Animal care and handling was performed according to the guidelines of the Canadian Council on Animal Care¹ and approved by the Facility Animal Care Committee of the Douglas Research Center. The VGLUT3^-/- mouse line used in the present study was obtained as previously described (Gras et al., 2008; Amilhon et al., 2010) and backcrossed with 10 generations (F10) of mice with a C57BL/6N background (99% homozygous). Heterozygous mice were bred to generate VGLUT3^-/- mice and wild type littermates.

The VGLUT3^flox/flox mice Slc17a8 mutant mouse line was established at Phenomin - iCS (Phenomin- Institut Clinique de la Souris-, Illkirch, France²). In brief, the targeting vector was constructed as follows (Figure 4A). A 0.8 kb fragment encompassing exon 2 was amplified by PCR (from C57Bl/6N ES cells genomic DNA) subcloned in a Phenomin-iCS proprietary vector. This Phenomin-iCS vector contains a LoxP site as well as a floxed and flipped Neomycin resistance cassette. A 4.5 kb fragment (corresponding to the 5′ homology arm) and 3.1 kb fragment (corresponding to the 3′ homology arms) were amplified by PCR and subcloned in step1 plasmid to generate the final targeting construct. The linearized construct was electroporated in C57Bl/6N mouse embryonic stem (ES) cells. After selection, targeted clones were identified by PCR using external primers and further confirmed by Southern blot with 5′ and 3′ external probes. Two positive ES clones were injected into BALB/cN blastocysts, and male chimeras derived gave germline transmission. Mice lacking VGLUT3 selectively in GABAergic neurons (VGLUT3^{V IAAT-Cre-flox/flox}) were generated by breeding VIAAT^-ires-cre mice (The Jackson laboratory, stock No: 016962) with VGLUT3^flox/flox mice.

Mice were housed in groups of three to four animals per cage under standard conditions: 22°C and a 12 h light/dark cycle (8:00–20:00 light period) with food and water provided ad libitum. All precautions were taken to minimize the number of animals used and their suffering they may have had to undergo.

CB1 autoradiography was performed as previously described (Herkenham et al., 1991). Briefly, fresh frozen brain sections (10 μm) were incubated for 3 h at RT° in a buffer containing Tris HCl (50 mM, pH7.4), bovine serum albumin (5%) and [³H] CP-55 940 (10 nM, NEN, Boston, MA, USA) in the absence or presence of AM251 (10 mM, Tocris Bioscience) to determine total and non-specific binding, respectively. Slides were washed in Tris HCl (50 mM) containing bovine serum albumin (1%, fatty acid-free) for three times (30 min each) at 4°C dipped in ice-cold distilled water. Dried sections were apposed to tritium-sensitive screen (BAS-TR Fuji Imaging) for a period of 7 days and developed with the FujiBAS 7000 PhosphoImager. Densitometry was analyzed with MCID software, using the standard curve generated from [³H]-standards. Specific binding measured in each structure was determined by subtracting the non-specific binding image from that of total binding.

In situ hybridization was performed as previously reported (Gras et al., 2008; Vigneault et al., 2015). Shortly, sense or antisense oligonucleotides specific for CB1 or for the floxed exon 2 of VGLUT3 (Tables 1, 2, generated by Oramacell) were labeled with [³⁵S]-dATP (GE Healthcare) using terminal deoxynucleotidyl transferase (GE Healthcare) to a specific activity of 1–3 × 10⁸ dpm/mg. Mouse brain fresh frozen sections (10 μm) were incubated with hybridization medium (Oramacell, Paris, France) containing 1.4 μl of the labeled oligonucleotides, washed, dried and exposed to a BAS-SR Fujifilm Imaging Plate for 5 days. Plates were scanned with a Fujifilm BioImaging Analyser BAS-5000.

