Decrease of SYNGAP1 in GABAergic cells impairs inhibitory synapse connectivity, synaptic inhibition and cognitive function

Haploinsufficiency of the SYNGAP1 gene, which codes for a Ras GTPase-activating protein, impairs cognition both in humans and in mice. Decrease of Syngap1 in mice has been previously shown to cause cognitive deficits at least in part by inducing alterations in glutamatergic neurotransmission and premature maturation of excitatory connections. Whether Syngap1 plays a role in the development of cortical GABAergic connectivity and function remains unclear. Here, we show that Syngap1 haploinsufficiency significantly reduces the formation of perisomatic innervations by parvalbumin-positive basket cells, a major population of GABAergic neurons, in a cell-autonomous manner. We further show that Syngap1 haploinsufficiency in GABAergic cells derived from the medial ganglionic eminence impairs their connectivity, reduces inhibitory synaptic activity and cortical gamma oscillation power, and causes cognitive deficits. Our results indicate that Syngap1 plays a critical role in GABAergic circuit function and further suggest that Syngap1 haploinsufficiency in GABAergic circuits may contribute to cognitive deficits.

L ong-term changes in the strength of synaptic transmission are thought to be critical both during brain development and for learning and memory throughout life. The Ras family GTPases, their downstream signalling proteins and upstream regulators are key biochemical cascades modulating synaptic plasticity. SYNGAP1 codes for a GTPase-activating protein (GAP) that physically interacts with the small GTPase Ras, which in turn acts in a cycle as a molecular switch with an active GTP-bound form and an inactive GDP-bound form 1,2 . Ras has a slow intrinsic GTPase activity, and GAPs such as SYNGAP1 negatively regulate Ras by enhancing the hydrolysis of GTP to GDP. The importance of SYNGAP1 in synaptic plasticity is exemplified by the fact that de novo mutations in the SYNGAP1 gene cause moderate or severe intellectual deficiency (ID) [3][4][5][6][7][8][9] . SYNGAP1 function has been mainly studied in excitatory neurons. For example, in primary neuronal cultures, SYNGAP1 functions to limit excitatory synapse strength by restricting the expression of the AMPA receptor (AMPAR) at the postsynaptic membrane 1,2,10,11 . In mice, Syngap1 haploinsufficiency causes abnormal synaptic plasticity as well as behavioural abnormalities and cognitive deficits [12][13][14][15] . Syngap1 þ / À mice are also characterized by enhanced excitatory synaptic transmission early in life and the premature maturation of glutamatergic synapses 16,17 . Thus, it has been proposed that glutamatergic synaptic alterations represent the main contributing factor for the occurrence of cognitive and behavioural deficits 16,17 .
During healthy cortical network activity, excitation is precisely balanced by GABAergic inhibition. Inhibitory activity not only regulates circuit excitability, but also restricts the temporal window for integration of excitatory synaptic inputs and resulting spike generation, thereby facilitating an accurate encoding of information in the brain 18 . In addition, GABAergic cells are implicated in generating temporal synchrony and oscillations among networks of pyramidal neurons, which are involved in complex cognitive functions, such as perception and memory 19,20 . Furthermore, GABAergic inhibition plays a critical role in modulating developmental plasticity in the young brain 21 . Highlighting the importance of GABA interneurons in cognitive functions, cortical circuits in several mouse models of ID and autistic-like behaviour show excitation/inhibition imbalance, which is due to alterations in glutamatergic or GABAergic neurotransmission, or more often, in both 16,[22][23][24][25][26][27] . Whether and to what extent Syngap1 haploinsufficiency affects GABAergic cell circuits, thus contributing to excitation/inhibition imbalance and cognitive abnormalities remains unclear.
Here, we examined the specific contribution of Syngap1 to the formation of perisomatic innervations by parvalbumin-positive basket cells, a major population of GABAergic neurons, by single-cell deletion of Syngap1 in cortical organotypic cultures. In addition, we generated mice with specific deletion of Syngap1 in GABAergic neurons generated in the medial ganglionic eminence (MGE) to assess its role in the establishment of mature GABAergic connectivity and mouse cognitive function in vivo. We found that Syngap1 strongly modulated the formation of GABAergic synaptic connectivity and function and that MGE cell-type specific Syngap1 haploinsufficiency altered cognition.
