Galanin neurons in the ventrolateral preoptic area promote sleep and heat loss in mice

The preoptic area (POA) is necessary for sleep, but the fundamental POA circuits have remained elusive. Previous studies showed that galanin (GAL)- and GABA-producing neurons in the ventrolateral preoptic nucleus (VLPO) express cFos after periods of increased sleep and innervate key wake-promoting regions. Although lesions in this region can produce insomnia, high frequency photostimulation of the POAGAL neurons was shown to paradoxically cause waking, not sleep. Here we report that photostimulation of VLPOGAL neurons in mice promotes sleep with low frequency stimulation (1–4 Hz), but causes conduction block and waking at frequencies above 8 Hz. Further, optogenetic inhibition reduces sleep. Chemogenetic activation of VLPOGAL neurons confirms the increase in sleep, and also reduces body temperature. In addition, chemogenetic activation of VLPOGAL neurons induces short-latency sleep in an animal model of insomnia. Collectively, these findings establish a causal role of VLPOGAL neurons in both sleep induction and heat loss.

A lmost a century ago, von Economo discovered that patients with injury to the rostral hypothalamus often had unrelenting insomnia 1 . Subsequent animal research showed that lesions of the preoptic area (POA) reduced sleep [2][3][4] whereas stimulation increased sleep [5][6][7][8] . However, the specific circuits constituting this preoptic sleep-inducing area remained unknown. Sherin et al. identified a cluster of neurons in the ventrolateral preoptic nucleus (VLPO) and scattered neurons in the adjacent ventromedial and dorsolateral preoptic area (the extended VLPO) of rats that express cFos during sleep and project to the wake-promoting tuberomammillary nucleus (TMN) 9,10 . About 85% of these neurons contain the inhibitory neurotransmitters galanin (GAL) and GABA 9,10 ; similar observations in other species confirmed that most sleep-active neurons in the VLPO and extended VLPO are galaninergic and form a single population, referred to here as the VLPO GAL neurons 11 . Lesions in this region in rodents reduce sleep by 40-50% [12][13][14] . In humans, loss-of-VLPO GAL neurons is associated with sleep fragmentation and fewer bouts of consolidated sleep 15 .
Although GAL is a specific marker for sleep-active neurons in the VLPO and extended VLPO in rats, in mice about 20% of all VLPO GAL cells are active during wake 11 . In addition, a recent optogenetic study found that photostimulation of POA GAL neurons at 10 Hz failed to increase sleep and surprisingly increased wake 16 . These results call into question whether POA GAL neurons promote sleep.
In addition to sleep regulation, the POA is also a key site for thermoregulation [17][18][19] .
Large preoptic lesions cause hyperthermia 3,20 , whereas photoactivation of specific POA neurons causes profound hypothermia (4-6°C) [21][22][23] . Interestingly, many sleep-active VLPO neurons are also activated by increased skin or body temperature (T b ), and may drive reductions in T b , demonstrating the tight coupling of sleep and heat loss pathways 24,25 . However, it is unknown whether VLPO GAL neurons contribute to thermoregulation.
To determine the specific role of VLPO GAL neurons in sleep and thermoregulation, we selectively activated and inhibited VLPO GAL neurons in mice. We find that activation of VLPO GAL neurons promotes sleep even in an animal model of insomnia. Conversely, inhibition of VLPO GAL neurons reduces NREM sleep and increases wake. Interestingly, chemoactivation of VLPO GAL neurons also causes profound hypothermia, indicating that VLPO GAL neurons can also promote heat loss.

Results
Distribution of GAL neurons in the POA. We first investigated the distribution of GAL neurons in the POA of GAL-IRES-Cre:: L10-GFP mice (Methods). In rats and humans, POA GAL neurons are predominantly found in three major clusters, including the VLPO and extended VLPO; the medial preoptic nucleus; and the supraoptic nucleus 11 . We found that in mice, a larger proportion of the VLPO GAL neurons is located in the dorsal extended VLPO. The cluster of GAL neurons in the medial preoptic nucleus is less dense than in rats, but there are many more small (15-20 µm) GAL-expressing neurons that extend along the wall of the third ventricle and into the median preoptic nucleus ( Supplementary Fig. 1). There are very few GALexpressing magnocellular supraoptic neurons at the levels of the POA studied here. Further, the entire POA in a mouse is so small that these cell groups, which are quite distinct in larger species, appear to be crowded together, with a tendency to overlap at the edges. Thus, identifying GAL-expressing cell groups involved in a specific function requires careful histological examination.
