Endosomal Phosphatidylinositol 3-Phosphate Promotes Gephyrin Clustering and GABAergic Neurotransmission at Inhibitory Postsynapses*♦

The formation of neuronal synapses and the dynamic regulation of their efficacy depend on the proper assembly of the postsynaptic neurotransmitter receptor apparatus. Receptor recruitment to inhibitory GABAergic postsynapses requires the scaffold protein gephyrin and the guanine nucleotide exchange factor collybistin (Cb). In vitro, the pleckstrin homology domain of Cb binds phosphoinositides, specifically phosphatidylinositol 3-phosphate (PI3P). However, whether PI3P is required for inhibitory postsynapse formation is currently unknown. Here, we investigated the role of PI3P at developing GABAergic postsynapses by using a membrane-permeant PI3P derivative, time-lapse confocal imaging, electrophysiology, as well as knockdown and overexpression of PI3P-metabolizing enzymes. Our results provide the first in cellula evidence that PI3P located at early/sorting endosomes regulates the postsynaptic clustering of gephyrin and GABAA receptors and the strength of inhibitory, but not excitatory, postsynapses in cultured hippocampal neurons. In human embryonic kidney 293 cells, stimulation of gephyrin cluster formation by PI3P depends on Cb. We therefore conclude that the endosomal pool of PI3P, generated by the class III phosphatidylinositol 3-kinase, is important for the Cb-mediated recruitment of gephyrin and GABAA receptors to developing inhibitory postsynapses and thus the formation of postsynaptic membrane specializations.

The formation of neuronal synapses and the dynamic regulation of their efficacy depend on the proper assembly of the postsynaptic neurotransmitter receptor apparatus. Receptor recruitment to inhibitory GABAergic postsynapses requires the scaffold protein gephyrin and the guanine nucleotide exchange factor collybistin (Cb). In vitro, the pleckstrin homology domain of Cb binds phosphoinositides, specifically phosphatidylinositol 3-phosphate (PI3P). However, whether PI3P is required for inhibitory postsynapse formation is currently unknown. Here, we investigated the role of PI3P at developing GABAergic postsynapses by using a membrane-permeant PI3P derivative, timelapse confocal imaging, electrophysiology, as well as knockdown and overexpression of PI3P-metabolizing enzymes. Our results provide the first in cellula evidence that PI3P located at early/ sorting endosomes regulates the postsynaptic clustering of gephyrin and GABA A receptors and the strength of inhibitory, but not excitatory, postsynapses in cultured hippocampal neurons. In human embryonic kidney 293 cells, stimulation of gephyrin cluster formation by PI3P depends on Cb. We therefore conclude that the endosomal pool of PI3P, generated by the class III phosphatidylinositol 3-kinase, is important for the Cbmediated recruitment of gephyrin and GABA A receptors to developing inhibitory postsynapses and thus the formation of postsynaptic membrane specializations.
Phosphoinositides, phosphorylated derivatives of phosphatidylinositol, are critical regulators of intracellular signaling, membrane traffic, and cell compartmentalization (1-3). The functional roles of phosphoinositide metabolism have been studied in great detail at the presynaptic terminal, where phosphoinositide turnover is of critical importance for synaptic vesicle (SV) 3 recycling and synapse function (4). Little, however, is known about specific roles of phosphoinositides at postsynapses.
Core components of most inhibitory GABAergic postsynapses are GABA A receptors (GABA A Rs), the cell adhesion protein neuroligin 2 (NL2), the scaffolding protein gephyrin, and the guanine nucleotide exchange factor collybistin (Cb) (5,6). During the formation of many GABAergic synapses, most notably those at neuronal somata, NL2 is thought to activate Cb, which promotes Cb membrane association and the subsequent recruitment of gephyrin and GABA A Rs (7)(8)(9). In vitro binding studies indicate that the pleckstrin homology (PH) domain of Cb specifically binds phosphatidylinositol 3-phosphate (PI3P) (9 -13), and this interaction is thought to be essential for the anchoring of NL2-gephyrin-Cb complexes at the postsynaptic plasma membrane of synapses that depend on Cb (7,9). The notion that PI3P binding by the PH domain is required for proper Cb function is supported by the observation that deletion of the PH domain (14), or the substitution therein of two arginine residues that are essential for PI3P binding (9,11), causes a marked reduction in the density of postsynaptic gephyrin clusters in hippocampal neurons. Furthermore, an emerging feature of Cb missense mutations in patients with epilepsy and intellectual disability appears to be impaired PI3P binding (13,15).
PI3P is enriched on early/sorting endosomes (16) and has important roles in membrane trafficking (17,18). In addition, some studies implicate class II phosphatidylinositol 3-kinase (PI3K) isoforms (PI3K-C2␣/␤/␥) in the receptor-triggered accumulation of PI3P at the plasma membrane of various cell types (19,20); however, unequivocal evidence for the presence of PI3P at the neuronal plasma membrane is presently lacking. Thus, it remains unclear whether and how PI3P might contribute to the Cb-mediated anchoring of gephyrin scaffolds beneath the postsynaptic membrane and whether the affinity and specificity of the Cb/PI3P interaction as determined in vitro correlate with a defined role of PI3P in the formation of inhibitory postsynapses.
In this study, we investigated the role of PI3P at developing GABAergic postsynapses in cultured hippocampal neurons by using a membrane-permeant PI3P derivative (21), time-lapse confocal imaging, electrophysiology, as well as knockdown and overexpression of PI3P-metabolizing enzymes. Our results indicate that a PI3P pool associated with early/sorting endosomes is important for the formation of Cb-dependent inhibitory postsynapses. We thereby provide the first demonstration that PI3P is a critical regulator of postsynaptic gephyrin and GABA A R clustering and that it is involved in the regulation of inhibitory postsynaptic strength in neurons.

