Presynaptic α2δ-2 Calcium Channel Subunits Regulate Postsynaptic GABAA Receptor Abundance and Axonal Wiring

Presynaptic α2δ subunits of voltage-gated calcium channels regulate channel abundance and are involved in glutamatergic synapse formation. However, little is known about the specific functions of the individual α2δ isoforms and their role in GABAergic synapses. Using primary neuronal cultures of embryonic mice of both sexes, we here report that presynaptic overexpression of α2δ-2 in GABAergic synapses strongly increases clustering of postsynaptic GABAARs. Strikingly, presynaptic α2δ-2 exerts the same effect in glutamatergic synapses, leading to a mismatched localization of GABAARs. This mismatching is caused by an aberrant wiring of glutamatergic presynaptic boutons with GABAergic postsynaptic positions. The trans-synaptic effect of α2δ-2 is independent of the prototypical cell-adhesion molecules α-neurexins (α-Nrxns); however, α-Nrxns together with α2δ-2 can modulate postsynaptic GABAAR abundance. Finally, exclusion of the alternatively spliced exon 23 of α2δ-2 is essential for the trans-synaptic mechanism. The novel function of α2δ-2 identified here may explain how abnormal α2δ subunit expression can cause excitatory–inhibitory imbalance often associated with neuropsychiatric disorders. SIGNIFICANCE STATEMENT Voltage-gated calcium channels regulate important neuronal functions such as synaptic transmission. α2δ subunits modulate calcium channels and are emerging as regulators of brain connectivity. However, little is known about how individual α2δ subunits contribute to synapse specificity. Here, we show that presynaptic expression of a single α2δ variant can modulate synaptic connectivity and the localization of inhibitory postsynaptic receptors. Our findings provide basic insights into the development of specific synaptic connections between nerve cells and contribute to our understanding of normal nerve cell functions. Furthermore, the identified mechanism may explain how an altered expression of calcium channel subunits can result in aberrant neuronal wiring often associated with neuropsychiatric disorders such as autism or schizophrenia.

Despite the increasing knowledge, it is still largely elusive how the individual ␣ 2 ␦ isoforms contribute to the formation and function of central synapses. The fact that ␣ 2 ␦ subunits constitute large extracellular and highly glycosylated proteins (Davies et al., 2007;Bauer et al., 2010) puts them in an ideal position for extracellular functions, as previously suggested (Fell et al., 2016;Wang et al., 2017). Indeed, a recent study provided evidence for an interaction of ␣ 2 ␦-1 with NMDARs in dorsal horn neurons (Chen et al., 2018) and presynaptic ␣-neurexins (␣-Nrxns) are potential candidates for mediating synaptic functions of ␣ 2 ␦ subunits (Tong et al., 2017;Brockhaus et al., 2018). However, the role of ␣ 2 ␦ subunits has until now only been addressed in excitatory glutamatergic synapses. This is surprising because ␣ 2 ␦ subunits are strongly expressed in GABAergic neurons (Cole et al., 2005;Schlick et al., 2010) and aberrant GABAergic signaling is primarily implicated in the etiology of the abovementioned neurological disorders (e.g., epilepsy; Bonansco and Fuenzalida, 2016). Finally, the brain robustly expresses ␣ 2 ␦-1, ␣ 2 ␦-2, and ␣ 2 ␦-3 ; however, whether and how the individual isoforms contribute to specific functions in neurons simultaneously expressing all isoforms is elusive. Therefore, to address ␣ 2 ␦ subunit-mediated synaptic functions, we analyzed the consequences of presynaptic ␣ 2 ␦ subunit expression in glutamatergic and GABAergic synapses. We show that presynaptic overexpression of ␣ 2 ␦-2 specifically induces the formation of mismatched glutamatergic synapses by transsynaptically recruiting postsynaptic GABA A Rs. Moreover, ␣ 2 ␦-2 recruits postsynaptic GABA A Rs also in GABAergic synapses and thus independently of the presynaptic neurotransmitter identity. Most importantly, the mismatched synapse formation is explained by an altered wiring of glutamatergic axons to GABAergic postsynaptic positions. Interestingly, the ␣ 2 ␦-2-induced GABA A R clustering is further upregulated in glutamatergic neurons lacking all three ␣-Nrxns. Therefore, ␣-Nrxns can modulate the presynaptic effect of ␣ 2 ␦-2; however, they are not required for the trans-synaptic role of ␣ 2 ␦-2 in recruiting GABA A Rs. Together, our findings prove that presynaptic ␣ 2 ␦-2 acts trans-synaptically on postsynaptic GABA A R. The fact that increased ␣ 2 ␦-2 expression triggers aberrant axonal wiring is particularly interesting in light of neurological disorders associated with axonal wiring defects and altered excitatory-inhibitory balance (Huang and Hsueh, 2015;Lee et al., 2017).

Materials and Methods
Breeding and genotyping procedures Animals. Animal procedures for WT BALB/c and ␣ 2 ␦ mutant mice were performed at the Medical University Innsbruck in compliance with government regulations and approved by the Austrian Federal Ministry of Science, Research and Economy (license numbers BMWFW-66.011/ 0113-WF/V/3b/2014 and BMWFW-66.011/0114-WF/V/3b/2014). Regular reports including the mouse numbers used for this project were given to the Austrian Federal Ministry of Science, Research and Economy (bmwfw). Animal experiments at the University of Münster [␣-Nrxn triple knock-out (TKO) and control mice] were performed in accordance with government regulations for animal welfare and approved by the Landesamt für Natur, Umwelt und Verbraucherschutz (LANUV, NRW, Germany, license numbers 84-02.05.20.11.209 and 84-02. 04.2015.A423). Mice were maintained at the central animal facilities in Innsbruck and Münster under standard housing conditions with food and water available ad libitum on a 12 h light/dark cycle.

Cell culture and transfection procedures
Lentiviral production. Lentiviruses were produced by transient transfection of confluent tsa201 cells with the lentiviral expression vectors containing pHR-␤A-eGFP, pHR-␤A-mcherry, or pHR-␤A-eGFP*␣ 2 ␦-2 in combination with psPAX2 ( packaging plasmid) and the pVSV (envelope plamid) using Metafectene (Biontex Laboratories). The following day, medium was changed to neuronal plating medium (NPM; consisting of MEM, 10% horse serum, 0.2% glucose, and 1 mM sodium pyruvate) and, after 24 and 48 h, supernatants containing the viruses were harvested, sterile filtered (0.20 m), aliquoted, and stored at Ϫ20°C. Cultured hippocampal and striatal neurons were infected immediately after plating with the lentiviral medium supernatant diluted 1:4 in NPM and incubated for 4 h in a humidified incubator (95% air and 5% CO 2 ) at 37°C. More details for infection procedures are given in the individual sections below.
Primary cultures of hippocampal neurons for fluorescence imaging. Lowdensity cultures of hippocampal neurons were obtained from 16.5-to 18-d-old embryonic BALB/c mice of either sex as described previously (Obermair et al., 2004;Kaech and Banker, 2006;Stanika et al., 2016;Folci et al., 2018). In brief, hippocampi were dissected in cold Hank's balanced salt solution (HBSS) following dissociation by 2.5% trypsin-EDTA treatment and trituration. Dissociated neurons were plated at a density of ϳ3500 or 7000 cells/cm 2 on 18 mm glass coverslips (#1.5; GML) coated with poly-L-lysine (Sigma-Aldrich) in 60 mm culture dishes. After allowing the neurons to attach for 3-4 h, coverslips were transferred neuronside down into a 60 mm culture dish containing a glial feeder layer. Maintenance of neurons and glia was done in serum-free neurobasal medium supplemented with Glutamax and B-27 (NBKO, all ingredients from Thermo Fisher Scientific). Three days after plating, Ara-C (5 M) was added to stop glial proliferation and 1/3 of the medium was replaced weekly with fresh maintenance medium. Plasmids were introduced into neurons at 6 days in vitro (DIV) with Lipofectamine 2000-mediated transfection (Thermo Fisher Scientific) as described previously (Obermair et al., 2004). For cotransfection of p␤A-eGFP plus p␤A-␣ 2 ␦ or mCherry plus p␤A-␣ 2 ␦, 1.5 g of total DNA was used at a molar ratio of 0.7:1. Control neurons from the same culture preparations were transfected with 1-2 g of p␤A-eGFP or mCherry, respectively. Labeling of cells for superresolution gated stimulated emission depletion (gSTED) microscopy (see Fig.  11A-C) was done by introducing mCherry or mCherry plus ␣ 2 ␦-2 at 6 DIV and p␤A-eGFP at 7 DIV with lipofection. Cells were processed for immunostaining experiments between 20 and 30 DIV.
For ␣-Nrxn TKO, WT and mutant mice of either sex were used for neuronal cultures derived from timed pregnant dams at embryonic day 17 (E17). Dissociated primary neurons were prepared in HBSS from single hippocampi as described previously (Neupert et al., 2015). Briefly, cell suspensions obtained after 0.25% trypsin treatment and trituration were plated onto 18 mm glass coverslips (Menzel-Glaeser) coated with poly-L-lysine (Sigma-Aldrich) at a density of 55,000 cells/coverslip. After 4 h at 37°C in plating medium (MEM, 10% horse serum, 0.6% glucose, and 1 mM sodium pyruvate), coverslips were inverted onto a 70 -80% confluent monolayer of astrocytes grown in 12-well plates (Falcon) and incubated in Neurobasal medium supplemented with B27, 0.5 mM glutamine, and 12.5 M glutamate. After 3 d, media were refreshed with Neurobasal medium supplemented with B27, 0.5 mM glutamine, and 5 M AraC. Cultures were maintained at 37°C in a humidified incubator with an atmosphere of 95% air and 5% CO 2 . Neurons were transfected at 14 DIV using lipofectamine (Thermo Fisher Scientific) and experiments were performed between 20 and 25 DIV.
