Bottom-Up versus Top-Down Induction of Sleep by Zolpidem Acting on Histaminergic and Neocortex Neurons

Zolpidem, a GABAA receptor-positive modulator, is the gold-standard drug for treating insomnia. Zolpidem prolongs IPSCs to decrease sleep latency and increase sleep time, effects that depend on α2 and/or α3 subunit-containing receptors. Compared with natural NREM sleep, zolpidem also decreases the EEG power, an effect that depends on α1 subunit-containing receptors, and which may make zolpidem-induced sleep less optimal. In this paper, we investigate whether zolpidem needs to potentiate only particular GABAergic pathways to induce sleep without reducing EEG power. Mice with a knock-in F77I mutation in the GABAA receptor γ2 subunit gene are zolpidem-insensitive. Using these mice, GABAA receptors in the frontal motor neocortex and hypothalamic (tuberomammillary nucleus) histaminergic-neurons of γ2I77 mice were made selectively sensitive to zolpidem by genetically swapping the γ2I77 subunits with γ2F77 subunits. When histamine neurons were made selectively zolpidem-sensitive, systemic administration of zolpidem shortened sleep latency and increased sleep time. But in contrast to the effect of zolpidem on wild-type mice, the power in the EEG spectra of NREM sleep was not decreased, suggesting that these EEG power-reducing effects of zolpidem do not depend on reduced histamine release. Selective potentiation of GABAA receptors in the frontal cortex by systemic zolpidem administration also reduced sleep latency, but less so than for histamine neurons. These results could help with the design of new sedatives that induce a more natural sleep. SIGNIFICANCE STATEMENT Many people who find it hard to get to sleep take sedatives. Zolpidem (Ambien) is the most widely prescribed “sleeping pill.” It makes the inhibitory neurotransmitter GABA work better at its receptors throughout the brain. The sleep induced by zolpidem does not resemble natural sleep because it produces a lower power in the brain waves that occur while we are sleeping. We show using mouse genetics that zolpidem only needs to work on specific parts and cell types of the brain, including histamine neurons in the hypothalamus, to induce sleep but without reducing the power of the sleep. This knowledge could help in the design of sleeping pills that induce a more natural sleep.


Introduction
Many healthy people who cannot sleep, as well as many with neurological or mental illness, take sedatives (Wafford and Ebert, 2008;Winsky-Sommerer, 2009;Nutt and Stahl, 2010;Mignot, 2013;Rihel and Schier, 2013). Zolpidem (Ambien), a nonbenzodiazepine GABA A receptor-positive modulator, is currently the most successful "sleeping pill" (Nutt and Stahl, 2010). In the United States alone, there were 53.4 million prescriptions for zolpidem in 2010 (Greenblatt and Roth, 2012). The drug decreases sleep latency, the time to the onset of NREM sleep (Arbilla et al., 1985;Gottesmann et al., 1998;Alexandre et al., 2008;Anaclet et al., 2012;Xu et al., 2014). After taking a 10 mg tablet of zolpidem, the average person goes to sleep after ϳ12 min (Greenblatt and Roth, 2012). Compared with natural (drug-free) NREM sleep, however, the NREM sleep induced by zolpidem has reduced power in most EEG frequencies (Landolt et al., 2000;Kopp et al., 2004;Alexandre et al., 2008). This reduced power may indicate that zolpidem-induced sleep is suboptimal compared with natural NREM sleep.
We previously used a pharmacogenetics method, based on mutated GABA A receptors, for probing how endogenous GABA inputs onto selected cell types generates behavior (Wulff et al., 2007;Wisden et al., 2009;. The mutation F77I in the ␥2 subunit (␥2I77) abolishes zolpidem binding to GABA A receptors (Buhr et al., 1997;Wingrove et al., 1997;Cope et al., 2004). In ␥2I77 lox mice, the zolpidem-insensitive ␥2I77 subunits can be swapped with zolpidem-sensitive ␥2F77 versions (Wulff et al., 2007). Here, using this method, we found that strengthening inhibition onto histamine neurons by zolpidem induces NREM sleep but does not reduce EEG power.

