Regulation of Gephyrin Assembly and Glycine Receptor Synaptic Stability*

Gephyrin is required for the formation of clusters of the glycine receptor (GlyR) in the neuronal postsynaptic membrane. It can make trimers and dimers through its N- and C-terminal G and E domains, respectively. Gephyrin oligomerization could thus create a submembrane lattice providing GlyR-binding sites. We investigated the relationships between the stability of cell surface GlyR and the ability of gephyrin splice variants to form oligomers. Using truncated and full-length gephyrins we found that the 13-amino acid sequence (cassette 5) prevents G domain trimerization. Moreover, E domain dimerization is inhibited by the gephyrin central L domain. All of the gephyrin variants bind GlyR β subunit cytoplasmic loop with high affinity regardless of their cassette composition. Coexpression experiments in COS-7 cells demonstrated that GlyR bound to gephyrin harboring cassette 5 cannot be stabilized at the cell surface. This gephyrin variant was found to deplete synapses from both GlyR and gephyrin in transfected neurons. These data suggest that the relative expression level of cellular variants influence the overall oligomerization pattern of gephyrin and thus the turnover of synaptic GlyR.

Fast neurotransmission at synapses depends on the enrichment of ionotropic receptors in the postsynaptic membrane. Scaffolding proteins present in the postsynaptic densities participate in the local and selective accumulation of most excitatory and inhibitory receptors in front of the corresponding transmitter release sites (1,2). Synaptic localization of clusters of inhibitory ␥-aminobutyric acid, type A and glycine receptors relies on gephyrin, initially discovered as a GlyR-associated extrinsic membrane protein (3,4). The pivotal role of gephyrin has been largely demonstrated by antisense experiments (5,6) and the use of knock-out mice (7,8). Gephyrin binds both GlyR via an 18-amino acid amphipathic helix within the M3-M4 cytoplasmic loop of the ␤ subunit (9, 10) and tubulin via a motif similar to Tau/MAP2 tubulin-binding domain (11,12). Therefore, gephyrin is functionally adapted to anchor GlyR at synapses via the cytoskeleton (8), but its physical link with ␥-aminobutyric acid, type A receptors is not yet understood.
Gephyrin has a modular structure resulting from the fusion of two genes of bacterial origin, encoding the enzymes MogA and MoeA that catalyze the biosynthesis of the molybdenum cofactor in Escherichia coli (13,14). These enzymes are homologous to the gephyrin G and E domains (N-and C-terminal domains), respectively, which flank a 170-residue highly variable region (linker domain or L domain). The recent determination of the tertiary and quaternary structures of the MogA, MoeA, and gephyrin G domains (15)(16)(17)(18) provides clues for delineating structure-function relationships for gephyrin. The binding site of the GlyR ␤ subunit has been mapped on a gephyrin-specific structure in the E domain crystal (19 -21). Isolated G and E domains form stable trimers and dimers, respectively, implying that gephyrin has two distinct oligomerization potentials. If these properties were both conserved in the gephyrin molecule, then oligomerization would build up a two-dimensional hexagonal gephyrin lattice (18,23). However, this model has to be reconciled with the observation that purified gephyrin is a trimer harboring a monomeric E domain (19,20).
Several gephyrin splice variants differing by the presence of distinct sequences (cassettes) have been characterized for rodent (12,23,24) and human (25) proteins. The variation in gephyrin L domain primary structure is speculatively correlated with the numerous gephyrin-interacting proteins (22,25) and raises the questions of the function of the cassettes (22,25,26). Until now, studies aimed at characterizing interactions have mainly focused on the more abundant gephyrin variant (clone P1, 23), which harbors two cassettes (C2 and C6) of nine known to be alternatively spliced in rat. In a previous study (24), we used in vitro assays of binding of gephyrin variants to the GlyR ␤ subunit M3-M4 cytoplasmic loop to investigate the effect of gephyrin variability on receptor recognition. This had led us to suggest that the presence of a cassette in the gephyrin G domain alters conformation in the N-terminal domain of gephyrin and results in reduced ␤ subunit binding. Such an effect remained to be explained, because of the exclusive binding ability of the E domain.
Gephyrin homophilic interactions, which rely on the ability of the molecule to form trimers and dimers via distinct domains, are thought to account for the clustering of gephyrin-interacting GlyR in the postsynaptic membrane. Because of the variable primary structure of gephyrin, we investigated here the effect of the cassettes within splice variants on * This work was supported by grants from the Association Française contre les Myopathies and from the Institut de Recherche sur la Moelle Epiniè re. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. □ S The on-line version of this article (available at http://www.jbc.org) contains supplemental text and figures. 1 Supported by the Institut National de la Santé et de la Recherche Mé dicale.
its dual function as scaffold protein for GlyR and in polymerization. We have designed novel molecular tools to analyze in vitro the GlyR binding properties of gephyrin and its oligomerization capability. We show that the cassette composition of gephyrin has no effect on a high affinity interaction with the ␤ subunit, provided that its gephyrin-binding site is presented within a closed M3-M4 cytoplasmic loop. In contrast, we found that 1) trimerization of gephyrin can be impaired by the insertion of a defined cassette (cassette 5) in the G domain and 2) gephyrin trimerization correlates with the ability to stabilize GlyR at the cell surface. In cultured neurons, the incorporation of a trimerization-defective gephyrin molecule into the postsynaptic gephyrin polymer interfered with the presence of GlyR at synapses. Altogether, these results indicate that the relative cellular expression levels of gephyrin variants influence the overall oligomerization pattern of gephyrin, thereby controlling GlyR number at synapses.