Primary cocultures of WT cortical and striatal neurons. Because the inclusion of glutamatergic neurons is required for the proper development of GABAergic medium spiny neurons (MSNs) in culture (Segal et al., 2003), we modified the protocol for hippocampal neurons and adapted the striatal-cortical coculture published previously (Penrod et al., 2011). Cocultures of cortical neurons and striatal MSNs were prepared from 16.5-to 18-d-old embryonic BALB/c mice of either sex as summarized in Figure 7. Briefly, fetuses were removed from the uterus, decapitated, and brains were dissected in cold HBSS. After separating the cerebral hemispheres and stripping away the meninges, hemispheres were placed medial surface up, showing the hippocampus as a clearly visible structure. A region of the prefrontal cortex was dissected, minced, and transferred to a 15 ml Falcon tube containing HBSS (see Fig. 7A). The remaining cortical tissue was peeled along the line of the hippocampus to reveal the striatum, which was scooped out using small curved scissors and transferred to a separate tube. Striatal and cortical tissue of at least 2 hemispheres was separately collected and dissociated by 2.5% trypsin-EDTA treatment and trituration as described for hippocampal neurons (see above). Plasmid DNA-lipid complexes were prepared according to the Lipofectamine 2000-mediated transfection protocol (Thermo Fisher Scientific) and ϳ2.4 ϫ 10 5 striatal neurons were transfected for 20 min in a 37°C water bath keeping the total volume to 1 ml with NBKO. Subsequently, the cell suspension was directly seeded on poly-L-lysine-coated glass coverslips within a 60 mm culture dish containing 4 ml of prewarmed NPM and striatal neurons were allowed to attach at 37°C. For the entire transfection procedure, dissociated cortical neurons were maintained in HBSS in a 15 ml tube at 37°C and occasionally swirled. After 2 h, transfection of striatal neurons was stopped by replacing the transfection-plating solution with 5 ml of fresh, prewarmed NPM and untransfected cortical neurons were seeded onto striatal neurons in a ratio of 2 (cortical neurons) to 3 (MSNs) at a total density of ϳ14,000 cells/cm 2 . For the immunostaining experiments and electrophysiological recordings shown in Figure 7, striatal neurons were plated at a density of 4000 or 7000 cells/cm 2 , respectively, and infected with a lentiviral pHR-␤A-eGFP construct. Following attachment of MSNs and viral infection for 4 h, untransfected cortical neurons were plated at a density of 2800 cells/cm 2 . Cortical cells were allowed to attach for 3-4 h and coverslips were transferred neuron-side down into a 60 mm culture dish containing a glial feeder layer. In cocultures for superresolution gSTED microscopy (see Fig. 11E), striatal and cortical neurons were labeled with eGFP and mCherry, respectively. For this purpose, striatal neurons were plated at a density of 7000 cells/cm 2 on poly-L-lysine-coated glass coverslips in a 60 mm culture dish containing 4 ml of prewarmed NPM. MSNs were virally infected using a lentiviral pHR-␤A-eGFP construct and allowed to attach for 3-4 h. In the meantime, ϳ2.0 ϫ 10 5 cortical neurons were transfected at 37°C in a water bath, keeping the total volume to 1 ml with NBKO. To this end, mCherry or mCherry plus ␣ 2 ␦-2 were introduced using the Lipofectamine 2000-mediated transfection protocol (Thermo Fisher Scientific). Total DNA amounts and molar ratios were the same as described above for hippocampal neurons. Infection of striatal neurons was stopped by replacing the virus-plating solution with 4 ml of fresh, prewarmed NPM. The cell suspension containing the cortical neurons and transfection solution was centrifuged for 5 min at 1000 rpm at 4°C. Finally, cortical neurons were resuspended in 1 ml of NBKO, seeded onto striatal neurons, and allowed to attach for 3-4 h before coverslips were transferred neuron-side down into a 60 mm culture dish containing a glial feeder layer. Ara-C treatment and maintenance of neurons and glia were as described above. Cells were processed for immunostaining and patch-clamp experiments at 21-27 and 13-14 DIV, respectively.
Neuronal cultures for paired recordings. Culture protocols were modified to enable electrophysiological recordings from pairs of isolated neurons. Poly-L-lysine solution (Sigma-Aldrich) was sprayed on glass coverslips, producing uniformly distributed single dots of neuronal substrate. Hippocampal neurons were seeded at a density of 880 cells/cm 2 and infected with pHR-␤A-eGFP*␣ 2 ␦-2. Cells were processed for electrophysiological experiments at 13-18 DIV.
To compare the relative expression of distinct ␣ 2 ␦ subunits in different preparations, data were normalized to the preparation with the highest Hprt1 expression.
Expression vectors and cloning procedures. To facilitate neuronal expression, all constructs were cloned into a eukaryotic expression plasmid containing a neuronal chicken ␤-actin promoter, p␤A. Cloning of all constructs was confirmed by sequencing (Eurofins Genomics).
For p␤A-␣ 2 ␦-1: Mouse ␣ 2 ␦-1 was cloned from genomic cDNA derived from mouse cerebellum. Primer sequences were selected according to GenBank NM-001110844. Briefly, the cDNA of ␣ 2 ␦-1 was amplified by PCR in three fragments. The forward primer used for amplifying fragment 1 introduced a NotI site and the Kozak sequence (CCTACC) upstream of the starting codon and the reverse primer used for amplifying fragment 3 introduced a KpnI and a SalI site after the stop codon. Fragment 2 (nt 1442-2564) was MfeI/BamHI digested and fragment 3 (nt 2335-3276) was KpnI/BamHI digested and co-ligated in the corresponding MfeI/KpnI sites of the p␤A vector, yielding an intermediate construct. Fragment 1 (nt 1-1575) was NotI/MfeI digested and coligated with the SalI/MfeI-digested intermediate construct containing fragments 2 and 3 and the NotI/SalI-digested p␤A vector, yielding p␤A-␣ 2 ␦-1 (GenBank accession number MK327276).
For p␤A-2HA-␣ 2 ␦-1: The putative signal peptide (aa1-24) was predicted using Signal P (SignalP 4.0: discriminating signal peptides from transmembrane regions (Petersen et al., 2011). A double hemagglutinin tag (2HA) followed by a TEV cleavage site was introduced between the third and fourth amino acids after the predicted signal peptide cleavage site of mouse ␣ 2 ␦-1; that is, residue F27. Introduction of this sequence did not alter the predicted cleavage site. Briefly, the cDNA sequence of ␣ 2 ␦-1 (nt 1-516) was PCR amplified with overlapping primers introducing the double HA tag and the TEV cleavage site in separate PCRs using p␤A-␣ 2 ␦-1 as a template. The two separate PCR products were then used as templates for a final PCR with flanking primers to connect the nucleotide sequences. The resulting fragment was then NotI/BglII digested and ligated into the corresponding sites of p␤A-␣ 2 ␦-1, yielding p␤A-2HA-␣ 2 ␦-1.
For p␤A-2HA-␣ 2 ␦-2 (v1): A putative signal peptide was not reliably predicted using Signal P (SignalP 4.0: discriminating signal peptides from transmembrane regions; Petersen et al., 2011); however, the highest prediction showed that the signal peptide comprises residues 1-64. The 2HA tag followed by a thrombin cleavage site was therefore introduced after the predicted signal peptide cleavage site of mouse ␣ 2 ␦-2; that is, residue A64. Introduction of this sequence did not alter the predicted cleavage site. Briefly the cDNA sequence of ␣ 2 ␦-2 (nt 1-761) was PCR amplified with overlapping primers introducing the double HA tag and the thrombin cleavage site in separate PCRs using p␤A-␣ 2 ␦-2 as a template. The two separate PCR products were then used as templates for a final PCR with flanking primers to connect the nucleotide sequences. The resulting fragment was then HindIII/AflII digested and ligated into ␣ 2 ␦-2 (v1).
For p␤A-2HA-␣ 2 ␦-3: The putative signal peptide (aa 1-28) was predicted using Signal P (SignalP 4.0: discriminating signal peptides from transmembrane regions; Petersen et al., 2011). The 2HA tag followed by a thrombin cleavage site was therefore introduced after the predicted signal peptide cleavage site of mouse ␣ 2 ␦-3; that is, residue D28. Introduction of this sequence did not alter the predicted cleavage site. Briefly, the cDNA sequence of ␣ 2 ␦-3 (nt 1-653) was PCR amplified with overlapping primers introducing the double HA tag and the thrombin cleavage site in separate PCRs using p␤A-␣ 2 ␦-3 as a template. The two separate PCR products were then used as templates for a final PCR with flanking primers to connect the nucleotide sequences. The resulting fragment was then NotI/BsrGI digested and ligated into the corresponding sites of p␤A-␣ 2 ␦-3, yielding p␤A-2HA-␣ 2 ␦-3.
For p␤A-eGFP*2HA-␣ 2 ␦-2 (v1): The eGFP coding sequence, followed by a stop codon and the Kozak sequence (CCTACC), were introduced by PCR between the Kozak sequence and the coding sequence of 2HA-␣ 2 ␦-2. The resulting fragment was then HindIII/PpuMI digested and ligated into the corresponding sites of p␤A-2HA-␣ 2 ␦-2, yielding p␤A-eGFP*2HA-␣ 2 ␦-2. This construct was then transfected in hippocampal cultures with lipofectamine to insure that both the eGFP and HA signals were present in transfected neurons.
For pHR-p␤A-eGFP: The cloning procedure to generate this plasmid was described previously (Subramanyam et al., 2009).

Immunocytochemistry and microscopy
Antibodies. Information on primary and secondary antibodies is summarized in Table 1.