PI3P Promotes Postsynaptic GFP Gephyring Clustering-To
assess the specific role of PI3P in the Cb-dependent clustering of gephyrin at inhibitory postsynapses, we focused on hippocampal neurons in culture, which receive glutamatergic and GABAergic synaptic inputs and show defects in GABAergic postsynaptic composition and function upon Cb deletion (22). We first aimed to experimentally increase the intracellular PI3P concentration in cultured neurons by applying a membranepermeable PI3P-acetoxymethyl (AM) ester derivative, its photoactivatable "caged" coumarin-PI3P-AM variant, and, as a negative control, its regioisomer PI4P-AM, all of which had previously been validated in HeLa cells (21) and were recently used for studying vesicular trafficking in muscle cells of X-linked centronuclear myopathy patients (23). Previous studies indicate a distribution of both coumarin-caged and -uncaged PIP-AM probes into most cellular membranes (21,24). However, despite their wide subcellular distribution, these probes demonstrated high structural specificity, presumably through their specific interaction with endogenous PIP-binding proteins. For example, it has been shown that PI3P-AM induces early endosome fusion in living cells and that the resulting fused endosomes were positive for the endogenous early endosome antigen 1 (EEA1) (21), a known marker of early endosomes that interacts specifically with PI3P via its FYVE finger domain (25). In contrast, compounds of identical chemical composition as PI3P-AM but structurally slightly different, such as PI4P-AM and the enantiomer of PI3P-AM, were unable to induce early endosome fusion (21). In cultured hippocampal neurons, the AM-modified phosphoinositide variants were efficiently accumulated, as demonstrated by adding 50 M couma-rin-PI3P-AM to the culture medium. Neurons incubated with coumarin-PI3P-AM for ϳ10 min displayed diffuse coumarin fluorescence (Fig. 1A), indicating efficient uptake and widespread distribution of the PI3P derivative into intracellular membranes.
For quantitative experiments, we transfected cultured hippocampal neurons at DIV 9 with GFP-gephyrin, incubated the transfected neurons 1 day later for 2 h with 50 M PI3P-AM (or PI4P-AM) or with the vehicle dimethyl sulfoxide (DMSO) only, and analyzed the resulting effects on GFP-gephyrin distribution (Fig. 1, B and C). This protocol allowed us to visualize newly synthesized intracellular gephyrin that is en route to the plasma membrane at early stages of inhibitory postsynapse formation (from DIV 9 to DIV 10). Under these conditions, only few dendritic GFP-gephyrin clusters were formed in the transfected neurons prior to treatment, and most GFP-gephyrin was diffusely distributed in the cells and accumulated in larger somatic structures (Fig. 1B, left). The subsequent addition of PI3P-AM for 2 h (Fig. 1, B, left and C), but not of PI4P-AM or DMSO only (Fig. 1C), led to an increase in the number of dendritic GFP-gephyrin clusters, to a concomitant reduction of the diffuse cytoplasmic GFP-gephyrin fluorescence, and to an apparent redistribution of GFP-gephyrin from the large cytoplasmic structures to numerous smaller clusters (Fig. 1B, left). Three-dimensional reconstruction of confocal image stacks and examination of internal structures in the somata of transfected neurons disclosed an enhanced accumulation of GFPgephyrin fluorescence within both somatic (Fig. 1B, YZ plane views) and perisomatic clusters. Quantification of the total numbers of GFP-gephyrin clusters in identified neurons before and after the 2-h treatment indicated an increase by 76.6% in the neurons treated with PI3P-AM but not in those treated with DMSO or PI4P-AM ( Fig. 1C; DMSO, 111.8 Ϯ 7.2%, n ϭ 10; PI3P-AM, 176.6 Ϯ 15.6%, n ϭ 10; PI4P-AM, 97.9 Ϯ 1.6%, n ϭ 10). Qualitatively similar effects were also observed with the coumarin-PI3P-AM derivative upon UV-light photoactivation and a 1-h incubation (Fig. 1A).
To test for PI3P effects on GFP-gephyrin recruitment to developing dendritic and perisomatic inhibitory postsynapses, we performed confocal live-imaging of GFP-gephyrin-transfected neurons prior to compound treatment (Fig. 2, A-C, left), live-imaging of the same neurons upon a 2-h treatment with DMSO, PI3P-AM, or PI4P-AM, respectively (Fig. 2, A-C, center), and confocal imaging of these neurons after fixation and post hoc staining for the vesicular inhibitory amino acid transporter (VIAAT), a marker of GABAergic presynapses (Fig. 2, A-C, right). This revealed an ϳ1.5-fold increase in the average fluorescence intensity of postsynaptic GFP-gephyrin clusters apposed to presynaptic VIAAT puncta in PI3P-AM-treated neurons but not in cells treated with DMSO only or PI4P-AM ( Fig. 2D; DMSO, 95.9 Ϯ 3.8%, n ϭ 70 postsynaptic clusters; PI3P-AM, 150.4 Ϯ 17.9%, n ϭ 70 postsynaptic clusters; PI4P-AM, 98.9 Ϯ 5.1%, n ϭ 35 postsynaptic clusters). We therefore conclude that membrane-permeant PI3P promotes gephyrin clustering at inhibitory postsynaptic sites. Notably, a considerable fraction of the GFP-gephyrin clusters seen after PI3P-AM treatment were not apposed to VIAAT but appeared intracellularly localized (Fig. 2B, right). Their rounded structures suggest that these clusters represented vesicle-associated GFP-gephyrin assemblies rather than amorphous intracellular aggregates, as seen upon overexpression in heterologous cells (26,27) or under particular in vivo conditions (28).