VLPO GAL neurons innervate wake and thermoregulatory centers. To determine whether VLPO GAL neurons influence brain regions that regulate sleep-wake behavior and T b , we unilaterally microinjected the VLPO of GAL-IRES-cre mice (n = 8) with a Cre-dependent adeno-associated viral vector (AAV) coding for channelrhodospsin-2 (ChR2) and mCherry (AAV8-EF1α-DIO-ChR2-mCherry, hereafter 'AAV-ChR2') [26][27][28] . Six weeks later, we immunolabeled the brain sections for mCherry. We analyzed the projections of 3 cases in which the bulk of the injection involved the three VLPO GAL populations, with little or no involvement of the periventricular, median, or medial preoptic nuclei. Similar to the pattern in rats 10 , we found a high density of mCherry-labeled terminals in wake-promoting regions, including the lateral hypothalamic area (LH), TMN, pedunculopontine tegmental nucleus (PPT), lateral and medial parabrachial nucleus (PB) and locus coeruleus (LC). We also observed intense mCherry-labeled terminal fields in the ventrolateral and lateral periaqueductal gray matter (lPAG) and the adjacent lateral pontine tegmentum, a key region for REM sleep regulation 29 . In addition, labeled axons were dense in the dorsal hypothalamic area (DHA) and dorsomedial hypothalamic nucleus (DMH), the raphe pallidus (RPa), and parapyramidal region (PPR) in the rostral medulla, regions essential for thermoregulation 17,18 . These results indicate that VLPO GAL neurons have projections that may allow them to influence many brain regions that regulate both sleep-wake behavior and T b .
In vitro, whole cell, current clamp recordings from ChR2expressing GAL neurons in POA slices indicated that 10 ms blue light pulses at stimulation frequencies ≤2 Hz evoked single action potentials and entrained VLPO neurons in a temporally precise manner with each light pulse evoking a full action potential (crossing 0 mV level) for the entire duration of a 60 s stimulation train (Fig. 2a, b). Similarly, in vivo, low frequency photostimulation (1 Hz, 10 ms duration) for 2 h increased cFos expression in VLPO and extended VLPO ChR2-expressing neurons. Identical stimulation in VLPO GAL -mCherry mice did not increase cFos expression (Fig. 1c).
During spontaneous sleep, VLPO neurons typically fire at only 2-5 Hz 25,32 , and we found that VLPO GAL neurons cannot entrain to higher frequency photostimulation in vitro. At stimulation frequencies >2 Hz, the peak of the photo-evoked excitatory response, fell below 0 mV, at which point it likely failed to elicit action potentials. While 75% of 4 Hz stimulations triggered action potentials, at 8 and 16 Hz, only the first light pulses evoked an action potential while the following light flashes evoked only excitatory potentials, indicating depolarization block (Fig. 2a, b).
To determine whether photoactivation affects sleep-wake behavior and T b , we photostimulated VLPO GAL neurons continuously for 2 h (from 21:00 h to 23:00 h; 2 h after dark onset) and recorded sleep-wake parameters and T b for 2 h before, during and after stimulation. Consistent with our hypothesis, photostimulation at 0.5, 1, 2, and 4 Hz substantially (>60%) increased NREM sleep when compared to sham stimulation in the same mice as well as photostimulation at the same frequencies in VLPO GAL -mCherry mice (Fig. 2c and Supplementary Table 1), but these stimulations did not affect REM sleep ( Supplementary  Fig. 2k). Behaviorally, mice with low frequency stimulation (0.5-4 Hz) reduced locomotor activity, assumed a sleep posture within their nest and transitioned into sleep. To determine whether the increase in NREM sleep depended on the duration of light pulses, we photostimulated VLPO GAL neurons at 1 Hz and varied the duration of light pulses (5, 10, 20, and 50 ms). We found that 10 ms pulses increased NREM sleep slightly more efficiently than either shorter or longer pulses ( Supplementary Fig. 2h).
AAV injections in VLPO GAL -ChR2 mice included GAL neurons in the VLPO (including the dorsal and medial extended VLPO), but also in the medial, median, and periventricular preoptic nuclei in most cases (Fig. 1b). However, a comparison of the different injection sites shows that the mice with the greatest increases in NREM sleep during stimulation (e.g., cases 832, 833, and 888) had injection sites involving the largest part of the VLPO core and the medial and dorsal extended VLPO, whereas mice with lesser increases in sleep (e.g., 830, 1141) typically had less extensive involvement of the VLPO and extended VLPO. In contrast, some mice with large increases in sleep had minimal involvement of the medial, median, or periventricular preoptic GAL neurons (e.g. 832, 888). In addition, in all of these experiments, the optical fiber tips were laterally placed in the dorsal extended VLPO, about 200-400 µm dorsal to the VLPO cluster ( Supplementary Fig. 2j). As in our previous work 23  We then analyzed sleep architecture and EEG power spectra in mice with 1 Hz stimulations with 10 ms pulse duration, as these parameters increased NREM sleep more efficiently. We found that the photo-evoked increase in NREM sleep was primarily due to more NREM sleep bouts rather than longer bouts (Supplementary Table 1), suggesting a role for VLPO GAL neurons in facilitating wake-to-NREM transitions. Interestingly, the EEG showed high voltage slow waves entrained to the frequency of stimulation, or its harmonics (i.e., 1, 2, and 4 Hz; Supplementary  Fig. 3) during wake, NREM sleep and REM sleep, resulting in an apparent increase in delta power (Fig. 2d, e and Supplementary Table 1). Identical light pulses in VLPO GAL -mCherry mice did not affect the EEG spectrum (Supplementary Table 1). The mechanism for entraining cortical EEG to VLPO firing may involve rhythmic inhibition of wake-promoting neurons.