High-resolution fluorescence microscopy. Immunolabeling of permeabilized or live-cell-stained neurons was performed as described previously (Obermair et al., 2004Stanika et al., 2016;Folci et al., 2018) and information on antibodies is given in Table 1. For PSD-95 labeling of eGFP-transfected neurons, cultures were fixed in 4% paraformaldehyde/4% sucrose (pF) in PBS for 20 min at room temperature, rinsed in PBS followed by a second fixation with absolute methanol for 10 min at Ϫ20°C. For all other antibody combinations, neurons were fixed with pF only. Subsequent to fixation and washing, cultures were incubated for 30 min in 5% normal goat serum in PBS containing 0.2% bovine serum albumin (BSA) and 0.2% Triton X-100 (PBS/BSA/Triton), enabling membrane permeabilization. Primary antibodies were applied overnight in PBS/BSA/Triton at 4°C and detected by fluorochrome-conjugated secondary antibodies incubated for 1 h at room temperature. For surface staining of HA-tagged ␣ 2 ␦ constructs, transfected neurons were incubated with rat-anti-HA antibody for 20 min at 37°C following rinsing in warm HBSS and fixation with pF for 10 min at room temperature. Subsequent washing and blocking steps were done with PBS and PBS/BSA, respectively, followed by 1 h incubation with fluorochrome-conjugated secondary goat-anti-rat Alexa Fluor 594 antibody. After rinsing and postfixation of cells in pF for 5 min, neurons were permeabilized by blocking with PBS/BSA/Triton. Primary mouse-anti-synapsin antibody was applied overnight at 4°C and detected with goat-anti-mouse Alexa Fluor 350 antibody. Coverslips were either mounted in p-phenylenediamine glycerol (Flucher et al., 1993) or DAKO fluorescence mounting medium (Agilent Technologies) for ␣-Nrxn knock-out and control cultures. Hippocampal cultures were mainly viewed with an Axio Imager microscope (Carl Zeiss) using a 63ϫ 1.4 numerical aperture (NA) oil-immersion objective lens and 14-bit grayscale images were recorded with a cooled CCD camera (SPOT; Diagnostic Instruments) using MetaMorph soft-ware (Molecular Devices). ␣-Nrxn knock-out cultures and MSNs were observed with a BX53 microscope (Olympus) using a 60ϫ 1.42 NA oilimmersion objective lens and 14-bit grayscale images were recorded with a cooled CCD camera (XM10; Olympus) using cellSens Dimension software (Olympus). To analyze presynaptic and postsynaptic differentiation cover glasses were systematically screened for contact sites between axons of presynaptic neurons [transfected with the fluorescent marker alone (eGFP or mCherry only) or cotransfected with the marker and target construct (eGFP or mCherry plus ␣ 2 ␦)] and dendrites or somas of nontransfected postsynaptic neurons. Images of randomly selected well differentiated cells were acquired with the same exposure and gain settings for all conditions within an experiment. These settings were initially adjusted to facilitate optimal visualization of peripheral cell compartments (axons and presynaptic boutons) also in weakly expressing neurons. Therefore, overly saturated neurons (based on eGFP and mCherry levels) were excluded from analysis and only cells with medium to low eGFP or mCherry expression were selected for further analysis. Images were analyzed with MetaMorph software (Molecular Devices) as described below. Figures were assembled in Adobe Photoshop CS6 and linear adjustments were done to correct black level and contrast.
To study the localization of excitatory synapses on postsynaptic dendrites of excitatory neurons (see Fig. 11B), hippocampal neurons labeled with mCherry and eGFP, respectively, were fixed in 4% pF in PBS for 20 min at room temperature, washed in PBS, and mounted in DABCO/glycerol. Fluorophores were recorded in the following sequence (excitation/detection range wavelength in nm): mCherry (presynaptic axon: mCherry only or mCherry ϩ ␣ 2 ␦-2; 584/600 -688; gate 0.20 -6 ns in confocal mode) and eGFP (postsynaptic dendrite: eGFP only, 490/508 -584 or 520 -584; gate 1-6.2 ns, STED laser at 592 nm). To enable clear visualization of dendritic structures, STED laser intensity was adjusted according to the eGFP signal. To study the localization of excitatory synapses on postsynaptic dendrites of inhibitory GABAergic neurons (see Fig. 11E), cortical neurons and MSNs were labeled with mCherry and eGFP, respectively, as described above. Primary mouse-anti-gephyrin (1:1000) was applied overnight in PBS/BSA/Triton at 4°C and detected with secondary goat-anti-mouse STAR440 (1:500). Fluorophores were recorded in the following sequence (excitation/detection range wavelength in nm): mCherry (presynaptic axon: mCherry only or mCherry ϩ ␣ 2 ␦-2; 580/ 600 -690; gate 0.5-6 ns in confocal mode), eGFP (postsynaptic dendrite: eGFP only, 470/508 -580; gate 1-6.0 ns, STED laser at 592 nm), and STAR440 (mouse-anti-gephyrin; 470/480 -570; gate 1-6.0 ns, STED laser at 592 nm). STED laser intensity was adjusted according to the eGFP signal to enable clear visualization of dendritic structures and to reduce cross-activation during the detection of STAR440. Channel dye separation was performed to further split the eGFP and STAR440 signals. Images were recorded with LAS AF software (Leica Microsystems), analyzed using MetaMorph software (Molecular Devices), and figures were assembled in Adobe Photoshop CS6 and linear adjustments were done to correct black level and contrast. LacZ reporter expression. Using the LacZ reporter function of the mutated Cacna2d3 gene (Cacna2d3 tm1Dgen ; (Neely et al., 2010), ␤galactosidase activity was analyzed in male ␣ 2 ␦-3 Ϫ/Ϫ knock-out mice. Eight-week-old mice were killed by CO 2 asphyxiation, decapitated, and brains were removed from the skull. Subsequently, hemispheres were separated with a sagittal cut along the midline, placed medial side down on a flat piece of thin acryl glass, and frozen in 2-methylbutan (Carl Roth) cooled to Ϫ80°C. Native samples were stored at Ϫ80°C until further processing and transferred to Ϫ20°C 1 d before sectioning. Twentymicrometer-thick sagittal sections of one hemisphere were obtained with a cryotome (NX50; Histocom), collected on polylysine-coated glass slides (Lactan), and stored at Ϫ20°C until further use. Sections were air dried at room temperature for 15 min, fixed with freshly prepared cold 4% pF and 0.2% glutaraldehyde (Sigma-Aldrich) for 10 min, and washed repeatedly in PBS for 30 min. X-Gal staining reaction was done with X-Gal (Boehringer Mannheim; 4% in N,N-dimethylformamide) diluted 1:40 in prewarmed X-Gal buffer (5 mM K 3 Fe(CN) 6 , 5 mM K 4 Fe(CN) 6 ⅐ 3H 2 O, 2 mM MgCl 2 in PBS; Sigma-Aldrich) at 37°C in a humidified incubator. Reaction was stopped after 24 h by several washes with PBS and rinsing in Milli-Q water (Millipore) at room temperature. Sections were counterstained with 0.2% eosin (Carl Roth) in 0.1 M CH 3 COOH in Milli-Q, dehydrated by a graded series of ethanol, followed by washing in xylene (Carl Roth) and mounting with Eukitt (Christine Gröpl, Tulln, Austria). Representative images were recorded with a BX53 microscope (Olympus) equipped with a SC100 color camera (Olympus). Eight-bit panorama photographs were created by scanning specimens with a 4ϫ 0.16 NA objective and using the manual multiple image alignment (MIA) function in cellSens Dimension software (Olympus).
Quantification of fluorescent clusters in single boutons. To analyze the effects of presynaptic expression of ␣ 2 ␦ subunits in GABAergic synapses of cultured MSNs, 14-bit grayscale images from triple-fluorescence labeling were acquired from the eGFP (green), GABA A R ␤2/3 (red), and vGAT (blue) channels. Images were analyzed with a custom programmed and semiautomated MetaMorph journal (Molecular Devices) as follows. Corresponding eGFP and vGAT images were superimposed, eGFP/vGAT-positive varicosities (putative GABAergic synapses) were randomly chosen as regions of interest (ROIs), and a region for background subtraction was selected. Axonal varicosities were defined as prominent swellings with higher fluorescence signal compared with the adjacent axonal shaft. Subsequently, eGFP, GABA A R, and vGAT grayscale images were thresholded to include the entire area of individual varicosities (eGFP) and synaptic clusters (GABA A R, vGAT). Selected ROIs were then transferred to the eGFP image and, by applying the "shrink region to fit" tool, automatic boundaries were drawn according to the threshold enabling only colocalized clusters to be analyzed. ROIs and background regions were further transferred to the thresholded GABA A R and vGAT channel images and ROI staining intensities were recorded. Therefore, the following parameters could be measured for the individual synapses in each of the channels in a blinded manner: eGFP threshold area as a measure for bouton size and average and integrated fluorescence intensities providing information on the size and intensity of clusters. To address the effects of presynaptic expression of ␣ 2 ␦ subunits in glutamatergic synapses of ␣-Nrxn TKO neurons, 14-bit grayscale images from triple-fluorescence labeling were acquired from the eGFP (green), GABA A R ␤2/3 (red), and vGLUT1 (blue) channels. Images were analyzed with a custom programmed and semiautomated journal as described above. Corresponding eGFP and vGLUT1 images were superimposed and eGFP/vGLUT1-positive varicosities (putative glutamatergic synapses) were randomly selected as ROIs. To analyze the effects of presynaptic expression of ␣ 2 ␦ subunits in glutamatergic synapses of cultured hippocampal WT neurons, 14-bit grayscale images from triplefluorescence labeling were acquired from the eGFP (green), GLUR1 (red), and GABA A R ␤2/3 /gephyrin (blue) channel. Images were analyzed in the same manner as mentioned above but without the use of a semiautomatic MetaMorph journal. For all experiments, background subtractions for individual channels and further data organization and analyses were done with Microsoft Excel. Mean values for individual cells were calculated from individual boutons and normalized to the control (eGFP only) or ␣ 2 ␦-2 condition within each culture preparation. Cumulative frequency distribution of GABA A R ␤2/3 fluorescence intensities of single boutons was done to visualize the effect of ␣ 2 ␦ subunits on the whole GABA A R population. Two to four independent culture preparations were analyzed and details are given in the respective figure legends.
Quantification of gephyrin content in mismatched and neighboring synapses. To compare gephyrin content of ␣ 2 ␦-2-induced mismatched synapses with endogenous clusters, 14-bit grayscale images from triplefluorescence labelings were acquired from the eGFP (green), GLUR1 (red), and gephyrin (blue) channel and analyzed with MetaMorph software (Molecular Devices). First, corresponding eGFP, GLUR1, and gephyrin images were superimposed and eGFP-positive varicosities expressing ␣ 2 ␦-2 (putative mismatched synapses) were selected as ROIs and a region was selected for background subtraction. eGFP, GLUR1, and gephyrin grayscale images were thresholded to include the entire area of individual varicosities (eGFP) and synaptic clusters (GLUR1, gephyrin). Regions were transferred to the eGFP image, shrunk to fit the threshold area, and further transferred to the thresholded gephyrin channel images. After measuring the integrated gephyrin intensity of eGFPpositive mismatched synapses, ROIs of 10 untransfected neighboring WT synapses located on the same dendrite were selected and regions were shrunk according to the gephyrin threshold. The GLUR1 staining served as orientation to outline dendritic morphology of untransfected neurons. Per image, one mismatched gephyrin cluster induced by ␣ 2 ␦-2 was compared with the mean of 10 endogenous clusters. Dots in graph represent values for individual boutons (mismatched synapses) and means of 10 endogenous clusters measured per image. Values were normalized to endogenous gephyrin intensities of neighboring synapses within each culture preparation. Three independent culture preparations were done and eight to 10 images per preparation were analyzed.
