Induction of GABAergic Postsynaptic Differentiation by α-Neurexins*

β-Neurexin and neuroligin cell adhesion molecules contribute to synapse development in the brain. The longer α-neurexins function at both glutamate and γ-aminobutyric acid (GABA) synapses in coupling to presynaptic calcium channels. Binding of α-neurexins to neuroligins was recently reported, but the role of the α-neurexins in synapse development has not been well studied. Here we report that in COS cell neuron coculture assays, all three α-neurexins induce clustering of the GABAergic postsynaptic scaffolding protein gephyrin and neuroligin 2 but not of the glutamatergic postsynaptic scaffolding protein PSD-95 or neuroligin 1/3/4. α-Neurexins also induce clustering of the GABAA receptor γ2 subunit. This synapse promoting activity of α-neurexins is mediated by the sixth LNS (laminin neurexin sex hormone-binding protein) domain and negatively modulated by upstream sequences. Although inserts at splice site 4 (S4) in β-neurexins promote greater clustering activity for GABA than glutamate proteins in coculture assay, α-neurexin-specific sequences confer complete specificity for GABA proteins. We further report a developmental increase in the ratio of -S4 to +S4 forms of neurexins correlating with an increase in glutamate relative to GABA synaptogenesis and activity regulation of splicing at S4. Thus, +S4 β-neurexins and, even more selectively, α-neurexins may be mediators of GABAergic synaptic protein recruitment and stabilization.

␤-Neurexin and neuroligin cell adhesion molecules contribute to synapse development in the brain. The longer ␣-neurexins function at both glutamate and ␥-aminobutyric acid (GABA) synapses in coupling to presynaptic calcium channels. Binding of ␣-neurexins to neuroligins was recently reported, but the role of the ␣-neurexins in synapse development has not been well studied. Here we report that in COS cell neuron coculture assays, all three ␣-neurexins induce clustering of the GABAergic postsynaptic scaffolding protein gephyrin and neuroligin 2 but not of the glutamatergic postsynaptic scaffolding protein PSD-95 or neuroligin 1/3/4. ␣-Neurexins also induce clustering of the GABA A receptor ␥2 subunit. This synapse promoting activity of ␣-neurexins is mediated by the sixth LNS (laminin neurexin sex hormone-binding protein) domain and negatively modulated by upstream sequences. Although inserts at splice site 4 (S4) in ␤-neurexins promote greater clustering activity for GABA than glutamate proteins in coculture assay, ␣-neurexinspecific sequences confer complete specificity for GABA proteins. We further report a developmental increase in the ratio of ؊S4 to ؉S4 forms of neurexins correlating with an increase in glutamate relative to GABA synaptogenesis and activity regulation of splicing at S4. Thus, ؉S4 ␤-neurexins and, even more selectively, ␣-neurexins may be mediators of GABAergic synaptic protein recruitment and stabilization.
Synaptic connections in the brain require precise alignment of neurotransmitter receptors on dendrites opposite transmitter release sites on axons. Neurexin and neuroligin cell adhesion molecules are thought to function in development and maintenance of the two main synapse types, excitatory glutamatergic and inhibitory GABAergic 3 synapses (reviewed in Refs. 1 and 2). Neuroligins, normally present on dendrites, alone are sufficient to induce presynaptic differentiation when presented to axons of cultured neurons (3)(4)(5). Conversely, neurexins, normally present on axons, alone are sufficient to induce postsynaptic differentiation when presented to dendrites of cultured neurons (6,7). These coculture studies suggest that neurexins and neuroligins play some role in synapse development in vivo, perhaps in protein recruitment and stabilization of synaptic complexes. There are 4 -5 neuroligins in mammals. Neuroligin 1 localizes primarily at glutamatergic synapses (8), and neuroligin 2 localizes primarily at GABAergic synapses (6,9). Six main neurexin isoforms are derived from three genes (1-3) and two promoters each (␣ and ␤) (10). The shorter ␤-neurexins bind neuroligins via their single LNS domain (11). Alternative splicing at multiple sites also contributes to neurexin and neuroligin diversity. In particular, the absence of the splice site 4 (S4) insert in ␤-neurexins and presence of the B insert in neuroligin 1 selectively promotes function at glutamatergic synapses (12)(13)(14)(15).
The importance of these protein families for human cognition is suggested by the linkage of rare mutations in neuroligins 3 and 4 to autism and mental retardation (16,17) and the association of variations in neurexin 1 with autism (18,19). Mice lacking the major neuroligins 1/2/3 exhibit defects in inhibitory and excitatory synaptic transmission and die within 24 h of birth (20). Selective loss of neuroligin 1 or 2 results in selective defects in glutamate or GABA synapses, respectively (21). Overexpression or knockdown of neuroligins in cultured neurons also affects synapse development, altering the function and ratio of excitatory and inhibitory synapses (22)(23)(24)(25). A role for neuroligin in development of neuronal cholinergic synapses has also recently been found (26).