EXPERIMENTAL PROCEDURES
Constructs and expression plasmids, protein expression, and purification are described in the supplemental materials (27,29,30).
In Vitro Binding Assays-Tagged gephyrin fragments and PH 3 chimeras in PBSM (5 mM ␤-mercaptoethanol in phosphate-buffered saline, pH 7.4) buffer were used immediately after elution from Ni-NTA-agarose. Two binding protocols were used. Solid phase binding assays (first protocol) were performed in ELISA plates. 96-Well microtiter plates (PVC plates; Costar) were coated with 50 l/well of 60 nM PH or GST chimera solution in 0.1 M Na 2 CO 3 /NaHCO 3 , pH 9.6. Nonspecific binding sites were blocked with BPBSM buffer (4 mg/ml bovine serum albumin, 10% (w/v) sucrose, 0.25% (w/v) gelatin, 0.25% (v/v) Tween 20, 5% (w/v) milk in PBSM). Then, unless otherwise stated, gephyrin fragments were incubated for 16 h at 4°C using 50 l/well of solutions at concentrations specified in the figure legends. After washing, bound gephyrin was detected with either anti-Myc or anti-T7 (Novagen) tag-specific antibodies and horseradish peroxidase-conjugated rabbit antimouse antibody. For quantitation, ABTS (Sigma-Aldrich) was used as substrate, and development was monitored in an ELISA reader (Bio-Tek). For solution assays (second protocol) of gephyrin protomer binding capacity, the tested fragments at the final concentration indicated in the figure legend were incubated in a final volume of 15 l for 16 h at 4°C in PBSMG buffer (PBSM containing 20% (w/v) glycerol). Then 50 l of 12-fold dilutions of reaction mixes were incubated at 25°C in Ni-NTA HisSorb strips (Qiagen) for capture. After washes with the same buffer, bound gephyrin species were detected separately with either anti-Myc, anti-T7, or anti-His6 (BD Biosciences Clontech) antibodies and horseradish peroxidase-conjugated antibody as above. Calibration of tag reactivity (see Fig. 4) was performed in normal Ni-NTA HisSorb strips (Qiagen), except for T7mycGL (2) and T7mycLE, which were adsorbed to strips pretreated with 20 mM EDTA in 0.1 M Na 2 CO 3 /NaHCO 3 , pH 9.6.
Cell Culture and Transfection-African green monkey kidney (COS-7) cells were plated on glass coverslips and grown in Dulbecco's modified Eagle's medium (Invitrogen) containing 10% fetal calf serum (Invitrogen) at 37°C and 7.5% CO 2 . For transfection, the experiments were performed on subconfluent cultures (60% confluency) using the FuGENE 6 (Roche Applied Science) method. Usually, 2 g of plasmid DNA were added to 35-mm dishes. Transient protein expression was allowed to proceed for up to 16 h at 37°C.
Spinal cord neurons were prepared from embryonic Sprague-Dawley rats as described previously (28). Routinely, the cells were plated at a density of 5.10 4 cells/cm 2 onto glass coverslips (Assistent, Winigor, Germany) and grown in neurobasal medium supplemented with B27, 2 mM glutamine, and antibiotics (Invitrogen, France) at 36°C in a 5% CO 2 atmosphere. They were transfected 8 -10 days after plating using the Lipofectamine2000 TM method (Invitrogen) according to the manufacturer's protocol. The cells were usually transfected with 2 g of plasmid DNA in 20-mm wells.
