The GABA-Glutamate supramammillary–dorsal Dentate Gyrus pathway controls theta and gamma oscillations in the DG during paradoxical sleep

Several studies suggest that neurons from the lateral region of the SuM (SuML) innervating the dorsal dentate gyrus (DG) display a dual GABAergic and glutamatergic transmission and are specifically activated during paradoxical (REM) sleep (PS). The objective of the present study is to fully characterize the anatomical, neurochemical and electrophysiological properties of the SuML-DG projection neurons and to determine how they control DG oscillations and neuronal activation during PS and other vigilance states. For this purpose, we combine structural connectivity techniques using neurotropic viral vectors (rabies virus, AAV), neurochemical anatomy (immunohistochemistry, in situ hybridization) and imaging (light, electron and confocal microscopy) with in vitro (patch clamp) and in vivo (LFP, EEG) optogenetic and electrophysiological recordings performed in transgenic VGLUT2-cre mice. At the cellular level, we show that the SuML-DG neurons co-releases GABA and glutamate on dentate granule cells and increase the activity of a subset of DG granule cells. At the network level, we show that the activation of the SuML-DG pathway increases theta power and frequency during PS as well as gamma power during PS and waking in the DG. At the behavioral level, we show that the activation of this pathway does not change animal behavior during PS, induces awakening during slow wave sleep and increases motor and exploratory activity during waking. These results suggest that the SuML-DG pathway is capable of supporting the increase of theta and gamma power in the DG observed during PS and plays an important modulatory role of DG network activity during this state.


Introduction
The supramammillary nucleus (SuM) is a thin structure overlying the mammillary bodies in the hypothalamus that provides substantial projections to many regions of the limbic system including the hippocampal formation (for review see McNaughton 2004, Vertes 2015). The SuM is composed of several populations of neurons that differ in their neurochemical content and the specificity of their connections (Swanson et al., 1982;Gonzalo-Ruiz et al., 1992;Leranth and Kiss, 1996;Kocsis et al., 2003;Borhegyi and Leranth, 1997;Haglund et al., 1984). In rats, many studies have described the projections from SuM neurons to the hippocampus (Segal & Landis, 1974;Pasquier & Reinoso-Suarez, 1976, 1978Wyss et al., 1979;Haglund et al., 1984;Saper et al., 1985;Vertes, 1992;Magloczky et al., 1994;Nitsch and Leranth 1996;Vertes & McKenna, 2000). We (Soussi et al., 2010) and others (Boulland et al., 2009) demonstrated in rats that SuM neurons from the most medial part of the SuM (SuMM; Paxinos and Watson, 1998), referred as SuMm by Swanson (1998), or SuMp by Pan and McNaughton (2004) innervating the inner molecular layer of the ventral dentate gyrus (DG) and CA2 pyramidal cells are glutamatergic (Soussi et al., 2010). In contrast, SuM neurons located in the lateral two-third region of the SuM (SuML), referred as SuMg by Pan and McNaughton (2004), innervate the supragranular layer of the dorsal DG (dDG) and to a lesser extend the ventral DG (vDG) and display a unique dual glutamatergic and GABAergic neurotransmitter phenotype (Bouland et al., 2009;Soussi et al., 2010). Indeed, these SuML neurons and their projections to the dDG co-express markers for both glutamatergic (vesicular glutamate transporter 2; VGLUT2) and GABAergic (glutamate decarboxylase 65, GAD65 and vesicular GABA transporter, VGAT) neurotransmission and establish asymmetric and symmetric synapses on DG cells (Soussi et al., 2010). The not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint connectivity and neurochemical properties of these different SuM-hippocampal pathways suggest that they may contribute differently to hippocampal dependent functions. Indeed, it has been shown that the SuM controls hippocampal theta rhythm (Kocsis and Vertes 1994, Vertes and Kocsis 1997, Kocsis and Kaminski 2006 and is involved in emotional learning and memory (Pasquier and Reinoso-Suarez 1976, Richmond et al 1999, Pan and McNaughton 2002, Santin et al 2003, Pan and McNaughton 2004, Shahidi et al. al., 2004. By combining retrograde tracing, neurotoxic lesion and FOS immunostaining, it was recently shown that SuML GABA/glutamate neurons are responsible for the activation of dDG granule cells during paradoxical sleep (PS) (Renouard et al., 2015 ;Billwiller et al., 2016). Such activation may play a key role in the previously reported beneficial effect of PS on learning and memory (Maquet et al., 2000). In addition, Pedersen and colleagues (2017), by using a chemogenetic approach in transgenic mice, showed that SuM neurons containing only VGLUT2 but not those containing VGAT and VGLUT2 play a crucial role in waking. Altogether these results suggest that glutamatergic neurons from the SuM known to innervate several limbic regions including the CA2/CA3a hippocampal region and the vDG are involved in normal wakefulness whereas the GABA/GLU SuML projecting to the dDG could be instrumental for PS function.
In this study we investigate how the SuML-dDG pathway control dDG neurons activity during PS and other vigilance states. For this purpose we combine innovative structural connectivity techniques using neurotropic viral vectors (rabies virus, AAV), neurochemical anatomy (immunohistochemistry, in situ hybridization) and imaging (light, electron and confocal microscopy) with in vitro (patch clamp) and in vivo (LFP, EEG) optogenetic and electrophysiological recordings performed in transgenic VGLUT2-cre mice in order: 1) to fully characterize the anatomical, neurochemical not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101 and electrophysiological properties of the SuML-dDG projection neurons in our mice trangenic model and 2) to determine the influences of this specific pathway on behaviors and associated oscillatory activities characterizing the different vigilance states as well as DG neuron activity. not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Animals
For this study, we used 30 transgenic "VGLUT2-cre" adult mice (25-30 g; aged 10-11 weeks) obtained by mating male and female homozygote Vglut2-ires-Cre mice from Jackson Laboratories (Strain name Slc17a6 tm2(cre)Lowl /J, stock #016963). All mice were bred in-house and maintained in standard cages, with food and water ad libitum, in a temperature-and humidity-controlled room under a 12 hr light/12 hr dark cycle. All the surgical and experimental procedures were performed according to the National Institutes of Health guidelines and the European communities Council Directive of 86/609/EEC and were approved by the University of Aix-Marseille and Lyon University Chancellor's Animal Research Committees.