Materials and Methods
Mice. All experiments were performed in accordance with the United Kingdom Home Office Animal Procedures Act (1986); all procedures were approved by the Imperial College Ethical Review Committee. All mice weighed between 19 and 30 g and were ϳ17 weeks old at the time of AAV injections. Both male and female mice were used, but no sex differences were observed and the data were pooled. The sleep-wake studies and drug administrations were started ϳ4 weeks after AAV injection (see below). Adult C57BL/6J mice were purchased from Harlan. The HDC-⌬␥2I77 mice were produced by crossing HDC-Cre (JAX stock #021198, RRID:IMSR_JAX:021198) and ␥2I77 lox -(zolpidem-insensitive) mice (JAX stock #021197, RRID:IMSR_JAX:021197) on C57BL/6J backgrounds, as described previously (Zecharia et al., 2012). In adult HDC-Cre mice, Cre recombinase expression is driven by the endogenous hdc gene and is found selectively in histaminergic neurons in the TMN, and mast cells in the rest of the brain; the knock-in HDC-Cre allele expresses functional HDC protein (Zecharia et al., 2012). In the ␥2I77 lox -mice, exon 4 of the GABA A receptor ␥2 subunit gene ( gabrg2), which encodes the critical I77 residue, is flanked by loxP sites (Wulff et al., 2007); deletion of exon 4 by Cre recombinase produces a null gabrg2 allele (Wulff et al., 2007(Wulff et al., , 2009aRovó et al., 2014). The baseline vigilance-state data (% Wake, NREM, and REM) recorded for a 2 h period, as determined by EEG/EMG scoring, for the mice in drug-free conditions are shown in Table 1 (see EEG recordings and sleep scoring).
AAV transgenes and AAV production. The AAV-panpromoter-flex-rev-Venus-2A-␥2F77 transgene was constructed from components of pAAV-CAG-promoter-Cre-2A-␥2F77, kindly provided by Zoltan Nusser, The data were recorded over a baseline period of 2 h. One-way ANOVA revealed no significant differences between the mouse types for Wake, NREM, or REM sleep.  (Klugmann et al., 2005). AAV was suspended in sterile PBS at 1:1 concentration. EEG recordings and sleep scoring. For nontethered EMG and EEG recordings, mice were fitted with Neurologger 2A devices (Anisimov et al., 2014). The recordings took place in home cages. Data were sampled at 200 Hz and downloaded offline and processed with Spike2 software (Cambridge Electronic Design, Spike2 Software, RRID:SCR_000903). The EEG was high-pass filtered (0.5 Hz corner frequency, Ϫ3 dB). The EMG was bandpass filtered (5-45 Hz, Ϫ3 db). Wake, non-REM, and REM were first automatically sleep-scored using previously described criteria (Costa-Miserachs et al., 2003). The EEG results were then manually verified.

Institute of Experimental
To calculate power spectra, segments of NREM identified after sleep scoring were concatenated and power spectra calculated using a Fast Fourier transform with a Hanning window function. Segments of data of at least 10 min were used. The power spectra were normalized as we described previously, such that the area under the saline controls for a given genotype was unity . Power changes were computed as differences in areas under the power spectra.
Behavioral experiments and drug administration. Mice were maintained on a 12 h light/12 h dark cycle with ad libitum food and water. All behavioral experiments took place during the "lights off" part of the cycle when the mice were most active. In all cases, we used paired comparisons where the animals served as their own controls, a withinanimals approach (crossover design). The experimenters were not blinded to treatment. Zolpidem (Tocris Biosciences) was dissolved in equimolar tartaric acid (BDH Chemicals) in 0.9% w/v saline.