PI3P Increases the Strength of GABAergic Postsynapses-Previous studies have demonstrated a strong interdependence between gephyrin and major GABA A R subunits in postsynaptic clustering (29,30). To determine the physiological consequences of the enhanced postsynaptic GFP-gephyrin clustering found after PI3P-AM treatment, we recorded GABAergic miniature inhibitory postsynaptic currents (mIPSCs) and evoked inhibitory postsynaptic currents (eIPSCs) in autaptic cultures of DIV 9 -10 GABAergic neurons from striata. In these cultures, in which isolated single neurons form synapses with themselves, pre-and postsynaptic effects of PI3P-AM on GABAergic neurotransmission can be distinguished with relative ease (31). Mean mIPSC amplitudes were significantly (by 83.4%) increased in PI3P-AM-treated neurons (61.6 Ϯ 4.7 pA, n ϭ 30) as compared with cells treated with DMSO only (33.6 Ϯ 2.3 pA, n ϭ 26; Fig. 3, A and B). In contrast, mIPSC frequencies were not significantly different between the two groups (PI3P-AM, 1.3 Ϯ 0.2 Hz, n ϭ 30; DMSO, 1.2 Ϯ 0.4 Hz, n ϭ 26; Fig. 3B). The larger mIPSC amplitudes and the unchanged mIPSC frequencies in PI3P-AM-treated neurons indicate an increased number of functional postsynaptic GABA A Rs, which is in line with the increase in postsynaptic GFP-gephyrin clustering observed upon PI3P-AM treatment (Fig. 2D). Furthermore, exogenous application of GABA, which activates both synaptic and extrasynaptic GABA A Rs, led to a significant increase of the mean current amplitude (by 82.4%) in PI3P-AM-treated neurons (1.82 Ϯ 0.24 nA; n ϭ 33) as compared and normalized to cells treated with DMSO only (1.00 Ϯ 0.1 nA; n ϭ 24; Fig. 3C).
As PI3P is present in endosomes at presynaptic terminals (32), we tested whether evoked neurotransmitter release is affected in PI3P-AM-treated striatal autaptic neurons. Although mean mIPSC amplitudes were strongly increased (  3, E, left and center). Likewise, the numbers of primed synaptic vesicles (PSVs; RRP charge/mIPSC charge) were not significantly different between PI3P-AM-and DMSO-treated neurons (PI3P-AM, 1245 Ϯ 216, n ϭ 21; DMSO, 969 Ϯ 207, n ϭ 18; Fig. 3E, right). Based on these results, the SV release probability (P vr ) in the PI3P-AM-treated neurons, calculated either by dividing the charge transfer during AP-evoked IPSCs by the RRP charge (Fig. 3F, left) or by dividing the number of SVs released per AP by the number of PSVs (Fig. 3F, right), was strongly reduced in the presence of PI3P-AM-treated as compared with DMSO-treated neurons. Together, our electrophysiological analyses of autaptic GABAergic neurons indicate that mIPSCs and eIPSCs are regulated differentially upon an increase of PI3P in endomembranes, likely due to an increased GABA A R density at the postsynapse and an impairment of SV recycling at the presynaptic terminal.

PI3P Increases the Size and Percentage of Endogenous Gephyrin and GABA A Rs at Inhibitory Postsynaptic
Sites-To study the effects of PI3P-AM during the early stages of inhibitory synapse formation, we immunostained the autaptic striatal neurons for VIAAT as a marker of GABAergic presynapses, and for gephyrin and the ␣2 subunit of GABA A Rs, which are co-enriched in GABAergic postsynapses (35). After treating the striatal autaptic cultures at DIV 9 for 2 h with DMSO only (Fig. 4A, left) or 50 M PI3P-AM (Fig. 4A, right), immunocytochemistry was performed as indicated. At this developmental stage, gephyrin and GABA A R-␣2 immunoreactive clusters were distributed at both extrasynaptic and postsynaptic sites, as indicated by their only partial apposition to VIAAT-positive puncta (Fig. 4B). PI3P-AM treatment did not significantly change the total number of dendritic VIAAT, gephyrin, or GABA A R-␣2 clusters, as compared with DMSO-only treated neurons (Fig. Together, these immunocytochemical results are consistent with the observed increases in mIPSC mean amplitudes and GABA-induced responses seen in PI3P-AM-treated striatal autaptic neurons (Fig. 3, B and C) and indicate that PI3P-AM increases the densities and sizes of synaptic gephyrin and GABA A R clusters.

PI3P-AM Treatment Substantially Reduces Evoked Excitatory Transmission but Has No Effect on Glutamatergic
Postsynapses-To further test whether the presynaptic changes produced by PI3P-AM treatment of autaptic GABAergic neurons contribute to the observed increase of their postsynaptic strength, we recorded mEPSCs and eEPSCs from glutamatergic neurons in autaptic DIV 9 -11 hippocampal cultures after a 2-h incubation with either DMSO only or 50 M PI3P-AM. Notably, here the amplitudes and frequencies of mEPSCs were not significantly different between these treatments (  1.11 Hz Ϯ 0.24 Hz, n ϭ 15; PI3P-AM, 1.33 Ϯ 0.18, n ϭ 21), indicating that the density of synaptic excitatory glutamate receptors is unchanged in PI3P-AM-treated neurons, as compared with DMSO-only treated cells. Furthermore, exogenous application of glutamate led to similar responses in both groups ( Fig. 5C; DMSO normalized, 1.00 nA Ϯ 0.11 nA, n ϭ 19; PI3P-AM, 1.02 nA Ϯ 0.10 nA, n ϭ 24); thus the total number of surface glutamate receptors (synaptic ϩ extrasynaptic) was not altered upon PI3P-AM treatment. In contrast, exogenous application of GABA resulted in a significantly larger (by 62%) mean current amplitude in PI3P-AM-treated neurons (1.62 Ϯ 0.18 nA; n ϭ 24) as compared and normalized to cells treated with DMSO only (1.00 Ϯ 0.07 nA; n ϭ 19; Fig. 3C). Thus, PI3P specifically increases the density of surface GABA A Rs, but not glutamate receptors, in autaptic hippocampal neurons.