Rabies virus
Four VGLUT2-cre mice underwent stereotaxic injection of rabies virus (RV) within the inner molecular layer of the dDG in order to obtain golgi-like retrograde-labeling of SuM neurons projecting to this region. The strain of RV used was the Challenge Virus Standard (CVS, 4.10 7 plaque-forming units/mL) (Bras et al. 2008;Ugolini 2010;Coulon et al. 2011). Only vaccinated personnels conducted these experiments at the appropriate biosafety containment level until the sacrifice of the animals as described below.
These animals were used to determine the neurotransmiter phenotype of SuM neurons projecting to dDG at the mRNA level, combining immunohistofluorescent detection of RV with fluorescent in situ hybridization detection of VGAT and VGLUT2 mRNAs. not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Adeno-Associated Virus (AAV) Double floxed Inverted ORF (DIO) vectors
Sixteen VGLUT2-cre mice underwent stereotaxic injection of the cre-dependent viral vectors: AAV5-EF1a-DIO-EYFP (3.5 10 12 virus molecules / ml; UNC Gene Therapy Center Vector Core; Dr Deisseroth) into the SuML. These VGLUT2-EYFP mice expressing the Yellow Fluorescent Protein (YFP) in VGLUT2 SuM neurons and their axon terminals were used 1) to determine the neurotransmitter phenotype of SuM neurons innervating the dDG at the protein level, by simultaneous immunohistofluorescent detection of EYFP labeled axon fibers and terminals, VGAT and VGLUT2 in dDG (n=3); 2) to determine the synaptic profile of these EYFP labeled axon terminals at the electron microscopy level (n=3); 3) as control animals (n=3) for in vitro optogenetic stimulation and patch clamp electrophysiological recordings; 4) as control animals (n=7) for in vivo optogenetic stimulation and electrophysiological recordings.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101 NpHR3 mice expressing the inhibitory opsin Halorhodopsin (NpHR3) were used for in vivo optogenetic inhibition and electrophysiological recordings.

Surgery
Mice were anesthetized by an intraperitoneal injection (i.p.) of Ketamin (50mg/kg) / Xylazine (5mg/kg) solution. If necessary, this anesthesia was repeated during the surgery. Animals were then secured in a stereotaxic frame (David Kopf instruments).
The body temperature of mice was monitored and maintained at about 37°C during the entire procedure by means of an anal probe and heating blanket respectively.
The head was shaved and sanitized with Betadine and 0.9% NaCl. Local anesthesia was performed by infiltration of the scalp with xylocaine (lidocaine hydrochloride 0,5%), and an ophthalmic gel was placed on the eyes to avoid drying. After scalp incision, holes were drilled in the skull, with antero-posterior (AP), medio-lateral (ML) and dorso-ventral (DV) coordinates based on Paxinos & Franklin's atlas (2005).

RV and AAV viral vector injections
The RV (200 nl) was pressure-injected unilaterally (n=2) or bilaterally (n=2) within the dDG inner molecular layer of VGLUT2-cre mice according to the following coordinates: AP = -2; ML = +/-1.5; DV = -1.7. Injections were performed by using a 33-gauge Hamilton syringe connected to a Micro4 injection pump system (World Precision Instruments). After completion of the injection procedures, the syringe was removed and the skin was sutured. Animals were treated with local anesthetic, returned to their cages kept at the appropriate biosafety containment level for a survival period of 38h to observe an optimal RV retrograde labeling for the dendritic arbor of SuM neurons. AAV5-EF1a-DIO viral vectors (500nl) were injected bilaterally not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Optrode and electrode implantation
After AAV injections, 15 VGLUT2-mice were implanted with an optrode in the left dDG (AP = -2; ML = -1.1; DV = -1.7) for optic stimulation of axon terminals originating from transfected SuML neurons and local field potential (LFP) recordings from the DG. The optrode consisted of an optic fiber (250 µm ∅, 0.39 NA; Thorlabs SAS) and the LFP electrode. The LFP electrodes consisted of two 45 µm diameter tungsten wires (California Fire Wire Company), twisted and glued together to form a rigid and solid structure, with the 2 ends of the wires separated by 100 µm from each other.
Two small screws (1 mm in diameter, Plastics One) each soldered to a wire were fixed on the skull. The first screw was fixed in the parietal part of the skull for EEG recording; the second screw, at the level of cerebellum as reference electrode. Two wire electrodes were inserted into the neck muscles for bipolar EMG recordings. All leads were connected to a miniature plug (Plastics One) that was cemented on the skull.
After completion of the surgery the skin was sutured. Mice were treated with local anesthetic, and an intramuscular injection of antibiotic (Baytril, 5mg/kg) to prevent any risk of infection. Mice were monitored until waking and replaced in their home cages for a survival period of 3 weeks. not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Tissue preparation for light microscopy
Animals were deeply anesthetized with ketamine and xylazine and transcardially perfused with 4% paraformaldehyde (PFA) in 0.12 M sodium phosphate buffer, pH 7.4 (PB). After perfusion, the brains were removed from the skull, post-fixed in the same fixative for 1 h at room temperature (RT), rinsed in PB, cryoprotected in 20% sucrose overnight, frozen on dry ice and sectioned coronally at 40 µm with a cryostat (Microm). The sections were rinsed in PB, collected sequentially in tubes containing an ethylene glycol-based cryoprotective solution and stored at -20°C until histological processing. One of every ten sections was stained with cresyl violet to determine the general histological characteristics of the tissue throughout the rostro-caudal extent of the brain. Selected sections were processed for 1) simultaneous detection of VGLUT2 mRNA, VGAT mRNA and RV; 2) simultaneous immunohistofluorescent detection of EYFP, VGLUT2 and GAD65 or VGAT and / or 3) immunohistochemical detection of cFos. Mm-S1c17a6 for detection of VGLUT2 mRNA and Mm-S1c32a1-C3 for detection of VGAT mRNA. After hybridization, sections were then processed for visualization using the RNA-scope Multiplex Fluorescent reagent Kit v2 (Advanced Cell Diagnostics) and the Tyramide Signal Amplification (TSA™) Plus Cyanine 3 and TSA Plus Cyanine 5 systems (Perkin Elmer).
After the RNAscope assay, section were rinsed in 0.02 M potassium phosphate- After several rinses in KPBS, all sections were coverslipped with Fluoromount (Electron Microscopy Sciences). The specimens were analyzed with a Zeiss laserscanning confocal microscope.