Fluorescence Microscopy-Immunofluorescence labeling was performed essentially as described (31) on cells that were fixed in 4% (w/v) paraformaldehyde. The primary antibodies were used at the following concentrations: mouse anti-GlyR ␣1 subunit: 1 g/ml (clone mAb2b; Synaptic System GmBH), mouse anti-c-Myc monoclonal antibody: 0.5 g/ml (clone 9E10; Roche Applied Science), rabbit anti-synapsin I antibody: 1 g/ml (Chemicon), rat anti-HA monoclonal antibody: 1 g/ml (clone 3F10; Roche Applied Science), rabbit anti-green fluorescent protein antibody: 1 g/ml (BD Biosciences Clontech). The secondary antibodies were: goat anti-mouse or antirabbit antibodies conjugated with fluorescein isothiocyanate, tetramethylrhodamine isothiocyanate, Cy3 or Cy5 (1:250 -1:400; Jackson Immunoresearch Laboratories), or Alexa 488 (Molecular Probes). Extracytoplasmic Myc tags were detected as described (31). Observations were made using the 63ϫ/1.32 and 100ϫ/1.3 objectives of a Leica DMR fluorescence microscope. Fluorescence pictures were usually acquired with a Leica DMR/HCS microscope (63ϫ immersion objective) (Nussloch, Germany) equipped with a CCD camera (Coolsnap; Princeton Instruments). Acquisition was done in black and white in 12-bit mode using appropriate filters, and quantification was done with Metamorph software (Princeton Instruments). GlyR and gephyrin clusters were quantified on digitalized images on neurons from two independent cultures. Integrated synaptic GlyRassociated fluorescence was measured on cells chosen randomly on the basis of Venus::Ge fluorescence. For each transfected neuron the GlyR-associated fluorescence intensity colocalized with synapsin I staining was measured on neurites using the gephyrin fluorescence as a mask, after subtraction of the background signal. All of the measurements were performed using Metamorph software. Statistical significance was determined by means of Student's t test using StatView Software (Abacus Concepts, Berkeley, CA). Fluorescence of clusters was determined on pictures acquired under a 100ϫ objective lens. When required, pictures were pseudocolored with Photoshop (Adobe Systems, San Jose, CA).
Endocytosis Test-The cells were incubated at 2°C with 1 g/m anti-c-Myc antibody for 30 min (prelabeling step) and then extensively washed (time 0 min of chase). They were returned to their initial culture medium pre-equilibrated at 37°C to allow endocytosis for 40 min. Low pH stripping was then performed (time, 40 min of chase) by incubating cells at 2°C in 0.2 M acetic acid, 0.5 M NaCl, pH 2.2, for 7 min. At any step, the cells were either processed for immunofluorescence staining or lysed for two-site ELISA. For two-site ELISA, washed and pelleted cells were solubilized in 50 mM Tris/HCl, 100 mM NaCl, 0.2% (w/v) Triton X-100, pH 8.0, for 30 min on ice. The cleared detergent extract was supplemented with 0.2% (w/v) gelatin, 0.05% (v/v) Tween 20 and incubated for 18 h at 4°C with rabbit anti-mouse antibody coated in 96-well microtiter plates (10 g/ml, 50 l/well). 9E10 IgG was detected with horseradish peroxidase-conjugated rabbit anti-mouse antibody and ABTS reaction as described above.

Interaction of Gephyrin with GlyR Subunit M3-M4 Loops-
The influence of the various domains and cassettes of gephyrin on its functional properties was analyzed with in vitro assays using recombinant proteins obtained after synthesis in E. coli and His 6 -mediated purification ( Fig. 1).
First, in an attempt to mimic their native conformation, we designed recombinant versions of GlyR cytoplasmic M3-M4 loops in which the N and C termini were brought close together. For this, the subunit loop sequences were substituted for the small loop connecting the anti-parallel ␤-strands 6 and 6Ј in the cytohesin PH domain, which contributes to hydrogen bonding with 5-phosphate in inositol-(1,3,4,5)tetrakis-triphosphate (33). The structure and stability of the PH domain are not altered by the insertion of the loop. 4 These chimeras can be considered as displaying the loop in a restricted conformation, and the 18-amino acid gephyrin-binding sequence, ␤gb, should be properly displayed within the ␤ loop (9). Second, a series of tagged, truncated forms of gephyrin in addition to full-length versions were constructed ( Fig. 1) and used for binding experiments. The insertion of a tobacco etch virus protease (TEV) cleavage site in some constructs offered the possibility of inves-tigating gephyrin protomer-protomer interactions in capture assays utilizing the His 6 tag. The appropriate immunochemical reactivity of the sequence tags, as well as the efficiency of TEV digestion, was verified (supplemental Fig. S1).
To characterize a possible differential binding of gephyrin-derived molecules ( Fig. 1) to the cytoplasmic loop of the GlyR ␤ subunit, the interaction of fragment LE was first compared with that of E, taken here as the reference binding entity. In the solid phase assay (first protocol, see "Experimental Procedures") using immobilized PH-␤Loop, maximal binding was reached after 3-4 h of incubation for both ligands and for all of the concentrations tested. In addition to the same kinetics, a saturable association of E and LE molecules to PH-␤Loop was also described by superimposable curves (Fig. 2A). The same K D value in the nanomolar range was estimated for binding to PH-␤Loop. In contrast, very little binding to GST-␤Loop could be detected over this concentration range. A similar and high affinity interaction was also observed with the various LE i variants (Fig. 2B), indicating that none of the cassette sequences (C3, C4, and C4Ј) present in the L domain affect the recognition of PH-␤Loop. Moreover, the association of full-length (GLE) molecules, 4 S. Eimer, C. Bedet, and C. Vannier, unpublished results. when tested with identical ligand to PH-␤Loop concentration ratios (Fig. 2C), revealed not only similar respective binding curves but affinities close, if not identical, to those of the corresponding LE fragment (Fig. 2, compare B and C). Therefore, PH-␤Loop binding to the E domain was not influenced by either G domain, implying that cassette C5 of Ge (2,4,5) does not alter the high affinity interaction between gephyrin and the ␤ loop. The GL construct, which contains neither the GlyR ␤ subunit-binding site nor the E subdomains involved in protomer interface within the E dimer (18), could be used as a control. The curves of Fig. 2D confirmed that G domains had no binding ability whatever the variants tested. In this assay, no binding to the ␣1 loop or to the ␣2 loop occurred (Fig. 2, B and C). These data indicate that the formation of a high affinity complex with the GlyR ␤ subunit cytoplasmic M3-M4 loop is not influenced by variations in the primary structure of gephyrin outside its E domain.