Analysis of synapse localization in gSTED images. Scans of mCherrylabeled boutons (control or ␣ 2 ␦-2) contacting eGFP-labeled postsynaptic dendrites were acquired from the eGFP (green, gSTED mode) and mCherry (red, confocal mode) channels. Axonal varicosities were visible as prominent swellings with higher fluorescence signal compared with the adjacent axonal shaft. Initially, linear contrast adjustments were done with LAS AF software (Leica Microsystems), thus enabling clear visualization of postsynaptic dendritic spines and the potential overlap of fluorescence signals, which was used to define the contact zone of putative synapses. Superimposed images were exported into a TIFF format and further analyzed using MetaMorph software (Molecular Devices). To determine the position of synapses on postsynaptic elements, two types of analysis were used. First, the postsynaptic contact sites of presynaptic boutons were manually classified as dendritic shafts, dendritic spines, or unclear localizations. Second, the distances from the postsynaptic dendritic shafts to the contact points with the presynaptic boutons were calculated as follows. First, the distances between the center of the dendritic shafts and the shaft surfaces were determined and measured. Second, the distances between the center of dendritic shafts and the contact zone (presynaptic bouton with postsynaptic dendrite) were determined and measured. Third, the actual distance from the dendritic shaft surface to the contact zone was determined by subtracting the individual center surface distances from the center contact zone distances. Dots in graph (Fig. 11C) represent values for individual boutons and means (lines) Ϯ SEM are shown. Two independent culture preparations were analyzed, of which one experiment was performed in a blinded manner.

Sequence alignment and homology modeling
Information on ␣ 2 ␦-2 splice sites was obtained from the Uniprot (Q6PHS9) and Ensemble (ENSMUSG00000010066) databases. Protein sequences of the distinct variants were aligned with CLUSTAL O (1.2.4) multiple sequence alignment tool. The high-resolution structure of ␣ 2 ␦-1 (PDB code: 5GJV; Wu et al., 2016) was used as a template to model the structure of ␣ 2 ␦-2 splice variants. The pdb files were generated with the Swiss-Model server (http://www.expasy.org/swissmod/SWISS-MODEL.html) and models were exported and further analyzed using UCSF Chimera (http://www.rbvi.ucsf.edu/chimera).

Experimental design and statistical analysis
According to the RRR principle, the number of mice used was kept to the minimum necessary for a statistical representative analysis, which was comparable to numbers reported in previous studies. Where indicated and possible, investigators were partially blinded during experiments and analyses. Moreover, analysis of presynaptic and postsynaptic proteins was done in a blinded manner as described above. Two to four independent culture preparations were analyzed per experiment and details on cell or bouton numbers are given in the respective figure legends. The graphs show values of individual cells/boutons (dots) and means (line) Ϯ SEM. The numbers (n) used to calculate SEMs were either given by number of animals used (qRT-PCR) or the number of analyzed cells/ boutons. Before statistical analysis, the distribution of datasets was evaluated with histograms. Whenever datasets were strongly skewed or normality was experimentally excluded (e.g., a large proportion of "0" values), raw data were log10 transformed and data distribution was reevaluated by histograms. For normally distributed raw or log10transformed datasets, unpaired t test or one-way ANOVA with Holm-Sidak post hoc test was used. Alternatively, when log10 transformation did not result in normally distributed data, the Mann-Whitney U test or Kruskal-Wallis ANOVA with Dunn's post hoc analysis was applied. Significance levels ( p-values) of statistical tests and post hoc analysis are presented in the respective figure legends. The model in Figure 3 was generated with Maya software (version 2018; Autodesk). Data, graphs and figures were organized, analyzed, and assembled using Microsoft Excel, GraphPad Prism 6, SigmaPlot (Systat Software), and Adobe Photoshop CS6.

Presynaptic overexpression of ␣ 2 ␦-2 induces the formation of mismatched synapses in excitatory hippocampal neurons
To date, the synaptic functions of ␣ 2 ␦ subunits have been identified in cells and tissues primarily expressing a single dominant ␣ 2 ␦ isoform (Eroglu et al., 2009;Pirone et al., 2014;Fell et al., 2016;Wang et al., 2017). To study the involvement of the individual ␣ 2 ␦ isoforms in glutamatergic synapse formation and differentiation, we homologously overexpressed mouse ␣ 2 ␦-1, ␣ 2 ␦-2, and ␣ 2 ␦-3 cDNAs in WT mouse hippocampal cultures together with soluble eGFP. To analyze the consequences on synapse differentiation, transfected neurons were immunolabeled against the vesicular glutamate transporter (vGLUT1), a marker for presynaptic excitatory synapses, and the postsynaptic GABA A R ␤2/3 subunit (GABA A R), which is typically absent in glutamatergic synapses. Immunofluorescence analysis revealed that, as expected, in control hippocampal neurons (eGFP only) axonal varicosities were positive for presynaptic vGLUT1 and negative for postsynaptic GABA A R labeling (Fig. 1A,B, first column). Surprisingly, presynaptic expression of ␣ 2 ␦-2 induced a robust localization of GABA A R clusters opposite vGLUT1positive presynaptic boutons (Fig. 1B, blue arrowheads). These apparently mismatched synapses are atypical because they are formed between excitatory nerve terminals and inhibitory postsynaptic receptors. Importantly, mismatched postsynaptic receptor localization was not observed opposite terminals expressing ␣ 2 ␦-1 or ␣ 2 ␦-3.
Excitatory and inhibitory neurotransmitter transporters can coexist in glutamatergic hippocampal and cerebellar mossy fiber terminals, possibly leading to the corelease of glutamate and GABA and the localization of postsynaptic GABA A Rs in spines (Bergersen et al., 2003;Münster-Wandowski et al., 2016). Therefore, we next tested whether vesicular GABA transporters (vGATs) are present in synapses overexpressing presynaptic ␣ 2 ␦ subunits. As expected and in contrast to neighboring untransfected GABAergic synapses (Fig. 1D, asterisks), presynaptic vGAT and postsynaptic GABA A Rs were not found in putative excitatory synapses expressing ␣ 2 ␦-1, ␣ 2 ␦-3, or eGFP only (Fig.  1C,D). Most importantly, presynaptic vGAT labeling was also absent in synapses expressing presynaptic ␣ 2 ␦-2, which again showed strong postsynaptic GABA A R clustering (Fig. 1D, ϩ␣ 2 ␦-2, blue arrowheads). Therefore, the mismatched localization of postsynaptic GABA A Rs cannot be explained by the exis-tence of inhibitory neurotransmitters within glutamatergic nerve terminals.
To further characterize our observation, we next quantitatively assessed the postsynaptic receptor composition of mismatched glutamatergic synapses by double labeling transfected hippocampal neurons with markers for excitatory and inhibitory synapses ( Fig. 2A-C). Immunofluorescence analysis revealed that, in control hippocampal neurons (eGFP only) postsynaptic AMPARs (GLUR1), but not GABA A Rs, were expressed opposite transfected presynaptic terminals, as expected for putative  glutamatergic synapses ( Fig. 2A, control). Consistent with our initial observations (Fig. 1), postsynaptic GABA A R fluorescent intensities were strongly and significantly increased opposite ␣ 2 ␦-2expressing presynaptic terminals (Fig.  2 A, D). The elevated GABA A receptor abundance ( Fig. 2A) was accompanied by a 76% reduction of GLUR1 clustering compared with control (Fig. 2E). Surprisingly, presynaptic expression of ␣ 2 ␦-3, which did not alter postsynaptic GABA A R clustering, also caused a significant reduction of GLUR1 receptors (66% reduction vs control). Finally, we also observed a slight but not significant tendency for a reduced GLUR1 receptor expression opposite ␣ 2 ␦-1-overexpressing boutons (Fig. 2E). We next tested whether gephyrin, a key organizer of GABAergic synapses that anchors, clusters, and stabilizes GABA A R and other postsynaptic elements of inhibitory synapses (Choii and Ko, 2015), was also recruited to mismatched GABA A R-positive synapses. Indeed, we found that the strong increase in GABA A R expression opposite ␣ 2 ␦-2-expressing terminals was accompanied by similarly increased gephyrin abundance, whereas it was missing from control and ␣ 2 ␦-1or ␣ 2 ␦-3-expressing synapses (Fig. 2 B, F ).
Presynaptic ␣ 2 ␦-2 recruits synaptic GABA A R subtypes Because immunofluorescence analysis suggested an effect on postsynaptic GABA A R clusters opposite presynaptic boutons, we next used superresolution gSTED microscopy to verify their subsynaptic localization (Fig. 3A). Postsynaptic GABA A R clusters appear closely opposed to mCherrypositive, ␣ 2 ␦-2-overexpressing presynaptic boutons: depending on the orientation of the imaged synapses, the GABA A R-labeling pattern is either visible as a thin line (side view; Fig. 3A, first column and Fig. 3B, top) or as a ring-like structure (planar view; Fig.  3A, second column and Fig. 3B, bottom). These labeling patterns support a confined localization of ␣ 2 ␦-2-recruited GABA A Rs in the postsynaptic membrane.