Quantification of co-localizing RV, VGLUT2 and VGAT mRNAs
Quantitative analysis was conducted to evaluate the extent of SuM neurons with direct projections to the dDG that co-express VGLUT2 and VGAT mRNAs. For this purpose, the number of triple neurons was determined for each animal (n=4), from 3 sections (120 µm apart from each other) across the antero-posterior extent of the not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint SuM. For each section, an image of the entire SuM region was obtained from a single confocal slice using the Tile Scan function with a 20x objective and sequential acquisition of the different wavelength channels to avoid fluorescent talk with ZEN software (Zeiss). The analysis was then performed with Neurolucida software (version 7, mbfBioscience) as follows: for each confocal image, all RV-labeled neurons were identified on the green channels and examined for colocalization of VGLUT2 mRNA in the blue channel and/or VGAT mRNA in the red one. Tripledouble-and single labeled neurons were tagged differently and counted by the software. A total of 380 RV-labeled neurons were analyzed.

Simultaneous immunohistofluorescent detection of EYFP, VGLUT2 and GAD65 or VGAT performed in VGLUT2-EYFP mice
Selected sections at the level of the dDG were processed using the MOM kit as were then mounted on superfrost-coated slides, dried overnight at RT and not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Quantification of co-localizing GFP, VGLUT2 and VGAT
Two quantification protocols were used in order to evaluate the extent of the different neurochemical phenotypes of axon terminals from the SuM innervating the dDG. In the first protocol previously described (Soussi et al., 2015), the densities of VGLUT2/VGAT and VGLUT2 only labeled terminals were assessed by the quantification of immunolabeling for VGLUT2/VGAT and VGLUT2 only, respectively.
These analyses were performed on 4 sections for each mouse. Single optical confocal images were acquired with Zeiss LSM 510 laser-scanning microscope and analyzed with the software provided by the microscope manufacturer (LSM 510 Zen, Zeiss). All images were acquired from the suprapyramidal and infrapyramidal blades of the dDG, using identical parameters. The percentages of VGLUT2 labeled terminals containing VGAT were estimated by the the Manders' coefficient (proportion of pixels for VGLUT2 also positive for VGAT) obtained with the JACoP co-localization Plugin for Image J, in the region of interest (ROI) which included granule cell layer (GCL) and the narrow zone superficial to the granule cells defined as the supragranular layer (SGL) following recommendations from Bolte & Cordelières (2006). For each channel, an identical bottom threshold was applied throughout the analyses, and only the pixels with a value above this threshold were counted. When a pixel had a value above the threshold in both channels, it was counted as double positive. The size and the shape of the ROI was the same for each confocal image. The average % of co-localization was calculated for each blade of the DG for each mouse. not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint In the second quantification protocol, we determined for each mouse (n=4), the relative percentages of triple-and double-labeled boutons for the GFP anterograde tracer and VGAT and ⁄ or VGLUT2 in the SGL and infragranular blade of the dDG from several z-stacks of 10 confocal slices, acquired with the 100X objective and a numerical zoom 8, in three different sections. The analysis was performed as previously described (Persson et al., 2006;Soussi et al., 2010) and following recommendations from Bolte & Cordelières (2006). For each z-stack, the confocal images obtained from separate wavelength channels (green, red and blue) were displayed side by side on the computer screen together with the images corresponding to colocalized pixels within each optical slice of the z-stack obtained with the colocalization highlighter plugin in ImageJ. The GFP-labeled boutons were identified in the green channel within a probe volume defined by the size of the confocal slice (19.38 µm · 19.38 µm) and the height of the z-stack (2 µm). Each bouton was examined for colocalization through the individual optical slices of the zstack. Single-, double-and triple-labeled boutons were counted using the Cell Counter plugin in Image J. The total number of GFP-labeled terminals analyzed in the two regions of interest was 400.

VGLUT2-ChR2 (n=4) and VGLUT2-NpHR3 (n=4) mice
Selected sections at the level of the DG were processed for immunohistochemistry according to previously described protocol (Esclapez et al. 1994). Sections were pretreated for 30 min in 1 % H2O2, rinsed in PB and KPBS, preincubated for 1 h in 3 % normal goat serum (NGS, Vector Laboratories) diluted in KPBS containing 0.3 % Triton X-100 and incubated overnight at RT in cFos rabbit polyclonal antiserum (1:20,000 ; Calbiochem) diluted in KPBS containing1 % NGS and 0.3 % Triton X-100. not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019.