GABAergic circuits comprise an astonishing variety of different cell types, exhibiting differences in molecular, morphological and electrophysiological properties 19 . These differences are particularly important in the light of recent discoveries suggesting that different GABAergic cell types are recruited by different behavioural events 19 . Among the different GABAergic neuron subtypes, the parvalbumin-expressing (PV þ ) basket cells comprise the largest subpopulation in cortical circuits 19 . Each PV þ basket cell innervates hundreds of neurons, with large, clustered boutons targeting the soma and the proximal dendrites of postsynaptic targets, an optimal location to control timing and frequency of action potential generation 19,36 . Such distinct features of PV þ basket cell innervations are achieved during the first postnatal month in rodents and are modulated by neural activity levels 35,[37][38][39] . We found that almost the totality of PV þ basket cells express SYNGAP1 in dissociated neuronal cultures ( Supplementary Fig. 1b) and thus we sought to investigate whether Syngap1 plays a role in the formation of the innervation of PV þ basket cells, by inducing single-cell Syngap1 deletion in cortical organotypic cultures.
To reduce Syngap1 expression in isolated PV þ basket cells and simultaneously label their axonal arbours at synaptic resolution, we used a previously characterized promoter region P G67 (ref. 37) to express either the Cre recombinase together with GFP (P G67 -GFP/Cre) or GFP alone (P G67 -GFP; control basket cells) in cortical organotypic cultures prepared from Syngap1 flox/flox mice 35,[37][38][39][40] . In organotypic cultures, PV þ basket cells initially display very sparse and simple axons, which develop into complex and highly branched arbours in the following 4 weeks, recapitulating the in vivo situation 37,38 . We chose to induce Syngap1 deletion in PV þ basket cells at Equivalent Postnatal day 10 (EP10 ¼ cultured prepared at Postnatal day 4 þ 6DIV) and collect the cultures at EP24 because extensive and stereotyped maturation of PV þ basket cell innervations occurs during this time window 37,[37][38][39][40][41] .
We investigated two aspects of PV þ basket cell axon innervation: (1) the extent of perisomatic innervation around single neuronal somata (terminal branching and perisomatic bouton density) and (2) the fraction of potentially innervated cells in the field (percentage of innervation). By studying the localization of pre-and post-synaptic markers and performing electron microscopy, we have previously shown that the vast majority of GFP-labelled boutons in our experimental condition represent presynaptic terminals 35,37,42 .
Whereas control basket cells at EP24 showed complex perisomatic innervations (Fig. 1a,b), Syngap1 knockdown in single-basket cells from EP10-24 induced a significant reduction in the number of both axonal branching and synaptic boutons innervating the target neurons (NeuN-positive cells; Fig. 1c-f: boutons/soma, 7.4±0.3 for control versus 5.2±0.5 for basket cells transfected with P G67 -GFP/Cre; Students t-test P ¼ 0.002). On the other hand, the percentage of potentially innervated neurons were not significantly different between the two groups ( Fig. 1g: 71%±2 for control versus 66%±2 for basket cells transfected with P G67 -GFP/Cre; Students t-test, P ¼ 0.125), suggesting that Syngap1 deletion specifically affected local synapse formation of PV þ basket cells but not their overall axonal growth. Syngap1 knockout at later age (EP16-24) caused a similar decrease in perisomatic bouton density ( Supplementary Fig. 2, boutons/soma, 7.4±0.3 for control versus 4.7±0.5 for basket cells transfected with P G67 -GFP/Cre; Students t-test P ¼ 0.00023). Altogether, these data demonstrate that Syngap1 promotes the formation of PV þ basket cell innervations in a cell-autonomous manner.
