Auxiliary α2δ1 and α2δ3 Subunits of Calcium Channels Drive Excitatory and Inhibitory Neuronal Network Development

VGCCs are multisubunit complexes that play a crucial role in neuronal signaling. Auxiliary α2δ subunits of VGCCs modulate trafficking and biophysical properties of the pore-forming α1 subunit and trigger excitatory synaptogenesis. Alterations in the expression level of α2δ subunits were implicated in several syndromes and diseases, including chronic neuropathic pain, autism, and epilepsy. However, the contribution of distinct α2δ subunits to excitatory/inhibitory imbalance and aberrant network connectivity characteristic for these pathologic conditions remains unclear. Here, we show that α2δ1 overexpression enhances spontaneous neuronal network activity in developing and mature cultures of hippocampal neurons. In contrast, overexpression, but not downregulation, of α2δ3 enhances neuronal firing in immature cultures, whereas later in development it suppresses neuronal activity. We found that α2δ1 overexpression increases excitatory synaptic density and selectively enhances presynaptic glutamate release, which is impaired on α2δ1 knockdown. Overexpression of α2δ3 increases the excitatory synaptic density as well but also facilitates spontaneous GABA release and triggers an increase in the density of inhibitory synapses, which is accompanied by enhanced axonaloutgrowth in immature interneurons. Together, our findings demonstrate that α2δ1 and α2δ3 subunits play distinct but complementary roles in driving formation of structural and functional network connectivity during early development. An alteration in α2δ surface expression during critical developmental windows can therefore play a causal role and have a profound impact on the excitatory-to-inhibitory balance and network connectivity. SIGNIFICANCE STATEMENT The computational capacity of neuronal networks is determined by their connectivity. Chemical synapses are the main interface for transfer of information between individual neurons. The initial formation of network connectivity requires spontaneous electrical activity and the calcium channel-mediated signaling. We found that, in early development, auxiliary α2δ3 subunits of calcium channels foster presynaptic release of GABA, trigger formation of inhibitory synapses, and promote axonal outgrowth in inhibitory interneurons. In contrast, later in development, α2δ1 subunits promote the glutamatergic neurotransmission and synaptogenesis, as well as strongly enhance neuronal network activity. We propose that formation of connectivity in neuronal networks is associated with a concerted interplay of α2δ1 and α2δ3 subunits of calcium channels.


Introduction
The transfer and processing of information in neuronal networks critically depend on structural and functional connections between neurons. Network connectivity is not static but evolves over time and reflects both genetically predetermined factors and the previously processed stimuli. The initial circuitry formation occurs during early development and is associated with the emergence of synaptic contacts, which serve as substrate for functional network interaction. During early development, spontaneous neuronal activity involving transient changes in intracellular calcium is necessary and sufficient for neuronal development, and powerfully drives the establishment of connectivity maps (Ben-Ari, 2001;Spitzer, 2006).
VGCCs (Ca V s) on presynaptic boutons play a crucial role in synaptic transmission by mediating the electrochemical conversion of electrical activity into vesicle release. VGCCs are multiunit complexes that consist of a mandatory pore-forming a1 subunit and auxiliary a2d and b subunits (Catterall, 2000;Arikkath and Campbell, 2003;Zamponi et al., 2015). In mammalian synapses, activation of mainly P/Q-type (Ca V 2.1) and N-type (Ca V 2.2) VGCCs on membrane depolarization results in rapid presynaptic calcium influx that triggers neurotransmitter release (Wheeler et al., 1994;Scholz and Miller, 1995;Cao and Tsien, 2010). Four a2d isoforms (a2d1-4) encoded by CACNA2D1-CACNA2D4 genes have been identified, with a2d1 and a2d3 being particularly abundant in the cerebral cortex and hippocampus (Klugbauer et al., 1999;Cole et al., 2005;Schlick et al., 2010). Expression of the b and a2d subunits increases the trafficking of the channel and modulates its biophysical properties at the surface (Arikkath and Campbell, 2003;Dolphin, 2012). For example, overexpression of a2d subunits triggers synaptic recruitment of VGCCs, enlargement of the presynaptic terminals, and facilitation of presynaptic release (Hoppa et al., 2012;Schneider et al., 2015), whereas downregulation of a2d subunits decreases the surface expression of a1 subunit and leads to the reduction of presynaptic structures and glutamate release (Dickman et al., 2008;Kurshan et al., 2009;Cordeira et al., 2014). Additionally, a2d1 and a2d3 subunits were shown to promote excitatory synaptogenesis in mammalian brain (Eroglu et al., 2009) and in Drosophila (Dickman et al., 2008;Kurshan et al., 2009), respectively.
Altered expression of a2d subunits has been implicated in the pathogenesis of several syndromes and diseases (Geisler et al., 2015;Zamponi et al., 2015). In particular, postinjury overexpression of a2d1 in sensory neurons is associated with hyperalgesia and chronic neuropathic pain and underlies the antiallodynic efficacy of gabapentinoids (Luo et al., 2001;Bauer et al., 2009;Patel et al., 2013). Null mutation of CACNA2D2 leads to global developmental delay, absence epilepsy, and cerebellar ataxia in mice (Barclay et al., 2001) and humans (Edvardson et al., 2013;Pippucci et al., 2013). Symptomatic convulsive epilepsy and intellectual disability were also reported in humans with aberration of the CACNA2D1 gene (Vergult et al., 2015). Furthermore, analyses of gene-disrupting mutations in individuals with autism highlighted CACNA2D3 among autism susceptibility genes (Iossifov et al., 2012;De Rubeis et al., 2014). Autism is a pervasive neurodevelopmental disorder diagnosed early in childhood and associated with aberrant brain connectivity (Folstein and Rosen-Sheidley, 2001;Freitag, 2007). Remarkably, autistic spectrum disorders are accompanied by epilepsy in up to 38% of affected individuals, which represents manifold higher incidence of epilepsy compared with the population average (Tuchman and Rapin, 2002;Levisohn, 2007).
Thus, converging lines of evidence suggest that a2d subunits are involved in the establishment and/or modulation of the excitation/inhibition ratio, but little is known about the mechanisms and the contribution of individual a2d isoforms to network connectivity and activity of central neurons. Therefore, in this study, we used acute upregulation and downregulation of the a2d subunits to dissect their impact on the formation of structural and functional connectivity, as well as on the balance between excitation and inhibition.

Ethics statement
All experimental procedures were conducted in accordance with the EU Council Directive 86/609/EEC and were approved and authorized by the local Committee for Ethics and Animal Research (Landesverwaltungsamt Halle, Germany).
Breeding and genotyping of mutant mice Animal procedures for control and a2d1 KO (a 2 d 1 À/À ) mice having a mixed 129J Â C57BL/6 background 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 #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 involving WT mice were performed in accordance with government regulations for animal welfare and approved by the Landesamt für Natur, Umwelt und Verbraucherschutz (license #84-02.05.20.11.209 and #84-02.04.2015. A423). Mice were maintained at central animal facilities in Innsbruck and Münster under standard housing conditions with food and water ad libitum at a 12 h light/dark cycle. The a2d1 À/À mutant mouse strain was previously generated and characterized (Fuller-Bicer et al., 2009;Patel et al., 2013;Mastrolia et al., 2017). Genotyping for the Cacna2d1 gene was done as published previously (Fuller-Bicer et al., 2009) with some modifications by use of standard PCR conditions (annealing at 52°C for 30 s). Primers: WT-F1: 59-GAGCTTTCTTTCTTCTGATTCCAC-39, mutant-F2: 59-CTGCACGAGACTAGTGAGACG-39, R: 59-ACATTCTCAAG ACTGTAGGCAGAG-39. Expected band sizes were 346 bp for WT (a2d1 1/1 ) and 635bp for KO (a2d1 À/À ) animals, respectively, and heterozygous mice showed both bands.
Transmission electron microscopy Brain tissue from WT control and a2d1 À/À mice was embedded in epon resin (Electron Microscopy Science). For embedding, anesthetized adult male mice were transcardially perfused with 25 ml of 2% glutaraldehyde (Roth) and 2% PFA (Merck) in 0.1 M PB at 37°C, and postfixed at 4°C overnight. Blocks of hippocampal tissue were contrasted in 1% osmium tetroxide for 2 h at room temperature. Following washes with distilled water and dehydrating, tissue was incubated with propylene oxide (Electron Microscopy Science) for 45 min, infiltrated with propylene oxide/ epon (1:1) for 1 h, in pure epon overnight, and hardened at 60°C for 24 h. Additional contrasting of thin sections from brains was done on Formvar-coated copper grids with a saturated solution of 12% uranyl acetate and lead citrate.
