GABA neurons in the ventral tegmental area regulate non-rapid eye movement sleep in mice

Sleep/wakefulness cycle is regulated by coordinated interactions between sleep- and wakefulness-regulating neural circuitry. However, the detailed mechanism is far from understood. Here, we found that glutamic acid decarboxylase 67-positive GABAergic neurons in the ventral tegmental area (VTAGad67+) are a key regulator of non-rapid eye movement (NREM) sleep in mice. VTAGad67+ project to multiple brain areas implicated in sleep/wakefulness regulation such as the lateral hypothalamus (LH). Chemogenetic activation of VTAGad67+ promoted NREM sleep with higher delta power whereas optogenetic inhibition of these induced prompt arousal from NREM sleep, even under highly somnolescent conditions, but not from REM sleep. VTAGad67+ showed the highest activity in NREM sleep and the lowest activity in REM sleep. Moreover, VTAGad67+ directly innervated and inhibited wake-promoting orexin/hypocretin neurons by releasing GABA. As such, optogenetic activation of VTAGad67+ terminals in the LH promoted NREM sleep. Taken together, we revealed that VTAGad67+ play an important role in the regulation of NREM sleep.


Introduction
Sleep or sleep-like behavioral quiescence is known to be one of the most ubiquitously observed phenomena across the animal kingdom, from nematodes to primates (Joiner, 2016;Siegel, 2008). Broadly, sleep consists of non-rapid eye movement (NREM) sleep and REM sleep in mammals (Siegel, 2008). While the physiological functions of either NREM sleep or REM sleep, or sleep as a whole, are intriguing and shrouded in mystery, sleep deprivation in humans and experimental animals causes severe cognitive impairment (Siegel, 2008). Pioneering studies discovered certain physiological functions of sleep that include clearing metabolic waste products and toxins from the brain (Xie et al., 2013), memory encoding, consolidation and erasure (Rasch and Born, 2013), synaptic homeostasis (Bushey et al., 2011), and energy conservation (Schmidt, 2014). However, a universal function of sleep that is relevant to all animals is yet to be revealed (Joiner, 2016). As animals remain largely isolated from sensory processing and goal-oriented activity during sleep, it is expected that the regulation of sleep, both NREM and REM, as well as arousal should be controlled by the central nervous system. Many brain areas and residing cellular subtypes have been shown to be critical in regulating sleep-wakefulness. For instance, orexin/hypocretin-producing neurons (orexin neurons) in the lateral hypothalamus (LH) project to and activate monoaminergic, cholinergic, and other peptidergic neurons as well as other orexin neurons to induce and maintain wakefulness (Brown et al., 2012;Inutsuka and Yamanaka, 2013;Sakurai, 2007;Scammell et al., 2017). Subsequently, these monoamine neurons inhibit sleep-active g-aminobutyric acid (GABA)-ergic neurons in the ventrolateral preoptic area (VLPO) in the hypothalamus to induce wakefulness (Saito et al., 2018;Saper et al., 2010). It is reported that some wake-active neurons also display activity during REM sleep (Brown et al., 2012;Scammell et al., 2017). Comparatively, NREM sleep is regulated by neurons that release classical fast neurotransmitters, including GABA. For example, circadian rhythms and/or homeostatic sleep pressures activate GABAergic neurons in the VLPO and median preoptic nucleus (MnPO), which in turn inhibit wake-promoting orexin/hypocretin, monoaminergic, and cholinergic systems (Scammell et al., 2017). While this flip-flop switch model of sleep-wake regulation is well established, recent studies have demonstrated a critical involvement of other brain areas and neuronal subtypes in regulating the transformations and subsequent maintenance of specific vigilances states (Liu et al., 2017;Oishi et al., 2017b).
Reinforcement learning, motivation, and locomotion, as well as the adaptation of responses to salient stimuli, all of which demand behavioral arousal, are critically regulated by a midbrain structure called the ventral tegmental area (VTA) in both rodents and primates (Arsenault et al., 2014;Fields et al., 2007). While one could also expect a critical role for VTA in the regulation of sleep/ wakefulness, it is only recently that VTA dopamine (DA) neurons have been reported to have a fundamental role in the maintenance of the awake state as well as in the consolidation of arousal in mice (Eban-Rothschild et al., 2016;Oishi et al., 2017a). However, the VTA contains considerable heterogeneity among the neuronal subtypes, which include GABAergic and glutamatergic neurons alongside DA neurons. Studies have reported that about 60-65% of VTA neurons are dopaminergic, whereas 30-35% are GABAergic, and 2-3% are glutamatergic neurons (Nair-Roberts et al., 2008;Pignatelli and Bonci, 2015).
As GABAergic neurons provide strong inhibition to the wake-and REM-active DA neurons in the VTA (Eban-Rothschild et al., 2016;Tan et al., 2012;van Zessen et al., 2012), it is probable that these GABAergic neurons in the VTA may also participate in sleep/wakefulness regulation. Moreover, GABA-mediated responses have been implicated in the modulation of the sleep/wakefulness cycle (Brown et al., 2012;Scammell et al., 2017). Here, we examined the role of glutamic acid decarboxylase 67 (Gad67)-positive neurons in the VTA on sleep/wakefulness by using AAV-aided whole-brain anterograde tracing, neural manipulations by chemo-and optogenetics, fiber photometry, as well as slice electrophysiology. Gad67 (encoded by Gad1 gene) is an isomer of an enzyme that synthesizes GABA from glutamic acid, suggesting that Gad67+ neurons are GABAergic neurons (Erlander et al., 1991). We revealed that Gad67+ neurons in the VTA (VTA Gad67+ neurons) are highly active during NREM sleep and send their axons to multiple brain areas that were previously reported to regulate sleep/wakefulness. Bidirectional manipulations of neuronal activity and fiber photometry recordings revealed that VTA Gad67+ neurons are active in and promote NREM sleep. Part of the NREM sleep-promoting effect of VTA Gad67+ is mediated through inhibition of wake-promoting orexin/hypocretin neurons in the LH.

