Hippocampal dentate gyrus coordinates brain-wide communication and memory updating through an inhibitory gating

Distinct forms of memory processing are often causally identified with specific brain regions, but a key facet of memory processing includes linking separated neuronal populations. Using cell-specific manipulations of inhibitory neuronal activity, we discovered a key role of the dentate gyrus (DG) in coordinating dispersed neuronal populations during memory formation. In whole-brain fMRI and electrophysiological experiments, we found that parvalbumin (PV) interneurons in the DG control the functional coupling of the hippocampus within a wider network of neocortical and subcortical structures including the prefrontal cortex (PFC) and the nucleus accumbens (NAc). In a novel object-location task, regulation of PV interneuron activity enhanced or prevented memory encoding and, without effect upon the total number of task activated c-Fos+ cells, revealed a correlation between activated neuronal populations in the hippocampus-PFC-NAc network. These data suggest a critical regulatory role of PV interneurons in the dentate gyrus in brain-wide polysynaptic communication channels and the association of cell assemblies across multiple brain regions.


Introduction
Memory is thought to be encoded through modifications in the weights of synaptic connections. The circuits enabled by these synaptic modifications, and the corresponding activated cell assemblies, form brain-wide memory engrams that hold specific memory (1) .
In the dentate gyrus (DG), a region of the hippocampal formation important for pattern separation and contextual learning (2)(3)(4) , activity-tagging via immediate early gene (c-Fos)dependent expression of channelrhodopsin, has been shown to recall a contextual fear memory upon light activation (5) . However, the same manipulation targeted to the CA1 region of the hippocampus was ineffective (5) , despite this region being required for contextual fear memory encoding and consolidation (6,7) . This dissociation raises the intriguing possibility of a hitherto unknown role of the DG in memory formation, even when the expression of memory requires a systems level interaction of multiple brain regions (8)(9)(10) . Key support for this idea comes from brain-wide imaging techniques which have revealed that long-term potentiation (LTP) of perforant pathway inputs to the DG not only causes measurable change in DG itself, but also enhances the functional coupling of a network of mesolimbic neocortical and subcortical structures important for memory formation (11)(12)(13) . Specifically, changes in activity measured using BOLD show striking changes in frontal cortex and other regions remote from the sites of plasticity.
Studies on memory formation have largely focussed on excitatory synapses and their potentiation following repeated or coincident activation as the fundamental cellular mechanism supporting the associations between information streams (14)(15)(16)(17)(18) . However, the contribution of inhibition has long been proposed as well (19) , with pharmacological interventions increasing inhibitory activity associated with impaired hippocampus-dependent memory acquisition (20,21) and vice versa during mildly decreased inhibitory activity (22)(23)(24). More recent work with cell-specific manipulations and electrophysiological recordings of identified cell populations, have unambiguously demonstrated transient periods of neuronal disinhibition during learning, causally linked to memory encoding and/or expression (reviewed in (25,26) . Even persistent reductions of PV+ cell inhibition in DG, measured as a decrease in PV staining, has been associated with enhanced spatial learning in the watermaze (27) . However, in the same study, a stronger PV+ inhibitory network was shown to develop in the course of memory formation as more stable memories formed (27) .
These data suggest that changes in the apparent tight coupling of excitation and inhibition (E-I) could be a fundamental network property that gates memory formation. If so, transient changes in E-I balance could lead to increased propagation of activity through regionally disbursed cell-assemblies. This would account for the brain-wide changes in BOLD seen in functional imaging.
These considerations led us to investigate the impact of selective manipulations of perisomatically innervating PV+ interneurons in the DG on memory formation and the potential brain-wide network activity that accompanies it. In combined fMRI and electrophysiological experiments, we found that pharmacogenetic control of PV-cell firing can enhance or preclude activity propagation within the hippocampus and also a brain-wide network of cortical and subcortical structures including the PFC and NAc. Disinhibition of DG enhanced the encoding of novel object-location associations whereas increased inhibition blocked it. Cell assemblies in the hippocampus-PFC-NAc network that were activated by learning were also bi-directionally controlled by DG PV cell activity. Interestingly, DG disinhibition that increased memory encoding concomitantly enhanced the functional coupling between c-Fos+ cell assemblies, but did not change the total number of activated granule cells, preserving the sparseness of activation and thus maintaining context specificity. Complementarily, the increase in PV-cell activity in the DG, decorrelated cell assemblies, without affecting their size, and prevented memory updating. Because manipulations of perisomatic inhibition were also shown to leave synaptic plasticity and dendritic integration unaltered in granule cells, our results suggest that PV-cells in DG are a major contributor of a system-level gating mechanism that orchestrates neuronal activity in separated brain regions.
In PV-Gi animals in vivo, evoked potentials in the DG following stimulation of the perforant pathway, the main entorhinal cortex (EC) input to the hippocampus (Fig. 1B), showed a significant increase in the amplitude of the population spike (PS) after CNO (i.p. 1 mg/Kg) administration (Fig. 1C), reflecting the facilitated firing of GC. Conversely, decreased PS amplitude in the DG (Fig. 1C) was recorded in PV-Gq animals upon CNO administration, indicative of the hindered GC firing. These effects were maximal 30 min after CNO administration and remained constant for the 3 hr of the remainder of recording (Supp.  1D) (28) . Importantly, the excitatory postsynaptic potentials (EPSPs), reflecting the synaptic responses and dendritic integration in GC, were unchanged in all groups (Fig. 1D).
This was clearly evidenced in the input-output curves relating synaptic inputs to spike output  and after CNO i.p. injection in Sham (grey), PV-Gi (yellow) and PV-Gq (blue) animals. Insets: representative PS waveforms (scale 2 ms, 4 mV). D, Comparison of PP stimulus-response curves of DG EPSP slope before and after CNO i.p. injection (same colour-code as before). Insets: representative EPSP waveforms (scale 1 ms, 1 mV). E, Input-Output (EPSP vs. PS) curve demonstrating increased (PV-Gi) and decreased (PV-Gq) granule cell output for equal synaptic inputs (same colour-code; non-linear regression fit, shadow represents 95% CI). F and G, LTP induction.