␣-Neurexins terminate in the same LNS domain, transmembrane region, and intracellular region with PDZ domain binding site as ␤-neurexins but contain additionally five LNS domains and three epidermal growth factor (EGF)-like domains organized into three modules. Analysis of triple knock-out mice for ␣-neurexin 1/2/3, leaving expression of the ␤-neurexins intact, revealed a surprising function in coupling presynaptic calcium channels to synaptic vesicle exocytosis (27). ␣-Neurexins function in transmitter release linked to N-and P/Q-type calcium channels at central nervous system synapses, in calcium-triggered exocytosis of secretory granules in endocrine cells, and to a lesser degree contribute to efficient neuromuscular transmission (28 -30). Intracellularly, ␣and ␤-neurexins bind CASK, Mint, syntenin, and synaptotagmin (31)(32)(33)(34). Extracellularly, ␣-neurexins presumably have unique binding partners to explain the calcium channel coupling phenotype that is not shared with ␤-neurexins. Known extracellular binding partners of ␣-neurexins include dystroglycan (35) and the secreted peptides neurexophilins (36), although the significance of these interactions is not well understood. Originally, it was reported that ␣-neurexins did not bind neuroligins (11), but while the current work was in progress it became clear that ␣-neurexins also bind some neuroligins (12). Neurexin-1␣ with or without the S4 insert binds to neuroligins 2 and 3 and the minor variant of neuroligin 1 lacking a B insert but not to the major variant of neuroligin 1 containing a B insert (Fig. 3 of Boucard et al. (12)). Neurexin 1␣ lacking the S4 insert was also reported to induce clustering of gephyrin and neuroligin 2 but not PSD-95 when presented on COS cells to dendrites of cultured hippocampal neurons (14).
We set out here to characterize the activity of ␣-neurexins in COS cell neuron coculture assays; that is, to test the ability of ␣-neurexins to induce glutamatergic and/or GABAergic postsynaptic differentiation in comparison with ␤-neurexins. We further explore the structural basis for the difference in synapse-promoting activity of ␣compared with ␤-neurexins and the regulated expression patterns of neurexins with an emphasis on regulation at the S4 splice site.

EXPERIMENTAL PROCEDURES
Primary Neuronal Culture and COS Cell Coculture-Dissociated primary hippocampal neuronal cultures were prepared from embryonic day 18 (E18) rats as previously described (37,38). Hippocampi were dissociated by trypsinization and trituration. Dissociated neurons were plated onto poly-L-lysine-coated glass coverslips in 60-mm culture dishes at a density of 300,000 cells/dish and cocultured over a monolayer of glia. After 2 days, cytosine arabinoside (5 M) was added to neuron cultures to prevent the overgrowth of glia. Cultures were maintained in serum-free minimal essential medium with N2 supplements, 0.1% ovalbumin, and 1 mM pyruvate, with replacing 1 ⁄ 3 of the media per dish once per week. Neurons were treated with 100 M 2-amino-5-phosphonopentanoic acid (APV, Research Biochemicals) beginning on day 7. COS-7 and HEK293 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and transfected with Lipofectamine 2000 (Invitrogen). The flat shape of COS cells was preferred for coculture compared with the rounder shape of HEK cells preferred for confocal analysis of surface association. Transfected COS cells were trypsinized 24 h after transfection, washed, and plated at 200,000 cells/dish onto neurons pre-grown for 8 -12 days in vitro (DIV). After 24 h of coculture, cells were fixed for 15 min in warm phosphate-buffered saline with 4% paraformaldehyde and 4% sucrose and permeabilized with 0.25% Triton X-100. For experiments involving immunocytochemistry for neuroli-gins in the cocultures, cells were fixed in Ϫ20°C methanol for 10 min.
Construction of Expression Vectors-Neurexin-1␤-cyan fluorescent protein (CFP) (ϩS4) was described previously (13). To generate neurexin-1␣-CFP, the N-terminal portion of rat neurexin 1␣ was cloned by RT-PCR and joined with the C-terminal portion of the mouse cDNA (BC047146; Open Biosystems) at the internal BstEII site and inserted in-frame into the pECFP-N1 vector (Clontech). Neurexin 2␣ (AK129239) and neurexin 3␣ (BC060719) were first corrected for apparent errors and then cloned into pECFP-N1. For neurexin 2␣, the Stratagene site-directed mutagenesis kit was used to restore CAG in place of a premature TAG at residue 1301. For neurexin 3␣, the splice site 1 insert consisting of an apparent duplication of exons 2 and 3 was removed by an overlap PCR method. To generate neurexin 2␤Ј and 3␤Ј, the LNS domain of neurexin 1␤ (residues 84 -261) was replaced with the equivalent residues of LNS6 of neurexin 2␣ or 3␣, altering the junctional amino acids GT to EF and amino acids EV to ST to facilitate cloning. Neurexin 1␣BC contained a deletion of residues 31-473, and neurexin 1␣C contained a deletion of residues 31-899 (numbering according to (39), each with the addition of an extra LV at the junction. All neurexin variants used in this paper have the insert at the splice site 4 position.
Immunocytochemistry and Imaging-Fixed neuron/COS cell cocultures were blocked with 10% bovine serum albumen (30 min; 37°C) and incubated with appropriate primary antibodies in phosphate-buffered saline with 3% bovine serum albumen (overnight; room temperature) and then with secondary antibodies (45 min; 37°C). Coverslips were mounted in elvanol (Tris-HCl, glycerol, and polyvinyl alcohol with 2% 1,4diazabicyclo(2,2,2)octane). For Figs. 1 and 2, cocultures were stained with anti-gephyrin (mAb7a, IgG1, 1:500; Synaptic Systems), anti-PSD-95 (6G6 -1C9, IgG2a, 1:500; Affinity Bioreagents), and anti-synapsin (rabbit, 1:1000; Chemicon) followed by secondary antibodies conjugated to Alexa 488, Alexa 568, and Alexa 647 (Molecular Probes), respectively. For  (40)) or ␥2 antibodies for 30 min in the neuronal medium plus 50 mM HEPES, pH 7.4, at room temperature, washed extensively, fixed, permeabilized, and then incubated with anti-synapsin followed by secondary antibodies. Fluorescence and phase contrast images were captured with a Photometrics Sensys-cooled CCD camera mounted on a Zeiss Axioplan microscope with a 63ϫ 1.4 numerical aperture oil objective using MetaMorph imaging software (Molecular Devices) and customized filter sets. Controls lacking specific antibodies confirmed no detect-␣-Neurexins Induce GABAergic Postsynaptic Differentiation able bleed-through between channels CFP, Alexa 488 (imaged through a yellow fluorescent protein filter set), Alexa 568, and Alexa 647. Images were acquired as grey scale from individual channels, and pseudo-color overlays prepared using Adobe Photoshop software. For quantification, sets of cells were cocultured and stained simultaneously. Images of transfected COS cells showing extensive contact with dendrites were taken in all fluorescence channels using the same exposure time for all constructs. For supplemental Fig. S4, optical sections were captured on an Olympus Fluoview FV500 confocal on a BX61W microscope with a 60ϫ 1.4 numerical aperture oil objective, 442-nm laser, and customized filter set.