Quantifying the spread of AAV transduction. For the HDC-␥2F77 mice, Venus-positive neurons were counted on fixed sections using ImageJ. Signal-emitting outliers (Ͻ15 m or Ͼ30 m diameter) were excluded, as were objects visually scored as incorrect (e.g., microglia). For the FC-␥2F77 and SC-␥2F77 mice, Cre-positive neurons were identified by immunohistochemistry, and the percentage area of the target region (FC or SC) was calculated, again using ImageJ (ImageJ, RRID:SCR_003070).
Brain-slice electrophysiology. We recorded spontaneous IPSCs from whole-cell, voltage-clamped, HDC neurons of the TMN and pyramidal neurons of the FC in acute slices. Brains were rapidly removed and immersed in ice-cold slicing ACSF (85 mM NaCl, 2.5 mM KCl, 1 mM CaCl 2 , 5 mM MgCl 2 , 1.25 mM NaH 2 PO 4 , 26 mM NaHCO 3 , 75 mM sucrose, 11 mM glucose, bubbled with 95% O 2 /5% CO 2 ). For the TMN, a tissue block was cut between the cerebellum and optic tract, and coronal sections were cut to a thickness of 250 m on a vibratome. For the FC, a tissue block was cut between the optic tract and ϳ1 mm behind the olfactory bulb. After slicing, the holding chamber was transferred to a 37°C heat block for 15 min, and slicing ACSF was gradually exchanged for recording ACSF (125 mM NaCl, 2.5 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 1.25 mM NaH 2 PO 4 , 26 mM NaHCO 3 , and 11 mM glucose, pH 7.4, bubbled with 95% O 2 /5% CO 2 ) over 40 min. Electrophysiological recordings were made at room temperature. We identified histaminergic neurons by the presence of hyperpolarization-activated currents (Ih), a transient outward current, and the spontaneous firing activity of the cells (Stevens et al., 2001). Virally transduced pyramidal neurons in the FC were found by Credependent expression of EGFP. Pyramidal neurons were identified primarily by morphology. For detection of IPSCs, we performed whole-cell recordings in voltage-clamp (Ϫ70 mV) using internal pipette solutions containing the following: 140 mM CsCl, 4 mM NaCl, 0.5 mM CaCl 2 , 10 mM HEPES, 5 mM EGTA, and 2 mM Mg-ATP; the pH was adjusted to 7.3 with CsOH.
Statistical analysis. For behavioral and EEG comparisons, and comparisons of IPSC decay times from the electrophysiology analysis, we used two-tailed paired t tests (Janusonis, 2009), and normality was assumed. Analyses were performed in OriginPro 8.6 or GraphPad Prism 4.03 (GraphPad Prism, RRID:SCR_002798).
In agreement with previous studies (Kopp et al., 2004;Alexandre et al., 2008), we found that the power of "NREM sleep" produced by zolpidem in C57BL/6J mice was lower than that found in natural NREM sleep. Zolpidem reduced power during NREM sleep (Fig. 3A) over the frequency range between 5 and 16 Hz (n ϭ 5, paired t test, t (4) ϭ 4.5, p ϭ 0.01). In ␥2I77 lox mice, zolpidem injection did not change the EEG power spectrum in either the waking (Fig. 3B) or NREM (Fig. 3C) states.