For better comparability with imaging and electrophysiological results, samples containing the stratum radiatum of the hippocampal CA1 region were investigated. Ultrastructural analysis was done with a transmission electron microscope (Libra 120, Carl Zeiss) at 80 kV, and images taken with a CCD camera (Tröndle). For quantifying the density of asymmetric synapses, tissue areas were reconstructed from panorama pictures (each composed of 9 individual images = 210 mm 2 ), and three panoramas were analyzed per genotype (n = 3 panoramas from 3 animals per genotype = 1890 mm 2 ). Asymmetric (Type 1) synapses were defined as contacts with a visible synaptic cleft, a distinct postsynaptic density, and at least three synaptic vesicles.
Cloning of lentiviral a2d ::HA overexpression constructs. For immunoreactive detection, a2d subunits were N-terminally labeled with a double hemagglutinin (HA)-tag. The extracellularly double HA-labeled (between aa 27 and 28) rabbit a2d1 construct was kindly provided by G.J.O. (Medical University Innsbruck). For the a2d3, the double HA-tag was inserted between aa 36 and 37 of mouse CACNA2D3 (provided by Prof. Norbert Klugbauer, Albert-Ludwigs-University Freiburg) (see Klugbauer et al., 1999) via a synthesized DNA fragment using the KpnI and BsrGI restriction sites. For cloning of lentiviral transfer plasmids for a2d overexpression, a pLenti vector of the third generation equipped with a neuron-specific synapsin promotor was used as backbone (pLenti-Synapsin-hChR2(H134R)-EYFP-WPRE; Addgene; plasmid #20945). The hChR2 insert was cut from this vector via the unique sites AgeI and BsrGI, and sticky ends were used for insert integration or filled up to blunt ends using Klenow Fragment (Thermo Fisher Scientific). The a2d1-2HA was enzymatically digested via the unique restriction sites NotI and SalI, filled up to blunt ends, and ligated into the lentiviral transfer vector. The a2d3-2HA was amplified via PCR and equipped with the unique restriction sites BsiWI (generating BsrGI overhang) and AgeI allowing sticky-end ligation into the target vector. Correct integration was determined by qualitative digestion and partial sequencing.
Constructs for shRNA-mediated a2d knockdown. For knockdown of the a2d1 subunit, siRNA target sequences corresponding to the a2d1 coding region (CACNA2D1, GenBank accession number NM_009784.2) (Obermair et al., 2005) were selected and tested for efficient knockdown. The siRNA was expressed as shRNA under the control of a U6 promoter (derived from the pSilencer1.0-U6 siRNA expression vector, Ambion) cloned into the pb A-eGFP plasmid . For lentiviral expression, a2d1 shRNA was cloned into pHR as previously described (Subramanyam et al., 2009). For knockdown of the a2d3 subunit, four 29mer shRNA constructs against rat Cacna2d3 (Gene ID 306243) cloned in lentiviral GFP vector (pGFP-C-shLenti Vector, catalog #TR30023) were ordered from OriGene Technologies (catalog #TL713428). Based on their specificity for rat and mouse a2d3, two of these constructs were tested for their knockdown efficiency where the construct "C" was evaluated to results in a reduction of a2d3 expression down to 40%-50%. As control for a2d -knockdown experiments, a noneffective 29-mer scrambled shRNA cassette cloned into pGFP-C-shLenti vector (catalog #TR30021; OriGene Technologies) was used.
Evaluation of shRNA-mediated a2d3 knockdown. The knockdown efficiency of the a2d3 shRNA constructs was tested on both the expression of HA-tagged a2d3 (transfected into HEK293T cells and rat hippocampal cultures) and the endogenous expression level of a2d3 in rat hippocampal cultures. Expression levels of HA-tagged a2d3 subunits were quantified via anti-HA immunostaining and Western blotting (rat anti-HA, 1:1000, Roche, catalog #11867423001, clone 3F10) or monoclonal mouse anti-HA-tag (1:1000; OriGene Technologies, catalog #TA180128) and polyclonal anti-HA-tag (1:1000; Synaptic Systems, catalog #245003), respectively. Furthermore, HA-tagged a2d3 subunits were used to evaluate the correct endogenous a2d3 bands targeted in Western blotting via polyclonal rabbit anti CACNA2D3 (1:1000; Thermo Fisher Scientific, catalog #PA5-87 802). Protocols used for immunocytochemical staining and Western blotting are described below.
Preparation of cell lysates. For the validation of a2d antibodies, HEK293T cells were transfected with HA-tagged a2d1 and a2d3 variants. Transfected cells as well as nontransfected HEK293T cells were processed for Western blot analysis 48 h after transfection. Cells were washed with ice-cold 1 Â PBS for 2 times, scraped off, collected, and centrifuged at 800 rpm for 10 min. Afterward, cells were lysed with lysis buffer (125 mM sodium chloride, 0.1% [w/v] SDS, 0.01% [v/v] Triton X-100, 50 mM Tris/HCl, pH 7.5) containing a protease inhibitor cocktail (cOmplete ULTRA Tablets, Sigma Millipore, catalog #05892791001, Roche). Lysates were cleared by centrifugation at 15,000 rpm for 15 min at 4°C and incubated for 10 min at room temperature with 4 Â loading buffer (40% [v/v] glycerol, 240 mM Tris/HCl, pH 6.
. Primary neurons were infected with the overexpression or knockdown constructs 7 d before harvesting. In general, cells were harvested at DIV21-DIV28, except the a2d 3 knockdown condition was harvested at DIV11-DIV12 where the a2d 3 expression was found to be most prominent. For sample collection, cells were washed with prewarmed 1Â PBS and directly lysed using 2Â sample buffer (1Â Tris/HCl, pH 6; 8. 4Â Tris/HCl, 500 mM Tris, containing a protease inhibitor cocktail (cOmplete ULTRA Tablets, Sigma Millipore, catalog #05892791001, Roche). Cells were then scraped and the lysate was pipetted up and down (at least 5 times) through a 30 G cannula. The lysate was then incubated for 1 h at 37°C and briefly spun down before gel loading.
For image acquisition, z stacks were acquired for 20 planes at 200 nm steps, using a spinning disk confocal microscopy system (Andor Technology) controlled by Andor iQ2 software. The microscope (BX51WI Olympus) was equipped with a CSU-X1 spinning disk (Yokogawa), an EMCCD camera (iXon1 897, Andor Technology), and 60 Â, NA 1.4 oil objective (Olympus) for synaptic density analysis or using a 20 Â, 0.8 NA oil objective (Olympus) to investigate axonal outgrowth.
Lentivirus production. For production of lentiviral particles, human embryonic kidney cells (HEK293T) cells were used for packaging and maintained in DMEM (Thermo Fisher Scientific) supplemented with 10% FCS (Thermo Fisher Scientific), 1% glutamine (Invitrogen), and 1 Â antibiotic-antimycotic (Invitrogen) at 37°C in a humidified atmosphere with 5% CO 2 and 95% air. The 30%-40% confluent HEK293T cells were triple-transfected with the second-generation helper plasmids: psPAX2 (Addgene, plasmid #12 260) and pVSV-G (Addgene, plasmid #8454), as well as the target gene-containing transfer vector in a molar ratio of 1:1:2. For the transfection of a 175 cm 2 cell culture flask, 80 mg of total DNA was pipetted to 1 ml solution A (500 mM calcium chloride) and mixed. Subsequently, 1 ml of solution B (140 mM sodium chloride, 50 mM HEPES, 1.5 mM disodium hydrogen phosphate, pH 7.05) was added. The mixture was incubated for 2 min at room temperature and was then pipetted to the medium of the cells for overnight incubation. The next day, the medium was replaced by DMEM supplemented with only 4% FCS only, 1% glutamine, and 1 Â antibiotic-antimycotic. On the following 2 d, the media was harvested, centrifuged at 2000 Â g for 5 min, and the supernatant was stored at 4°C. Both harvests were pooled, filtered through a 0.45 mm filter, and centrifuged at 20,000 rpm for 2 h at 4°C. Afterward, the supernatant was removed and the pellet resuspended in DMEM (supplemented with 10% FCS, 1% glutamine, and 1 Â antibiotic-antimycotic) on a shaker at 300 rpm and room temperature for 1 h.