Image Analysis-For quantification for Figs. 3 and 4, images were randomized so that the experimenter was blind to the transfection group. The area for measuring was defined by the perimeter of the transfected COS cell. Images of the presynaptic and postsynaptic proteins were thresholded. For each postsynaptic protein cluster, a region was drawn around each cluster, and the area and total gray value was measured. Thresholded synapsin was measured through postsynaptic protein regions to determine which clusters were synaptic. Postsynaptic protein clusters that were apposed to synapsin were excluded from the final quantification. Analysis was performed using MetaMorph and Microsoft Excel. All data are reported as mean Ϯ S.E.
For quantification for Fig. 6, cocultures were assessed blind to the transfection group. On each coverslip all neurexin-CFPtransfected COS cells having significant contact with neuronal dendrites based on phase contrast were assessed (up to 40 cells/ coverslip). The area associated with each chosen COS cell was then visually scored as either positive or negative, positive indicating the presence of any clusters of postsynaptic protein (PSD-95, gephyrin, or neuroligin 2) that lacked associated staining for synapsin.
RNA Isolation and RT-PCR-For RNA extraction from brain, either two hippocampi or roughly 50 mg of cortex was dissected from rats at embryonic day 18 (E18), postnatal day 11 (P11), or adult and rapidly homogenized in 1 ml of Trizol reagent (Invitrogen), and RNA was prepared according to the manufacturer's protocol. For RNA extraction from hippocampal cultures, roughly 1 ϫ 10 6 cells that were either untreated or chronically treated with 100 M APV were harvested by briefly washing in cold phosphate-buffered saline and either scraping directly in Trizol reagent or by using the RNeasy mini kit (Qiagen) according to the manufacturer's protocol. First-strand cDNA synthesis was performed on 3 g of total RNA for brain tissue samples or 300 ng of total RNA for hippocampal culture samples in a total volume of 50 l using 500 units of SuperScript III reverse transcriptase (Invitrogen) and oligo(dT) primer. Extension of cDNA was performed at 50°C for 60 min. The cDNA was then treated with 4 units of ribonuclease H at 37°C for 30 min. PCR amplification was carried out on 0.5 l of each above cDNA template in a total volume of 20 l using 1 unit of TaqDNA Polymerase (Invitrogen). Thermal cycling parameters were denaturation at 95°C for 30 s, annealing at 57°C for 30 s, extension at 72°C for 40 s. PCR was performed for 25-40 cycles, and the entire reaction was loaded on 2% agarose gel for electrophoresis. After staining and visualization with ethidium bromide, semiquantitation of band intensities was performed using ImageJ software. Two independent RNA preparations were made from independent animals or hippocampal cultures, and RT-PCR was performed in triplicate. All data are reported as the mean Ϯ S.E.

Specific Induction of GABAergic Postsynaptic Differentiation
by ␣-Neurexins-We determined here the synaptogenic activity of neurexin 1␣, 2␣, and 3␣ in COS cell neuron coculture (Figs. 1 and supplemental Fig. S1). The sixth LNS domain of ␣-neurexins, in common with ␤-neurexins, mediates binding to specific neuroligins (12). Thus, for comparison, we assayed in the same experiments the sixth LNS domain of neurexin 1␣, 2␣, and 3␣ each in the context of the neurexin 1␤-flanking sequences, corresponding to neurexin 1␤ itself, and constructs termed neurexin 2␤Ј and neurexin 3␤Ј (similar to neurexins 2␤ and 3␤ but with the regions flanking the LNS domain derived from 1␤; Figs. 2 and supplemental Fig. S2). We expressed the neurexin variants tagged intracellularly with CFP in COS cells, cocultured these with hippocampal neurons, and assayed for ability to induce postsynaptic differentiation (as in Refs. 6 and 13). The ability of each neurexin to induce clustering of the inhibitory postsynaptic scaffolding protein gephyrin or the excitatory postsynaptic scaffolding protein PSD-95 in contacting dendrites was determined. Co-labeling for the presynaptic marker synapsin was used to identify the few endogenous interneuronal synaptic clusters that may underlie the transfected COS cells since these bona fide synapses label for PSD-95 or gephyrin plus synapsin. Additional clusters of PSD-95 or gephyrin lacking apposed synapsin immunofluorescence were specifically associated with COS cells expressing neurexins and are referred to as "induced clusters" throughout this paper. Such non-synaptic clusters of PSD-95 or gephyrin were not observed associated with control COS cells expressing membrane-associated CFP (mCFP; Fig. 1B). Induced clusters of PSD-95 or gephyrin were frequently but not always localized near the edges of neurexin-expressing COS cells, presumably where the COS cells come into closest contact with the neighboring neurons. Most often, the inducing neurexin was highly thus, clusters appear turquoise or purple, respectively, in the color overlays (arrows). In contrast, clusters of gephyrin lacking associated synapsin appear red in the color overlay and were induced by all ␣-neurexins (arrowheads). By alignment with the phase contrast images, induced clusters of gephyrin can be seen corresponding to sites of neuronal process contact with the neurexin-expressing COS cells and were frequently but not always localized near the edges of ␣-Neurexins Induce GABAergic Postsynaptic Differentiation FIGURE 2. ␤-Neurexins induce clustering of the inhibitory synaptic scaffolding protein gephyrin and the excitatory synaptic scaffolding