DG PV cells bidirectionally control memory updating
We then investigated whether the gating of GC's output, keeping an intact synaptic and dendritic function, had any impact on memory formation. We used the hippocampaldependent novel object location task (NOL) (29,30) . In this two-phase task, mice first learn new spatial cues (2 identical objects) in an otherwise familiar context during a single 10 min exploration trial. They are then allowed to retrieve the encoded memory of object locations 24 h later in the same familiar context with 1 of the 2 objects displaced. If memory is intact, this elicits a relative exploratory preference towards the moved object (See Methods and Supp. Fig. 2A). Activity of PV cells in the DG was modulated as before (i.p. 1 mg/Kg CNO) either 90 min before the exploration trial (encoding phase), 10 min after the exploration trial when the animal is returned to its home-cage (consolidation phase) or 90 min before the retrieval test 24 h after encoding (retrieval phase) ( Fig. 2A). The results demonstrate a highly significant and bidirectional effect of modulating PV-inhibition in the DG during memory encoding (Fig. 2B), but not during consolidation (Fig. 2C) or retrieval (Fig. 2D). Decreasing perisomatic inhibition during encoding resulted in enhanced performance 24 h later, while increasing the inhibitory tone prevented encoding (Fig. 2B). Locomotor and anxiety measures during field exploration and the elevated plus maze, respectively, were indistinguishable between groups (Supp. Fig. 2C and E). An important control was to repeat the NOL task in the same group of animals and substitute CNO administration by its vehicle (saline) in the encoding phase; comparable memory encoding was observed in all groups ( Fig. 2E).
One hypothesized role for the DG in memory formation is its contribution to pattern separation (2-4) , a function that would theoretically benefit from the known sparse activation of GCs (2,3) . Compatible with this role, enhanced object location discrimination was observed in PV-Gi animals injected with CNO as the difficulty of detecting the magnitude of object displacement was made harder (Fig. 2F). A small environmental change not ordinarily detected in control conditions was noticed and effectively assimilated into memory when PV-cells were inhibited. However, previous work manipulating the activity of somatostatin positive interneurons in the DG showed an enlargement of the recruited GC assembly (31) , which would work against the sparsity of GC activation and their capacity to discriminate patterns. To clarify this issue, we used cellular imaging of activity-dependent c-Fos expression, and asked whether specific PV-interneuron activation or inactivation had an impact on the number of cells recruited in the local assembly. The baseline sensitivity of this assay is shown by the number of c-Fos+ cells in the DG being increased by object exploration in the NOL task compared to those of home-cage mice (Fig. 2G). The logically appropriate comparisons are between the number of c-Fos+ GCs in CNO injected PV-Gi, PV-Gq animals and CNO injected PV-Sham controls that revealed a constant size of the recruited GC assembly across conditions (Fig. 2H). As a control, in the same animals, we validated the successful DREADDs manipulation showing enhanced activation of PV+ interneurons in PV-Gq animals, and vice versa in PV-Gi (Supp. Fig. 3). Thus, although GC spiking probability is radically changed, the number of c-Fos+ cells in the DG seems to be regulated by the synaptic input, unaltered here, rather than overt neuronal firing. In addition, neither The number of c-Fos+ neurons in CA1 and the DG largely covaried in all groups (Pearson correlation coefficients of 0.94, 0.92 and 0.88, for sham, PV-Gi and PV-Gq, respectively), but the number of c-Fos+ cells recruited in CA1 per active DG cFos+ cell was increased by perisomatic disinhibition, consistent with enhanced activity propagation from DG to CA1 and in parallel with enhanced memory encoding. In contrast, increased perisomatic inhibition was sufficient to decrease this ratio and prevent memory formation.  Table 2. *#p≤0.05, **##p≤0.01, ***p≤0.001, ****####p≤0.0001.