Although, we observed slow waves entrained to the photostimulation frequency during wake and REM sleep (Fig. 2d), the different states were still clearly distinguishable both electrophysiologically and on video. Moreover, the increase in NREM sleep during photostimulation of VLPO GAL neurons was followed by a period of 'rebound-wake', i.e. an increase in wake across the 2 h after stimulation (Supplementary Fig. 2a-g), suggesting that the induced sleep fulfilled the homeostatic qualities of natural sleep. Collectively, these observations indicate that VLPO GAL neurons promote NREM sleep mainly by initiating sleep and increasing transitions from wake into NREM sleep.
Importantly, and in contrast to low frequency stimulation, higher frequency photostimulation (8 and 16 Hz) of VLPO GAL neurons did not increase NREM sleep (Fig. 2c). Our in vitro recordings showed depolarization block with high frequency stimulations, and in vivo, 16 Hz photostimulation actually increased wakefulness and exploratory activity similar to a prior study 16 in 7 out of 10 mice (three mice which did not display wake after 8 or 16 Hz stimulations had comparatively smaller preoptic injections). Thus, high frequency photostimulation of VLPO GAL neurons may induce depolarization block, disinhibiting the arousal systems and paradoxically increasing wakefulness.

Photoinhibition of VLPO GAL neurons decreases NREM sleep.
To study whether VLPO GAL neurons are necessary for sleep, we injected AAV-CAG-Flex-ArchT-GFP ('AAV-ArchT') 33 bilaterally into the VLPO of GAL-IRES-Cre mice ('VLPO GAL -ArchT mice') as above (Fig. 3a). AAV-ArchT resulted in specific expression of ArchT-GFP in VLPO GAL neurons, consistent with the distribution of GAL neurons in GAL-IRES-Cre::L10-GFP reporter mice (Supplementary Fig. 1). In the 7 mice used for the experiments, the AAV injections covered an average of 47.2% (range: 37.8-68.5%) of the VLPO. In vitro, whole-cell current clamp recordings from ArchTexpressing GAL neurons in VLPO slices showed that yellow/ orange laser light for 5 min hyperpolarized VLPO GAL neurons and completely prevented action potential firing during the illumination period (Fig. 3b). After the photoinhibition, action potentials resumed at the pre-illumination rate in all neurons (Fig. 3b).
We then measured the sleep-wake effects of photoinhibition of VLPO GAL neurons (5 min continuous light every 30 min for 24 h) (Fig. 3c, d and Supplementary Fig. 4a-d). After~2 min latency, the percentage time spent in wake gradually climbed from about 47% to about 60% at the end of the 5 min photoinhibition. Wake levels remained in that range for about 5 more min, before gradually returning to baseline. We compared the percentages of sleep-wake states during the 5 min periods immediately prior ('pre'), during ('during'), and after ('post') laser inhibition or sham inhibition. Over the entire 24-h period, the time spent in NREM sleep was decreased during the 5 min post laser inhibition when compared to the pre-laser inhibition (14.8 ± 2% reduction;  Table 2). These changes in NREM sleep and wake were mainly due to longer wake bouts (Supplementary Table 2), suggesting that photoinhibition reduced initiation of NREM sleep. REM sleep levels did not differ (Fig. 3c). In contrast to photoinhibition, sham inhibition had no effect on wake, NREM or REM sleep in the same mice ( Fig. 3d; Supplementary  Fig. 4b,d). These data indicate that acute inhibition of VLPO GAL neurons increases wake by reducing transitions into NREM sleep.

Chemogenetic activation induces sleep and hypothermia.
Although optogenetics offers many advantages, photostimulation may induce firing patterns that differ from natural patterns. We therefore sought to corroborate our optogenetic findings with chemogenetic experiments, which are also more suitable for longterm stimulation. We bilaterally injected AAV8-hsyn-DIO-hM3Dq-mCherry ('AAV-hM3Dq') 34,35 into the VLPO of GAL-Cre mice ('VLPO GAL -hM3Dq mice'; Fig. 4a) and implanted a telemetry transmitter to record EEG, EMG, T b , and locomotion 35 . AAV-hM3Dq injections resulted in specific expression of hM3Dq (as evidenced by mCherry labeling) in POA GAL neurons (Fig. 4b, c). In some mice, the injections were sufficiently laterally placed that they involved neurons only in the VLPO and extended VLPO, avoiding the medial, median, and periventricular preoptic nuclei. We therefore examined the responses in that group separately, to see whether they differed from the cases where the more medial, small celled preoptic GAL-expressing neurons were also involved. Injection of AAV-hM3Dq in WT littermates (not expressing Cre) resulted in no mCherry expression. We also used GAL-IRES-Cre mice injected with AAV-DIO-mCherry in the VLPO (VLPO GAL -mCherry mice) as negative controls.