MGE-specific Syngap1 knockdown impaired PV cell connectivity. Our data suggest that reducing Syngap1 expression has a direct impact on the formation of PV þ basket cell innervations in vitro. To investigate the role of Syngap1 in PV þ cell circuits in vivo, we generated mice that were heterozygous or homozygous for the Syngap1 flox allele and hemizygous for the Tg(Nkx2.1-Cre) transgene. This approach allowed the conditional deletion of Syngap1 in cortical, 43 hypothalamic 43 and mesencephalic 44 (striatum) GABAergic interneurons originating from the MGE as early as embryonic day 10.5 and throughout adulthood 45 . In mouse, Nkx2.1-expressing MGE precursors produce most of PV þ and Somatostatin (SST) þ cortical -G F P p G A D 6 7 -C R E / G F P p G A D 6 7 -G F P p G A D 6 7 -C R E / G F P   Fig. 3; percentage of GFP þ cells expressing PV or SST, 74 ± 7% and 25 ± 6%, respectively, n ¼ 3 mice). Further, the large majority of cortical neurons immunopositive for PV also expressed GFP (85 ± 6%, n ¼ 3 mice), as previously reported 47 .
Together with the observation that the number of cortical perisomatic boutons formed by PV þ baskets was reduced in Tg(Nkx2.1-Cre);Syngap1 flox/ þ (Figs 2 and 3), this result demonstrated that Syngap1 haploinsufficiency impaired GABAergic synapse formation and function of MGE-derived interneurons.
MGE-specific Syngap1 knockdown affected gamma oscillations. PV þ cells represent the large majority of cortical GABAergic interneurons generated by MGE-derived precursors. Correlative, causal and computational evidence indicates that gamma oscillations in the 30-80 Hz range depend on synchronous activity of PV þ cells. 48,49 Since we found that PV þ cell connectivity was reduced in both cortex and hippocampus of Tg(Nkx2.1-Cre);Syngap1 flox/ þ mice (Fig. 2), we hypothesized that PV þ cell synaptic impairment would lead to changes in gamma oscillatory activity in these mice.
Mice of all genotypes spent significantly more time with an unfamiliar mouse (stranger 1) relative to the empty wire cage ( Supplementary Fig. 5, two-way ANOVA Po0.05), indicating that social approach behaviour was normal in Syngap1 þ / À and Tg(Nkx2.1-Cre);Syngap1 flox/ þ relative to their respective control littermates. Next, we tested whether social novelty preference was affected in these mice by assessing the preference for a second stranger mouse. Syngap1 þ / þ and Syngap1 flox/ þ mice spent significantly more time in the chamber containing the novel mouse (stranger 2) than in the chamber with the familiar one (stranger 1; Fig. 7c Finally, we tested spatial working memory by measuring spontaneous alternation in the T-maze. Spontaneous alternation was strongly altered in both Syngap1 þ / À and Tg(Nkx2.1-Cre); Syngap1 flox/ þ relative to their respective control littermates ( Fig. 7e; 67±6% for Syngap1 þ / þ versus 41±11% for Syngap1 þ / À , Student's t-test P ¼ 0.035, and 64 ± 7% for Syngap1 flox/ þ versus 33.333±11% for Tg(Nkx2.1-Cre); Syngap1 flox/ þ , Student's t-test P ¼ 0.038), suggesting impaired spatial working memory in both mutant lines. On the other hand, we found that Tg(Nkx2.1-Cre) mice showed no statistical difference in locomotion, anxiety, spontaneous alternation and social recognition behaviour ( Supplementary Fig. 6). All together, these data show that specific inactivation of Syngap1 in MGE-derived interneurons recapitulates at least in part the behavioural and cognitive alterations observed in mice with germ-line Syngap1 haploinsufficiency.