In a previous work (24), we showed that a differential ␤ loop binding of gephyrin variants could be measured using GST pull-down assays. In an attempt to better characterize the effect of the structure surrounding the ␤␥␤ gephyrin-binding sequence of ␤ loop, we probed the association of Ge (2,4,5) and Ge(2) with PH-␤Loop, GST-␤Loop, and PH-␣1␤␥␤Loop using the solid phase assay. As observed with E and LE, the strength of the Ge/GST-␤Loop complex was strikingly low as compared with that of Ge/PH-␤Loop (supplemental Fig. S2A). However, in contrast to the interaction between gephyrins and PH-␤Loop, the distinct ability of the two gephyrin variants to bind GST-␤Loop (as exhibited in GST pull-down assays), was reproduced here, because Ge(2) could be captured more efficiently than Ge(2,4,5) by immobilized GST-␤Loop (supplemental Fig. S2A, inset). This difference was not observed when ␤␥␤ was inserted in the PH-␣1␤␥␤Loop chimera, which binds gephyrin as strongly as PH-␤Loop (supplemental Fig. S2B). This suggests that structural elements promoting gephyrin recognition are maintained in the PH-M3-M4 loop proteins and lost in GST-␤Loop. PH domain-based carriers of the ␤␥␤ sequence thus provide high affinity binding sites for both gephyrin variants. The present results (Fig. 2C) demonstrate that gephyrins Ge (2,4,5) and Ge(2) bind with similar affinities to PH-␤Loop. Furthermore, we show that high affinity binding FIGURE 3. Size exclusion chromatography of gephyrin fragments. A, dotblots of eluted fractions reacted with anti-His antibody after chromatography of HisT7-tagged G, GL, LE, and E fragments from variants Ge(2) and Ge (2,4,5). B, chromatograms obtained by plotting densities of spots shown in A. The exponential calibration curve shown is derived from the calculated linear fit between the K av and logarithm of molecular mass of standards. The molecular mass of the main species eluted beyond aggregates found in the void volume (V 0 , dotted line) is indicated with each peak. The molecular masses predicted for analyzed HisT7-tagged constructs were: G(2), 24.5 kDa; GL(2), 52.9 kDa; G(2,5), 25.9 kDa; GL(2,4,5), 55.8 kDa; E, 50.2 kDa; and LE, 65.8 kDa. to ␤␥␤ is dependent on structural elements that are only maintained in a closed M3-M4 loop conformation.
Size Exclusion Chromatography of Gephyrin Fragments-We then investigated whether cassette composition could affect the quaternary structure of the molecule. Based on reports from others (19), the more stable oligomeric state of purified fulllength gephyrin (GLE) is a trimer behaving as a 300-kDa assembly, in agreement with the crystal structure of the G domain (16,17) but suggesting that the E domain does not possess a conformation suitable for dimerization. Here, only the truncated gephyrin versions were used, potentially able to form either trimers or dimers. Polypeptides were analyzed immediately after purification to minimize self-aggregation. Size exclusion chromatography of fragments corresponding to Ge (2,4,5) and Ge(2) and detected by dot-blotting is illustrated (Fig. 3A). The elution profiles obtained for G(2) and GL(2) were consistent with those of homogeneous trimers with peaks at 81 and 178 kDa, respectively (Fig. 3B). This was not the case for G(2,5) and GL (2,4,5), mainly eluted at volumes nearly corresponding to the size of the monomers (peaks at 27 and 60 kDa, as compared with 25.9 and 55.8 kDa, respectively). This demonstrated a lack of trimer formation by the G domain of Ge (2,4,5). A striking difference was also observed in the behavior of the E and LE fragments. E, which is a dimer at concentrations in the micromolar range (20), was indeed recovered in a single peak (123 kDa), consistent with the size of the dimerized polypeptide (50.2 kDa/monomer). In contrast, LE was recovered as two distinct entities: predominantly a monomeric form (68 kDa, predicted: 65.8) and a dimer (145 kDa), clearly eluting ahead of the E dimer. This result suggested that the L domain does not favor the formation of an LE dimer. These results indicate that trimer formation by the G domain of gephyrin can be prevented by the insertion of cassette C5 and support the notion that E domain dimerization is impeded by the L linker domain.