Antisense cRNA riboprobes were obtained from the transcription of a PCR template amplified with primers containing the T7 promoter followed by the vesicular inhibitory amino acid transporter (VIAAT) sequence (forward primer 5′-AATTAACCCTCACTAAAGGGAGCCAGGGCCTGCAGATGGAC-3′ and reverse primer 5′-TAATACGACTCACTATAGGGTCGCTGGGCTGCTGCATGTT-3′) or glutamic acid decarboxylase (GAD) sequence (forward primer 5′-AATTAACCCTCACTAAAGGGAGAGGAGCGGATCCTAATACTACC-3′ and reverse primer 5′-TAATACGACTCACTATAGGGAGATCCATGAGAACAAACACGGG-3′) or the VGLUT3 sequence (forward primer 5′-AATTAACCCTCACTAAAGGGAGAAAAACAGGACTGGGCTGACCC-3′ and reverse primer 5′-TAATACGACTCACTATAGGGAGAGAGACCAAGGTCCATATTCCC-3′). GAD and VIAAT or VGLUT3 riboprobes were labeled with UTPs coupled to fluorescein and digoxigenin, respectively (Roche Applied Science, Indianapolis, IN, USA). Cerebral coronal cryosections (10 μm) were fixed with 4% formaldehyde and hybridized as previously described (Gras et al., 2002). Sections were incubated with anti-fluorescein antiserum coupled with horseradish peroxidase (1:250, 1 h RT, Roche Applied Science). The signal was amplified with the TSA-plus-biotin kit (Perkin Elmer, Waltham, MA, USA). GAD or VIAAT RNA was visualized with Neutravidin Oregon Green (Invitrogen) at 488 nm. After the horseradish peroxidase was inactivated with a glycine solution, the brain slices were incubated in the blocking solution for 30 min at RT. The slices were then incubated with anti-DIG coupled with horseradish peroxidase (1:2500, Roche Applied Science) for 1 h at RT. The TSA-plus-Cyanine 3 kit (Perkin Elmer, 10 min at room temperature) was used to detect the VGLUT3 transcript under 555 nm-excitation fluorescence light. The slices were mounted with Fluoromount-G (Southern Biotech, Birmingham, AL, USA) and analyzed with an AxioObserver.Z1 microscope (Carl Zeiss, Germany).

Immunofluorescence experiments were performed as previously described (Amilhon et al., 2010). Coronal mouse brain sections that included the ventral hippocampus were incubated with VGLUT3 guinea-pig antiserum [1:3000 (Amilhon et al., 2010)], VIAAT rabbit antiserum (1:3000 (Dumoulin et al., 1999)), gephyrin mouse antiserum (1:1000, Synaptic System, Göttingen, Germany), mGLUR4 rabbit antiserum (1:100, Invitrogen, Life Technologies Inc., Burlington, ON, Canada, cat. # 51-3100), mGLUR7 rabbit antiserum (1:1000, homemade), CB1 rabbit antiserum (1 : 500, generous gift from Zsolt Lenkei, UMR8249, ESPCI-ParisTech. (Thibault et al., 2013)) and DAPI to stain nuclei. Primary antisera were detected with Alexa-fluor-488- or Alexa-fluor-555-conjugated secondary antibodies (1:2000, Molecular Probes Inc., USA). Images were acquired using an AxioObserver.Z1 microscope connected to an Apotome.2 module (Carl Zeiss, Germany). For 3D reconstruction, pictures were captured using a Leica TSC SP2 upright laser scanning confocal microscope (Leica MicroSystems, Germany) equipped with a 63×, 1.4 NA oil immersion objective with the pixel size set to 60 nm and a z-step of 130 nm. Images were deconvoluted using a Maximum Likelihood Estimation algorithm with Huygens 3.6 software (Scientific Volume Imaging). The background intensity was estimated from the average of the voxels with the lowest intensity, and the signal-to-noise ratio was assigned an estimated value of 20. Labeled terminals were quantified using the JaCop plugin of ImageJ (Bolte and Cordelieres, 2006). Co-localization analyses were performed by correlating the intensities of pairs of pixels in the two different channels as previously described (Jaskolski et al., 2005) using the co-localization color map plugin of ImageJ.

VGLUT3 was detected with an electron microscope with a post-embedding immunogold method as previously described (Gras et al., 2002, 2008). Sagittal brain sections of the hippocampus were incubated with guinea pig anti-VGLUT3 antiserum (1:1000). Immunolabeling was detected with goat anti-guinea pig IgGs conjugated to gold particles (1.4 nm in diameter; Nanoprobes, NY, USA, 1:100). After post-fixation (1% glutaraldehyde), the immunogold signal was enhanced with a silver enhancement kit (HQ silver, Nanoprobes; NY, USA). Ultrathin sections were examined with a transmission electron microscope (EM 912 OMEGA, ZEISS).