Potential mechanisms explaining an ␣ 2 ␦-2-induced mismatched synapse formation So far, our results show that the specific expression of a single ␣ 2 ␦ subunit isoform, ␣ 2 ␦-2, in presynaptic glutamatergic terminals triggers a mismatched localization of postsynaptic GABA A Rs, which was accompanied by a strong reduction of postsynaptic AMPARs. The underlying mechanism could be explained by three hypotheses (Fig. 5): First, it is known that an elevated expression of ␣ 2 ␦ subunits increases presynaptic calcium channel abundance and current densities (Hoppa et al., 2012;Geisler et al., 2015) and thus likely enhances synaptic transmission (Zhou and Luo, 2015;Chen et al., 2018). Therefore, postsynaptic GABA A Rs could be recruited opposite these glutamatergic boutons in an attempt to suppress excessive glutamatergic excitation (Shrivastava et al., 2011). If this hypothesis applies, then GABA A R abundance at inhibitory synapses should not change upon ␣ 2 ␦-2 overexpression in GABAergic presynaptic terminals (Fig. 5A). Second, presynaptic ␣ 2 ␦-2, an extracellular protein extending far into the synaptic cleft, may contribute to the anchoring of postsynaptic GABA A Rs by a trans-synaptic mechanism. Such a specific role of ␣ 2 ␦-2 should be independent of the synapse type and thus GABA A R abundance should be affected when ␣ 2 ␦-2 is overexpressed in both glutamatergic and GABAergic synapses (Fig.   5B). Independent of the two hypotheses, the observed reduction in GLUR1 labeling opposite ␣ 2 ␦-2-expressing glutamatergic terminals (Fig. 2) could be explained by GABA A Rs competing for AMPAR slots within dendritic spines. Alternatively, however, presynaptic ␣ 2 ␦-2 could also trans-synaptically induce an aberrant axonal wiring by guiding glutamatergic axons to GABAergic postsynaptic locations along dendritic shafts (Fig. 5C). In this scenario, the reduced GLUR1 expression would be a secondary effect because no GLUR1 receptors are to be expected in postsynaptic GABAergic synapses. To distinguish between these hypotheses, we next studied the consequence of presynaptic ␣ 2 ␦-2-overexpression on the molecular composition of GABAergic synapses.

MSNs express three presynaptic ␣ 2 ␦ isoforms
The cultures of hippocampal neurons used for the analysis of experiments presented in Figures 1, 2, 3, and 4 contain only 5-10% GABAergic interneurons (Obermair et al., 2003). Consequentially transfected GABAergic neurons are rarely observed. . Presynaptic expression of ␣ 2 ␦-2 induces the recruitment of synaptic GABA A R subtypes. Representative immunofluorescence micrographs of cultured hippocampal neurons cotransfected with distinct ␣ 2 ␦ subunits and soluble eGFP. Transfected permeabilized neurons (20 -30 DIV) were double stained for vGLUT1 and different postsynaptic GABA A R subtypes. Colocalization of fluorescence signals within eGFP-filled axonal varicosities (arrowheads, axons are outlined with dashed lines) was analyzed using line scans. A, B, Immunofluorescence analysis identified intensely fluorescent clusters of the GABA A R subunits ␣ 1 , ␣ 2 , ␤ 3 , and ␥ 2 opposite ␣ 2 ␦-2 expressing glutamatergic (vGLUT1-positive) nerve terminals (A). In contrast, these postsynaptic GABA A R subtypes are absent opposite putative glutamatergic synapses expressing eGFP only (control), ␣ 2 ␦-1, or ␣ 2 ␦-3 (B, micrographs depict examples for ␣ 1 ). C, D, Labeling of the GABA A R subunits ␣ 3 , ␣ 4 , ␤ 2 , and ␦ displayed weak and mainly extrasynaptic immunoreactivity in all conditions. Note that all GABA A R subtypes presented for ␣ 2 ␦-2 were also analyzed in hippocampal neurons expressing eGFP only, ␣ 2 ␦-1, and ␣ 2 ␦-3. Representative images of two independent cultures are shown. Scale bars, 3 m. Therefore, to quantitatively address the role of ␣ 2 ␦ subunits in GABAergic synapses, we established cultures of striatal neurons, which consist of ϳ95% inhibitory MSNs. Quantitative RT-PCR revealed that differentiated monocultured MSNs (24 -25 DIV) expressed three neuronal ␣ 2 ␦ isoforms (␣ 2 ␦-1, ␣ 2 ␦-2, and ␣ 2 ␦-3) similar to the adult mouse striatum (Fig. 6A). Whereas ␣ 2 ␦-2 and ␣ 2 ␦-3 expression levels were similar in cultured neurons, the higher abundance of ␣ 2 ␦-3 transcripts in 8-week-old striatal tissue was consistent with previous studies (Cole et al., 2005). The strong expression of ␣ 2 ␦-3 was confirmed on cryosections of a mouse line carrying a lacZ reporter gene in the Cacna2d3 locus using ␤-galactosidase histochemistry (Fig. 6B). The ␣ 2 ␦-4 iso- Figure 5. Potential mechanisms explaining the observed mismatched synapse formation of glutamatergic nerve terminals expressing ␣ 2 ␦-2. A, Compensatory mechanism. Elevated expression of ␣ 2 ␦ subunits increases presynaptic calcium channel abundance and current densities and thus glutamate release. Therefore, GABA A Rs could be recruited to the dendritic spine in an attempt to compensate for excessive excitatory synaptic activity. If this is the case, then GABA A R abundance at inhibitory synapses should not change upon ␣ 2 ␦-2 overexpression in GABAergic presynaptic terminals. B, ␣ 2 ␦-2 may be involved in trans-synaptically anchoring postsynaptic GABA A Rs. In this scenario, GABA A R abundance should be increased when ␣ 2 ␦-2 is overexpressed in both glutamatergic and GABAergic synapses. C, Trans-synaptic function of ␣ 2 ␦-2 could also induce aberrant axonal wiring by guiding glutamatergic axons to GABAergic postsynaptic locations positioned along dendritic shafts. This is in contrast to the normal situation in which glutamatergic synapses are generally formed on dendritic spines.
form, which is predominantly present in the retina, was hardly detectable and thus not considered for the following experiments. For studying presynaptic ␣ 2 ␦ subunit expression in differentiated GABAergic neurons, we adapted a coculture system of GABAergic MSNs with glutamatergic cortical neurons (Penrod et al., 2011) (Fig. 7A). Lentiviral infection of MSNs with soluble eGFP just before starting the coculture allowed discriminating them from excitatory cortical neurons (Fig. 7B) based on their eGFP fluorescence. Immunolabeling of presynaptic and postsynaptic markers for excitatory (vGLUT1, PSD-95) and inhibitory (vGAT) synapses demonstrated that glutamatergic and GABAergic synapses formed properly on dendritic spines and shafts, respectively (Fig. 7C-E). Moreover, patch-clamp recordings of spontaneous mEPSCs and mIPSCs in MSNs further confirmed the functionality of glutamatergic and GABAergic synapses in 14 DIV cocultures (Fig. 7F ). Finally, live-cell labeling of MSNs cotransfected with HA-tagged ␣ 2 ␦ subunits and soluble eGFP revealed that all ␣ 2 ␦ isoforms are expressed on the surface of presynaptic axonal varicosities and thus can in theory contribute to synaptic functions in GABAergic neurons (Fig. 6C).

Presynaptic overexpression of ␣ 2 ␦-2 upregulates postsynaptic GABA A Rs in inhibitory synapses
To distinguish between our hypotheses (Fig. 5), we coexpressed individual ␣ 2 ␦ isoforms together with soluble eGFP in differentiated GABAergic MSNs. As expected, immunofluorescence analysis revealed that, in control MSNs (eGFP only), axonal varicosities were typically positive for presynaptic vGAT and postsynaptic GABA A R labeling (Fig. 8 A, B). Postsynaptic GABA A Rs were similarly localized opposite vGAT-positive terminals expressing the individual ␣ 2 ␦ isoforms (Fig. 8B). However, quantitative analysis of GABA A R fluorescence intensities of individual cells revealed that GABA A R labeling was strongly (ϳ2-fold) and significantly increased opposite boutons expressing the ␣ 2 ␦-2 isoform (Fig. 8C). Furthermore, presynaptic expression of ␣ 2 ␦-3 caused a 20% decrease of GABA A R labeling intensity compared with control (Fig. 8C). Plotting the cumulative frequency distribution of GABA A R fluorescence intensities of the single synapses demonstrates that presynaptic ␣ 2 ␦ subunit isoforms differentially regulate postsynaptic GABA A R abundance in MSNs (Fig.  8D). GABA A R clusters were increased in size and intensity in the majority of synapses overexpressing ␣ 2 ␦-2 (blue line), causing a shift of the entire population to the right toward higher fluorescence intensities compared with control (eGFP only, green line). In contrast, ϳ24% and ϳ40% of all boutons expressing ␣ 2 ␦-1 (yellow line) and ␣ 2 ␦-3 (magenta line), respectively, lacked postsynaptic GABA A R labeling and the remaining population was left shifted toward smaller clusters. The slightly but significantly reduced GABA A R clustering opposite ␣ 2 ␦-3-overexpressing boutons (Fig. 8C) was associated with smaller presynaptic boutons (27% smaller than control; Fig. 8E) and decreased presynaptic vGAT content (46% lower than control; Fig. 8F ). Conversely, presynaptic expression of ␣ 2 ␦-2, which strongly increased postsynaptic GABA A R clustering, affected neither bouton size nor vGAT labeling intensity.
Similar to mismatched glutamatergic synapses (see above), we used superresolution gSTED microscopy to visualize the subsynaptic localization of postsynaptic GABA A R clusters in cultured MSNs (Fig. 3C). MSNs were cotransfected with ␣ 2 ␦-2 and soluble mCherry or mCherry only (control). In both conditions, postsynaptic GABA A R clusters appeared closely opposed to mCherrypositive presynaptic boutons, thus confirming their localization in the postsynaptic membrane. Moreover, similar to hippocam-pal neurons, the GABA A R pattern was either visible as a thin line (side view; Fig. 3C, first and third column and Fig. 3B, top) or a ring-like structure (planar view; Fig. 3C, second and fourth column and Fig. 3B, bottom). Importantly, gSTED imaging further confirmed that presynaptic expression of ␣ 2 ␦-2 increased GABA A R density at the corresponding postsynaptic side.