Quantification of cFos immunolabeled neurons
The number of cFos labeled neurons was calculated in the DG GCL of the right and left (ipsilateral to the optic stimulation) hemispheres in VGLUT2-EYFP control (n=7), VGLUT2-ChR2 (n=4) and VGLUT2-NpHR3 (n=4) mice. These analyses were performed using a computer-assisted system connected to a Nikon 90i microscope and the Neurolucida software (MicroBrightField). A total of 4 sections (400 µm apart from each other) surrounding the optrode site were analyzed for each animal. In each section the GCL was delineated and all neurons labeled for cFos were plotted. The software calculated the total number of labeled neurons in each hemisphere for each animal. The average total number of labeled neurons / hemisphere ± SEM was calculated for each group of control VGLUT2-EYFP, VGLUT2-ChR2, and VGLUT2-NpHR3 mice. Statistical analysis was performed by Statview software using Wilcoxon Rank Sum Test.
not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Tissue preparation for electron microscopy
Three VGLUT2-EYFP mice were perfused intracardially with a fixative solution containing 4% PFA and 0.1% glutaraldehyde in 0.12 m PB. After perfusion, the brain was removed from the skull, post-fixed in the same fixative overnight at 4°C and rinsed in PB for 1.5 h. Blocks of the forebrain were sectioned coronally at 60 µm with a vibratome. Pre-embedding immunolabeling for GFP sections at the level of dDG were pre-treated for 15 min in 1% sodium borohydride prepared in PB and in Durcupan resin and polymerized at 56°C for 24 h (Zhang & Houser, 1999). Labeled regions of the DG that contained the molecular and granule cell layers were trimmed from the sections, re-embedded on capsules filled with polymerized Durcupan and further polymerized at 56°C for an additional 24 h.
Ultrathin sections from the most superficial face of the blocks were cut on an ultramicrotome. Serial sections were picked up on nickel mesh grids and stained with not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint uranyl acetate and lead citrate. Sections were examined and photographed with a JEOL electron microscope.

Whole-cell voltage-clamp recordings
Slices were submerged in a low-volume recording chamber and continuously superfused with 32-34°C ACSF at 5ml/min perfusion rate. For each mouse, four slices containing the dDG were selected for patch clamp recordings. DG neurons were visualized by infrared video microscopy using an upright microscope (SliceScope, Scentifica Ltd). Patch pipettes were pulled from borosilicate glass tubing (1.5 mm outer diameter, 0.5 mm wall thickness) and filled with an intracellular not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.  Figure 4K). At this holding potential the recorded currents not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint were essentially generated by glutamate receptors. Then the glutamate component was abolished by NBQX and D-AP5, and the GABA component was revealed at +10mV holding potential. This PSC was completely inhibited by bicuculline application that confirmed the GABA receptor origin of these remaining currents (see also Figure 4J).

Double immunohistofluorescent labeling for Biocytin and GFP
After recordings slices were processed for simultaneous detection of the biocytin- In vivo electrophysiology: optic stimulation, LFP and EEG recordings not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint All mice were placed for 7 days in a recording box in order to get them used to the recording conditions. The recording box was ventilated, as well as electrically and sound isolated. The temperature was regulated at 21°C, and a 12 hr light/12 hr dark cycle imposed. Mice were accustomed to the cable connecting them to the recording device. The recording cable connected the micro-connector implanted on the head of the animal to a collector, which ensured the continuity of the recorded signals without hindering the movements of the mouse. At the end of this habituation, the control recordings begun. EEG and EMG recordings were digitized at 1kHz, amplified 5000 times with a 16 channels amplifier (A-M System) and collected on a computer via a CED interface using Spike 2 software (Cambridge Electronic Design). The signal was band-pass filtered online between 1 and 300 Hz for EEG, and between 10 and 100 Hz for EMG. The 50 Hz signal was removed with a notch filter. The EEG and LFP signals were acquired by monopolar derivation (differential between the recording electrode and the reference electrode located above the cerebellum). The EMG bipolar signals were calculated by measuring the differential between the two EMG electrodes. Mice were recorded for 24 h of baseline followed by optogenetic manipulation.

In vivo optogenetic stimulation
Optical stimulations were delivered via a patch cable connected to either a 100 mW 473 or 532 nm laser diode (Laserglow). Stimulations were performed during 4 days: in the first 3 days, mice were stimulated during one specific vigilance state par day: waking (WK), slow wave sleep (SWS) or PS. Each day, stimulations were delivered during the same circadian period (10 AM-2 PM). Stimulations were applied 10 s after the occurrence of a stable WK, SWS or PS event as detected by real time not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019.

Analysis of the sleep wake states
Polysomnographic recordings were visually scored by 5 s epochs for WK, SWS and PS as previously described (Sapin et al. 2009). Hypnograms were obtained by using a custom Matlab script. For each animal, the number of awakenings during SWS and PS optogenetic stimulations was counted and expressed as percentage of the total number of stimulations.