CA1 output is strongly regulated by DG PV cells
We then used in vivo multi-site electrophysiological recordings to investigate intrahippocampal functional connectivity (Fig. 3A). In control conditions, stimulation of the perforant pathway with a single electrical pulse produced the well-known activity propagation from DG to CA1 in the trisynaptic circuit (ECDGCA3CA1), generating only small synaptic currents in the stratum radiatum ( Fig. 3C and D) that were alone insufficient to drive action potential firing in CA1 pyramidal neurons ( Fig. 3C and E) (32) .
However, during PV cell inhibition in the DG, the same stimulation triggered strong synaptic responses and robust firing ( Fig. 3B to E). Increasing the activity of PV-cells produced the opposite effect ( Fig. 3B to E). The relevant control here was to quantify separately the disynaptic propagation from CA3 to CA1 (ECCA3CA1), bypassing the DG, to check on the absence of differences within or between groups before and after CNO administration These results reveal that facilitated firing in CA1 is causally influenced by specific disinhibition of the DG relay. However, it was not only the consequence of increased GC firing, since the amplitude of the DG PS did not fully explain CA1 activity (Fig. 3G). We selected stimulation trials with comparable activation of GCs (same rage of PS amplitudes) across all experimental groups and quantified the corresponding CA1 propagation, and found that the effective activity transfer from DG to CA1 was largely increased in CNO injected PV-Gi animals (Fig. 3G). What might be the basis of this additional effect? In these experiments, it was observed that the firing delay of GCs was decreased (Supp. Fig. 5B and C) and its synchrony increased (Supp. Fig. 5D and E). Inclusion of synchrony as co-variable significantly explained the evoked CA1 EPSP (F(1,80)=4.13, p= 0.046), indicating a contribution to the propagation to CA1 and a role of the precise timing of GC firing. Data represent mean ± SEM. G, Activity transfer from the DG to CA1. DG PS amplitude is plotted against their trisynaptically evoked EPSPs in CA1 (solid lines represents non-linear regression fit, shadow represents 95% CI). All statistical values are detailed in Supp. Table 3. *p≤0.05, ****p≤0.0001.

