosis of GABA_ARs. Therefore, TNFα appears to engage a mechanism shown previously to mediate constitutive endocytosis of GABA_ARs through phospho-dependent regulation of the β3–AP2 interaction (Kittler et al., 2005; Smith et al., 2012). A similar mechanism of constitutive endocytosis has been described based on phosphorylation of the γ2 subunit at Y365/367 (Kittler et al., 2008). Although we do not rule out TNFα-dependent regulation of γ2 phosphorylation, we suspect that this does not play a major role because we can completely block the effects of TNFα via serine/threonine phosphatase inhibition with okadaic acid, whereas phosphorylation of Y365/367 is regulated by tyrosine phosphatases (Brandon et al., 2001). Similarly, although phosphorylation of GABA_ARs can also affect channel gating (Brandon et al., 2002) and therefore could account for the decrease in mIPSC amplitude, we suspect this does not play a major role because the majority of the effect requires dynamin GTPase activity.

In culture experiments, we observed a decrease in mIPSC frequency in response to TNFα treatment, yet immunostaining for GAD65 revealed no effect on puncta density or size, suggesting no change in the number of GABA release sites or in GABA synthesis. We conclude that the decrease in event frequency is likely to be regulated by a decrease in release probability and perhaps to some extent by low-amplitude events falling below the detection threshold at higher concentrations of TNFα (>50 ng/ml). Although we did not observe an effect of TNFα treatment on the frequency of mIPSCs recorded from CA1 pyramidal neurons in acute slices, it is possible that the effect is nonetheless present at inhibitory synapses onto other hippocampal neuron subtypes. For example, TNFα controls astrocyte-mediated presynaptic modulation at excitatory synapses onto dentate gyrus neurons (Santello et al., 2011).

TNFα is an important mediator of homeostatic synaptic plasticity and is upregulated during prolonged periods of neuronal inactivity, when it is required for the inverse regulation of AMPAR and GABA_AR trafficking (Stellwagen and Malenka, 2006). Our findings imply that such homeostatic downregulation of synaptic GABA_AR content is mechanistically distinct from the heterosynaptic depression of inhibitory synapses (inhibitory long-term depression; Castillo et al., 2011), which requires either calcineurin (Lu et al., 2000) or PP2A (Lu et al., 2010) rather than PP1 to dephosphorylate the receptor.

TNFα is strongly upregulated in a number of neurological disorders, including epilepsy, autism spectrum disorder (ASD), and neurodegenerative diseases (Vezzani and Granata, 2005; Collins et al., 2012; Montgomery and Bowers, 2012; Onore et al., 2012). Mice lacking TNFR1 display reduced seizure susceptibility in the hippocampus, suggesting that TNFR1-dependent signaling enhances excitability in pathological states (Balosso et al., 2005). Similarly, impairment in GABA_AR function achieved by knocking out the β3 subunit produces electroencephalographic abnormalities and seizures (DeLorey et al., 1998; Liljelund et al., 2005), and epilepsy is associated with a downregulation of gephyrin and GABA_ARs (González et al., 2013). Deficits in GABAergic neurotransmission are also implicated in ASD (Chao et al., 2010; Han et al., 2012), which is often accompanied by immune dysregulation and elevated levels of TNFα (Chez et al., 2007; Malik et al., 2011). It has been proposed that an elevation of the excitatory/inhibitory ratio, which is regulated by TNFα (Stellwagen et al., 2005), is a major cause of ASD (Rubenstein and Merzenich, 2003; Yizhar et al., 2011). In summary, we propose that our findings provide a mechanistic link between the inflammatory component of certain neurological diseases and phenotypes associated with deficits in GABAergic neurotransmission.