LFP and EEG analysis
LFP and EEG signals were analyzed using a custom Matlab script using the Chronux toolbox. The time-frequency spectrograms were computed with the same toolbox and expressed in arbitrary units. The mean power spectral density in the 10 s before the stimulation was compared to that in the 10 s during the stimulation (sliding window: 1s), in order to obtain a mean spectral power ratio (PR) ± SEM. The frequency not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
To analyze the evolution of LFP and EEG theta and gamma bands during optogenetic stimulation, the mean PR of these spectral bands and the respective 95% confidence intervals were calculated for VGLUT2-EYFP and VGLUT2-ChR2 animals from 10 seconds before to 10 seconds after the photostimulation. In order to compute these intervals, we used a bootstrap procedure, which allows creating artificial groups from the original data, with replacement. The mean of each artificial group derived from the original data was then computed. This operation was repeated 10000 times and the 95% confidence interval was the 5 th and the 95 th percentile of the means of the randomly constructed samples.
Finally, during WK and PS the peak of theta frequency (6-12 Hz) in the 10 s before and during the optogenetic stimulation was identified in VGLUT2-EYFP and VGLUT2-ChR2 animals.

EMG analysis
In VGLUT2-EYFP and VGLUT2-ChR2 mice EMG signals during WK were analyzed using a custom Matlab script. The mean EMG value in the 10 s before the stimulation was compared to that in the 10 s during the stimulation, in order to obtain a mean EMG ratio ± SEM. In the 2 groups of animals we performed a sequential analysis per 0.5s on the mean of the absolute EMG values from 20 seconds before to 10 seconds after the photostimulation. The respective bootstrap 95% interval (computed as described above) was calculated for each sequential value. not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

VGLUT2-Cre transgenic mice
Three populations of VGAT and/or VGLUT2 mRNA-containing neurons were observed at all antero-posterior levels of the SuM. These included a population of large neurons co-expressing VGAT and VGLUT2 mRNA (Fig.1 A,D, arrows); and two populations of smaller neurons containing either VGAT mRNA (Fig.1 A,D, arrowheads) or VGLUT2 mRNA ( Fig.1 A,D, blue). The distribution as well as the density of these populations differed significantly within the SuM. Almost all VGAT/VGLUT2 mRNAs -containing neurons were clustered around and above the mammillary tract (mt) (Figure 1 A, D) in a region similar to that described in rats as the grandicellular SuM (SuMg) by Pan and Mc Naughton (2004) and that is included in the lateral SuM (SuML) region of Paxinos and Franklin's mouse Atlas. The numerous singly labeled VGLUT2 mRNA -expressing cells were small and mostly located in the most central area (Fig. 1A) termed the parvicellular SuM (SuMp) by Pan & Mc Naughton (2004) corresponding to the SuMm in Paxinos and Franklin's mouse Atlas. Some were also intermingled with the larger VGAT/VGLUT2 mRNAscontaining cells in the SuMg. Few neurons expressed VGAT mRNA only. These neurons were scattered within the SuML and SuMM.
In summary, our results indicate that in mice like in rats a large number of neurons co-expressing markers of GABAergic and glutamatergic transmission are located in the SuML region.

Distribution of SuM neurons with dorsal dentate gyrus (dDG) projections
Unilateral or bilateral injections of rabies-virus (RV) were performed into the DG of not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint the dorsal hippocampal formation (Fig.1B, insert) to evaluate the distribution and proportion of SuM neurons that project to this structure. In all these mice, the RV tracer injection was located either in the granule cell layer or inner molecular region of the dorsal blade of the dentate gyrus. Within the SuM, all these animals displayed retrogradely labeled neurons exclusively in the region of SuML (SuMg) immediately dorsal to the mt (Fig. 1B, E, H). Virtually none were found in the SuMM (Fig. 1B).
Many of these neurons displayed a large soma with several labeled proximal dendrites (Fig. 1E). Therefore in mice, SuM neurons innervating the dDG are located in the SuML as in rats (Soussi et al., 2010).

Neurotransmitter phenotype of SuM neurons projecting to the dDG revealed by simultaneous detection of rabies-virus retrograde tracer, VGAT mRNA, and VGLUT2 mRNA
We next investigated whether the SuML neurons projecting to the dDG express markers of GABAergic and glutamatergic transmissions. Sections processed for simultaneous detection of VGAT mRNA, VGLUT2 mRNA and RV ( Fig. 1 A-I) showed that almost all RV retrogradely-labeled neurons co-express VGAT and VGLUT2 mRNAs ( Fig. 1 A-I, arrows). These data were confirmed by quantitative analysis performed on twelve sections (4 mice; 3 sections per mouse) showing that 99% (n= 378 out of 380 neurons) of the RV-labeled neurons co-expressed VGAT and VGLUT2 mRNAs. The percentages of these triple-labeled cells were the same (99%) for the three antero-posterior levels of the SuM analyzed (range 98%-100%).