Synaptosomes from hippocampus were prepared, and the LS1 fractions containing synaptic vesicles (SVs) were isolated as described previously (Zander et al., 2010). The LS1 fractions were incubated with ^∗superparamagnetic beads (Dynabeads Pan Mouse IgG, Invitrogen, Life Technologies Inc., Burlington, ON, Canada) coupled to anti-synaptophysin, anti-VGLUT1 or anti-VIAAT antisera (all from Synaptic Systems, Göttingen, Germany). Beads without primary antibodies were used as negative controls. Bead-bound SVs were dissolved in Laemmli buffer, and the resultant protein pattern was analyzed by SDS-PAGE/Western blot (Zander et al., 2010) using anti-VGLUT1, anti-VIAAT, anti-synaptophysin and anti-VGLUT3 (all from Synaptic Systems, Göttingen, Germany). Immuno-signals were visualized by enhanced chemiluminescence. Densitometric analyses were performed using LabImage 1D 2006 (Kapelan Bio-Imaging Solutions) and quantified using standard curves obtained from the initial LS1 fraction.

Vesicular [³H]glutamate uptake analyses were performed as described previously (Zander et al., 2010; Frahm et al., 2015). LS1-immunoisolated SVs were bound to M-280 sheep anti-mouse beads (Invitrogen). Beads coupled with non-specific mouse IgG (Santa Cruz Biotechnologies) were used as negative controls. Uptake assays were performed with immunoisolated SVs and initiated by the addition of L-glutamate (49.5 μM, Sigma–Aldrich) and [³H]-glutamate (0.5 μM, Hartmann Analytic GmbH, Germany). Uptake was stopped after 10 min at 25°C, and unbound radioactivity was removed by centrifugation (435,000 × g, 10 min).

Mice (23–35 days old) were anesthetized with isoflurane and decapitated. Their brains were quickly removed and placed in an oxygenated (95% O2/5% CO₂ gas mixture) icy slicing solution containing (in mM): 2.5 KCl, 0.1 CaCl₂, 4 MgCl₂, 1.25 KH₂PO₄, 25.2 sucrose, 26 NaHCO₃ and 10 glucose (Manseau et al., 2010). Coronal slices (300 μm) containing the ventral hippocampus were cut with a vibratome (vt1200, Leica Microsystems Inc., Canada) and placed in a slice saver filled with oxygenated artificial cerebrospinal fluid [ACSF, 125 mM NaCl, 2.5 mM KCl, 2 mM CaCl₂, 2 mM MgCl₂, 1.25 mM NaH₂PO₄, 26 mM NaHCO₃ and 25 mM glucose (Manseau et al., 2010)]. The slices were incubated for at least 1 h at RT prior to experimentation.

For patch-clamp experiments, evoked and spontaneous miniature inhibitory postsynaptic potentials (eIPSCs and mIPSCs, respectively) were recorded in CA1 pyramidal neurons. Recordings were performed using the whole cell configuration of the patch-clamp technique with borosilicate glass electrodes (tip resistance: 2.5–3.5 MΩ) filled with a “high chloride” intracellular solution (70 mM K-gluconate, 70 mM KCl, 2 mM NaCl, 2 mM MgCl₂, 10 mM HEPES, 1 mM EGTA, 5 mM QX-314, 2 mM Mg-ATP, and 0.3 mM Na-GTP (Manseau et al., 2010)). The E_Cl estimated using the Nernst equation was approximately -16 mV. Action potentials were blocked with QX-314 (5 mM). Under these conditions, monosynaptically elicited eIPSCs in a pyramidal neuron held at -70 mV were represented by an inward current. Fast glutamatergic transmissions were blocked with DNQX and AP5 for all patch-clamp recordings. The perfusion of bicuculline (5 μM) completely abolished the recorded inward current, which confirmed the GABAergic nature of the postsynaptic events (see Supplementary Figure S1). An inhibitory square pulse was applied before each stimulation to measure the input resistance and assess the quality of the patch. Recordings were excluded if the input resistance varied more than 20%. Although the theoretical ECl^- was estimated to be approximately -16 mV by the Nernst equation, the measured ECl^- was approximately -8 mV on the IV curve of GABAergic currents (n = 5, see Supplementary Figure S1). The membrane potential was clamped at -70 and -80 mV for eIPSC and mIPSC recordings, respectively. Spontaneous mIPSCs were recorded in the presence of tetrodotoxin (TTX, 0.5 μM). The eIPSCs were evoked monosynaptically by placing a borosilicate glass pipette in the stratum pyramidale 50 μm away from the recorded neuron. The signals were amplified with a Multiclamp 700B patch-clamp amplifier (Axon Instruments, Foster City, CA, USA), sampled at 20 kHz and filtered at 10 kHz using a Multiclamp 700B commander (Axon Instruments). The data were analyzed using pClamp (Axon Instruments) and Origin (Microcal Software, Northampton, MA, USA) software. A custom-written software was used for the mIPSC analysis (Manseau et al., 2010).