To assess changes in sleep-wake behavior and T b with chemoactivation of VLPO GAL neurons, we injected CNO (0.3 mg/kg, i.p.) or saline (vehicle) at the onset of the dark period and early in the light period. Administration of CNO at dark onset (19:00 h) increased NREM sleep 71% over the next 4 h when compared to saline, while CNO injection at 10:00 h increased NREM sleep 21% during the first 12 h (three bins of 4 h each) (Fig. 5a, b and Supplementary Fig. 5). This increase in NREM sleep was due to a substantial increase in the number (but not the duration) of NREM sleep bouts, further demonstrating that VLPO GAL neurons promote NREM sleep by increasing wake-to- +0.14 mm  Fig. 5e-h,). Unlike photostimulation, which entrained the EEG to the frequency of the laser bursts, CNO did not alter the EEG spectrum during NREM sleep ( Supplementary Fig. 5c,d). In contrast to the increase in NREM sleep, REM sleep was suppressed by 66% and 71% during the first 8 h post CNO in the dark and light periods, respectively, and this REM sleep suppression was accompanied by longer latencies to enter REM sleep (Fig. 5c, d). None of these sleep-wake changes were observed in control GAL-IRES-Cre mice that received an AAV encoding only mCherry ( Supplementary Fig. 6).
In addition to sleep-wake changes, chemoactivation of hM3Dq-expressing VLPO GAL neurons markedly reduced T b during both the light and dark periods. T b began to drop 5-10 min after CNO injections, fell~4-6°C within~2 h, and then returned to baseline levels after 12-14 h (Fig. 5e, f). As this hypothermia was not observed with photostimulation which predominantly activated GAL neurons in the lateral POA (optical fiber tips were mainly dorsal to the VLPO cluster; Supplementary  Fig. 2j), we examined whether differences in the injection sites in optogenetic and chemogenetic experiments could contribute to this hypothermic response. Examination of the injection sites that caused the greatest fall in temperature showed that many of them were confined to the core and dorsal extended VLPO, with little or no involvement of the medial, median, or periventricular preoptic GAL neurons (e.g., cases 757, 808). As expected, these cases were also among those showing the greatest increase in  (Fig. 5g, i). Conversely, the percentage change in REM sleep negatively correlated with the magnitude of hypothermia (Fig. 5h,  j). Although we did not test whether the same individual neurons cause both the fall in T b and changes in sleep, it appears that both responses are caused by activation of a single population of VLPO GAL neurons.
We were also concerned, in light of recent reports that CNO may be converted to clozapine in vivo 36 , whether CNO injection could cause hypothermia by itself, but it failed to do so in GAL-IRES-Cre mice injected with AAV-mCherry ( Supplementary  Fig. 6j,l). It is therefore more likely that the profound hypothermia is due to differences in firing rate/pattern of VLPO GAL neurons with chemoactivation (vs. photoactivation) that caused different downstream responses such as release of other neurotransmitters/peptides.
Chemoactivation of VLPO GAL neurons at different ambient temperatures. Sleep and thermoregulation are closely intertwined in many ways [37][38][39][40] . For example, T b normally falls about 1°C during sleep in mice 41 . On the other hand, severe hypothermia may promote arousal, fragment NREM sleep, reduce EEG delta power during NREM sleep, and suppress REM sleep [37][38][39]42 . We therefore hypothesized that chemoactivation of VLPO GAL neurons would more strongly increase the quantity and quality of sleep (more consolidated sleep bouts, increased EEG delta power during NREM, and more REM sleep) if the mice were provided with 'thermal comfort' by exposure to an ambient temperature (T a ) that would blunt the fall in T b . To test this hypothesis, we chemogenetically activated VLPO GAL neurons in mice housed at thermoneutral (29°C) or warm T a (36°C, empirically determined to prevent the fall in T b , Fig. 6a).
As predicted, CNO injections produced larger increases in NREM sleep at both 29°C and 36°C compared to 22°C or injections of saline ( Fig. 6c; Supplementary Fig. 7). Surprisingly, a T a of 29°C, which is generally thought to be near thermoneutral for mice, did not prevent the hypothermia due to CNO, and only at a T a of 36°C was the CNO-induced hypothermia completely prevented (Fig. 6b). The amount of NREM sleep after CNO at T a 36°C was substantially greater and the NREM sleep latency was shorter than with CNO injections at T a 22°C (Fig. 6c, d) while the mice at T a of 29°C showed intermediate responses. CNO also increased NREM EEG delta power at T a 36°C (Fig. 6e) when compared to CNO at T a 29°C, suggesting deeper sleep at higher T a .
In contrast, CNO given at dark onset suppressed REM sleep at all T a (Fig. 6f). Whether this was due to activation of VLPO GAL neurons is difficult to determine as REM sleep is generally low during the dark period, and high T a alone can suppress REM sleep 43 . To address these issues, we administered CNO or saline to VLPO GAL -hM3Dq mice at 29°C during the light period when REM sleep is more abundant. We found that CNO decreased REM sleep during the first 4 hours at 29°C, but REM sleep rose quickly during the next 12 h until it was fully recovered. Cumulative REM sleep amounts in the 16 h post CNO were higher than at T a 22°C and were comparable to post-saline data at T a 22°C or T a 29°C (Fig. 6g). These results indicate that a T a of 29°C, which does not prevent hypothermia or initial REM sleep loss, can rescue REM sleep amounts by enhancing REM sleep rebound after hypothermia.