In contrast to its lack of an effect on mEPSCs, PI3P-AM treatment abolished eEPSCs recorded from autaptic hippocampal neurons almost completely (Fig. 5E). Notably, 65% of the PI3P-AM-treated cells showed no detectable eEPSCs, whereas the remaining cells (9 of 26 tested) displayed markedly reduced mean eEPSC amplitudes, as compared with DMSO controls (Fig. 5E, center; DMSO, 1.95 nA Ϯ 0.38 nA, n ϭ 20; PI3P-AM, 0.28 nA Ϯ 0.04, n ϭ 9/26). Analysis of the postsynaptic responses triggered by hypertonic sucrose solution indicated that the RRPs of SVs were similar between both groups ( Fig. 5F; DMSO, 0.16 Ϯ 0.03 nC, n ϭ 17; PI3P-AM, 0.18 Ϯ 0.02 nC, n ϭ 25). Accordingly, P vr in the PI3P-AM-treated neurons, calculated by dividing the charge transfer during AP evoked EPSCs by the RRP charge, was strongly reduced as compared with DMSO-treated neurons ( Fig. 5G; DMSO, 6.14 Ϯ 0.81%, n ϭ 17; Procedures"). Cells were fixed after treatment and stained with gephyrin-, VIAAT-, and GABA A R-␣2-specific antibodies, as indicated. Scale bars, 10 m. B, overlays and single channels of the corresponding boxed areas in A, at higher magnifications, as indicated. Note increases in the size of synaptically localized gephyrin (geph) and, to a lesser extent, ␣2 subunit-containing GABA A Rs in neurons treated with PI3P-AM, as compared with control cells. C, quantifications of the densities of VIAAT, gephyrin, and GABA A R ␣2 immunoreactive puncta per 40 m dendritic length (left), of the percentages (center) ,and the mean sizes (right) of synaptically localized dendritic gephyrin and GABA A R ␣2 clusters in PI3P-AM-treated (green) and control (gray) autaptic neurons. Bars correspond to counts on randomly selected dendrites of 20 individual neurons from three independent treatments per condition. PI3P-AM, 0.54 Ϯ 0.13%, n ϭ 16/25). Together, these data indicate that the impairment of SV exocytosis seen after PI3P-AM treatment is not caused by or associated with changes in the density of glutamate receptors in the postsynaptic neuron. Together, these electrophysiological analyses of both GABAergic and glutamatergic autaptic neurons demonstrate a specific increase of surface and synaptic GABA A Rs, but not glutamate receptors, upon PI3P-AM treatment.

Phosphatidylinositol 3-Kinases and Phosphatases Involved in the Regulation of PI3P Levels during GABAergic Synapse Formation-Earlier reports have shown that individual PI3K
inhibitors differentially affect the PI3P pools in endosomal and plasma membrane compartments, which indicate that multiple different enzymes are involved in the generation of PI3P (36,37). Current consensus is that PI3K-C3 is responsible for the production of the constitutive endosomal pool of PI3P at early/ sorting endosomes (17), whereas class II PI3K isoforms (PI3K-C2␣/␤/␥) are implicated in both the synthesis of PI(3,4)P 2 and the agonist-mediated accumulation of PI3P at the plasma membrane (18 -20). Specifically, the ␣-isoform has been demonstrated to preferentially synthesize PI(3,4)P 2 both in vitro and at plasma membrane endocytotic pits in vivo, a site where PI(3,4)P 2 is required for membrane constriction prior to endocytic vesicle fission (18,38). Of the three known class II PI3K isoforms, PI3K-C2␣ and PI3K-C2␤ are expressed in the mammalian brain, whereas PI3K-C2␥ is not (39,40).
To determine which PI3K enzymes might be relevant for the generation of the PI3P pool involved in postsynaptic gephyrin clustering, we conducted miRNA-based knockdown experiments targeting PI3K-C3, PI3K-C2␣, and PI3K-C2␤ in dissociated rat hippocampal neurons. Knockdown efficiencies of the different miRNAs were determined by Western blotting of lysates of Rat2 fibroblasts (ATCC CRL-1764; a fibroblast cell line of rat origin) stably expressing the corresponding miRNAs or control miRNAs. This led to the identification of miRNAs, which reduced the expression levels of PI3K-C3, PI3K-C2␣, and PI3K-C2␤ by about 79, 83, and 83%, respectively (Fig. 6, A  and B). To focus on postsynaptic defects resulting from downregulating the PI3P-specific kinases and to minimize confounding effects of parallel presynaptic changes, we performed Ca 2 PO 4 transfections with vectors expressing GFP and a given miRNA at low transfection efficiencies (1-2%) and concentrated our immunocytochemical analysis on the sparse morphologically identifiable pyramidal neurons coexpressing GFP and the respective miRNAs. Cultured rat hippocampal neurons were transfected at DIV 4, and the effects of down-regulating the corresponding kinases on gephyrin clustering were analyzed at DIV 14. We detected a significant reduction by ϳ33% in the density of dendritic gephyrin immunoreactive puncta in neurons expressing the miRNA directed against PI3K-C3 but not in neurons expressing the miRNAs directed against PI3K-C2␣ or PI3K-C2␤, as compared with neurons transfected with the control miRNA (Fig. 7, A and B;  . Notably, expression of the PI3K-C3 miRNA had no effect on the density of dendritic VIAAT-immunoreactive puncta ( Fig. 7C; control miRNA, 17.9 Ϯ 0.7 puncta/40-m dendritic length, n ϭ 13; PI3K-C3 miRNA, 18.0 Ϯ 1.3 puncta/ 40-m dendritic length, n ϭ 12), indicating that the number of contacting GABAergic presynapses had not changed. EEA1 is a protein containing a FYVE finger domain that binds to PI3P and targets EEA1 to early/sorting endosomes (25). In agreement with the involvement of PI3K-C3 in generating the PI3P pool localized on early/sorting endosomes, down-regulation of PI3K-C3 by miRNA reduced the density of EEA1 immunoreactive puncta in the dendrites (Fig. 7F; control miRNA, 22.6 Ϯ 1.8 puncta/40-m dendritic length, n ϭ 10; PI3K-C3 miRNA, 11.8 Ϯ 1.7 puncta/40-m dendritic length, n ϭ 10) but not the somata ( Fig. 7E; control miRNA, 18.6 Ϯ 1.2 puncta/100 m 2 , n ϭ 10; PI3K-C3 miRNA, 16.3 Ϯ 2.7 puncta/100 m 2 , n ϭ 10) of transfected neurons (Fig. 7, D-F).