Discussion
Our data show that Syngap1 is required for the proper terminal axonal branching and bouton formation of cortical GABAergic PV þ basket cells, both in organotypic cortical cultures and in vivo. Further, MGE-restricted Syngap1 haploinsufficiency causes deficits in cortical evoked-inhibitory synaptic transmission and gamma oscillations and specific impairments in social novelty preference and working memory, in adult mice. In contrast, a recent study did not observe any cognitive deficits in Gad2-cre;Syngap1 flox/ þ mice 17 . One possible explanation for this discrepancy is that Nkx2.1 and Gad2 may be expressed at different developmental time points. In fact, Nkx2.1 is already expressed in all MGE-derived GABAergic interneurons by E10.5. (ref. 45) On the other hand, while Gad2 transcription starts early during embryogenesis 50 (around E11.5), it is possible that not all GABAergic cells express it at the same time. For example, GAD65, which is coded by Gad2, is barely expressed before P6 in rodent somatosensory cortex 51 . Syngap1 haploinsufficiency appears to disrupt the function of excitatory circuits mainly by affecting their early synaptic development 16,17 . Our observations suggest that Syngap1 is also required during the early phases of the development of GABAergic circuit connectivity.
Another major difference is that, while Gad2 is expressed by most GABAergic interneurons, Nkx2.1 is expressed only by interneurons derived from the MGE, which include PV þ and SST þ interneurons but not, for example, those positive for the vasoactive intestinal polypeptide (VIP). Recent results demonstrate that VIP-expressing neurons form a disinhibitory microcircuit that is conserved across cortical regions 52 . In particular, VIP interneurons tend to inhibit most SST þ and a fraction of PV þ interneurons. Inhibition of these interneurons in turn disinhibits principal cells in vivo, providing a form of gain control. 52 It is possible that removing Syngap1 in all interneurons may allow homeostatic synaptic adjustments that are not otherwise engaged when Syngap1 is removed only in MGE-derived interneurons. It will therefore be important to assess whether and how Syngap1 regulates the synaptic connectivity of distinct GABAergic interneuron types other than PV þ basket cells.
Disruption of excitatory/inhibitory balance is emerging as a common theme for many neurodevelopmental disorders 53,54 . However, the critical challenge is to understand how specific circuit alterations resulting from mutations affect the development of cognitive abilities. Our study suggests new avenues of research for the understanding of mechanisms involved in SYNGAP1-related ID. For example, our results showed that Syngap1 plays a role in the formation of perisomatic synapses by PV þ basket cells. GABAergic neurotransmission from these cells promotes network gamma oscillations 48,55 . Cortical gamma oscillations increase in power during tasks that require complex processing of sensory information, attention, working memory and cognitive control, suggesting that gamma oscillations are crucial for cognition 56 . Interestingly, substantial evidence indicates that individuals with specific psychiatric disorders, such as schizophrenia or autism, show lower power of gamma oscillations induced during the performance of cognitive tasks compared with healthy individuals 57,58 . Here, we showed that MGE-specific Syngap1 haploinsufficiency leads to reduced gamma oscillation power during exploration, which in turn may contribute to the occurrence of cognitive deficits. It will be interesting to explore the presence of alterations in brain oscillations in patients carrying SYNGAP1 mutations. Furthermore, cortical GABAergic inhibition, in particular the one mediated by PV þ cells, has been suggested to play a key role in critical period plasticity in the developing brain 21,59 . Critical periods represent heightened epochs of brain plasticity during childhood, during which experience can produce permanent, large-scale changes in neuronal circuits. By regulating critical period plasticity, PV þ cell-mediated inhibition may influence how experience shapes brain wiring during early life. As a consequence, altered PV cell connectivity and function could drive cortical neuronal circuits along abnormal developmental trajectories, which could then contribute to cognitive dysfunction. Ultimately, the understanding of SYNGAP1 function in specific inhibitory and excitatory circuits might lead to the development of tailored therapies.
MGE-derived interneurons include PV þ basket cells and SST þ interneurons 60 , thus alterations in the synaptic connectivity of this latter group may also contribute to the cognitive deficits that were observed in Tg(Nkx2.1-Cre); Syngap1 flox/ þ mice. Further, it is very likely that altered GABAergic signalling results in either causal or compensatory alterations in non-GABA circuits as well, which can further play a role in cognitive dysfunctions in mutant mice. The main challenges will be to understand when these alterations arise and whether and to what extent cognitive deficits are due to either altered neural circuit development leading to diverse changes in brain and behaviour and/or a state of continuous synaptic dysfunction in adulthood.