protein PSD-95. COS cells expressing the neurexin variants tagged intracellularly with CFP were overlaid on hippocampal neurons pre-grown for 8 -10 days in culture. After 1 day of coculture, cells were fixed and immunolabeled for PSD-95, gephyrin, and synapsin; full field overlays are shown of PSD-95 (green) plus synapsin (blue) or gephyrin (red) plus synapsin (blue). Endogenous inter-neuronal synapses have apposed clusters of postsynaptic PSD-95 or gephyrin and presynaptic synapsin; thus, clusters appear turquoise or purple, respectively, in the color overlays (arrows). In contrast, clusters of PSD-95 or gephyrin lacking associated synapsin appear green or red, respectively, in the color overlays and were induced by all ␤-neurexins (arrowheads). By alignment with the phase contrast images, induced clusters of PSD-95 and gephyrin can be seen corresponding to sites of neuronal process contact with the neurexin-expressing COS cells and were frequently but not always localized near the edges of transfected COS cells, presumably where the COS cells come into closest contact with the neighboring neurons. Although the induced clusters of gephyrin and PSD-95 appear to overlap at low resolution, at higher resolution they usually resolve into separate clusters side by side (enlarged regions at the bottom of triple overlay of PSD-95 (green), gephyrin (red), and synapsin (blue)). All neurexin constructs used in this paper contain the splice site 4 insert. See also supplemental Fig. S2 for single channel larger field of view and phase contrast images corresponding to the cropped fields shown here for neurexin 2␤-CFP. Scale bar, 10 m.
transfected COS cells, presumably where the COS cells come into closest contact with the neighboring neurons. Clusters of PSD-95 lacking associated synapsin were not induced by any ␣-neurexins. B, in sister coculture experiments, control COS cells expressing mCFP did not induce clustering of either PSD-95 or gephyrin. All neurexin constructs used in this paper contain the splice site 4 insert. See also supplemental Fig. S1 for single channel larger field of view and phase contrast images corresponding to the cropped field shown here for neurexin 2␣-CFP. Scale bar, 10 m.

␣-Neurexins Induce GABAergic Postsynaptic Differentiation
expressed over the entire COS cell and not visibly clustered, although rarely neurexin clusters on the COS cell were observed associated with dendritic protein clusters (e.g. Fig. 3 of Graf et al. (6)).
All three ␣-neurexins induced prominent clustering of gephyrin but no detectable clustering of PSD-95 in contacting dendrites (Figs. 1 and supplemental Fig. S1). In contrast, like neurexin 1␤ (13), neurexin 2␤Ј and 3␤Ј induced prominent clustering of both gephyrin and PSD-95 (Figs. 2 and supplemental Fig. 2). Although the induced clusters of gephyrin and PSD-95 appear to overlap at low resolution, at higher resolution they usually resolve into separate clusters side by side (supplemental Fig. S2 and Graf et al. (6)). Quantitation of clusters of PSD-95 or gephyrin lacking synapsin was performed on random sets of COS cells expressing each neurexin variant in comparison with the base-line clusters associated with COS cells expressing only mCFP. This quantitation confirmed a significant induction of PSD-95 clusters only by the three ␤-neurexin constructs but none of the ␣-neurexins and induction of gephyrin clusters by all neurexin variants tested (Fig. 3, p Ͻ 0.05 or p Ͻ 0.005 as indicated, n ϭ 32-35 cells each). For induction of gephyrin clustering, the ␣-neurexins exhibited weaker activity than the ␤-neurexin constructs but were still 2-7-fold above the background of mCFP. In this same set of experiments showing differences in the ability to induce postsynaptic protein clustering, expression levels per COS cells did not differ among constructs (average relative CFP intensity values for each neurexin were 92 for 1␣-CFP, 103 for 1␤-CFP, 85 for 2␣-CFP, 97 for 2␤Ј-CFP, 113 for 3␣-CFP, and 110 for 3␤Ј-CFP; ANOVA p Ͼ 0.1).
The same type of coculture assay can be used to test the recruitment of neuroligin 1/3/4 (using an antibody that recognizes a common epitope) compared with neuroligin 2 to contact sites of dendrites with neurexin-expressing COS cells. Neuroligin clusters induced by neurexin contact were again differentiated from endogenous interneuronal synaptic clusters by the absence of associated synapsin immunoreactivity. We again observed a differential activity between ␣ compared with ␤ neurexins. Although the area of induced clusters of the neuroligins tends to be larger than that of gephyrin or PSD-95, covering a larger contact zone rather than being split into clusters more typical of a synaptic size (Ref. 49 and compare Figs. 1, 2, and 4), we could detect no significant clustering of neuroligin 1/3/4 by the ␣-neurexins (Fig. 4). All three ␣-neurexins induced prominent clustering of neuroligin 2 but not neuroligin 1/3/4 in contacting dendrites. In contrast, like neurexin 1␤ (13), neurexin 2␤Ј and 3␤Ј induced prominent clustering of both neuroligin 2 and neuroligin 1/3/4. Although this neuroligin 1/3/4 antibody recognizes all three neuroligins (41), neuroligins 1 and 3 are present at similar levels, but neuroligin 4 is not detectable in neonatal rodent tissue (20). Furthermore, neuroligin 1 is mainly expressed in the ϩB splice form (14), and neuroligin 1 ϩB binds neurexin 1␤ with 2-6-fold greater affinity than neuroligin 3 (15). Thus, the major variant clustered by ␤-neurexins but not ␣-neurexins and recognized by the neuroligin 1/3/4 antibody in this study is likely neuroligin 1 ϩB.