Lentivirus titration. The working dilution of virus suspension was determined by test infection of dissociated EXVIII-EXIX rat cortical cultures seeded on coverslips and incubated in Neurobasal medium (Thermo Fisher Scientific) at 37°C in humidified atmosphere with 5% CO 2 and 95% air. The cortical cultures were infected at DIV2 with dilutions of the viral particles from 1:50 to 1:5000, and incubated overnight. On the following day, the medium containing the virus was exchanged with the conditioned Neurobasal stored before infection. At DIV11, the cells were stained for a2d expression via the HA-tag (rat anti-HA; Sigma Millipore, catalog #11867423001), cortical glial cells using anti-GFAP antibody (rabbit anti-GFAP; Synaptic Systems, catalog #173002), and total cell number via DAPI staining (0.5 mg/ml; Sigma Millipore, catalog #D9542). For this purpose, the cells were fixed for 5 min with 4% PFA (preheated to 37°C) and subsequently permeabilized with 0.3% Triton-X/PBS for 2 min at room temperature. Then, cells were washed three times for 10 min at room temperature using washing buffer (1Â PBS; 2% BSA; 25 mM glycine) and incubated with the primary antibodies mentioned above at concentration of 1:1000 and DAPI 1:200 for 1 h at room temperature. Afterward, three washing steps were done, followed by the incubated with the secondary antibodies (1:1000): anti rat-Alexa-488 (Thermo Fisher Scientific, catalog #A11006) and antirabbit-Cy5 (Dianova, 111-175-144) for 1 h at room temperature in the dark. After three final washings, the coverslips were mounted with Mowiol (9.6 g Mowiol; 24 ml H 2 O; 24 g glycerol; 48 ml 0.2 M Tris/HCl, pH 8.5). Images were acquired with an Axio Imager.A2 microscope (Carl Zeiss) equipped with a CoolSNAP MYO CCD camera (Photometrics) and a 20Â Plan-Apochromat oil objective (NA = 1.40, Carl Zeiss) using the VisiView (Visitron Systems) software. Images were acquired as stacks of 10 frames that were subsequently averaged and used for quantification in ImageJ (National Institutes of Health).
Heterologous expression of calcium channel subunits. Transient expression of tagged VGCCs in HEK293T was achieved by cotransfection of constructs for tagged a1 subunits together with b 3-and a2d1/3-encoding constructs at a 1:1:1 ratio using the FuGENE X-tremeGENE 9 DNA transfection reagent (Roche) according to the manufacturer's protocol. Transiently transfected cells were measured 48-72 h after transfection. Current amplitudes .1 nA were considered to result from successful cotransfection of all three subunits (a1, b 3, and a2d1/3) as confirmed further by simultaneous detection of the GFP-tag fused to b 3 and the extracellular HA epitope in a2d1/3::HA (data not shown).
Preparation, transfection, and infection of dissociated neuronal cultures. Dissociated hippocampal cultures were prepared from Wistar rat (Charles River; RRID:RGD_8553003) and glutamic acid decarboxylase 67 (GAD67)::GFP mouse embryos (EXVIII) as described previously (Kaech and Banker, 2006). Briefly, cell suspensions obtained after dissociation with trypsin were plated onto poly-L-lysine-coated 18 mm glass coverslips (Menzel-Glaeser) at a density of 30,000 cells per coverslip. After 1-2 h in DMEM plus FBS at 37°C, five coverslips were transferred into a 35 mm dish containing a 70%-80% confluent monolayer of astrocytes in Neurobasal medium supplemented with B27 and 5 mM glutamine. Cultures were incubated at 37°C in humidified atmosphere with 5% CO 2 and 95% air. At DIV3, AraC was added to the cells to a final concentration of 1.4 mM.
For multichannel recordings, suspension of dissociated hippocampal cells (750,000 cells/ml) was plated on poly-D-lysine-coated 60-electrode microelectrode arrays (MEAs) with interelectrode distance 200 mm (MultiChannel Systems). After plating, all cultures were incubated at 37°C in humidified atmosphere (95% air and 5% CO 2 ) in serum-free Neurobasal medium (Thermo Fisher Scientific). Throughout the lifespan, cultures were covered by semipermeable membranes (ALA-MEM, MultiChannel Systems) to avoid evaporation of the medium, which was partially replaced on a weekly basis.
Throughout the study, two infection protocols were applied. The first protocol was used to dissect the effects of a2d subunits on the network activity during development and involved infection at distinct developmental stages (after first, second or third week in vitro) followed by recording of spontaneous activity 1 or 2 weeks later as specified in the text. The second protocol was used for analysis of long-term structural and functional consequences of upregulation or downregulation of a2d subunits, as well as of lentiviral GFP expression. For this purpose, infection was performed during first week in vitro at DIV2-DIV4, and the data were acquired within the period of DIV7 to DIV24. Given the data on infection rate at different virus dilutions (data not shown), viral constructs were diluted in conditioned cultured media used for infections in the ratio 1:1000.
Transfection of neurons was conducted at DIV3-DIV4 using the calcium phosphate method. Before transfection, cells were placed in a 12-well dish with 1 ml 37°C Optimem media (Thermo Fisher Scientific). To prepare the precipitate, 150 ml of transfection buffer (in mM as follows: 274 NaCl, 10 KCl, 1.4 Na 2 HPO 4 , 15 glucose, 42 HEPES, pH 7.04-7.1) was added dropwise to a solution containing 5 mg of DNA and 200 mM CaCl 2 , under gentle stirring. The resulting mix was placed for 15 min at room temperature; 60 ml of the mix was added per well, and neurons were placed in the incubator for 30-60 min. Medium was exchanged for 2 ml 37°C prewarmed Neurobasal medium, followed by 2 times exchanging 1.5 ml. After this procedure, cells were finally placed back in the stored dishes in conditioned culture media.
Compounds and treatments. To test the contribution of Ca V 2.2 and Ca V 2.1 channels to mEPSCs and mIPSCs, specific calcium channel blockers v -agatoxin IVA (200 nM) or v -conotoxin GVIA (1 mM) (both from Alomone Labs), respectively, were applied to the bath solution. The changes of mPSCs frequency were analyzed ;7 min after the toxin application. Contribution of high voltage-activated calcium channels to spontaneous neurotransmitter release was estimated by application of 100 mM CdCl 2 .
Synaptic density analysis. The analysis of synaptic density was conducted using custom-written routines for ImageJ software (National Institutes of Health). In rat hippocampal cultures infected with pLenti-syn-a2d1::HA or pLenti-syn-a2d3::HA, as well as in noninfected sister controls, immunolabeling of Bassoon and either Homer1 or Gephyrin was conducted for identification of presynaptic and postsynaptic sites of excitatory or inhibitory synapses, respectively. For each individual scan, puncta with the mean fluorescence exceeding arbitrary threshold value (2 SDs computed across the FOV) were detected and stored as sets of ROIs corresponding to individual presynaptic or postsynaptic compartments. Next, a segmented line was drawn by a trained user along individual dendrites from soma to the most distal point that could be reliably detected. Particularly in rather mature cultures with extensive dendritic branching, identification of individual dendrites was aided by additional MAP2 immunolabeling. The selected dendritic ROIs were straightened and the number of colocalized presynaptic and postsynaptic puncta per micrometer was calculated. For each preparation, the data obtained in a2d -overexpressing cultures were further normalized to the mean value obtained in control sister cultures (taken as 100%).
Axonal outgrowth analysis. Axonal outgrowth was investigated either in WT rat hippocampal neurons transfected with the volume marker GFP or using hippocampal cultures of the GAD67-GFP mouse line (kindly provided by Prof. O. Stork, Otto-von-Guericke University of Magdeburg) to specifically examine GABAergic interneurons. In both cases, the GFP signal was enhanced via anti-GFP staining (Thermo Fisher Scientific, catalog #A6455). Neuronal dendrites were labeled using anti-MAP2 (Synaptic Systems, catalog #188004). The axons of GFPpositive neurons were identified based on the morphology of the neurites and negativity for MAP2 immunofluorescence.