Results
GABAergic neurons in the VTA project to brain areas involved in the regulation of sleep/wakefulness  Table  showing the comparative scoring of hrGFP signals across different brain areas. Abbreviations-AcbC: Nucleus accumbens core, AcbSh: Nucleus Figure 1 continued on next page the VTA, we counted a total of 636 ± 122 neurons per animal (n = 4 mice). Among them, 63.5 ± 1.8% were TH-positive neurons (DA neurons) and 36.1 ± 1.8% were hrGFP-positive neurons. Only 0.4 ± 0.1% were co-labeled with hrGFP and TH ( Figure 1c). hrGFP was distributed not only in the soma, but also in the axons. We could even anterogradely trace axons to reveal projection sites since hrGFP emits a strong fluorescence (Figure 1d-f). Along with local innervations, we found long-range projections of Gad67+ neurons in the VTA (VTA Gad67+ neurons) throughout the brain. Among these sites, the lateral hypothalamus (LH) and the central nucleus of the amygdala (CeA) were densely innervated (Figure 1e-f). Moderate projections were found in the nucleus accumbens (NAc), ventral pallidum (VP), parafascicular thalamic nucleus (PF), periaqueductal gray (PAG), ventral nucleus of the lateral lemniscus (VLL), dorsal raphe nucleus (DR), and pontine reticular nucleus (PnO). These brain areas are also reported to be involved in the modulation of sleep/wakefulness, suggesting that VTA-Gad67+ neurons might play a role in this regulation (Brown et al., 2012).
Chemogenetic activation of VTA Gad67+ neurons induced NREM sleep with high delta power Next, to reveal whether VTA Gad67+ neurons contribute to the regulation of sleep/wakefulness, we activated these neurons by means of pharmacogenetics (chemicogenetics), using designer receptors  Ventral nucleus of the lateral lemniscus, VP: Ventral pallidum, Xi: Xiphoid thalamic nucleus and xscp: Decussation of the superior cerebellar peduncles. DAPI (4 0 ,6-diamidino-2-phenylindole) staining was used to label nuclear DNA and also to assist understanding of the anatomical position of VTA Gad67+ neuronal projections. The online version of this article includes the following source data and figure supplement(s) for figure 1: Source data 1. Source data for Figure 1c.   exclusively activated by designer drugs (DREADD). We bilaterally injected a Cre-inducible AAV virus to express either hM3Dq-mCherry or mCherry into the VTA of Gad67-Cre mice (Figure 2a- We then confirmed the function of hM3Dq by applying its ligand clozapine-N-oxide (CNO) to acute brain slices while recording neuronal activity (Figure 2-figure supplement 1c). As expected, CNO application significantly increased the firing frequency of hM3Dq-expressing, but not mCherry-expressing, VTA Gad67+ neurons (Figure 2-figure supplement 1d-e, hM3Dq: 286 ± 61%, n = 8 cells; mCherry: 110 ± 8%, n = 6 cells, p=0.02, unpaired t-test). Next, to analyze the effect of CNO-induced activation of VTA Gad67+ neurons in sleep/wakefulness states, electroencephalogram (EEG) and electromyogram (EMG) electrodes were implanted in Gad67-Cre mice (Figure 2a). After recovery from the surgery and behavioral habituation (see Materials and methods), either saline or CNO (1 mg/kg) were administered intraperitoneally (i.p.) just before the onset of the dark period (at 8:00 pm). CNO administration resulted in a significantly reduced time spent in wakefulness and increased time spent in NREM sleep (also known as slowwave sleep) in the hM3Dq-mCherry expressing mice, but not in mCherry-expressing mice ( Figure 2c-d, hM3Dq: n = 6 mice, mCherry: n = 4 mice). The CNO-induced increase in NREM sleep lasted for at least 4 hr after CNO administration ( Figure 2d, % change from saline in hM3Dq-mCherry-expressing mice: wakefulness 22 ± 2, NREM 259 ± 18, REM 30 ± 8, vs saline NREM, p=3.0e-4, paired t-test). Interestingly, the delta power (1-5 Hz) during NREM sleep was significantly increased in the CNO-injected group compared to the NREM sleep in the saline-injected control group (mean relative delta power for 4 hr post-injection: hM3Dq-saine: 83 ± 3%, hM3Dq-saine: 138 ± 8%, p=0.001, paired t-test), suggesting that VTA Gad67+ neurons might be a critical regulator of slow-wave in NREM sleep (