Finally, we tested for the ability of each neurexin construct to induce clustering of the two other major components of GABAergic synapses, GABA A receptors and dystroglycan. The major synaptic GABA A receptor subunit ␥2 (42), like gephyrin, was clustered by all ␣-neurexins and ␤-neurexins tested. One example of induced clustering of GABA A receptor ␥2 is shown for neurexin 2␣ in Fig. 5. Qualitatively, the relative clustering ability of the different constructs for GABA receptor ␥2 appeared to parallel that for gephyrin (data not shown). Unlike GABA A receptor ␥2, which is mainly synaptic, GABA A receptor ␣5 subunit is mainly extrasynaptic, contributing to tonic GABAergic signaling (40,43). In a coculture experiment of neurons with COS cells expressing neurexin 2␣-CFP, immunostaining of sister coverslips revealed obvious induced nonsynaptic clustering of GABA A receptor ␥2 but no detectable clustering of GABA A receptor ␣5 (supplemental Fig. S3). As previously described, endogenous GABA A receptor ␣5 was detected more diffusely along dendrites compared with ␥2, which was clustered at synapses. Some inhomogeneity in ␣5 immunoreactivity was observed along many dendrites, but at sites unrelated to COS cell contacts, and strong clusters were never observed. Thus, the effects of neurexins are selective for clustering GABA A receptor ␥2 but not ␣5 subunit. Because supplemental Fig. S3 was performed with live cell primary antibody incubation, this also shows that the ␣-neurexin-induced clusters of GABA A receptor ␥2 are on the dendrite surface. Dystroglycan is present at only a subset of mature GABA synapses (44,45) but was also reported to bind directly to ␣-neurexins (35). However, in the coculture assay we did not observe clear induced clustering of dystroglycan by any neurexin, ␣ or ␤, on neurons at 2-4 weeks in culture.
Structural Basis for the Differential Synaptogenic Activity of ␣-Neurexins Versus ␤-Neurexins-The C-terminal region of neurexin-1␣ is identical to that of neurexin-1␤ from LNS6 on, and yet we show here that the synapse-promoting activity of neurexin 1␣ is weaker and more specific for GABAergic postsynaptic pro- . Quantitation of induced clustering of gephyrin but not PSD-95 by ␣-neurexins. Quantitation from random transfected COS cells was performed to assess the total integrated intensity of associated PSD-95 or gephyrin clusters that did not overlap with synapsin; values were normalized to the integrated intensity associated with COS cells expressing membrane-targeted CFP in sister cultures (gray lines at 1). This quantitation confirmed significant clustering of gephyrin by all neurexin variants tested but of PSD-95 only for the ␤-neurexin constructs and not the ␣-neurexins (p Ͻ 0.005 (**) and p Ͻ 0.05 (*) compared with mCFP by t test; n ϭ 32-35 cells per construct combined from 2 independent cultures).

␣-Neurexins Induce GABAergic Postsynaptic Differentiation
teins than that of neurexin 1␤ in the coculture assay. Previous work has shown that a construct containing only the region common to neurexin 1␣ and 1␤ (deleting the ␤-specific sequence from neurexin 1␤) had postsynaptic clustering activity equal to that of neurexin 1␤ (6). Thus, the difference in synaptogenic activity between neurexin 1␣ and 1␤ is due to an inhibitory effect of ␣-specific sequences rather than an enhancing effect of ␤-specific sequences. To explore the basis of this differential activity, we made sequential deletions from the mature N terminus of neurexin 1␣, leaving the signal sequence for proper surface expression (Fig. 6). All deletion and mutant neurexin-CFP constructs assayed in this paper appeared to reach the cell surface as efficiently as neurexin 1␣ and neurexin 1␤, as assayed by confocal optical sections through transfected HEK293 cells (supplemental Fig. S4). Furthermore, expression levels did not vary significantly among neurexin-CFP constructs in the COS cell expression studies (average relative CFP intensity values for each neurexin were 111 for 1␣-CFP, 95 for 1␣BC-CFP, 102 for 1␣C-CFP, 103 for 1␣D1176A-CFP, and 89 for 1␤-CFP; n ϭ 15-21, ANOVA p Ͼ 0.1).
Deleting module A containing LNS1, LNS2, and the intervening EGF-like domain had no effect; the ability of this construct termed neurexin 1␣BC to cluster postsynaptic proteins was indistinguishable from that of full-length neurexin 1␣ (Fig. 6, A and B). In contrast, further deleting module B containing LNS3, LNS4, and the intervening EGF-like domain resulted in increased synapse promoting activity. The ability of this construct, termed neurexin 1␣C, to cluster GABAergic postsynaptic proteins gephyrin and neuroligin 2 was intermediate between that of neurexin-1␣ and neurexin-1␤. However, although deletion of modules A and B together significantly increased clustering of gephyrin and neuroligin 2 compared with full-length neurexin 1␣, it did not increase clustering of PSD-95 (Fig. 6). Thus, LNS5 and/or the third EGF-like domain within module C apparently inhibit the ability of neurexin 1 to cluster PSD-95, contributing synapse-type specificity.