In order to acquire the whole axons, 3-9 scans per neuron were taken. The analysis was done for each single image; values of images from the same cell were then integrated to obtain the total length and total number of branches per given axon. The measurement of axonal length and branching was performed by reconstructing the axons of acquired cells by using the Fiji plugin Simple Neurite Tracer (Longair et al., 2011). The recording and analysis were conducted by a trained person in a blinded manner to exclude the bias in estimate of different conditions. For each preparation, the data obtained in a2d -overexpressing cultures were further normalized to the mean value obtained in control sister cultures.
Whole-cell electrophysiological recordings. Recordings of recombinant VGCCs have been described previously (Brockhaus et al., 2018). In brief, patch-clamp recordings from transfected tsA-201 cells were done 3-5 d after plating. The bath solution (32°C) contained the following (in mm): 115 NaCl, 3 CaCl 2 , 1 MgCl 2 , 10 HEPES, glucose, 20 TEA-Cl, pH 7.4 (300 6 5 mOsm/ kg osmolality). Patch pipettes with a resistance of 2-4 MV when filled with pipette solution containing the following (in mm): 125 Cs-methane sulfonate, 20 TEA-Cl, 5 EGTA, 2 MgCl 2 , 10 HEPES, 4 Na 2 -ATP, 0.5 Na-GTP, pH 7.4 (285 6 5 mOsm/kg osmolality). Whole-cell calcium currents were recorded with an EPC 10 USB Double patch-clamp amplifier and Patchmaster software (HEKA Elektronik). Signals were filtered at 3 kHz and digitized at 10 kHz. Cells were held at À80 mV in whole-cell configuration, series resistance, and membrane capacitance determined and compensated online. Leak currents were subtracted online using a P/5 protocol. Recordings for each condition were done on cells from at least three independent experiments. Current-voltage relationships were obtained by 50 ms voltage pulses from a holding potential of À80 mV to voltages between À40 mV and 70 mV in 10 mV increments with 6 s intervals. Current densities were calculated from currents normalized to whole-cell capacitance. Steady-state inactivation properties were measured by evoking currents with a 500 ms test pulse to 20 mV after 2 s voltage displacement (prepulse) from 20 mV to À80 mV in 10 mV increments (for further details, see Brockhaus et al., 2018).
Somatic whole-cell voltage-clamp recordings of spontaneous mEPSCs and mIPSCs were performed between DIV7 and DIV21 in cultured rat hippocampal neurons. Primary hippocampal cultures were constantly perfused with extracellular solution containing the following (in mM): 145 NaCl, 2.5 KCl, 2 MgCl 2 , 2 CaCl 2 , 10 HEPES, and 10 D-glucose (pH 7.4 adjusted with NaOH), supplemented with 0.1 mM TTX, 5 mM AP5, and 5 mM bicuculline (to record mEPSC) and 0.1 mM TTX, 5 mM AP5, and 10 mM DNQX (to record mIPSCs). Patch pipettes from borosilicate glass had a pipette resistance of 2-4 MV when filled with the intracellular solution of the following composition (in mM): 130 KCl, 2 MgCl 2 , 0.5 CaCl 2 , 1 EGTA, 40 HEPES (pH 7.25 adjusted with KOH). Before experiments, 1 mM ATP and 0.1 mM GTP were added, and pH was readjusted to 7.2-7.3 with KOH. Only patches with a series resistances ,15 MV were analyzed. In all recordings, the membrane potential was clamped at À70 mV.
Individual mEPSCs and mIPSCs were detected using a peak detection algorithm of MiniAnalysis 6.0 software (Synaptosoft), which measured the peak amplitude, as well as rise and decay times. Amplitude threshold values were set at 3 times the root mean square of the baseline noise amplitude. All detected events were visually inspected and verified by a trained experimenter. The events were collected after 1-2 min after commencement of recording when the frequency of miniature currents was stabilized.
Recording and analysis of neuronal network activity. The neuronal network activity in high-density rat hippocampal cultures grown on MEAs was sampled at 10 kHz using MEA1060INV-BC system (MultiChannel Systems) at 37°C in a humidified atmosphere with 95% air and 5% CO 2 . The analysis was conducted using Spike2 software (Cambridge Electronic Design) on 10-min-long intervals for each culture at each time point. The threshold-based (67 SDs of spike-free noise) detection of spikes in high pass-filtered records (gain 300 Hz) was followed by identification of bursts (!5 spikes with interspike interval 100 ms). Channels with the mean firing rate lower than arbitrary minimum (0.01 spike/s) were considered as nonspiking in given session and discarded from further analyses. The mean firing rate was calculated separately for each active channel (electrode) in each individual culture. Network burst (NB) analysis was conducted as described previously (Bikbaev et al., 2015). NB was defined as a non-zero period of correlated (synchronous) bursting in two or more channels. For each NB, participating channels were ranked according to their temporal order of recruitment into given NB, forming vector (1, ..., n), where n denotes the rank of the last recruited channel (i.e., the size of given NB; n ! 2). The mean burst onset lag reflecting the synchronicity of bursting onset in remote network locations was calculated for each NB as DT on ¼ , where T i denotes the burst onset time in channel with rank i within given NB.

Statistics
To avoid potential bias of results, neuronal cultures grown on coverslips or MEAs were generally randomized before treatments. Additionally, experimental procedures and treatments, as well as separate experimental routines (acquisition, analysis, and interpretation) were conducted in a blind manner by different researchers where possible. The statistical effects of experimental treatments on analyzed parameters were evaluated using protected parametric and nonparametric (Kruskal-Wallis) ANOVA followed by post hoc tests as specified in the text. Pairwise comparisons were conducted using Student's t test or Mann-Whitney U test. Treatment of data and statistical analysis were performed using Prism software (GraphPad) and Statistica data analysis system (Statsoft). Factorial effects and differences were considered significant at p , 0.05. Data are presented as mean 6 SEM.

Results
Constitutive KO of the a2d1 subunit in vivo leads to reduction of excitatory synaptic density Since a2d1 and a2d3 subunits are both abundant in the hippocampus in vivo and in cultured neurons (Klugbauer et al., 1999;Cole et al., 2005;Schlick et al., 2010), we chose to characterize their functional effects on network activity and connectivity in hippocampal neurons as a standard model preparation. Investigations of constitutive a2d1 KO mice have shown that the chronic loss of a2d1 subunits has massive impact on structure and density of synapses at least in the cortex (Risher et al., 2018). To first examine whether hippocampal glutamatergic synapses also undergo changes in the constitutive KO model of the a2d1 subunit, we used transmission electron microscopy. We found changes in both numbers and spine morphology of asymmetric (presumably excitatory) synapses (Fig. 1A-D), with synapse density being reduced by 32% compared with WT (Fig. 1E). Quantitative immunoblotting of brain lysates from WT and a2d1 À/À mice demonstrated that deletion of the a2d1 subunit generally did not alter the overall expression levels of various presynaptic marker proteins, including the pore-forming subunit of Ca V 2.1 channels (Fig.  1F). These results confirm and extend recently reported alterations of cortical synapses in the same KO mouse model (Risher et al., 2018). However, the constitutive KOs of a2d isoforms are associated with severe phenotypes (Striessnig and Koschak, 2008), such as diabetes in the a2d1 KO mice (Felsted et al., 2017), which might obscure more specific a2d functions and complicate the distinction between direct and compensatory effects. To brace against this possibility and to be able to alter expression of a2d1 and a2d3 at defined time points during development, we mostly used lentivirus-mediated overexpression and knockdown to address the role of these auxiliary subunits in defining the connectivity of neuronal networks. a2d1 and a2d3 affect neuronal network activity in distinct developmental windows To address the central question whether a2d1 and/or a2d3 affect synaptogenesis differently and may interfere with the balance between excitation and inhibition, we infected rat hippocampal cultures with lentiviral particles carrying HA-tagged a2d1 or a2d3 subunits. The HA-tag was introduced shortly after the N-terminus of the protein (Fig. 2A). The expression, surface delivery ( Fig. 2B-D) and impact of tagged a2d subunits on current properties Ca V 2.1 and Ca V 2.2 channels were tested. Tagged a2d1 or a2d3 subunits had no impact on the current density or voltage-dependent inactivation of channels tested by expression of Ca V 2.1 or Ca V 2.2 with the b 3 subunit and tagged or untagged a2d subunits in HEK293T cells (current density: a2d1, HA 37.2 6 12.4 pA/pF, n = 14; nontagged 33.0 6 8.9, n = 15; a2d3, HA 58.0 6 15.3, n = 19; nontagged 57.4 6 14.7, n = 16), or Ca V 2.2 (a2d1, HA 27.7 6 7.3 pA/pF, n = 11; nontagged 31.0 6 3.8, n = 10; a2d3, HA 140.3 6 27.7, n = 12; nontagged 115.0 6 21.5, n = 13; half-maximal steady-state inactivation of Ca V 2.1: a2d1, HA À26.6 6 2.3 mV, n = 9; nontagged À30.7 6 3.1 mV, n = 13; a2d3, HA À20.8 6 1.7 mV, n = 13; nontagged À24.0 6 1.3 mV, n = 15; Ca V 2.2: a2d1, HA À44.0 6 1.7 mV, n = 12; nontagged À44.7 6 1.5 mV, n = 12; a2d3, HA À37.0 6 2.3 mV, n = 12; nontagged À37.4 6 1.2 mV, n = 12). Antibodies against a2d1 or a2d3 subunits were suitable for biochemical detection of the proteins in Western blot analysis, but not for evaluation of the surface expression of a2d subunits in live immunocytochemical experiments (Fig. 2B,C,E,G). Comparison of the a2d protein levels in infected cultures to the endogenous level in control sister cultures revealed that total expression of a2d1 or a2d3 was significantly increased by 36% or 160%, respectively (Fig. 2E-G). These evaluating experiments encouraged us to use the viral expression of the a2d subunits to probe whether they have a specific impact on neuronal network development and activity.