Optogenetic inhibition of VTA Gad67+ neurons induced wakefulness
Since activation of VTA Gad67+ neurons resulted in increases in NREM sleep with increases in delta wave power, we next examined the selective inhibition of VTA Gad67+ neurons, which might be expected to increase wakefulness. To test this, we used an acute inhibition strategy with optogenetics. An inhibitory anion channel, anion channelrhodopsin-2 (ACR2, Genbank accession no. KP171709) (Mohammad et al., 2017), was expressed in VTA Gad67+ neurons ( Figure 3a  . We first confirmed the function of ACR2 employing in vitro electrophysiology. Three weeks after injection of AAV (expressing either ACR2-2A-mCherry or mCherry) into the VTA of Gad67-Cre mice, we prepared acute brain slices including the VTA and performed cell-attached recordings from mCherry-expressing neurons. Blue light (6.8 mW/mm 2 ) was able to completely silence the spontaneous activity of ACR2-2A-mCherry-expressing VTA Gad67+ neurons (n = 10 cells), whereas light irradiation on mCherry alone-expressing neurons had no such effect (n = 7 cells) (Figure 3-figure supplement 1d-f). Next, using these two groups of mice, we implanted fiber optics at a diameter of 400 mm into the VTA along with EEG and EMG electrodes (Figure 3a and b). After recovery and habituation, continuous blue light for 5 s was illuminated every 15 min for 24 hr ( Figure 3c). Interestingly, blue light illumination immediately induced wakefulness from NREM sleep, but not from REM sleep, in mice expressing ACR2-2A-mCherry (n = 6 mice, Figure 3d-f, Video 1). No such effect was observed in mice expressing mCherry alone (n = 5 mice, Video 2). However, as the light-induced influences on NREM sleep and wakefulness showed an extended effect after the cessation of light (Figure 3d-e), with behaviors taking around 60 s to return to the basal state, we sought to identify whether optogenetic inhibition of VTA Gad67+ neurons also causes prolonged wakefulness. We, therefore, isolated the trials depending on sleep-wakefulness states just before light illumination in the cases of wakefulness, NREM, or REM sleep (with the same state lasting !30 s before light illumination). Surprisingly, we found that optogenetic inhibition of VTA Gad67+ neurons in the state of wakefulness prolonged the time spent in wakefulness in all sorted trials compared to behavior of the control group ( Figure 3f, ACR2: 35 ± 5 s, mCherry: 8 ± 8 s; p=0.02, unpaired t-test). Again, REM sleep was not affected. Therefore, these data showed that in vivo optogenetic inhibition of VTA Gad67+ neurons promoted and sustained wakefulness in mice. This result clearly suggested that VTA Gad67+ neurons have a role in the regulation of not only NREM sleep but also wakefulness. Next, we tested whether brief (5 s) optogenetic inhibition of VTA Gad67+ neurons can induce arousal even under conditions of high homeostatic sleep pressure. To test this, mice were sleepdeprived for 4 hr, starting at light onset, and were then allowed to experience recovery sleep for 30 min ( Figure 4a). Sleep-deprived animals usually display extended NREM sleep because of high homeostatic sleep pressure. Moreover, the slow-wave activity in NREM sleep increases during recovery sleep (Lancel et al., 1992). However, to our surprise, even under such a higher sleep pressure condition, optogenetic inhibition of VTA Gad67+ neurons could successfully and immediately induce wakefulness in all trials (Figure 4b and c). Once again, induced-wakefulness displayed an extended effect after cessation of light, whereby it took 54 ± 14 s to return to NREM sleep. Taken together, these results suggest that VTA Gad67+ neurons might be involved in the initiation and maintenance of physiological NREM sleep.

VTA Gad67+ neurons showed the highest population activity during NREM sleep
Our chemogenetic activation and optogenetic inhibition studies suggest that the in vivo activity of VTA Gad67+ neurons might change across brain states with putatively higher activity during NREM sleep. To test this hypothesis, we recorded the population activity of VTA Gad67+ neurons using fiber photometry (Inutsuka et al., 2016). A Cre-inducible AAV expressing the fluorescent calcium indicator GCaMP6f  was unilaterally injected into the VTA of Gad67-Cre mice (n = 8 mice; Figure 5a and . Notably, VTA Gad67+ neurons began to increase their activity before wake-to-NREM transitions (mean DF/F: Wake: 2.9 ± 0.4%, NREM: 3.8 ± 0.4%, p=2.5e-6) and decrease their activity before NREM-to-REM (mean DF/F: NREM: 5.0 ± 0.5%, REM: 2.7 ± 0.3%, p=2.4e-5) and NREM-to-wake (mean DF/F: NREM: 3.8 ± 0.5%, wake: 3.0 ± 0.4%, p=2.4e-4) transitions. However, the changes in signal from REM-to-wake (mean DF/F: REM: 2.9 ± 0.4%, wake: 3.7 ± 0.5%, p=0.02) was comparatively less significant and occurred only after the onset of state transition. Most interestingly, the population activity of VTA Gad67+ neurons was found to be completely contrary to DA neuronal activity in the VTA (Dahan et al., 2007;Eban-Rothschild et al., 2016), further that delta wave power activity increases after CNO administration compared to the saline control. (g) Summary of the delta wave power change during NREM sleep after saline or CNO injection. Traces in dark color indicate mean value, while lighter color indicates EEG spectrum of each mice injected with saline (black) and CNO (red). Inset shows the mean relative delta power for 4 hr after either saline or CNO administration. Data are shown as the mean ± SEM (hM3Dq: n = 6 mice; mCherry: n = 4 mice). *p<0.05, ***p<0.001, (d) two-way ANOVA followed by Tukey post hoc, (e) twotailed paired Student's t-test (n = 8 mice). The online version of this article includes the following source data and figure supplement(s) for figure 2: Source data 1. Source data for Figure 2c-e and 2g.