Haploinsufficiency of the NF1 gene, which also codes for a RasGAP, causes neurofibromatosis type 1, a complex neuro-cutaneous condition that is associated with cognitive impairment and autism spectrum disorders. Decrease of Nf1 induces deficits in hippocampal-dependent spatial learning, amygdala dependentsocial learning and prefrontal cortex-dependent working memory by increasing GABA release in these different areas of the brain through the activation of the Ras-ERK pathway [61][62][63][64] . The increase of Ras thus appears to produce opposite effects on GABA neurotransmission depending upon whether it is associated with Syngap1 or Nf1 haploinsufficiency. How can we explain this paradox? One possibility is that Syngap1 and Nf1 affect Ras activity at different time points during the development of GABAergic circuits. As discussed above, Syngap1 haploinsufficiency might impact the establishment of GABAergic connectivity by acting during early synaptic development. Nf1 might on the other hand modulate GABA release in mature synapses. Alternatively, it is possible that SYNGAP1 and the Nf1 product, neurofibromin, function in distinct subcellular compartments of GABAergic neurons and thus regulate different processes. SYNGAP1 is associated with NMDA receptor complexes at the postsynaptic membrane of glutamatergic neurons 1,2,29 , raising the possibility that it also interacts with such receptors in GABAergic neurons 65 . In contrast, neurofibromin has been shown to localize to smooth vesiculotubular elements, cisternal stacks and multivesicular bodies in the cell body and dendrites of Purkinje cells, but not with the plasma membrane 66 . All together, these observations suggest that SYNGAP1 and neurofibromin modulate the development and function of GABAergic cells by regulating differently the spatial and/or temporal activity of Ras.
In summary, our study suggests that Syngap1 haploinsufficiency affects GABAergic circuit connectivity and function, which in turn may contribute to cognitive alterations. Further studies will be required to determine whether pharmacological interventions during development can rescue these deficits. Primary cortical neuron culture. Cortical neurons from E18 rat embryos were dissociated mechanically and plated on dishes treated with poly-D-lysine (0.1 mg ml À 1 ; Sigma-Aldrich, St. Louis, MO). Neurons were cultured overnight in attachment medium: MEM (Invitrogen) supplemented with 1 mM sodium pyruvate (Invitrogen), 0.6% D-glucose (Sigma-Aldrich) and 10% horse serum. The medium was replaced the next day with maintenance medium: Neurobasal-A medium (Invitrogen) supplemented with 2% B27 (Invitrogen) and 1% L-glutamine (Invitrogen). The maintenance medium was replaced with fresh medium at 3DIV. Subsequently, the medium was replaced every 6 days with maintenance medium until collecting the cells. Cultures were incubated in a humidified incubator at 37°C with a 5% CO 2 -enriched atmosphere during 21 days then were fixed in 100% methanol ( À 20°C) for 5 min. The fixation was stopped by adding 0.3 M glycine for 5 min. Neurons were permeabilized with 0.25% Triton in phosphate buffer pH 7.4, washed and blocked for 0.5 h in 3% BSA at room temperature. Cultures were then incubated for 1 h with anti-GAD67 antibody (mouse, 1:200, Abcam #26116) or anti-PV antibody (mouse, 1:500, Sigma, #P3088) and anti-SYNGAP1 antibody (rabbit, 1:200, Abcam #3344), followed by incubation with Alexa 488-conjugated IgG (goat anti-rabbit, 1:400, Invitrogen #A-11008) and Alexa 555-conjugated IgG (goat anti-mouse1:400, Invitrogen #A-21422) for 0.5 h at room temperature (RT), then mounted in Prolong (Invitrogen #P-7481) and stored at 4°C. Images were acquired using a confocal microscope (LEICA TCS SPE or Leica TSC SP8).

Analysis of basket cells in vitro.