Activation of POA GAL neurons attenuates sleep-onset insomnia. To explore the translational relevance of our findings for insomnia patients, we examined whether chemoactivation of VLPO GAL neurons can improve sleep in an animal model of insomnia. Acute exposure to a novel environment (e.g. transfer to the new cage) reduces sleep in mice for 3-4 h, akin to sleep-onset insomnia in humans 44 . Therefore, we administered saline or CNO to VLPO GAL -hM3Dq mice at 10:00 h (3 h into the light period; at 22°C) and immediately transferred them to a new cage (similar to home cage) with fresh bedding and nesting material (Fig. 7a). CNO injections increased NREM sleep about threefold during the first 4 h compared to saline treatment ( Fig. 7b and Supplementary Fig. 8c) and also significantly shortened NREM sleep latencies (Fig. 7c and Supplementary Fig. 8d). In addition, the hyperthermia usually associated with exposure to a new cage was attenuated by CNO ( Fig. 7d and Supplementary Fig. 8g). On the other hand, the hypothermia observed after CNO treatment in the home cage was still present, albeit with a lower magnitude. These findings indicate that activation of VLPO GAL neurons can attenuate sleep-onset insomnia and hyperthermia induced by the stress of a novel environment.

Discussion
Although, the role of the POA in sleep has been recognized for decades, the neurochemical identity of sleep-promoting neurons has remained controversial. Based on cFos expression and projections to the TMN 9-11 , GAL neurons in the VLPO (including the extended VLPO) have been suspected to play a critical role in the regulation of NREM sleep. Although lesions of the VLPO in rats dramatically reduced sleep time [12][13][14] , and loss-of-VLPO GAL neurons in humans with age is associated with fragmentation and loss-of-consolidated sleep 15 , it remained unclear if VLPO GAL neurons actually promote sleep 16 . We find that selective At first glance, our results may appear to contradict a recent report by Chung et al. that 10 Hz optogenetic activation of VLPO GAL neurons causes wakefulness, rather than sleep 16 . However, VLPO sleep-active neurons generally fire no faster than 2-5 Hz, even during deep NREM sleep 16,25,32 . We find that VLPO GAL neurons fire reliably with low frequency photostimulation (0.5-4 Hz) in vitro, and in vivo photostimulation at these frequencies clearly increases NREM sleep. In contrast, higher frequency stimulation (8 or 16 Hz) causes depolarization block in vitro, and in vivo these frequencies either decrease or have no effect on NREM sleep. We suspect that the wakefulness with 10 Hz photostimulation in the prior study 16 was caused by conduction block and acute silencing of VLPO GAL neurons.
Interestingly, Chung et al. also photostimulated POA neurons at 10 Hz that had transported ChR2-AAV retrogradely from the TMN, and found that they promote sleep 16 . Most of these are   likely to have been VLPO GAL neurons, as about 85% of preoptic neurons retrogradely labeled from the TMN in rats contain GAL 10 . However, the remaining 15% leaves room for additional neurons projecting to the TMN that do not express GAL. Consistent with this idea, most VLPO neurons projecting to the TMN fired in the range of 2-6 Hz during sleep and 1 Hz during wake 16 .
However, other TMN-projecting POA neurons fired at rates of up to 12 Hz, especially during REM sleep. These faster firing neurons may produce neurotransmitters other than GAL and may be activated when photostimulated at 10 Hz, but they would not have been included in our study, which was limited to VLPO GAL neurons.
As VLPO GAL neurons contain GABA and other neuropeptides besides GAL 10 , it is possible that sleep and T b changes induced by activation of VLPO GAL neurons may be mediated by neurotransmitters or neuropeptides other than GAL. General photoactivation of POA GABAergic neurons produced only a minor increase in sleep 45 or even wakefulness 16 , although there are other GABAergic neurons in the VLPO region that do not contain GAL 16,46 . On the other hand, photoactivation of only GABAergic POA neurons that project to the TMN (85% of which contain GAL) produced sleep 16 . These experiments suggest that there are several populations of GABAergic neurons in and around the POA (e.g., the basal forebrain GABAergic arousal neurons are just lateral to the VLPO core), some of which may promote sleep while others promote wake 25,32 . Further investigation is needed to identify the specific neurotransmitter in GAL neurons involved in these functions.
An important question is whether the VLPO GAL neurons help to initiate or to maintain NREM sleep. Neurotoxic lesions of the VLPO severely shortened NREM sleep bouts 12,13 , but animals transitioned into NREM sleep at a normal frequency 13 . This was interpreted as the VLPO being critical for NREM sleep maintenance rather than initiation. However, in those chronic experiments, animals likely had an abnormally high sleep drive due to accumulated sleep loss, and this high sleep pressure may have increased transitions into NREM sleep. We found that VLPO GAL photoactivation and chemoactivation increased transitions into NREM sleep but did not lengthen NREM sleep bouts. Similarly, chemoactivation reduced NREM sleep latency in an insomnia model where the drive for wake in mice is extremely high. These findings suggest a critical role for VLPO GAL neurons in the initiation of NREM sleep.