␣-Nrxns modulate postsynaptic GABA A R abundance set by presynaptic ␣ 2 ␦-2 Thus far, our results revealed that presynaptic ␣ 2 ␦-2 strongly induced postsynaptic GABA A receptor clustering in both glutamatergic and GABAergic synapses, suggesting that ␣ 2 ␦-2 acts trans-synaptically according to our hypothesis outlined in Figure  5. This means that ␣ 2 ␦-2 could either directly interact with postsynaptic receptors or may act via other synaptic organizer molecules such as ␣-Nrxns. Presynaptic ␣-Nrxns are synaptic cell adhesion molecules whose general importance in a variety of synaptic functions is well established (Reissner et al., 2013;Südhof, 2017). Several lines of evidence implicate ␣-Nrxns as interesting candidates for mediating a possible indirect interaction between ␣ 2 ␦-2 and GABA A Rs. First, ␣-Nrxns have been linked functionally to high-voltage-dependent Ca 2ϩ channels, which has strong effects on presynaptic neurotransmitter release (Missler et al., 2003;Chen et al., 2017). Second, ␣-Nrxns have been implicated as organizing molecules at both excitatory glutamatergic and inhibitory GABAergic synapses by promoting clustering of postsynaptic AMPARs (Heine et al., 2008;Aoto et al., 2013Aoto et al., , 2015 and GABA A Rs (Kang et al., 2008). Third, the presynaptic complex of ␣-Nrxn and its ligand neurexophilin-1 is characteristic for inhibitory interneurons and can induce postsynaptic GABA A R recruitment when ectopically expressed at excitatory synapses (Born et al., 2014). Fourth, neurexin-1␣ has recently been shown to modulate P/Q-type (Brockhaus et al., 2018) and N-type (Tong et al., 2017) calcium channels by cooperating with ␣ 2 ␦ subunits. Therefore, to test the possibility that upregulation of postsynaptic GABA A Rs depends on ␣-Nrxns, we expressed ␣ 2 ␦-2 in glutamatergic synapses lacking all three ␣-Nrxn isoforms (Missler et al., 2003). We transfected eGFP alone or in combination with ␣ 2 ␦-2 or ␣ 2 ␦-3 in hippocampal cultures from WT and ␣-Nrxn TKO mice. Immunofluorescence analysis revealed that, as expected, in control hippocampal neurons, axonal varicosities were typically positive for presynaptic vGLUT1 but negative for postsynaptic GABA A R labeling ( Fig.  9 A, C, WT control). Consistent with our previous observations, presynaptic expression of ␣ 2 ␦-2 induced the formation of mismatched synapses (Fig. 9 A, C, WTϩ␣ 2 ␦-2), whereas expression of ␣ 2 ␦-3 had no effect ( p ϭ 0.64; Fig. 9 A, C, WTϩ␣ 2 ␦-3). Strikingly, expression of presynaptic ␣ 2 ␦-2 in ␣-Nrxn TKO neurons induced postsynaptic GABA A R clustering opposite glutamatergic boutons more strongly than in WT (Fig. 9 A, C, ␣-Nrxn TKOϩ␣ 2 ␦-2). Quantitative analysis revealed that this effect was three times higher compared with the WTϩ␣ 2 ␦-2 condition, suggesting that the propensity of ␣ 2 ␦-2 in inducing mismatched synapses was enhanced in the absence of ␣-Nrxns. In fact, increased GABA A R abundance was already visible in nontransfected ␣-Nrxn TKO synapses in which the presence of endogenous ␣ 2 ␦-2 may be responsible for a significantly increased baseline level of mismatched GABA A Rs ( p ϭ 0.023; Fig. 9 A, C). This is an important result because it suggests that no overexpression of ␣ 2 ␦-2 is needed to induce postsynaptic GABA A R clustering, but an imbalance between ␣ 2 ␦-2 and ␣-Nrxns. These differential effects can be best compared by blotting the cumulative frequency distribution of GABA A R fluorescence intensities of all individual synapses analyzed (Fig. 9D). Compared with WT (control, green line), the entire population of synapses was rightshifted toward larger GABA A R clusters in ␣-Nrxn TKO neurons (dashed green line). Presynaptic expression of transfected ␣ 2 ␦-2 strongly shifted the distributions toward larger GABA A R clusters in WT (ϩ␣ 2 ␦-2, blue line) and ␣-Nrxn TKO neurons (TKOϩ ␣ 2 ␦-2, dashed blue line). Presynaptic expression of ␣ 2 ␦-3 had no effect in WT neurons, which, due to the lack of mismatched synapses, basically displayed only background GABA A R intensity (ϩ␣ 2 ␦-3, magenta line). Interestingly, overexpression of ␣ 2 ␦-3 in ␣-Nrxn TKO neurons reduced the elevated GABA A R clusters to control levels (TKOϩ␣ 2 ␦-3, dashed magenta line). This possibly indicates that overexpressed, abundant ␣ 2 ␦-3 blocked the effect by endogenous ␣ 2 ␦-2 on baseline GABA A R intensity in TKO synapses. Together, our data suggest that presynaptic ␣ 2 ␦-2 potently induced mismatched synapse formation with GABA A R clustering in the absence of ␣-Nrxns. This demonstrates that the transsynaptic role of ␣ 2 ␦-2 in recruiting postsynaptic GABA A Rs is independent of ␣-Nrxns. However, the fact that GABA A R abundance was already increased in ␣-Nrxn TKO synapses without overexpression of ␣ 2 ␦-2 and that the abundance could be differentially modulated by presynaptic ␣ 2 ␦-2 and ␣ 2 ␦-3 suggests a cooperative and trans-synaptic activity of ␣ 2 ␦ subunits and ␣-Nrxns in fine-tuning postsynaptic GABA A R levels.
To distinguish between these two possibilities, we focused more closely on the position of the mismatched synapses along the postsynaptic neurons. First, it is important to emphasize that ␣ 2 ␦-2 induced the upregulation of postsynaptic GABAergic components only at the sites contacted by the presynaptic axon without affecting the neighboring synapses. This is best visualized by the observation that gephyrin fluorescence intensity was 4.5 times higher in ␣ 2 ␦-2-induced mismatched synapses ( Fig.  10A-C, asterisks in A and B) compared with neighboring endogenous GABAergic synapses situated on the same dendrite, which are formed by axons from untransfected neurons ( Fig. 10A-C,  arrowheads in A and B). This finding further supports a local and synapse-specific effect of presynaptic ␣ 2 ␦-2 rather than a global alteration in the expression of GABAergic postsynaptic components. Importantly, ␣ 2 ␦-2-overexpressing glutamatergic axons preferentially formed synapses at typical GABAergic postsynaptic locations, namely linearly aligned along the dendritic shaft (Fig.  10D, right column, blue arrowheads), in contrast to control axons, which preferably contacted dendritic spines (Fig. 10D, left column, white arrowheads). Therefore, to determine the exact position of mismatched synapses on postsynaptic dendrites, we established hippocampal cultures in which presynaptic (control or ␣ 2 ␦-2) and postsynaptic neurons were transfected with mCherry (red) and eGFP (green), respectively ( Fig. 11A-C). gSTED microscopy allowed precise visualization of the contact point between the presynaptic mCherry-positive axon and the postsynaptic eGFP-positive dendrite (Fig. 11 A, B). As expected for glutamatergic excitatory synapses, the vast majority of axonal varicosities was located on dendritic spines in the control condition (80% spine, 13% shaft, 7% unclear; Fig. 11 B, C, top). In contrast, presynaptic expression of ␣ 2 ␦-2 significantly shifted the preferential contact points to dendritic shafts (24% spine, 65% shaft, 11% unclear; 2 test: p Ͻ 0.001). This finding was further  sketch, B). Consistent with our initial observation (see Fig. 1), presynaptic expression of ␣ 2 ␦-2 (blue arrowheads) induces postsynaptic GABA A R localization opposite vGLUT1-positive nerve terminals in WT neurons. Most importantly, in ␣-Nrxn TKO synapses, the expression of ␣ 2 ␦-2 strongly induced postsynaptic GABA A R localization (blue asterisks). C, Postsynaptic GABA A Rs clusters (integrated intensities) are significantly increased opposite glutamatergic nerve terminals expressing ␣ 2 ␦-2 compared with control (eGFP only). This effect is much stronger in ␣-Nrxn TKO synapses compared with WT neurons (3-fold higher GABA A R fluorescence intensity). D, Cumulative frequency distribution further reveals that the vast majority of glutamatergic boutons are positive for postsynaptic GABA A Rs in WTϩ␣ 2 ␦-2 (88%) and TKOϩ␣ 2 ␦-2 (97%) and that the population is shifted toward larger and more intense clusters. However, this analysis also demonstrates that mismatched GABA A Rs already form opposite ␣-Nrxn TKO synapses, although at a lower overall intensity. Values for individual cells (dots) and means (lines) Ϯ SEM are shown. Values were normalized to the WT control within each culture preparation. Data from two independent culture preparations and 12-19 cells were analyzed within each condition. Statistics: C, ANOVA on log10-transformed data with Holm-Sidak post hoc analysis: genotype: F (1,86) ϭ 5.3, p ϭ 0.023; transfection: F (2,86) ϭ 49.0, p Ͻ 0.001; genotype ϫ transfection: F (2,86) ϭ 3.7, p ϭ 0.03; post hoc: *p Ͻ 0.05, ***p Ͻ 0.001. Asterisks in graphs indicate the significant difference compared with control. Scale bar, 3 m. corroborated by a highly significant reduction of the mean distance from the contacting presynaptic boutons to the postsynaptic dendritic shafts in ␣ 2 ␦-2-transfected compared with control axons (mean distance Ϯ SEM: control: 0.92 Ϯ 0.05 m; ␣ 2 ␦-2: 0.54 Ϯ 0.05 m; Mann-Whitney U test: p Ͻ 0.001; Fig. 11C, bottom). Importantly, applying the same methodological approach for the coculture of GABAergic MSNs and glutamatergic cortical neurons (Fig. 11D, also see Fig. 7) revealed that axonal varicosities from control cortical neurons generally contacted dendritic spines of postsynaptic inhibitory MSNs (total of 65 synapses: 83% spine, 1.5% shaft, 15.5% unclear: Fig. 11 E, F, top). In contrast, the presynaptic expression of ␣ 2 ␦-2 shifted the preferential contact points of cortical boutons to dendritic shafts of MSNs (total of 51 synapses: 19.6% spine, 72.6% shaft, 7.8% unclear; Fig. 11 E, F, bottom, gephyrin labeling and position at dendritic shaft). These findings demonstrate that the ␣ 2 ␦-2-induced aberrant wiring is independent of the identity of the postsynaptic neuron.
Together, our results provide strong evidence that overexpression of presynaptic ␣ 2 ␦-2 induces the aberrant wiring of glutamatergic axons to postsynaptic GABAergic locations, resulting in a mismatch of presynaptic excitatory neurotransmitters and postsynaptic inhibitory receptors.