To evaluate whether the effects of PI3K-C3 miRNA downregulation can be rescued by exogenously added PI3P-AM, we transfected hippocampal cultures at DIV 4, e.g. prior to the onset of inhibitory synapse formation, with the PI3K-C3 miRNA or the control miRNA, and subsequently treated them at DIV 8 with 50 M PI3P-AM or DMSO for 2 h prior to fixation and immunocytochemical analysis. Again, the PI3K-C3 miRNA-expressing neurons showed a significant reduction by ϳ54% in the density of dendritic gephyrin cluster neurons, as compared with cells expressing the control miRNA, when treated with DMSO only (Fig. 7, G and H, top/left; control miRNA, 6.06 Ϯ 0.71 puncta/40-m dendritic length, n ϭ 16; PI3K-C3 miRNA, 2.76 Ϯ 0.32 puncta/40-m dendritic length, n ϭ 17). Furthermore, we detected an ϳ37% reduction in the size of dendritic gephyrin clusters in the PI3K-C3 miRNA-expressing neurons, as compared with control miRNA expressing cells (Fig. 7, G and H, bottom/left; control miRNA, 0.217 Ϯ 0.014 m 2 , n ϭ 16; PI3K-C3 miRNA, 0.137 Ϯ 0.011 m 2 , n ϭ 17). In neurons treated for 2 h with PI3P-AM, the difference in endogenous gephyrin cluster densities between PI3K-C3 and control miRNA-expressing cells was similar to that observed with DMSO only-treated cells (Fig. 7, G, and H,   differ significantly between the PI3K-C3 miRNA and control miRNA-expressing neurons but in both cases were larger than upon DMSO treatment (Fig. 7H, bottom/right; control miRNA, 0.286 Ϯ 0.022 m 2 , n ϭ 17; PI3K-C3 miRNA, 0.231 Ϯ 0.019 m 2 , n ϭ 17). Thus, consistent with the results obtained from striatal autaptic neurons (Fig. 4), hippocampal neurons also show a significant change in the mean size, but not density, of endogenous gephyrin clusters after a 2-h treatment with PI3P-AM. Furthermore, these findings demonstrate that the comparatively short incubation with PI3P-AM rescues the reduction in mean gephyrin cluster size, but not density, observed upon prolonged PI3K-C3 miRNA expression.
We next wondered whether raising intracellular PI3P levels by overexpression of recombinant PI3K-C3 similarly would result in enhanced gephyrin clustering, as observed upon PI3P-AM treatment. Hence, we expressed the GFP-tagged human orthologue of PI3K-C3, GFP-hVps34, or as a control GFP from DIV 7 to 14 in cultured hippocampal neurons and then analyzed the resulting effects on the clustering of endogenous gephyrin by immunostaining. In contrast to the results obtained with a 2-h PI3P-AM treatment, prolonged overexpression of GFP-hVps34 led to significant increases in the density (GFP, 10.9 Ϯ 1.1 puncta/100 m 2 , n ϭ 15; GFP-hVps34, 17.1 Ϯ 1.5 puncta/100 m 2 , n ϭ 15) and size (GFP, 0.30 Ϯ 0.03 m 2 , n ϭ 562 clusters; GFP-hVps34, 0.54 Ϯ 0.08 m 2 , n ϭ 851 clusters) of perisomatic, but not dendritic, gephyrin clusters (Fig. 8, A and B). The reasons for the opposing effects on gephyrin perisomatic and dendritic clusters observed upon prolonged hVps34 expression are currently unknown but might reflect different functions of PI3K-C3 within the cell or differences in the compartmental contents of PI3P-related kinases and phosphatases (see under "Discussion"). In either case, these results further confirm that increases in intracellular PI3P levels lead to enhanced gephyrin clustering.
The ability of PI3P to serve dynamic functions in membrane trafficking and protein sorting relies on the tight spatial and temporal control of the kinases and phosphatases involved in its generation and degradation. Members of the myotubularin phosphatase family, MTMs, are the principal PI3P and PI(3,5)P 2 3-phosphatases. Notably, a pool of membrane-associated PI3K-C3 is bound to MTM1, which competes with GTPases for PI3K-C3 binding (2). This indicates a complex interplay of small GTPases, MTMs, and PI3K-C3 in the tuning of endosomal PI3P levels. Furthermore, the activity of certain inositol polyphosphate 5-phosphatases can lead to an enrichment of PI3P in endosomes or at the plasma membrane. For example, ectopic expression of the 72-kDa 5-phosphatase (72-5ptase; also called pharbin) leads to the depletion of PI(3,5)P 2 and the accumulation of PI3P at the plasma membrane, and thereby it promotes the plasma membrane translocation of the glucose transporter 4 in adipocytes (41). This 72-5ptase contains a C-terminal CAAX motif for membrane targeting.
Characteristics of Synaptic PI3P Pools-Our finding that the down-regulation of PI3K-C3, but not of PI3K-C2␣ or PI3K-C2␤, causes a significant reduction in gephyrin cluster density on the dendrites of rat hippocampal neurons indicates that the PI3P pool required for gephyrin clustering is located on early/ sorting endosomes, because PI3K-C3 is the kinase responsible for the generation of the constitutive PI3P pool on these endosomal compartments (17). To investigate whether gephyrin associates with PI3P-containing membranes during the initial stages of cluster formation, we transfected rat hippocampal neurons for 1 day (DIV 9 -10) with vectors expressing monomeric red fluorescent protein (mRFP)-gephyrin and GFP-2ϫFYVE, a probe consisting of two tandem PI3P-binding FYVE finger domains that binds PI3P specifically and with high affinity in vitro and in vivo (16). Newly clustered mRFP-gephyrin partially colocalized with GFP-2ϫFYVE at both extrasynaptic and inhibitory postsynaptic sites (Fig. 10A), as indicated by a 25% coapposition of the two proteins to presynaptic VIAAT puncta (Fig. 10, B and C).