DG PV-cells gate long-range communication channels in the brain
The change in CA1 firing after a cell-specific alteration of DG inhibition potentially has major implications because activity in hippocampal long-range circuits largely relies on CA1 output. What then might be the systems-level consequences of the bi-directional control of DG physiology by PV-cells? A series of brain-wide functional magnetic resonance imaging (fMRI) experiments in control, PV-Gi and PV-Gq animals was performed (Fig. 4).
In control animals, stimulation of the perforant pathway ( Fig. 4A and B and Supp. Fig. 6A and B) activated structures of the hippocampal formation including the DG, CA3, CA1 and subiculum, with little or no extra-hippocampal propagation (12,33) , and with no effect of CNO (1 mg/Kg i.p.) (Fig. 4C to G). However, inhibition of PV interneurons in PV-Gi animals given CNO allowed hippocampal activity to propagate to cortical areas in the medial temporal lobe (including the entorhinal and perirhinal cortices), the PFC (including the prelimbic, infralimbic, orbitofrontal and cingulate cortices) and the retrosplenial cortex ( Fig.   4C and F). Hippocampal activity also reached subcortically to the striatum, notably the NAc ( Fig. 4C and G). Conversely, increasing PV-inhibitory tone in PV-Gq mice resulted in fMRI activation maps with no extra-hippocampal propagation (Fig. 4C, E to G and Supp. Fig. 6).
To validate the fMRI findings, we performed simultaneous electrophysiological recordings in the medial PFC (Fig. 4H), one of the fMRI-identified extra-hippocampal structures activated during DG disinhibition. Electrophysiological potentials evoked by stimulation of the perforant pathway were only seen when the activity of the PV-cells was decreased (Fig. 4H). Therefore, downregulating the perisomatic inhibitory tone in the DG enhances the effective connectivity from the hippocampus to the PFC. To extend this finding to additional brain regions and, importantly, under anesthesia-free conditions, we measured the covariation of c-Fos expression between the hippocampus, PFC and NAc in animals performing in the NOL task. Activation in these brain regions has been shown to be necessary during memory encoding in spatial memory tasks (34)(35)(36)(37) and we demonstrated that encoding in the NOL task indeed activates the three brain regions in comparison to homecage exploration ( Fig. 2G and Supp. Fig. 7). The important analysis here was the covariation between c-Fos+ cell assemblies, demonstrating enhanced coupling between the hippocampus, PFC and NAc in the PV-Gi animals and decreased coupling in the PV-Gq animals (Fig. 4I). Interestingly, the number of C-Fos+ cells in the NOL task across conditions and regions was constant (Fig. 4J), highlighting the effect of the experimental manipulation on the coordination between the activated cell assemblies, rather than their activation per se.

Discussion
The key new finding is the identification of a systems-level gating mechanism in the DG and operated by PV-interneurons, permitting effective hippocampal coordination of cell assemblies in a distributed brain network. DG disinhibition increased functional connectivity between neuronal populations in the hippocampus, associative and prefrontal cortical areas and NAc and enhanced the encoding of novel object-context associations and context discrimination, while the opposite was found when PV-cell activity was increased. We support these findings in whole-brain fMRI data and targeted electrophysiological recordings in anesthetized animals, as well as cognitive evaluation and c-Fos network analysis in awake and freely moving animals. Because the brain receives a continuous bombarding of sensory information, we propose that the capacity of the DG to functionally couple or decouple a large network of brain regions relevant for learning (8)(9)(10) , might provide the flexible mechanisms required to select information and assimilate it into memory, or discard it, preserving the integrity of the memory base.