Because of the default absence of external inputs, the development of cultured neuronal networks is rather stereotypical and culminates in developmental arrest on maturation after ;28 DIV (van Pelt et al., 2004;Bettencourt et al., 2007;Bikbaev et al., 2015). As a consequence, the spontaneous network activity emerging in neuronal cultures faithfully reflects solely intrinsic formation and maturation of the network connectivity ( Fig. 3A-C) without being influenced or masked by external sensory inputs. Therefore, three cohorts of cultures grown on 60-channel MEAs were infected after 7, 14, or 21 DIV, and the spontaneous activity was recorded ;1 week after infection (Fig. 3D). We found that upregulation of a2d subunits differentially affected the mean firing rate. Depending on the infection time point, a2d1 and a2d3 subunits showed opposite (all p , 0.001, one-way ANOVA; Fig. 3E-I) effects. Upregulation of a2d3 during second developmental week increased the neuronal firing almost fourfold by DIV14 compared with age-matched control or a2d1-overexpressing cultures (both p , 0.001, Duncan's test; Fig.  3E,G). In contrast, a2d3 overexpression after DIV14 strongly suppressed neuronal firing to 21 6 4% by DIV21, compared with the mean values in controls (p , 0.001, Duncan's test). Overexpression of a2d1 had no impact by DIV14 but consistently increased the mean firing rate after DIV14 compared with corresponding values in controls or a2d3-overexpressing cultures at DIV21 (p , 0.01 and p , 0.001, respectively; Duncan's test), with the difference being even more pronounced at DIV28 (both p , 0.001, Duncan's test; Fig. 3F,G). Figure 1. Constitutive KO of the a2d 1 subunit of calcium channels results in a smaller number of asymmetric synapses in the CA1 area of the hippocampus. A, B, Representative areas of panorama images of the CA1 area from WT (A) and a2d 1 À/À (B) mice. Red arrows point to identified asymmetric (presumably excitatory) synapses. Scale bar, 250 nm. C, D, Representative spinous synapses from CA1 of WT (C) and a2d 1 À/À (D) mice. Blue represents postsynaptic spines. Scale bar, 250 nm. E, The mean number of synapses is significantly lower in a2d 1 À/À mice (68.0 6 1.9%, n = 9 images) compared with WT controls (100.0 6 4.7%, n = 9 images). F, The constitutive KO of the a2d 1 subunit generally does not alter presynaptic protein composition in a2d 1 À/À mice, compared with WT animals. VGlut1, vesicular glutamate transporter 1, GAD65, glutamic acid decarboxylase isoform 65, SNAP-25, synaptosome-associated protein 25 kDa, CASK, calcium/Calmodulin-dependent serine protein kinase, Rab3A, Ras-related protein Rab-3A. *p , 0.05, ***p , 0.001. Means and n values are given in Extended Data Figure 1-1. Next, we examined the impact of a2d overexpression on the functional connectivity. For this purpose, we analyzed the occurrence rate of NBs and the burst onset lag, which reflect episodes of functional network interaction between remote neuronal clusters and synchronization of their bursting activity across the network (Bikbaev et al., 2015). At DIV14, we observed no significant change in the mean NB rate on upregulation of a2d subunits. Intriguingly, functional network interaction at DIV21 was strongly enhanced on a2d1 overexpression, whereas upregulation of the a2d3 subunit led to dramatic suppression of NBs (Fig. 3H). Remarkably, the effect of a2d3 upregulation on the synchronicity of the bursting onset was reversed during the third week in vitro: the burst onset lag was shorter at DIV14, but longer at DIV21 in comparison with respective values in age-matched controls (both p , 0.01, Duncan's test; Fig. 3I). In a2d1overexpressing cultures, the burst onset lag was shorter than in controls at DIV14 and DIV21 (p , 0.01 and p , 0.001, respectively; Duncan's test).
Thus, we found that overexpression of a2d1 and a2d3 differentially changes the spontaneous neuronal firing and network interaction in a development-dependent manner. These results indicate that upregulation of the a2d subunits indeed alters the excitatory-to-inhibitory balance in developing hippocampal networks and raise the question how upregulation of a2d1 and a2d3 affects the transmission in excitatory and inhibitory synapses.
a2d1 subunit selectively enhances presynaptic release in excitatory and a2d3 in inhibitory synapses The enhancement of neuronal firing in a2d1-overexpressing cultures (Fig. 3G) could potentially reflect a reported earlier increase in glutamate release and synapse structure (Hoppa et al., 2012;Schneider et al., 2015) or be caused by a decreased release of GABA. Similarly, the a2d3induced suppression of the network activity after DIV14 indicated a shift in the excitatory-to-inhibitory balance due to either enhanced GABA release or reduced release of glutamate. To clarify this, we measured mEPSCs and mIPSCs in neurons overexpressing either a2d1 or a2d3 subunits. To enable recordings in developing neurons, in the following experiments, the primary hippocampal cultures were infected during first developmental week at DIV2-DIV4. Subsequently, mEPSCs and mIPSCs were recorded in the presence of TTX, APV, and either DNQX or bicuculline, respectively, at three time points between DIV7 and DIV21 (Fig. 4A). No significant effect of a2d upregulation on miniature currents was observed in 1-week-old cultures. At DIV14 and DIV21, the mean mEPSC frequency was higher in cultures overexpressing the a2d1, but not a2d3, compared with control values at corresponding time points (p , 0.01 and p , 0.05, respectively; Dunn's test; Fig. 4B,C). In striking contrast, upregulation of the a2d3, but not a2d1, strongly increased the mean mIPSC frequency at DIV14 and DIV21 compared with agematched controls (p , 0.001 and p , 0.05, respectively; Dunn's test; Fig. 4E,F). The amplitude of miniature currents was not affected by a2d overexpression compared with control values at any time point (Fig. 4D,G). However, the mIPSC amplitude at DIV21 was significantly smaller in a2d3-overexpressing cultures compared with cultures overexpressing the a2d1 subunit (p , 0.05; Dunn's test; Fig. 4G). The stochastic opening of high VGCCs accounts for ;50% of mEPSCs and mIPSCs (Goswami et al., 2012;Williams et al., 2012;Ermolyuk et al., 2013). Therefore, the pronounced effect of a2d overexpression on the mEPSCs and mIPSCs after DIV14 (Fig. 4C,F) strongly suggested a bigger contribution of VGCCs to spontaneous release. Indeed, we found that acute blockade of VGCCs by cadmium (Cd 21 ) strongly decreased the frequency of miniature currents in 2-week-old cultures overexpressing the a2d1 (mEPSCs: p , 0.001, Mann-Whitney test) or a2d3 (mIPSCs: p , 0.001) subunit to respective control levels obtained in the presence of Cd 21 from noninfected cultures (Fig. 4H).