To exclude the possibility that the observed colocalization of mRFP-gephyrin with GFP-2ϫFYVE results from a random signal-match due to overexpression of the two proteins in dendritic shafts, we performed time-lapse confocal microscopy to monitor the movement of mRFP-and GFP-positive particles over a period of 30 min. This revealed that mRFP-gephyrin clusters remained stable over the observation period, whereas the GFP-2ϫFYVE fluorescence associated with early/sorting endosomes showed extensive movement, particularly in the dendrites of transfected neurons (Fig. 7D and supplemental Movie 1). However, the GFP-2ϫFYVE puncta that colocalized with mRFP-gephyrin and were apposed to VIAAT-positive presynapses, as indicated by post hoc staining for VIAAT upon time-lapse imaging and fluorescence intensity plot analysis (Fig. 10, D, bottom, and E), showed less movement over time as compared with those that were mRFP-gephyrin negative and not apposed to VIAAT (Fig. 10D and supplemental Movie 1). Notably, prolonged expression of GFP-2ϫFYVE in cultured neurons (DIV 4 -11) led to a significant reduction of endogenous gephyrin clusters in the dendrites, as compared with neurons expressing GFP alone (Fig. 11, A and B; untransfected, 11.4 Ϯ 0.5 puncta/40-m dendritic length, n ϭ 26; GFP, 10.3 Ϯ 0.5 puncta/40-m dendritic length, n ϭ 26; GFP-2ϫFYVE, 7.8 Ϯ 0.6 puncta/40-m dendritic length, n ϭ 26). We attribute this to a competitive masking of membrane-associated PI3P by high levels of GFP-2ϫFYVE (16,42). Together, the results presented above support the view that PI3P is present within endosomal membrane domains that are closely associated with inhibitory postsynapses and hence are restricted to lateral diffusion.
PI3P-AM-induced Redistribution of Gephyrin Depends on Cb-Different lines of evidence suggest that the stimulation of neuronal gephyrin clustering by PI3P-AM involves Cb. First, Cb is the only component of inhibitory postsynapses known to specifically bind this phosphoinositide (10 -12, 15). Second, Cb deletions and mutations impairing PI3P binding have been demonstrated to impair Cb-dependent gephyrin clustering in cultured neurons (9 -11). Third, several collybistin mutations in humans causing X-linked intellectual disability are associated with a loss of, or reduced, PI3P binding capacity (10,13,15). We therefore examined whether the observed effect of PI3P-AM on gephyrin clustering depends on Cb by using HEK 293 cells expressing both GFP-gephyrin and Cb. This heterologous coexpression system was chosen because of the follow- ing: (i) it has proven highly reliable for the identification of proteins that are implicated in gephyrin clustering, including Cb (43), NL2 (7), and its homolog NL4 (44) as well as the Cbinteracting GTPase TC10 (8); and (ii), most importantly, it displays a stringent dependence on Cb of gephyrin cluster formation (14,43).

Discussion
This study provides the first direct demonstration of a key role of PI3P in the regulation of postsynaptic gephyrin clustering and inhibitory synaptic strength in neurons. Specifically, by using cultured hippocampal neurons transiently expressing GFP-gephyrin, we show that increasing intracellular PI3P concentrations with the membrane-permeant derivative PI3P-AM suffices to promote the formation of new gephyrin clusters and to increase the gephyrin content of inhibitory postsynapses by enhancing the membrane recruitment of gephyrin from the cytosol and intracellular vesicle-like structures. In autaptic GABAergic neurons derived from the striatum, the increase in postsynaptic gephyrin induced by PI3P-AM correlated with an increase in postsynaptic GABA A Rs, as illustrated by larger mIPSC amplitudes but unaltered mIPSC frequencies and by significant increases in the number and mean size of synaptic gephyrin and GABA A R clusters. Furthermore, current responses to externally applied GABA were significantly larger, showing that the number of GABA A Rs at the cell surface had increased. A very similar increase in GABA-induced responses was found in autaptic hippocampal glutamatergic neurons upon PI3P-AM treatment. However, here PI3P-AM had no effect on the densities of surface and postsynaptic glutamate receptors, as indicated by the unaltered sizes of glutamate-induced currents and mEPSC amplitudes. Thus, PI3P-AM does not simply enhance plasma membrane incorporation of receptor-containing endosomes at postsynaptic sites but specifically up-regulates the formation of inhibitory, but not excitatory, postsynapses.
In both GABAergic and glutamatergic autaptic neurons, PI3P-AM treatment resulted in marked decreases in eIPSC and eEPSC amplitudes without altering RRP sizes; this is consistent with a markedly reduced probability of presynaptic neurotransmitter release. Based on these findings, we conclude that PI3P negatively affects the fusion or exocytosis of SVs at both inhibitory and excitatory presynapses. Currently, the reasons for the presynaptic effects of PI3P-AM are not clear, but it is conceivable that higher local PI3P concentrations might compete with and reduce the phosphatidylinositol 4,5-bisphosphate-regulated binding of Ca 2ϩ to synaptotagmins and thereby inhibit SNARE-dependent synaptic vesicle fusion, as reported previously (45). Further work will be required to elucidate the precise role of PI3P in presynaptic terminals.