Gephyrin Oligomer Formation via the G and E Domains-Interactions between either G or E domains of gephyrin were analyzed using a solution assay (second protocol, see "Experimental Procedures"). This assay was designed to measure the potential exchange of protomers between oligomeric forms of gephyrin at equilibrium. Briefly, His-tagged (capture molecules) and Myc-tagged (ligand molecules) fragments of gephyrin (Fig. 1A) were coincubated (at concentrations in the micromolar range), prior to binding to Ni-NTA-coated wells of ELISA plates. The epitope tags associated with the wells were assayed, and the recovery of the Myc tag was taken as the indication of capture and ligand molecule association. His-tagged G or GL fragments from Ge (2,4,5) and Ge(2) (capture molecules) were mixed with Myc-tagged GL (2) (ligand molecule) at varying concentration ratios (Fig. 4, A and B). In these experiments, when T7mycGL (2) was incubated with either HisT7G(2) orHisT7GL (2), the Myc tag was recovered as complexed with the His tag according to a saturable process (Fig. 4A). The progressive decrease in the reactivity of the His tag, despite its obligatory capture, likely results from steric hindrance in antibody binding (compare anti-His 6 reactivity toward trimeric HisT7GL (2) and monomeric HisT7LE80 in Fig. 4D). Therefore, the partial loss of measurable His tag with increasing T7mycGL (2) concentrations is consistent with the appearance of HisT7GL(2) 2 T7mycGL(2) and HisT7GL(2)T7mycGL(2) 2 trimers during incubation. Taken as an internal standard in the test, the amount of captured T7 tag remains constant, as expected if protomers can be exchanged between trimers. When HisT7G(2,5) was used as the capture molecule, no binding of the Myc tag could be detected (Fig. 4B). This confirmed that trimer formation cannot occur between G(2,5) domains and implies that G(2,5) and G(2) cannot form heterotrimers.
A similar experiment was performed using mixes of MycLE with either HisT7E or HisT7LE. It revealed that whereas little capture of mycLE by HisT7E could be obtained, none occurred with HisT7LE (Fig. 4C). These results indicated that LE exists mainly as a monomer. These data therefore support the notion that, whatever its cassette composition, the L domain acts as an inhibitor of the E-E interaction in the solution assay. These experiments corroborated the monomeric state of the LE molecule observed upon size exclusion chromatography.
Ge (2,4,5) and Ge(2) have distinct oligomerization abilities. Therefore, the possibility arises that the gephyrin/receptor coclusters (supplemental Fig. S3, C2 and D2) were not similar with regard to the extent of GlyR bridging. To examine whether the stability of cell surface GlyR was influenced by the nature of gephyrin, transfected COS-7 cells were subjected to an endocytosis test of surface GlyR molecules. The respective typical diffuse or punctate distribution pattern of ␣1␤gb after such labeling is illustrated (Fig. 5, A-C). Coclusters of cell surface GlyR and either Ge (2) or Ge (2,4,5) were formed. After a 40-min chase period at 37°C, the acid wash-resistant labeling was only found in endocytic vesicles or cisternae (Fig. 5, D-F, and supplemental  Fig. S4). At this stage, however, labeled GlyR was still present at the surface of cells not exposed to an acid wash (not shown). The intracellular label (Fig. 5, E and F, arrowheads) was abundant in cells coexpressing Ge(2,4,5), and their fluorescence pat-tern was reminiscent of that found in cells expressing only ␣1␤gb (Fig. 5D). In contrast, very little labeling of intracellular vesicles was observed in cells coexpressing Ge(2) (supplemental Fig. S4, compare E1 and F1). Whether Ge (2,4,5) or Ge(2) was expressed, clusters of gephyrin could be observed beneath the plasma membrane (Fig. 5, E2 and F2, arrows) after the low pH treatment. Plasma membrane gephyrin clusters, devoid of GlyR labeling, were of smaller size in the case of Ge (2,4,5) as compared with Ge(2) (see Fig. 5, E and F, insets). Colocalization of Ge (2,4,5) and internalized GlyR was also observed (Fig. 5, F3,  crossed arrows).
The extent of ␣1 or ␣1␤gb subunit internalization was measured by ELISA of the receptor-bound anti-Myc antibody. (Fig. 5G). The results of two sets of duplicate representative experiments are shown, demonstrating that GlyR-Ge(2) interaction through ␤gb largely contributed to maintain the receptor at the cell surface because ␣1␤gb expressed alone was internalized as rapidly as the gephyrin- non interacting ␣1 in the presence of Ge(2) (Fig. 5G1). Internalization of ␣1␤gb in the presence of Ge (2,4,5) occurred as rapidly as that of ␣1 (Fig. 5G2). Altogether, our data demon-strate that the interaction of cell surface GlyR with Ge (2,4,5), in contrast to that involving Ge(2), was not sufficient to prevent internalization. Consequently, the cell surface stability of GlyR, in addition to binding to gephyrin, is also determined by the oligomeric state of gephyrin itself.