VTA Gad67+ neurons directly inhibited wake-promoting orexin neurons in the lateral hypothalamus
Dense projections were observed from VTA Gad67+ neurons to a well-known sleep-wake regulatory brain region, the lateral hypothalamus (LH), where wake-active and wake-promoting orexin (LH orexin ) neurons are exclusively located. Thus, we wondered whether VTA Gad67+ neurons mediate their sleep-promoting effect through the inhibition of LH orexin neurons. To test this, we generated a bigenic orexin-Flippase (Flp); Gad67-Cre mouse, in which orexin neurons exclusively express Flp recombinase and Gad67+ neurons express Cre recombinase (Figure 6a-c) (Chowdhury et al., 2019). We injected a Cre-inducible AAV expressing the blue light-gated cation channel channelrho-dopsin2 (E123T/T159C; ChR2) (Berndt et al., 2011) in the VTA as well as a Flp-inducible AAV expressing tdTomato in the LH of orexin-Flp; Gad67-Cre mice (Figure 6a-c). In slice recordings from VTA Gad67+ neurons expressing ChR2, blue light flashes (6.8 mW/mm 2 ) through an objective lens could depolarize and significantly increase spontaneous firing frequency to approximately 650% compared with before light illumination (Figure 6d-f, n = 5 cells, p=0.004 vs either pre or post, oneway ANOVA followed by post-hoc Tukey). Next, we recorded spontaneous firings from tdTomatopositive neurons (orexin neurons) in the LH by loose cell-attached recordings, and the nerve terminals of VTA Gad67+ neurons in the LH were activated by illuminating blue light pulses (Figure 6g and To reveal the mechanism of inhibition of orexin neurons, we performed additional electrophysiological experiments. We performed whole-cell voltage clamp recordings from orexin neurons at À60 mV holding potential (mV hold ) to record post-synaptic currents. Activation of nerve terminals of VTA Gad67+ neurons in the LH (blue light pulse, duration of 5 ms) induced a post-synaptic current (PSC) in 8 out of 11 cells. These lightinduced PSCs were blocked by gabazine (10 mM), a GABA A receptor antagonist (Figure 7a-c, aCSF: À253 ± 70 pA, gabazine: À9 ± 3 pA, n = 8 cells, p=1.6e-10, paired t-test). This result suggests that GABA is involved in generating the light-induced PSCs in LH orexin neurons. The average synaptic delay from light onset was recorded as 6.2 ± 1.0 ms (Figure 7d). To rule out the effect of glutamate, we blocked both AMPA (a-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid) and NMDA (N-Methyl-D-aspartic Figure 3 continued EEG, EEG power spectra, and EMG during optogenetic inhibition in different vigilance states (wake, NREM, and REM sleep). Vigilance states are indicated by colored bars above the EEG traces. (e) Probability of vigilance state before and after light illumination in all recorded trials of ACR2-2A-mCherry or mCherry-alone expressing mice. Blue and red lines indicate mean probability of each vigilance state, ACR2-2A-mCherry (n = 6 mice) and mCherry (n = 5 mice). (f) Light illumination during wakefulness, NREM, or REM sleep was isolated from subfigure e (ACR2-2A-mCherry = 6 mice; and mCherry = 5 mice). Each vigilance state lasted for at least 30 s before light illumination was isolated. SEM is indicated as the lighter color band. The online version of this article includes the following source data and figure supplement(s) for figure 3: Source data 1. Source data for Figure 3e and 3f.  Video 1. Optogenetic inhibition of VTA Gad67+ neurons during NREM sleep, REM sleep, and wakefulness. Blue light (475 ± 18 nm, 10 mW, 5 s) was illuminated in the VTA area in each vigilance state to inhibit ACR2expressing VTA Gad67+ neurons. EEG, EMG, and light stimulation signals appear at the top of the window.

VTA Gad67+ neurons mediate a sleep-promoting function via the lateral hypothalamus
Finally, to test whether the VTA Gad67+ to LH projection participates in the induction of NREM sleep, we performed in vivo optogenetic activation of VTA Gad67+ nerve terminals in the LH. We expressed either ChR2-EYFP (n = 7 mice) or hrGFP (n = 5 mice) in the VTA Gad67+ neurons in a Cre-dependent manner using Gad67-Cre mice. Two weeks later, we bilaterally implanted fiber optics at a diameter of 400 mm above the LH (Figure 8a and b). After recovery and habituation, blue light (20 Hz, 10 mW) was applied for 60 min in the dark period when mice were awake and active (Figure 8c). Interestingly, blue light illumination into the LH resulted in decreased wakefulness and increased NREM sleep in the ChR2-expressing mice, whereas hrGFP-expressing mice showed no such effects (Figure 8c-e, in ChR2-expressing mice: wakefulness (in min) Pre: 55.0 ± 2.2, Stimulation: 29.2 ± 1.7, Post: 53.3 ± 2.1; p=7.3e-8 vs Pre, p=2.7e-7 vs Post, NREM (in min) Pre: 4.3 ± 1.8, Stimulation: 29.9 ± 1.8, Post: 6.0 ± 1.9, p=0 vs Pre, p=5.5e-8 vs Post, one-way ANOVA followed by post-hoc Tukey). Again, the REM sleep remained unaffected (Figure 8c-e, in ChR2-expressing mice: REM (in min) Pre: 0.7 ± 0.5, Stimulation: 0.9 ± 0.3, Post: 0.7 ± 0.4, p=0.94 vs Pre, p=0.95 vs Post, one-way ANOVA followed by post-hoc Tukey). These results suggest that VTA Gad67+ neurons promote NREM sleep, at least in part, through their projection to the LH.