Boucard et al. (12) found that LNS6 was the only LNS domain of neurexin 1␣ able to bind neuroligin when tested individually or in modules. We showed previously that a single point mutation to a predicted calcium binding residue D137A of this LNS domain in the context of neurexin 1␤ abolishes binding to neuroligins and abolishes synaptogenic activity (13). We, thus, tested whether this single point mutation in LNS6 in the context of full-length neurexin (i.e. 1␣ D1176A) affects synaptogenic activity. Indeed, neurexin 1␣ D1176A completely lacked the ability to cluster postsynaptic proteins (Fig. 6B). Thus, LNS6 is essential for mediating postsynaptic protein clustering in neurexin 1␣, and the upstream sequences in modules B and C, regardless of additional LNS domains, FIGURE 4. ␣-Neurexins induce clustering of neuroligin 2 but not neuroligin 1/3/4. A, COS cells expressing the neurexin variants tagged intracellularly with CFP (blue) were overlaid on hippocampal neurons pre-grown for 10 -12 days in culture. After 1 day of coculture, cells were fixed and immunolabeled for neuroligin 1/3/4 (with an antibody that recognizes all three) or neuroligin 2 (red) and synapsin (green). Whereas the ␤-neurexin constructs induced clustering of neuroligin 1/3/4 and neuroligin 2 in contacting dendrites (red or pink not associated with green), the ␣-neurexins induced clustering of neuroligin 2 but not neuroligin 1/3/4. Scale bar, 10 m. B, cocultures were scanned, and the COS cells with the strongest apparent associated clusters were imaged and quantitated. The total integrated intensity of associated neuroligin that did not overlap with synapsin was normalized to a value of 100 for neurexin 1␤. Significant differences among constructs were found (p Ͻ 0.001 ANOVA; n ϭ 4 -11 cells per construct).
Expression Patterns of Neurexins-Neurexin expression is brain-specific (39). At the cellular level the six major neurexin forms 1-3 ␣ and ␤ show fairly broad overlapping expression patterns in brain such that most neurons express multiple neurexins (46). How the expression patterns change with development has not been reported for the mammalian central nervous system. We addressed this question here with emphasis on the S4 insert given its importance in regulating the synaptogenic activity of ␤-neurexins (12)(13)(14). By RT-PCR, we could readily detect all six neurexin forms 1-3 ␣ and ␤ from E18 to adult in rat hippocampus and cortex (Fig. 7A). There was a consistent increase in the percentage of all neurexins lacking the S4 insert compared with containing the S4 insert through development in both hippocampus and cortex. The difference in the ratio of splice variants was greatest for neurexin 3 ␣ and ␤, for which the percent lacking the S4 insert changed from Ͻ5% at E18 to 35-50% at P11 (Fig. 7B). Neurons from E18 rat grown in culture for 7-22 DIV showed a similar trend with the percent of neurexins lacking the S4 insert increasing with development (Fig. 8).
To assess potential activity regulation, we grew hippocampal neurons in culture under control conditions or chronically in the presence of the NMDA receptor antagonist APV. Semiquantitative RT-PCR in comparison to GAPDH revealed no significant difference in the total level of each neurexin mRNA (ϪS4 and ϩS4 forms combined) due to the presence or absence of APV (data not shown). However, blockade significantly increased the percentage of neurexin 2␤ lacking the S4 insert (from ϳ20 to ϳ40%) while reducing the percentage of neurexins 3␣ and 3␤ lacking the S4 insert (Fig. 8B). Thus, NMDA receptor activity differentially regulates splicing of neurexins at the S4 splice site.

DISCUSSION
We show here that ␣-neurexins promote GABAergic but not glutamatergic postsynaptic specialization. Specifically, ␣-neurexins expressed on COS cells cluster gephyrin and GABA A receptor but have no detectable effect on the distribution of PSD-95 in contacting dendrites of cultured hippocampal neurons (Figs. 1-3 and 5). This specificity presumably arises at least in part from the specific clustering of neuroligin 2 but not neuroligins 1/3/4 on the contacting dendrites by ␣-neurexins (Fig.  4). The postsynaptic clustering activity of ␣-neurexins is dependent on LNS6, being abolished by a single point mutation in the predicted calcium binding residue D1176A (Fig. 6). The central region of ␣-neurexin containing LNS 3-5 and two EGFlike domains inhibits the synaptogenic activity of LNS6 in the context of full-length ␣-neurexin (Fig. 5). Finally, we report that splicing of all neurexins at the S4 site is developmentally regulated, with levels of neurexins lacking the S4 insert low just before birth and increasing during the first two postnatal weeks in rat cortex and hippocampus (Fig. 7). Major differences between neurexins 1, 2, and 3 were not observed in these assays except for activity regulation of splicing at the S4 site, with differential regulation in 2␤ compared with 3␣,␤ (Fig. 8).