Reduced synaptic transmission in aberrantly wired synapses
Excitatory and inhibitory synapses require matched presynaptic and postsynaptic specializations to enable efficient and synchronous neurotransmission. To assess the functional consequences of ␣ 2 ␦-2 overexpression in presynaptic boutons of aberrantly wired glutamatergic synapses, we performed whole-cell patchclamp recordings on synaptically connected pairs of isolated neu-rons grown on poly-L-lysine dots (Fig. 12A). We first investigated whether AMPAR-mediated evoked EPSCs (eEPSCs) between individual pairs of hippocampal neurons were altered by the lentiviral overexpression of ␣ 2 ␦-2 in the presynaptic neuron. Isolated synaptic currents were measured in the postsynaptic neuron in response to an action potential induced in the presynaptic neuron. The presynaptic action potential followed a given waveform recorded from WT hippocampal neurons (Fig. 12B). This experimental setup allowed to directly compare synaptic transmission in glutamatergic control synapses (presynaptic cell ϭ control, postsynaptic cell ϭ ␣ 2 ␦-2 overexpressing) and mismatched synapses (presynaptic cell ϭ ␣ 2 ␦-2 overexpressing, postsynaptic cell ϭ control). In each of seven neuronal pairs recorded, the eEPSC amplitude was lower in the mismatched compared with the control pair (on average reduced by 48%) (Fig. 12C). To discriminate whether the eEPSC reduction is the result of the reduced postsynaptic AMPA receptor expression associated with the aberrantly wired axons or if presynaptic ␣ 2 ␦-2 reduces release probability, we next measured synaptic paired-pulse response ratios (PPRs) (Katz and Miledi, 1968;Zucker and Regehr, 2002). eEPSCs were recorded in the postsynaptic neuron in response to two action potentials elicited in the presynaptic neuron at different interstimulus intervals. As previously reported for hippocampal pyramidal neurons, calculated PPRs displayed facilitation (Fig. 12D) with a decline in the magnitude at increasing intervals (Dobrunz et al., 1997). Overexpression of ␣ 2 ␦-2 in presynaptic neurons slightly increased facilitation at 25 and 50 ms compared with the initial 10 ms interval. The strong reduction of eEPSC amplitudes is thus consistent with a reduced number of functional synapses caused by the aberrant wiring of glutamatergic ␣ 2 ␦-2-expressing boutons with GABAergic postsynaptic sites. Conversely, the only moderately altered paired-pulse facilitation (PPF) indicates that the decreased amplitude of eEPSCs was not caused by a reduced presynaptic release probability. The observed slightly increased PPF may reflect an increased abundance of presynaptic calcium channels caused by the overexpression of ␣ 2 ␦ subunits, as has been previously demonstrated (Hoppa et al., 2012).

Lack of exon 23 in ␣ 2 ␦-2 splice variants mediates the trans-synaptic effect on GABA A Rs
Our results revealed a specific trans-synaptic function of only one of the three expressed ␣ 2 ␦ isoforms, namely ␣ 2 ␦-2. Sequence comparisons of the previously known alternatively spliced regions (Hobom et al., 2000) identified three potential splice sites in ␣ 2 ␦-2 (Fig. 13A, scheme): inclusion or absence of exon 23 and two alternative splice sites at exons 30 and 38. Using homology modeling, we explored the potential implication of these sequence differences on the structural fold of ␣ 2 ␦-2 (Fig. 14) based on the recently suggested ␣ 2 ␦-1 cryo-EM structure (Wu et al., 2016). Inclusion of exon 23 resulted in the formation of an extra loop that might disrupt an ␣-helix (Figs. 13B, 14B), whereas alternative splicing of exons 30 and 38 predicted only small changes in the secondary structure (Fig. 14C,D). Therefore, to test whether alternative splicing at these sites is involved in the transsynaptic function of ␣ 2 ␦-2, we cloned and compared three splicing variants (␣ 2 ␦-2-v1, ␣ 2 ␦-2-v2, and ␣ 2 ␦-2-v3, ␣ 2 ␦-2-v1 ϭ original construct; Fig. 13A). To determine the potential of the splice variants in inducing mismatched synapses, we transfected WT hippocampal neurons with the three distinct ␣ 2 ␦-2 variants together with soluble eGFP. Strikingly, only the splice variants lacking exon 23 (␣ 2 ␦-2-v1 and ␣ 2 ␦-2-v3) induced the clustering of postsynaptic GABA A Rs opposite transfected glutamatergic nerve terminals (Fig. 13C, ⌬Exon 23). The fact that ␣ 2 ␦-2-v1 and ␣ 2 ␦-2-v3 both induced mismatched synapse formation excludes a major role of exons 30 and 38, which differ in ␣ 2 ␦-2-v1 and ␣ 2 ␦-2-v3. In contrast, ␣ 2 ␦-2-v2, which contains the exon 23 insert, failed completely to induce mismatched synapse formation (Fig. 13C, ϩExon 23), indicating that the trans-synaptic function Figure 11. Presynaptic ␣ 2 ␦-2 induces the aberrant wiring of glutamatergic axons to dendritic shafts. A, D, To analyze the position of ␣ 2 ␦-2-induced mismatched synapses on postsynaptic dendrites of excitatory (A, hippocampal neurons) and inhibitory neurons (D, MSNs), presynaptic (control or ␣ 2 ␦-2) and postsynaptic neurons were labeled with mCherry (red) and eGFP (green), respectively (for details, see Materials and Methods). The magnified insets in A and D demonstrate the close association of presynaptic axonal varicosities (mCherry) with postsynaptic dendritic spines (eGFP), as expected for excitatory spine synapses. Using gSTED microscopy, the distance from the dendritic shaft to the contact zone of each synaptic bouton was measured (A, arrows and lines in the magnified inset). B, E, gSTED microscopy confirms the preferential location of excitatory synapses on dendritic spines of both glutamatergic or GABAergic neurons (white arrowheads and sketch in B) and, most importantly, suggests an aberrant wiring of putative glutamatergic axons expressing ␣ 2 ␦-2 with postsynaptic sites along dendritic shafts (blue arrowheads and sketch in B), as typically observed for GABAergic synapses. C, Corresponding postsynaptic location of each contacting bouton was categorized as spine (green), shaft (gray), or as unclear (white; top) and the distance from the dendritic shaft to the contact point was measured (bottom). Although the vast majority of synaptic contact points in the control condition are located on dendritic spines (80%), presynaptic expression of ␣ 2 ␦-2 shifted the preferential contact points to dendritic shafts (65%, top). This is further confirmed by a significantly decreased contact point to shaft distance of ␣ 2 ␦-2-expressing compared with control synapses. Values for individual boutons (dots) and means (lines) Ϯ SEM are shown. Dot colors (bottom) show the respective categorized synapse position (spine, green; shaft, gray; and unclear, white) presented in the top. Data from two independent culture preparations and 118 (control) and 75 (ϩ␣ 2 ␦-2) boutons were analyzed. Statistics: C, top: 2 test: 2 (2) ϭ 63.4, ***p Ͻ 0.001; bottom: Mann-Whitney U test: ***p Ͻ 0.001. E, F, gSTED microscopy (E) and high-resolution fluorescence microscopy (F ) revealed that, similar to hippocampal neurons, presynaptic expression of ␣ 2 ␦-2 in cortical neurons shifted the preferential contact points to dendritic shafts of GABAergic MSNs, as indicated by gephyrin labeling and the dendritic position (blue arrowheads and line scan in F ). Representative micrographs of two independent cultures are shown. Scale bars, 50 m (overview), 3 m (insets, A, D), and 2 m (B, E, F ).
of ␣ 2 ␦-2 on GABA A Rs is transcriptionally regulated. Together, our results have uncovered a highly specific role of the ␣ 2 ␦-2 subunit of voltage-gated calcium channels in clustering postsynaptic GABA A Rs, an important process that is further regulated by the presence of ␣-Nrxns and by alternative splicing.

Discussion
The experiments described here reveal that presynaptic expression of a single splice variant of ␣ 2 ␦-2 trans-synaptically regulates postsynaptic GABA A R abundance. Therefore, presynaptic ␣ 2 ␦-2 modulates postsynaptic GABA A R clustering independently of the presynaptic neurotransmitter, induces an aberrant wiring of glutamatergic axons with postsynaptic GABAergic sites of both glutamatergic and GABAergic neurons, and alters synaptic transmission in mismatched synaptic pairs. These results demonstrate for the first time that presynaptic ␣ 2 ␦ subunits are able to trans-synaptically regulate synapse specification and axonal wiring.
␣ 2 ␦-2 induces the formation of mismatched synapses via a trans-synaptic mechanism Excitatory and inhibitory synapses are characterized by their distinct molecular composition and subcellular localization in neurons. The proper alignment of presynaptic and postsynaptic specializations is a hallmark of synapse formation and a prerequisite for efficient and synchronous excitatory and inhibitory synaptic transmission (Lardi- Studler and Fritschy, 2007). Only a small number of studies have reported mismatched synapses in vitro. In micro-island cultures of autaptic hippocampal neurons, the vast majority of GABA A Rs and gephyrin clusters are located opposite glutamatergic terminals, suggesting that postsynaptic compartments can be recruited without the proper presynaptic input (Rao et al., 2000;Christie et al., 2002). Moreover, in low-density cultures of hippocampal and cerebellar granule neurons, mismatches of presynaptic and postsynaptic proteins can form during early developmental stages, but are essentially absent in differentiated cultures (Brünig et al., 2002;Anderson et al., 2004;Studler et al., 2005). In contrast to these studies, the mismatched synapses reported in our study are specifically induced by the presynaptic expression of ␣ 2 ␦-2 and are independent from neurotransmitter input, suggesting that ␣ 2 ␦-2 operates via a different mechanism.