Discussion
By employing anterograde tracing and localization of brain-wide neural projections, bidirectional neuronal manipulations, fiber photometry, slice electrophysiology, as well as sleep recordings, we provide multiple lines of evidence in favor of our claim that VTA Gad67+ neurons regulate NREM sleep in mice. GABAergic neurons constitute a significant part of the VTA (Nair-Roberts et al., 2008;Pignatelli and Bonci, 2015) and help to regulate the function of DA neurons residing nearby (Tan et al., 2012;van Zessen et al., 2012). Dysregulation of signaling pathways in the VTA is associated with drug abuse and several other psychiatric disorders including schizophrenia, bipolar Video 2. mCherry control of optogenetic inhibition during NREM sleep, REM sleep, and wakefulness. Blue light (475 ± 18 nm, 10 mW, 5 s) was illuminated in the VTA area in each vigilance state in mCherry-expressing Gad67-Cre mice. EEG, EMG, and light stimulation signals appear at the top of the window.
https://elifesciences.org/articles/44928#video2  Figure 4 continued on next page disorder, and major depressive disorder (Winton-Brown et al., 2014;Wulff et al., 2010). Moreover, irregular sleep-wake timing and architectures are recognized as common co-morbidities in many neuropsychiatric and neurodegenerative diseases (Wulff et al., 2010). Therefore, the relationship between neurochemical signaling in the VTA and the regulation of sleep/wakefulness poses an interesting point of study. However, classical lesioning experiments suggest that cats with reduced dopamine levels exhibit decreased behavioral arousal but no significant change in electro-cortical waking (Jones et al., 1973). It is only very recently that investigators have shown an interest in understanding the role of the VTA in the regulation of sleep/wakefulness (Eban-Rothschild et al., 2016;Oishi et al., 2017a;Yang et al., 2018). However, not much scientific literature has been published focusing on the functional importance of GABAergic neurons in the VTA. Therefore, our findings on the role of these neurons in sleep/wakefulness regulation will provide a conceptual and systematic framework for the association between sleep and psychiatric disorders and will generate opportunities to study VTA-related dysregulation in mental disorders.
van Zassen and colleagues reported that in vivo optogenetic activation of GABAergic neurons in the VTA in mice disrupts reward consummatory behavior (van Zessen et al., 2012). In addition, Shank et al. reported that dose-and time-related selective ablation of GABAergic neurons in the VTA in rats increased spontaneous locomotor activity (Shank et al., 2007). These studies are consistent with the hypothesis that GABAergic neurons in the VTA play an important role in the regulation of behavior. We now argue that one reason for such disruption in behavior might be promotion of NREM sleep by selective activation of GABAergic neurons in the VTA.
Using bidirectional chemogenetic manipulations as well as neurotoxic lesions in rats, a recent study found that neurons in the rostromedial tegmental nucleus (RMTg), also known as the GABAergic tail of the VTA, are essential for physiological NREM sleep (Yang et al., 2018). Although Yang et al. did not identify neuronal subtypes involved in the RMTg, their results might be related to our findings. Interestingly, VTA Gad67+ neurons in our study are located throughout the VTA, but at a somewhat higher density toward the caudal parts of the VTA. More recently, Takata et al. reported that GABA neurons in the ventral medial midbrain/pons, which includes the VTA region, regulate sleep/wake cycles by modulating DA neurons (Takata et al., 2018). These GABA neurons should include VTA Gad67+ neurons. Indeed, GABAergic neurons regulating NREM sleep might be distributed across both VTA and RMTg.
Chemogenetic activation of VTA Gad67+ neurons induced NREM sleep accompanied by higher delta power (slow wave) compared with control conditions (Figure 2g), suggesting that VTA Gad67+ neurons might play a critical role in the generation of slow wave in NREM sleep. Recently, Oishi et al. reported that activation of either the cell bodies of GABAergic neurons in the core of NAc or their axonal terminals in the VP evoked slow wave sleep (Oishi et al., 2017b). In addition, activation of GABAergic neurons in the basal forebrain, which includes the VP, produced wakefulness, whereas their inhibition induced sleep (Anaclet et al., 2015). These facts suggest that inhibition of GABAergic neurons in the VP is a critical pathway to generate slow waves in NREM sleep. We also found that VTA Gad67+ neurons moderately project to the VP. Therefore, we reasoned that VTA Gad67+ neurons projecting to the VP might be involved in the generation of slow wave sleep.
Population activity recordings across vigilance states shows that DA neurons in the VTA exhibit higher activity in REM sleep versus either wake or NREM sleep (Eban-Rothschild et al., 2016). On the contrary, VTA Gad67+ neurons exhibit a completely opposite activity pattern from that of DA neurons across vigilance states, with highest activity during NREM sleep ( Figure 5). This suggests an existing functional interaction between DA and Gad67+ neurons in the VTA. Using in vivo single unit recordings in rats, Lee et al. found wake-and REM-active VTA GABA neurons, suggesting that there might be several types of VTA GABA neurons (Lee et al., 2001). Our fiber photometry data showed that VTA Gad67+ neurons exhibit weak activity even during wakefulness. This might suggest that    (Yu et al., 2019). Although this study similarly found a NREM-sleep promoting role of VTA GABA neurons, the in vivo activity of VTA GABA neurons was quite different from our observations with the highest activity observed during wake and REM sleep. One difference between studies was that Yu et al. used Vgat-Cre mice (Cre recombinase is targeted to the slc32a1 gene) and we used Gad67-Cre mice to target GABAergic neurons. Our in situ hybridization results showed that Gad67-positive neurons represent a small subset of Vgat-positive neurons in the VTA. Again, this difference further indicates the existence of different populations of GABAergic neurons in the VTA.
Optogenetic inhibition of VTA Gad67+ neurons induced immediate wakefulness from NREM sleep, but not from REM sleep, suggesting the importance of uninterrupted neuronal activity of VTA Gad67+ neurons for the maintenance of NREM sleep. Both chemogenetic activation and optogenetic inhibition data suggest that VTA Gad67+ neurons might not play a decisive role in the physiological regulation of REM sleep. Interestingly, inhibition of VTA Gad67+ neurons prolonged wakefulness (Figure 3f). This result might suggest that VTA Gad67+ neurons also regulate levels of wakefulness. This is consistent with data showing that VTA Gad67+ neurons displayed weak activity in wakefulness in terms of population activity. This idea is also supported by observed increases in spontaneous locomotor activity following selective ablation of VTA GABA neurons in rats (Shank et al., 2007). These facts might suggest that an improvement in alertness and ability to maintain wakefulness require the suppression of activity of VTA GABA neurons.
One possible cellular mechanism underlying NREM sleep promotion by VTA GABA neurons is via inhibition of DA neurons residing in the VTA. In addition to this, our results showed that direct inhibition of wake-promoting LH orexin neurons might contribute to the induction of NREM sleep. Projection-specific activation of VTA GABA neuron nerve terminals using optogenetics indicated that VTA GABA -to-LH represents a pathway responsible for inducing NREM sleep. It will also be fascinating to study how VTA GABA neurons are regulated. Using an optimized trans-synaptic retrograde tracing approach, Faget and colleagues recently labeled afferent neurons to DA, GABA, or glutamate neurons in the VTA and found that these populations receive qualitatively similar inputs, with dominant and comparable projections from three brain areas known to be critical for sleep/wakefulness regulation: LH, raphe, and ventral pallidum (Faget et al., 2016). Here we report that VTA Gad67+ neurons project to those areas, suggesting the existence of a mutual interaction with these brain areas to regulate sleep/wakefulness. Moreover, while many brain areas including the cholinergic basal forebrain and brain stem, histaminergic posterior hypothalamus, serotonergic raphe nucleus, as well as the noradrenergic locus coeruleus are important in sleep/wake regulation (Brown et al., 2012), we speculate a critical role for VTA Gad67+ neurons in sleep/wakefulness-regulating circuitry as these neurons modulate diverse targets, including DA neurons in the VTA (Tan et al., 2012).
In conclusion, our study elucidated that VTA GABA neurons regulate NREM sleep in mice. These neurons might be a possible target for therapeutic intervention in treating sleep-related as well as neuropsychiatric disorders.       Continued on next page Figure 6 continued The online version of this article includes the following source data and figure supplement(s) for figure 6: Source data 1. Source data for Figure 6f.