Potentiation of GABA inputs onto histaminergic neurons by zolpidem induces and maintains NREM sleep
Having demonstrated that ␥2I77 lox mice do not enter NREM sleep after systemic zolpidem administration, we next made several areas of the brains of ␥2I77 lox mice zolpidem-sensitive using the ␥2I77 to ␥2F77 subunit switch. The first target was histamine neurons. Previously, we generated and studied HDC-⌬␥2I77 mice, which are mice with a deletion of the ␥2I77 subunit from histaminergic neurons in the TMN, obtained by crossing HDC-Cre and ␥2I77 lox mice (Zecharia et al., 2012). The histaminergic neurons of these HDC-⌬␥2I77 mice lack IPSCs (Zecharia et al., 2012). Because the HDC-⌬␥2I77 mice still had Cre recombinase expressed in their histaminergic neurons, we could implement a "restorative genetics" strategy and put the ␥2F77 subunit back into the neurons from which the ␥2I77 version was deleted. We introduced the zolpidem-sensitive ␥2F77 subunit into the histaminergic neurons of HDC-⌬␥2I77 mice using a Cre recombinase flex switch-dependent transgene (Atasoy et al., 2008), flex-rev-4 Figure 2. EEG spectra and sleep scoring for zolpidem-induced sleep in C57BL/6 (␥2F77) mice compared with ␥2I77 lox mice. A, EEG power spectra for C57BL/6J mice injected with saline or 5 mg/kg zolpidem. The spectra are aligned in register with the /␦ power ratio, the root mean square electromyogram (RMS EMG), the primary EEG, and the sleep scoring assignments (Wake, NREM, REM). Arrow indicates the time of saline or zolpidem injection. B, EEG power spectra for ␥2I77 lox mice injected with saline or 5 mg/kg zolpidem. All alignments of traces are as above in A.  (Fig. 3A) over the frequency range between 5 Hz and 16 Hz (n ϭ 5, paired t test, t (4) ϭ 4.5, p ϭ 0.01). B, EEG power spectrum of ␥2I77 lox mice (n ϭ 6) in the waking state following saline intraperitoneal injection (black) or zolpidem (5 mg/kg i.p.; red). C, In ␥2I77 lox mice, zolpidem does not influence the power spectra during NREM sleep. Typical epochs of EEG trace are shown. Calibration: A, C, 200 V, 500 ms.
␥2F77-2A-Venus, packaged into AAV capsids (Fig. 4A). This AAV-flex-rev-␥2F77-2A-Venus was bilaterally injected into the ventral TMN area of adult HDC-⌬␥2I77 mice to generate HDC-␥2F77 mice (Fig. 4B). The flex switch in the AAV transgene ensured that ␥2F77 expression was restricted to Cre-positive neurons (Fig. 4A). These mice had bilateral expression of the ␥2F77-2A-Venus transgene in their TMN area, confined to HDC neurons in the ventral parts of the TMN (Fig. 4 B, C), in an ϳ700 m anterior-to-posterior segment (Fig. 4D). The mean number of AAV-transduced neurons in HDC-␥2F77 brains, as assessed by Venus expression, was compared with a count performed on a brain from an HDC-Cre x Rosa-YFP reporter mouse cross. We hypothalamus, detected by EGFP immunocytochemistry. E, Spontaneous IPSCs recorded from histaminergic neurons in acute brain slices prepared from the posterior hypothalamus of C57BL/6J (␥2F77), ␥2I77 lox , and HDC-␥2F77 mice before (black) and after 10 M zolpidem (red) application. Traces represent peak-normalized, superimposed average waveforms. For the C57BL/6J (␥2F77) data, 87 events were averaged before zolpidem and 61 events were averaged after zolpidem. For the ␥2I77 lox data, 124 events were averaged before zolpidem and 167 events were averaged after zolpidem. For the HDC-␥2F77 data, 59 events were averaged before zolpidem and 90 events were averaged after zolpidem. Graphs represent the mean weighted decay times before and after zolpidem application for the different groups of mice. **p ϭ 0.002 (paired t test).