For long term potentiation/depression (LTP/LTD) recordings, field excitatory postsynaptic potentials (fEPSPs) were recorded in the stratum radiatum of the medial CA1 region with a borosilicate glass electrode (tip resistance: 2.7–3.3 MΩ) filled with ACSF at RT. The fEPSPs were evoked every 20 s with a 80 μsec stimulation at afferent Schaffer collateral fibers delivered through a stimulus isolator (WPI) with a bipolar tungsten-based electrode. The stimulation intensity was adjusted to obtain amplitudes close to 50% of the maximum response. The fEPSP baseline was monitored for 20 min. The synaptic transmission was potentiated either by a high-frequency tetanic stimulation (HFS) consisting of three trains of 100 pulses 20 s apart at a frequency of 100 Hz or a theta-burst stimulation (TBS) consisting of 3 trains 20 s apart of 10 bursts of 5 pulses at 100 Hz at a burst frequency of 5 Hz. The fEPSP slope is defined as the slope of the linear fit of the portion of the field curve between 10 and 60% of maximum peak amplitude. The synaptic transmission was depressed by a low-frequency stimulation consisting of 900 pulses at either 1 Hz (LFS-1 Hz) or 3 Hz frequency (LFS-3 Hz). When stated, bicuculline (5 μM) was added to the ASCF throughout the experiment to block GABA_A receptor-mediated fast inhibitory transmission. Long-term potentation (LTP) or long-term depression (LTD) are defined as the comparison of the mean fEPSP slope over the last 10 min of recording after stimulation (30–40 min. for LTP, 50–60 min. for LTD) with the mean fEPSP slope over the 10 min prior to stimulation.

Intact hippocampi were isolated from 21- to 25-day-old mice as previously described (Khalilov et al., 1997; Tretter et al., 2012; Amilhon et al., 2015). Briefly, the hippocampi were dissected and placed in an oxygenated high-sucrose solution before use. Recordings were made with 4.0 mM KCl ACSF filled borosilicate pipette (Goutagny et al., 2009). Local field potentials (LFP) recorded in the CA1 stratum radiatum of the temporal part of the hippocampus were filtered online (0.1–500 Hz) and sampled at 5 kHz. Rhythmic oscillations ranging from 3 to 12 Hz were considered to represent theta oscillations.

With the exception of bicuculline (Sigma–Aldrich) and tetrodotoxine (TTX, Alomone, Israel), all drugs were purchased from Ascent Scientific (Abcam Biochemicals), including D-(-)-2-Amino-5-phosphonopentanoic acid (AP5), 6,7-Dinitroquinoxaline-2,3-dione (DNQX), (2S)-2-Amino-2-[(1S,2S)-2-carboxycycloprop-1-yl]-3-(xanth-9-yl)propanoic acid (LY341495), (4-Methoxyphenyl)-(6-methoxy-quinazolin-4-yl)amine (LY456236), 2-Methyl-6-(phenylethynyl) pyridine (MPEP) and (RS)-α-Methylserine-O-phosphate (MSOP) and WIN55,212-2 mesylate.

Statistical analyses were performed using Prism 4 software (Graph Pad). Each statistical test was appropriately chosen for the relevant experimental design. To compare two groups, a two-tailed paired Student t-test or a non-parametric Mann–Whitney test was performed according to normality of the group distributions. For multiple group comparisons, a one- or two-way ANOVA or repeated measure ANOVA was performed followed by a post hoc Bonferroni comparison of the group means. Data are expressed mean ± SEM. ^∗P < 0.05, ^∗∗P < 0.01, ^∗∗∗P < 0.001. A descriptive table of all the statistics is provided in the Supplementary Tables S1–S5.

In the hippocampus, VGLUT3 is expressed by CCK-positive and more rarely by calbindin-positive interneurons (Somogyi et al., 2004; Klausberger and Somogyi, 2008). We first determined the percentage of GABAergic interneurons expressing VGLUT3. A double FISH technique was used to label the mRNA encoding VGLUT3 and VIAAT (Figure 1A) or glutamic acid decarboxylase (GAD, the GABA-synthesizing enzyme, Figure 1B). We found that 8.7 ± 0.3% of all GAD expressing GABAergic neurons in the hippocampus expressed VGLUT3 (Figure 1B). Similar results were obtained with VIAAT-expressing GABAergic neurons (not shown). Once synthesized, VGLUT3 is transported to the axon terminals surrounding the somata of neurons in the pyramidal layer of the hippocampus (Figures 1C,D). We determined by three-dimensional non-biased counting that 98 ± 1% of VGLUT3-positive terminals co-expressed VIAAT but only 13 ± 2% of VIAAT-positive terminals contained VGLUT3 in the CA1 pyramidal layer of the hippocampus. Electron microscopy showed that VGLUT3-immunoglod puncta were present over synaptic vesicle clusters in axon terminals forming axo-somatic symmetrical synapses (Figure 1E). VGLUT3-positive terminals co-localized with gephyrin (Figure 1F), a major scaffolding protein found at inhibitory synapses (Freund and Buzsaki, 1996). These results demonstrated that only a minority of functional GABAergic synapses in the pyramidal cell layer expressed VGLUT3.