Animals
All experimental protocols that involved animals were approved by the Institutional Animal Care and Use Committees, Research Institute of Environmental Medicine, Nagoya University, Japan (Approval number #18232, #18239). All efforts were made to reduce the number of animals used and also to minimize the suffering and pain of animals. Animals were maintained on a 12 hr light-dark cycle (lights were turned on at 8:00 am), with free access to food and water.

Immunohistochemistry
Under deep anesthesia with 0.65% pentobarbital sodium solution (Kyoritsu Seiyaku Corporation, Tokyo, Japan) diluted with saline (1.0 ml/kg body weight), mice were subjected to serial transcardial perfusion first using ice-cold saline (20 ml) and then ice-cold 4% formaldehyde solution (20 ml, Fujifilm Wako Pure Chemical Industries, Ltd., Osaka, Japan). The brain was then gently collected and post-fixed with 4% formaldehyde solution at 4˚C overnight. Later, the brain was subsequently immersed in phosphate-buffered saline (PBS) containing 30% sucrose at 4˚C for at least 2 days. Coronal sections of either 40 or 80 mm thickness were made using a cryostat (Leica CM3050 S; Leica Microsystems, Wetzlar, Germany; or Leica VT1000 S, Wetzlar, Germany), and slices were preserved in PBS containing 0.02% of NaN 3 at 4˚C until stained. For staining, coronal brain sections were immersed in blocking buffer (1% BSA and 0.25% Triton-X in PBS), and then incubated with primary antibodies (TH: Millipore, Massachusetts, 1/1000 dilution; DAT: Frontier Institute Co. Ltd., Hokkaido, Japan, 1/1000 dilution, Japan; DsRED: Santa Cruz Biotechnology, Heidelberg, Germany, 1/1000 dilution; GFP: Fujifilm Wako Pure Chemical Corporation, Osaka, Japan, 1/1000 dilution; orexin-A: Santa Cruz Biotechnology, 1/1000 dilution) at 4˚C overnight. For Gad67 staining, slices were incubated with anti-Gad67 antibody (Millipore, 1/500 dilution in blocking buffer without Triton-X) at 4˚C for 4 days. After washing by blocking buffer three times, the brain sections were then incubated with secondary antibodies for 1 hr at room temperature. After washing with the same blocking solution three times, slices were stained by DAPI (Thermo Fisher Scientific, Waltham, MA) across several experiments. Slices were mounted in 50% glycerol solution and examined with an epifluorescence microscope (BZ-9000, Keyence, Osaka, Japan or IX71, Olympus, Tokyo, Japan).

Anterograde tracing and localization of brain-wide neural projections
A Cre-inducible AAV carrying the hrGFP gene (AAV(9)-CMV-FLEX-hrGFP; 100 Â 1 nl, 6.0 Â 10 12 copies/ml) was unilaterally injected into the VTA of Gad67-Cre mice. Three weeks post-injection, animals were perfused-fixed and brain slices of 80 mm thickness were made serially from the anterior to the posterior part of the brain using a vibratome (Leica VT1000 S, Wetzlar, Germany). After DAPI staining, slices were serially mounted and images were taken using an epifluorescence microscope (BZ-9000, Keyence, Osaka, Japan or IX71, Olympus, Tokyo, Japan). Images were taken using an identical configuration in the microscope and were then analyzed using ImageJ (US National Institute of Health) software. Projection scorings were made in all visible projection sites, except for the VTA, by first selecting the most innervated brain region and comparing other areas to that region.