Photostimulation of VLPO GAL neurons did not alter REM sleep whereas chemoactivation strongly suppressed this state. While photostimulation drives neuronal activity rhythmically, chemoactivation likely makes neurons more responsive to natural inputs, and this could drive different downstream responses that impact REM sleep. Alternatively, hypothermia and thermal stress during chemogenetic activation of VLPO GAL neurons at room temperature may also have suppressed REM sleep because (a) REM sleep loss with chemoactivation of VLPO GAL neurons positively correlated with the degree of hypothermia, and (b) REM sleep was recovered with exposure to a thermoneutral T a (29°C). This is unsurprising as changes in T a and T b disrupt REM sleep much more than NREM sleep 37,39,43,[47][48][49][50] and REM sleep is maximal at a thermoneutral T a (27-32°C in mice) and decreases with T a above or below these levels [51][52][53] .
The severity of hypothermia with chemoactivation of VLPO-GAL neurons demonstrates an important role for these neurons in thermoregulation in addition to sleep-wake regulation. In our previous work in rats 10 , we found that VLPO neurons project to the raphe pallidus nucleus, an area that mediates hyperthermic responses, and that most of the neurons contributing to this projection are in the region of the dorsal extended VLPO 20 .
McAllen and colleagues found that neurons in this region, which they called the caudolateral POA to distinguish it from the median preoptic nucleus (their "rostromedial" POA), promote heat loss via tail artery vasodilation 54,55 . Currently, it is unknown whether the neurons contributing to this pathway are specifically galaninergic, but our results suggest that at least some of them are. NREM sleep is usually associated with a small decrease in T b (~1°C) and wakefulness is associated with an increase. In contrast, higher T b can promote NREM sleep (e.g. low grade fever) and hypothermia causes arousal [56][57][58][59] . However, in our chemogenetic experiments, we observed a substantial increase in NREM sleep despite dramatic hypothermia, indicating that both responses were consequences of VLPO GAL activation and that increased NREM sleep was not secondary to the hypothermia. The additional increase in NREM amounts and decrease in sleep latency observed when activating VLPO GAL neurons in a warm environment suggests that hypothermia (and the associated thermal discomfort) caused some arousal effects at room temperature.
Thermal and sleep changes induced by VLPO GAL activation highlight the strong link between these two functions as most sleep-active POA cells in rats are also warm-sensitive and their activation may drive thermoregulatory responses 24,25,51,57 . Local preoptic warming (which activates warm-sensitive neurons) can promote NREM sleep and reduce T b , whereas local cooling does the opposite 8,51,56 . Moreover, chemoactivation of POA neurons that expressed cFos during recovery sleep, resulted in similarly profound hypothermia 60 suggesting that sleep and thermoregulatory mechanisms are strongly coupled in the same neurons in the POA. Our present results indicate that VLPO GAL neurons play a key role in this coupling.
The anatomical projections of VLPO GAL neurons suggest pathways through which they can promote NREM sleep and heat loss. Conditional tracing showed that VLPO GAL neurons in mice project to targets similar to VLPO neurons in rats 10 . Both densely innervate wake-promoting regions, including the LH, TMN, and LC and moderately innervate the lateral PB and the PPT. Thus, NREM sleep-promotion by VLPO GAL neurons may be mediated by simultaneous inhibition of multiple-wake promoting regions. On the other hand, VLPO GAL stimulation may cause hypothermia through projections to the DHA, DMH, lateral PAG, RPa and PPR, which are considered critical nodes for heat production and retention. Inhibition of these structures by VLPO GAL neurons may reduce T b by suppressing thermogenesis and promoting vasodilation.
Collectively, our optogenetic studies demonstrate that activation of VLPO GAL neurons within their typical physiological firing range causes sleep. Additionally, chemoactivation of VLPO GAL neurons also produced a drastic fall in T b. However, the relative degree to which VLPO GAL neurons contribute to sleep vs. thermoregulation may depend on their firing patterns, projections and environmental conditions. Finally, the deep hypothermia seen with chemoactivation of VLPO GAL neurons, which is far beyond that seen with normal wake-sleep cycles, suggests a role in torpor. Torpor is a deeply hypothermic and unresponsive state, with a slow wave EEG, seen in mammals (from mice to bears) when food is too scarce for homeothermy 61,62 . Torpor permits low metabolism, which can sustain life when food is unavailable. The relationship between torpor and sleep has been debated 63-65 , but our observations suggest that prolonged activation of VLPO neurons, as with chemoactivation, may push VLPO GAL neurons into a firing state that fosters torpor. Thus, rather than just serving as a wake-sleep switch, the VLPO GAL neurons may serve as a switch with three positions: wake, sleep, and torpor.