Gephyrin Can Displace GlyR from Synaptic Loci-The reduced cell surface stability of GlyR when associated with small clusters of Ge (2,4,5) in COS-7 cells suggested that the latter could interfere with Ge(2) to maintain the receptor under a clusterized form in neurons. This hypothesis deserved to be tested directly in neurons because, very likely, several of the variants identified in the adult spinal cord are synthesized in cells undergoing synaptogenesis (supplemental Fig. S5). The influence of Ge (2) and Ge (2,4,5) variants on synaptic GlyR was thus investigated in transfected spinal cord neurons expressing Venus-tagged full-length molecules (Fig. 6). After transfection GlyR immunoreactivity was always predominantly found as spots associated with the presynaptic immunoreactivity of Synapsin I (as illustrated in Fig. 6C). When expressed alone, both Ge(2) and Ge (2,4,5) were able to populate synaptic domains, as revealed by fluorescent clusters colocalizing with endogenous synaptic GlyR shortly after transfection (Fig. 6, A and B). In contrast to Ge(2), a high proportion of the Ge(2,4,5) remained diffusely localized in the cytoplasm. This observation was consistent with the failure of Ge (2,4,5) to polymerize and its reduced ability to be recruited to submembranous, GlyR-associated aggregates formed in COS-7 cells. The effect of Ge(2) and Ge (2,4,5) on the fate of synaptic GlyR was therefore examined in neurons in which Venus-tagged Ge(2) was expressed alone or coexpressed with either untagged Ge(2) or Ge (2,4,5) for an 18-h period (Fig. 6, C-E). The immunofluorescence labeling of GlyR and of synapsin I confirmed that Venustagged Ge(2), predominantly colocalized with GlyR (Fig. 6, C and D, arrows in enlarged fields) at synaptic loci, did not modify the receptor subcellular distribution whether or not its untagged version (Ge(2)) was coexpressed (Fig. 6, compare C and D). In contrast, GlyR localization was dramatically altered when Venus-tagged Ge(2) and untagged-Ge (2,4,5) were coexpressed (Fig. 6E). Typically, a diffuse distribution pattern of the Venus fluorescence over the entire cytoplasmic compartment was now observed and superimposed on that of Ge (2,4,5) expressed in the same cell (supplemental Fig. S6). This redistribution was accompanied with a disappearance of GlyR clusters from the postsynaptic domains. The stability of GlyR under the above conditions was scored by counting clusters of both Venus-tagged gephyrin and endogenous receptor ␣1 subunit over individ-ual dendrites (Fig. 6F). Clearly, the synthesis of Ge (2,4,5) in transfected cells was responsible for an almost parallel loss of both cluster populations even in the presence of a coexpressed pool of Ge (2). The intrinsic fluorescence of GlyR clusters remaining after a 18-h period under these conditions was measured (Fig. 6G) but showed no significant change. These results favor the notion that Ge (2,4,5), which can bind both Ge(2) (and endogenous gephyrin) and GlyR, was acting as a dominant-negative variant in the formation of gephyrin/GlyR coclusters. This could only result from the titration by Ge(2,4,5) of its two binding partners, which acquired the diffuse distribution expected for small complexes. Altogether these data show that Ge (2,4,5), which exhibits a limited oligomerization potential, can compete with Ge(2) for the binding to GlyR, leading to a loss of synaptic receptor upon expression.

DISCUSSION
In this work, we demonstrate that: 1) the ability of gephyrin to stabilize GlyR at the cell surface is correlated with its trimerization and 2) gephyrin unable of trimerization displaces both gephyrin and GlyR from postsynaptic loci, when expressed in cultured neurons. These results suggest that modifications of the relative cellular expression level of gephyrin variants influence the turnover and number of GlyR at synapses.