In situ RNA hybridization using RNAscope
Under deep anesthesia with 0.65% pentobarbital sodium solution (Kyoritsu Seiyaku Corporation, Tokyo, Japan) diluted with saline (1.0 ml/kg body weight), mice were transcardially perfused using ice-cold saline (20 ml) and then using ice-cold 4% formaldehyde solution (20 ml, Fujifilm Wako Pure Chemical Industries, Ltd., Osaka, Japan). The brain was then gently removed, post-fixed with 4% paraformaldehyde solution overnight at 4˚C and was subsequently immersed in phosphate-buffer containing 30% sucrose at 4˚C for at least 2 days. Coronal brain sections of 25 mm thickness were made using a cryostat (Leica CM3050 S; Leica Microsystems) and were mounted on glass slides (SMAS-01, Matsunami, Japan), and fixed in 4% paraformaldehyde for 60 min. The slides were then treated by RNAscope multiplex fluorescent v2 (#323100, Advanced Cell Diagnostics, Hayward, CA) according to the RNAscope standard protocol. In short, slides were incubated with hydrogen peroxide at RT for 10 min, followed by boiling with target retrieval reagent at 98~102˚C for 5 min, and protease digestion at 40˚C in a HybEZ hybridization oven (Advanced Cell Diagnostics) for 30 min. Subsequently, slides were incubated at 40˚C with target probes in hybridization buffer for 2 hr, AMP1 in hybridization buffer for 30 min, AMP2 in hybridization buffer for 30 min, and AMP3 in hybridization buffer for 15 min. After each hybridization step, slides were washed with wash buffer twice at room temperature. For multiplex detection, equimolar amounts of target probes, AMP1, AMP2 and AMP3 of each amplification system were used. Sequences of target probes, AMP1,AMP2,and AMP3 are proprietary (vGAT,GAD1 and mCherry,Advanced Cell Diagnostics,Hayward,CA). For fluorescent detection, the RNA probes were conjugated to Opal 520, 620, or 690 with the HRP and TSA Plus fluorophores system (Perkin Elmer, Waltham, MA). Slices were mounted with Pro-Long Gold Antifade Mountant (Thermo Fisher Scientific, Waltham, MA) and observed via a confocal microscope (LSM 710 Zeiss, Oberkochen, Germany).

Surgery for EEG-EMG and/or optogenetics, fiber photometry
Procedures for implanting EEG and EMG electrodes for polysomnographic recording experiments were adapted from the previously published protocol (Tabuchi et al., 2014). Briefly, virus-injected mice were implanted with EEG and EMG electrodes under isoflurane anesthesia. Immediately after surgery, each mouse received an i.p. injection of 10 ml/kg of an analgesic solution containing 0.5 mg/ml of Carprofen (Zoetis Inc, Parsippany-Troy Hills, NJ). Mice were singly housed for 7 days during the recovery period. Mice were then connected to a cable in order to allow them to move freely in the cage as well as to be habituated to the recording cable for another 7 days.
For fiber-guided optogenetic experiments, virus-injected mice received a surgical implantation of single fiber optic cannula (400 mm; Lucir Inc, Japan), along with EEG-EMG electrodes, above the VTA (AP À3.3 to À3.6 mm; ML 0.4 to 0.6 mm; DV À3.75 mm). For fiber photometry experiments, virus-injected mice received surgical implantation of a single guide cannula (400 mm; Thorlabs Inc) just above the VTA (AP À3.3 mm; ML 0.4 to 0.5 mm; DV À4.0 mm) to target VTA Gad67+ neurons. These mice were also implanted with the EEG-EMG electrodes following the protocol described above.
For projection-specific optogenetic activation of VTA Gad67+ nerve terminals in the LH, virusinjected mice received bilateral implantation of fiber optic cannula (400 mm; Kyocera Corporation, Japan), along with EEG-EMG electrodes, above the LH at a stereotaxic co-ordinate of AP À1.4 mm; ML ± 0.9 mm; DV À5.0 mm.

Vigilance state determination
EEG and EMG signals were amplified (AB-610J, Nihon Koden, Japan), filtered (EEG at 1.5-30 Hz, and EMG at 15-300 Hz), digitized (at a sampling rate of 128 Hz), and recorded (Vital Recorder, Kissei Comtec Co., Ltd, Japan) from individual habituated mice. Recorded signals were then analyzed to identify vigilance states using SleepSign (Kissei Comtec) software. Vigilance state identification was assisted by an infrared sensor as well as by video monitoring through a CCD video camera (Amaki Electric Co., Ltd., Japan) during both the light and dark periods (Kissei Comtec). Video recording during the dark period was aided by infrared photography (Amaki Electric Co., Ltd., Japan). EEG and EMG data were automatically scored in epochs (every 4 s) and classified as wake, REM sleep, or NREM sleep. All auto-screened data were examined visually and corrected. The EEG analysis yielded power spectra profiles over a 0~20 Hz window with 1 Hz resolution for delta (1-5 Hz), theta (6-10 Hz), alpha (11-15 Hz), and beta (16-20 Hz) bandwidths. The criteria for determining vigilance states were the same as the protocol described elsewhere (Tabuchi et al., 2014): briefly, (i) wake (low EEG amplitude with high EMG or locomotion score), (ii) NREM sleep (low EMG and high EEG delta amplitude), and (iii) REM sleep (low EMG as well as low EEG amplitude with high theta activity, and should be followed by NREM).
In vivo recordings and data analysis of neuronal activity using fiber photometry In vivo population activity of the VTA Gad67+ neurons was recorded using a silica fiber of 400 mm by implanting the fiber just above the VTA. Details of the fiber photometric recordings are described  (Inutsuka et al., 2016). Briefly, the fiber photometry system (COME2-FTR/OPT, Lucir, Tsukuba, Japan) utilizes a custom-made single silica fiber of 400 mm diameter to deliver excitation light and to detect fluorescence from GCaMP6f, simultaneously. Blue excitation light (465 nm, 0.5 mW at the tip of the silica fiber) was produced by a high-power LED system (PlexonBright OPT/LED Blue_TT_FC, Plexon, Dallas, TX). The LED-emitted excitation light was reflected by a dichroic mirror and coupled to the silica fiber (400 mm diameter) through an excitation bandpass filter (path 472 ± 35 nm). GCaMP6f-emitted green fluorescence was collected by the same silica fiber passed through a bandpass emission filter (path 525 ± 25 nm) and guided to a photomultiplier (PMTH-S1M1-CR131, Zolix instruments, Beijing, China). The fiber photometry signal was recorded by Vital Recorder (Kissei Comtec Co., Ltd, Japan) along with the EEG/EMG signals. Fiber photometry signals were collected at a sampling frequency of 128 Hz and the software averaged every 10 samples to minimize fluctuations and noise.
After recording and sleep analysis, the fiber photometry signal was outputted along with the EEG and EMG signals as a text file of raw data. For each experiment, the photometry signals at all data points were motion averaged and were then converted to DF/F by DF/F(t) = (F(t)-F min )/F min . We recorded the signals in the light period as nocturnal animal mice usually show multiple transitions among different vigilance states during the light period. All mice were subjected to at least two recording sessions with at least a 2 day interval in between each session to allow photobleaching recovery. We separated all sleep-state transitions that last at least for 1 min before and after the state change. All the sessions were selected after the photometry signal became stable, as we observed a decay of photometry signal at the beginning of the recordings.