In central synapses, the neurotransmitter release is triggered predominantly by Ca V 2.1 and Ca V 2.2 (Wheeler et al., 1994;Scholz and Miller, 1995;Cao and Tsien, 2010), but their Figure 3. Upregulation of a2d subunits strongly affects the neuronal network activity in age-dependent manner. A, An example of rat hippocampal culture grown on 60-channel MEA (MAP2 immunostaining of naive mature culture at DIV35). B, C, Traces of activity (top) and corresponding raster representation of detected spikes (bottom) for the same two channels at DIV15 (B) and DIV35 (C, recorded immediately before immunostaining shown in A). Right, Peristimulus histogram for spikes with channel 1 taken as a trigger. The location of electrodes 1 and 2 is indicated by arrows in A. Note the difference in scaling for spike counts. D, A timeline of infection (green triangle) and recording (orange triangle) in different cohorts of rat hippocampal cultures infected during second (recorded after DIV14; control n = 6 MEAs, a2d 1 n = 5, a2d 3 n = 5), third (recorded after DIV21; all groups n = 8), or fourth (recorded after DIV28; control n = 10, a2d 1 n = 7, a2d 3 n = 6) weeks in vitro. E, F, Representative raster plots of spontaneous neuronal network activity recorded in developing (E, DIV14; scale bar, 60 s) and mature (F; DIV28; scale bar, 10 s) cultures. Thirty of 60 channels from each array are shown. G, Overexpression of the a2d 3 subunit strongly enhances the mean firing rate at DIV14 but suppresses it at DIV21, whereas the a2d 1 upregulation-induced enhancement of neuronal activity is evident later in development (DIV21-DIV28). H, Overexpression of a2d 1 or a2d 3 subunits during the third week in vitro is associated with opposite effects on functional network interaction at DIV21. I, Overexpression of a2d 3 improves synchronization of bursting activity across the network during, but not after, the second developmental week, whereas upregulation of a2d 1 consistently decreases the burst onset lag. *p , 0.05, **p , 0.01, ***p , 0.001. Means and n values are given in Extended Data Figure 3-1. abundance at excitatory and inhibitory presynaptic terminals may vary (Iwasaki et al., 2000). To clarify whether the elevation of the mEPSC and mIPSC frequency by a2d1 and a2d3 subunits involves distinct subpopulations of presynaptic VGCCs, we performed additional patch-clamp recordings in the presence of isoform-specific channel blockers. In a2d1-overexpressing cultures, the blockade of Ca V 2.2 by v -conotoxin GVIA did not abolish the increase in the mean mEPSC frequency, but the blockade of Ca V 2.1 by v -agatoxin IVA reduced the mEPSC frequency (p , 0.001, Dunn's test) to a level observed in control cultures treated with agatoxin (Fig. 4I). In contrast, we found that the a2d3 overexpressioninduced increase in mIPSC frequency was abolished by conotoxin (p , 0.01, Dunn's test), but not by agatoxin (Fig. 4J), compared with control values obtained in the presence of respective toxins.
These results revealed a selective impact of the a2d1and a2d3 calcium channel subunits on the spontaneous neuro-transmitter release in excitatory and inhibitory synapses. Furthermore, we found that facilitation of the spontaneous glutamate release by a2d1 is predominantly mediated by Ca V 2.1, whereas a2d3-driven enhancement of GABA release involved mainly Ca V 2.2 calcium channels.
shRNA-mediated knockdown of a2d1 and a2d3 subunits mirror the effects of overexpression on neurotransmitter release and network activity To rule out possible artifacts of overexpression or lentiviral infection and verify that the effects on neurotransmitter release and the neuronal firing are caused by the overexpression of a2d subunits, we acutely knocked down the a2d1 and a2d3 subunits using specific shRNAs.
For the a2d1 subunit, both live anti-HA labeling of HA-tagged a2d1 subunits and Western blot analysis of the total a2d1subunit population demonstrated strong downregulation in neurons (Fig. 5A-C). Since the most pronounced effect of the a2d1 overexpression on the glutamate release was observed at DIV14 (Fig. 4B, C), we recorded mEPSCs in neurons at DIV14 on a2d1 downregulation, as well as in neurons infected with GFP-expressing lentiviral particles that served as lentiviral infection control (Fig. 5D). We found that shRNAinduced a2d1 knockdown markedly reduced the mEPSC frequency compared with noninfected controls (p , 0.05, Mann-Whitney test; Fig. 5E,F). No effect of lentiviral expression of GFP on the mEPSC frequency or amplitude was found (Fig. 5E-H).
Since the strongest effect of a2d1 overexpression on the network activity was observed at DIV28 (Fig. 3G), in an additional set of 3-week-old cultures, we induced upregulation or downregulation of the a2d1 subunit and assessed spontaneous neuronal firing 1 week later (Fig. 5I). In control cultures, no significant change of the firing rate was observed between DIV21 and DIV28. The upregulation of a2d1 enhanced neuronal firing (p , 0.001, Duncan's test), whereas the a2d1 knockdown led to suppression of the mean firing rate compared with values in control and a2d1-overexpressing cultures (p , 0.05 and p , 0.001, respectively; Duncan's test; Fig. 5J,K).
Similar experiments were conducted using shRNA constructs to knock down the a2d3 subunit. Evaluation of the construct demonstrated a robust suppression of a2d3 subunit expression by 50%-60% in HEK cells (p , 0.001, Mann-Whitney test; Fig.  6A,B) and primary hippocampal cultured neurons (p , 0.05; Fig.   Figure 4. Overexpression of a2d 1 and a2d 3 subunits selectively increases the frequency of neurotransmitter release in excitatory and inhibitory synapses, respectively. A, A timeline of infection (green triangle) and electrophysiological recordings (orange triangles). B, Representative traces of mEPSCs recorded at DIV14 in control and a2d 1-and a2d 3-overexpressing cultures. C, D, The mean frequency (C) and the amplitude (D) of mEPSCs in a2d 1-and a2d 3-overexpressing cultures. E, Representative traces of mIPSCs recorded at DIV14 in control and a2d 1-and a2d 3-overexpressing cultures. F, G, The mean frequency (F) and the amplitude (G) of mIPSCs in a2d 1-and a2d 3-overexpressing cultures. H, The increase in the mEPSC and mIPSC frequency by a2d 1 and a2d3 subunits, respectively, is caused by bigger contribution of high voltage-activated VGCCs as demonstrated by Cd 21 -induced reduction to respective values obtained in controls in the presence of Cd 21 . I, J, The effects of a2d 1 and a2d 3 overexpression on the frequency of mEPSCs (E) and mIPSCs (F) are mediated by P/Q-and N-type calcium channels, respectively. CNTX, conotoxin, AGTX, agatoxin. *p , 0.05, **p , 0.01, ***p , 0.001. Means and n values are given in Extended Data Figure 4-1. 6C,D). Furthermore, the quantification of the a2d3 expression level in neuronal cultures demonstrated a significant reduction on shRNA-mediated knockdown both in HEK cells (p , 0.001; Fig. 6E,F) and in neurons (p , 0.05; Fig. 6G-I).
Functional analysis of the a2d3 knockdown demonstrated that higher frequency of spontaneous GABA release and the enhance neuronal network activity in young a2d3-overexpressing cultures were indeed caused by upregulation of this auxiliary subunit. We found that the frequency of mIPSCs was markedly decreased on a2d3 knockdown, but not GFP expression, compared with controls (p , 0.05, Mann-Whitney test; Fig. 6J-L). The amplitudes of mIPSCs where not affected in any of the groups (Fig. 6M,N). Finally, a comparison of the spontaneousactivity recorded under control conditions or on a2d3 upregulation or downregulation (Fig. 6O) revealed that shRNA-mediated a2d3 knockdown resulted in suppression of spontaneous neuronal firing compared with values in control or a2d3-overexpressing cultures (p , 0.05 and p , 0.001, Duncan's test; Fig. 6P,Q).