Our knockdown experiments targeting different PI3K isoforms in hippocampal neurons indicate an involvement of PI3K-C3, the enzyme that generates the constitutive pool of PI3P on early/sorting endosomes, in postsynaptic gephyrin clustering. In contrast, PI3K-C2␣ and PI3K-C2␤, which are involved in both the synthesis of PI(3,4)P 2 (38) and the de novo biosynthesis of PI3P at the plasma membrane (19,20,37,46), appear not to be involved in postsynaptic gephyrin accumulation. These findings lead to the unexpected conclusion that the pool of PI3P on endomembranes, and not PI3P generated at the plasma membrane, is of crucial relevance for postsynaptic Cb-dependent gephyrin clustering. In addition, we consistently found the following: (i) miRNA-mediated down-regulation of PI3K-C3 in cultured hippocampal neurons caused an ϳ35% reduction in the density of dendritic gephyrin clusters as compared with control cells; (ii) a 2-h treatment with PI3P-AM rescued the reduction in mean size, but not density, of dendritic gephyrin clusters observed in neurons overexpressing the PI3K-C3 miRNA; and (iii) overexpression of hVps34, the human homologue of PI3K-C3, had opposing effects on gephyrin perisomatic and dendritic clusters, compared with a 2-h treatment of the cells with PI3P-AM. Based on these data, one might speculate that PI3P may not be the only signaling lipid involved in the Cb-dependent accumulation of gephyrin at postsynapses. In agreement with this idea, a recent study indicates broader PIP specificities for the short Cb splice variant, which lacks the N-terminal SH3 domain, than those determined for an open-conformation mutant (W24A/E262A) of the SH3 domain-containing isoform (12). Furthermore, Vps34 has been shown to exist in two complexes in both yeast and mammalian cells, which not only have opposing effects but can potentially negatively regulate one another (18). Finally, phosphoinositide interconversion by additional kinases or phosphatases present at different subcellular compartments might partially restore the endosomal PI3P pool in the absence of PI3K-C3. In line with the latter interpretation, overexpression of the PI(3,5)P 2 -specific phosphatase 72-5ptase/pharbin, which is predicted to increase PI3P levels in the transfected neurons, caused a significant increase in the size of perisomatic gephyrin clusters. This result supports the notion of a complex interplay between additional kinases, phosphatases, and PI3K-C3 in the tuning of endosomal PI3P levels and gephyrin clustering.
Of note in this context, a comprehensive siRNA screen employing a functional cell-based assay for gephyrin clustering in HeLa cells as a readout identified PI3K-C3 and PI3K-C2␥ as candidate kinases involved in the stabilization of GFP-gephyrin clusters (47). To examine the involvement of Cb in the enhancement of gephyrin clustering by PI3P-AM, we employed a similar cell-based assay that relied on a HEK 293 cell line, which inducibly expresses GFP-gephyrin upon TET application and thereby allows us to investigate the clustering of newly synthesized gephyrin independently of other neuronal proteins. In agreement with previous reports (9 -11, 13, 15), our results obtained with this heterologous expression system clearly show that PI3P-AM increases gephyrin cluster formation only upon coexpression of Cb.
The importance of early endosomal PI3P pools for gephyrin clustering, which we demonstrate here, extends our understanding of the mechanisms underlying Cb activation and the subsequent nucleation of gephyrin scaffolds (Fig. 13). Our find-ings support the view that, at least at early stages of synaptogenesis, PI3P-enriched early/sorting endosomes contribute to the membrane recruitment of Cb-gephyrin complexes. Indeed, a significant fraction of 2ϫFYVE-labeled early/sorting endosomes was closely associated with inhibitory postsynapses and apposed to both VIAAT and gephyrin in DIV 10 cultured hippocampal neurons. Furthermore, prolonged overexpression of the 2ϫFYVE probe significantly reduced the density of gephyrin clusters along dendrites of cultured hippocampal neurons, presumably by masking PI3P on endomembanes. Based on these results, we propose that the interplay between gephyrinand Cb-interacting proteins and phosphatases is crucial for interconversion of PI3P and thereby the trafficking of Cbgephyrin complexes toward the plasma membrane, where the latter might be stabilized through interactions with additional proteins, such as NL2 and NL4 (7,9). In line with this idea, a recent report identified a mechanism for phosphoinositide interconversion from PI3P to PI4P at endosomes en route to the plasma membrane and showed that this interconversion enables the subsequent recruitment of the exocyst tethering complex (23). Interestingly, the small GTPase TC10, which interacts in its active GTP-bound state with Cb (8), also binds to one of the components of the exocyst complex, Exo70, and promotes translocation of Exo70 to the plasma membrane (48).
In line with our finding that PI3K-C3 is important for gephyrin clustering at inhibitory postsynapses, overexpression of hVps34, the human homologue of PI3K-C3, in hippocampal neurons led to significant increases in the density and size of perisomatic gephyrin clusters. Similarly, overexpression of the SH3 domain-deficient Cb isoform, which strongly binds PI3P (9 -11), causes increases in the density and size of perisomatic gephyrin clusters (49,50). This indicates that the interaction of Cb and gephyrin with PI3P-containing endosomes is of particular importance for the regulation of the size of perisomatic synapses, which typically produce a stronger inhibitory drive than the smaller synapses localized on dendrites (51). Furthermore, mutations in both PI3K-C3 and gephyrin have been implicated in bipolar disorder and schizophrenia (52,53), and both phosphoinositides and gephyrin are potential targets for the therapeutic effects of lithium in bipolar disorder (52,54). Thus, our findings also provide a new view on signaling pathways related to neuropsychiatric disorders, linking them to PI3K-C3 function, PI3P generation, and postsynaptic gephyrin clustering.
In summary, this study identifies PI3P as an efficient regulator of Cb-dependent gephyrin and GABA A R clustering and indicates the involvement of early/sorting endosomes during the membrane activation process (55) that is required for gephyrin nucleation and scaffold formation at the postsynaptic plasma membrane of developing inhibitory synapses.