Mesoscopic and macroscopic consequences of DG disinhibition
Our results link the effect of the E-I balance at the cellular level in the DG circuit (GC-PV interactions) to its mesoscopic (intrahippocampal) and macroscopic (brain network) levels. A first characteristic is the regulation of GC neuronal firing without affecting dendritic currents and synaptic plasticity (Fig. 1C-G), as expected form the perisomatic innervation of GCs by PV+ interneurons (40,41) . In addition, GCs firing became more synchronous (less jitter) and less delayed with respect to the synaptic input (Supp. Fig. 5B-E) during PV-cell inhibition. This effect might have a large impact on intrahippocampal effective connectivity, since it has been shown that the relative timing of the CA3 and EC inputs onto the apical dendrite of the CA1 pyramidal neurons determines the pyramidal cell firing probability (42,43) . Since activity and timing in the di-synaptic pathway to CA1 (ECCA3CA1) in our experiments were shown to be unaltered ( Fig. 3B and 3F), as it was likely the case in the direct temporoammonic pathway (ECCA1) far from the DG manipulation, the change in relative timing in these two pathways relative to the tri-synaptic pathway It is known that PV+ interneurons are fundamental network elements generating gamma oscillations (45,46) and organizing brain rhythms (47) . Communication between brain regions is thought to occur when the oscillatory activity in connected populations is coherent or phase locked (48) and it has been shown that PV-cell inhibition is able to induce phase resetting (49) . Therefore, we hypothesize that the activation level of PV-cells in the DG sets the phase of ongoing oscillations facilitating or precluding subsequent information exchange in the network. Supporting this hypothesis, a recent electrophysiological analysis of hippocampal gamma and theta activity in CA1 and DG during novelty exploration and memory guided behaviour, shows increased theta synchronization between regions associated to theta-gamma interactions (50) . In this study, gamma activity, representing E-I circuit interaction, was associated to theta-phase shifts and synchronization between regions (50) . Therefore, in addition to the increased gain of the CA1 output, we speculate that longrange activity synchronisation contributes to the enhanced functional coupling during DG disinhibition. This might be a special role of PV circuits in the DG, since inhibition of PVcells in CA1 or the medial PFC disrupts the timing of hippocampal ripples and cortical spindles, respectively, decreasing their temporal coincidence and impairing memory consolidation (7) . Overall, we have shown that a relatively simple and constrained manipulation in the local DG circuit has a large impact on the macro-scale organization of functional connectivity, supporting the view of the DG as a critical node in the brain network for memory formation (12,13) .

Binding cell assemblies, updating the memory
Learning new context associations activates cell assemblies in the DG, CA3, CA1, PFC and NAc (Fig. 2G-H, Fig. 3I-J and Supp. Fig. 7). However, this activation was not sufficient for memory formation. Equally sized but functionally unbound (uncorrelated) assemblies, found during increased DG inhibition, were associated with memory failure. In contrast, enhanced functional binding between comparable cell assemblies, obtained in our experiments during DG disinhibition, was associated with improved memory outcomes. This result indicates that experience-activated cell assemblies need to be integrated into systemslevel circuits to encode the memory, and point to the DG as a critical node in this network function. A second function of this circuit mechanism can be found in the preservation of stored memories from continuous overwriting, in this case by decorrelating or decoupling brain assemblies. The threshold for memory updating would thus be set by the instantaneous E-I balance in the DG, a prediction that should be tested in future experiments. The dynamic control of PV-cell activity levels in the DG appears as a critical gate for memory updating.
Our results circumscribe this DG function to the earliest stage of memory formation, at the time of initial stimulus encoding, with no detectable contribution during memory consolidation or retrieval (Fig. 2). This result is compatible with a role of the PV cells of the DG in the coordination of the initial associations between dispersed cell assemblies necessary to build up the circuits supporting a stable memory engram, likely providing the scaffold for subsequent stabilization through the consolidation process (9,10,34,51). This could be the mechanism that leads to the tagging of PFC cell assemblies reported in contextual fear conditioning and that subsequently mature during the consolidation process (52,53) .
In the electrophysiological results we have shown that dendritic activity and plasticity in response to EC inputs is preserved during PV-cell activity manipulations (Fig. 1) and, therefore, the association of medial and lateral EC inputs in the dendrites of GCs in the behavioural experiments was likely not affected. This association, sometimes referred to as DG binding (54) , is fundamentally different from the PV-operated DG binding mechanism that we introduced here. In our case, the binding refers to the association of multiple cell assemblies across brain regions. Both binding mechanisms, however, may contribute to pattern separation in the DG due to multiple feature binding, since the more exact and comprehensive the memory representation of an experience is, the easier it would be to discriminate from previous stored experiences with overlapping features. The contribution of systems-level binding to pattern separation was unveiled by the experiments in which we showed that GCs activation sparsity, a fundamental DG property for pattern separation (2)(3)(4) , was not affected by PV-cell inhibition, but context discrimination was however enhanced.