So far, these data revealed a selective impact of the a2d1 as well as a2d3 calcium channel subunit on the presynaptic neurotransmitter release in excitatory and inhibitory synapses. Given these findings, next we asked whether the elevated frequency of miniature currents on upregulation of a2d subunits reflects corresponding changes in the number of glutamatergic and/or GABAergic synaptic contacts.  The a2d1 subunit was reported earlier to trigger excitatory synaptogenesis in mouse retinal ganglion cells and cortical neurons (Eroglu et al., 2009), but it remained unknown whether a2d3 plays a similar role in central synapses. To clarify this, we labeled the presynaptic scaffold protein Bassoon and the postsynaptic scaffold protein Homer1 or Gephyrin to identify glutamatergic and GABAergic synapses, respectively. The immunolabeling was conducted in hippocampal cultures 2-3 weeks after infection at DIV14-DIV24 (Fig. 7A). Using colocalization of presynaptic and postsynaptic markers distributed along dendrites (Fig. 7B,C), we evaluated the density of synaptic contacts per mm (for details, see Materials and Methods).
The transfection-induced overexpression of a2d subunits triggers accumulation of presynaptic proteins via increased surface expression of VGCCs (Hoppa et al., 2012;Schneider et al., 2015), which we could not reveal in lentiviral infected cultures. This, in turn, leads to recruitment of presynaptic scaffold components when expressed in combination with the a1 subunit (Davydova et al., 2014;Schneider et al., 2015). To verify that, we assessed the fluorescence intensity of several key presynaptic proteins in hippocampal cultures transfected either with a2d1-HA or a2d3-HA. The transfection-induced overexpression allowed us to distinguish the HA-positive transfected synapses and HAnegative puncta of nontransfected neurons embedded into the same network. Indeed, upregulation of either a2d1 or a2d3 led to an enhanced accumulation of Bassoon and RIM (Fig. 7F,G; both p , 0.001) that was more pronounced for Bassoon on a2d1 upregulation (p , 0.001, Bonferroni test). Furthermore, the upregulation of a2d1 or a2d3 significantly increased the fluorescence of VGlut1, indicating a structural change of excitatory synapses ( Fig. 7G; both p , 0.001). Remarkably, the fluorescence of VGAT, an inhibitory synapse-specific marker, was 38 6 9% higher only in a2d3-overexpressing neurons compared with control or a2d1-overexpressing neurons (p , 0.01 and p , 0.001, respectively; Fig. 7G). Consistent with previous reports (Hoppa et al., 2012;Schneider et al., 2015), upregulation of a2d1 or a2d3 subunits increased the synaptic abundance of Ca V 2.1 (both p , 0.001) and Ca V 2.2 (p , 0.001 and p , 0.01, respectively; Fig.  7G). No differences between a2d1or a2d3-overexpressing neurons in the fluorescence intensity of either Ca V 2.1 or Ca V 2.2 were found.
These findings demonstrate that upregulation of a2d1 or a2d3 subunits in rat hippocampal neurons triggers the glutamatergic synaptogenesis, hence corroborating previous reports (Dickman et al., 2008;Eroglu et al., 2009;Kurshan et al., 2009). Moreover, we found that upregulation of the a2d3, but not a2d1, subunit increases the number of GABAergic synapses in hippocampal cultures already 2 weeks in vitro.
a2d3 selectively promotes axonal outgrowth and branching in inhibitory neurons Apart from mediating the synaptic inhibition, GABA is directly involved in a variety of fundamental processes, such as neuronal migration, differentiation, and axonal outgrowth, that take place before the formation of functional synapses (Owens and Kriegstein, 2002;Huang et al., 2007). Given the a2d3-specific effect on the GABA-dependent inhibitory postsynaptic currents (Fig. 4F) and the inhibitory synaptogenesis (Fig. 7E), next we examined whether upregulation of this subunit is associated with enhanced axonal outgrowth, as it was shown for a2d2 subunit in the spinal cord (Tedeschi et al., 2016). Therefore, we first looked at rat hippocampal cultures, which were infected with a2d1-HA or a2d3-HA subunits at DIV2-DIV4 and additionally transfected at DIV4 with GFP as a volume marker to aid identification of individual neurons and their processes within the network. At DIV9-DIV10, cultures were stained for MAP2 to label the dendritic arbor of individual neurons. Subsequently, the length and branching of axons, which were detected by GFP-positive but MAP2-negative signal, were analyzed using Scholl analysis and Simple Neurite Tracer plug-in for Fiji software (Longair et al., 2011) for semiautomatic reconstruction of cells (for details, see Materials and Methods). We found no significant effect of a2d1 or a2d3 upregulation on the mean axonal length, nor were the number of branching points markedly affected. However, individual values obtained in a2d3-overexpressing neurons were distributed within substantially broader range (length min-max 15%-394%, mean 121 6 20%; branches min-max 32%-402%, mean 124 6 20%; n = 23 neurons), compared with control (length min-max 36%-208%, mean 100 6 11%; branches min-max 32%-229%, mean 100 6 12%; n = 21) or a2d1-overexpressing (length min-max 15%-211%, mean 86 6 13%; branches min-max 11%-192%, mean 77 6 11%; n = 21) neurons. We assumed that such heterogeneity in the dataset might reflect a mixture of values obtained in excitatory and inhibitory neurons. Therefore, we proceeded with the analysis of the axonal outgrowth and branching specifically in interneurons.
In young neurons, GAD67 is a rate-limiting enzyme responsible for up to 90% of GABA synthesis in the brain (Asada et al., 1997). In order to unequivocally identify and quantify individual interneurons, we prepared hippocampal cultures from mice expressing GFP under control of GAD67 promoter (GAD67:: GFP). Cultures underwent the infection at DIV2-DIV4 and fixation at DIV9, followed by immunostaining for MAP2 to visualize the dendritic arbor as previously described. Subsequently, the length and the number of axonal branches were quantified exclusively for GFP-positive cells (i.e., for GAD67-positive interneurons) (Fig. 8A-C). In a2d1-overexpressing interneurons, the mean axon length and the number of branches did not significantly differ from respective values obtained in control noninfected cultures. In contrast, axons of a2d3-overexpressing interneurons were significantly longer and branched more extensively, compared with controls or a2d1-overexpressing cultures (both p , 0.001 Dunn's test; Fig. 8D,E; see Extended Data Fig. 8-1). These data demonstrated that upregulation of auxiliary a2d3 subunit of calcium channels promotes the axonal outgrowth specifically in inhibitory GABAergic interneurons.
Together, our findings demonstrate that the a2d1 or a2d3 calcium channel subunits play an important role in several aspects of early circuitry formation in neuronal networks. The expression of both a2d1 and a2d3 favors the formation of synaptic connectivity. However, we found that the impact of the a2d3 subunit is inhibitory cell type-specific, with a2d3 upregulation being associated with enhanced GABA release, formation of inhibitory synapses, and axonal outgrowth in interneurons. Furthermore, we found that such synapse type-specific impact of a2d1 and a2d3 on the neurotransmitter release is associated with their functional preference for distinct VGCC isoforms.

Discussion
This study characterizes the differential impact of a2d1 and a2d3 auxiliary subunits of VGCCs on structural and functional properties of developing hippocampal neurons. To overcome the limitations and side effects of constitutive KO of individual subunits of calcium channels (Striessnig and Koschak, 2008), in this work we used lentiviral overexpression of a2d subunits in cultured neuronal networks. We found that both a2d1 and a2d3 can trigger excitatory synaptogenesis in hippocampal neurons, whereas upregulation of only a2d3 subunit increases inhibitory synapse number and enhances presynaptic GABA release. Using hippocampal cultures prepared from GAD67::GFP mice, we found that a2d3 overexpression also promotes the axon outgrowth in young interneurons. Together, these findings shed new light on the earlier reported functional redundancy of a2d1 and a2d3 despite pronounced structural differences between these isoforms (Klugbauer et al., 1999;Dolphin, 2013), and show their differential but complementary roles in early circuitry formation.