We next assessed whether VGLUT3 and VIAAT were co-localized within the same SVs. After immunoisolation of VGLUT1- or VIAAT-containing SVs, VGLUT3 was recovered only from VIAAT-isolated SVs (Figures 1G,H). Moreover, VIAAT-positive SVs were able to accumulate [³H]glutamate (VIAAT, n = 4; IgG, n = 4, Student t-test, unpaired, P = 0.0028, t = 2.65, Figure 1I). These results demonstrated that VGLUT3 was co-expressed with VIAAT in SVs and that these SVs could readily concentrate glutamate, and might have the ability to release both glutamate and GABA.

GABAergic inputs inhibit individual pyramidal cells and synchronize large groups of excitatory neurons. PV-positive basket cells may play an important role in generating hippocampal rhythms, and CCK-positive basket cells may modulate network activity in a behaviorally and environmentally contingent manner (Bliss and Collingridge, 1993; Klausberger and Somogyi, 2008; Lapray et al., 2012). We examined whether the absence of VGLUT3 significantly affected theta rhythms and long-term synaptic plasticity, two important phenomena involved in learning, memory and emotional processing.

Theta oscillations were recorded in the CA1 area of the ventral hippocampus from a complete preparation in vitro (Figures 2A,B) (Khalilov et al., 1997; Amilhon et al., 2015). These theta oscillations are generated intrinsically in the hippocampus by the delicate interactions between interneurons and pyramidal cells (Khalilov et al., 1997; Amilhon et al., 2015). The frequency of theta oscillations was significantly lower in the VGLUT3^-/- mice than in WT littermates throughout recordings during the first 5–10 min [WT (n = 12): 7.4 ± 0.6 Hz and VGLUT3^-/- (n = 14): 4.8 ± 0.4 Hz, Mann–Whitney test, P = 0.0002, U = 9.5] or recordings during the last 95–100 min (WT: 4.4 ± 0.3 Hz and VGLUT3^-/-: 3.5 ± 0.3 Hz, Mann–Whitney test, P = 0.0054, U = 29.5, Figure 2B). However, no significant differences in theta power between groups were observed (not shown). Therefore, the absence of VGLUT3 impaired the excitatory-inhibitory balance required for normal oscillatory rhythmic activity in the hippocampus.

Long-term plasticity of synaptic transmission is a molecular and cellular process that underlies learning and memory through the fine-tuning of synaptic strength (Cooke and Bliss, 2006; Omiya et al., 2015). Using CA1 field potential recordings and extracellular stimulations we examined whether VGLUT3 loss modulates the expression of LTP and LTD. Under conditions of normal glutamate and GABA transmission, tetanic high-frequency stimulation (HFS; see Materials and Methods) of Schaffer’s collateral-CA1 connections led to a long-term potentiation of +24 ± 5% of field EPSP (fEPSP) slope in WT mice (n = 9, Wilcoxon test, P = 0.0039, W = -45, Figure 2C). In VGLUT3^-/- mice no LTP was significantly induced by HFS stimulation (+5 ± 7%, n = 10, Wilcoxon test, P = 0.4922, W = -15, Figure 2C). Therefore, the absence of VGLUT3 resulted in a significant difference in HFS-induced LTP between genotypes (Mann–Whitney test, P = 0.0350, U = 19).

When fast GABAergic transmission was suppressed with application of the GABA_A receptor antagonist bicuculline (5 μM), in WT mice HFS-induced LTP was of +41 ± 2% (n = 6, Wilcoxon test, P = 0.0313, W = -21, Figure 2D) and in VGLUT3^-/- mice was of: +45 ± 10% (n = 6, Wilcoxon test, P = 0.0313, W = -21, Figure 2D). Hence, in the absence of GABAergic transmission, HFS-LTPs were of comparable amplitude in WT and VGLUT3 null mice (Mann–Whitney test, P = 0.3095, U = 11).