Slices were fixed and immunostained as in ref. 40. To visualize basket cell targeted neurons, we used anti-NeuN (monoclonal antibody, 1:400; Millipore Bioscience Research Reagents, #MAB377), followed by incubation with Alexa Fluor 633-conjugated goat anti-mouse IgG (1:400; Invitrogen A-21053). Only one basket cell was acquired from each organotypic culture. Non-overlapping fields of axonal innervation from well-isolated basket cells were acquired with a Â 63 glycerol-immersion objective (NA 1.3; Leica) using a confocal microscope (LEICA TCS SPE) and them analyzed with Neurolucida software (MicroBrightField). Axon branch complexity around a single pyramidal cell soma was quantified as the average number of intersections between a basket cell axon and each one of 9 Sholl spheres traced with 1 mm increment from the centre of the pyramidal cell soma. Bouton density around each pyramidal cell soma was defined as the total number of GFP þ boutons identified in a radius of 9 mm from the centre of the pyramidal cell soma. For each basket cell, we analyzed 15-20 pyramidal neurons. The percentage of pyramidal somata innervated by basket cells was defined in a confocal stack by the number of somata contacted by the GFP þ axon divided by the total number of somata. At least two fields were imaged for each basket cell and the results were averaged. For these experiments, the number of basket cells was used as independent replicate (N). Data are presented as mean ± s.e.m. Imaging and quantification were performed independently by two different operators. All quantification was done blind to the genotype or the experimental conditions.
Analysis of perisomatic innervation in vivo. P45 mice were anesthetized and perfused transcardially with 4% paraformaldehyde (PFA 4%) in phosphate buffer, pH 7.4. Brains were removed, immerged in PFA 4% overnight then cryoprotected in 30% sucrose (in sodium-phosphate buffer pH 7.2) and finally embedded in optimal cutting temperature Tissue Tek. Coronal sections (40 mm) from Syngap1 þ / À , Tg(Nkx2.1-Cre);Syngap1 flox/ þ and their control littermates of both sexes were cut using a cryostat (Leica VT100). Brain sections were blocked in 10% normal goat serum (NGS) and 0.3% Triton X-100 for 2 h at RT. Slices were then incubated overnight at 4°C with anti-Parvalbumin (rabbit 1:8,000, Swant, #PV25) and anti-NeuN antibodies (mouse 1:400, Chemicon, #MAB377) followed by incubation with Alexa 488-conjugated anti-rabbit IgG and Alexa 633-conjugated anti-mouse IgG (1:400, Invitrogen, A-11008 and A-21053, respectively) for 2 h at RT. After washing, slices were mounted in Vectashield mounting medium (Vector). Confocal images were taken using a Â 63 oil objective (Leica NA 1.3). Scans from each channel were collected in multiple-track mode and subsequently merged. For each animal, three to four coronal sections containing the somatosensory cortex and an equal number containing the CA1 region of the hippocampus were imaged. Stacks were acquired with 1 mm steps, exported as TIFF files and analyzed with ImageJ software. We analyzed the intensity of PV-positive puncta around layer 5 pyramidal cell somata because the size of their apical dendrite rendered their identification fairly easy, therefore reducing the variability in the analysis. For each identified neuronal soma, intensity was quantified in the optical section where the soma profile was larger, in particular an area 2 mm distal from the edge of a neuron profile was traced and outlined, and staining intensity within this area was measured. Results of at least 30 NeuN-positive somata from the somatosensory cortex of each mouse were averaged. For these experiments, the number of animals was used as independent replicate (N). Data are presented as mean ±s.e.m. Imaging and quantification were performed independently by two different operators. All quantification was done blind to the genotype. Viral vector and stereotaxic injections. The pAAV.CAGGS.flex.ChR2.tdTomato. SV40 was a gift from Scott Sternson (Addgene plasmid # 18917) and was produced as AAV2/9 serotype by the UPenn Vector Core Facilities (University of Pennsylvania, Philadelphia). Mice aged postnatal day 24-28 (P24-P28) were used for all surgeries. Bilateral viral injections were performed at the following stereotaxic coordinates: 1.5 mm from bregma, 1.5 mm lateral from midline and 0.70 mm vertical from cortical surface. Surgical procedures were standardized to minimize the variability of AAV injections. To ensure minimal leak into surrounding brain areas, injection pipettes remained in the brain for B5 min after injection before being slowly withdrawn. The final injected volume for the AAV was 0.5 ml. We waited 2 weeks for maximal viral expression. The titre for the viruses was B2.69 Â 10 12 genome copies per ml.