Concluding remarks
We have presented imaging, electrophysiological and behavioural data that overall demonstrate a critical role of the DG in coordinating neuronal activity in a large network of brain structures known to be fundamental for memory formation. Using cell-specific manipulations we have shown that this systems-level gating mechanism can be operated by PV+ interneurons. We propose that through this simple mechanism, the DG fulfils two complementary functions, to select relevant information for updating memory, and to discard redundant information preserving the memory base from continuous overwriting. This mechanism merits further investigation in pathological conditions in which the proper updating of the memory base might be compromised. For instance, while a balanced increase in the inhibitory tone may prevent inefficient memory overwriting, an exacerbated or dysregulated DG inhibitory tone may render memories fixed, even after contingencies have changed, which might be potentially relevant for conditions such as Posttraumatic Stress Disorder (55) . More generally, our results unveiled a new mechanism to control activity propagation in a complete network by regulating activity at a single node. Finding influential nodes in the brain network can help us develop methods for retuning maladaptive network dynamics (13) .

Animals
All animal experiments were approved by the Animal Care and Use Committee of the Instituto de Neurociencias de Alicante (Alicante, Spain) and comply with the Spanish law (53/2013) and European regulations (EU directive 2010/63/EU).
In total, 143 mice, both males (n=82) and females (n=61), with two months of age at the beginning of the experiments, were randomly assigned to the different experimental groups (see below). Ten additional animals were used but excluded due to poor behavioural performance (less than 8 seconds of object exploration in the allocated 10 minutes of the novel object location task). No differences between sexes were found and data were pooled.
Coordinates for targeting the injections in the hilus of the DG, from Bregma, were -2mm AP, ± 1.4 mm LM, +2mm DV (56) . After opening the skin, we opened a 700µm Ø trepan with a

Novel Object Location task (NOL).
After a period of handling, mice were introduced into an empty arena (50x50x30 cm) with spatial cues, and softly illuminated (luxes: 23 in the center ± 2 in the corners), and were allowed to freely explore it for 2 periods of 5 minutes (habituation phase). Twenty-four hours later we introduced two identical objects in the arena (located in opposite corners, 13'5 cm away from the walls) and the animals were allowed to explore them (familiarization or encoding phase). Familiarization with the objects was terminated when an animal reaches 20 seconds of accumulative exploration of both objects, or after 10 minutes in the arena (30) .
Animals exploring the objects less than 8 s in the 10 min of the familiarization phase were removed from the study (a priori exclusion criteria). Twenty-four hours later, animals entered in the same arena with the same objects, but one of them was displaced to a new location (another corner; 13'5 cm away from the walls and 15 cm away from the other object) and we again let the animals to explore under the same criteria (test or retrieval phase) (Supp. Fig. 2A and B). In one experiment (Fig. 2F), the object displacement was reduced to 10 cm to increase the task difficulty. We set the time that the animals explore the object in the new location divided by the total time of exploration of both objects, as an index of the mice spatial memory. Using this behavioural protocol, and injecting CNO or its vehicle in different moments, we could modulate the activity of PV interneurons during the encoding phase (injecting CNO or saline 90' before the familiarization phase; n=38), during the consolidation (injecting CNO or saline 10 minutes after the familiarization phase; n=42) or during the retrieval (injecting CNO or saline 90' before the test phase; n=41) ( Fig. 2A). For encoding and control experiments, same animals were used. Different groups were used for consolidation and retrieval experiments.
In addition to the objects exploration, we measured and quantified locomotor parameters as movement velocity, distance travelled and side vs centre arena preference (Supp. Fig. 2B and C).