Theta burst stimulation (TBS; see Materials and Methods) is a pattern of activity of milder intensity but considered more physiologically relevant than tetanic HFS. We wondered whether the loss of VGLUT3 would also affect TBS-induced LTP. TBS induced a significant LTP in WT mice of +16,6 ± 5% (n = 9, Wilcoxon test, P = 0.0039, W = -49, Figure 2E) as well as in VGLUT3^-/- mice: +16,8 ± 4% (n = 8, Wilcoxon test, P = 0.0078, W = -36, Figure 2E). The LTP elicited in both groups were very similar in amplitude (Mann–Whitney test, P > 0.99, U = 36). These results suggest that the loss of VGLUT3 affected specific patterns of activity rather than blunting synaptic plasticity in general.

To further investigate this hypothesis, we investigated how VGLUT3 deletion altered LTD. LTD is generally induced by low frequency patterns of inputs between 0.5 and 10 Hz. The application of a low-frequency stimulation of 900 pulses at 1 Hz (LFS-1 Hz) led to a synaptic depression in WT group of -14,4 ± 2,6% (n = 6, Wilcoxon test, P = 0.0313, W = 21, Figure 2F). It also induced a synaptic depression in VGLUT3^-/- mice: -9,8 ± 3,3% (n = 9, Wilcoxon test, P = 0.0195, W = 39, Figure 2F). LTDs were comparable in amplitude between WT mice and VGLUT3^-/- mice (Mann–Whitney test, P = 0.3153, U = 18). We then applied a low-frequency stimulation consisting of 900 pulses at a 3 Hz frequency (LFS-3 Hz). After LFS-3 Hz stimulation, a significant depression of -13,1 ± 3,6% was observed in WT mice (n = 7, Wilcoxon test, P = 0.0313, W = 26, Figure 2G). However, LFS-3 Hz application failed to induce significant LTD in the VGLUT3^-/- mice: -2,6 ± 2,6% (n = 7, Wilcoxon test, P = 0.2969, W = 14, Figure 2G). The difference between LTD levels was statistically significant at 3 Hz (Mann–Whitney test, P = 0.0379, U = 8).

The theoretical curve shown on Figure 2H summarizes the long-term plasticity results. Levels of LTP and LTD were plotted in regard to the intensity of the stimulation protocol. These results suggest that the removal of VGLUT3 induces a metaplastic shift of synaptic plasticity toward lower frequencies of induction. Taken together, these results demonstrate the impact of VGLUT3 on the excitatory-inhibitory balance and demonstrate that VGLUT3 is an important player in the regulation of theta oscillations and synaptic plasticity.

We next investigated how VGLUT3 regulated the excitatory-inhibitory balance of the hippocampal network. In CCK-basket cells, VGLUT3-dependent glutamate could act directly as a neurotransmitter of excitatory synaptic transmission and/or indirectly as a neuromodulator of the inhibitory synaptic transmission.

To test VGLUT3-dependent glutamate activity on basal glutamatergic synaptic transmission, we recorded fEPSPs at various stimulation intensities. We built up an input/output curve of fEPSP slopes obtained in the CA1 region of WT and VGLUT3^-/- mice (Figure 3A). The average maximal amplitude of fEPSPs obtained in each genotype was similar (Mann–Whitney test, P = 0,6182, U = 43, not shown). When normalized to the maximal amplitude of each slice, the mean slope recorded at each intensity of stimulation was also similar between WT and VGLUT3^-/- mouse groups [RM-ANOVA, P = 0.6765, F(1,18) = 0,1799, Figure 3A]. This result suggests that VGLUT3-dependent glutamate did not directly regulate fast glutamatergic transmission.

We then investigated whether VGLUT3-dependent glutamate was a neuromodulator of GABAergic transmission in CA1 pyramidal neurons (Figures 3B–G). First we measured mIPSCs under the blockade of action potentials with TTX and of fast synaptic AMPA/Kainate and NMDA glutamate receptors with DNQX (20 μM) and AP-5 (25 μM), respectively. In VGLUT3^-/- mice, the mean mIPSCs amplitude was reduced by 27 ± 7% (WT: -20.2 ± 1.4 pA, n = 27 and VGLUT3^-/-: -15.8 ± 1.4 pA, n = 25, Mann–Whitney test, P = 0.0200, U = 210, Figures 3B–E). This observation suggested that a vesicular synergy between glutamate and GABA loading in SVs was lost in the absence of VGLUT3 (El Mestikawy et al., 2011).