Distribution and neurotransmitter phenotype of fibers and axon terminals originating from SuM neurons innervating the dorsal dentate gyrus
To further characterize the neurochemical properties of SuML neurons innervating not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint the dDG, AAV-5-DIO-EYFP was injected bilaterally in the SuML (Fig. 2 A) allowing specific EYFP labeling of VGLUT2 neurons and of their axon fibers innervating the hippocampus including the dDG (Fig.2 A). Sections of these VGLUT2-EYFP mice were processed for simultaneous immunohistofluorescent detection of EYFP, VGLUT2 and VGAT or GAD65. All mice injected in the SuML displayed numerous anterograde -EYFP labeled fibers and axon terminals in the dDG (Fig.2 B arrowheads) on both sides of the hippocampal formation. Labeled fibers and axon terminals were also present in the CA2 ⁄CA3a region of the hippocampus in particular when the injection sites involved more posterior levels of the SuML (Fig. 2B arrowheads).
In the dDG, EYFP axonal fibers and terminals were mainly located in the supragranular layer of the dorsal and ventral blades of the DG (Fig. 2 B showed that 90% (range 82% to 99%) of VGLUT2 labeled terminals were labeled for VGAT with no major differences between the infragranular blade (89%; range 82% to 99%) and the supragranular blade (91%; range 83% to 99%). Further quantification of EYFP labeling revealed that 98 % (range 96% and 100%) of the axon terminals contained both VGAT and VGLUT2 confirming that all EYFP-containing axon terminals originating from SuML neurons contained both makers of GABA and glutamate neurotransmissions. not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Electron microscopy analysis of synaptic contacts established by EYFPcontaining axon terminals originating from SuML neurons
In the supragranular region of the dDG, axon terminals from SuML neurons labeled for EYFP displayed diffuse electron-dense labeling that contrasted with adjacent unlabeled cellular compartments. These labeled axon terminals formed synaptic contacts on the soma (Fig.3 A,B,D,E,F,I) and dendritic profiles (Fig. 3 C,G,H) of presumed granule cells (GCs). These axon terminals were often very large boutons that displayed one (Fig. 3 A,B,D,E) or more (Fig 3 F) synaptic zones with relatively thin post-synaptic densities (arrowheads) characteristic of symmetric synapses.
Other synaptic contacts formed by these labeled terminals displayed thicker postsynaptic densities characteristic of asymmetric synapses (Fig. 3 A,B,C; arrow). We also observed large labeled "en passant" boutons establishing symmetric synapses on the dendrites of a presumed granule cell (Fig.3 G,H). Finally, some axon terminals from SuML neurons formed both symmetric (arrowhead) and asymmetric (arrow) synapses on the soma or the proximal dendrite of two different neighboring GCs (Fig.3, A,B).
Together these data demonstrate that in mice all SuM neurons innervating the dDG belong to a single population of large neurons located in the SuMg region of the SuML. These cells display a dual neurochemical phenotype for GABA and glutamate neurotransmissions and establish symmetric (presumably inhibitory) and asymmetric (presumably excitatory) synapses on the GCs of the dDG.

Co-release of glutamate and GABA at the SuML-dentate granule cells synapses
The neurochemical profile of SuML neurons projecting to the dDG strongly suggested that they co-release GABA and glutamate at the SuML-dentate granule cells synapses. To test this hypothesis, we performed patch clamp recordings of dDG not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint granule cells layer using optogenetic stimulation in hippocampal slices obtained from VGLUT2-ChR2 mice (5 mice, 3 sections per mouse, 16 neurons) and VGLUT2-EYFP control mice (3 mice, 3 sections per mouse, 10 neurons) (Fig. 4). The light stimulation of SuML-DG fibers and axonal terminals expressing ChR2 (Fig. 4 A) evoked a fast inward and a slower outward synaptic current in fourteen out of sixteen neurons recorded at a holding potential of -30 mV (Fig. 4D, 4F, 4H). Post-hoc immunodetection of biocytin-filled neurons showed that all these recorded neurons correspond to GCs (Fig. 4C, 4E, 4G, 4I). They were equally distributed in the apex and the upper and lower blades of dDG granule cell layer (Fig. 4B). These GCs were additionally recorded at different holding potentials. At -10 mV and +10 mV (close to the reversal potential of the glutamatergic receptor-mediated currents), only the outward (positive going) synaptic current was observed (Fig. 4D, see also 4F, 4H).
From -70 mV to -50mV (close to the reversal potential of GABA-A receptor-mediated currents) only the inward (negative going) synaptic current was observed (Fig. 4D, see also 4F, 4H). At intermediate holdings, from -30 to -20mV, the light-evoked synaptic currents displayed both inward and outward components (Fig. 4D, F, H).
Using a pharmacological approach, we comfirmed the nature of these currents. Bath application of a mixture of AMPA/Kainate and NMDA receptor antagonists (NBQX 10µM and D-AP5 40µM) abolished the inward component (Fig. 4F, K, L, red trace red) while GABA-A receptor blockers inhibited the outward component (Fig. 4H, Jgabazine ; K, L -bicuculline, green trace). Quantitative analysis performed in 5 DG neurons (Fig. 4K) further illustrated that in regular ACSF the repetitive light pulses (5 ms, 0.05 Hz) evoked PCS of relatively stable amplitude (1.00 ± 0.26 normalized, Fig.   4K, L, left hand side red trace is an example of an averaged response of one neuron). The application of glutamate receptor blockers (10µM NBQX + 40µM D-not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint AP5) reduced the peak amplitude by 84% (0.16 ± 0.15, p<0.01, n=5, paired Wilcoxon test). The remaining response seen in 4 from 5 neurons was probably due to GABA A mediated current because at -70mV the driving force for chloride driven currents is close to but not zero (Fig. 4K, L, red trance in the middle). Therefore in case of GABA massive release some inward current is still possible. Indeed a switch to Vh=10mV revealed a large PSC response to light stimulation (Fig. 4 L, green trace in the middle) which had a stable amplitude (1.00 ± 0.20 normalized), and was subsquently reduced to 12% (0.12 ± 0.16, p<0.01, n=5, paired Wilcoxon test) by the addition of 10µM bicuculline to the ACSF already containing GluR blockers (Fig. 4L, green trace at left). In 3 out of 5 neurons, bicuculline completely abolished the synaptic response to light pulses after 6 minutes of drug perfusion. In two neurons the small remaining current was probably due to the fact that bicuculline is a competitive antagonist and can be displaced when large quantities of GABA are released. All these results indicate that the inward synaptic current component is mediated by glutamate and the outward component by GABA. Importantly, the disappearance of the inward (glutamatergic) component induced by the light stimulation in the presence of NBQX and D-AP5 antagonists (red trace) did not affect the outgoing (GABAergic) currents ( Fig. 4F). In two GCs (Fig. 4I), the disappearance of the glutamate response (incoming current recorded at -30 mV) in the presence of NBQX and AP5 (red trace) unmasked the GABAergic component (outgoing current) abolished after addition of gabazine (green trace) (Fig. 4J).
The very short latency of the glutamatergic inward and GABAergic outward currents recorded in a GC after light stimulation of ChR2 containing axon terminals, and the persistence of GABAergic outflow current in the presence of NBQX and AP5 indicates that the GABAergic current results from direct light stimulated GABA not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint release and not from the excitation of DG GABAergic neurons by light stimulated glutamate release. Together, these results suggest that activation SuML axon terminals produce monosynaptic glutamatergic and GABAergic currents on their targets. Finally, we verified that light stimulation (5 ms, 10%, 50% 90% max power LED) of fibers and axon terminals expressing EYFP on slices of control VGLUT2-EYFP mice (n = 3) did not evoke any response in the recorded GCs (n = 10) (Fig. 4M, N).