found HDC-Cre x Rosa-YFP mice had ϳ8000 neurons in which the HDC promoter was active in the TMN area (data not shown). This could be an overestimate if some of the Rosa-YFP expression originates from the HDC-Cre gene turning on and off during development. In the adult HDC-␥2F77 mice, where Venus expression can only be seen if the HDC-Cre gene is active in the adult, ϳ7000 Venus-positive cells could be detected. This number will be an underestimate because it cannot be expected that all histaminergic neurons in HDC-⌬␥2I77 mice would be transduced by the AAV-flex-rev-␥2F77-2A-Venus virus (see below). On the other hand, counts of cells immunoreactive for histamine estimated that there were 2500 -3500 such neurons in the mouse hypothalamus (Parmentier et al., 2002), and 3800 in the rat determined by staining with histidine decarboxylase antibodies (Ericson et al., 1987). Given the difference in sensitivity between the genetic and primary immunoreactive detection methods, our estimate of histaminergic neuronal number is approximately in the same range. The extent of AAV transduction in HDC-␥2F77 mice was also ascertained by whole-cell voltage-clamp recordings in acute slices made from the posterior hypothalamus. In the ventral TMN region, we found that 10 of 16 neurons (62.5%) had restored IPSCs that resembled IPSCs recorded from C57BL/6J neurons (e.g., Fig. 4E). Of these 16 cells, the remaining 6 still had no IPSCs (data not shown). Presumably, these six neurons had not been transduced by AAV-flex-rev-␥2F77-2A-Venus virus and were still HDC-⌬␥2I77 knock-out cells (Zecharia et al., 2012). We confirmed that the HDC-␥2F77 histaminergic neurons with rescued IPSCs also had restored zolpidem sensitivity. Zolpidem (10 M) applied to C57BL/6 (␥2F77 ) histaminergic cells slowed the IPSC decay by 48 Ϯ 8%, from 30 Ϯ 3 ms (control) to 44 Ϯ 3 ms (zolpidem) (n ϭ 5 cells, paired t test, t (4) ϭ 7.5, p ϭ 0.002; Fig.  4E, left); by contrast, 10 M zolpidem applied to ␥2I77 lox histaminergic neurons had no effect on the IPSC decay, being ϳ23 Ϯ 4 ms (control) and 24 Ϯ 4 ms (zolpidem) (n ϭ 5 cells, paired t test, t (4) ϭ 0.7, p ϭ 0.27; Fig. 4E, middle). In contrast, 10 M zolpidem applied to HDC-␥2F77 TMN neurons slowed the IPSC decay by ϳ46 Ϯ 7% from 24 Ϯ 3 ms (control) to ϳ35 Ϯ 5 ms (zolpidem) (n ϭ 6 cells, paired t test, t (5) ϭ 5.7, p ϭ 0.002; Fig. 4E, right), the same magnitude of response obtained by applying zolpidem to C57BL/6J histaminergic neurons (Fig. 4E, left).

The FC also can contribute to zolpidem-induced sleep induction but not maintenance
In some circumstances, the FC could help initiate sleep (see Discussion). To test whether zolpidem might work partly through this route to induce sedation, we made the FC of ␥2I77 lox mice selectively zolpidem-sensitive by genetically swapping zolpidemsensitive ␥2F77 subunits into ␥2I77 lox frontal cortical neurons, so generating FC-␥2F77 mice (Fig. 6 A, B). For this, we coinjected bilaterally a mixture of two AAVs into the FC: AAV-Cre-2A-␥2F77 and AAV-flex-rev-EGFP. This produced cotransduced neurons. The swap works as follows. From the AAV-Cre-2A-␥2F77 transgene, the Cre recombinase destroys production of functional ␥2I77, and the zolpidem-sensitive ␥2F77 subunit replaces it; the second AAV expresses EGFP only if Cre recombinase is present, and thus marks neurons that have been transduced with AAV-Cre-2A-␥2F77 (Fig. 6A). We visualized the transduced area by serial sectioning and then immunocytochemistry with GFP antisera (Fig. 6C,D). In all injections, the spread of transduced cells reached rostral almost to the olfactory bulb and caudal as far as the primary motor cortex (Fig. 6D). AAV-Cre-2A-␥2F77 transgene expression was found in both the frontal motor cortex and the prefrontal cortical areas (including prelimbic cortex, orbital areas, primary and secondary motor cortices). Maximally 20% of any given cortical area was transduced, mostly restricted to the deep layers of cortex, V and VI (Fig. 6C,D). To confirm that the ␥2I77 to ␥2F77 swap had produced zolpidem sensitivity in pyramidal neurons of FC-␥2F77 mice, we made acute slices of M1 and M2 frontal neocortex and performed whole-cell voltage-clamp recordings on pyramidal neurons from layers 5 and 6. The neurons expressing the AAV-Cre-2A-␥2F77 transgene were identified by their primary EGFP signal. Zolpidem (1 M) slowed the decay of IPSCs by 63.5 Ϯ 14% in EGFPexpressing neurons, from 15.4 Ϯ 1.8 ms (control) to 24.5 Ϯ 2.2 ms (zolpidem) (n ϭ 6, paired t test, t (5) ϭ 6.4, p ϭ 0.001) (Fig. 6E,  right), which was similar to its effect on the IPSCs of C57BL/6 pyramidal neurons (n ϭ 6, paired t test, t (5) ϭ 8.8, p ϭ 0.0003; Fig.  6E, left); by contrast, the decay of IPSCs was unchanged by zolpidem (13.2 Ϯ 2.2 ms) in ␥2I77 lox neocortical pyramidal neurons compared with control solution (12.8 Ϯ 2.2 ms) (n ϭ 5, paired t test, t (4) ϭ 1, p ϭ 0.4; Fig. 6E, middle).