Mixed synapses with glutamatergic and GABAergic properties have also been detected in different brain areas, indicating a functional role in vivo. For example, postsynaptic GABA A Rs are present opposite presynaptic glutamatergic mossy fiber axon terminals in the cerebellar cortex and hippocampus (Nusser et al., 1996a,b;Bergersen et al., 2003). Physiologically, these receptors may be involved in fine-tuning excitatory synaptic transmission in response to GABA spillover from surrounding synapses (Isaacson et al., 1993). Although we cannot completely rule out GABA spillover from adjacent neurons, it is important to note that the ␣ 2 ␦-2-induced mismatched synapses do not express presynaptic vGAT. Therefore, the recruitment of postsynaptic GABA A Rs is entirely independent of presynaptic GABA release. These findings are consistent with a recent report showing that, even in the absence of presynaptic glutamate release, dendritic spines are formed and AMPARs are present in glutamatergic synapses (Sigler et al., 2017).
boutons predominantly contact postsynaptic dendritic spines (Peters, 2002), whereas GABAergic nerve terminals mostly connect to the shaft of the dendrite (Markram et al., 2004). Nevertheless, converging lines of evidence demonstrate morphologically mixed synapse types in different parts of the brain. In the adult neocortex, for example, two distinct types of GABAergic synapses are positioned either on the shaft or spine of the dendrite, which are suggested to have different roles in shaping dendritic activity (Chen et al., 2012). It is hypothesized that these constellations facilitate the shunting of excitatory currents, thereby balancing excitatory and inhibitory neurotransmission (Nusser et al., 1996a;Root et al., 2014). Along these lines, we proposed that the underlying mechanism of ␣ 2 ␦-2-induced mismatched synapses could be explained by a compensatory upregulation of postsynaptic GABA A Rs due to increased excitatory transmission upon ␣ 2 ␦ overexpression (Zhou and Luo, 2015;Chen et al., 2018). However, several of our observations argue against this theory. First, using distinct experimental paradigms, we show here that ␣ 2 ␦-1 and ␣ 2 ␦-3, which both enhance excitatory synaptic transmission and increase presynaptic release probability (Hoppa et al., 2012), do not induce postsynaptic clustering of synaptic GABA A R subtypes. Second, the effect of ␣ 2 ␦-2 is independent of the presynaptic neurotransmitter identity, thus providing compelling evidence for a trans-synaptic function. Third, solely ␣ 2 ␦-2 splice variants lacking exon 23 specifically induced the recruitment of postsynaptic GABA A Rs opposite glutamatergic nerve terminals. Fourth, the effect of ␣ 2 ␦-2 is restricted to transfected synapses, emphasizing a synapse-specific role rather than a global alteration of GABAergic postsynaptic components.
Evidence for an ␣ 2 ␦-2-mediated transsynaptic signaling cascade To date, surprisingly little is known on the specific signals imperative for the proper alignment of inhibitory synapses and the targeting of GABA A Rs and gephyrin to the postsynaptic compartment. In vitro cultured neurons provide an excellent and reduced model system allowing us to address basic cell biological mechanisms. Most importantly, they are less prone to potential compensatory mechanisms compared with the in vivo situation (Dong et al., 2007;Brown et al., 2016;Chen et al., 2017). Likewise, overexpression of proteins in their native environment provides a powerful experimental tool but needs to be carefully discussed, particularly in relation to the specific physiological roles. Therefore, it is conceivable that an aberrantly high accumulation of presynaptic ␣ 2 ␦-2 proteins may induce artificial synaptic connections. Nevertheless, our study provides several lines of evidence supporting a presynaptic role of ␣ 2 ␦-2 in the trans-synaptic alignment of synapses and receptor clustering. First, we observed that presynaptic expression of a single splice variant of ␣ 2 ␦-2 trans-synaptically regulates postsynaptic GABA A R abundance. Second, overexpression of ␣ 2 ␦-2 specifically induces an aberrant wiring of glutamatergic axons to postsynaptic GABAergic sites independently of the postsynaptic neuron type (Fig. 11). Third, whereas mismatched synapses lack glutamate receptors, they contain different synaptic GABA A R subtypes and gephyrin. Importantly, the strongly reduced GLUR1 expression (Fig. 2) in the mismatched and aberrantly wired synapses is a secondary effect because no AMPARs are to be expected in postsynaptic GABAergic synapses (Fig. 5C, model). Fourth, gephyrin abundance is higher in ␣ 2 ␦-2-induced mismatched synapses compared with untransfected endogenous GABAergic synapses. Considering that gephyrin immobilizes synaptic GABA A R clusters in hippocampal neurons Figure 14. Splicing sites of ␣ 2 ␦-2 and implications on secondary structure. A, Using homology modeling, we tested the potential consequences of alternative splicing on the structure prediction of three distinct ␣ 2 ␦-2 splice variants (v1-v3) based on the high-resolution structure of ␣ 2 ␦-1 (PDB code: 5GJV; Wu et al., 2016). Although some of the longer loops generally seemed to be quite flexible and differently orientated in all three models (gray arrows), alternative splicing of exons 23, 30, and 38 resulted in differences between secondary structure elements in distinct variants (boxed regions in B-D). B, Higher magnification showing that the inclusion of exon 23 in ␣ 2 ␦-2-v2 suggests the formation of an extra loop (arrowhead and green selection in overlay), leading to the disruption of an ␣-helix present in ␣ 2 ␦-2-v1 and ␣ 2 ␦-2-v3 (see also Fig. 13). C, Structure modeling further proposes that the insertion of three base pairs in exon 30 of ␣ 2 ␦-2-v3 causes the formation of a short ␣-helix in the depicted loop, which is absent in ␣ 2 ␦-2-v1 and ␣ 2 ␦-2-v2. D, Moreover, models implicate that alternative splicing of exon 38 affects the length of a ␤ sheet at this position, which is longer in ␣ 2 ␦-2-v1 and ␣ 2 ␦-2-v2. (Jacob et al., 2005), we provide strong evidence that gephyrin recruitment occurs secondary to synaptic GABA A R clustering.
However, it is still unknown whether the interaction with postsynaptic receptors is mediated by a direct trans-synaptic action of ␣ 2 ␦-2 or if it may be indirectly mediated by the interaction with other primary synaptic organizers (see also discussion in Fell et al., 2016). Several studies implicate the trans-synaptic cell adhesion molecules ␣-Nrxns as the most interesting candidates for mediating a possible indirect interaction between ␣ 2 ␦-2 and GABA A Rs (Missler et al., 2003;Kang et al., 2008;Born et al., 2014;Tong et al., 2017). Here, we demonstrate that presynaptic ␣ 2 ␦-2 also potently induces mismatched synapse formation in the absence of ␣-Nrxns. Although this finding supports an ␣ 2 ␦-2mediated trans-synaptic effect and possibly a direct transsynaptic interaction of ␣ 2 ␦-2 with postsynaptic GABA A Rs, it does not exclude the potential involvement of other cell adhesion molecules not addressed in our study (Graf et al., 2004;Lardi-Studler and Fritschy, 2007).
The effect of overexpressed ␣ 2 ␦-2 was more pronounced in ␣-Nrxn TKO synapses compared with WT control synapses. We observed that the baseline expression of postsynaptic GABA A Rs in TKO was already significantly increased, which shows that no overexpression of ␣ 2 ␦-2 is needed for inducing postsynaptic GABA A R clustering. This finding may indicate that the presence of ␣-Nrxns negatively influences GABA A R recruitment by ␣ 2 ␦-2. Alternatively, the loss of ␣-Nrxns could be either overproportionally compensated by the still present ␤-neurexins (Missler et al., 2003) or shift the modulation of postsynaptic receptor abundance to the neuroligin-2/Slitrk3 or the neuroligin-2/MDGA1 complexes (Pettem et al., 2013;Li et al., 2017). Independently of the detailed underlying mechanism, our findings support a cooperative action of ␣ 2 ␦ subunits and neurexins in fine-tuning synaptic functions (Tong et al., 2017;Brockhaus et al., 2018).

Future implications and potential relevance to neurological disorders
One important finding presented here is that only ␣ 2 ␦-2 splice variants lacking exon 23 induced a mismatched localization of postsynaptic GABA A Rs. This is interesting for several reasons. First, the presynaptic effect of ␣ 2 ␦-2 lacking a single spliced exon is highly specific and thus argues against a simple unspecific effect caused by homologous overexpression of ␣ 2 ␦ isoforms in our experimental paradigm. Second, in light of discussing potential interaction mechanisms (see above), it is tempting to speculate that ␣ 2 ␦-2 interacts with its partners via the region surrounding the lacking exon 23, which will be tested in future experiments. Third, our findings show that the trans-synaptic function of ␣ 2 ␦-2 may be independent of the calcium channel complex. This is further supported by a previous study showing that heterologous coexpression of different ␣ 2 ␦-2 splice variants with various ␣ 1 subunits caused similar effects on calcium channel current densities and activation/inactivation kinetics (Hobom et al., 2000). Fourth, and most importantly, alternative splicing has also been described for mouse and human genes encoding for ␣ 2 ␦-1 and ␣ 2 ␦-4 (Lana et al., 2014;Bacchi et al., 2015). This is particularly interesting considering the previously identified associations between ␣ 2 ␦ subunits with neurological disorders (Chioza et al., 2009;Edvardson et al., 2013;Pippucci et al., 2013;Vergult et al., 2015;Butler et al., 2018;Valence et al., 2018). Two distinct splice site mutations in CACNA2D2, for example, lead to the truncation of the protein, causing epilepsy, dyskinesia, and cerebellar atrophy in one patient (Pippucci et al., 2013), whereas the other mutation is associated with congenital ataxia (Valence et al., 2018). These clinically observed phenotypes are similar to the previously described phenotype of Cacna2d2 mutant or null mouse models (Barclay et al., 2001;Brodbeck et al., 2002;Ivanov et al., 2004) and may be caused by the loss of functional GABAergic synapses and the consequential increased excitatory activity. Moreover, ␣ 2 ␦ subunits are potential risk genes for autism spectrum disorders (Iossifov et al., 2012;De Rubeis et al., 2014) and schizophrenia (Purcell et al., 2014;Moons et al., 2016), diseases classically associated with altered neuronal connectivity (Belmonte et al., 2004;Fornito and Bullmore, 2015). The findings presented here provide a possible mechanistic explanation for how altered ␣ 2 ␦ subunit expression may be linked to neurological disorders. However, for the future understanding of the precise underlying disease mechanisms, it will be necessary to reveal whether all ␣ 2 ␦ subunit isoforms can act as similarly specific trans-synaptic organizers in distinct synapses. Finally, our findings allow hypothesizing that presynaptic ␣ 2 ␦ subunits are critical determinants of synapse specificity and connectivity. Therefore, alterations in the balance of presynaptic ␣ 2 ␦ isoforms may underlie synaptic plasticity and, conversely, pathological axonal wiring observed in neurological disorders.