Electrophysiology. All the following experiments were done by investigators blind to the genotype until after data analysis was completed.
Optogenetics: Slices were prepared as described above. Whole-cell patch clamp recordings were obtained from visually identified pyramidal neurons in somatosensory cortex layer 5 using borosilicate pipettes (3)(4)(5) and an upright microscope (Olympus BX50WI), equipped with a water immersion long-working distance objective ( Â 40, Nomarski optics) and an infrared video camera. Care was taken to record from brain slices showing similar dense AAV tranfection, that is, tdTomato expression. The internal solution contained 120 mM CsMeSO3, 5 mM NaCl, 1 mM MgCl 2 , 10 mM HEPES, 10 mM diNa-phosphocreatine, 2 mM ATP-tris, 0.4 mM GTP-tris, 0.5 mM spermine and 2 mM QX-314-Cl. Light-evoked IPSCs were recorded by holding the cell at 0 mV in the presence of 10 mM CNQX and 50 mM DL-APV. IPSCs were evoked by brief pulses of blue light (470 nm, custom-made LED system) delivered to the whole slice every 30 s via a light guide positioned above the slice. Data was acquired using a Multiclamp 700A amplifier (Molecular Devices) and digitized using Digidata1440A and pClamp 10 (Molecular Devices). Recordings were low-pass filtered at 3 kHz and digitized at 20 kHz. Series resistance was regularly monitored during experiments, and data were included only if the holding current and series resistance were stable. Light-evoked IPSCs were analyzed using Clampfit 10 (Molecular devices). For statistical analysis, the number of animals was used as independent replicates (N).
EEG recordings and gamma oscillation analysis: To analyze power in the gamma frequency band (30-80 Hz), we monitored P80-P100 mice by continuous video-EEG recordings after implantation of electrodes in the hippocampus and the overlying cortex. The insertion of electrodes was performed under isoflurane anaesthesia, while the head of the mouse was immobilized in a stereotaxic frame. A stainless steel bipolar electrode (Plastics-1 Inc., Roanoke, VA, USA) was positioned at the following coordinates with reference to Bregma: AP ¼ À 1 mm, ML ¼ À 1 mm and DV ¼ 1.5 mm for one electrode in the hippocampus and 0.5 mm for the one in the cortex. Following a 5-day recovery period, EEG was recorded with a Stellate systems 32-channel video-EEG recording unit for 2 weeks. Mice were monitored during 6-hour sessions (afternoon and night). An operator blind to the genotype extracted EEG traces during mouse exploratory behaviour spanning 30-120 s. A total of 10 min of active exploration was selected for each mouse.
EEG recordings from the cortex and hippocampus were first analyzed to detect time periods of strong theta oscillations, which were indicative of stable recordings during which the animal was exploring. To detect theta oscillations, we used the spectrogram method using discrete short-time Fourier transforms. More specifically, data sampled at 200 Hz were first low-pass filtered by a 100th order finite impulse response filter. Spectrograms were then built from the data set and separated into 1 s intervals, to which a Hamming window was applied. The discrete Fourier transforms were then evaluated over 2 14 points with zero padding. Finally, for each time slot, the algorithm identified oscillations with the highest intensity in the theta frequency band (4)(5)(6)(7)(8)(9)(10)(11)(12) and that lasted at least 1 s. The power spectral density was normalized in amplitude from 0 to 1 and equalized with an exponential function with exponent 0.15 to improve its contrast to random noise. This analysis resulted in a list of theta oscillations for each extracted EEG recording. For each time period of theta oscillations, we then calculated the average power in the gamma frequency band. Analysis was performed in Matlab (7.11.0) (The Mathworks, Natick, MA, USA). Results are expressed as mean ± s.e.m. The investigator was blind to the genotype until after data analysis was completed.