The baseline sleep-wake parameters of FC-␥2F77 mice did not differ from ␥2I77 lox or C57BL/6J mice (Table 1). We tested whether making FC neurons selectively zolpidem-sensitive was sufficient to produce a behavioral response in FC-␥2F77 mice (Fig. 7). We administered zolpidem (5 mg/kg) to FC-␥2F77 mice and found that this reduced latency to NREM sleep (Fig. 7A). FC-␥2F77 mice had a 50 Ϯ 14% shorter latency to NREM sleep (zolpidem: 13 Ϯ 4 min) compared with those injected with saline  (paired t test). B, Time spent in NREM sleep in the first 45 min after saline and zolpidem in HDC-␥2F77 mice. **p ϭ 0.002. B, Inset, Normalized EEG power spectra from HDC-␥2F77 mice following zolpidem (5 mg/kg i.p.; red) compared with their natural NREM sleep spectra (blue). Typical epochs of EEG trace are shown. C, EEG power spectra for HDC-␥2F77 mice injected with saline or 5 mg/kg zolpidem. The spectra are aligned in register with the /␦ power ratio, the root mean square electromyogram (RMS EMG), the primary EEG, and the sleep scoring assignments (Wake, NREM, REM). Arrow indicates the time of saline or zolpidem injection.

Discussion
We have previously suggested that sedatives produce sleep by interacting with the NREM sleep-inducing circuitry, changing activity in the hypothalamic and brainstem circuits that globally govern arousal (Nelson et al., 2002;Lu et al., 2008;Zhang et al., 2015). We show here that this seems to be the case for zolpidem, too. By using a pharmacogenetic method that probes endogenous GABA tone, we found that selectively augmenting the active GABA input onto hypothalamic histamine neurons by systemic zolpidem administration decreased NREM sleep latency and enhanced sleep time but without reducing power in the EEG. As well as revealing a potential site for zolpidem's sleeppromoting actions in vivo, our pharmacogenetic findings support the hypothesis that the initiation of natural NREM sleep could arise by increased and sustained inhibition onto histaminergic neurons (Nitz and Siegel, 1996;Sherin et al., 1996Sherin et al., , 1998. Clinical features of zolpidem mimicked in mice with brain regions selectively zolpidem-sensitive Positive GABA A receptor modulators are often good at inducing sleep (Lancel and Steiger, 1999;Winsky-Sommerer, 2009;Nutt and Stahl, 2010;Rye et al., 2012). Zolpidem's pharmacokinetics make it effective for treating insomnia: it maximally occupies its receptor sites minutes after entering the blood, causing sleep quickly, but its short plasma half-life limits "hangovers" (Benavides et al., 1988). In controlled clinical settings, zolpidem's main effect on people is to reduce sleep latency; but overall zolpidem performs no better than placebo in sleep maintenance, wake time after sleep onset, or number of awakenings (Greenblatt and Roth, 2012). By these measures, zolpidem's key clinical action, reduction of sleep latency, is mimicked by increasing inhibition onto histaminergic neurons. However, the NREM sleep induced by zolpidem in humans and wild-type rodents does not entirely resemble natural sleep because "zolpidem sleep" has diminished power in the EEG compared with natural NREM sleep for frequencies Ͼ5 Hz in rodents (Kopp et al., 2004;Alexandre et al., 2008), and most frequencies in humans (Landolt et al., 2000). It is not clear whether this diminished EEG power is a good or bad feature of zolpidem-induced sleep. But in the HDC-␥2F77 and FC-␥2F77 mice, the power of zolpidem-evoked NREM sleep was the same as natural NREM sleep, so these "power-decreasing" effects of zolpidem must originate in other brain areas. This knowledge may be useful for designing sedatives that produce a more natural sleep.