In addition, the frequency of mIPSCs was increased by 32 ± 7.6% in the VGLUT3^-/- mice (1.8 ± 0.2 Hz in VGLUT3^-/- compared to 1.2 ± 0.1 Hz in WT littermates, Mann–Whitney test, P = 0.0438, U = 217.5, Figure 3F). These results indicate that VGLUT3 regulates both amplitude and frequency of GABA mIPSCs.

To further understand the changes observed in GABAergic synaptic transmission, we quantified electrically evoked IPSCs (eIPSCs) in WT and VGLUT3^-/- mice. Evoked IPSCs were elicited monosynaptically by the direct activation of GABAergic somata and terminals while the fast synaptic AMPA/Kainate and NMDA glutamate receptors were blocked with DNQX (20 μM) and AP-5 (25 μM), respectively (Supplementary Figure S1). The input-output curve of GABA eIPSCs in CA1 pyramidal cells showed that the maximal eIPSC amplitude in pyramidal neurons was significantly increased in the VGLUT3^-/- mice (+68 ± 24%, n = 14, relative to WT mice, n = 16, Mann–Whitney test, P = 0.0151, U = 63, Figure 3G).

These results demonstrate that VGLUT3-dependent glutamate acts as a neuromodulator that strongly inhibits GABAergic transmission.

To confirm the GABAergic origin of this VGLUT3-dependent regulatory mechanism, we developed mutant mice selectively lacking VGLUT3 in GABAergic neurons. The exon 2 of Slc17a8, the gene coding for VGLUT3, was flanked by LoxP sites (Figure 4A). VGLUT3^flox/flox mice were bred with VIAAT-CRE mice. In situ hybridization was performed with antisense oligonucleotides targeting exon 2 of VGLUT3 in the hippocampus of WT and VGLUT3^{V IAAT-Cre-flox/flox} mice (Figures 4B,C, left hemisphere). These experiments showed that VGLUT3 exon 2 was absent from the hippocampus of VGLUT3^{V IAAT-Cre-flox/flox} mice.

VGLUT3 protein was then detected by immunoautoradiography. We found a significant decrease of VGLUT3 expression [TWO-WAY ANOVA, P < 0.0001, F(5,48) = 68.15] in VGLUT3^{V IAAT-Cre-flox/flox} mice. In these mutant animals, the loss of VGLUT3 in the entire hippocampus was 42 ± 10% (Bonferroni post-test, p < 0.05, t = 2.792, Figure 4D) and 54 ± 7% for the CA1 region (Bonferroni post-test, p < 0.01, t = 3.363). The analysis of CA1 layers showed specific decreases of VGLUT3 content in the pyramidal layer (-84 ± 3%, Bonferroni post-test, p < 0.001, t = 9.197), the stratum oriens (-59 ± 6%, Bonferroni post-test, t = 2.914, p < 0.05) but not in the stratum radiatum (-32 ± 13%, Bonferroni post-test, p > 0.05, t = 1.456) nor the hippocampal fissure (-5 ± 15%, p = Bonferroni post-test, p > 0.05, t = 0.499). These results were further confirmed at higher magnification with immunofluorescent detection of VGLUT3 in the hippocampus of WT mice and VGLUT3^{V IAAT-Cre-flox/flox} mice (Figure 4E).

Input-output curves of eIPSCs were then done to compare the GABAergic synaptic transmission in VGLUT3^{V IAAT-Cre-flox/flox} mice and in WT littermates. The maximal amplitude of eIPSCs was significantly increased by 55.2 ± 22.7% in the VGLUT3^{V IAAT-Cre-flox/flox} mice (Mann–Whitney test, P < 0.0028, U = 21, n = 10, Figures 4E,F).

This result demonstrates that VGLUT3-dependent glutamate release by GABAergic basket cells is a potent inhibitory modulator of GABAergic transmission.

As recently documented in the amygdala (Albayram et al., 2016), we found that VGLUT3-positive basket cells from the hippocampus neurons expressed CB1 endocannabinoid receptors (CB1R) (Figure 5A). Activation of CB1R strongly inhibits GABA release (Ferraguti et al., 2005). Hence, we investigated the status of CB1R in mice lacking VGLUT3 (Figures 5A–D). We found that neither CB1R transcripts (Mann–Whitney test, P = 0.4206, U = 8, Figure 5B), nor its binding sites (Mann–Whitney test, P = 0.3095, U = 7, Figure 5C) were modified in the CA1 pyramidal layer of VGLUT3^-/- mice. We next evaluated the effects of WIN55,212-2 (1 μM), a CB1R agonist, on GABA transmission. WIN55,212-2 inhibited GABA eIPSCs by 40 ± 8% in WT mice and by 45 ± 5% in VGLUT3^-/- mice, showing no significant alteration of CB1R signal transduction in the absence of VGLUT3 (Mann–Whitney test, P = 0.7209, U = 28, Figure 5D). These results demonstrate that VGLUT3-dependent inhibitory regulation of GABAergic transmission is independent of the CB1R signaling pathway.