Effect of optogenetic stimulation of SuML axon terminals innervating the dDG on behavior, LFP and EEG spectral content
We analysed the effects of light stimulation of SuML axon terminals innervating the dDG on behavioral states, spectral content of the LFP recorded in dDG and cortical EEG in VGLUT2-ChR2 (n=4) and control VGLUT2-EYFP mice (n=4).
The increase in EMG was due to an increase in animal movements (Fig. 5A, D). It was associated with a slight but not significant increase in theta power in the dDG LFP (Fig. 5B, 6A) and the EEG (Fig. 7A) compared to control VGLUT2-EYFP mice.
In contrast to SWS, light stimulation of SuML axon terminals in the dDG of VGLUT2-ChR2 mice during PS did not induce significantly more awakening of VGLUT2-ChR2 mice compared to control VGLUT2-EYFP animals (p=0.3865; Fig.5 L).
In addition, gamma was significantly increased both in the dDG LFP and the EEG during the stimulation in VGLUT2-ChR2 mice compared to control VGLUT2-EYFP mice (LFP: VGLUT2-ChR2= 1.1±0.02, VGLUT2-EYFP= 1.0±0.01, p=0.0209; EEG: not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Effect of light inactivation of SuML axon terminals innervating the dDG on behavior, LFP and EEG spectral content
The effects of light stimulation of SuML axon terminals innervating the dDG on the spectral features of LFP recorded in dDG, cortical EEG and associated behaviors during WK, SWS and SP were assessed in VGLUT2-NpHR3 mice (n=4) as illustrated in Fig. 8 and in control VGLUT2-EYFP mice (n=3). No statistical difference of the spectral content of the LFP and EEG was observed during light inactivation of SuML axon terminals in the dDG of VGLUT2-NpHR3 mice compared to control mice. The only exception was that VGLUT2-NpHR3 mice compared to control VGLUT2-EYFP mice displayed a significant increase of the LFP sigma PR (VGLUT2-NpPH3: 1.11±0.02, VGLUT2-EYFP: 1.03±0.00, p=0.0339; Fig 8).

Effect of optogenetic activation of SumL axon terminals innervating dDG on cFos expression
Mouse brains were processed for immunohistochemical detection of cFos in order to assess the effect of SuML axon terminal activation on DG cell activity. Whereas only a few neurons were labeled for cFos in the GCL of the DG in control VGLUT2-EYFP mice (n=4) (Fig 9 A, B), numerous neurons strongly labeled for cFos were observed in the granule cell layer of the dDG in VGLUT2-ChR2 mice (n=4) (Fig 9 C, D).
not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Neuroanatomical and "in vitro" experiments showing the dual glutamate-GABA feature of the SuML-DG pathway
Our results first establish at the neuroanatomical level that in mice as in rats (Soussi et al., 2010), all SuM neurons innervating the dDG display a dual GABAergic and glutamatergic neurotransmitter phenotype. We further demonstrate that these neurons correspond to a population of SuM cells located just dorsal to the mammillary tract that likely correspond to grandicellular neurons described in rat within the SuML region (Paxinos and Watson 1998) or SuMg (Pan and McNaughton 2004). In addition, our EM data show that SuML terminals form asymmetric (presumed excitatory) synapses onto some GCs and symmetric (presumed inhibitory) synapses onto others as previously described in rat (Boulland et al., 2009;Soussi et al., 2010). We further found in this study that one axon bouton from a SuML neuron can form an asymmetric (glutamate-preferring) synapse on a GC and a symmetric (GABA-preferring) synapse on another GC. In agreement with these neuroanatomical results, our in vitro electrophysiological experiments show that optical stimulation of SuM axon terminals innervating the dDG induce co-release of GABA and glutamate on almost all dentate GCs in line with two recent studies (Pedersen et al 2017;Hashimotodani et al 2018). Hashimotodani et al (2018) further showed that such co-transmission of GABA and glutamate induce net excitatory effects on GCs and potentiate GC firing when temporally associated with perforant path inputs. In line with such hypothesis, after in vivo optic stimulation of SuML axon terminals innervating the dDG in VGLUT2-ChR2 mice, we found that a significant number but not all GCs neurons below the optic fiber were labeled with cFos.
Therefore, these cFos labeled GCs could constitute a population of GCs that are not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint simultaneously activated by the optic stimulation of SuML axon terminals and perforant path inputs.