Methods
Mice. We used male GAL-IRES-cre transgenic mice on a C57BL/6J background (generated by Dr. Münzberg) 28,66 and their wild-type (WT) littermates (aged 8-12 weeks at the start of the experiments). Eutopic expression of Cre was verified by immunostaining for the GAL peptide 28 and by crossing GAL-Cre mice with a reporter line expressing green fluorescent protein (GFP) and comparing the expression pattern to that of Gal mRNA in the Allen Brain Atlas 28,66 . Mice were maintained under standard vivarium conditions (12 h: 12 h light-dark cycle with lights on at 07:00 h; 22 ± 1°C ambient temperature). Care of the animals met National Institutes of Health standards, as set forth in the Guide for the Care and Use of Laboratory Animals, and all protocols were approved by the BIDMC Institutional Animal Care and Use Committee. Viral vectors. We obtained Cre-dependent adeno-associated viral vectors for optogenetic stimulation (AAV8-EF1α-DIO-ChR2(H134R)-mCherry), optogenetic inhibition (AAV8-CAG-Flex-ArchT-GFP), chemogenetic stimulation (AAV8-hSyn-DIO-hM3Dq-mCherry) as well as control viral vectors (AAV8-EF1α-DIO-mCherry) from the Vector Core at the University of North Carolina, USA. All viral vectors had titer concentrations of 3-6 × 10 12 vector genomes per milliliter. The specificity of these viral vectors has been confirmed in various Cre lines including GAL-IRES-Cre mice used in this study 35,66 . In our hands, we observed that 88.6% of ChR2-expressing neurons were positive for Gal mRNA (by in situ hybridization using RNAScope), as well as 88.5% of ArchT-expressing neurons and 90.0% of hM3Dq-expressing neurons.
Optogenetic stimulation and inhibition. Mice were connected to the EEG/EMG wires and optical patch cords >2 weeks prior to recordings to facilitate proper habituation and natural sleep patterns. For photoactivation experiments, we used a function generator (PCGU1000, Velleman Instruments, USA) to stimulate both, VLPO GAL -ChR2 mice (n = 10) and VLPO GAL -mCherry mice (n = 6) with 0.5, 1, 2, 4, 8, and 16 Hz light pulses (10 ms per pulse, 10 mW light output at the fiber tip; the order of stimulation frequencies was randomized) for 2 h (21:00-23:00 h) and recorded EEG/EMG, video, body temperature, and locomotion data for 2 h prior to, during and 2 h after stimulation time. For photoinhibition experiments, we applied continuous pulses of orange/yellow light for 5 min every 30 min for 24 h and recorded EEG/EMG signals and video. In the sham inhibition condition the laser was not switched on, but the function generator still produced 5 min pulses every 30 min. EEG and EMG signals were amplified (A.M systems, USA), digitized and recorded using vital recorder software (Kissei Comtek, Japan) 13 . T b and LMA were recorded telemetrically using Dataquest A.R.T 4.1. software (Data Sciences International., USA) 31,35 . The order of stimulation frequencies for the photoactivation experiments was counterbalanced between animals and >3 days of rest were allowed between stimulation sessions to avoid habituation to induced sleep patterns. To prevent light leakage into the cage, we used shielded optical patch cords to connect the implanted optical fibers (200 µm diameter, ThorLabs) to the DPSSL laser diodes (473 nm, Laserglow, Canada) and employed heat-shrink tubing as well as black nail polish to further shield all optical connections and even the dental cement on the implant itself 68 .
Chemogenetic stimulations. Four weeks after the surgery, VLPO GAL -hM3Dq mice and VLPO GAL -mCherry mice were injected with either vehicle (saline) or CNO (0.3 mg/kg; Sigma, USA) and telemetric recordings of sleep-wake behavior, T b and LMA were performed for 24 h after each injection using Dataquest ART 4.1 software (Data Sciences International, USA) 35  For the novel cage experiments, mice were transferred immediately after i.p. injections to another cage (similar to the home cage to which the mice were habituated for at least a week) with new bedding and nesting material. These injections and cage transfer were conducted at 10:00 h. Saline and CNO injections and experimental conditions were randomized and there was at least 1 week between two CNO injections or two T a /novel cage conditions in the same mice. All recordings were performed in an environmental chamber (Powers scientific, USA) where T a was controlled within 0.5°C precision.
Histology. Upon completion of data collection, mice in the optogenetic stimulation experiments underwent a 1 Hz stimulation protocol for 2 h and were then deeply anesthetized (chloral hydrate; 700 mg/kg body weight) and transcardially perfused with PBS followed by 10% formalin. Mice in the chemogenetic experiments received CNO (0.3 mg/kg, i.p.) and were perfused after 3 h. Brains were cut into 3 series of 40 µm sections with one series immunolabeled for cFos (as a marker of neuronal activity; 1°antibody-rabbit (Rb) anti-cFos; Oncogene Sciences; cat. no: 4188; 1:30000 dilution) and DsRed (to label hM3Dq-mCherry or ChR2-mCherry-expressing neurons; 1°antibody-Rb-Anti-DsRed; Clontech, USA; cat.no: 632496; 1:10,000 dilution) 30,35 . Similarly, the brain sections from mice in the optogenetic inhibition experiments were labeled for cFos and GFP (to label the ArchT-expressing neurons; 1°antibody-Rb-Anti-GFP; Thermo Fisher Scientific cat. no: A11122; 1:10,000 dilution). cFos immunoreactivity was visualized with 0.06% DAB solution with 0.01% hydrogen peroxide, 0.01% nickel ammonium sulphate and 0.005% cobalt chloride, resulting in black nuclear staining. mCherry/ GFP immunoreactivity was then added in a second DAB staining step without nickel and cobalt, resulting in brown labeling of mCherry/GFP-expressing neurons. No mCherry/GFP labeling was observed in areas of the brain that did not receive AAV injections indicating the specificity of the DsRed and GFP antibodies.