A major conclusion from this study is that ␣-neurexins are completely specific for promoting GABA but not glutamate postsynaptic differentiation in coculture assays. ␤-Neurexins containing the S4 insert were also previously shown to be more active for promoting GABA than glutamate postsynaptic differentiation in coculture assays (13,14). Chih et al. (14) did not observe induction of PSD-95 clusters by either neurexin 1␤ containing the S4 insert or neurexin 1␣ lacking the S4 insert and suggested that ␣-neurexins have a similar GABAergic specificity for synaptogenic activity as ␤-neurexins containing the S4 insert. However, we show here and in Graf et al. (13) that neurexin 1␤ containing the S4 insert does cluster PSD-95 significantly above the background level assessed by mCFP expression in COS cells. In the same experiment (Figs. 1-3) where we observe significant clustering of PSD-95 by all ␤-type neurexin constructs (all ϩS4 variants), we find no significant clustering of PSD-95 by ␣-neurexins. It is possible that we have observed these more subtle distinctions due to differences in experimental conditions from Chih et al. (14) including differences in construct expression level or neuron culture style. In experiments where we tried to bias toward seeing any effect of ␣-neurexins on distribution of glutamatergic postsynaptic proteins, assaying the more strongly clustered neuroligins rather than the scaffolding proteins, we could still find no effect of ␣-neurexins on distribution of neuroligin 1/3/4 (Fig. 4, noting that the major variant recognized by the common antibody in this experiment is thought to be neuroligin 1 ϩB as detailed under "Results"). In the same experiment, the ␣-neurexins could induce robust clustering of neuroligin 2 (Fig. 4). Previous evidence shows that neuroligin 2 is normally clustered only at GABA synapses (9, 13) and neuroligin 1 primarily at glutamate synapses (8). Subcellular localization of endogenous neuroligins 3 and 4 has not been reported, but the distribution of yellow fluorescent protein-tagged forms of neuroligins 3 and 4 is similar to that of neuroligin 1 and not 2 (6). We conclude here that ␣-neurexins induce clustering of the GABAergic proteins gephyrin, GABA A receptor and neuroligin 2, but not of the glutamatergic protein PSD-95 or neuroligin 1/3/4. These differences in ability to cluster specific postsynaptic components agree well with the binding selectivity of neurexins to neuroligin splice variants as reported recently (12). The majority of neuroligin 1 in the brain is present in the form with the B insert (14). ␤-Neurexins with the S4 insert have lower binding affinity for neuroligin 1 with the B insert than without the B insert (12)(13)(14). However, this difference is not an absolute effect, since in the most sensitive binding assay incubating soluble neurexin forms to neuroligins expressed on the surface of HEK293 or COS cells, ␤-neurexins with the S4 insert still bound specifically to cells expressing neuroligin 1 with the B insert compared with untransfected control cells or compared with binding of an LNS deletion variant of neurexin (12,13). In contrast, in the same experiment binding of ␣-neurexins with or without the S4 insert to neuroligin 1 with the B insert was not detected (12). In short, all ␤ neurexins bind neuroligin 1 ϩB but ␣ neurexins do not. Thus, we conclude that the binding to neuroligin 1 with the B insert and ability to cluster glutamatergic postsynaptic proteins in coculture occurs in parallel. Furthermore, the addition of the S4 insert to ␤-neurexins reduces but does not abolish these activities, whereas the presence of the additional sequences in the longer ␣-neurexins abolishes these activities.
Regulation of neurexin splicing has not yet been rigorously studied, although differential regulation can occur among mammalian brain regions (46) and during development in other systems (47,48). Specifically with respect to the S4 site, differences in ratios of ϩS4 compared with ϪS4 variants of selected ␣or combined ␣plus ␤-neurexins have been reported among adult rat brain regions (11) and during devel-FIGURE 6. Postsynaptic protein clustering by neurexin-1␣ requires LNS6 and is negatively modulated by upstream sequences. A, neurexin 1␤, 1␣, and derivatives of 1␣ mutated in a predicted calcium binding residue in LNS6 (1␣D1176A), lacking module A (1␣BC), or lacking modules A and B (1␣C) were tagged intracellularly with CFP (blue) and tested for postsynaptic protein clustering activity in the COS cell-neuron coculture assay. Activity was considered positive if the contacting dendrites exhibited any clusters of PSD-95, gephyrin, or neuroligin 2 (red) lacking synapsin (green). Neurexin 1␣D1176A had no detectable clustering activity. For PSD-95, all neurexin-1␣ derivatives also had no detectable clustering activity; PSD-95 clustering was only observed in response to neurexin 1␤ (compare these images with Figs. 1, 2, and 4). For gephyrin and neuroligin 2, the clustering activity of neurexin 1␣BC was indistinguishable from that of 1␣. In contrast, the clustering activity of neurexin 1␣C was higher than that of 1␣ and intermediate between that of 1␣ and 1␤. Scale bar, 10 m. B, quantitation of the percentage of expressing COS cells exhibiting non-synaptic clusters in contacting dendrites confirmed these differences in relative synaptogenic activity (p Ͻ 0.001 by ANOVA; *, p Ͻ 0.001 by t test compared with neurexin 1␣; n ϭ 4 cultures with 10 -78 cells scored each). PDZBD, PDZ binding domain; TM, transmembrane domain. CH, glycosylated region.

␣-Neurexins Induce GABAergic Postsynaptic Differentiation
opment and neurotrophin exposure in chick sympathetic neurons (47). Here we show that the percentage of each ␣ and ␤ neurexin lacking the S4 insert increases between E18 and P11 in rat cortex and hippocampus. This developmental increase in the more glutamate-selective ϪS4 ␤ variants correlates with the developmental increase in glutamatergic synaptogenesis. The earlier expression of the more GABA-selective ϩS4 ␤ variants correlates with the earlier onset of GABAergic synaptogenesis (50,51). Neurexin 3 ␣ and ␤, which showed the greatest developmental regulation, showed a similar increase in ϪS4 variants with development of E18 hippocampal neurons in dissociated culture from DIV 7 to 14 (Fig. 8). Interestingly, in the culture system where we could manipulate activity, we found that NMDA receptor blockade differentially regulates splicing, decreasing ϪS4 variants of 3␣,␤ and increasing ϪS4 variants of 2␤. The significance of this differential activity regulation is not yet clear but may be important in combination with regulation of splicing at site 5, which is predicted to generate secreted forms of neurexin 3 but not 1 or 2 (52). No obvious developmental changes were observed for overall expression levels of ␣-neurexins, perhaps reflecting more their calcium channel coupling function at many synapse types rather than their additional GABA-specific roles.