Presynaptic group III mGLURs are found on GABAergic interneuron terminals. Their activation efficiently depresses GABA release onto CA1 pyramidal cells (Shigemoto et al., 1997; Fitzjohn et al., 1998; Somogyi et al., 2003). However, the source of endogenous glutamate responsible for this inhibition is unknown. We hypothesized that the increased GABA transmission observed in VGLUT3^-/- mice (see Figures 3F,G, 4F,G) was caused by the loss of an inhibitory control normally mediated by VGLUT3-dependent glutamate release and mGLUR activation. To test this hypothesis we investigated whether the broad-spectrum mGLUR antagonist LY341495 (75 μM) was able to stimulate GABAergic transmission (Shannon et al., 2005). In accordance with our hypothesis, the amplitude of eIPSCs was markedly increased by LY341495 in the WT mice (+61 ± 21%, n = 10, P < 0.01) but not in the VGLUT3^-/- mice (n = 11, Mann–Whitney test, P = 0.0002, U = 2, Figure 5E). Furthermore, LY341495 (75 μM) increased the frequency of spontaneous mIPSCs only in WT mice (+32 ± 8%, n = 11 compared with -3 ± 5%, n = 11 in VGLUT3^-/- mice, Mann–Whitney test, P = 0.0002, U = 19, Figures 5F,G) but had no effect on the amplitude of mIPSCs in both groups (0 ± 5%, n = 11 in WT group and 3 ± 7%, n = 11 in VGLUT3^-/- group, Mann–Whitney test, P = 0.7802, U = 41). Taken together these results show for the first time that VGLUT3-positive basket cells provide the glutamate necessary for pre-synaptic mGluR activation and the subsequent depression of GABAergic transmission onto CA1 pyramidal cells.

We then pharmacologically explored which subtype(s) of mGluR was (were) involved in this pathway (Figure 5H). Neither the group-I mGluR negative allosteric modulators LY456236 [500 nM (Gasparini et al., 1999)] and MPEP [10 μM (Kingston et al., 1998)] nor the group-II mGluR antagonist LY341495 [5 nM (Thomas et al., 1996)] had significant effect on eIPSCs amplitude in WT mice: 0 ± 7% (n = 6, Student t-test, paired, P = 0.7290, t = 0.37, df = 5) and +2 ± 11% (n = 9, Student t-test, paired, P = 0.5392, t = 0.64, df = 8), respectively. However, the group-III mGluR antagonist MSOP [300 μM (Kogo et al., 2004)] increased eIPSCs amplitude by 35 ± 10% (n = 8, Student t-test, paired, P = 0.0087, t = 3.60, df = 7). As expected, MSOP did not increase eIPSCs amplitude in VGLUT3^-/- mice (n = 7, Student t-test, paired, P = 0.8324, t = 0.22, df = 6). These results strongly suggest that VGLUT3-dependent glutamate inhibit GABAergic transmission by activating pre-synaptic group-III mGluRs.

To further confirm this hypothesis we examined whether mGLURs and VGluT3 were co-localized on perisomatic basket cell terminals in the CA1 pyramidal layer. Among the group-III mGluRs, mGluR4, mGluR7 and mGluR8 are known to be present on presynaptic elements of the hippocampus (Shigemoto et al., 1996; Fremeau et al., 2002; Somogyi et al., 2003). Using immunofluorescence, we found that mGluR4a (but not mGLUR7) was expressed in a subset of VGLUT3-immunopositive interneurons (Figures 5I–L). Indeed, some pyramidal cells were surrounded by VGLUT3- and mGluR4-positive boutons (Figure 5L). Observations along the Z-axis confirmed that VGLUT3 and mGluR4 were co-localized and not superimposed (Figure 5L3). Interestingly, most of the VGLUT3-positive boutons surrounding pyramidal cells co-expressed mGLUR4 (stars); however, in other cases, VGLUT3- and mGluR4-positive terminals were juxtaposed (arrow heads). Taken together, these results suggest that VGLUT3-containing terminals are physically co-localized or closely localized with mGluRs to modulate the release of GABA (proposed model Figure 6).