In situ hybridization using RNAScope. We processed one series of sections for labeling with GAL mRNA by in situ hybridization as well as for mCherry/GFP by immunochemistry. We mounted our brain sections on Superfrost Plus slides in RNAs-free conditions and dried them in −20°C overnight. After warming, we further dried the slides in an oven for 30 min at 40°C and then performed the RNAscope hybridization using a RNAScope Multiplex Flourescent Reagent Kit V2 (Catalog #323100, Advanced Cell Diagnostics, Hayward, CA). We followed the provided instructions and pretreated the sections with hydrogen peroxide for 20 min at room temperature and then performed an antigen retrieval procedure by placing the slides in a steamer (at >99°C) for 5 min. We then dehydrated the sections in 100% alcohol and air-dried them for 5 min. Next, we treated the sections with protease reagent (Protease III, RNAscope) for 30 min at 40°C. After rinsing in sterile water, we incubated the sections in the RNAscope probe for Galanin-C1 (RNAscope® Probe-Mm-Gal;Cat No. 400961 Advanced Cell Diagnostics) for 2 h at 40°C for the hybridization step. Following the hybridization procedure, we performed three amplification steps at 40°C (AMP1-FL and AMP2-FL: 30 min each; AMP3-FL: 15 min) and subsequently incubated the sections in HRP blocker for 15 min. We then further incubated the sections in TSA plus Fluorescein/Cy5 fluorophore (Catalog # NEL741001, Perkin Elmer) for 30 min to visualize the GAL mRNA (Channel 1 at 488 nm). Since the mCherry or GFP fluorescence from the viral transfection was quenched by these procedures, we performed an additional immunolabeling step to re-establish the mCherry/GFP signals. We incubated the sections in Rabbit anti-DsRed/anti-GFP antibody (1:7500) at 4°C overnight, then washed them in PBS (2 × 2 min) and incubated them in a secondary antibody (Alexa fluor 555/488 Donkey anti-Rabbit, Thermo Fisher Scientific, Cat. no: A-31572 and A-21206) for 2 hours at room temperature. Finally, we washed the slides one more time before drying and then cover-slipped them with Vectashield mounting medium (Vector Laboratories, Catalog # H-1200).
Data analysis. EEG/EMG recordings were scored manually as 10 s epochs (optogenetic experiments) or 12 s epochs (chemogenetic experiments) into wake, NREM or REM sleep using SleepSign software (Kessei Comtec, Japan) 12,13,35 . Percentages of wake, NREM and REM sleep, bout numbers and average bout durations of individual sleep-wake states were calculated for 2 h before, during and after the photostimulations, or every 4 h or 2 h for 24-h after saline or CNO injections depending upon the experiment. EEG power spectra, specifically delta and theta power during individual sleep-wake states in the corresponding periods were also calculated. Mean T b was calculated every 5 min in all experiments except in Fig. 6 where 2 h means are presented. NREM and REM sleep latencies were calculated as the time taken to the first NREM and REM sleep episode, respectively, from the time of injection. Experimenters scoring sleep-wake data were blinded to the conditions.
In vitro chemogenetic activation of VLPO GAL neurons. GAL-Cre mice (n = 5) were injected with AAV-hM3Dq into the VLPO and 6 weeks after the injections, VLPO slices were collected and whole-cell current clamp recordings were performed from a total of five neurons 27,35 . After 15 min of stable recordings, artificial cerebrospinal fluid (aCSF) containing CNO (1 µM solution) was perfused through the chamber and recordings continued for another 5 min before washed out by aCSF 27,35 .
Statistical analysis. All statistical analyses were performed using GraphPad Prism (version 7.03; GraphPad Software, USA). For optogenetic experiments, we compared sleep-wake and T b data from 2 h during and 2 h immediately after laser stimulations at various frequencies in VLPO GAL -ChR2 mice and VLPO GAL -mCherry mice using two-way repeated measures (RM) ANOVAs followed by Sidak's post hoc comparisons; whereas, the EEG Power spectra were analyzed using Mann-Whitney tests. In vitro data were compared using a one-way ANOVA. For chemogenetic experiments, we compared sleep-wake and T b data after saline and CNO treatment from VLPO GAL -hM3dq mice and VLPO GAL -mCherry mice using two-way RM ANOVAs followed by Sidak's post hoc comparisons. Sleep latencies were analyzed using a Wilcoxon matched-pairs signed-rank test. Sample sizes for optogenetic and chemogenetic mouse cohorts were determined after pilot experiments with 2-3 mice revealed the effect size and variance between mice.