Behavioural studies. All animals were kept under a light-dark cycle (12 h light-12 h dark) in a temperature and humidity controlled room. Food and water were available ad libitum. Room lights were kept low during all procedures. For all experiments, a camera was mounted above the arena; images were captured and transmitted to a computer running the Smart software (Panlab, Harvard Apparatus). The sequence of animals tested was randomized by the genotype. Care was taken to test litters containing both the genotypes specific to the breeding. Results are presented as mean ±s.e.m.
Spontaneous open-field activity. The open-field test was performed as previously described 14 . Each subject (P25) was gently placed in the centre of the open-field, allowed to freely explore undisturbed for 10 min, after which the animal was removed, the arena cleaned with 70% ethanol and dried before testing the next animal. Locomotor activity was indexed as the total distance travelled (m).
Elevated plus maze. Experiments were performed as described in ref. 67. The test session lasted 5 min during which the animals (P27) were allowed to explore the maze freely before being removed and returned to their home-cage. The Three-chamber. A three-chamber arena was used to assess the social recognition performance of the mice 68 . The tested animal (P36) was placed in the middle of the central chamber and allowed to explore all the chambers for 10 min. During this habituation session, small wire cages were present, one in each opposite chamber. After habituation, an unfamiliar conspecific of the same sex and age (Stranger 1) was placed inside a small wire cage whereas the other remained empty. The tested animal was allowed to freely explore the three chambers of the apparatus for 10 min. At the end of this 10 min, a new unfamiliar mouse of the same sex and the same age (Stranger 2) was placed in the previously unoccupied wire cage and the tested mouse was examined for an additional 10 min to assess preference for social novelty. Stanger 1 and stranger 2 animals originated from different home cages and had never been in physical contact with the tested mice or between each other. Social novelty preference was evaluated by quantifying the time spent by the tested mice in each chamber during the third 10 min session.
T-maze. A discret-two-trials spontaneous alternation paradigm in a T-maze apparatus was used to assess the working memory 69 . Each test consisted of a two free-choice trials separated by a 1 min inter-trial interval (ITI) for 3 consecutive days. At the beginning of each trial, the animals (P38-40) were placed in the starting arm, allowed to freely move until one arm was chosen. The animals were left in the choice arm for 10 s before being removed and placed in the home-cage for 50 s. Then, the individuals were allowed to perform another free-choice trial. If the mice entered in the opposite arm than the first trial, the responses were considered as an alternation.
Statistical analysis. Data were expressed as mean±s.e.m. unless otherwise specified in the legends. Normality tests were performed for all data analysed. Differences between two groups were assessed with the Student's t-test for normally distributed data or with the Mann-Whitney rank sum test for not-normally distributed data. Differences between multiple groups were assessed with one-way ANOVA with either Dunn's or Tukey's post hoc test for non-normally distributed and normally distributed data, respectively. Two-way ANOVA with Sidak's multiple comparison post hoc test was used for the detection of differences in the novelty social recognition test. Two-way Repeated Measure ANOVA with Bonferroni's multiple comparison post hoc test was used for analysing light-evoked IPSCs, considering N of mice as independent replicates. mIPSC frequency and amplitude were represented with Tukey boxplots (median ± 1.5 Â interquartile range; box represents 1st-3rd quartile). In the same graph, a cross represented mean mIPSC frequency and amplitude values.
In addition, all individual cell values were reported, using different symbols for different animals. mIPSC frequency and amplitude were analysed using a LMM, modelling animal as a random effect and genotype as fixed effect. We used this statistical analysis because we considered the number of mice as independent replicates and the number of cell in each mouse as repeated measures. For gamma oscillation power analysis, averages from the two genotypes were compared with Wilcoxon rank sum tests since values were not normally distributed. Finally, twotailed power analysis was performed for all experimental data, using alpha ¼ 0.05 and desired power ¼ 0.8. Statistical analysis was performed using Prism 5.0 (GraphPad Software), SigmaStat 3.5 (Systat) and the function lmer implemented in the R package lme4 specifically for LMM.
Three animals were excluded from the analysis of electrophysiological recordings because the genotype was ambiguous (n ¼ 1 mouse) or tdTomato fluorescence was not detectable after AAV injection (n ¼ 2 mice).
Data availability. Detailed statistics and data that support the findings of this study are available from the corresponding authors on request.