Gephyrin Interacts with GlyR Independently of Its Structure-Gephyrin is likely the main regulator of receptor organization. The identification of the gephyrin E domain as bearing the GlyR-binding site has been largely substantiated recently (19,20,25). In this study we designed a new tool to probe GlyRgephyrin interaction. A soluble pleckstrin homology domain was used as a carrier for isolated GlyR large cytoplasmic loops to maintain their native tertiary structure. We present evidence that a high affinity interaction occurs between the large gephyrin E domain and the cytoplasmic loop of the GlyR ␤ subunit when presented as the PH-␤Loop chimera. ELISA-based assays revealed PH-␤Loop binding characteristics consistent with surface plasmon resonance experiments reported by Sola et al. (19), indicating dissociation constants in the nanomolar range. Our data now establish that ␤ subunit association with the E domain is not modulated by cassettes encountered in the gephyrin L domain studied here. Notably, the E domain displayed either alone or in truncated or full-length molecules had unchanged binding capacity and affinity for PH-␤Loop. C-E, triple labeling of Venus::Ge(2), GlyR, and Synapsin I in representative cells transfected with either pC-Venus::Ge(2) (C ), pC-Venus::Ge(2) and pC-Ge(2) (D), or pC-Venus::Ge(2) and pC-Ge(2,4,5) (E ). The cells were fixed 18 h after transfection. Higher magnifications of merged fields selected in C, D, and E are shown on the right. Note that in the presence of Ge (2,4,5), gephyrin/GlyR coclusters (arrows) disappear from synaptic sites identified by Synapsin I. F, quantitative analysis of GlyR and fluorescent gephyrin clusters in cells transfected by the indicated gephyrin combination and after a 18-h expression time. The asterisk refers to Venus tagging of either Ge (2) or Ge (2,4,5). The mean number of clusters (ϮS.E.) from two separate experiments is expressed per 100 m of neurite length. A significant reduction in number of Ge clusters was found when comparing cells expressing Ge(2)*ϩGe(2) with those expressing Ge(2)*ϩGe (2,4,5) (p Ͻ 0.05, n ϭ 13 and 14, respectively). When cells that expressed Ge(2)* alone and Ge(2,4,5)* alone were compared, the results were similar (p Ͻ 0.05, n ϭ 20 and 10 cells, respectively). The number of GlyR clusters was also reduced in both comparisons (p Ͻ 0.05). Note that the lower number of GlyR clusters as compared with synaptic gephyrin clusters under control conditions is explained by the selective recognition of the ␣1subunit, which is present in ϳ40% of glycinergic synapses in neurons taken 8 -10 days after plating. G, quantification of GlyR cluster-associated fluorescence (mean Ϯ S.E.) in experiment of Fig. 6F. (p ϭ 0.2, n ϭ 103 clusters and 32 clusters for Ge(2)*ϩGe (2) and Ge(2)*ϩGe(2,4,5), respectively). Bars, 10 m.
This indicates that the E domain association with ␤ subunit is autonomous.
We have found no detectable effect of cassette C5, when present in the G domain of full-length gephyrin, on the PH-␤Loop binding. Clearly, the measurement of the strength of the interaction between gephyrin and its binding site on ␤ subunit, ␤gb, depends on the experimental design (see Ref. 24). Schrader et al. (20) also showed that affinities almost 2 orders of magnitude higher could nevertheless be detected by isothermal titration calorimetry, corroborating th4e surface plasmon resonance data of Sola et al. (19) performed with the same GST-␤Loop chimera. Our data favor the notion that the ␤gb sequence is not displayed properly when the entire ␤ subunit M3-M4 loop is fused by its N terminus to a leader protein. In contrast, it becomes a strong binding site in vitro and in vivo ( Fig. 2 and supplemental Fig.  S2), when the fusion protein mimics the topology of the ␤ subunit loop. We propose that ␤gb is optimally recognized by gephyrin in a folded, constrained structure provided by proteins such as the native ␤ subunit, the ␣1␤␥␤ subunit, and PH-␤Loop or PH-␣1␤␥␤ loop, but not GST-␤Loop. The possibility of measuring strong interactions with chimera of the 49-residue sequence is plausibly due to the absence of additional and misfolded polypeptide impeding recognition.
Trimerization and Dimerization Are Controlled by Gephyrin Structure According to Distinct Mechanisms-Intrinsic structural elements govern the oligomerization of gephyrin. On the one hand, G domain trimerization can be prevented in a splice variant harboring cassette C5, and on the other hand, dimer formation by the E domain is impeded by the central L domain. The basic molecular form of full-length gephyrin is a highly stable trimer as recently determined by dynamic light scattering and chemical cross-linking (19,20). Such quaternary structure agrees well with the strength of the interaction between protomers in purified G domain trimers and suggests that the E domain, which otherwise forms dimers when isolated, is in a metastable conformation within the full-length polypeptide (19,20). These quaternary structures of gephyrin are in both cases consistent with their behavior of soluble proteins when expressed in transfected cells.
We now provide evidence of a lack of trimer formation by Ge (2,4,5) resulting from the presence of cassette C5. This cassette, which has been found following cloning of several rat gephyrin variants, was also identified in the human gene and attributed to exon 5 (24,34). Actually, a mere alteration of the trimer interface might alter the formation of gephyrin oligomers (16). We propose that the Ge(2,4,5) monomer results from the disruption of the highly conserved ␣-helix 4 upon insertion of the 13 residues of splice cassette C5 between residues Glu 98 and Ala 99 (16,17,23). Interestingly, we found that monomeric full-length Ge (2,4,5) still binds GlyR ␤ subunit, in agreement with evidence that dimerization of the E domain is not required for GlyR binding (20).