Elevated plus maze (EPM)
We tested potential changes in anxiety induced by the modulation of PV-cell activity in the DG, using the elevated plus maze (EPM). One week after the NOL task, animals were injected either with CNO or saline, and placed 90' later in a EPM with two open and two closed arms (Supp. Fig. 2D). Animals were video-recorded and its performance tracked with commercial software (Smart Video Tracking Software, Panlab, Barcelona, Spain). We In PV-Gi animals, in which CNO administration induced a large increase in the PS amplitude, the EPSP was measured more distal from the granule cell soma layer (in the outer third of the molecular layer), giving smaller estimations. This is a common procedure to avoid the volume conducted artefact of the PS that would have prevented the precise measure of the EPSP slope.

Functional MRI
For the fMRI experiments, mice were anaesthetized with 1.4 g/kg of urethane and implanted with custom made carbon fibre MRI-compatible electrodes as described previously (44) . Briefly, individual 7 µm diameter carbon fibres (Goodfellow Cambridge T2-weigthed anatomical images were collected using a rapid acquisition relaxation enhanced sequence (RARE): FOV, 25x25 mm; 12 slices; slice thickness, 1mm; matrix, 192x192; TEeff, integrated combiner and preamplifier, and no tune/no match, was employed in combination with the actively detuned transmit-only resonator (Bruker BioSpin MRI GmbH, Germany).
Functional MRI data were analysed offline using our own software developed in MatLab (The MathWorks Inc.), which included Statistical Parametric Mapping packages (SPM8, www.fil.ion.ucl.ac.uk/spm). After linear detrending, temporal (0.015-0.2 Hz) and spatial filtering (3x3 Gaussians kernel or 1.5 sigma) of voxel time series, a general linear model (GLM) was applied. Functional maps were generated from voxels that had a high significant component for the model (p < 0.01) and were clustered together in space. ROIs were extracted from mice brain atlas (56)  Quantitative analysis of c-Fos positive nuclei was performed offline on 12-bit grey scale images acquired in a Leica DM4000 fluorescence microscope at 10x/0.25 dry objective using a Neurolucida software (MBF Bioscience, Williston, VT USA). The ROIs were manually delineated following the Franklin and Paxinos mouse brain atlas (56) . Analysis were performed using Icy Software (57) in a semi-automatic manner. The threshold for detection of positive nuclei was set for each brain region, setting average nuclei size and a signal/noise ratio higher than 23%, according to Rayleigh criterion for resolution and discrimination between two points. Animals with misplaced viral infections were removed from all the analysis.

Statistical analysis
The statistical analysis was done using GraphPad Prism 7 software (GraphPad Software Inc., La Jolla, CA, USA) or SPSS v20 software (IBM, New York, USA). After an exploratory analysis for the presence of outlier values, we checked the skewness and kurtosis of the data before testing their statistical distribution. Parametric-test requirements, including normality (D'Agostino-Pearson test and Shapiro-Wilk test) and homoscedasticity (F of Levene test) were tested. All the data fulfilled parametric criteria, unless otherwise specified.
In most analysis we applied two-way ANOVA, with a group factor with 3 levels (Sham, PV-Gi and PV-Gq), and a time factor with 2 levels (before and after CNO injection). In case of c-Fos analysis, we applied one-way ANOVA with group as factor with the same 3 levels as above. We applied a Sidak post hoc analysis for multiple comparisons with adjusted alpha.