Zolpidem can induce NREM sleep by selectively inhibiting histaminergic neurons
Despite having a 20-fold higher affinity at ␣1␤␥2-containing GABA A receptors (Pritchett and Seeburg, 1990), which are the most widely expressed and abundant type of GABA A receptors in the brain (Pritchett et al., 1989;Wisden et al., 1992;McKernan and Whiting, 1996), zolpidem (5 mg/kg) induces sleep through the ␣2␤␥2 and/or ␣3␤␥2 GABA A receptors (Kopp et al., 2004). The ␣1-containing receptors are, instead, responsible for the decrease in EEG power across most frequencies Ͼ5 Hz in zolpidemevoked sleep (Kopp et al., 2004). We might also expect that zolpidem's effects, such as sleep, result from additive slowing of IPSCs on cell types with ␣2 and/or ␣3 subunits throughout the brain. But this is not the case. Prolonging IPSCs on just histaminergic neurons is enough to induce and maintain sleep, although not to the full extent generated by zolpidem in wild-type C57BL/6J mice. Zolpidem is probably effective at histamine neurons because of their hub-like nature and their ability to promote arousal and wakefulness (Haas and Panula, 2003). Although there are relatively few histamine neurons, between 3000 and 7000 in the mouse, their axons ascend and descend from the TMN, coursing throughout the brain, coreleasing histamine and GABA to give balanced arousal (Wada et al., 1991;Haas and Panula, 2003;. Thus, acutely inhibiting the "histamine hub" by zolpidem will cause histamine levels to fall throughout the brain and sleep to ensue. This fits with previous pharmacological data that infusing GABA agonists into the TMN area induces sleep (Lin et al., 1989;Nitz and Siegel, 1996), and that GABA/galanin neurons in the lateral preoptic neurons, which send axons to the TMN, increase their activity during sleep (Sherin et al., 1996(Sherin et al., , 1998. Histaminergic neurons principally express ␣1␤3␥2 and ␣2␤3␥2 GABA A receptors (Fritschy and Mohler, 1995;Sergeeva et al., 2002;Zecharia et al., 2009Zecharia et al., , 2012May et al., 2013). Thus, these ␣2-containing GABA A receptors on the histaminergic neurons are likely candidates for a part of zolpidem's sleep-inducing actions in vivo. The ␣1-containing GABA A receptors that cause zolpidem to reduce EEG power must be on other types of neurons elsewhere.

Zolpidem can initiate sleep top-down from the FC
We found that zolpidem can act in the frontal neocortex to reduce sleep latency, although the effect was not as large as for the histaminergic neurons, and sleep time was also not increased. Other data also link the frontal and preFC and behavioral sleep: sleep can initiate top-down if the FC is stimulated at 4 Hz (Penaloza-Rojas et al., 1964; Lineberry and Siegel, 1971); slow waves initiate in frontal neocortex (Massimini et al., 2004;Vyazovskiy et al., 2009); and in human aging, atrophy of the medial prefrontal cortex correlates with disrupted NREM slow waves (Mander et al., 2013).