Sleep deprivation
Mice expressing ACR2-2A-mCherry in VTA Gad67+ neurons were used for optogenetic inhibition in Figure 3 and were subjected to sleep deprivation. Mice were submitted to complete sleep deprivation, through the gentle handling method, which consists of keeping the animal awake by gently touching them with a soft brush if behavioral signs of sleep were observed. Mice were sleepdeprived for 4 hr, starting at light onset. Mice were then allowed to experience recovery sleep for 30 min. Optogenetic inhibition was performed during the recovery sleep.

In vitro electrophysiology
After the recovery period, acute brain slices were transferred to a recording chamber (RC-26G; Warner Instruments, Hamden, CT). The recording chamber was equipped with an upright Figure 8 continued illumination). Vigilance states determined by EEG and EMG signals are indicated by colored bars. (d and e) Probability of vigilance state (d) and total time spent (e) of either ChR2-or hrGFP-expressing mice. Blue and green lines indicate mean probability of each vigilance state in ChR2 (n = 7 mice) and hrGFP (n = 5 mice) group, respectively. Data are represented as mean ± SEM. ***, p<0.001 (vs Pre); +++ , p<0.001 (vs Post). p values were calculated by one-way ANOVA followed by post-hoc Tukey test. The online version of this article includes the following source data for figure 8: Source data 1. Source data for Figure 8d and 8e. fluorescence microscope (BX51WI; Olympus, Tokyo, Japan) stage and was superfused with a 95% O 2 and 5% CO 2 -gassed bath solution at a rate of 1.5 ml/min using a peristaltic pump (Dynamax; Rainin, Oakland, CA). An infrared camera (C3077-78; Hamamatsu Photonics, Hamamatsu, Japan) was installed in the fluorescence microscope along with an electron multiplying charge-coupled device camera (Evolve 512 delta; Photometrics, Tucson, AZ) and both images were separately displayed on monitors. Micropipettes of 4-6 MW resistance were prepared from borosilicate glass capillaries (GC150-10; Harvard Apparatus, Cambridge, MA) using a horizontal puller (P-1000; Sutter Instrument, Novato, CA). Patch pipettes were filled with KCl-based internal solution (in mM: 145 KCl, 1 MgCl 2 , 10 HEPES, 1.1 EGTA, 2-Mg-ATP, 0.5 Na 2 -GTP; pH 7.3 with KOH) with osmolality between 280-290 mOsm. Electrophysiological properties of cells were monitored using an Axopatch 200B amplifier (Axon Instrument, Molecular Devices, Sunnyvale, CA). Output signals were low-pass filtered at 5 kHz and digitized at a sampling rate of 10 kHz. Patch clamp data were recorded through an analog-to-digital (AD) converter (Digidata 1550A; Molecular Devices) using pClamp 10.2 software (Molecular Devices). Voltage clamp recordings were performed at a holding potential of À60 mV, unless otherwise stated. Blue light at a wavelength of 475 ± 18 nm and yellow light at a wavelength of 575 ± 13 nm were generated by a light source that used a light-emitting diode (Spectra Light Engine; Lumencor, Beaverton, OR) and guided to the microscope stage with a 1 cm diameter optical fiber. Brain slices were illuminated through the objective lens of the fluorescence microscope.

In vitro calcium imaging
Gad67+ neurons were identified via green fluorescence of GCaMP6f. Excitation light of 475 ± 18 nm (6.8 mW/mm 2 ) was emitted into the brain slice containing VTA through the objective lens of a fluorescence microscope. The light source (Spectra light engine) was controlled by Metamorph software (Molecular Devices). GCaMP6f fluorescence intensity was recorded continuously using Metamorph software at a rate of 1 Hz with 100 ms of exposure time. To synchronize calcium imaging and patch clamp recording, pClamp software was triggered by the TTL output from Metamorph software. Metamorph data were analyzed by setting the region of interest (ROI) on GCaMP6f-expressing VTA Gad67+ neurons and the DF/F was calculated from the average intensity of the ROI. Finally, DF/F values for 10, 20, and 50 Hz were normalized to the DF/F values for corresponding 100 Hz frequencies.

Data analysis and presentation
Immunostaining data were analyzed and processed with ImageJ (US National Institute of Health) and BZ-X Analyzer (Keyence BZ-X710 microscope). Electrophysiological analysis was performed with either Clampfit10 (Molecular Devices, Sunnyvale, CA) or Minianalysis software (Synaptosoft Inc, Decatur, GA). Analysis of EEG-EMG data was performed using SleepSign software (Kissei Comtec) and data were outputted as text files. Further analyses were performed using Microsoft Excel. Electrophysiological data were saved as American Standard Code for Information Interchange (ASCII) files and further data calculations were performed in Microsoft Excel. Graphs were generated in Origin 2017 (OriginLab, Northampton, MA) using data from Excel. Statistical analysis was also performed with Origin 2017. Graphs were generated using Canvas 15 (ACD Systems, Seattle, WA). The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.