Experimental Procedures
Plasmids and Oligonucleotide Sequences-The GFP-gephyrin and mRFP-gephyrin constructs used have been described previously (56,57). The pEGFP-N1 vector was purchased from Clontech. The GFP-2ϫFYVE construct was provided by Dr. Marino Zerial (Dresden, Germany), and vectors encoding mCherry-MTM1-CAAX and mCherry alone (38) were provided by Dr. Volker Haucke (Berlin, Germany). A plasmid encoding the Myc-tagged 72-5ptase (41) was provided by Dr. Christina A. Mitchell (Victoria, Australia), and the plasmid encoding GFP-hVps34 was provided by Dr. Matthias Wymann (Basel, Switzerland). The pcDNA 6.2-GW/EmGFP and pcDNA 6.2-GW/EmGFP-miR-neg.Cntrl plasmid were included in the BLOCK-iT polymerase II miR RNAi expression vector kit (Invitrogen). The negative control sequence without 5Ј-overhangs was 5Ј-GAAATGTACTGCGCGTGGAGACGTTTTG-GCCACTGACTGACGTCTCCACGCAGTACATTT-3Ј. The following single-stranded DNA oligonucleotides encoding the pre-miRNAs of PI3K-C3, PI3K-C2␣, and PI3K-C2␤ used in this study were purchased from Invitrogen, Germany: PI3K-C3: forward, 5Ј-TGCTGTCAAGAAGGTACAAAGATCCTGTT-TTGGCCACTGACTGACAGGATCTTTACCTTCTTGA-3Ј, and reverse, 5-CCTGTCAAGAAGGTAAAGATCCTGT-CAGTCAGTGGCCAAAACAGGATCTTTGTACCTTCTT-GAC-3Ј; PI3K-C2␣, forward, 5Ј-TGCTGTACAAATGATGG-TTCAAGGTGGTTTTGGCCACTGACTGACCACCTTGA-CATCATTTGTA-3Ј, and reverse, 5Ј-CCTGTACAAATGAT-GTCAAGGTGGTCAGTCAGTGGCCAAAACCACCTTGA- FIGURE 13. Tentative model for the role of PI3P-containing early/sorting endosomes in the assembly of the gephyrin scaffold at postsynaptic membranes. A, intracellularly, gephyrin trimers bind to Cb in its closed conformation. At postsynaptic sites, the open conformation is achieved and maintained by the interaction of NL2, NL4, or the ␣2 subunit of GABA A Rs with the SH3 domain of Cb (7,44,59). In addition, Cb binds in its open conformation, via its PH domain, to PI3P and further PIPs located at the plasma membrane (9,12,15). B, particularly at early developmental stages and at extrasynaptic sites, GABA A Rs undergo constitutive endocytosis and are either rapidly recycled back to the cell surface or targeted to lysosomal degradation (5,60). C, Cb (via its SH3 domain (59)) and gephyrin (5) interact with GABA A Rs also at extrasynaptic sites, thereby limiting their diffusion and facilitating their "trapping" to postsynaptic sites. D, at GABA A R-containing early/sorting endosomes, both interactions of gephyrin and Cb with GABA A Rs, as well as binding of Cb's PH domain to PI3P, induce the open conformational state of Cb, thereby increasing the rates of incorporation of receptor-associated scaffold to extrasynaptic or postsynaptic sites. At this stage, levels of gephyrin-and Cb-interacting proteins as well as of PI3K-C3 and additional kinases/phosphatases involved in the generation/degradation of PI3P might be crucial for determining PI3P levels and its interconversion rates and thereby the trafficking of Cb-gephyrin complexes toward the plasma membrane, where the latter might be stabilized through interactions with NL2 and NL4 (see text).
of 120,000 cells/ml in Neurobasal medium supplemented with B27 (Gibco), GlutaMax (Gibco), and penicillin/streptomycin (Roche Applied Science). Neurons were transfected at different DIV, as indicated under "Results," using the CalPhos TM mammalian transfection kit (Clontech). Immunocytochemistry with the antibodies listed above was performed as described previously (11). Images were collected with an inverse Leica DMIRE2 microscope equipped with a ϫ63 oil-immersion objective and connected to a Leica TCS SP2 AOBS confocal laser scanning setup (Leica Microsystems) or with an Axio-Imager Z1 equipped with a Zeiss apochromat ϫ63 objective and an Apotome module (Zeiss). Acquired images were processed identically using the ImageJ software package (rsb.info. nih.gov). Single channels were recorded using the same standardized threshold levels. Subsequently, a binary image was generated, and immunoreactive puncta were counted automatically by the ImageJ software, as described previously (8,9).
PI3P-AM/PI4P-AM Treatment and Time-lapse Imaging of Transfected Hippocampal Neurons-Neurons were cultured on poly-L-lysine-coated 35-mm glass bottom ibidi plates (ibidi GmbH) and transfected as described above. The synthesis of caged PI3P-AM, PI3P-AM, and PI4P-AM has been described previously (21). PIP-AM derivatives were dissolved in DMSO to make a stock solution of 50 mM. To achieve optimal dispersion of the AM compounds, appropriate amounts of the 50 mM stock solutions or DMSO only were mixed with Pluronic F-127 (Invitrogen) at a 1:0.5 (v/v) ratio and added to the imaging medium (Neurobasal-A, phenol red-free, Invitrogen, Karlsruhe, Germany) to a final concentration of 50 M of the AM compounds. The final concentration of DMSO was 0.1%. The neuronal cultures were washed once and kept in imaging medium during the subsequent procedures. Image stacks of transfected hippocampal neurons were taken with an inverse Leica DMIRE2 microscope equipped with a ϫ63 oil-immersion objective and connected to a Leica TCS SP2 AOBS (acoustooptical beam splitter) confocal laser scanning setup (Leica Microsystems) 10 min prior to treatment. Media were replaced by imaging media containing DMSO/pluronic, 50 M PI3P-AM/pluronic, or 50 M PI4P-AM/pluronic, and the cultures were incubated at 37°C in 5% CO 2 for another 2 h. Subsequently, new image stacks of the same identified neurons were collected, or the cultures were processed directly for immunocytochemistry, as described previously (8).
Cultures of Hippocampal and Striatal Autaptic Neurons-Micro-island cultures of hippocampal and striatal neurons were prepared and cultured as reported previously (34,58). Briefly, astrocytes for autaptic cultures were obtained from mouse cortices dissected from P0 wild type animals and enzymatically digested for 15 min at 37°C with 0.25% (w/v) trypsin/ EDTA (Gibco, Life Technologies, Inc.). Astrocytes were plated in T75 culture flasks in DMEM (Gibco, Life Technologies, Inc.) containing 10% fetal bovine serum (Gibco), penicillin (100 units/ml)/streptomycin (100 g/ml; Roche Applied Science), MITO (BD Biosciences), and grown for 7-10 DIV. Subsequently, astrocytes were trypsinized and plated at a density of ϳ30,000 cells/well onto 32-mm diameter glass coverslips, which had been coated with agarose (Sigma) and stamped using a custom-made stamp to generate 200 ϫ 200 m substrate