Pharmacogenetic manipulation of GABA inputs versus receptor knock-outs, acute versus chronic
There are several caveats to consider when interpreting our results. The first point is that acute inhibition of one area in the brain could affect circuit dynamics in other areas (Otchy et al., 2015). The brain's dense interconnectivity could cloud, rather than reveal, the function of the inhibited region and so "transient circuit manipulations may have their own interpretive difficulties" (Otchy et al., 2015). This could indicate that zolpidem does not normally induce sleep by enhancing inhibition on histamine neurons but only does so in this particular artificial situation whereby the histamine neurons are made uniquely sensitive to zolpidem in the HDC-␥2F77 mice. The second point is that dif- Figure 8. EEG power spectra and sleep scoring for zolpidem-induced sleep in FC-␥2F77 mice. EEG power spectra for FC-␥2F77 mice injected with saline or 5 mg/kg zolpidem. The spectra are aligned in register with the /␦ power ratio, the root mean square electromyogram (RMS EMG), the primary EEG, and the sleep scoring assignments (Wake, NREM, REM). Arrow indicates the time of saline or zolpidem injection. ferent results are often produced by chronic or acute ablations (Wisden et al., 2009;Otchy et al., 2015). Genetic ablation of the GABA A receptor ␥2 subunit from histaminergic neurons did not affect normal sleep over a 24 h period, although it did produce the more subtle effect of preventing the mice settling down and going to sleep in a new environment; in other words, removing synaptic GABA A receptors from histaminergic neurons lengthened the latency to NREM sleep (Zecharia et al., 2012), and this fits with our new data that, going in the opposite direction, enhancing IPSCs with zolpidem on these neurons shortens the latency to NREM sleep. Nevertheless, it remains remarkable that fast GABA input to the histamine neurons is dispensable for controlling the basic sleep-wake cycle. Similar to our results on histamine neurons, we found that chronic ablation versus acute pharmacoge-netic modulation of GABA inputs on cerebellar Purkinje neurons also produced different results: mice with zolpidem-sensitive GABA A receptors selectively expressed in Purkinje neurons had acute ataxia after being given zolpidem, and so we concluded that ongoing GABA input onto Purkinje cells modulates motor control (Wulff et al., 2007;Wisden et al., 2009); by contrast, knocking out the ␥2 subunit selectively and permanently from Purkinje cells, and the consequent removal of fast synaptic responses to GABA, did not produce overt ataxia, but only a subtle deficit in limb coordination (Vinueza Veloz et al., 2015). To explain the large difference in behavioral phenotype produced by acute zolpidem modulation of Purkinje cells and chronic ablation of fast inhibitory input, we hypothesized that the cerebellar circuitry with chronically removed synaptic GABA input on Purkinje cells had undergone adaptation (Wulff et al., 2007;Wisden et al., 2009). We think the weak phenotypes produced by ␥2 subunit ablation from Purkinje cells and histaminergic cells, and the contrasting strong phenotypes obtained by acute manipulation with zolpidem are analogous: the pharmacogenetic "zolpidem method" unmasks the acute role for GABA in modulating histaminergic neurons, whereas HDC-⌬␥2 mice have undergone compensatory changes. Acute zolpidem manipulation in HDC-␥2F77 mice produces the "true" result.
In conclusion, zolpidem has rather subtle effects on synaptic IPSCs. Typically, it prolongs them by ϳ50%. We might have expected that zolpidem induces sleep by potentiating IPSCs everywhere in the brain; the net effect would be behavioral sleep. But instead we have shown that zolpidem can induce sleep by strengthening GABA signaling on just one cell type (histamine neurons). Normally, the NREM sleep induced by zolpidem does not resemble natural sleep; the drug produces a lower power in most frequencies of the EEG during NREM sleep. But via histamine neurons, zolpidem can induce sleep without reducing the EEG power of the sleep. This knowledge could help design drugs that induce a more natural sleep.