The inhibition of E domain dimerization in full-length gephyrin results either from an inappropriate conformation or from the masking of the interface, which is defined between protomers in the crystal structure (19). Altogether, available data and our findings support the notion that the two types of oligomerization events within holo-gephyrin are not simultaneous. Regulation of dimerization is therefore likely to be a mechanism whereby formation of GlyR clusters is achieved. A, the gephyrin polymer represented as a branched structure (see text) is composed of trimer-and nontrimer-forming gephyrins which differ in their G domain (black). For the sake of clarity the L domain is not represented, and E domains are all drawn as light blue triangles. B, assembly and disassembly of the lattice. Assuming that the E domain is metastable and that gephyrin lacking cassette C5 is a stable monomer (see text), the effect of Ge (2,4,5) and Ge(2) harboring or not harboring cassette C5, respectively, is illustrated. Trimeric gephyrins devoid of cassette C5 could be added to the lattice as a whole via their E domain (part 1). They could also behave as donors of protomers either inserted in the polymer via their G domain only (part 2) or via both G and E domains following two exchange reactions (part 3). As a consequence, the building block of the lattice is not necessarily the trimeric gephyrin. In contrast, gephyrins harboring cassette C5 would be exchanged through their E domain only (part 4) because of their failure to heterotrimerize. Gephyrin variants can exit or enter the polymer (lattice), complexed or not with partners. Newly incorporating E domains are in dark blue or yellow. Note the polymer disruption following insertion of Ge (2,4,5) and its extending organization with Ge(2). Furthermore, the control of dimerization would prevent inappropriate lattice formation outside synapses.
Control Receptor Cell Surface and Synaptic Stability by Gephyrin Oligomerization-We found in a previous study (32) that gephyrin contributes to the accumulation of GlyR in the plasma membrane when interaction is permitted. Such a situation may in part rely on coclustering events similar to those already reported for Kv1.4 and Kv4.2 channels, which undergo reduced internalization when complexed with postsynaptic differentiation-95 (35,36). The distinct effects of Ge (2,4,5) and Ge(2) on endocytosis of bound GlyR support two conclusions: 1) a GlyR-interacting gephyrin variant unable to trimerize is not sufficient to stabilize GlyR at the cell surface and 2) formation of complexes between GlyR and gephyrin variants with altered oligomerization can lower the half-life of this cell surface receptor. Therefore, in addition to receptor-gephyrin interaction, the stability of GlyR in the plasma membrane also depends on the extent of gephyrin oligomerization. As a matter of fact, gephyrin variant Ge (2,4,5), which can transiently populate postsynaptic loci in neurons, is responsible for the removal of GlyR from these sites. The parallel loss of postsynaptic cassette 5-lacking gephyrin provides evidence that Ge (2,4,5) acts as a dominant-negative molecule titrating both receptor and gephyrin within synapses. The disappearance of GlyR from synapses could therefore result from the formation of small GlyR-Ge(2,4,5) complexes able to undergo diffusion in the plasma membrane and subsequent internalization (37).
Postsynaptic Scaffold and Regulation of GlyR Clustering-Our findings now allow us to propose a model for the formation and plasticity of the inhibitory postsynaptic lattice (Fig.  7). The general organization (18,22) of the hexagonal gephyrin lattice remains. However, variants of gephyrin allow the construction of a dynamic lattice, which influences GlyR clustering. It is assumed that E domains within the lattice can be engaged in dimers for a majority of gephyrin variants because of the low frequency of splicing in this region (however, see Ref. 12). This is not necessarily the case because dimerization could be modulated by specific partner proteins or phosphorylation. Based on our data, a disruption of the hexagonal oligomerization pattern results from a failure of the trimerization process. In the presence of a mixture of the two types of gephyrin, Ge (2,4,5) and Ge (2), this mechanism would contribute to a regulated steady state size of a nonhexagonal, branched polymer (Fig. 7). Moreover, changes in the relative expression level of these variants would result in disassembly or assembly of the scaffold (Fig. 7).
The gephyrin lattice assembly/disassembly substantiated here has consequences for the local density as well as for the dynamics of its partners and is consistent with the diffusionretention mechanism underlying clustering of GlyR (31,37,38). In particular, gephyrin may shuttle between the lattice and the cytosolic phase in a GlyR-bound state. Our data and model thus challenge the hypothesis that simple incorporation of gephyrins harboring C5 in a lattice might selectively reduce the density of GlyR at synaptic sites but not that of ␥-aminobutyric acid receptor (39). Instead, the association of GlyR with C5-containing gephyrin present in the lattice is possible, but impairment of dimerization would drive redistribution of such a complexed receptor toward the extrasynaptic space. The cassette-dependent oligomerization of gephyrin has mechanistic implications for both the control of GlyR clustering at synapses and dynamic events in the postsynaptic differentiation. As a conclusion, we provided evidence that the functional heterogeneity of gephyrin with regard to lattice formation creates an additional, nonexclusive mechanism regulating the presence, density, and dynamics of the inhibitory postsynaptic differentiation components.