Throughout the experiments, we implemented two infection protocols. Lentiviral infection at different developmental time points, namely, after first, second, or third week in vitro (Fig.  3D), demonstrated that a2d subunits alter neuronal firing and network interaction in a development-dependent and subunitspecific manner. Given the isolation of neuronal cultures from external sensory inputs that drive network activity already in the early postnatal period (Khazipov et al., 2004), the suppression of activity on a2d3 upregulation (Fig. 3G) indicated a prevalence of inhibition over excitation. In contrast, a2d1 upregulation after the second week in vitro consistently enhanced the network activity and demonstrated a shift toward excitation on the network level. Thus, these results show that a2d1 and a2d3 are intimately involved into the establishment and modulation of the excitation/inhibition balance.
To characterize the long-term consequences on neurotransmitter release, in the rest of experiments, the infection was performed during the first week in vitro and the data were acquired within the period of DIV7 to DIV24. This protocol revealed that a2d1 overexpression selectively enhances sponta-neous presynaptic glutamate release without affecting the spontaneous release of GABA (Fig. 4C,F), whereas the knockdown of this subunit led to impairment of glutamate release (Fig. 5E,F). Such selectivity of a2d1 in facilitation of release in excitatory synapses is consistent with previously shown localization of a2d1 primarily in excitatory presynaptic terminals in the hippocampus (Hill et al., 1993;Bian et al., 2006;Nieto-Rostro et al., 2014) and corroborates recent reports on the positive correlation between surface expression of a2d1 and the mEPSC frequency (Cordeira et al., 2014;Zhou and Luo, 2015). Notably, higher frequency of spontaneous glutamate release in 2-week-old a2d1overexpressing neurons (Fig. 4C) was not accompanied by higher synaptic density (Fig. 7D), suggesting that the elevation of the release probability precedes the synaptogenic function of a2d1.
One of the central findings of our study is the a2d3 overexpression-induced increase in the frequency of spontaneous GABA release (Fig. 4E,F), which was accompanied by the higher density of inhibitory synapses (Fig. 7E). Surprisingly, we found that the a2d3 upregulation also increases the excitatory synapse density in rather mature 3-week-old cultures (Fig. 7D) without affecting the mEPSC frequency (Fig. 4B,C). Electrical activity per se in immature networks is necessary and sufficient for synaptogenesis and early circuitry formation (Ben-Ari, 2001;Spitzer, 2006) and can potently influence the development of GABAergic synapses (Chattopadhyaya et al., 2007). The enhancement of the network activity observed on overexpression (Fig. 3G), but not downregulation (Fig. 6P,Q), after DIV7 in cultures grown on MEAs could therefore indirectly trigger the formation of surplus glutamatergic synapses. The GABA synthesis and signaling begin already at embryonic stages; thus, GABA acts as a trophic factor influencing fundamental developmental processes before it becomes a principal inhibitory neurotransmitter (Owens and Kriegstein, 2002;Ben-Ari et al., 2007;Huang et al., 2007). Although still debated in the literature, GABA in immature neurons can exert an excitatory action so that binding to GABA A receptors results in membrane depolarization. In particular, the GABA A receptor-mediated depolarization in young neurons was shown to be sufficient for VGCC activation (Leinekugel et al., 1995;LoTurco et al., 1995;Ganguly et al., 2001) and required for formation and/or maintenance of GABAergic synapses (Oh et al., 2016). Intriguingly, we observed a dramatic change in the effect of a2d3, but not a2d1, overexpression on neuronal firing depending on the developmental stage (Fig. 3G). A reversal from enhancing spontaneous network activity at DIV14 to its suppression at DIV21 likely reflected the switch to hyperpolarizing GABA action and/or formation of functional inhibitory synapses (Fig. 7E), which requires binding of GABA to GABA A receptors followed by aggregation of postsynaptic Gephyrin puncta (Oh et al., 2016). In line with increased GABA release (Fig. 4F), we found that the number of Bassoon puncta colocalized with Gephyrin already by the end of second developmental week was bigger in a2d3-overexpressing cultures than in controls (Fig. 7E). These outcomes corroborate previous reports showing an important role of spontaneous Ca 21 transients in regulation of the neurite outgrowth and branching (Ciccolini et al., 2003) and structural maturation of synapses (Choi et al., 2014). Despite our data on the inhibitory synapses number (Fig. 7E) and GABA release (Fig. 4E,F), the finding that upregulation of a2d3 enhances the axonal growth and branching specifically in interneurons (Fig. 8D,E) provided additional evidence for the a2d3-specific modulation of GABArelated functions. Importantly, the outgrowth was promoted selectively in interneurons positive for GAD67, which plays major role in GABA synthesis (Asada et al., 1997), as well as in maturation of perisomatic inhibition and elimination of excessive excitatory synapses (Nakayama et al., 2012).
The a2d subunits are known to support the trafficking of the pore-forming a1 subunit of calcium channels (Dolphin, 2012). Our results, that the elevation of frequency of neurotransmitter release in 2-week-old neurons is abolished by VGCC isoformspecific blockers (Fig. 4I,J), demonstrated a preference of a2d1 and a2d3 for trafficking of Cav2.1 in glutamatergic and Cav2.2 in GABAergic synapses, respectively. Agatoxin and gabapentin, but not conotoxin, were reported earlier to induce an identical nonadditive decrease in K 1 -triggered Ca 21 influx (Fink et al., 2000), indicating that a2d1-mediated contribution to calcium signaling is sensitive to Cav2.1 blockade. Furthermore, our finding that a2d1 modulates the release in excitatory synapses preferentially via P/Q-type channels is in line with strong reduction of spontaneous (Bomben et al., 2016) and evoked (Mallmann et al., 2013) release of glutamate reported in Ca V 2.1 KO mice. Although, P/Q-type channels were shown to induce synaptic recruitment of Bassoon (Davydova et al., 2014), in our study the accumulation of Bassoon was evident only in inhibitory synapses and only on a2d3 overexpression. Since both evoked and spontaneous GABA release require presynaptic accumulation of VGCCs (Williams et al., 2012), the latter supports previous reports (Hoppa et al., 2012;Schneider et al., 2015) and suggests that both a2d1 and a2d3 subunits can serve as rather universal cargos of the pore-forming a1 subunit. The a2d isoform-specific recruitment of P/Q-or N-type channels can therefore be related to different roles a2d subunits play in the neuronal network development. Indeed, the dominant role of the a2d3 subunit in the early and the a2d1 subunit in the late development matches expression profiles of P/Q-and N-type calcium channels (Scholz and Miller, 1995;Iwasaki et al., 2000;Fedchyshyn and Wang, 2005). Furthermore, the effects of the a2d1 on the mEPSC frequency and of the a2d3 on the mIPSC frequency were more pronounced in the presence of conotoxin and agatoxin, respectively, compared with values in respective groups obtained without toxins. The latter finding corroborates the concept of functional competition of VGCC isoforms in presynaptic active zone (Cao and Tsien, 2010;Davydova et al., 2014). In this context, the surface interaction with synaptic adhesion molecules, such as a-neurexin (Missler et al., 2003), which has been suggested in several systems (Tong et al., 2017;Brockhaus et al., 2018), can be an important contributing factor for such specificity of a2d subunits in network development.
Our data provide support for the reported association of CACNA2D1 and CACNA2D3 genetic aberrations with autism (Iossifov et al., 2012;De Rubeis et al., 2014;Vergult et al., 2015) and the high comorbidity of epilepsy in individuals with autism (Tuchman and Rapin, 2002;Levisohn, 2007). By fostering the GABAergic signaling, the a2d3 subunit effectively drives the early network activity that is crucial for the initial circuitry formation. The impact of the a2d1 subunit becomes prominent later in development and is rather restricted to glutamatergic signaling. One interaction partner for this action could be a-neurexin, which, together with a2d1, facilitates the trafficking of Ca V 2.1 VGCCs to presynaptic terminals (Brockhaus et al., 2018), whereas a2d3 may play an opposite role (Tong et al., 2017). Altered expression of a2d1 or a2d3 can therefore cause a chronic imbalance between excitation and inhibition that is rather characteristic for autism spectrum disorders (Rubenstein and Merzenich, 2003;Nelson and Valakh, 2015). As a consequence, impairment of a2d -mediated functions during critical developmental periods can trigger in affected individuals devastating maladaptive changes on the network level and potentially lead to global aberrations in the brain connectivity (Baron-Cohen and Belmonte, 2005;Courchesne and Pierce, 2005) and the neural information processing (Belmonte et al., 2004).