Analysis here of the single point mutation D1176A in the predicted calcium binding residue of LNS6 indicates that LNS6 is necessary for GABAergic synaptogenic activity of neurexin-1␣; no other LNS domains are active (Fig. 6). Again, this coculture result is in good agreement with the binding data of Boucard et al. (12). These authors found that LNS6 and a short upstream region is the minimal sequence necessary and sufficient from ␣-neurexin for binding neuroligin lacking the B insert. The combination of this published binding data and our coculture studies strongly suggest that ␣-neurexins cluster gephyrin and GABA A receptor through clustering neuroligin 2 on dendrites via binding of neuroligin 2 to LNS6. Although the ␣-neurexin-specific region upstream of LNS6 presumably interacts with other proteins and functions in coupling presynaptic calcium channels to release (28), our deletion analyses show that this region of ␣-neurexin inhibits the synaptogenic activity of the C-terminal portion containing LNS6 (Fig.  6). The inhibitory activity did not reside in module A but in modules B and C. These upstream regions may either participate in an intramolecular interaction inhibiting the activity of LNS6 or may bind an additional inhibitory factor.
More generally, in the coculture assays all neurexins induce clustering of GABAergic proteins to the same or a greater extent than that of glutamatergic proteins (e.g. comparing the increase in integrated intensity relative to the background on mCFP cells (Fig. 3) or comparing total area of induced clusters since non-normalized intensity would not be directly comparable (13,14)). This equal or greater activity for promoting GABA postsynaptic development applies to all neurexins tested to date, 1-3, ␣ and ␤, and containing or lacking the S4 insert. Although all splice forms of ␣-neurexins have not been tested, given that LNS6 of the ␣-neurexins bears the synaptogenic activity, it is highly unlikely that any forms not yet tested would have stronger activity for inducing glutamatergic than GABAergic postsynaptic protein clustering. These data are consistent with the idea that the neurexin-neuroligin system may be more important for the development and maintenance of GABA synapses than glutamate synapses. Support for this idea comes from the finding that inhibitory synaptic transmission is more severely impaired than excitatory synaptic transmission in P0 brainstem of the neuroligin 1/2/3 triple knockout mice (20). Spontaneous glutamatergic transmission frequency was reduced 4-fold and a evoked failure rate unaffected, whereas spontaneous GABAergic/glycinergic transmission frequency was reduced 10-fold and evoked a failure rate markedly increased. At the same time, survival of the individual neuroligin knockouts but neonatal death of the triple knockout (20) indicates some redundancy in function rather than an absolute requirement for neuroligin 2 function at GABA synapses. In addition to the calcium channel coupling phenotype, ␣-neurexin 1/2/3 triple knock-out mice had a 2-fold reduction in the density of type II symmetric GABAergic synapses in neonatal brainstem with no change in density of type I asymmetric glutamatergic synapses (27). A 30% reduction in type II symmetric synapses but not type I asymmetric synapses was also found in adult neocortex of neurexin 1/2␣ and 2/3␣ double knock-out mice (53). It was not clear whether this loss of inhibitory synapses reflected a direct function of ␣-neurexin in GABA synapse development or an indirect effect of reduced synaptic activity due to the presynaptic calcium FIGURE 8. Alternative splicing at the S4 site is regulated by NMDA receptor activity in hippocampal cultures. Neurons from E18 rat hippocampus were cultured in the chronic presence (ϩ) or absence (Ϫ) of the NMDA receptor antagonist APV. Cultures were harvested at the indicated ages, and RT-PCR was performed to assess the relative levels of neurexin variants containing or lacking inserts at splice site 4. A, representative agarose gel electrophoresis of RT-PCR products. G, GAPDH loading control. DIV, days in vitro. B, quantitative analysis revealed a significant increase in the percentage of neurexin 2␤ and decrease in the percentage of neurexin 3␣ and 3␤ lacking the S4 insert with chronic NMDA receptor blockade. Statistical analyses represent differences between untreated cells and age-matched APV-treated cells as determined by ANOVA with Tukey's post hoc test; *, p Ͻ 0.05; †, p Ͻ 0.005; n ϭ 6. Semiquantitative RT-PCR in comparison with GAPDH did not reveal any significant differences in total expression level of neurexin 1␣, 1␤, 2␣, 2␤, 3␣, or 3␤ (ϩ and Ϫ S4 forms combined) between control cells or cells chronically treated with APV (data not shown).
channel coupling defect. Based on the coculture results presented here, we suggest that the loss of inhibitory synapses in these mice may reflect a direct function of ␣-neurexins in linking GABAergic presynaptic and postsynaptic components. The apparent requirement for ␣-neurexin LNS6 to bind neuroligins for coculture activity (D1176A in Fig. 6 combined with Refs. 12 and 13) suggests that the function of ␣-neurexins in GABA synapse formation and/or stabilization occurs via binding neuroligins. Although we could not observe clustering of dystroglycan by neurexins in the coculture assay, and GABA synapses can form in the absence of dystroglycan (45,54), the association of ␣-neurexins with dystroglycan (35) may also contribute to stabilization of a postsynaptic complex at some GABA synapses (44). Members of the SynCAM, ephrin, Eph receptor, and netrin G ligand families are reported to trigger presynaptic and/or postsynaptic differentiation specifically for glutamate and not GABA synapses (55)(56)(57)(58). The stronger functional association of the neurexin-neuroligin complex with GABA compared with glutamate synapses in coculture and in the knock-out mice may reflect a greater redundancy of synaptic organizing molecules at glutamate synapses. In Drosophila, which lack ␤ neurexins and have less redundancy in general, null mutations in the single ␣ neurexin gene lead to a reduction in numbers of central and neuromuscular synapses, altered ultrastructure, and defects in synaptic transmission and associative learning (59,60).