Effect of optogenetic stimulation of SuML-DG fibers on vigilance states
Activation of SuML axon terminals projecting to the DG induces an awakening effect in mice when performed during SWS but not during PS. It has been shown by Renouard et al (2015) that the SuML-DG pathway is active during PS and therefore its stimulation during PS might not induce WK because this pathway is already engaged and its overactivation might therefore not to be sufficient to awaken the animal. In contrast, when the stimulation occurs during SWS, an awakening is induced likely because the path is normally inactive during this state. It can be proposed that the stimulation of the DG granule cells by the SuML induces the reactivation of memories and subsequently of structures involved in the exploration leading to an awakening of the animal during SWS. It might be due also to the fact that there is no muscle atonia during this state compared to PS. In line with such hypothesis, stimulation during WK induces increased motor and exploratory activity.
Such result is in line with the literature since an increase in exploratory activity was reported when stimulating with blue light dDG neurons expressing ChR2 (Kheirbek et al., 2013).

Optogenetic stimulation of SuML-DG fibers increases gamma and theta
Our study further demonstrates that activation of SuML axon terminals innervating the DG during PS increases theta power and frequency as well as gamma power in the DG LFP and to a minor extent in the EEG. Activation of the SuML fibers during WK also induces an increase of theta frequency and power in the DG LFP and EEG not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint associated with an increased locomotion. During SWS, the activation of the SuML-dDG pathway induced awakening and a switch from delta to theta activity and an increase in gamma power both in DG LFP and EEG. It has been previously shown that the SuM can exert significant modulatory control of the theta rhythm (Vertes and Kocsis, 1997). A large percentage of SuM neurons discharge rhythmically, in phase with theta McNaughton, 1991, Kocsis and and this activity is independent of that occurring in the hippocampus.
Indeed, neurons in the SuM continue to fire bursts in the theta range frequency after lesion or pharmacological inactivation of the medial septum (Kirk and McNaughton, 1991) known to abolish theta rhythmic activity in the hippocampus. Further, electrical stimulation or carbachol injections in the SuM synchronously drive theta phaselocked cells in both septum (Bland et al., 1994) and hippocampus (Colom et al., 1987). In addition, the SuM controls the frequency and amplitude of theta (Kirk and McNaughton, 1993). These results indicated that the SuM plays a role in theta occurrence but experiments were performed in anesthetized animals and did not specifically study the SuML-DG pathway like in the present study. Indeed, SuM neurons also project directly to the medial septum (Vertes and McKenna, 2000) and can influence theta through the latter structure as well.
Our results also show that the frequency at the peak of theta is higher during PS than in WK in basal conditions. Further, we found out that optical stimulation of SuML-DG fibers during PS and WK induced a similar increase of the frequency at the peak of theta. However, no effect on theta was observed after inactivation of the fibers by halorhodopsin. It is likely that it is due to the fact that the inactivation was restricted to a small portion of the fibers located beneath the optic fiber. Indeed, lesion of the SuM induced a decreased of theta power specifically during PS (Renouard et al., 2015). not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint On the other hand, lesion of all neurons or specific inactivation of the GABAergic neurons of the medial septum strongly decreases theta during PS and WK (Mitchell et al., 1982, Green and Arduini, 1954, Boyce et al., 2016. Our data also show that stimulation of the SuML-DG fibers induces an increase in gamma power in the dDG and also in the EEG both during PS and WK. Interestingly, Montgomery et al. (2008) found by coherence analysis that dentate/CA3 theta and gamma power and synchrony was significantly higher during PS compared with active WK and that, in contrast, gamma power in CA1 and CA3-CA1 gamma coherence showed significant decreases during PS. These and our results strongly suggest that the medial septum GABAergic neurons induces theta both during WK and PS whereas the increase of theta and gamma power and theta frequency occurring in the DG during PS compared to WK is induced by the projection from the SumL.

Functional role of the SumL-DG pathway
It has been previously shown that GCs of the DG are instrumental for spatial discrimination (McHugh et al., 2007). In particular, it has been shown that small populations of GCs (2-4%) representing memory engrams are specifically activated when the animals are exposed to a specific context. These cells are reactivated each time the animal is re-exposed to the same context (Schmidt et al., 2012).
Optogenetic activation in a different context of a DG engram activated during contextual fear conditioning induces freezing (Liu et al., 2012). Conversely, inactivation during contextual fear memory recall of the DG GCs activated during encoding decreases freezing (Denny et al., 2014). These results indicate that activation of memory engrams composed of DG GCs cells is necessary for spatial learning. Interestingly, in these studies, the mean number of neurons labeled with not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 21, 2019. . https://doi.org/10.1101/584862 doi: bioRxiv preprint cFos or Arc in one section (35 µm) of the DG after contextual fear conditioning was in the 30-60 range. During stimulation of the SuML fibers in the DG, a mean of 67 DG cells were expressing cFos in the DG per section (40 µm). Finally, Renouard et al. (2015) counted a mean of 68 Arc-and 66 cFos-labeled neurons in one DG section after PS hypersomnia in rats. Therefore, approximatively the same number of DG GCs cells is activated during encoding of a contextual fear memory, stimulation of the SuML fibers in the DG and PS hypersomnia. First, the fact that a similar number of cells are activated when stimulating SuML terminals as during PS hypersomnia suggests that the activation seen during PS is likely due to activation of the SuML-DG pathway. Second, the fact that the a similar number of cells are activated during a memory task as during PS hypersomnia and stimulation of the SuML fibers suggest that memory engrams could be activated in these two conditions. Therefore, the induction of an active behavior when the stimulation is made during WK could be due to the activation of memory engrams. Since only a limited number of GCs cells are activated during PS and SuML terminals stimulation despite the fact that a large number of GCs neurons seems to be innervated by SuML axon terminals, it also suggests that activation of these cells is due to a conjunction of the SuML input with another excitatory input. It can be proposed that the medial entorhinal input is involved since it is the main excitatory afferent to the DG involved in its activation during memory encoding (Sasaki et al., 2015). Such hypothesis remains to be tested using optogenetic manipulation of the activated neurons during PS. not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.